Senior SeriesEd\tor: leff Lasser Editorial Assistant; Erikka Adams Senior Marketing Manager: KellyMay Production Editor...
2459 downloads
12933 Views
8MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Senior SeriesEd\tor: leff Lasser Editorial Assistant; Erikka Adams Senior Marketing Manager: KellyMay Production Editor: RobertaShernnn Composition Btyer: Linda Cox Manufacturing Buyer: JoAnn e Sweeney Irrc' Services, Editorial production Servicesand Electronic Composition: Prtlllshtrs'L)esignorul Production Cover Administrator: Kristina Mose-Libon For relatedtitlesand supportmaterials,visitour online catlloq 'l[ \\ \\ \\ 'lhlttngtnan'com' Copyright a 2007,2004 PearsonEducation,Inc' or utilized in A1lrights resened.No part of the materialprotectedby this copr r-ightt]ottcr-mav be rcproduced information an,v or bv recorclinc' inq, including mechanical, or electronic means, Photo.(rF\ an,ufor,-,ror by any I r\iner. storageand retrievll system,rvithoutrvrittenpermissionfrpm th. !lf \ Iiiht and Bacon, To obtain permission(s)to use material from this work, pleasesubnrit .'.\\ rittE'nr!'qriestto Alh'n rL:rI!'.'rrc':tto 617-848-7320' \, f,rr t,t Ib 0ll NlA Bostotr, Street, Arlington L)epartment,75 Permissions for some sitesto have Betweenthe time Websiteinfbrmation is gatheredand then publishe.l.it i: rlot t-tttusual fhct'ublisherwouldappreciate r'rrLri:. closed.Also,thetranscriptionofURLscanresultintlpographical notification where theseerrors occur so that they may be correcteclirl sub\!'qrtenteditions. Cataloging-in-Publicationdata unavailableat presstime. 0-205-48.137,9
BRIEF CONT ENT S
PART ONE
Foundations
CHAPTERt
Doing Social Research
CHAPTERz
Theory and Social Research
CHAPTERs
Ethics in Social Research
cHAPTER+
Reviewingthe Scholarly Literature and Planninga Study
CHAPTERS
Qualitative and Quantitative Measurement
cHAPTERe
Qualitative and Quantitative Sampling
PARTTwo
ConductingQuantitativeResearch
CHAPTERZ
Survey Research
CHAPTER8
ExperimentalResearch
CHAPTER9
Nonreactive Researchand Secondary Analysis
CHAPTER1 o
PART THREr
Analysisof Quantitative Data
Conductin g Qualita t iv e Re s e a rc h
cHAPTER 1l
Field Rese a rch
cHAPTER I2
Historical-Comparative Research
CHAPTER 13 Analysisof QualitativeData
PAR T FO U R Writing a Rese a rchRe p o rt cHAPTER 14
Writing the ResearchReport
CONT ENT S
Preface
;;-;
xl
Ethicsand the Scientific Community 59
-- -'"-'-
Ethicsand the Sponsorsof Research 61
;-;
## Foundations
Politicsof Research
63
Value-Freeand Objective Research 64
CHA P T E R I
Doing SocialResearch
Conclusion
l
6G
Introduction
2 Alternativesto SocialResearch How ScienceWorks 7 Steps in the Researchprocess .l
Dimensionsof Research Conc lus ion 21
3 9
Introduction
O
69 Usingthe Internet for Social Research 80
Theory and SocialResearch
23
24
What ls Theory?
24
C oncl usi on
Qualitativeand Quantitative Measurement 108
44
45
Why Measure?
CHA P T E R 3
Ethicsin SocialResearch Why Be Ethical?
48 48
Power Relations 49 EthicallssuesInvolvingResearch Participants S0
9l
106
C H A P TE R 5
Introduction
Introduction
gg
QuantitativeDesignlssues 26
The AspectsofTheory Zg The Three Major Approaches to Social Science 41 Conc lus ion
Qualitativeand euantitative Orientations toward Research g4 QualitativeDesignlssues
The Parts ofTheory
The DynamicDuo
69
Literature Review
C HA P T E R 2
lntroduction
C H A P TE R 4
Reviewingthe ScholarlyLiteratureand Planninga Study 68
47
'l
09 I 09
Quanti tati veand eual i tati ve Measurement II0 Partsof the Measurement Process 1 'l 't R el i abi l i tyand V al i di ty
IIs
A Guideto Quantitative Measurement I 2l vil
VIII
CONT ENT S
lndex Construction Sc a l e s
126
Resultsof ExperimentalResearch:Making 219 C ompari sons
128
Conclusion
A Word on Ethics
138
C oncl usi on
221
222
C H A PT E R 5 C H A P TE R 9
Qualitativeand Quantitative S amp lin g 140 fntroduction
141
N o n p ro b a b i l i tyS a m p l i ng
225
fntroduction
141
Pro b a b i l i tyS a m p l i n g Conclusion
NonreactiveResearchand Secondary Analysis 224 NonreactiveMeasurement
145
227
Content Analysis
164
225
ExistingStatistics/Documents and 236 SecondaryAnalysis lssuesof lnference and Theory Testing 244
PA R T T W O
l:i::::: Conducting Quantitative Research
Conclusion
C H A P TE R I O
C H AP T ER 7
SurveyResearch l n tro d u c ti o n
Analysisof Quantitative 247 Data
155 167
' r5 8
The Logicof SurveyResearch
169
Constructingthe Questionnaire Typesof Surveys:Advantagesand Disadvantages 1 86 lnterviewing
Dealingwith Data
248
R esul tsw i th One V ari abl e More Than Two Variables lnferentialStatistics
196
C oncl usi on
197
251 257 263
268
272
PART THREE
C H A PT E R 8
ExperimentalResearch Introduction
248
fntroduction
Resultswith Two Variables
190
The EthicalSurvey Conclusion
245
200
ii? ConductingQualitative Research
2O4
C H A P TE R I 1
201
Random Assignment
2O2
ExperimentalDesignLogic fnternal and ExternalValidity PracticalConsiderations
212 219
FieldResearch fntroduction
275 27 5
C ON TEN TS
The Logicof FieldResearch
Z7g
c hoos inga s i te a n d' Ga ining ---.....D
Access
Relationsin the Field
Leavingthe Field FocusGroups
Introduction
2gS
Observingand CollectingData The Field ResearchInterview
287 296
300
Other Techniques 339 Softwarefor eualitative Data Conclusion 342
3OZ
CHA P T E R I 2
3O4
Data and Evidencein Historical Context 312 317
\{riting
the Research Report
Introduction
The ResearchReport
Glossary Bibliography
Ethics
Name Index 325
344
Conclusion
Equivalencein Historical-Comparative Research 322 Conclusion
34O
C H A P TE R 1 4
steps in a Historical-comparativeResearch Project 31 0
325
32g
PART Fo u R iti\ Writinga Research Report
The Logicof Historical-Comparative Research iO5
comparative Research
ComparingMethods of Data Analysis 328 AnalyticStrategiesfor Qualitative Data 335
Historical-comparative Research 3O4 fntroduction
327
329
Coding and Concept Formation
2gg
EthicalDilemmasof Field Research 301 Conclusion
C H A P TE R I 3
Analysisof eualitative Data
280
rx
Subject Index
359 36'l 377 391 3g3
344
343
CHAPTER 1
DoingSocialResearch
Introduction Alternatives to Social Research Authority Tradition C o m m o nSe n se Me d i aMy th s P e rs o n aEl x p eri ence How ScienceWorks Sc i e n c e T h e S c i e n ti flCommuni c ty The Scientific Methodand Attitude JournalArticlesin Science Steps in the ResearchProcess Dimensionsof Research Useof Research Pu rp o s e o fa Study T i m eD i me n s i on i n R esearcn D a taC o l l e c ti o n Techni ques Conclusion
PART ONE
,/ F OUND A TION S
IN T R OD U C T ION Socialresearchis all arounclus. Educators,go\,ernment officials, businessmanagers,hutntrrt serviceproviders,and health care professionals regularlyuse socialresearchmethods anclfindings.Peopleusesocialresearchto raisechildren, reducecrime, irnprclvepublic health,sellproducts,clrjust understandone'slil-e.Reportsof researchappearon brclaclcast ne$,sprograms,it.t popuiar magazines,in nen,spapers, and on the lnternet. Researchfindings can aftbct people'sdaily livesand public policies.For example,I recentll' heard a debateregrrrclirrg a U.S.lederaigovernment program to off-erteenagerssexualabstin e n c e c o u n s e l i n g .A h i gh-l evel government official argued for sucl-rcounselingancl stror"rgl,v opposed offering teensbirth control inforirati o n . A n i n d e p e n c l e n th eal th admi ni strator noted that there is no scientiflcevidenceshou,ing that abstinence-onlycounselingrvorks.Iie saidthat 80 percentoftcens arc alreadysexually activebythe ageof 18,thereforeit is essentialto provide birth control information. Hc prointed to ma n y re s e a rc hs tu c l i esshorvi ngthi rt bi rth prcsnJncv control instruction for tecnslecluce's ratesand the spreadof scrr.rallrtr'.rr.rsnrittcti t1i:e a s e sT. h e g o v e rn mc n ta Lrsti ncr.tcc or.i l r'.rrl rocatereiied on rnoral persurrsionLrcc.rr.rsc hc ir.r11 no researcheviclencc.lcieoloqr,iirith, .urcipoliti c s s h a p ema n \' g o v e rn n rentprogranrsrather than solid researchcvidcnce,br,rtgood socialrcsearchcan help all of us make inlbrnred decisions. The evidence also expltrins wh,y m.ury programs fail to accornplishmr.rchor m;ry do more harm than gooc1. This book is about socialresearch.In simple tenxs, researchis a way of going about finding answersto questions.Prof-essors, prof-essional researchers, practitioners,and studentsin many fields cor-rdr.rct researchto scekansr\rersto questions about the socialworld. You probably already have some notion clf whirt social reseal'ch entails.First,Iet me end sornepossiblemiscor.t-
When i askedstudentsin my classes ceptior.rs. ufiat they think socialreseirrchentails,they gave the lbllolr,ir-rgans\\rers: r r r r r r
r
r
It is basedon firctsalone; there is no theory juclgrlrent. or prersonal Onlr.expertsu,ith a Ph.D. degreeor college professorsread it or do it. It meansgoing to the library and finding a lot of magazinearticlesor bookson a topic. It is r,r,hensomeonehangsaround a group and observes. It means conducting ir controlled experinlent. Socialresearchis clrawinga sampleof peoto cornple and givingthem questionnaires plete. It is looking up lots of statistical tables from oftrcial government and intbrn-ratior-r reports. Llsecomputersto create To do it, onc r.r.rr-Lst charts,and graphs. stati sti cs,
The flrst t\\'o .urs\versare wrong, and the of what constitutessoothe-r's clcscribe or.rl.,.part to confuseone part cia] r'escarch.It is r-rr.rlvise rr ith the n.hole. P eopl e conduct soci al resear cht o lear n nerv about the social world; or to sorr-rething hunches,or beliefs carefullydocunrentguesses, about it; or to refinetheir understandingofhow the sociaiworld works. A researchercombines tl-reoriesor ideaslr,ith facts in a careful, systematic rvay and usescreativity. He or she learns to orgirnizeand plan carefully and to selectthe appropriatetechniqr-re to addressa specifickind of alsomust treatthe people question.A rescarcher in a studi,in ethicaland moral ways.In addition, a researchern-rustfirlly ilnd clearlycommunicate the resultsofa study to others. Socialreseirrchis ir processin which people combine a set of principles,outlooks,and ideas (i.c.,methodology)rvith a collectionof specific practi ces,techni ques,and strateg ies( i. e. , a rnethoclof inquiry) to produceknorvledge.It is
C H AP TE R 1 ,/ D OIN C SOC IA L R E S TdR C -
an exciting processof discovery,but it requires persistence,personalintegrity, tolerancefor ambiguity, interaction with others,and pride in doing qualiry work. Reading this book cannot transform you into an expert researcher,but it can teachyou to be a better consumerof researchresults,help you to understandhow the researchenterpriseworks, and prepareyou to conduct small researchprojects.After studying this book, you will be aware of what researchcan and cannot do, and why properly conductedresearchis important.
position of authority saysit is true or bec.ri:.c.. is in an authoritativepublication, )'ou are relvrn:on authority as a basisfor knowledge.Relr'inc on the wisdom of authoritiesis a quick, sirtrl.]3. and cheapway to learn something.Authoi-itie. often spend time and effort to learn son.rething. and_you can benefit from their experienceand worK. There are also limitations to relying on allthority. First, it is easyto overestimatethe expertise of other people.You may assumethat thev are right when they are not. History is full of past expertswhom we now seeasbeing misinformed. For example>some "experts" of the past n-reasured intelligenceby counting bumps on the AL T E RNA T I V E S T O S OC IA L skull; other "experts" usedbloodletting to try to R E S E A RCH cure diseases.Their errors seem obvious norl,, Unlessyou are unusual, most of r,vhatyou know but can you be certain that today's expertswill about the socialworld is not basedon doins sonot becometomorrow's fools?Second,authorrcial research.You probably learnedmost of i,hat ties may not agree,and all authoritiesmay not be you know using an alternativeto socialresearcl-r. equally dependable.Whom should we believeif It is basedon what your parentsand other people authorities disagree?Third, authorities may (e.g.,friends, teachers)have told you. You also speakon fields they know little about or be plaru have knowledgebasedon your personalexperiwrong. An expert who is very informed about ences,the books and magazinesyou have read, one area may use his or her authority in an unand the movies and televisionyou havewatched. related area. Also, using the halo effect (disYou may also useplain old "common sense." cussedlater), expertisein one areamay spill over More than a collection of technioues,social illegitimatelyto be authority in a totally different researchis a processfor producing knowledge.It area.Have you everseentelevisioncommercials is a more structured, organized,and systematic where a movie star useshis or her fame as auprocessthan the alternativesthat most of us use thority to convinceyou to buy a car?We r.reecl tcr in daily life. Knowledge from the alternativesis ask:Who is or is not an authority? often correct, but knowledge basedon research An additionalissueis the misuseof authoris more likely to be true and have fewer errors. ity. Sometimes organizations or indii'iduals Although researchdoesnot alwaysproduce pergive an appearanceof authority so thev ciut coltfect knowledge,compared to the alternativesit is vince others to agree to something that thel' rnuch lesslikely to be flawed. Let us review the might not otherwiseagreeto. A relateclsituation alternativesbefore examining socialresearch. occurswhen a personwith little trairringi.rndexpertiseis named as a "seniorfbllon"'or."adiunct scholar" in a private "think trrnk" r,ith .in inrAuthority pressivename, such as the Center tbr the Study You have acquired knowledge from parents, of X or the Institute on Y Research. Somethink teachers,and expertsaswell as from books, teletanks are Iegitimateresearchcenters,but rnany vision, and other media. When you accept are mere fronts createdby,.rveirlthv special-intersomething as being true becausesomeone in a estgroupsto engagein advocao.politics.Think
o a RT O\E
\)rl
o\5
t.r1rk:a,ii t.r.,ri .it]\r)uai.t"scholar"to facilitare ti rc :r...' . r' i r..l .,r.i a a c p ti n qthe pcrsonas al t auti t!j t.i i ' .L rtt.:tt l :rL l c .l rt real i ty,the perS onmay ir(rtir.r,.I .rir, l...rl!'\pertise.lAlso, too much re:r.r:ta !)i !' t ,tu th ()fl ti c sc a n b e dangeroustO a dei ]]r,.i .rti c .o c re t)' .E x p e rtsmay promote i deas il.l.rt:trr'nqthentheir own power and position. \\'ltcrt rrc'.lcCeptthe authority of experts,but do rr)t kuo\\' l.rolvthey arrived at their knowledge, rlc Ltrscthe ability to evaluatewhat the experts savancilosecontrol of our destiny. T ra d i ti o n Peoplesometimesrely on tradition fbr knowledge.Tradition is a specialcaseof authoritythe authority of the past. Tradition means you acceptsomethingasbeing true because"it's the \\ray things have always been." For example, my father-in-law saysthat drinking a shot of u'hiskeycures a cold. When I askedabout hrs statement,he said that he had learnedit from his father when he was a child, and it had come down from past generations.Tradition lvas the basisof the knowledge for the cure. Here is an examplefrom the socialworld: Many peoplebelieve that children who are raised at home by their mothers grow up to be better adjustedand have fewer personalproblems than those raised in other settings.People "know" this, but how did they learn it? Most acceptit becausethey believe (rightly or wrongly) that it rvastrue in the past or is the way things have alwaysbeen done. Sorne traditional social knowledge begins as simple prejudice.You might rely on tradition rvithor-rtbeing fully awareof it with a belief such as "Peoprlefrom that side of the trackswill never arnoLlntto anvthing" or "You never can trust that tvpe oi person"or "That's the way men (or u,omen) are." El,enif traditional knowledgewas oncetrue, it can becomedistortedas it is passed on, and soon it is no longer true. Peoplemay cling to traditional knowledgewithout real understanding;they assumethat becausesomething may haveworked or been true in the past, it rvill continueto be true.
C ommon S ense You knorv a lot about the socialworld from your everydayreasoningor cornmon sense.You rely on what everyoneknows and what "just makes sense."For example,it "just makessense"that murder rates are higher in nations that do not have a death penalty, becausepeople are less likely to kill if they face execution for doing so. This and other widely held commonsensebeliefs, such as that poor youth are more likely to commit deviantactsthan thosefrom the middle classor that most Catholics do not use birth control,are false. Comrnon senseis valuable in daily living, but it irilolvslogicalfallaciesto slip into thinking. For example,the so-calledgambler'sfallacysays: "If I hai'e a long string of lossesplaying a lottery, the nert tine I play, my chancesof winning lvill be better."In terms of probabilityand the facts, this is Ialse.Also, cclmmon sensecontainscor.rtradictory ideastl-ratoften go unnoticed because people use the ideas irt different times, such as "opposites attract" and "birds of a feather flock together."Common sensecan originate in tradition. It is useful and sometimescorrect,but it alsocontainserrors,misinformation,contradiction, and prejudice.
Media Myths Television shows, movies, and newspaper and magazinearticiesare important sourcesof information. For example, most people have no contactwith criminalsbut leam about crime by watching televisionshows and movies and by reading newspapers.However, the television portrayalsof crime, and of many other things, do not accuratelyreflect socialreality. The writers who create or "adapt" images from life for television sholvsand movie scripts distort reality either out of ignorance or becausethey rely on authority, tradition, and common sense. Their primary goal is to entertain,llot to representreality accurately.Although many journaliststry to presenta realisticpicture of the world,
C H AP TE R 1 /
they m us t wr it e s to ri e si n s h o rt ti m e p e ri o d s rvith limited information and within editorial eLridelines. Unfortunately, the media tend to perpetuate the mlths of a culture. For example,the media show that most people ivho receivewelfare are Black (actually, most are White), that most peoplewho are mentally ill are violent and dangerous (only a small percentageactually are), and that most peoplewho are elderlyare senile and in nur s ing ho m e s (a ti n y rn i n o ri ty a re ). AIso, massmedia "hype" can createa f-eelingthat a major problem existswhen it may not (seeBox Li). People are n-risledby i.isual imagesmore easilythan other forms of "lying"; this means that storiesor stereotypesthat appear on film and televisioncan havea porverful effecton people. For example,television repeatecllvshorvs low-income,inner-city,AfrictrnAntericanr.outl-r using illegal drugs. Eventuallr.,nlost peol)lg "know" that urban Blacksuse illegalclrLres at a higher rate than other groups in the Lrrritccl States,even though this notion is false. Competing interestsuse the rnedia to rvin public support.2Public relationscampaignstry to alter what the public thinks about scientific findings, making it difficult for the public to judge researchfindings. For exarnple,a large majority of scientific research supports the global rvnrrnir-rgthesis (i.e., pollutants from industrialization and massivedeforestationare raising the earth's temperatureand lvill cause drarnaticclimate changeand bring about environmental disasters).l'he scientificevidenceis growing and getsstrongereachyear.l'he media give equal attention to a few dissenterswho question global r,r,arming,creating the impressi o n in t he public n ti n c l th a t " n o o n e re a l l y knows" or that scientistsare undecidedabout the issueof global warming. The rnedia sources fail to mention that the dissentersrepresentless th an 2 per c ento1' a l ls c i c rrti s tso ,r th l t rn o s td i s sentingstudiesare paid for by heavilypolluting industries.The industriesalsospendmillions of dol l ar st o public iz eth e fi n d i n g s b e c a u s eth e i r goalis to cleflectgrowing criticismand delayen-
D OIN C S OC IA L R E S E A R C T
ls Road Rage a Media Myth? Americanshear a lot about road rage.Newsweekmagazine,Timemagazine,and newspapersin most major c i t i e s h a v e c a r r i e d h e a d l i n e sa b o u t i t . L e a d i n gn a tional politicalofficialshave held public hearingson it, and the federalgovernmentgives millionsof dotlars in grants to law enforcementand transportation departmentsto reduceit. Today, even psychologists specialize in thisdisorder. The term road rage first appearedin I 988, and by 1997, the print mediawere carryingover 4,000 articlesper year on it. Despitemediaattention about "aggressivedriving" and "anger behind the wheel," there is no scientificevidencefor road rage.The term i s n o t p r e c i s e l yd e f i n e d a n d c a n r e f e r t o a n y t h i n g from gunshots from cars,use of hand gestures,running bicyclistsoff the road,tailgating,and evenanger o v e r a u t o r e p a i rb i l l s !A l l t h e d a t a o n c r a s h e sa n d a c c i d e n t s s h o w d e c l i n e sd u r i n g t h e p e r i o d w h e n r o a d r a g e r e a c h e da n e p i d e m i c . Perhapsmediareportsfueled perceptionsof road rage. After hearingor readingabout road rage and havinga labelfor the behavior,people began to notice rude drivingbehaviorand engagedinselective observation. We will not know for sure until it is properry s t u d i e d ,b u t t h e a m o u n to f s u c h b e h a v i o rm a y b e u n changed.lt may turn out that the nationalepidemic of road rage is a widely held myth stimulatedby reports in the massmedia.(For more information,see Michael Fumento, "Road Rage versus Reality," AtlanticMonthly[August 1 998].)
vironrnentalregulations,not to advanceknowredge. Newspapersoffer l.roroscopes, ar.rdtelevision programsor nroviesreport on supertratural powers, E,SP(extrasensoryperception), LIFC)s (uni denti fi ed fl ,vi ng obj ects), ancl angel s or ghosts.Althor-rghr.roscientifice".icience existsfor such,betrveen2-5and -50percer-rt of the U.S.pubIic acceptsthern astrue, anclthe percentage with
6
pA RToNE ,/ F o u N D A T to N s
such beliefshas beengrowing over time as the entertainment media give the phenomenon more prominence.3
tive to featuresthat confirm what we think, but ignorefeaturesthat contradictit. For example,I believetall peopleareexcellentsingers.This may be becauseof stereotypes, what mi mother told me, or whatever.I observetall peopleand,with_ Personal Experience out awareness, pay particular attention to their If somethinghappensto you, if you personally singing.I look at a chorusor top vocalistandno_ seeit or experienceit, you acceptit astrue. per_ ticethosewho aretall. Withoui realizingit, I no_ sonalexperience,or ..seeingis te[eving,,, hasa tice and rememberpeople and situat[ns that strong impact and is a powerful sJurce of reinforcemy preconceivedideas.psychologists knowledge.Unfortunately,personalexperience found that peopletend to ..seekout" and dirtort c.an.l9.1d you astray.Somethingsimilarto an op_ their memoriesto make them more consistent tical illusion or miragecunoci.rr. What appears with what they alreadythink.a ! true.may actuallybe due to a slight oi dir_ A third error is prematureclosure.It often tortion in judgment. The powei of"r.o, immediacy operateswith and reinforcesthe first two errors. and direct personalcontaciis very strong.Even Premature closure occurs when you feel you knowing that, people fall for illusions."Many havethe answerand do not needio listen,seek peoplebelievewhat they seeor personallyexpe_ information, or raisequestionsany longer.Un_ riencerather than what very carefullydesiened fortunately,most of us arealittJelazvor"set a lit_ research hasdiscovered. tle sloppy.We take a few piecesof'evid"ence or The four errorsofpersonalexperiencerein_ look at eventsfor a short while and then think forceeachother and canoccurin other areas, as we haveit figured oul We look for eyidenceto well. They are a basis for misleading people confirm or rejectan ideaand stop when a small through propaganda, cons or fraudl magic, amount of evidenceis present.In a word, we stereotyping,and some advertising.The mtst jump to conclusions.For example,I want to frequentproblemis overgeneratization;it occurs learnwhetherpeoplein my town support Mary when someevidencesupportsyour belief, but Smith or |on Van Horn for mayor.t uit ZOp.o_ you falselyassumethat it appliesto many other ple; t 6 saytheyfavor Mary,2 areundecided,and situations,too. Limited generalization be only2 favorlon, soI stopthereandbelieveMary appropriate;under certainconditions,-uv a small will win. amount of evidencecan explain a larger situa_ Anothercommonerror is thehalo tion. The problem is that many peoplJgeneral_ ffict; itis r1eovergeneralize yhen from what we acceptas ize far beyond limited evidence.For eiample, being highly positive or prestigiousand let its over the years,I haveknown five blind p"opl.. strong reputation or prestige..rub off' onto All of them werevery friendly. Can t conclude other areas.Thus,I pick up a report by a person that all blind people are friendly? Do the five from a prestigiousuniveisity,sayHan ard or peoplewith whom I happenedtohave personal CambridgeUniversity.I assumethat the author experiencewith representall blind peopie? is smartand talentedand that the report will be The seconderror, selective obirvition, oc_ excellent.I do not makethis assumption abouta curswhenyou takespecialnoticeof somepeople reportby someonefrom UnknownUniversity.I or eventsand tendto seekout eviderr.ethut con_ form an opinion and prejudgethe report #d fir1s whal you alreadybelieveu"a ig"o." .orr_ may-notapproachit by consideringits own mer_ tradlctorFlnformation.peopleoften focuson or rts alone.How the variousalternativesto social observeparticularcasesor situations,especially researchmight addressthe issueof laundry is when they fit preconceivedideas.W. uri ,*i_ shownin Table1.1.
CHA P TE R 1 ,/ D OIN G S OC IA L R ES E A R C H
TABL E I . 1
A lt er n a ti v e s to S o c i a l Research
processes. This suggeststhat we examinethe meaningof scienceand how its works. Science
Authority
Expertssaythat as children, femalesare taught to make, select,mend,andcleanclothing as part of a femalefocuson physicalappearance andon caringfor childrenor othersin a family.Womendo the laundry basedon theirchildhood preparation.
Tradition
Womenhavedonethe laundrv for centuries,so it is a continuation of whathas happened for a longtime.
CommonSense Menjust arenot as concerned aboutclothingas muchas women,so it only makessense that womendo the laundrv moreoften. MediaMyth
Television commercials show womenoften doing laundryand enjoyingit, so they do laundry becausethey think it's fun.
Personal Experience
My motherand the mothersof all my friendsdid the laundry. My femalefriendsdid it for their boyfriends,but neverthe other wayaround.lt just feelsnatural for the womanto do it.
HOW SCIENCEWORKS Although it builds on someaspectsof the alternativewaysof developingknowledge,scienceis what separates socialresearch.Socialresearch involvesthinking scientificallyabout questions about the socialworld and following scientific
suggestsanimageof testtubes, The term science computers,rocketships,and peoplein white lab coats.Theseoutwardtrappingsarea part ofscience,especiallynatural science(i.e.,astronomy, biology, chemistry,geology,and physics,),that dealswith the physicaland materialworld (e.g., plants, chemicals,rocks, stars,and electricity). The socialsciences,such as anthropology,psychology,politicalscience,and sociology,involve the study of people-their beliefs,behavior,interaction,institutions,and so forth. Fewerpeople associatethesedisciplineswith the word Scienceis a socialinstitution and a way science. to produceknowledge.Not everyoneis well informed about science.For example,a 2001survey found that about only one-third of U.S. adults could correctly explain the basicsof sclence." Scientistsgatherdatausingspecialized techniquesand usethe datato supportor rejecttheories. Data are the empirical evidence or information that one gatherscarefullyaccording to rules or procedures.The data can be quantitative(i.e., expressedas numbers) or qualitative(i.e., expressedas words, visual images,sounds,or objects).Empiricaleyidence refersto observationsthat peopleexperience throughthe senses-touch,sight,hearing,smell, and taste.This confusespeople,becauserecannotusetheir senses searchers to directly observemany aspectsof the social world about which they seekanswers(e.g.,intelligence,attitudes,opinions, feelings,emotions,power, auhavemanyspecialized thority, etc.).Researchers techniquesto observeand indirectly measure suchaspectsof the socialworld. The Scientific Community Sciencecomesto life through the operation of the scientificcommunity,which sustainsthe as-
PART ONE , / FO UNDATI O NS
sumptions,attitudes,and techniouesof science. Thescientificcommunityisa collectionof people who practicescienceand a setof norms,6ehaviors,and attitudesthat bind them together.It is a professionalcommunity-a groupof interacting peoplewho shareethicalprinciples,beliefsand values,techniquesandtraining,and careerpaths. For the most part, the scientificcommunity includesboth thenaturalandsocialsciences.6 Many people outside the core scientific community usescientificresearchtechniques.A rangeof practitionersand techniciansapply researchtechniquesthat scientistsdevelopedand refined.Many usethe researchtechniques(e.g., a survey)without possessing a deepknowledge of scientificresearch.Yet, anyonewho usesthe techniques or resultsofsciencecando sobetter if they also understand the principles and processes of the scientificcommunity. Theboundariesof the scientificcommunity and its membershiparedefinedloosely.Thereis no membershipcard or masterroster.Many peopletreata Ph.D.degreein a scientificfield as an informal "entryticket" to membershipin the scientificcommunity. The ph.D., which stands for doctorate of philosophy, is an advanced graduatedegreebeyond the master'sthat prepares one to conduct independentresearch. Someresearchers do not haveph.D.sand not all thosewho receivePh.D.senter occupationsin which they conduct research.They enter many occupationsand may haveother responsibilities (e.g.,teaching,administration,consulting,clinical practice,advising,etc.).In fact, about onehalf of the peoplewho receivescientificph.D.s do not follow careersasactiveresearchers. At the coreof the scientificcommunity are researchers who conduct studieson a firll-time or part-time basis,usuallywith the help of assistants.Many researchassistants aregraduatestudents,and someare undergraduates. Working asa researchassistantis the waythat most scientistsgain a real graspon the detailsof doing research.Collegesand universitiesemploy most membersof the scientificcommunity's core. Somescientistswork for the governmentor pri-
:'
_- --:
vate industry in organizationssuch as the National Opinion ResearchCenter and the Rand Corporation.Most, however,work at the approximately200 researchuniversitiesand instituteslocatedin a dozenadvancedindustrialized countries. Thus, the scientific community is scatteredgeographically, but its memberstend to work togetherin smallclusters. Howbigis the scientificcommunity?This is not an easyquestionto answer.Usingthe broadestdefinition (including all scientistsand those in science-related professions,such as engineers),it includesabout 15percentof the labor'+, force in advancedindustrialized countries. A betterway to look at the scientificcommunity is to examinethe basicunit of the largercommunity: the discipline(e.g.,sociology,biology,psychology,etc.).Scientistsaremost familiar with a ." particular disciplinebecauseknowledgeis specialized.Compared to other fields with advancedtraining,the numbersarevery small.For example,eachyear,about 500 peoplereceive Ph.D.sin sociology,16,000receivemedicaldegrees,and 38,000receivelaw degrees. A disciplinesuch as sociologymay have about 8,000activeresearchers worldwide.Most researchers completeonly two or three studies in their careers,whereasa small number of highly activeresearchers conductmany dozens ofstudies.In a specialtyor topic area(e.g.,study of the death penalty, social movements,divorce), only about 100 researchers are very active and conduct most research studies. Aithough researchresultsrepresentwhat humanity knows and it hasa major impact on the lives of many millions of people,only a small number of peopleare actuallyproducing most new scientificknowledge. The Scientific Method and Attitude You have probably heard of the scientific method,and you may be wonderinghow it fits into all this. The scientificmethodis not one singlething; it refersto the ideas,rules,techniques, and approaches that the scientificcommunity
.I C H A P T E R , / D O I N C S O C I A LR E S E A R C H
uses.The method arisesfrom a looseconsensus within the community of scientists.It includesa way of looking at the world that placesa high valueon professionalism, craftsmanship,ethical integrity,creativity,rigorousstandards,and diligence.It alsoincludesstrongprofessional norms such as honestyand uprightnessin doing research,great candor and opennessabout how one conducteda study,and a focuson the merits ofthe research itselfandnot on any characteristicsof individualswho conductedthe study.
regularly reject half of the submissions.Thus, severalexperiencedresearchers screena journal articlebasedon its meritsalone,and publication represents the study'stentativeacceptance by the scientificcommunity as a valid contribution to knowledge.Unlike the authorsof articlesfor the popular magazinesfound at newsstands, scientistsarenot paid for publishingin scholarlyjournals.In fact, they may haveto pay a small feeto help defraycostsjust to havetheir papersconsidered. Researchers are huppy to make their research available to their peers (i.e., other scientistsand researchers)through scholarly fournal Articles in Science journals.The articlecommunicatesthe resultsof Considerwhat happensonce a researcherfin. a study that a researchermight have devoted ishesa study.First,he or shewritesa detaileddeyearsof his or her life to, and it is the way rescription of the study and the results as a searchers gain respectand visibility amongtheir researchreport or a paperusinga specialformat. professionalpeers.Likewise,the reviewersare Often,he or shealsogivesan oralpresentationof not paid for reviewingpapers,but considerit an the paper beforeother researchers at a conferhonor to be askedto conductthe reviewsand to enceor a meetingof a professionalassociation carryrout one of the responsibilitiesof being in and seekscommentsand suggestions. Next, the the scientificcommunity.Thescientificcommuresearcher sendsseveralcopiesto the editor ofa nity imparts great respectto researcherswho scholarlyjournal. Eacheditor, a respectedrepublish many articlesin the foremostscholarly searcherchosenby otherscientiststo overseethe journals becausetheseresearchersare directly journal, removesthe title page,which is the only advancingthe scientific community's primary place the author's name appears, and goal-the sends the accumulationof carefullydeveloped . articleto severalreviewers.The reviewersarereknowledge. A researchergains prestige and spectedscientistswho haveconductedstudiesin honor and a reputationas an accomplishedrethe samespecialtyareaor topic. The reviewers searcherthrough suchpublications. do not know who did the studn and the author You may neverpublishan articlein a scholof the paper doesnot know who the reviewers arly journal, but you will probably read many sucharticles.It is important to seehow they are ' are. This reinforcesthe scientificprinciple of judging a study on its merits alone.Reviewers a vital componentin the systemof scientificreevaluatethe researchbasedon its clarity, origisearch.Researchers activelyreadwhat appearsin nality, standardsof good researchmethods,and thejoumalsto learnaboutnewresearchfindings advancingknowledge.Theyreturn their evalua- and the methodsusedto conducta study.Eventions to the editor,who decidesto rejectthe patually, the new knowledgeis disseminatedin per, askthe author to reviseand resubmitit, or textbooks,new reports,or public talks. acceptit for publication.It is a very careful,cautious methodto ensurequality control. The scholarlyjournals that are highly reSTEPSIN THE RESEARCH spectedandregularlyreadby mostresearchers in PROCESS a field receivefar more papersthan they canpublish.Theyacceptonly 10to 15percentof submitSocialresearch proceedsin a sequence ofsteps, ted manuscripts.Even lower-ranked iournals althoughvariousapproaches to researchsuggest
10
P A RToNE ,/ F o u N D A T to N s
slightly different steps.Most studiesfollow the sevenstepsdiscussed here.To beginthe process, you selecta topic-a generalareaof study or issue,suchas domesticabuse,homelessness, or powerful corporateelites.A topic is too broad for conductinga study.This makesthe next step crucial.You must then narrow down the topic, or focusthe topic into a specificresearchquestion for a study (e.g.,"Are peoplewho marry youngermore likely to engagein physicalabuse of a spouseunder conditionsof high stressthan thosewho marry older?").As you learn about a topic and narrow the focus,you should review past research,or the literature,on a topic or question.You alsowant to developa possible answer,or hlpothesis,andtheorycanbe important at this stage. After specifringa researchquestion,you have to developa highly detailedplan on how you will carry out the study.This third steprequiresthat you decideon the manypracticaldetails of doing the research(e.g.,whetherto usea surveyor qualitativeobservingin the field, how many subjectsto use,etc.).It is only after completing the design stagethat you are ready to gatherthe data or evidence(e.g.,askpeoplethe questions,recordanswers,etc.).Onceyou have very carefirllycollectedthe data,your next stepis to manipulateor analyzethedata.This will help you seeany patternsin it and help you to give meaningto or interprefthe data (e.g.,"People who marry young and grew up in families with abusehave higher rates of physical domestic abusethan those with different family histories").Finally,you must inform othersbywriting a report that describesthe study'sbackground, how you conductedit, andwhatyou discovered. The seven-step processshownin Figure1.1 is oversimplified. in practice, you will rarely completeone steptotally then leaveit behindto move to the next step.Rather,the processis interactive in which the steps blend into each other.What you do in a laterstepmay stimulate you to reconsiderand slightlyadjustyour thinking in a previousone.The processis not strictly linearandmay flow backand forth beforereach-
FIcURE t . I Stepsin the Research Process
ing an end.The sevenstepsarefor one research project;it is one cycleofgoing through the steps in a singlestudyon a specifictopic. Scienceis an ongoingenterprisethat builds on prior researchandbuildsa larger,collectively createdbody of knowledge.Any one study is a smallpart of the much largerwholeof science.A singleresearchermay be working on multiple researchprojectsat once,or severalresearcher may collaborateon one project. Likewise,one projectmay resultin onescholarlyarticleor several,and sometimesseveralsmallerprojectsare reportedin a singlearticle.
DIMENSIONS OF RESEARCH
Three yearsafter they graduatedfrom college. Tim and Sharon met for lunch. Tim asked Sharon,"So,how is your newjob asa researche for SocialData, Inc.? What are you doing?" Sharonanswered,"Right now I'm working on an appliedresearchprojecton daycarequalityin which we'redoinga cross-sectional surveytoget descriptivedatafor an evaluationstudy."sharon
touchedon four dimensionsof socialresearchas shedescribedher researchon daycare. Socialresearchcomesin severalshapesand sizes.Beforeyou begina study,you will needto makeseyeraldecisionsaboutthe specifictypeof researchyou are going to conduct.Researihers need to understandthe advantagesand disadvantagesof each type, although most end up specializingin doing one tFpe.We can think of the typesasfitting into one of the categoriesin eachof four dimensionsof research. The first dimensionis a distinction in how researchis used,or betweenappliedandbasicresearch.Thenextis the purposeofdoing research, or its goal,to explore,describe,or explain.The nexttwo dimensionsaremorespecifiqhow time is incorporated into the study design,and the specificdatacollectiontechniqueused. The dimensionsoverlap,in that certaindimensionsareoften found together(e.g.,the goal of a studyanda datacollectiontechnique).Once you learn the dimensions,you will beginto see how the particularresearchquestionsyou might want to investigatetend to be more compatible with certainwaysof designinga study and collectingdata.In addition, being awareof the dimensions of researchwill make it easier to understandthe researchreportsby others.
Bnsic Research. Basicsocialresearchadvances fundamentalknowledgeabout the socialworld. Basicresearchers focuson refuting or support_ ing theoriesthat explain how the socialworld operates,what makesthingshappen,why social relations are a certain way, and why society changes. Basicresearchis the sourceofmost new scientificideasand waysof thinking about the world. Many nonscientists criticize basic re_ searchand ash "What good is it?" and consider it to be a wasteof time and money.Althoughba_ sicresearchoften lacksa practicalapplicationin the short term, it provides a foundation for knowledge that advancesunderstanding in many policy areas,problems,or areasof study. Basicresearchis the sourceof most of the tools, methods,theories,and ideasabout underlying clus-esof how peopleact or think usedby ap_ plied researchers. It providesthe major bieak_ throughsthat significantadvances in knowledge; it is the painstakingstudy of broad questions that has the potential of shifting how we think abouta wide rangeof issues.It mayhavean impact for the next 50 yearsor century.Often, tne applicationsof basicresearchappearmanyyears or decadeslater. Practicalapplicationsmay be apparentonly aftermanyaccumulatedadvances in basicknowledgebuild over a long time pe_ riod. For example,in 1984,Alec Jeffreys,a ge_ neticist at the Universityof Leicesterin England, Use of Research wasengagedin basicresearchstudyingthe evo_ For over a century sciencehashad two wings. lution ofgenes.As an indirect ,rsom-e accidentialsideefresearchers adopt a detached,purely scifect of a new technique he developed,he entific, and academicorientation;others are discovereda wayto producewhat is now callhumore activist, pragmatic, and interventionist man DNA "fingerprints"or uniquemarkingsof oriented. This is not a rigid separation.Rethe DNA of individuals.This wasnot his inient. searchers in the two wingscooperateand mainHe evensaidhe would haveneverthoughtof the tain friendly relations.Someindividuals move teghlique if DNA fingerprintshadbeenhis goal. from one wing to anotherat different stagesin Within 10 yearsapplied usesof the technique their careers.In simpleterms,someresearchers weredeveloped.Today,DNA analysisis a widiiv concentrateon advancinggeneralknowledge usedtechniquein criminalinvestigations. over the long term, whereasothers conduct studiesto solvespecific,immediateproblems. Applieil Research, Applied socialresearchisdeThosewho concentrateon examiningthe funsignedto addressa specificconcernor to ofi[er damentalnature of socialreality are engagedin solutionsto a problem identified by an embasicresearch. ployer,club,agenry,socialmovement,or orga-
12
P A RToNE / F o u N D AT Io N s
nization.Applied socialresearchers are rarely concernedwith building, testing,or connecting to a largertheory, developinga long-term generalunderstanding,or carryingout a large-scale investigationthat might spanyears.Instead,they usuallyconduct a quick, small-scalestudy that providespracticalresultsfor use in the short term (i.e.,next month or next year).For example, the student government of University X wantsto know whetherthe number of UniversityX studentswho arearrestedfor drivingwhile intoxicatedor involvedin autoaccidentswill declineifit sponsorsalcohol-freepartiesnextyear. Applied researchwould be most applicablefor this situation. Peopleemployedin businesses, government offices,healthcarefacilities,socialserviceagencies,political organizations,and educationalinstitutionsoften conduct appliedresearchand usethe resultsin decisionmaking.Applied researchaffectsdecisionssuch as the following: Shouldan agencystarta newprogramto reduce the wait time before a client receivesbenefits? Shoulda policeforce adopt a new tlpe of responseto reducespousalabuse?Shoulda political candidateemphasizehis or her standon the environmentinsteadof the economy?Shoulda companymarket a skin careproduct to mature adultsinsteadof teenagers? The scientific community is the primary consumerof basicresearch.The consumersof appliedresearchfindings are practitionerssuch as teachers,counselors,and socialworkers, or decisionmakerssuch as managers,agencyadministrators,and public officials.Often, someone other than the researcher who conducted the studyusesthe results. Applied researchresultsarelesslikely to enter the public domain in publicationsand may be availableonly to few decisionmakersor practitioners.This meansthat appliedresearchfindings often are not widely disseminatedand that well-qualifiedresearchers rarelygetto judgethe quality of appliedstudies. The decisionmakerswho usethe resultsof an applied study may or may not use them
wisely.Sometimesdespiteseriousproblemswith a study'smethodologyand cautionsfrom the repoliticiansuseresultsto justiSrcutting searchers, programsthey dislike or to advanceprograms they favor. Becauseapplied researchoften has immediateimplications or involyescontroversial issues, it often generatesconflict. One famousresearcher,William Whyte (1984),encounteredconflict over findings in his applied researchon a factory in Oklahoma and on restaurantsin Chicago.In the first case,the managementwasmore interestedin defeatinga union than in learningabout employmentrelations; in the other, restaurantowners really soughtto makethe industry look good and did not want findingson the nitty-gritty of its operationsmadepublic. Applied and basicresearchers adopt different orientationstoward researchmethodology (seeTable1.2).Basicresearchers high emphasize methodologicalstandardsand try to conduct near-perfectresearch.Applied researchers must make more tradeoffs.They may compromise scientificrigor to get quick, usableresults,but compromiseis never an excusefor sloppy research.Applied researchers try to squeezeresearchinto the constraintsofan appliedsetting and balancerigor againstpracticalneeds.Such balancingrequiresan in-depth knowledgeof researchand an awareness ofthe consequences of compromisingstandards.
Typesof Applied Research. There are many specifictypesof appliedresearch.Here,you will learn about three major types:evaluation,action, and socialimpact assessment.
EvaluationResearch Study. Evaluationresearc study is applied researchdesignedto find out whethera program, a new way of doing something, a marketing campaign,a policy, and so forth, is effective-in other words, "Does it work?"The most widelyusedtlpe of appliedresearchis evaluationresearch.T This type of researchis widely used in large bureaucratic organizations(e.9.,businesses, schools,hospi-
q HAPT ER I
,/ D OIN C SOC IA L R E S E A R C H
r3
Basicand Applied SocialResearchCompared
l. Research is intrinsically satisfying and judgmentsare by other sociologists. 2. Research problemsand subjectsare selected with a great dealof freedom. 3. Research isjudgedby absolutenormsof scientificrigor,and the higheststandardsof scholarship are sought. 4. The primaryconcernis with the internallogic and rigor of researchdesign. 5. The drivinggoal is to contributeto basic, theoreticalknowledge. 5. Success comeswhenresultsappearin a journaland havean impacton others scholarly in the scientific community.
.l
. Research is part of a job and isjudgedby sponsorswho areoutsidethe discipline of sociology.
2. Research problemsare "narrowlyconstrained" to the demandsof employerc or.,ponro.r. 3. The rigorand standards ofscholarship depend on the usesof results.Research can be ,,quick and dirty" or may matchhighscientific standards. 4. The primaryconcernis with the abilityto generalize findingsto areasof interestto sponsors. 5. The drivinggoal is to havepracticalpayoffsor usesfor results. 5. Successcomeswhenresultsare usedbv sponsorsin decisionmaking.
Source: Adapted fromFreeman (1984:572-573\. andRossi
tals, government, large nonprofit agencies) to demonstrate the effectivenessof what they are doing. An evaluation researcherdoes noi use techniques different from those of other social researchers.The difFerencelies in the fact that decision makers, who may not be researchers themselves,define the scope and purpose of the research.Also, their objective is to use results in a practical situation.S Evaluation research questions might include: Does a Socratic teaching technique improve learning over lecturing? Does a law-enforcement program of mandatory arrest reduce spouseabuse?Does a flextime program increase employee productivity? Evaluation researchers measure the effectivenessof a program, policy, or way of doing something and often use several researchtechniques (e.g.,survey and field). Ifit can be used, the experimental technique is usually preferred. Practitioners involved with a pol-
lcy or program may conduct evaluation research for their own information or at the recuest of outside decision makers. The decision Luk.., may place limits on the research by fixing boundaries on what can be studied and by determining the outcome of interest. This often createsethical dilemmas for a researcher. Ethical and political conflicts often arise in evaluation researchbecausepeople can have opposing interests in the findings. The findings of researchcan affect who getsor keepsa job, it can build political popularity, or it may help promote an alternative program. people who are personally displeasedwith the findings may attack the researcheror his or her methods. Evaluation research has severallimitations: The reports ofresearch rarely go through a peer review process,raw data are rarely publicly available, and the focus is narrowed to select inputs and outputs more than the full processbvwhich
14
P A RToN E ,/ F o u N D A T to N s
a programaffectspeople'slives.In addition,decisionmakersmay selectively useor ignoreevaluation findings. Action ResearchStudy. Action researchis applied researchthat treatsknowledgeasa form of power and abolishesthe divisionbetweencreating knowledgeandusingknowledgeto engagein political action.Thereareseveraltypesofaction research,but most sharefive characteristics: ( 1) the peoplebeing studiedactivelyparticipatein the researchprocess;(2) the researchincorporatesordinaryor popularknowledge;(3) the researchfocuseson issuesof power; ( ) the researchseeksto raiseconsciousness or increase awareness ofissues;and (5) the researchis tied directlyto a plan or programof political action. Action researchtendsto be associated with a social movement,political cause,or advocacyfor an issue.It canbe conductedto advancea range of political positions.Someaction researchhas an insurgentorientationwith goalsof empowering the powerless, fighting oppressionandinjustice, and reducing inequality. Wealthy and powerfrrlgroups or organizationsalso sponsor and conductactionresearchto defendtheir status,position,and privilegesin society. Most actionresearchers areexplicitlypolitical,not valueneutral.Becausethe primary goal is to affectsociopoliticalconditions,publishing resultsin formal reports,articles,or booksis secondary.Most actionresearchers alsobelievethat knowledgedevelopsfrom directexperience, particularly the experienceof engagingin sociopolitical action. For example,most feminist researchis action research.It hasa dual mission:to createsocialchangebytransforminggenderrelationsand to contributeto the advancementofknowledge. A feministresearcher who studiessexualharassment might recommendpolicy changesto reduceit aswell asto inform potentiaivictims so they can protect themselveiand defendtheir rights. At times, researchers will explain study resultsin a public hearingto try to modi$, new policiesor laws.The authorsof a study on do-
mesticviolencethat will be discussedshortly as an explanatorystudy example(Cherlin et al., 2004)testifiedin the United StatesSenate.The studyfindingsand the testimonyhelpedto alter marriagepromotion provisionsin a 2005welfarereform law.e
SocialImpactAssessment Research Study. A researcherwho conductssocialimpactassessme (S1A) estimatesthe likely consequencesof a planned intervention or intentional changeto occurin the future.It may be part of a largerenvironmentalimpact statementrequiredby gov-I ernment agenciesand usedfor planning and making choicesamong alternativepolicies.He or sheforecastshow aspectsof the socialenvironment may changeand suggests waysto mitigatechangeslikely to be adversefrom the point ofview of an affectedpopulation.Impactsarethe differencebetweena forecastof the future with the project or policy and without the project or policy. For example,the SIA might estimatethe ability of a localhospitalto respondto an earthquake,determinehow housing availabilityfor the elderlywiil changeif a major new highwayis built, or assess the impact on collegeadmissions if students receive interest-freeloans. Researcherswho conduct SIAs often examine a rangeof socialoutcomesand work in an interdisciplinaryresearchteam to estimatethe social outcomes.The outcomesinclude measuring "quality oflife" issues,such as accessto health care,illegal drug and alcohol use,employment opportunities,schoolingquality,teenpregnancy rates,commuting time and traffic congestion, availability of parks and recreationfacilities, shoppingchoices,viablecultural institutions, crime rates,interracialtensions,or socialisolation. Thereis an internationalprofessionalassociation for SIA researchthat advancesSIA techniquesand promotesSIA by governments, corporations,and other organizations. Social impact assessments are rarely required, but a few governmentsmandatethem. For example,in New South Wales,Australia,a registeredclub or hotel cannot increasethe
CHAPTER1 , / D o I N c s o c I A L R E S E A R c H
number of poker machinesunlessthe Liquor AdministrationBoardin the DepartmentGaming and Racingapprovesan SIA for the club or hotel. The SIA enablesthe board to assess the likely local community impact from increasing the number of poker machines.The format includesa matrix that allowsthe board to identify the socialand economicimpacts,positive and negative,financialor nonfinancial,quantifiedor qualitative. In New Zealand,the Gambling Act of 2003requiresan SIA beforeexpandinggambling. In one2004studyin New Zealandfor the AucklandCity Council,it noted that 90 percent of New Zealand'sadultsgamble,10percentgamble regularly(oncea week or more often), and about 1 percentareproblemgamblers,although this variesby age,income, and ethnicity. The SIA recommendedlimiting the locationsof new gamblingvenues,monitoring their usage,and tracing the amount of gambling revenuesthat are returnedto the community in variousways (e.g.,clubs,trusts,etc.).It containeda matrix with social (e.g,arrests,divorce,domesticvio-
I5
lence), economic (e.g., unemployment, banliruptcy, tourism expansion), and cultural impacts (e.g.,time awayfrom other leisure activin') listed by their effect on all gamblers, problem gamblers, the local community, and the region.lo
Purpose ofa Study
If you asksomeonewhy he or sheis conducting a study,you might geta rangeof responses: "My bosstold me to"; "It wasa classassignment";"I wascurious"; "My roommatethought it would be a good idea." Thereare almostasmany reasonsto do researchasthereareresearchers. Yet, the purposesof socialresearchmaybe organize into threegroupsbasedon whatthe researcher is trying to accomplish-explore a new topic, describea socialphenomenon,or explain why somethingoccurs.Studiesmay havemultiple purposes(e.g.,both to exploreandto describe but oneof threemajor purposesis usuallydominant (seeBox 1.2).
Exploratory
Descriptive
Explanatory
r
Becomefamiliarwith the basicfacts,setting,and concerns,
I
Providea detailed,highly accuratepicture.
r
Test a theory'spredictionsor principle.
r
Createa generalmental pictureof conditions.
Locatenewdatathat contradictpast data.
r
r
r
Formulateand focus questionsfor futureresearch.
Createa set ofcategoriesor classifytypes.
r
r
Elaborateand enricha theory'sexplanation. Extenda theoryto newissues or topics.
r
Ceneratenewideas, conjectures, or hypotheses. Determinethe feasibilitvof conductingresearch.
Clarifya sequenceof stepsor stages.
r
t
Supportor refutean explanation or prediction.
I
Documenta causalprocess or mechanism.
r
Linkissues or tooicswitha generalprinciple.
r
Reporton the background or contextofa situation.
I
Determinewhichof several explanations is best.
r r
Developtechniquesfor measuring and locatingfuture data.
16
PART oNE , / FO UNDATI O NS
Exploration. Perhapsyou haveexploreda new topic or issuein order to learnaboutit. If the ishadwritten about suewasnew or no researchers In exploratoryrebeginning. it, you beganat the new areato fora examines search.a researcher he or she can questions that mulate precise research Exploratory research. addressin future A studies' of in a sequence stage maybe the first exploratory an conduct need to researchermay studyin orderto know enoughto designand executea second,more systematicand extensive the "what?"question:"What study.It addresses is this socialactivityreallyabout?" Many higher-educationofficials are concerned about college students' low retention rates,especiallystudentsfrom minority-disadvantagedsocialbackgrounds.For example,of Latinoswho enroll in college,80 percentleave without receivinga degree.Officialsseekwaysto reducedropouts and increasethe chancesthat studentswho begin collegewill stay until they earn a degree.Garzaand Landeck(2004)conductedan exploratorystudy of over 500 Latino studentsat a collegealong the Texas-Mexico border who had dropped out. They wanted to learn the influencing factors and rationalesin studentdecisionmaking.Theauthorsdiscovered that the primary factorsand rationaleswereunrelatedto teachingquallty or universityservices. Instead,the studentswho droppedout hadbeen overwhelmedby personalproblemsor had seriousdifficultieswith familyor job responsibilities. Suchfactorswere a major reasongivenby over 80 percentofthe studentswho droppedout. tend to usequalitaExploratoryresearchers tive dataand not be weddedto a specifictheory or researchquestion.Exploratoryresearchrarely yieldsdefinitive answers.If you conduct an exploratory study, you may get frustrated and feel it is rlifficult becausetherearefew guidelinesto fdlow. Everything is potentially important, the step6are mt wdl defined, and the direction of fteqrn$y. You needto be creirffi.h'.GF -:&d, edoPt an inrd f-r'k; on-;c srra's of rcffrure"dqbcJ
drMm"
Description, Perhapsyou havea more highly developedideaabout a socialphenomenonand presents want to describeit. Descriptiveresearch a picture of the specificdetailsof a situation,social setting,or relationship;it focuseson "how?" and "who?" questions:"How did it happen?" "Who is involved?"A greatdeal of socialreuse searchis descriptive.Descriptiveresearchers field most data-gatheringtechniques-surveys, research,content analysis,and historical-comparativeresearch.Only experimentalresearchis iess often used.Much of the social research found in scholarlyjournals or usedfor makin$ policy decisionsis descriPtive' Descriptiveand exploratoryresearchoften blur togetherin practice'In descriptiveresearch, a researcherbeginswith a well-definedsubject and conductsa study to describeit accurately andthe outcomeis a detailedpictureof the subject. The resultsmay indicatethe percentageof peoplewho hold a particularview or engagein specificbehaviors-for example,that 8 percent of parentsphysicallyor sexuallyabusetheir children. A descriptivestudy presentsa picture of typesofpeople or ofsocial activities. Stack,Wasserman,and Kern (2004)conducteda descriptivestudy on pornographyuse on the Internet by peoplein the United States. They found that the greatestuserswere those with weak socialbonds' More specifically,the typesofpeoplewho wereadult usersofpornography tended to be maleswith unhappy marriages and weak ties to organizedreligion. Pornographyuserswerealsomore likely to have engagedin nonconventionalsexualbehavior (i.e.,had an extramaritalaffair or engagedin paid sex)but not other forms of deviance,such asillegaldrug use.
Explanation When you encounter an issue that is well recognizedand havea descriptionof it, you might beginto wonderwhy thingsarethe identifiesthe way they are.Explanatoryresearch sourcesof socialbehaviors,beliefs,conditions, and events:it documentscauses'teststheories and providesreasons.It builds on exploratory
CHAPTER1 , / D O I N G S O C I A L R E S E A R C H
and descriptiveresearch.For example,an exploratory studydiscoversa new typeofabuseby parents;a descriptiveresearcher documentsthat 10percentofparentsabusetheir childrenin this new way and describesthe kinds of parentsand conditionsfor which it is most frequent;the explanatoryresearcher focuseson why certainparents are abusingtheir children in this manner. Cherlin, Burton, Hurt, and Purvin (2004) explained instability in marriageor cohabitation using a woman'spast experiencewith sexualor physicalabuse.They testedthe hypothesisthat women with a history of abusewould be less likely marry than thosewithout such histories. The authors reasonedthat those who were abusedhavefewersocialsupportsand resources to resistor avoid abusivepartners,and they are more likely to harbor feelingsof self-blame, guilt, and low self-esteem that inhibit the formation of healthyromantic relationships.An abusive experiencealso createsgreateremotional distanceand a hesitancyto make long-term commitments.Using quantitativeand qualitative datagatheredin low-incomeneighborhoods in threecities-Boston, Chicago,and SanAntonio-they found that adult womenwho had experiencedpast abusewere less likely to be married,and thosewith multiple forms of abuse weremost likely to remainsingle.It appearsthat women without a past history of abusewho found themselvesin an abusiverelationshipas ,an adult were likely to withdraw from it, but 'womenwho had beenabusedaschildren were lesslikely to leaveand tendedto enterinto a seriesof unstable,transitoryrelations.
17
over severaltime points (longitudinal).euantitative studiesgenerallylook at many cases,people, or units, and measurelimited featuresabout them in the form of numbers.By contrast,a qualitative study usually involves qualitative data and examinesmany diversefeaturesof a small number of casesacrosseither a short or long time period(seeFigure1.2). Cross-SectionalResearch. Most social researchstudiesarecross-sectional; theyexaminea singlepoint in time or takea one-timesnapshot approach.Cross-sectional researchis usuallythe simplestand leastcostly alternative.Its disadvantageis that it cannotcapturesocialprocesses or change.Cross-sectional researchcan be exploratory, descriptive,or explanatory but it is most consistentwith a descriptiveapproachto research.The descriptive study by Stack, Wasserman,and Kern (2004) on pornography usewascross-sectional, basedon a nationalU.S. surveyconductedin 2000. Longitudinal Reseqrch. Researchersusing longitudinalresearch examinefeaturesof people or otherunits at morethan onetime. It is usually more complexand costlythan cross-sectional research,but it is alsomore powerful and informative.Descriptiveand explanatoryresearchers uselongitudinalapproaches. Let us now look at the three main types of longitudinal research: time series,panel,and cohort.
Time-Series Study. A time-series studyislongitudinal researchin which a researchergathers the sametypeof information acrosstwo or more time periods.Researchers can observestability Time Dimensionin Research or changein the featuresofthe units or cantrack An awareness of how a study usesthe time diconditions over time. The specificindividuals mensionwill helpyou reador conductresearch. may changebut the overallpattern is clear.For This is becausedifferentresearchquestionsor isexample,therehasbeena nationwidesuweyof a suesincorporatetime difFerently.Somestudies large sampleof incoming freshmanstudents givea snapshotof a single,fixed time point and since1966.Sinceit began,over 11 million stuallowyou to analyzeitin detail(cross-sectional). dentsat more than 1,800collegesparticipated. Other studiesprovide a moving picturethat lets The fall 2003surveyof276,449 studentsfound you follow events,people,or socialrelations manyfactsandtrends,suchasonly 34percentof
PART oNE /
l8
Flc U RE 1 .2
FoUNDATI oNS
The Time Dimensionin SocialResearch
CROSS-SECTIONAL: Observea collectionof peopleat one time.
a l . R P - ^@
z. E@ # F
sT|l
##=d eEE= i
7|FtrffiFR€
P
February 2007
TIME SERIES: Observedifferentpeopleat multipletimes'
*e%[w"MIMK wffililwwwwl& wffil#wwffilwff 2010
1990
1970
1950
PANEL: Observe the exact same people at two or more times
2006
1996
1986
COHORT: Observepeoplewho sharedan experienceat two or moretimes'
1987
Married in 1967
CASE STUDY: Observea small set intenselyacrosstime.
--*
G. €
-9.
Q
*\G
F+ F +# E Hr G g #nE E EEFFtr - E 2005+2007
CHAPTERI
enteringfreshmenstudiedsix or more hoursper week.This was the lowestlevel sincethe question wasaskedin 1987(when itwas47 percint). Yet, alcohol consumptionwas down. In 2003, 44.8percentreporteddrinking beer,which rep_ resenteda steadydecline from73.7 percentin 7982.In2003,freshmenweremore inierestedin keepingup with politics. The 33.9percentwho saidit wasvery important to staypolitically in_ formed was up from a low of 2g.l percentin 2O00,and 22.5 percentsaidthey discussed poli_ tics regularl/, up from 19.4 percent in )OOZ (whighhad beenthe highestsincea low point in 1993).Thesefiguresare still far lower than the 60.3percentwho expressed an interestin politics in 1966,or the one-third who discussed politics regularlyin 1968.The importanceof familyhas steadilyincreasedover the years,with 74.gper_ cent of studentscalling it essentialo, ,,"ry1-_ portant.This is up from the low point oi SS.g percentin 1977when the questionwas first asked.However,religiousinvolvementdeclined. The percentageof studentswho attendedreli_ giousserricesregularlywasat its lowestlevelin 35 years. In addition, the percent claiming "none" asa religiouspreferencereacheda record high of 17.6percent,comparedto a recordlow of 6.6percentin 1966.Another trend overthe past two decades hasbeena steadygrowthin opposition to the deathpenalty.Nearlyonein thiee incoming students advocated ending capital punishment. This is the highestscoresincei9g0 , I (w!g" itwas 33.2percent), althoughthe percent withholding an opinion wasfar higherejrher in time; it exceeded 60 percentin the tgZO.ll
/
D O I N C S O C I A LR E S E A R C H
I9
term panel studiescan clearlyshowthe impact of a particularlife event.For example,Oesterle, fohnson,and Mortimer (2004)examinedpanel data from a longitudinal study that began in 1988with 1,000ninth-gradestudentsenrolled in the St.Paul,Minnesota,public schooldistria and lookedat volunteeringactivitiesduring late adolescenceand young adulthood, covering nine yearsfrom age l}-t9 (1992)to age26_27 (2000).They found that volunteeringat an ear_ lier stagestrongly affectedwhether one volun_ teeredat a later stage.Also, peoplewho devoted full time to working or parenting at an earlier stage(18-19yearsold) werelesslikelv to volun_ teerat a later stage(26-27 yearsold) than those whosemajor activitywasattendingschool.
CohortStudy. A cohortstudy is similar to a panelstudy,but ratherthan observingthe exact samepeople,the study focuseson a categoryof peoplewho sharea similar life experieniein a specifiedtime period. Researchers examinethe categoryas a whole for important featuresand focuson the cohort,or categorynot on specific individuals.Commonlyusedcohortsincludeall peopleborn in the sameyear (called birth co_ horts),all peoplehired at the sametime, and all people who graduatein a given year. Unlike panelstudies,researchers do not haveto find the exactsamepeoplefor cohort studies;rather, theyneedonly to identifythosewho experienced a commonlife event.In a studyof GenerationX in the United States,AndolinaandMayer e003) focusedon the cohort ofpeople born between 1967and 1974. They comparedt0 birth cohorts at different time periods over severaldecades, PanelStudy. Thepanelstudyis apowerfirltyp. tracing questionsacross24 years.The authors of longitudinalresearchin which the researciier found that White Xers are distinct in their observesexactlythe samepeople,group, or orsupport for school racial integration and for ganization acrossmultiple time points. It is governmentactionto enforcesuchefforts,com_ more difficult to conductthan time-seriesre_ paredto other birth cohorts, but not in their at_ search.Panelresearchis formidableto conduct titudes toward employment opportunities or and very costly.Trackingpeopleovertime is ofaffirmativeaction. Despitegreatergeneralsup_ ten difficult becausesomepeopledie or cannot port than other cohortsfor equalitythrough in_ be located.Nevertheless, the resultsof a well-detegration,it doesnot extendto issuesbeyondthe signedpanelstudyareveryvaluable.Evenshort_ schoolyard.
20
P A Rr oNE ,/ F o U N D AT T o N S
CsseStudies. In cross-sectionaland longitudinal research. a researcher examines features on many people or units, either at one time period or across time periods, and measures several common features on them, often using numbers. In case-studyresearch,a researcher examines, in depth, many features of a few casesover a duration of time with very detailed, varied, and extensive data, often in a qualitative form. The researchercarefirlly selectsa few key casesto iilustrate an issue and study it (or them) in detail and considers the specific context of each case. This contrasts with other longitudinal studies in which the researchergathersdata on many units or cases,then looks for general patterns in the massof numbers. For example, Snow and Anderson (1992) conducted a case study on homeless people in Austin, Texas. It provided a wealth of details about the lives and conditions of homeless people, identified severaltypes of homeless people, outlined the paths by which they became homeless, and discussedseveralprocessesthat kept them homeless.This casestudy used many Qpes of detailed qualitative and quantitative data, with exploratory descriptive, and explanatory phasesto reveal a great amount of unexpected and new information.12 Data Collection Techniques Social researcherscollect data using one or more specific techniques. This section givesyou a brief overview of the major techniques. In later chapters, you will read about these techniques in detail and learn how to use them. Some techniques are more effective when addressing specific kinds of questions or topics. It takes skill, practice, and creativity to match a researchquestion to an appropriate data collection technique. The techniques fall into two categories based on whether the data being gathered are quantitative or qualitative. Quantitative Data Collection Techniques, Techniques for quantitative data collection in-
clude experiments, surveys, content analyses, and existing statistics. Experiments. Experimental researchclosely follows the logic and principles found in natural science research;researcherscreate situations and examine their efFectson participants. A researcherconducts experiments in laboratories or in real life with a relatively small number of people and a well-focused researchquestion. Experiments are most effective for explanatory research. In the tlpical experiment, the researcher divides the people being studied intol two or more groups. He or she then treats both groups identicallS except that one group but not the other is given a condition he or she is interested in: the "treatment." The researcher measures the reactions ofboth groups precisely. By controlling the setting for both groups and giving only one group the treatment, the researcher can conclude that any differenees in the reactions of the groups are due to the treatment alone. Surveys. A survey researcherasks people questions in a written questionnaire (mailed or handed to people) or during an interview and then records answers. The researcher maniprrlates no situation or condition; he or she simply asksmanypeople numerous questions in a short time period. Typically, he or she then summarizes answersto questions in percentages,tables, or graphs. Researchersuse survey techniques in descriptive or explanatory research.Surveysgive the researchera picture of what many people think or report doing. Survey researchersoften use a sample or a smaller group of selectedpeople (e.g., 150 students),but generalizeresultsto a larger group (e.g., 5,000 students) from which the smaller group was selected.Survey research is very widely used in many fields. Content Analyses. A contenLtanalysis is a technique for examining information, or content, in written or symbolic material (e.g., pictures, movies, song lyrics, etc.). In content analysis, a
c HAPTER 1 , / D O I N C S O C T A LR E S E A R C H
researcherfirst identifiesa body of material to (e.g.,books,.t.*rpup"rr, films, etc.)and a-nalyze then createsa systemfor-reiording specificas_ pectsof it. The systemmight include counting how often certain words or themesoccur. Fi_ nalln the researcherrecordswhat wasfound in the material.He or sheoften measuresinforma_ tion in the contentasnumbersandpresentsit as tablesor graphs.This techniqueletsa researcher discoverfeaturesin the contentoflargeamounts of materialthat might otherwisego unnoticed. Researchers can use content analysisfor ex_ ploratoryand explanatoryresearch,tut primar_ ily it is usedfor descriptiveresearch.
21
in detail.The researcher getsto know personally the_people being studied,may condlct open_ ended and informal interviews,and takesde_ tailed notes on a daily basis.After leaving the field site, the researchercarefully rereadi the notes and prepareswritten reports.Field re_ searchis usedmost oftenfor expioratoryand de_ scriptivestudies;it is rarelyusedfor explanatory research.
Historical-ComparativeResearch. Historical_ comparativeresearchers examjneaspectsof social life in a pasthistoricaletuo. u.ros differentcul_ tures.Researchers who usethis techniquemay focus on one historicalperiod o. r.rr.ril, .o-_ ExistingStatistics. In existingstatisticsresearch, pareone or more cultures,or mix historicalpe_ a researcherlocatespreviouslycollectedinfor_ riods and cultures. Like field research.^ u mation,often in the form of governmentreports researchercombines theory building/testing or previouslyconductedsuryeys,then reorga_ with data collection and beginswith a loose$ nizesor combinesthe information in new ways formulated question that is iefined during the to addressa researchquestion.Locatingsources researchprocess. Researchers often gatlier a canbe time consuming,so the researcher needs wide arrayof evidence,including existiig statis_ to considercarefullythe meaningof what he or tics and documents(e.g.,novels,officialieports. she finds. Frequently,a rer"arih", does not books,newspapers,diaries,photographs,and know whether the information of interest is maps)for study.In addition,theymav makedi_ availablewhen he or shebeginsa study.Some_ rect observations and conduct interviews.His_ times,the existingquantitativeinformaiion con_ torical-comparative researchcanbe exploratory sists of stored surveys or other data that a descriptive,or explanatoryandcanblend types. researcherreexaminesusing various statistical procedures.Existingstatisticsresearchcan be usedfor exploratory descriptive,or explanatory purposes,but it is most frequentlyusedfor de_ C ON C L U S ION scriptiveresearch. This chapter gave you
an overview ofsocial re_ search.You saw how social researchdiffers from Qualitative Data Collection Techniques. the ordinary ways of learning-knowing about Techniquesfor qualitativedata collectioi in_ the social world, how doing researchis basedon clude fi eld researchand historical-comparative scienceand the scientific community, and about research. several tlpes of social research based on its dimensions (e.g.,its purpose, the technique used FieldResearch.Most field researchers conduct to gather data, etc.).The dimensions of research casestudieslooking at a small group of people loosely overlap with each other. The dimensions oyer a length of time (e.g.,weeki, mbnt^hs, of social research are a kind of ..road map', to years).Afield researcher beginswith a looseiy help you make your way through the terrain of formulatedidea or topic, selectsa socialgroup social research. In the next chapter, we turn to or natural setting for study, gains accesiand social theory. You read about it a little in this adoptsa socialrole in the setting,and observes chapter. In the next chapter, you will learn how
22
P A RToNE / F o u N D A T ro N s
theoryand researchmethodswork togetherand aboutseveraltypesof theory.
social impact assessmentstudy social research suwey research time-series study
Key Terms action research study applied social research basic social research casestudy cohort study cross-sectional research data descriptive research empirical evidence evaluation research study existing statistics research experimental research explanatory research exploratory research field research halo effect historical comparative research longitudinal research overgeneralization panel study premature closure qualitative data quantitative data scientific community scientific method selective observation
E n dn o t e s 1. See Rampton and Stauber (200I:247-277 and
30s-306).
andmedia. on advocates 2. SeeBest(2001:15) Board(2002:735-739). 3. SeeNationalScience 4. Schacter(2001)providesa summaryof memoryl issues. 5. NationalScienceBoard(2002:739). 6, Discussionsof the scientificcommunity can be found in Coleand Gordon (1995),Crane(1972), Hagstrom(1965),Merton(1973),Mulkay(1991), andZiman(1999). 7. SeePatton(2001)and Weiss(1997)for a more in evaluadetaileddiscussionof recentadvances tion research. 8. Beck( 1995)providesa usefuloverview' 9. SeeHerringand Ebner(2005)on the useofdomesticviolencestudyfindings. 10. SeeAdams Q004) for more information on the Auckland City study. ll. Seethe websiteat www.gseis.ucla.edu/heriihe html. 12. AIsoseeSnowandAnderson(1991)for a discussion of the case-studymethod in their study of . homelesspeople.Also seeGeorgeand Bennett (2005)on the case-study methodgenerally.
Theory andSocialResearc
Introduction What ls Theory? BlameAnalysis The Parts of Theory Conceots Assumptions Relationshios The Aspects ofTheory Directionof Theorizing RangeofTheory Levelsof Theory Formsof Explanation The Three Major Approaches to Social Science PositivistApproach InterpretiveApproach CriticalApproach The Dynamic Duo Conclusion
I
24
PART ONE , / FO UNDATI O NS
in mind threethings about how socialscientific theorieswork. First, socialtheoriesexplain recurringpatterns,not uniqueor one-timeevents. Supposeyou want to makesenseof the hostility For example,they are not good for explaining betweenpeopleof differentraces.Trying to unwhy terrorists decided to attack New York's derstandit, you aska teacher,who responds: 11,2001,but World TradeCenteron September they canexplainpatterns,suchasthe conditions Most raciallyprejudicedpeoplelearnnegathat generallyleadto increasedlevelsoffear and aboutanotherracialgroup tive stereotlpes feelingsof patriotism in a people.Second,social from their families,friends.andothersin not partheoriesareexplanationsfor aggregates, If theylack their immediatesurroundings. are collectionsof ticular individuals.Aggregates sufficientintimatesocialcontactwith memmany individuals,cases,or other units (e.g., bersof the groupor intenseinformationthat schools,families,clubs,cities,na't businesses, theyremain contradictsthosestereotypes, tions, etc.).A socialtheory rarely can explain prejudiced. why fosephinedecidedto major in nursing This makessenseto you becauseit is consis- rather than engineering,but it can explainwhy tent with what you know about how the social femalesmore than malesin generalchoosenursworld works.This is an exampleof a small-scale ing over engineeringas a major. Third, social theoriesstatea probabiliry chance,or tendency usewhen socialtheory, a tpe that researchers for eventsto occur, rather than statethat one conductinga study. eventmust absolutelyfollow another.For examWhat do you think of when you hear the ple, insteadof stating that when someoneis word theory?Theory is one of the leastwell unabusedas a child, that personwill alwayslater derstoodterms for studentslearningsocialscience.My students'eyelidsdroop if I begina class abusehis or her own children, a theory might abusedurstatethat whensomeoneexperiences by saying,"Today we are going to examinethe theoryof . . ." The mentalpicturemanystudents ing his or her childhood,that personwill tendto or is morelikely to becomean abusiveparent have of theory is something that floats high when an adult.Likewise,it might statethat peoamongthe clouds.My studentshavecalledit "a ple who did not experiencechildhood abuse tangled mazeof jargon" and "abstractionsthat might becomeabusiveparents,but they areless areirrelevantto the realworld." likely to than someonewho has experience Contrary to theseviews,theory has an imabuseasa child. portant role in researchand is an essentialally for the researcher. Researchers usetheorydifferently in varioustypesof research,but sometl?e of theory is presentin most socialresearch.It is lessevidentin appliedor descriptivethan in baWHAT IS THEORY? sic or explanatoryresearch.In simpleterms,reinterweavea storyaboutthe operation In Chapter L, socialtheorywasdefinedasa syssearchers tem ofinterconnectedabstractionsor ideasthat of the socialworld (the theory) with what they condensesand organizesknowledgeabout the observewhen they examineit systematically(the socialworld. It is a compactway to think of the data). socialworld. Peopleare constantlydeveloping Peoplewho seekabsolute,fixed answersfor a specificindividual or a particular one-time newtheoriesabouthow the world works. Somepeopleconfusethe history of socia eventmay be frustratedwith scienceand social theories.To avoid frustration,it is wiseto keep thought, or what greatthinkerssaid,with social INTRODUCTION
CHAPTER2 , / TH E O R YA N D S O C I A L R E S E A R C H
theory. The classicalsocial theorists (e.g., Durkheim, Weber,Marx, and Tonnies) played an important role in generatinginnovativeideas. They developedoriginal theoriesthat laid the foundationfor subsequentgenerationsof social thinkers.Peoplestudythe classicaltheoristsbecausethey providedmany creativeand interrelated ideasat once.They radicallychangedthe waypeopleunderstoodand sawthe socialworld. We study them becausegeniuseswho generate many original, insightfirl ideasand fundamentally shift how pdoplesaw the socialworld are rare. At timespeopleconfusetheorywith a hunch or speculativeguessing.Theymay say,"It's only a theory" or ask,"What'syour theoryabout it?" This lax useof the term theorycauses confusion. Suchguessingdifi[ersfrom a serioussocialtheory that has been carefullybuilt and debated over many yearsby dozensof researchers who found support for the theory'skey parts in repeatedempiricaltests.A relatedconfusionis when what peopleconsiderto be a "fact" (i.e., light a matchin a gasoline-filledroom and it will explode)is what scientistscall a theory (i.e.,a theory of how combining certain quantitiesof particular chemicalswith oxygenand a level of heat is likely to producethe outcomeof explosiveforce). Peopleuse simple theorieswithout making them explicit or labelingthem as such. For example,newspaperarticlesor televisionreports on socialissuesusuallyhaveunstatedsocial theoriesembeddedwithin them. A news reporton the difficultyof implementinga school desegregation plan will contain an implicit theory aboutracerelations.Likewise,politicalleadersfrequentlyexpresssocialtheorieswhen they discusspublic issues.Politicianswho claim that inadequateeducationcausespoverty or that a declinein traditional moral valuescauseshigher crime ratesare expressingtheories.Compared to the theoriesof socialscientists,such laypersons'theoriesare lesssystematic,lesswell formulated, and harder to test with empirical evidence.
25
Almost all researchinvolvessometheon',so the questionislesswhetheryou shouldusetheory than how you should use it. Being explicit aboutthe theorymakesit easierto readsomeone else'sresearchor to conduct your own. An awarenessof how theory fits into the research processproducesbetter designed,easierto understand,andbetterconductedstudies.Most researchersdisparageatheoretical or "crude empiricist"research. BlameAnalysis Blameanalysisis a type of counterfeitargument presentedasif it werea theoreticalexplanation. It substitutesattributing blame for a causalexplanationthat is backedby supportingempirical evidence.Blamebelongsto the realmof making moral, legal,or ideologicalclaims.It implies an intention, negligence,or responsibilityfor an eventor situation (usuallyan unfavorableone). It shifts the focus from Why did it occur?to Who is responsible? Blame analysisassumes there is a party or source to which a fixed amount of responsibilitycan be attached.The goalof inquiry is to identifi a responsibleparty. Often, somesourcesare exemptedor shielded. This may be the injured party, membersof a sympatheticaudience,or a sacredvalue or principle. Blameanalysiscloudsdiscussionbecauseit confusesblame with cause;it givesan account (or story) insteadof a logical explanationwith intervening causalmechanisms;and it fails to exploreempirical evidencefor and againstseveral alternativecauses.Blameanalysisfirst presentsan unfavorableeventor situation.It could be a bank is robbed, a group is systematically paid lessin the labor force,or traffic congestion is terriblein an urban area.It next identifiesone or more responsibleparties,then it providesselectiveevidencethat shieldscertainpartiesor sources (e.g., employrnent conditions, the choicesavailableto the underpaidgroup,transportation poliry, and land cost).l
26
PART ON E , / FO UNDATI O NS
measureheightor compareit. A heightof zerois possible,and height can increaseor decreas overtime. As with manywords,we usethe word Concepts in severalways.Height is usedin the expression the heightof the battle,the heightof thesummer All theoriescontain concepts,and conceptsare andtheheightoffashion. the building blocks of theory.2A conceptis an The word heightrefersto an abstractidea. asa symbolor in words.Natural ideaexpressed its soundand its written form with in syrnbolic We associate scienceconceptsareoften expressed forms,suchasGreekletters(..g., 6) or formulas that idea. There is nothing inherent in the soundsthat make up the word and the idea it (e.g.,s = d/t; s= speed,d = distance,f = time). The connectionis arbitrary' but it is represents. as Most socialscienceconceptsare expressed words. The exotic symbolsof natural science still useful.Peoplecan expressthe abstractidea conceptsmakemany peoplenervous,asthe use to one anotherusingthe symbolalone. Conceptshavetwo parts:a symbol(word or of everydaywords in specializedsocialscience term) and a definition.We learn definitions in conceptscancreateconfusion. manyways.I learnedthe word heightandits dethe distinction I do not want to exaggerate finition frqm myparents.I learnedit asI learned aswords and conbetweenconceptsexpressed to speakand was socializedto the culture. My assymbols.Words, after all, are ceptsexpressed parentsnevergaveme a dictionarydefinition. I symbols,too; they are symbolswe learn with learnedit through a diffrrse,nonverbal,informal language.Heightis a conceptwith which you are process.Myparentsshowedme manyexamples alreadyfamiliar.For example,I cansaythe word I observedand listenedto othersusethe word; I heightor write it down; the spokensoundsand written words are part of the Englishlanguage. usedthe word incorrectlyand was corrected The combinationof lettersin the soundsyrnbol- and I used it correctly and was understood Eventually,I masteredthe concept. izes,or standsfor, the ideaof a height.Chineseor This exampleshowshow peoplelearn conthe Frenchwordhauteur, t]":'e Arabic characters, ceptsin everydaylanguageand how we share Germanword h1he,the Spanishword alturaconcepts.Supposemy parentshad isolatedme all symbolizethe sameidea. In a sense,a languageis merelyan agreementto representideas from televisionand other people,then taught by sounds or written charactersthat people me that the word for the ideaheightwaszdged.I learnedat somepoint in their lives.Learning would havehad difficulty communicatingwith conceptsand theoryis like learninga language.3 others.Peoplemust sharethe termsfor concepts and their definitionsif they areto be of value. andyou usethem Conceptsareeverywhere, Everydaylife is filled with concepts,but all the time. Height is a simple conceptfrom many havevagueand uncleardefinitions.Likeeverydayexperience.What doesit mean?It is easytousethe conceptofheight,but describing wise,the values,misconceptions,and experiencesof peoplein a culture may limit evgryday the conceptitselfis difficult. It representsan abconcepts.Socialscientistsborrow conceptsfrom stractideaaboutphysicalrelations.How would you describeit to a very young child or a crea- everydayculture,but they refinetheseconcepts and add new ones. Many conceptssuch as ture from a distantplanetwho wastotally unfamiliar with it? A new concept from a social sexism,life-style,peergroup, urban sprawl, and socialclassbeganas precise,technicalconcepts theory may seemjust as alien when you encounterit for the first time. Height is a charac- in socialtheorybut havediffrrsedinto the larger cu]tureand becomelessprecise. teristicofa physicalobject,the distancefrom top We createconceptsfrom personalexperito bottom. All people,buildings, trees,mounence,creativethought,or observation.The clastains,books,and so forth havea height.We can
THE PARTS OF THEORY
L HAT IEK
sicaltheoristsoriginatedmany concepts.Exam_ ple conceptsincludefamily system,genderrole, socinlization, self-worth,frustration, aid.disptaced aggression. Someconcepts,especiallysimple,concrete conceptssuchas bookor height,can be defined through a simplenonverbalprocess.Most social scienceconceptsaremore complexand abstract. They ared1fin9dby formal, diitionary_typede_ finitions that build on other .orr..pir.'ti may seemodd to useconceptsto define other con_ cepts,but we do this all the time. For example,I definedheighfasa distancebetweentop u.rdbot_ tom. Top,bottom,anddistanceare alfconcepts. We often combine simple,concreteconceprs from ordinary experienceto createmore ab_ stractconcepts.Heightis more abstractthantop or bottom.Abstractconceptsrefer to uspect,of the world we do not directly experience.They orga.nize thinking and extendunderstandingof reality. Researchers definescientificconceptsmore preciselythan thosewe use in daily discourse. Socialtheory requires well-definedconcepts. Thedefinitionhelpsto link theorywith research. A valuablegoal ofexploratory research,and of most good research,is to clarify and refinecon_ cepts.Weak,contradictory,oi uncleardefini_ tions of concepts restrict the advance of knowledge.
Z /
IH TOR Y AN D S OC IA L R E S E A R C H
27
conceptareamountof income,tunperatare,delr_ sity,ofpopulation,yearsof schoolinganddesrecof violence.Theseare calledvariablis, and,"r", *ilf read about them in a later chapter.Other con_ ceptsexpresstypesof nonvariablephenomena (e.q., \lyeaucracy,family, rwolution, homeles, andcold).Theoriesuseboth kinds ofconcepts.
ClassificationConcepts. Someconceptsare simple;they haveone dimensionandvary along a singlecontinuum. Others are complex;the| have multiple dimension, o, You can break complex concepts -urryiubparts. into a iet of simple,or single-dimension,.orr..ptr. For ex_ ample,Rueschemeyer and associatis(1992:43_ 44) statedthat democraryhasthreedimensions: (1) regular,freeelectionswith universalsuftage; (2) an electedlegislativebody that controls sov_ ernmenqand (3) freedomof expression*J ur_ sociation.The authors recognizedthat each dimensionvariesby degree.Tiey combinedthe dimensionsto createa set of typesof regimes. Regimesverylow on all threedimensionsir. to_ talitarian,thosehigh on all three are democra_ cies, and ones with other mixes are either authoritarianor liberaloligarchies. Classificationr arepartwaybetweena single, simpleconceptand a theory.aTheyhelpto oriu_ nizeabstract,complexconcepts.To createa n"ew classification, a researcher logicallyspecifiesand combinesthe characteristics of'simpler con_ ConceptClusters. Conceptsarc rarelyusedin cepts.You canbestgraspthis ideaby looking at isolation. Rather, they form interconnected some examples. groups,ot conceptclusters.This is true for con_ The idealtypeis awelJ.-known classification. ceptsin everydaylanguage aswell asfor thosein Idealtypesarepure, abstractmodelsthat define socialtheory.Theoriescontaincollectionsof as_ the essence of the phenomenonin question. sociatedconceptsthat areconsistentand mutu_ They arementalpicturesthat definethe central ally reinforcing. Together,they form a web of aspectsofa concept.Idealtypesarenot explana_ meaning.For example,if I want to discussa con_ tions because they do not tell why or how^som._ ceptsuchasurbandecay,Iwillneeda setofas_ thing occurs.Theyaresmallerthan theories,and sociated concepts (e.g., urban expansion, researchers usethem to build a theory.They are economicgrowth, urbanization,suburbs,center broader, more abstractconceptsthai brini to_ city, revitalization,masstransit, and,racial mi_ gether severalnarrower, more concretecon_ norities). cepts. Qualitative researchersoften use ideal Someconceptstake on a rangeof values, typesto see how well observablephenomena quantities,or amounts.Examplesoithis kind of matchup to the idealmodel.For eximple, Max
28
PART o NE , / FoUNDATI oNS
Weber developedan ideal type of the concept bureaucracy. Many peopleuseWeber'sidealtype (seeBox 2.I). lt distinguishesa bureaucracy from other organizationalforms (e.g.,social movements,kingdoms,etc.).It alsoclarifiescriticalfeaturesofa kind oforganizationthat people once found nebulousand hard to think about. No real-life organizationperfectlymatchesthe ideal type, but the model helps us think about and studybureaucracy. Scope. Conceptsvary by scope. Some are highly abstract,someareat a middle levelof abstraction,and some areat a concretelevel (i.e., they are easyto directly experiencewith the sensessuch as sight or touch). More abstract conceptshavewider scope;that is, they can be
r
lt is a continuousorganization governedby a system of rules.
r
Conductis governedby detached,impersonal rules.
r
There is divisionof labor,in which differentoffices are assigneddifferent spheresof competence.
I
Hierarchical authorityrelationsprevail;that is, lowerofficesare undercontrolofhigherones.
r
Administrative actions,rules,and so on are in writingand maintainedin files. r Individuals do not own and cannotbuy or sell their offices. r
Officialsreceivesalariesratherthan receivingdirect paymentfrom clientsin orderto ensureloyalty to the organization.
r
Propertyof the organization is separatefrompersonalpropertyof officeholders.
Source: Adapted fromChafetz (197872).
used for a much broader range of specific time points and situations. More concrete concepts are easy to recognize but apply to fewer situations. The concepts skin pigmentation, casting a ballot in an election,and agebasedon the date on a birth certificateare lessabstract and more concrete than the concepts racial group, democracy and maturity. Theories that use many abstract concepts can apply to a wider range of social phenomena than those with concrete concepts. An example of a theoretical relationship is: Increasedsize createscentralization, which in turn createsgreater formalization . Size,centralizatioi; and formalization are very abstract concepts. They can refer to features of a group, organization, or society. We can translate this to say that as anorganization or group getsbigger, authority and power relations within it become centralized and concentrated in a small elite. The elite will tend to rely more on written policies, rules, or laws to control and organize others in the group or organization. When you think explicitly about the scope of concepts, you make a theory stronger and will be able to communicate it more clearly to others.
Assumptions
Conceptscontain built-in assumptions,statementsaboutthe natureof thingsthat arenot observable or testable. We accept them as a necessary startingpoint. Conceptsand theories build on assumptionsabout the nature of human beings,socialreality, or a particular phenomenon.Assumptionsoften remainhiddenor to deepenhis unstated.Onewayfor a researcher her understanding of a concept is to identifr or which it is the assumptionson based. For example,the conceptbookassumesa systemof writing, peoplewho can read,and the existenceof paper.Without such assumptions the ideaof abook makeslittle sense.A socialscienceconcept,sueh as racialprejudice,restson severalassumptions.Theseinclude peoplewho make distinctionsamong individualsbasedon their racialheritage,attachspecificmotivations
CHAPTER2 , / T H E O R YA N D S O C I A L R E S E A R C H
and characteristicsto membershipin a racial group,and makejudgmentsaboutthe goodness of specificmotivationsand characteristics. If racebecameirrelevant,peoplewould ceaseto distinguish among individuals on the basisof race,to attachspecificcharacteristics to a racial group,and to makejudgmentsaboutcharacteristics.Ifthat occurred,the conceptof racialprejudicewould ceaseto be usefulfor research.All conceptscontainassumptionsaboutsocialrelations or how peoplebehave. Relationships Theoriescontainconcepts,their definitions,and assumptions. More significantly,theoriesspecify how conceptsrelateto one another.Theoriestell us whether conceptsare relatedor not. If they are related,the theory stateshow they relateto eachother.In addition,theoriesgivereasonsfor why the relationshipdoesor doesnot exist.It is a relationship,suchastEconomicdistressamong theWhite populationcausedan increasein mob violenceagainstAfrican Americans.When a researcherempiricallytestsor evaluatessuch a relationship,it is calledahypothesis. After many carefirltestsof a hypothesiswith data confirm the hypothesis,it is treated as a proposition.A propositionis a relationshipin a theoryin which the scientificcommunity startsto gain greater confidenceand feelsit is likely to be truthful.
THE ASPECTSOF THEORY Theory can be baffling becauseit comesin so many forms. To simplify, we can categorizea theory by (1) the direction of its reasoning,(2) the levelof socialrealitythat it explains,(3) the forms of explanationit employs,and (4) the overallframeworkof assumptionsand concepts in which it is embedded.Fortunately,all logically possiblecombinationsof direction, level, explanation,and frameworkare not equallyviable.Thereare only about half a dozenserious contenders.
29
Direction of Theorizing Researchers approachthebuilding andtestingof theoryfrom two directions.Somebeginwith abstractthinking. They logicallyconnectthe ideas in theory to concreteevidence,then test the ideas againstthe evidence.Others begin with specificobservationsof empiricalevidence.On the basisof the evidence,they generalizeand build towardincreasinglyabstractideas.In practice, most researchers are flexibleand useboth approachesat various points in a study (see Figure2.1). Deductive. In a deductiveapproach,you begin with an abstract,logical relationship among concepts,then movetoward concreteempirical evidence.You may haveideasabout how tne world operatesand want to test theseideas against"hard data." Weitzerand Tuch (2004,2005)useda deductiveapproachin a study ofperceptionsof police misconduct.They began with Group Position theory (a middle-rangetheory discussedlater) within the conflict theory framework (seeRangeof Theorylaterin this chapter). Group position theorystatesthat dominant and subordinateracial-ethnicgroupsarein competition for resourcesand statusin a multiethnic societythat hasa racialhierachy,and suchcompetition af[ectsracialbeliefsand attitudes.Dominant groups believe they are entitled to privilegesand a position of superiority,and they fear losing their privileges.Subordinategroups believetheir position can be enhancedif they challengethe existing order. The authors deduced that group competition extendsbeyond attitudesto perceptionsofsocialinstitutions,especiallyinstitutions of socialcontrol such as policing. They arguedthat subordinategroup members (i.e., Blacksand Latino/Hispanics) would preceivepolicemisconduct(measuredas unjustified stops of citizens,verbal abuseby police,an excessive useofforce, and policecorruption) differentlythan membersof the dominant group (Whites).The authorsthought that perceptionsoperatedvia three mechanisms:
30
PART ON E , / FO UNDATI O NS
FIG URE 2. I
Deductiveand InductiveTheorizing InductiveApProach
DeductiveApproach
o iheoreticat
o fheoretical
Middle-Range Theory
Middle-Range Theory
Hypothesis Testing
Grounded Hypothesis, EmpiricalGeneralization Theorizing
EmpiricalSocial Reality
personalencounterswith the police; reports of police encountersby friends,family, or neighbors;and noticingand interpretingnewsreports about police activity. In thesethree areas,they predictedthat non-Whiteswould interpretnegativeeventsor reportsasstrongevidenceofserious and systematicpolice misconduct. By constrast,Whites would tend to ignore or dismiss such eventsor reports or seethem as isolated incidents.Data from a national surveyof U.S. metropolitan areas(over 100,000population) supportedpredictionsof the theory. Inductive. ifyou use an inductiveapproach, you begin with detailed observationsof the world and move toward more abstractgeneralizations and ideas.When you begin, you may haveonly a topic and a few vagueconcepts.As you observe,you refine the concepts,develop and identifr prelimiempirical generalizations, nary relationships.You build the theory from the ground up.
ConceptFormation, EmpiricalGeneralization
EmpiricalSocial RealitY
Duneier (1999)usedan inductiveapproach in his studyof life on the sidewalk.He notedthat in much of socialscience,both quantitativesecondaryanalysisresearchand qualitativefield research, a researcher develops a theoretical understandingonly after datahave been collected.He stated,"I beganto get ideasfrom the thingsI wasseeingandhearingon the street"(p. who adopt an inductive 341).Many researchers approachusegroundedtheory.Groundedtheory is part of an inductive approachin which a researcherbuilds ideasand theoreticalgenetaliza tions basedon closelyexaminingand creatively thinking about the data (seeBox 2'2). A tesearchercreatesgrounded theory out of a processof trying to explain,interpret, and render meaningfrom data.It arisesfrom trying to accountfor, understand,or "make senseof'the that evidence.Duneier(1999:342)hassuggested the processis similar to seeingmany symptoms and later arriving at a diagnosis(i.e.,a storythat explainsthe sourceof the symptoms).
CHAPTER 2
/
TH E OR Y A N D S OC IA L R ES E A R C H
31
Croundedtheoryis a widelyusedapproachin quallt is not the only approachand it is itativeresearch. Crounded thenot usedby all qualitativeresearchers. research methodthat usesa sysoryis"aqualitative tematicset of proceduresto developan inductively and (Strauss derivedtheoryabouta phenomenon" Corbin,1990:24\. The purposeof groundedtheory is to build a theory that is faithful to the evidence.lt is a methodfor discoveringnewtheory.In witha it, the researcher unlikephenomena compares He or sheseesmiviewtowardlearningsimilarities. cro-leveleventsasthe foundationfor a moremacrolevelexplanation. Croundedtheory sharesseveral goalswith more positivist-oriented theory. lt seeks that is theorythat is comparable with the evidence p r ec is eand r igoro u sc, a p a b l eo f re p l i c a ti o na, n d generalizable. A groundedtheoryapproachpursues
by makingcomparisons acrosssogeneralizations cialsituations. Qualitative researchersuse alternativesto offer groundedtheory.Somequalitativeresearchers an in-depthdepictionthat is true to an informant's They excavatea singlesocialsituationto worldview. the microprocesses that sustainstablesocial elucidate Thegoalofother researchers isto provide interaction. a veryexactingdepictionof eventsor a setting.They analyzespecificeventsor settingsin orderto gaininsightinto the largerdynamicsof a society.Stillother applyanexistingtheoryto analyze researchers specific historsettingsthat they haveplacedin a macro-level ical context.They show connectionsamongmicroleveleventsand betweenmicro-level situationsand largersocialforcesfor the purposeof reconstructing the theoryandinformingsocialaction.
Range ofTheory
easyto test or observe.It is calleda generalization becausethe pattern operatesacrossmany time periodsand socialcontexts.The finding in the study on Internet pornographydiscussedin ChapterI that unhappilymarriedmen aremore likely than happily married men to useInternet porn is an empiricalgeneralization.
Social theories operate with varying ranges.One source of the confusion about theories involves the range at which a theory operates.At one end are highly specific theories with concrete concepts of limited scope.At the opposite end are whole systems with many theories that are extremely abstract. As part of the task of theory building, veri$'ing, and testing, a researcher connects theoretical statements of different ranges together, like a seriesof different-sized boxes that fit into one another or a set ofRussian dolls. Empirical Generalizqtion. An empirical generalization is the least abstract theoretical statement and has a very narrow range. It is a simple statement about a pattern or generalization among two or more concrete concepts that are very close to empirical reality. For example, "More men than women choose engineering as a collegemajor." This summarizesa pattern between gender and choice of college major. It is
Midille-RangeTheory. Middle-rangetheories areslightlymore abstractthan empiricalgeneralizationsor a specifichypothesis.A middlerangetheory focuseson a specificsubstantive topic area (e.g., domestic violence,military coups,studentvolunteering),includesa multiple empiricalgeneralization,and builds a theoreticalexplanation(seeForms of Explanation later in this chapter).As Merton (1967:39) stated,"Middle-rangetheoryis principallyused in sociologyto guideempiricalinquiry." A middle-rangetheory usedin a study discussedin Chapter I saidthat girls who sufferphysicalor sexualabuseexperienceself-blameand guilt feelingsthat inhibits them from developinga healthysocialnetworkor formingstableromantic
32
P A Rr oN E / F o U N D AT T o N S
relationships,andthat thesefactorsleadto them stayingsingleor experiencinggreatermarital instabilitywhen they becomeadults. TheoreticalFramewoflcs, A theoreticalframework (alsocalleda paradigmor theoreticalsystem) is more abstract than a middle-range theory.Figure2.1showsthe levelsand how they
Theoretical Framework Kalmijn. Structuralfunctionalismholds that the processes of industrialization andurbanization change humansocietyfroma traditionalto a modernform.In this processof modemization, socialinstitutionsand practices evolve.Thisevolutionincludes thosethat fill peopleto culthe socialsystem's basicneeds,socialize turalvalues,and regulatesocialbehavior.Institutions that filledneedsandmaintained the socialsystemin a traditionalsociety(suchas religion)are superseded by modernones(suchasformalschooling). Weitzerand Tuch. Conflicttheory holdsthat establishedsocial,political,and legalinstitutionsprotect the dominantor privilegedgroupsof a society.Major institutionsoperatein waysthat containor suppress the activitiesof nondominantgroups in society,especiallyif they challengeor threatenthe established social-economic Thus,conflict hierarchy. betweenthe dominantandsubordinate socialgroups is reflectedin how majorinstitutionsoperate,especiallyinstitutionsthat are chargedwith maintaining order and engagedin formalsocialcontrol,suchas lawenforcement. Middle-RangeSubstantiveTheory Kalmijn. A theory of intermarriage patternsnotes that youngadultsin modernsocietyspendlesstime in small,localsettings,wherefamily,religion,and communityall have a strong influence.Instead, youngadultsspendincreasing amountsof time in schoolsettings.In these settings,especiallyin col-
areusedin inductiveand deductiveapproaches makeprecisedisto theorizing.Fewresearchers tinctions amongthe rangesof theorizing.They rarely use a theoretical framework directly in empirical research.A researchermay test parts of a theory on a topic and occasionallycontrast partsof the theoriesfrom differentframeworks. Box2.3illustratesthe variousdegreesof abstrac-
lege,they haveopportunitiesto meetother unmarried people.ln modernsociety,educationhas become a major socializationagent. lt affectsfuture earnings,moralbeliefsand values,and leisureinterpartnersless ests.Thus,youngadultsselectmarriage on the basisof sharedreligiousor localtiesand more on the basisof commoneducationallevels.
WeiEerandTuch. Group-positiontheory usesgroup competitionover materialrewards,power,and status to explainintergroupattitudesand behaviors.Each group perceivesand experiencesreal or imagined threatsto its socialpositiondifferently.Membersof a dominantgrouptendto viewpoliceorgovemmentactions takento defendits interestsas beingfair or favorable,whereasmembersof subodorinategroups tend to seethe sameactionsnegatively. EmpiricalGeneralization once marriedotherswith simiKalmijn. Americans lar religiousbeliefsandaffiliation.Thispracticeis being replacedby marriageto otherswith similarlevels of education. Weitzerand Tuch. Non-Whitesexperiencemore negativeinterpersonalencounterswith police and tend to interpret mediareports about police misconductasevidenceof seriousandsystematicproblems with the police. By contrast,Whites have differentpoliceencountersor interprettheir encountersand mediareportsabout policeactions morefavorably.
CHAPTER2 , / TH E O R YA N D S O C I A L R E S E A R C H
tion with Kalmijn's study of changing marriage partner selection(seealso page40). Sociology and other social scienceshave several major theoretical frameworks.s The frameworks are orientations or sweeping ways of looking at the social world. They provide collections of assumptions, concepts, and forms of explanation. Frameworks include theories for many substantiveareas(e.g.,theories of crime, theories of the family, etc.). Thus, there can be a structural functional theory, an exchange theory and a conflict theory of the family. Theories within the same framework share assumptions and major concepts. Some frameworks are oriented more to the micro level; others focus more on macro-level phenomena (seeLevelsof Theorynext). Box2.4 shows four major frameworks
33
in sociologyand briefly describesthe key conceptsand assumptions of each. Levelsof Theory Socialtheoriescan be divided into three broad groupingsby the levelof socialrealitywith which they deal.Most of us devotethe majority of our time to thinking aboutthe micro leveloireality, the individualswe seeand interactwith on a dayby-daybasis.Micro-leveltheorydealswith small slicesof time, space,or numbersof people.The conceptsareusuallynot very abstract. Braseand fuchmond(2004)useda microleveltheory about doctor-patientinteractions and perceptions.The theorystatedthat physican attire affectsdoctor-patientinteractions.It sug-
;,,.Kffi Structural Functionalism
Symboliclnteractionism
MajorConcepts. System,equilibrium,dysfunction, divisionof labor
MajorConcepts. Sell reference group,role-playing, perception
KeyAssumptions. Societyis a systemof interdependentpartsthat is in equilibrium or balance. Over time,societyhasevolvedfrom a simpleto a complex type,whichhashighlyspecialized parts.The partsof societyfulfilldifferentneedsor functionsof the social system. A basicconsensus on valuesor a valuesvstem holdssocietytogether.
KeyAssumptions.Peopletransmitand receivesvmb o l i cc o mmuni cati on w henthey soci al l yi nteract. Peoplecreateperceptions of eachotherand social settings.Peoplelargelyact on their perceptions. How peoplethink about themselves and othersis basedon theirinteractions. Conflict Theory
ExchangeTheory (also RationalChoice) MajorConcepts. Opportunities,rewards,approval, balance, credit KeyAssumptions.Humaninteractionsaresimilarto economictransactions.Peoplegive and receiveresources(symbolic,socialapproval,or material)and try to maximize their rewardswhileavoidingpain,expense,and embarrassment. Exchange relationstend to be balanced. lfthey areunbalanced, personswith creditcandominateothers.
Major Concepts. Power,exploitation,struggle,inequality,alienation KeyAssumptions.Societyis madeup of groupsthat haveopposinginterests.Coercionand attemptsto gainpowerareever-present aspectsof humanrelations.Thosein powerattemptto hold on to their powerby spreadingmythsor by usingviolenceif necessary.
34
P A RToNE ,/ F o u N D A T to N s
would producegainsbecauseotherworkersand gestedthat a patient makesjudgmentsabout a governmentauthoritieswould supporttheir acphysican'sabilitiesbasedon attire and that a pations. tient'strust-opennesstoward a physicanis alsoafthe operationof Macro-lweltheoryconcerns fected. It said that perceptions of physican suchassocialinstitutions,enauthorityincreasedwith traditionalprofessional larger aggregates tire cultural systems,and wholesocieties.It uses formal attire over informal attire, but that trustmore conceptsthat areabstract. opennesswasinfluencedin the oppositedirection Marx's study (1998)on racein the United asauthority. Thirty-eight male and 40 femalereSouthA.frica,andBraziluseda macro-leve States, searchparticipants rated their perceptionsof theory.He wantedto explainthe conditionsthat same- and opposite-gendermodels who were led Blackpeopleto engagein protestto gain firll identifiedasbeingmedicaldoctors,butwho were of na-. citizenshiprightsandheexaminedpatterns wearing different attire. Findings showedthat a { tional racialpoliticsin threecountiesacrosstwo white coatandformal attire areclearlysuperiorto centuries.His theorysaidthat protestresultedin casualattire in establishingphysicanauthority, asexpected. an interaction between(1) race-basedpolitical but it did not reducetrust-openness mobilizationand (2) nationalgovernmentpoliMeso-lweltheorylinksmacroand micro levciesof racialdomination(i.e.,apartheidin South elsand operatesat an intermediatelevel.TheoAfrica, Jim Crow lawsin southernUnited States, ries of organizations,socialmovements,and and no legalized race-baseddomination in communitiesareoften at this level. Roscignoand Danaher(2001)usedmeso- Brazil).Policiesof racial domination developed from practicesof slavery exploitation,and disleveltheoryin a studyon the i930slabor movecrimination that justified White superiority. The ment among southern textile workers. The policiesreinforcedspecificracialideologiesthat researchers useda theory of movementsubculshapednational developmentduring the twentiture and political opportunity to explaingrowing labor movement strength and increased eth century. A critical causalfactor was how nationalpolitical elitesusedthe legalizeddominastrikeactivityamongworkersin one industryin a regionofthe United Statesacrossseveralyears. tion of Blacksto reducedivisionsamongWhites. In nationsthat had largeregionalor classdiviThey expectedstrike activity to grow asthe result sions among Whites, national elites tried to of a strongmovementsubculturethat carrieda increaseWhite backingfor the nationalgovernmessage of injusticeand a "political opportument by creatinglegalizedforms of racial dominity" or the expectationamongpeoplethat collectiveactionat a particulartime would produce nation. Over time, such legalizeddomination froze racialdivisions,which promoted a senseof positiveresults.Their studyshowedthat a techamongBlacks. racialidentity and consciousness nologicalinnovation (i.e.,the spreadof new raThe strongsenseof racialidentity becamea key dio stations with songs and discussionsof resourcewhenBlacksmobilizedpoliticallyto deworking conditionsand unfair treatment)conmand full citizenshiprights.Legalizedracialdomtributed to the growth of a subcultureof moveination also intensifiedthe Blacks'protest and ment solidarity among the textile workers and directedit againstthe national governmentasthe fosteredself-identityasa worker who had comsocietalinstitution that reinforcedtheir experimon interestswith the othertextileworkers.The enceof racialinequality. technologicalinnovationand eventsin thepolitical environment (i.e., union organizersand speeches by the Presidentof the United States) Forms of Explanation alsocreateda politicalopportunity for theworkPrediction and Explanation. A theory's priers.The workersbelievedthat collectionaction mary purposeis to explain.Many peoplecon(i.e.,strike)wasnecessary to achievejusticeand
CHAPTER2 /
T H E O R YA N D S O C I A L R E S E A R C H
35
fusepredictionwith explanation.Therearetwo probablyacceptinvolvesa theory about the ro_ meaningsor usesof the term explanation.Re_ tation of the earth and the position of the sun, searchers focuson theoretical explanation,alogthe star of our solarsystem.lnthis explanation, icalargumentthat tellswhysornethingoccurs.It the sun only appearsto rise. The sun doesnot refersto a generalrule or principle. ih"r. u." u move; its apparentmovement dependson the researcher'stheoretical urgum.rrt or connec_ earth'srotation. We are on a planet that both tionsamongconcepts.Thesecondtypeof expla_ spinson its axisand orbitsarounda starmillions nation, ordina-ryexplanation,makessomething of miles away in space.All three explanations clearor describessomethingin a way that illusl make the sameprediction: The sun'riseseach tratesit and makesit intelligible.Forexample,a morning.As you cansee,a weakexplanationcan good teacher"explains,,in the ordinury *rrr". produce an accurateprediction.A good expla_ *r typesof explanationcanblendtogether. nation dependson a well_developeJtheory Ilr: and This occurswhen a researcherexplaini (i.e., is confirmed in researchby empirical observa_ intelligible)his or her explanation(ire.,a tions. 3e: logicalargumentinvolving tr.ory). Prediction is a statement that something CausalExplanation. Causalexplanation, the will occur.It is easierto predict than to explain, most commontypeof explanation,is used when andan explanationhasmore logicalpowei than the relationshipis one o].u.rr" and effect. We predictionbecause good explariations alsopre_ useit all the time in everydaylanguage, which dlct. An explanationrarelypredictsmore than tendsto be sloppyand ambiguous."wf,ui do _. one outcome,but the sameoutcomemay be meanwhen we saycause? Foiexample,you may predictedby opposingexplanations.Although saythat povertycauses crimeor thailooseness in it-is lesspowerfirl than explanation,many peo_ moralscilusesan increasein divorce.This does ple areentrancedby the dramaticvisibility of a not tellhow or why the causalprocessworks. prediction. Researchers try to be more preciseand exact A gambling exampleillustratesthe differ_ when discussingcausalrelations. encebetweenexplanationand prediction. If I Philosophershavelong debatedthe idea of enter a casinoand consistentlyand accurately cause.Somepeoplearguethat causalityoccurs predictthe next cardto appearor the next num_ in theempirical world, but it cannotbe'proved. ber on a roulettewheel,if mt U. ,.rrrutiorr"i. t Causalityis "out there" in objective,.uliry arrd maywin a lot of money,at leastuntil the casino researchers can only try to find evidencefor it. officials realizeI am alwayswinning urra."pel Othersarguethat causalityis only an idea that me.Yet,my methodof makingthepiedictions is existsin the human mind, a mental construc_ mo-reinterestingthan the faci thaf I can do so. tion, not something ..real,,in the world. ihis Tellingyou what I do to predict the next cardis secondposition holds that causalityis only a more_fascinating than beingableto predict. _convenientway of thinking about ihe rvorta. Hereis anotherexample.you know that the Without enteringinto the lengthyphilosophical sun "rises" eachmorning. you can predict that debate,manyresearchers pursuecausalrelation_ at some everymorning, wheiher or not ships. -time, cloudsobscureit, the sun will rise.But why is You needthreethingsto establishcausality: this so?One explanationis that the Great irl f" temporalorder,association, and the eliminatiln carriesthe sun acrossthe sky on its back. an_ of plausible alternatiyes.An implicit fourth other explanationis that a god setshis arrow condition is an assumptionthat u .uur"t-..tu_ ablaze,.whichappearsto ui as the sun, and tionship makes re.tseo, fits with U.ouj", snootsrt across,!. rt y. Fewpeopletodaybelieve ur_ sumptions or a theoreticalframework. Let us theseancientexplanations.Theexplanaiion you examinethe threebasicconditions.
36
pA RToN E ,/ F o u N D AT to N s
The temporal order condition means that a cause must come before an effect. This commonsense assumption establishesthe direction of causality: from the cause toward the effect. You may ask, How can the cause come after what it is to affect?It cannot, but temporal order is only one of the conditions needed for causality. Temporal order is necessarybut not sufficient to infer causality. Sometimes people make the mistake of talking about "cause" on the basis of temporal order alone. For example, a professional baseballplayer pitches no-hit gameswhen he kisseshis wife just before a game. The kissing occurred before the no-hit games. Does that mean the kissing is the causeof the pitching performance? It is very unlikely. As another example, race riots occurred in four separatecities in 1968, one day after an intense wave of sunspots. The temporal ordering does not establish a causal link between sunspots and race riots. After all, all prior human history occurred before some specific event. The temporal order condition simply eliminates from consideration potential causesthat occurred later in time. It is not always easy to establish temporal order. With cross-sectionalresearch,temporal order is triclcF. For example, a researcher finds that people who have a lot ofeducation are also less prejudiced than others. Does more education causea reduction in prejudice? Or do highly prejudiced people avoid education or lack the motivation, self-discipline, and intelligence needed to succeedin school?Here is another example. The students who get high grades in my classsay I am an excellent teacher. Does getting high gradesmake them happy, so theyreturn the favor by saylng that I am an excellent teacher (i.e., high gradescausea positive evaluation)? Or am I doing a great job, so students study hard and learn a lot, which the grades reflect (i.e., their learning causesthem to get high grades)?It is a chicken-or-egg problem. To resolve it, a researcher needs to bring in other information or design researchto test for the temporal order. Simple causal relations are unidirectional. operating in a single direction from the causeto
the effect. Most studies examine unidirectional relations. More complex theories speci$r reciprocal-effect causal relations-that is, a mutual causal relationship or simultaneous causality. For example, studying a lot causesa student to get good grades, but getting good grades also motivates the student to continue to study. Theories often have reciprocal or feedback relationships, but these are difficult to test. Some researcherscall unidirectional relations nonrecursive and reciprocal-effect relations recursive. A researcher also needs an association for causality. Two phenomena are associatedif thei. occur together in a patterned way or appear to act together. People sometimes confuse correlation with association. Correlation has a specific technical meaning, whereasassociationis a more generalidea. A correlation coefficient is a statistical measure that indicates the amount of association, but there are many ways to measure association. Figure 2.2 shows 38 people from a lower-income neighborhood and 35 people from an upper-income neighborhood. Canyou seean associationbetween race and income level? More people mistake associationfor causality than confuse it with temporal order. For example, when I was in college,I got high gradeson the exams I took on Fridays but low grades on those I took on Mondays. There was an association between the day of the week and the exam grade, but it did not mean that the day of the week causedthe exam grade. Instead, the reason was that I worked 20 hours each weekend and was very tired on Mondays. As another example, the number of children born in India increased until the late 1960s,then slowed in the 1970s.The number of U.S.-made cars driven in the United Statesincreaseduntil the late 1960s,then slowed in the 1970s.The number of Indian children born and the number of U.S. cars driven are associated:They vary together or increaseand decrease at the same time. Yet there is no causal connection. By coincidence, the Indian government instituted a birth control program that slowed the number of births at the same time that Americans were buying more imported cars.
CHAPTER2 , / THE O R YA N D S O C I A L R E S E A R C H
Association of Income and Race
FIG U R E 2 .2
UpperIncome
Lower Income
A
OO
a
0 AI rC (-) \ I J rJl
io o o
O
fifi ffAfi.H, fii
o' i*i\q;i i:njH l{lfifi nfi^q.B"qfr flqi
?*$ m? 3
Qr T.
37
t ffi,i.in a?
/l IV
t
.l iii lt t?
v'fifr f,fl
Ifa researchercannot find an association, a causal relationship is unlikely. This is why researchersattempt to find correlations and other measuresof association.Yet, a researchercan often find an associationwithout causality.The association eliminates potential causesthat are not associated, but it cannot definitely identify a cause.It is a necessarybut not a sufficient condition. In other words, you need it for causality, but it is not enough alone. An association does not have to be perfect (i.e., every time one variable is present, the other also is) to show causality. In the example involving exam gradesand days of the week, there is an association if on 10 Fridays I got 7 As, 2 Bs, and 1 C, whereas my exam grades on 10 Mondays were 6 Ds, 2 Cs, and 2 Bs. An association exists' but the days ofthe week and the exam gradesare not perfectly associated.The race and incomelevel association shown in Figure 2.2 is also an imperfect association. Eliminating alternatives means that a researcher interested in causality needs to show that the effect is due to the causal variable and not to something else. It is also called no spuriousnessbecausean apparent causal relationship that is actually due to an alternative but unrec-
ognized cause is called a spurious relationship, which is discussedin Chapter 4 (seeBox 2.5). Researcherscan observetemporal order and associations.They cannot observe the elimination of alternatives. They can only demonstrate it indirectly. Eliminating alternatives is an ideal because eliminating all possible alternatives is impossible. A researchertries to eliminate major alternative explanations in two ways: through built-in design controls and by measuring potential hidden causes.Experimental researchers build controls into the study design itself to eliminate alternative causes.They isolate an experimental situation from the influence of all variables except the main causalvariable. Researchersalso tryto eliminate alternatives by measuring possible alternative causes.This is common in survey research and is called controlling for another variable. Researchersuse statistical techniques to learn whether the causal variable or something elseoperateson the effect variable. Causal explanations are usually in a linear form or state causeand effect in a straight line: A causesB B causesC C causesD. The study by Braseand Richmond (2004) on doctor-patient interactions discussedearlier
38
PART ONE /
FO UNDATI O NS
As I wasdrivinghomefrom the universityone day, I hearda radio newsreport about genderand racial biasin standardized tests.A personwho claimedthat biaswasa majorproblemsaidthat the tests should be changed.SinceI workin the fieldof educationand disdainracialor genderbias,the report caughtmy attention.Yet, as a socialscientist,I criticallyevaluated the newsstory.The evidencefor a biascharge wasthe consistentpatternofhigherscoresin mathematicsfor malehigh schoolseniorsversusfemale high schoolseniors,and for European-background studentsversusAfricanAmericanstudents.Wasthe causeof the pattern of differenttest scoresa bias built into the tests? When questionedby someonewho had designedthe tests,the personchargingbias lackeda crucialpieceof evidenceto supporta claimof test
useda causalexplanation;it saidphysicanattire causes certaintypesofpatient perceptions. The studybyWeitzerand Tuch (2004,2005)on police misconduct cited earlier used a causalexplanation. The causewas a person's group position and competitivepressurewith other groups.Theseare causallylinked to police encounters,eitherdirectlyor indirectly,and interpretionsof newsreports,which differ by group position.The policeencountersand the interpretationsofnews reports causevery different perceptionsof police misconduct.We can restatethe logic in a deductivecausalform: If the proposition is true, then we observecertain things in the empirical evidence.Good causal explanationsidentift a causalrelationshipand specifi'acausalmechanism.A simplecausalexplanationis: X causesYor Yoccursbecauseof X, whereX and Yare concepts(e.g.,earlymarriage and divorce). Some researchersstate causalityin a predictiveform: If X occurs,then Yfollows.Causalitycanbe statedin manyways:
bi as:the educati onalexperi ence of students.l t turnsout that girlsand boystakedifferentnumbers and types of mathematics coursesin high school Cirls tend to take fewer math courses.Amongthe girlswho completethe samemathematics curricu lum as boys,the genderdifferencedissolves.Likewise, a large percentageof African American attend raciallysegregated,poor-qualityschoolsin i nner ci ti es or i n i mpoveri shed rural areas.For AfricanAmericans who attend high-qualitysuburbanschoolsand completethe samecourses,racia differencesin test scoresdisappear.This evidence suggeststhat inequalityin educationcausestest score differences.Although the tests may have problems,identifyingthe real causeimpliesthat changing the testswithoutfirst improvingor equalizingeducationcouldbe a mistake.
Xleadsto Y,Xproducesy, XinfluencesY,Xis relatedto Y, the greaterXthe higher Y. Here is a simplecausaltheory:A risein unemploymentcausesan increasein child abuse. The subjectto be explainedis an increasein the occurrenceof child abuse.What explainsit is a rise in unemployment.We "explain" the increasein child abuseby identifring its cause.A completeexplanationalso requireselaborating the causalmechanism.My theorysaysthat when peoplelosetheir jobs, they feel a loss of selfworth. Once they lose self-worth,they become easilyfrustrated, upset, and angry. Frustrated peopleoften expresstheir angerby directingviolencetoward thosewith whom they haveclose personalcontact(e.g.,friends,spouse,children, etc.).This is especially true if they do not understandthe sourceof the angeror cannotdirect it towardits true cause(e.g.,an employer,government policy,or "economicforces"). The unemploymentand child abuseexample illustratesa chain of causesand a causal
CHAPT ER 2
mechanism.Researchers can testdifferentparts of the chain. They might test whetherunemployment ratesand child abuseoccur together, or whetherfrustratedpeoplebecomeyiolent toward the peoplecloseto them.A typicalresearch strategyis to dividea largertheoryinto partsand testvariousrelationshipsagainstthe data. Relationships betweenvariablescanbe positive or negative.Researchers imply a positiverelationship if they say nothing. A positive relationshipmeansthat a higher value on the causalvariablegoeswith a higher value on the effectvariable.For example,the more education a personhas,the longerhis or her life expectancF is. A negativerelationshipmeansthat a higher value on the causalvariablegoeswith a lower value on the effect variable. For example,the more frequentlya couple attendsreligiousservices,the lower the chancesof their divorcing eachother.In diagrams,a plus sign (+) signifies a positiverelationshipand a negativesign (-) signifiesa negativerelationship.
,/ T HE OR Y A N D S OC IA L R E S E A R C H
39
immigrants from the Philippines,Korea, Taiwan, and China in Los Angelesand found that socialnetworksmatchedand sortedimmigrants with jobs. New immigrants with limited languageand job skills soughtemploymenteither with a co-ethnicemployeror through informal social ties (i.e., they consulted experienced fiiends, relatives,and acquaintances and asked them to be intermediaries).Network usersexpandedjob opportunitiesbeyondemployersin their own ethnic group. Thus, ethnic network tieswere"bridgeties" (i.e.,they helpedimmigrantsgetjobs beyondtheir ethnic community by usingco-ethnicswho alreadymadethe transition to mainstreamemplo;rment).Over time, as languageand job skills improved, theseimmigrantsmoved on to mainstreamjobs. Immigrantslackingsocialties,in limited networks,or who worked for co-ethnicsfound it difficult to get a mainstreamjob. Thus, a person'snetwork location,accessto alarge and diversenetwork, and use of network ties is what facilitatedobtaining a mainstreamjob. Structural Explanation. A structuralexplanaStructuralexplanationsare alsousedin serlonis usedwith threetypesof theories:network, quence theory. The panel study on volunsequential,and functional theories.Unlike a teerismby Oesterle,|ohnson, and Mortimer causaleffectchain,which is similar to a stringof (2004) discussedin Chapter 1 employs seballslined up that hit one anothercausingeach quencetheory.The authorsuseda "life course" to bouncein turn, it is more similar to a wheel perspectivein which the impact of an event with spokesfrom a centralidea or a spiderweb happeningat one phaseof a person'slife differs in which eachstrandforms part of the whole.A what it would havebeenif the samehappenedat researcher making a structuralexplanationuses other phases,and early eventsgenerallyshape a set of interconnectedassumptions,concepts, eventsin later phases.The authors noted that and relationships.Insteadof causalstatements, the transition to adulthood is a critical stage he or sheusesmetaphorsor analogiessothat rewhen a personlearnsnew socialrolesand adult lationships"make sense."The conceptsand reexpectations.They found that the amountsand lations within a theory form a mutually typesofvolunteer activity in the last stagethey reinforcingsystem.In structuralexplanations,a observed(age 26-27) was strongly influenced researcher specifiesa sequence ofphasesor idenby suchactivitiesat prior stagesofa person'slife tifies essentialparts that form an interlocked (age18-19).Peoplewho volunteeredat an early whole. stagetendedto volunteer at later stages.Those Structuralexplanationsareusedin network who did not volunteerat an earlystageor who theory. Sanders,Nee, and Sernau(2002) exdevoted full time to working or parenting at plainedAsian immigrant job seekingwith netother prior stages(18-19 yearsold) were less work theory. They used interview data on likely to volunteer at a later stage(26-27 years
PART ONE
,/ F OUNDAT IONS
old). Thus, later eventsflowed from an interconnectedprocessin which earlier stagesset a course or direction that pointed to specific eventsin a later stage. Additionally, structural explanationsare usedin functionaltheory.6Functionaltheorists explain an eventby locating it within a larger, ongoing,balancedsocialsystem.They often use biologicalmetaphors.Theseresearchers explain somethingby identifying its function within a larger systemor the need it firlfills for the system.Functionalexplanationsarein this form: "I occursbecauseit servesneedsin the systemM." Theoristsassumethat a systemwill operateto stayin equilibrium and to continueovertime. A functionaltheoryof socialchangesaysthat, over time, a socialsystem,or society,movesthrough developmentalstages,becoming increasingly differentiatedand more complex. It evolvesa specialized divisionoflabor anddevelopsgreater individualism. These developments create greaterefficiencyfor the systemasa whole.Specializationand individualism createtemporary disruptions.Thetraditionalwaysof doingthings weaken,but new social relations emerge.The systemgenerates newwaysto fulfill functionsor satisfyits needs. Kalmijn (1991)useda functionalexplanation to explain a shift in how people in the United Statesselectmarriagepartners.He relied on secularizationtheory, which holds that ongoing historicalprocesses of industrialization and urbanizationshapethe developmentof society. During thesemodernizationprocesses, peoplerely lesson traditional ways of doing things. Religiousbeliefs and local community ties weaken,as doesthe family's control over young adults.Peopleno longer live their entire lives in small, homogeneouscommunities. Young adultsbecomemore independentfrom their parentsand from the religious organizations that formerly playeda critical role in selectingmarriagepartners. Societyhasa basicneedto organizethe way peopleselectmarriagepartnersand find partnerswith whom they sharefundamentalvalues.
In modern society,people spend time away from small local settingsin school settings.In theseschoolsettings,especiallyin college,they meet other unmarried people.Educationis a major socializationagentin modern society Increasingly,it affectsa person'sfuture earnings, moral beliefs and values,and ways of spendingleisuretime. This explainswhy there hasbeena trend in the United Statesfor people to marry lesswithin the samereligion and increasinglyto marry personswith a similar level of education.In traditional societies,the family and religious organizationservedthe function of socializingpeopleto moral valuesand linking them to potential marriagepartnerswho held similar values.In modern society,educational institutions largelyfulfill this function for the socialsystem.
Interpretive Explanation The purposeof an interpretiveexplanationis to fosterunderstanding. The interpretive theorist attempts to discoverthe meaningof an eventor practiceby placingit within a specificsocialcontext.He or shetries to comprehendor mentally graspthe operationof the socialworld, aswell asgeta feel for somethingor to seethe world asanotherperson does. Becauseeach person's subjective worldview shapeshow he or she acts,the researcherattemptsto discernothers' reasoning and view of things. The processis similar to decodinga text or work of literature.Meaning comesfrom the context of a cultural symbol system. Duneier's(1999)study of sidewalklife in New York City discussedearlierin this chapter usedan interpretiveexplanation.An interpretive explanation is also illustrated by Edelman, Fuller, and Mara-Drita's(2001)study of how companiesadoptedpoliciesrelatedto diversity issuesin the early1990s-that is, affirmativeaction and equalopportunity. The authorsexaminedwhat managerssaid,or their rhetoric,about diversityconcerns.Rhetoricincludedvarious statementsaboutdiversitymadeby professiona managers,businessschoolprofessors,and con-
CHAPTER2 , / T H E O R YA N D S O C I A L R E S E A R C H
4l
sultants in professional workshops, meetings, T HE TH R E E MA ' OR A P P R OA C H E S specializedmagazines,and electronic forums. T O S OC IA L S C IE N C E Edelman and colleagues(2001) found that managerstook legal ideas,terms, and concepts We began this chapter by looking at small-scale and converted them into ones that fit into their parts of a theory (i.e., ideas or concepts). We or ganizational setting. Professional managers moved toward larger aspectsof social theory, converted vague legal mandates and terms that and arrived at major theoretical frameworks in were basedon ideasabout racial discrimination the last section. Now, we move to an even a and ending injustice. They interjected their broader, more abstract level of the linkage beown views, values, training, and interests and tween theory and research-fundamental approduced slightly different ideas and proceproaches to social science. It involves issues dures. Management rhetoric changed legal sometimes called m etq - m ethodological (i. e., beideas from taking specific actions to end yond or supersizedmethodological concerns) racial-ethnic or gender discrimination and and blurs into areasofphilosophy that studies changed them into a "new idea" for effective what science means. We only brieflv touch on corporate management. The "new idea" was the issueshere, but we cannot ignore them bethat corporations benefit from a culturally dicausethey affect how people do social research verse workforce. Simply put, diversity is good studies. for company profits. They consolidated various About 45 years ago, a now famous philosostudiesand discussionson how to improve corpher of science,Thomas Kuhn, argued that the porate operations around the new idea-a soway science develops in a specific field across cially heterogeneous workforce is more time is based on researcherssharing a general creative, productive, and profitable. approach, or paradigm. A paradigm is an inteThe authors created a theory of "managerigrated set of assumptions, beliefs, models of doalization of law" from their data. This theory ing good research,and techniques for gathering statesthat professional managers operate in a and analyzing data.It organizescore ideas,theocorporate environment. They will not simply retical frameworks, and researchmethods. Kuhn take ideas and mandates created in a governobserved that scientific fields tend to be held toment-legal environment and impose them digether around a paradigm for a long period of rectly onto a corporation's internal operations. time. Very few researchersquestion the paraIn fact, on the issue of affirmative action, many digm, and most focus on operating within its corporate officials saw the legal ideas and regeneral boundaries to accumulate new knowlqirirements as hostile or alien. So the managers edge. On rare occasions in history, intellectual converted, or translated, the legal ideas into an difficulties increase,unexpected issuesgrow, and acceptable fslm-6ns acceptable from a mantroubling concerns over proper methods multiagerial point of view. They used new forms to ply. Slowly, the members of a scientific field shift move their corporations in a direction that in how they seethings and switch to a new parawould comply with the legal requirements. This digm. Once the new paradigm becomesfirlly esis an interpretive explanation because the autablished and widely adopted, the process of thors explained a social event (i.e., corporations accumulating knowledge begins anew. embracing programs and rhetoric to favor Kuhn's explanation covered how most scicultural diversity) by examining how the manencesoperate most of the time, but some fields agerssubjectively constructed new ways oflookoperate with multiple or competing paradigms. ing at, thinking about, and talking about the This is the casein severalofthe social sciences. diversity issue (i.e., they constructed a new This greatly bothers some social scientists, and interpretation). they believe having multiple paradigms hinders
I'A I( I ()N L
/
IOUNDA
II( ) N5
the growth ofknowledge.Theyseemultiple paradigmsasa signof the immaturity or underdevelopmentof the "science"in the socialsciences. Some believe all social scienceresearchers shouldembracea singleparadigmand stop using alternativesto it. Other socialscientistsacceptthe coexistence of multiple paradigms.They recognizethat this canbe confusingand often makescommunicating difficult amongthosewho usea different approach.Despitethis, they arguethat eachsocial scienceparadigmprovidesimportant kinds of knowledgeand insights,so to drop one would limit what we can learn about the socialworld. Thesesocialscientistsnotethat no one definitely cansaywhichapproachis "best"or evenwhether it is necessary or highly desirableto haveonly one paradigm.Soinsteadof closingoff an approach that offersinnovativewaysto study sociallife and gain insight into human behavior,they arguefor keepinga diversityofapproaches. In this section,we will look at three fundamental paradigmsor approachesusedin social science. Eachapproachhasbeenaroundfor over 150yearsand is usedby many highly respected professionalresearchers. Theseapproachesare unequal in terms of the number of followers, quantity ofnew studies,and typesofissuesacrdressed.Often, peoplewho strongly adhereto one approachdisagreewith researchers who use another,or seethe otherapproaches asbeingless valuableor less"scientific"than their approach. Although adherentsto eachapproachmay use variousresearchtechniques,theories,and theoretical frameworks,researchers who adopt one approachtend to favor certainresearchtechniques,theories,or theoreticalframeworksover others.The threeapproachesarepositivism,interpretive,and critical; eachhas internal divisions,offshoots,and extensions, but theseare the coreideasof the threemajor approaches. Positivist Approach Positivismis themostwidelypracticedsocialscience approach,especiallyin North America.
Positiyismseessocialscienceresearchasfundamentallythe sameasnatural scienceresearch;it assumesthat socialreality is madeup of objeccanprecisely tive factsthat value-freeresearchers measureand usestatisticsto testcausaltheories. companies bureaucraticagencies, Large-scale and many peoplein the generalpublic favor a positivist approachbecauseit emphasizesgetting objectivemeasuresof "hard facts" in the form of numbers. Positivistsput a greatvalueon the principle of replication,evenif only a fewstudiesarereplicated.Replicqtionoccurswhen researchersor othersrepeatthe basicsofa study and get identical or very similar findings. Positivists emphasizereplication and the ultimate test of knowledge.This is becausetheybelievethat diflooking at the samefactswill get ferentobservers the sameresults if they carefully speci!' their ideas,preciselymeasurethe facts,and follow the standardsof objectiveresearch.When many yield similar studiesby independentresearchers findings, confidencegrows that we accurately capturedtheworkingsof socialrealityandtherefore scientificknowledgeincreases. Ifa researcherrepeatsa studyand doesnot getsimilarfindings,oneor more of five possibilities may be occurring:(1) the initial studywas an unusualfluke or basedon a misguidedunderstandingof the socialworld; (2) important conditionswerepresentin the initial study,but no one was awareof their significanceso they were not specified;(3) the initial study, or the repeatof it, wassloppy-it did not includevery (4) the initial study,or precisemeasures; carefi.rl, the repeatof it, wasimproperlyconducted-researchersfailed to closely follow the highest standardsfor proceduresand techniques,or failed to be completelyobjective;or (5) the repeatedstudywasan unusualfluke. The positivist approachis nomothetic;it meansexplanationsuselaw or law-like principles.Positivistsmay useinductiveand deductive inquiry, but the ideal is to develop a general causallaw or principle then uselogical deduction to specifrhow it operatesin concretesitua-
CHAPTER2 , / TH E O R YA N D S O C I A L R E S E A R C H
tions. Next, the researcher empiricallytestsoutcomespredictedby the principlein concretesettings usingvery precisemeasures.In this way, a generallaw or principlecoversmanyspecificsituations.For example,a generalprinciple says that when two socialgroupsare unequaland competefor scarceresources,in-group feelings and hostility toward the other groupsintensify, and the competinggroupsarelikely to engagein conflict. The principle appliesto sportsteams, countries,ethnicgroups,families,and other social groupings.A researchermight deducethat in citieswith high levelsof interracialinequality, when jobs becomemore scarceand therebyincreaseeconomic competition, eachgroup will expressgreaterhostility about the other racial groups, and intergroup conflict (e.g., riots, demonstrations,violent attacks)will increase. The vast majority of positivist studiesare quantitative,and positivistsgenerallyseethe experiment as the idealway to do research.positivist researchers also use other quantitative researchtechniques,suchassurveysor existing statistics,but tend to seethem as approximations of the experimentfor situationswherean experimentis impossible.Positivistresearchers advocatevalue-freescience,seekprecisequantitative measures,test causaltheorieswith statistics,and believein the importanceof replicating studies. Interpretive Approach The interpretiveapproachis alsoscientific,but its seesthe idea of "scientific" differently from positivism.Unlike thepositivistapproach,interpretive researchers saythat human sociallife is qualitativelydifferentfrom other thingsstudied by science.This meansthat socialscientistscannot just borrow the principlesof sciencefrom the natural sciences. Instead,they believeit is necessary to createa specialtype ofscience,one basedon the uniqueness of humansandonethat canreallycapturehuman sociallife. Most researchers who use an interpretive approachadopt a versionofthe constructionist
43
view of social reality. This view holds that human social life is based less on objective, hard, factual reality than on the ideas,beliefs, and perceptions that people hold about reality. In other words, people socially interact and respond basedas much, if not more, on what theybilieve to be real than what is objectively real. This means that social scientistswill be able to understand social life only if they study how people go about constructing social reality. As people grow up, interact, and live their daily lives, they continuously create ideas, relationships, symbols, and roles that they consider to be meaningful or important. Theseinclude things such as intimate emotional attachments, religious or moral ideals,beliefs in patriotic values,racial-ethnic or gender differences, and artistic expressions. Rarely do people relate to the objective facts of reality directly; instead, they do so through the filter of these socially constructed beliefs and perceptions. What positivists and many people view to be objective facts (e.g., a person's height), interpretive researcherssay are only at the trivial surface level of social iife. Or, the "facts" are images/categoriesthat humans created (i.e., I am two meters tall) and we "forget,' that people originated the images/categoriesbut now treat them as being separate from people and objectively real. Interpretive researchersare skeptical ofthe positivist attempts to produce precise quantitative measures of objective facts. This is because they view social reality as very fluid. For most humans, social reality is largely the shifting perceptions that they are constantly constructing, testing, reinforcing, or changing and that have become embedded in social traditions or institutions. For this reason, interpretive researchers tend to trust and favor qualitative data. Theybelieve that qualitative data can more accurately capture the fluid processesof social reality. In addition, they favor interpretive over causal forms of theory (see discussion earlier in this chapter). Interpretive researchersare not likely to adopt a nomothetic approach,but insteadfavor
44
P A RToNE ,/ F o u N D A T ro N s
an idiographic form of explanation and use inductive reasoning. Idiographic literally means specific description and refers to explaining an aspectof the social world by offering a highly detailed picture or description of a specific social setting, process, or t'?e of relationship. For example, qualitative researchersdo not seereplication as the ultimate test of knowledge. Instead, they emphasize verstehenor empathetic understanding. Verstehenis the desire of a researcher to get inside the worldview of those he or she is studying and accurately represent how the people being studied seethe world, feel about it, and act. In other words, the best test of good social knowledge is not replication but whether the researcher can demonstrate that he or she really captured the inner world and personal perspective of the people studied.
because it profoundly shapes much of human action. The critical approach has an activist orientation and favors action research. Praxis is the ultimate test of how good an explanation is in the critical approach. It is a blending of theory and concrete action; theory informs one about the specific real-world actions one should take to advancesocial change,and one usesthe experiencesofengaging in action for social changeto reformulate the theory. All the approachesseea mutual relationship between abstract theory and concrete empirical evidence, but the critical approach goesfurther and tries to dissolve the gap between abstract theory and the empirical experiences of using the theory to make changes in the world.
Critical Approach
THE DYNAMIC DUO
The critical approachsharesmany featureswith an interpretiveapproach,but it blendsan objective/materialistwith a constructionistview of social reality. The key feature of the critical approachis a desireto put knowledgeinto action and a beliefthat researchis not valuefree. is the creationofknowledge,and peoResearch ple regularlyuseknowledgeto advancepoliticalmoral ends.This givesdoing socialresearcha strongconnectionto political-moralissues.The researchercan decideto ignore and help those with power and authority in society,or advance socialjusticeand empowerthe powerless. the multilayCritical approachemphasizes erednatureofsocialreality.On the surfacelevel, thereis oftenillusion,m1th,anddistortedthinking. The critical approachnotesthat peopleare often misled, are subjectto manipulatedmessages,or hold falseideas.Yet, beneaththe surface level at a deeper,often hidden level lies "real" objectivereality.Part ofthe taskofsocial researchis to strip awaythe surfacelayerof illusion or falsehood.Although a researcherwants to seebeyondthis layer,he or shedoesnot entirely ignoreit. Suchan outer layeris important
You haveseenthat theory and researchare interrelated.Only the naive,new researchermistakenly believes that theory is irrelevant to researchor that a researcherjust collectsthe who attemptto proceedwithdata.Researchers out theory may wastetime collectinguseless data.They easilyfall into the trap of hazy and vaguethinking, faulty logic, and impreciseconcepts.They find it difficult to convergeonto a crisp researchissueor to generatea lucid account of their study'spurpose.They also find themselvesadrift as they attempt to designor conductempiricalresearch. Thereasonis simple.Theoryframeshowwe look at and think about a topic. It givesus concepts,providesbasicassumptions,directsus to waysfor the important questions,and suggests us to make senseof data.Theory enablesus to connecta singlestudy to the immensebaseof knowledgeto which other researcherscontribute. To use an analogy,theory helps a researcherseethe forestinsteadofjust a single awareness tree. Theory increasesa researcher's of interconnectionsand of the broader significanceofdata (seeTable2.1).
CHAPTER2 /
TABLE 2.'l
MajorAspectsandTypes of SocialTheory
Direction
lnductiveor deductive
Level
Micro,meso,or macro
Explanation Causal,interpretive,or structural Abstraction
Empirical generalization, middle range,framework,or paradigm
Theory hasa placein virtually all research, but its prominencevaries.It is generallyless central in applied-descriptiveresearchthan in basic-explanatoryresearch.Its role in appliedand descriptiveresearchmay be indirect. The conceptsareoften moreconcrete,andthe goalis not to creategeneralknowledge.Nevertheless, researchersuse theory in descriptiveresearchto refine concepts,evaluateassumptionsof a theory, and indirectlytesthypotheses. Theorydoesnot remainfixedovertime; it is provisionaland opento revision.Theoriesgrow into more accurateand comprehensive explanations about the make-up and operation of the socialworld in two ways.They advanceastheorists toil to think clearlyand logically,bur this effort haslimits. The way a theorymakessignificant progressis by interactingwith research findings. The scientificcommunity expandsand alters theoriesbasedon empirical results.Researchers who adopta more deductiveapproach usetheoryto guidethe designof a studyand the interpretationof results.They refute,extend,or modify the theory on the basisof results.As researchers continueto conductempiricalresearchin testing a theory, they developconfidencethat somepartsof it aretrue. Researchers may modify somepropositionsof a theoryor reject them ifseveralwell-conductedstudieshave negativefindings.A theory's core propositions
TH E O R YA N D S O C I A L R E S E A R C H
45
and central tenets are more difficult to test and are refuted less often. In a slow process, researchersmay decide to abandon or change a theory as the evidence against it mounts over time and cannot be logically reconciled. Researchers adopting an inductive approach follow a slightly different process.Inductive theorizing begins with a few assumptions and broad orienting concepts. Theory develops from the ground up as the researchersgather and analyze the data. Theory emerges slowly, conceptby conceptand proposition by proposition in a specific area. The processis similar to a long pregnancy. Over time, the concepts and empirical generalizationsemerge and mature. Soon, relationships become visible, and researchersweavetogether knowledge from dif[erent studies into more abstract theory.
CONCLUSION In this chapter,you learnedabout socialtheory-its parts, purposes,and t1pes.The dichotomy betweentheory and researchis an artificial one.The valueof theory andits necessity for conductinggood researchshould be clear.Researchers who proceedwithout theory rarely conduct top-quality researchand frequentlyfind themselves in a quandary.Likewise, theoristswho proceedwithout linking theoryro researchor anchoringit to empiricalreality are in jeopardyof floatingoffinto incomprehensible speculationand conjecture.You arenow familiar with the scientificcommunity,the dimensionsof research, andsocialtheory.
Key Terms association assumption blame analysis causalexplanation classificationconcept concept cluster
P A R I (JN L
/
IUUNUA
deductive approach empirical generalization functionaltheory grounded theory ideal type idiographic inductive approach macro-level theory meso-leveltheory micro-level theory negative relationship nomothetic paradigm positive relationship praxis prediction proposition replication verstehen
II( JN)
E n dn o t e s
1. SeeFelson(1991),FelsonandFelson(1993)'and Logan(1991)for a discussionof blameanalysis. 2. For more detaileddiscussionsof concepts,see Chafetz(1978:45-61),Hage(1972:9-85)' Kaplan (1964'34-80), Mullins (197l:7-18), Reynolds (I97I), and Stinchcombe(1968,1973)' how sociologicalexpla3. Turner (1980)discussed nation and theorizingcan be conceptualizedas translation. 4. Classificationsare discussedin Chafetz(1978: 63-73) andHage(1972). 5. Introductions to alternativetheoreticalframe- & works and socialtheoriesare providedin Craib (1984), Phillips (1985:44-59),and Skidmore
(re7e). 6. An introductionto functionalexplanationcanbe found in Chafea (I97 8:22-25).
TR3
Ethicsin SocialResearch
Introduction Why Be Ethical? ScientificMisconduct Unethicalbut Legal Power Relations Ethical lssues Involving Research Participants Originsof ResearchParticipantProtection PhysicalHarm,Psychological Abuse,and LegalJeopardy Other Harmto Participants Deception InformedConsent SpecialPopulations and New Inequalities Privacy,Anonymity,and Confidentiality MandatedProtectionsof Research Participants Ethics and the Scientific Community Ethics and the Sponsors of Research Whistle-Blowing Arrivingat Particular Findings Suppressing Findings Concealing the TrueSponsor Politics of Research Value-Free and Objective Research Conclusion
48
PART ONE /
FO UNDATI O NS
INTRODUCTION Ethicsincludethe concerns,dilemmas'and conflicts that ariseover the proper way to conduct research.Ethicshelp to definewhat is or is not legitimateto do, or what "moral" researchprocedureinvolves.This is not as simple asit may appear,becausethere are few ethical absolutes and only agreed-uponbroad principles.These principlesrequirejudgment to apply and some may conflict with othersin practice.Many ethrcalissuesaskyouto balancetwo values:the pursuit of knowledgeand the rights of research participantsor of othersin society.Socialrebalancepotential benefits-such as searchers advancingthe understandingof sociallife, improving decisionmaking, or helping research participants-against potential 6s515-5u6}ra5 privacy,or democralossof dignity,self-esteem, tic freedoms. confront many ethical Socialresearchers how to act. They must decide dilemmas and professional obligationto be havea moral and participants areunaware ifresearch ethical,even about ethics. ofor unconcerned Many areasof professionalpracticehave ethicalstandards(e.g.,journalists,policedepartments,businesscorporations,etc.),but the ethical standardsfor doing socialresearchareoften stricter.To do professionalsocialresearch,you must both know the proper researchtechniques (e.g.,sampling)and be sensitiveto ethicalconcerns.This is not alwayseasy.For centuries, moral, legal,and political philosophersdebated regularlyface. the issuesresearchers It is difficult to appreciatefully the ethical until you dilemmasexperiencedby researchers until waiting actuallybegin to do research,but preto need You the middle of a studyis too late. ethical pareyourselfaheadof time and consider concernsasyou designa study so that you can build sound ethicalpracticesinto a study'sdesign.In addition,by developingsensitivityto ethicalissues,you will be alertto potentialethicalconcernsthat canariseasyou makedecisions rthile conducting a study.Also, an ethicalaware-
nesswill help you better understandthe overall process. research Ethicsbeginand end with you, the individual socialresearcher.A strong personalmoral is the bestdefenseagainst codeby the researcher unethicalbehavior.Before,during, and after hasopportuniconductinga study,a researcher ties to, andshould,reflecton the ethicsof researchactionsand consulthis or her conscience Ultimately, ethicalresearchdependson the integrity of an individual researcher.
WHY BE ETHICAL?
Given that most peoplewho conduct socialresearchare genuinelyconcernedabout others, you might ask,Why would any researcherever act in an ethicallyirresponsiblemanner?Most unethicalbehavioris due to a lack of awarenes to take ethical and pressureson researchers facepressuresto build a shortcuts.Researchers career,publish new findings, advanceknowledge,gain prestige,impressfamily and friends, hold on to a job, and so forth. Ethicalresearch will take longerto complete,costmore money' be more complicated,and be lesslikely to produceunambiguousresults.Plus,therearemany opportunitiesin researchto act unethically'the odds of gettingcaughtare small,and written ethicalstandardsarein the form ofvague,loose principles. The ethicalresearchergetsfew rewardsand wins no praise.The unethical researcher'if caught,facespublichumiliation, a ruinedcareer and possiblelegalaction. The best preparation for ethicalbehavioris to internalizea sensitivity to ethical concerns,to adopt a seriousprofessionalrole, and to interact regularlywith other Moreover,the scientificcommunity researchers. demandsethicalbehaviorwithout exceptions. Scientific Misconduct
that fund Theresearchcommunity and agencies researchopposea type of unethicalbehavio
CHAPT E R3 , / E T H I C SI N S O C I A L R E S E A R C H
calledscientificmisconduct;it includesresearch fraud and plagiarism.Scientificmisconductoc_ curs when a researcherfalsifiesor distorts the dataor the methodsof datacollection,or plagia_ rizesthe work of others.It alsoincludessienifi_ cant, unjustified departuresfrom the genlraily acceptedscientificpracticesfor doing and re_ porting on research. Research fraud occurswhen a researcherfakesor inventsdatathat he or she did not really collect, or fails to honestly and firllyreport howhe or sheconducteda study.Al_ thoughrare,it is considereda veryseriousviolation. The most famouscaseof researchfraud was that of Sir Cyril Burt, the father of British educationalpsychology.Burt died in I97I asan esteemedresearcherwho was famous for his studieswith twins that showeda geneticbasisof intelligence.In 1976,it was discoveredthat he had falsifieddata and the namesof coauthors. Unfortunately,the scientificcommunity had beenmisledfor nearly30years.More recently,a socialpsychologistwasdiscoveredto havefabri_ cateddata for severalexperimentson sexbias conductedat Harvard Universityin the tg9Os. Plagiarismoccurswhen a..raur.ir., ..steals,' the ideasor writings of anotheror usesthem with_ out citing the source.plagiarismalso includes stealingthe work of anotherresearcher, an assis_ tant, or a student,and misrepresentingit as one'sown. Theseareseriousbreachesof ethical standards.l Unethicalbut Legal Behaviormay be unethicalbut legal (i.e.,not breakany law). A plagiarismcaseillustratesthe distinction betweenlegal and ethicalbehavior. The AmericanSociologicalAssociationdocu_ mentedthat a 1988book without any footnotes by a deanfrom EasternNew Mexico University containedlarge sectionsof a 197gdissertation that a sociologyprofessorat Tufts University y9t.. Copyingthe dissertarionwasnot illegal;it did not violatecopyrightlaw becausethe sJcioogist'sdissertationdid not havea copyrightfiled with the U.S. government.Neverthii.ri it *u,
FIcURE 3.1
TypoloryofLegaland MoralActionsin Social Research
49 r
ETHICAL LEGAL
Yes
No
Yes
Moral and Legal
Legalbut lmmoral
No
lllegal but Moral
lmmoral and lllegal
clearly unethical according to standards ofpro_ fessional behavior.2 (SeeFigure 3.1 for relations between legal and moral actions.)
POWER RELATIONS A professionalresearcherand the researchpar_ ticipantsor employee-assistants are in a rela_ tionship of unequal power and trust. An experimenter,surveydirector,or researchinves_ tigator haspower over participantsand assis_ tants,and in turn, theytrust his or herjudgment and authority. The researcher,scredeitials, training, professionalrole, and the placeofsci_ encein modernsocietylegitimatethepowerand make it into a form of expert authoiity. Some ethical issuesinvolve an abuseof power and trust. A researcher's authority to conductsocial researchand to earn the trust of othersis ac_ companiedalwaysby an unyielding ethical re_ sponsibilityto guide,protect,and overseethe interestsof the peoplebeingstudied. When looking for ethical guidance,re_ searchers arenot alone.Theycanturn to a num_ ber of resources: professionalcolleagues, ethical advisorycommittees,institutionalreviewboards or human subjectscommitteesat a collegeor in_ stitution (discussedlater), codesof ethicsbv professionalassociations (discussed later in this chapter),and writings on ethicsin research.The largerresearchcommunity firmly supportsand upholds ethicalbehavior,evenif an-individual
PART ONE
/
F OUNDAT ION5
others in Nazi Germany,and similar "medical experiments"to test biological weaponsby |apanin the 1940s.In theseexperiments,terrible tortures were committed. For example,people wereplacedin freezingwaterto seehow long it ETHICAL ISSUESINVOLVING took them to die,peoplewerepurposelystarved RESEARCHPARTICIPANTS to death,peoplewereintentionallyinfectedwith and limbs were severedfrom Have you everbeen a participant in a research horrible diseases, onto others.3 childrenand transplanted study?If so,how wereyou treated?More attenof Suchhuman rightsviolationsdid not occur tion is focusedon the possiblenegativeefflects only long ago.In a famouscaseof unethicalreresearchon thosebeing studiedthan any other search,the TuskegeeSlphilis Study,alsoknown ethical issue,beginningwith concernsabout biomedicalresearch.Acting ethicallyrequires asBadBlood,thePresidentof the United States! balancethe valueofadvancing admitted wrongdoing and formally apologized that a researcher knowledgeagainstthe valueof noninterference in 1997to the participant-victims. Until the reportcauseda scanin the lives of others. Either extreme causes 1970s,when a newspaper problems.Giving researchparticipantsabsolute dal to erupt, the U.S. Public Health Service sponsoreda study in which poor, uneducated rights of noninterferencecould makeempirical abAfrican Americanmen in Alabamasufferedand researchimpossible,but giving researchers died of untreatedsyphilis,while researcher solute rights of inquiry could nullifr Participants'basichuman rights. The moral question studiedthe severephysicaldisabilitiesthat apjustified pear in advancedstagesofthe disease.The unbecomes:When, if ever,areresearchers ethicalstudybeganin l929,beforepenicillinwas in riskingphysicalharm or injury to thosebeing or availableto treat the disease,but it continued great them embarrassment causing studied, long aftertreatmentwasavailable.Despitetheir inconvenience,violating their privary, or frightunethical treatment of the people, the reeningthem? wereableto publishtheir resultsfor 40 The law and codesof ethicsrecognizesome searchers study ended in 1972,but a formal or The years. cause unnecessarF prohibitions: Never clear apologytook another25 Yeats." irreversibleharm to subjects;secureprior volUnfortunately,the BadBlood scandalis not untaryconsentwhenpossible;and neverunnecDuring the Cold War era,the U.S.govharmfirl unique. release or degrade, humiliate, essarily periodicallycompromisedethicalrewas that ernment individuals about specific information for military and political goals. principles words, search purposes. In other research for collected you shouldalwaysshowrespectfor the research In 1995,reports revealedthat the goYernment authorizedinjectingunknowingpeoplewith raparticipant. Theseare minimal standardsand (e.g., does dioactivematerialin the late 1940s.In the 1950s' What interpretation to are subject governmentwarned EastmanKodak and the situation?)' mean in a specific unnecessary other film manufacturersabout nuclearfallout from atomic teststo preventfoggedfilm, but it Origins of ResearchPafticipant did not warn nearbycitizensof health hazards Protection in the 1960s,the U.S. army gaveunsuspecting soldiersLSD (a hallucinogenicdrug),causingseparticiConcernover the treatmentof research widely recogrious trauma. Today,researchers gross violapants aroseafter the revelationof fundamental of two violations to be nize these of name tions of basic human rights in the harm and obphysical Avoid principles: ethical were science.The most notorious violations consent.s informed tain and "medical experiments"conductedon Iews
is ultimatelyresponsibleto do what is researcher ethicalin specificsituations.
CHAPTER3 , / E T H I C SI N S O C I A L R E S E A R C H
PhysicalHarm,Psychological Abuse, and Legalfeopardy Socialresearchcan harm a researchparticipant in severalways:physical,psychological, andlegal harm, aswell asharm to a person'scareer)reputation, or income. Different typesof harm are morelikely in othertypesof research(e.g.,in experimentsversus field research).It is a researcher'sresponsibilityto be awareof all types ofpotential harm and to takespecificactionsto minimize the risk to participantsat all times. PhysicalHarm. Physicalharm is rare.Evenin biomedicalresearch, wherethe interventioninto a person'slife is much greater,3 to 5 percentof studiesinvolved any person who sufferedany harm.6A straightforwardethicalprincipleis that researchers should never causephysicalharm. An ethicalresearcher anticipatesrisksbeforebeginning a study,including basicsafetyconcerns (e.g.,safebuildings,furniture, and equipment). This meansthat he or shescreensout high-risk subjects(thosewith heart conditions,mental breakdown,seizures,etc.) if great stressis involvedand anticipatespossiblesourcesofinjury or physicalattackson researchparticipantsor assistants. The researcheracceptsmoral and legal responsibilityfor injury due to participation in researchandterminatesa projectimmediately if he or she can no longer fully guaranteethe physicalsafetyof the people involved (seethe Zimbardostudyin Box 3.1). PsychologicalAbuse, Stress,or Lossof SelfEsteem, The risk of physicalharm is rare,but socialresearchers can place peoplein highly stressful,embarrassing,anxiety-producing,or unpleasantsituations.Researchers want to learn about people'sresponses in real-life,high-anxiety-producing situations,so they might place peoplein realisticsituationsof psychological discomfort or stress.Is it unethicalto causediscomfort? The ethics of the famous Milgram obedience studyarestill debated(seeBox 3.1). Some say that the precautionstaken and the knowledgegainedoutweighedthe stressandpo-
5l
tential psychologicalharm that researchparticipants experienced.Others believethat the extreme stressand the risk of permanentharm weretoo great.Suchan experimentcould not be conductedtodaybecauseof heightenedsensitivity to the ethicalissuesinvolved. Socialresearchers havecreatedhigh levelsof anxietyor discomfort.They haveexposedparticipantsto gruesomephotos; fuk"ly told male studentsthat they havestrongfemininepersonality traits; falselytold studentsthat they have failed; createda situation of high fear (e.g., smokeenteringa room in which the door is locked); asked participantsto harm others; placedpeoplein situationswheretheyfacesocial pressureto denytheir convictions;and had participantslie, cheat,or steal.TResearchers who study helpingbehavioroften placeparticipants in emergencysituationsto seewhetherthey will lend assistance. For example,Piliavin and associates(1969)studiedhelpingbehaviorin subways by faking someone'scollapseonto the floor. In the field experiment,the riders in the subwaycarwereunawareof the experimentand did not volunteerto participatein it. The only researchers who might evenconsiderconductinga studythat purposelyinduces great stressor anxiety in researchparticipants arevery experiencedand take all necessary precautionsbeforeinducing anxietyor discomfort. The researchers shouldconsultwith otherswho have conducted similar studies and mental healthprofessionals astheyplan the studn They should screenout high-riskpopulations(e.g., thosewith emotionalproblemsor weakhearts), and arrangefor emergenryinterventions or termination of the researchif dangeroussituations arise.Theymust alwaysobtainwritten informed consent(to be discussed)beforethe research and debrief the people immediatelyafterward (i.e.,explainany deceptionand what actually happenedin the study). Researchers should nevercreateunnecessary stress(i.e.,beyondthe minimal amountneededto createthe desiredeffect) or stressthat lacksa very clear,legitimate researchpurpose. Knowing what "minimal
PART ONE
/
F OUNDA IION5
study(Milgram,1963' StanleyMilgram'sobedience 1 965, 1 974) attemptedto discoverhow the horrorsof the Holocaustunderthe Naziscouldhaveocthe strengthof socialpressure curredby examining to obey authority.After signing"informedconsent in riggedrandomseforms,"subjectswereassigned, "teacher" wasthe while a confederate lection,to be a "pupil."The teacherwasto test the pupil'smemory of word lists and increasethe electricshocklevelif The pupilwaslocatedin a the pupilmademistakes. nearbyroom,so the teachercouldhearbut not see wasclearlylabeled the pupil.The shockapparatus voltage.As the pupil mademistakes with increasing sheor he alsomade andthe teacherturnedswitches, present pain. researcherwas The noisesasifin severe and madecommentssuchas "You must go on" to the teacher.Milgramreported,"subjectswere observedto sweat,tremble,stutter, bite their lips, into their flesh'These groanand dig theirfingernails were characteristicrather than exceptionalre(Milgram sponsesto the experiment" , 1963:375). of subjectswhowouldshockto danThe percentage higherthan expected. gerouslevelswasdramatically arose over the useof deceptionand Ethicalconcerns the extreme emotional stress experiencedby subjects. 1975) tea' (Humphreys, In LaudHumphreys's roomtradestudy(a study of malehomosexualencountersin publicrestrooms),about 100 menwere observedengagingin sexualactsas Humphreyspretended to be a "watchqueen"(a voyeur and lookout). Subjectswerefollowedto their cars,and their licensenumbersweresecretlyrecorded.Namesand wereobtainedfrom policeregisterswhen addresses One year Humphreysposedas a marketresearcher.
later,in disguise,Humphreysuseda deceptivestory about a health surveyto interviewthe subjectsin their homes.Humphreyswascarefulto keepnames in safetydepositboxes,and identifierswith subject advancedknowlnameswereburned.He significantly who frequent"tearooms"and edgeof homosexuals overturnedpreviousfalsebeliefsabout them.There has beencontroversyover the study:The subjects deceptionwasused;andthe names neverconsented; could havebeenusedto blackmailsubjects,to end prosecution. or to initiatecriminal marriages, ln the Zimbardoprison experiment(Zimbardo' 1972, 1973; Zimbardoet al.,1973, 1974), nale groups: studentsweredividedinto two role-playing volunguardsand prisoners.Beforethe experiment, teer studentsweregivenpersonalitytests,and only Volunteers thosein the "normal"rangewerechosen. weretold that signedup for two weeks,andprisoners and would have they would be undersurveillance no physical that but rights suspended, civil some prisonin the baseln a simulated abusewasallowed. ment of a StanfordUniversitybuilding,prisoners in standarduniforms (dressed weredeindividualized and calledonly by their numbers)and guardswere and reflective militarized(with uniforms,nightsticks, Guardsweretold to maintaina reasonsunglasses). abledegreeof order and served8-hour shifts,while prisonerswere lockedup 24 hoursper day. Unexpectedly,the volunteersbecametoo caughtup in becamepassiveand disorgatheir roles.Prisoners arbitrary,and nized,whileguardsbecameaggressive, By the sixthday,Zimbardocalledoff dehumanizing. the experimentfor ethicalreasons.The risk of perharm, harm,andevenphysical manentpsychological wastoo great.
amount" means comeswith experience.It is best to begin with too little stress,risking a finding of no effect, than to create too much. It is always wise to work in collaboration with other researcherswhen the risk to participants is high, becausethe involvement of severalethically sen-
sitive researchersreducesthe chancesof making an ethical misjudgment. Researchthat induces great stressand anxiety in participants also carries the danger that experimenters will develop a callous or manipulative attitude toward others. Researchers
CHAPT ER 3
havereportedfeelingguilt and regretafterconducting experimentsthat causedpsychological harm to people.Experimentsthat placesubjectsin anxiety-producing situationsmay produce significantpersonaldiscomfort for the ethicalresearcher. LegalHarm. A researcheris responsiblefor protectingresearchparticipantsfrom increased risk of arrest.If participationin researchincreasesthe risk of arrest,few individualswill trust researchers or be willing to participatein future research.Potentiallegalharm is one criticismof Humphreys's1975tearoomtradestudy (seeBox3.1). A related ethical issueariseswhen a researcherlearnsof illegalactivitywhen collecting data.A researcher must weighthe valueof protecting the researcher-subject relationshipand the benefitsto future researchers againstpotential seriousharm to innocent people.The researcherbearsthe cost of his or her judgment. For example,in his field researchon police,Van Maanen(1982:114-115) reportedseeingpolice beatpeopleand witnessingillegalactsand irregular procedures,but said,"On and following thesetroublesomeincidentsI followedpolice custom:I kept my mouth shut." Field researchersin particular can face difficult ethicaldecisions.For example,when studyinga mentalinstitution,Taylor(19S7)discoveredthe mistreatmentand abuseof inmates by the staff.He had two choices:Abandon the studyand callfor an immediateinvestigation,or keepquiet and continuewith the study for several months, publicizethe findings afterwards, and then becomean advocateto end the abuse. After weighingthe situation,he followedthe latter courseand is now an activistfor the rightsof mentalinstitution inmates. In some studies,observing illegal behavior may be central to the researchproject. If a researcher covertly observes and records illegal behavior,thensuppliestheinformationtolaw-enforcementauthorities,heorsheisviolatingethical standards regarding researchparticipants and is
,/ E TH IC S IN SOC IA L R E S E A R C H
53
undermining future social research.At the same time, a researcherwho fails to report illegalbehavior is indirectly permitting criminal behavior. He or shecouldbe chargedasan accessorytoa crime. Cooperation with law-enforcement offi cialsraises the question, Is the researchera professionalscientist who protects research participants in the processof seekingknowledge, or a free-lanceundercover informant who is really working for the police trying to "catch" criminals? Other Harm to Participants Researchparticipants may face other types of harm. For example, a survey interview may create anxiety and discomfort if it askspeople to recall unpleasant or traumatic events. An ethical researchermust be sensitiveto any harm to participants, consider precautions, and weigh potential harm against potential benefits. Another type of harm is a negative impact on the careers,reputations, or incomes of research participants. For example, a researcher conducts a survey of employees and concludes that the supervisor's performance is poor. As a consequence,the supervisorlosesher job. Or, a researcherstudies homelesspeople living on the street. The findings show that many engage in petty illegal acts to get food. As a consequence,a city government "cracks down" on the petty illegal acts and the homeless people can no longer eat. What is the researcher'sresponsibility? The ethical researcherconsidersthe consecuencesof research for those being studied. The general goal is not to cause any harm simply because someone was a researchparticipant. However, there is no set answer to such questions.A researchermust evaluateeachcase,iveighpotential harm against potential benefits, and bear the responsibility for the decision. h-_^_r:^uecePtlon Has anyone ever told you a half-truth or lie to get you to do something? How did you feel about it? Social researchersfollow the ethical principle of wluntary consent:Never force any-
54
P A Rr oNE / F o u N D A T ro N s
one to participate in research,and do not lie to anyone unless it is necessaryand the only way to accomplish a legitimate research purpose. The people who participate in social researchshould explicitly agree to participate. A person's right not to participate can be a critical issue whenever the researcherusesdeception, disguisesthe research.or usescovert researchmethods. Social researcherssometimes deceive or lie to participants in field and experimental research. A researcher might misrepresent his or her actions or true intentions for legitimate methodological reasons.For example, if participants knewthe true purpose, theywould modifr their behavior, making it impossible to learn of their real behavior. Another situation occurs when accessto a researchsite would be impossible if the researchertold the truth. Deception is never preferable if the researchercan accomplish the same thing without using deception. Experimental researchersoften deceivesubjects to prevent them from learning the hypothesisbeing tested and to reduce "reactive effects" (seeChapter 8). Deception is acceptableonly if a researcher can show that it has a clear, specific methodological purpose, and even then, the researcher should use it only to the minimal degree necessary.Researcherswho use deception should always obtain informed consent, never misrepresent risks, and always explain the actual conditions to participants afterwards.You might ask, How can a researcherobtain prior informed consent and still use deception? He or she can describe the basic procedures involved and conceal only specific information about hlpotheses being tested. Sometimes field researchersuse covert observation to gain entry to field researchsettings. In studies of cults, small extremist political sects, illegal or deviant behavior, or behavior in a large public area, it may be impossible to conduct research if a researcher announces and discloses her or his true purpose. Ifa covert stance is not essential,a researchershould not use it. If he or she does not know whether covert accessis necessary then a strategy of gradual disclosure may
be best.When in doubt,it is bestto err in the directionof disclosingone'strue identity and purpose.Covertresearchremainscontroversial,and feelthat all covertresearchis many researchers unethical.Eventhosewho acceptcovertresearch asethicalin certainsituationssaythat it should be usedonlywhen overt observationis impossible. Wheneverpossible,the researchershould inform participantsof the observationimmediately afterwardsand give them an opportunity to expressconcerns. Deceptionand covertresearchmay increase mistrustand qfnicismaswell asdiminish public respectfor socialresearch.Misrepresentationin field researchis analogousto being an undercoveragentor governmentinformer in nondemocratic societies.The use of deceptionhas a long-term negativeeffect.It increasesdistrust among peoplewho are frequentlystudied and makesdoing socialresearchmore difficult in the long term. lnformed Consent A fundamentalethicalprinciple of socialresearchis: Never coerceanyoneinto participating; participationmustbevoluntary at all times. Permissionaloneis not enough;peopleneedto know what they arebeingaskedto participatein sothat theycanmakean informeddecision.Participants can becomeawareof their rights and what theyaregettinginvolvedin whenthey read and sign a statementgiwng informedconsentan agreementby participants stating they are willing to be in a study and they know something aboutwhat the researchprocedurewill involve. Governmentsvary in the requirement for informed consent.The U.S.federalgovernment doesnot require informed consentin all researchinvolving human subjects.Nevertheless, shouldgetwritten informed consent researchers unlesstherearegoodreasonsfor not obtainingit (e.g.,covertfield research, useofsecondarydata, etc.) asjudgedby an institutional reviewboard (IRB) (seethe laterdiscussion of IRBs).
CHAPT ER 3
Informed consent statements provide spe_ cific information (seeBox 3.2). A generalstaie_ ment about the kinds ofprocedures or questions involved and the uses of the data are sufficient for informed consent. Studies suggestthat par_ ticipants who receive a full informed conienr statement do not respond differently from those who do not. If anlthing, people who refused to sign such a statement were more likely to guess or answer"no response"to questions. It is unethical to coerce people to partici_ pate,,including offering them special benefits that they cannot otherwise attain. For example, it is unethical for a commanding officer to orider a soldier to participate in a stud|, for a professor to require a student to be a research zubiect in order to passa course, or for an employer to ex_ pect an employee to complete u ,urr.y'u, u .or_
Informedconsentstatementscontainthe following: 1. A brief descriptionof the purposeand proce_ dureofthe research, includingthe expecteddu_ rationofthe study 2. A statementof any risksor discomfortassoci_ atedwith participation 3. A guaranteeof anonymityand the confidential_ ity of records 4. The identificationof the researcherand of whereto receiveinformationabout subiects' rightsor questions aboutthe study 5. A statementthat participation is completelyvol_ untaryandcanbe terminated at anvtimewith_ out penalty
,/ E TH IC S IN SOC IA L R E S E A R C H
55
dition of continued employment. It is unethical even if someoneother than the researcher(e.g., an employer) coercespeople (e.g.,employees)to participate in research. Full disclosurewith the researcher,sidentifi_ cation helps to protect research participants againstfraudulent researchand to protectiegit_ imate researchers.Informed consentlessensthe chance that a con artist in the guise of a re_ searcherwill defraud or abusepeople. It also re_ duces the chance that someone will use a boeus researcheridentity to market products or obtlin personal information on people for unethical purposes. Legally, a signed informed consent state_ ment is optional for most survey, field, and sec_ ondary data research,but it is often mandatrferiments was being videotaped.During the experiment' For example,during an experimenton reasontne viaeJequipmentfailedto work for one subing ability, subjectsbecomebored and sleepy iect. If it had iailed to work for more than one lower.Another exampleis *"d, u, a result,'score .subiector had worked for only part of the sesan experimenton the stylesof children'splaybesion, the experiment would have had instrutweengrades1 and 6. Playstylesare affectedby pioblems.(By the way,subjectstook physicil, emotional' and maturation changes -"rriutiott to deliverbad newsonly if they thought iong.. ',hu,o..o, asthe childrengrow older,insteadof the| were doing so publicly-that is' being or in addition to the effectsof a treatment'DevideotaPed') signswith a pretestand control group help re,elarchersdelermine whether maturation or Mortatity. Mortality, or attritiort, ariseswhen history effectsarepresent,becauseboth experithe ,o-. ,rr61..tsdo not continue throughout mental and control grouPswill show similar means experiment.Althoughthe wotd mortality changesovertime' meanthat subjects death,it doesnot necessarily havedied. Ifa subsetofsubjectsleavespartway itself Testing. Sometimes,the pretestme-asure through an experiment, a researchercannot affectsin experiment. This testingeffectthteatk ro*"*hether ihe resultswould havebeendifmorethan the treatensinternalvaliditybecause ferenthad the subjectsstayed'For example'a rement aloneaffectsthe dependentvariable'The searcherbeginsa weight-lossprogram with 50 Solomon four-group designhelps a researcher subjects.Atlhe end of the program' 30 remain' detecttestingeffects.For example,a researcher eacirof whom lost 5 poundswith no sideeffects eivesstudentsan examinationon the first dayof The 20 who left could havedifferedfrom the 30 Ilass.The courseis the treatment'He or shetests who stayed,changingthe results'-Maybethe learningby givingthe sameexamon the lastday and oronru- was effeclivefor those who left' of class.If subjectsrememberthe pretestquesp"tin"f *itnat"w after losing 25 pounds' 9: tions andthis affectswhat theylearned(i'e'' paid hapsthe programmadesubjectssickand forced attentionto) or how theyansweredquestionson shouldnotice and rethem to [uit' Researchers the posttest,a testingeffectis present'Iftesting
c HApr ER 8 , / E X p E R T M E N T ARLE S E A R c H
port the number of subjects in each group during pretests and posttests to detect this threat to internal validity. Statistical Regression, Statistical regressionis not easy to grasp intuitively. It is a problem of extreme values or a tendency for random errors to move group results toward the average.It can occur in two ways. One situation arises when subjects are unusual with regard to the dependent variable. Becausethey begin as unusual or extreme, subjects are unlikely to respond further in the same direction. For example, a researcherwants to see whether violent films make people act violently. He or she chooses a group of violent criminals from a high-securityprison, givesthem a pretest, shows violent films, then administers a posttest. To the researcher's shock, the prisoners are slightlyless violent after the film, whereasa control group of prisoners who did not seethe film are slightly more violent than before. Because the violent criminals began at an extreme, it is unlikely that a treatment could make them more violent; by random chance alone, they appear lessextreme when measured a second timi.2 . A second situation involves a problem with the measurement instrument. If many research participants score very high (at the ceiling) or very low (at the floor) on a variable, random chance alone will produce a changebetween the pretest and the posttest. For example, a researchergives 80 subjects a test, and 75 getperfect scores.He or she then gives a treatmenl to raise scores. Becauseso many subjects already had perfect scores,random errors will reduce the group averagebecausethose who got perfect scores can randomly move in only one direction-to get some answerswrong. An examination of scores on pretests will help researchers detect this threat to internal validity. Diffusion of Treatment or Contamination. Dffision of treatment is the threat that research participants in different groups will communicate with each other and learn about the other's
Zls
treatment. Researchers avoid it by isolating groupsor havingsubjectspromisenot to reveal anythingto otherswho will becomesubjects. For example,subjectsparticipatein a day-long experimenton a new way to memorizewords. During a break,treatment-groupsubjectstell thosein the control group aboutthe newwayto memorize,which control-groupsubjectsthen use.A researcherneedsoutsideinformation, suchaspostexperimentinterviews,with subiects to detectthis threat. ExperimenterExpectancy. Although it is not alwaysconsidereda traditional internal validity problem, the experimenter'sbehavior,too, can threatencausallogic.3A researcher maythreaten internal validiry not by purposefullyunethical behavior but by indirectly communicating experimenter expectancy to subjects.Researchers may be highly committedto the hypothesisand indirectly communicatedesiredfindings to the subjects.For example,a researcherstudiesthe effectsof memorizationtraining on student learningability, and also seesthe gradetranscriptsof subjects.The researcherbelievesthat studentswith highergradestend to do better at the training and will learn more. Through eye contact,tone ofvoice,pauses,and othernonver_ bal communication, the researcherunconsciouslytrains the studentswith higher grades more intensely;the researcher'snonverbalbe_ havior is the oppositefor studentswith lower grades. Here is a way to detectexperimenterex_ pectancy.A researcherhires assistantsand teachesthem experimentaltechniques.The as_ sistantstrain subjectsandtesttheir learningability. The researchergives the assistantsfake transcriptsand recordsshowingthat subjectsin onegroup arehonor studentsand the othersare failing,althoughin factthe subjectsareidentical. Experimenterexpectancyis presentif the fake honor students,asa group,do much betterthan the fakefailing students. The double-blindexperimenris designedto control researcher expectancy. In it, peoplewho
2 16
R E S E A R cH P A RTT w o / c o N D U c rl N c QU A N TITA TIvE
havedirect contactwith subjectsdo not know the detailsof the hypothesisor the treatment.It is doubleblind becauseboth the subjectsand thosein contactwith them areblind to detailsof the experiment(seeFigure8.3). For example,a wantsto seeif a new drug is effective. researcher Using pills of three colors-green, yellow, and pink-the researcherputs the new drug in the yellowpill, puts an old &ug in the pink one,and makesthe greenpill aplacebo-afalsetreatment that appearsto be real (e.g.,a sugarpill without who givethe pills anyphysicaleffects).Assistants and recordthe effectsdo not know which color
contains the new drug. Only another person who doesnot dealwith subjectsdirectly knows which coloredpill containsthe drug and it is he or shewho examinesthe results. External Validity and Field Experiments
Evenif an experimentereliminatesall concerns about internal validity, externalvalidity remains a potentialproblem.Externalvalidityis the ability to generalizeexperimentalfindingsto eYents and settingsoutsidethe experimentitself. If a *
or Ordinary,. An lllustrationof Single-Blind, Double-BlindExperiments: and Double-BlindExPeriments
Flc U RE 8. 3
ExPeriment Single-Blind
o
qfrq o0
o
7N fr fr fififrfifrfifrfrfr o o o o
SubiectsWho Are Blindto TrueHypothesis
o
q
Experimenter
P q
Double-BlindExPeriment
.o
o e o o
q Bq fr qfrq fififififrflfrfrm oo
SubjectsWho Are Blindto TrueHypothesis
R ELS E A R cH c H Ap rE R8 ,/ E X pE R TME N TA
study lacks externalvalidity, its findings hold true only in experiments,making them useless to both basicand appliedscience.
217
person encounters people who contradict the stereotype,especiallyifthe others are respected. They used both a laboratory experiment (with a two-group, posttest-only design) and a field exReactivity. Researchparticipantsmight react periment. Past studies focused on out-group stereot)?es, but the authors wanted to examine differentlyin an experimentthan they would in real life becausethey know they are in a study; the hypothesis for an in-group, women. In the laboratory experiment, experimenters randomly this is calledreactivity.The Hawthorneffict is a specifickind of reactivity.4The name comes assignedfemale subjects to view either (1) a set photographs and biographies of 16 famous from a seriesof experimentsby Elton Mayo at the Hawthorne,Illinois, plant of Westinghouse women leadersor (2) photos and descriptions of Electricduring the 1920sand 1930s.Researchers 16 flowers. The experimenters used deception and told subjects the study was about testing modified many aspectsof working conditions (e.g.,lighting, time for breaks,etc.) and meamemory. The dependent variable was attitudes and beliefs about women and was measured sured productivity. They discoveredthat prowith a implicit Association Test (IAT). The reductivity roseaftereachmodification,no matter what it was.This curiousresultoccurredbecause sults showed that subjects associatedgendered the workersdid not respondto the treatment first names (e.g.,|ohn vs. Emily) with leadership but to the additional attention they received or follower traits (e.g., assertive and sympafrom beingpart of the experimentand knowing thetic). A high IAT score indicated that a subthat they were being watched.Later research ject viewed women more than men as having questionedwhetherthis occurred,but the name leadership more than supportive traits. The reis used for an effect from the attention of researchers also used a scale on beliefs about A relatedeffectis the effectof some- women. They found support for the hypothesis searchers. that exposure to famous women in leadership thing new,which maywearoffover time. positions increased IAT scores,compared to exposure to neutral information about flowers. Fielil Experiments. Sofar, this chapterhasfoThe field experiment had a pretest and a posttest cusedon experimentsconductedunderthe controlled conditionsof a laboratory.Experiments but no random assignment. Subjectswere feare'alsoconductedin real-life or field settings males who attended two colleges in the same wherea researcherhaslesscontrol over the extown. One was a coeducational college and the perimental conditions.The amount of control other had all female students. Subjects were recruited from first-year classesat the beginning varieson a continuum.At one end is the highly controlled laboratory experiment,which takes of the academic year and completed the IAT placein a specialized measure, the beliefs about women scale, and a settingor laboratory at the which takes general campus questionnaire. The experioppositeendis thefield experimenr, placein the'field'-in naturalsettingssuchasa menters documented that the all-female college subwaycar, a liquor store,or a public sidewalk. had more females in administrative and faculty leadership positions. Pretest IAT scores were Subjectsin field experimentsare usuallyunawarethat they are involved in an experiment very similar, with subjects from coeducational college having slightly lower scores.This helped and react in a natural way. For example,rethe experimenters to check for possible selection searchers havehad a confederatefakea heartattack on a subwaycar to seehow the bystanders bias. Subjectswere contacted one year later and asked to complete the same measuresas prereact.5 sented in the posttest. Experimenters watched Dasguptaand Asgari (2004)testedthe hypothesisthat stereotypical very carefully for experimental mortality since a beliefsweakenwhen
2 18
pA RTT w o ,/ c o N D U c rtN c QU A N Tl rA rl vER E S E A R cH
somestudentsstoppedattendingcollegeor did not completelater surveys.The IAT scoresfor subjectsat the coeducationalcollegedeclined (i.e.,theywerelesslikely to seefemalesashaving leadershiptraits), whereasthe IAT scoresfor subjectsat the all-female college greatly increased.In addition, the experimentersfound that the more femaleteachersa studenthad at eithercollege,the higherthe posttestIAT scores, the casefor math and sciandthis wasespecially encescourses.Thus,exposureto womenin leadership positions caused the IAT scores to increase,whereasthe absenceofsuch exposure' if anlthing, loweredthe scores. (2005)useda field Von Larr and colleagues experimentto test the well-knov,n contacthypothesisthat saysintergroup contact reduces racial-ethnic prejudice as people replacetheir stereotl?eswith personalexperience,although this happenssolong asthe contactinvolvespeople of equalstatuspursuingcommon goalsin a cooperativesettingand is approvedby authorities.In addition,informal contactin which people get to know about out-group membersas alsoreducesout-groupprejudice. acquaintances The experimenttook placeat UCLA, wherethe studentbody is very racially and ethnicallydiverse.Unlesstheypreselecta roommate,incoming studentsarerandomlyassignedroommates. About 20 percentof studentschoosea roommate and the rest are randomly assigned.The authorsmeasuredstudentbackgroundand attitudesamong nearly 3,800new incoming students using a panel design acrossfive time periods-before collegeentry (summer 1996) and during the spring of eachof the next four years(1997-2000)with surveys(20-minutetelephone interviews).The dependentvariablewas the students' racial-ethnic attitudes and included questions about roommates, other friends, interracial dating, multiculturalism, sgnbolic racism, and feelingsabout various racial-ethnic groups. Thesewere the experiment's pretestand multiple posttestmeasures. Experimenterswatchedvery carefullyfor exper-
imental mortality, sincesomestudentsstopped attendingcollege,leftcollegedormitories,or did not completethe later surveys.They testedthe hlpothesesthat studentswho wererandomlyassignedto live with an out-group member (the independentvariable)developedlessprejudicial attitudestoward membersof that out-group. They found that comparedto pretestrneasures' prejudicialattitudesdeclinedaspredictedby the contacthypothesiswith one exception.Apparently having an Asian American roommate worked in the opposite way and actually increasedprejudice,especiallyamong the White students. Experimentercontrol relatesto internaland externalvalidity. Laboratoryexper'imentstend to havegr eaterinternalvalidity but lower external validity; that is, they arelogicallytighter and Fieldexbettercontrolled,but lessgeneralizable. perimentstend to havegreaterexternalvalidity but lower internalvalidity;that is, they aremore but lesscontrolled.Quasi-experigeneralizable mental designsare common in field experiments.Table8.3 summarizesthreatsto internal andexternalvalidity.
TABTE 8.3
Maiorlnternaland External Validity Concerns
bias Selection History effect Maturation Testing Instrumentation mortality Experimental Statisticalregression Diffusionof treatment expectancy Experimenter
Hawthorneeffect
CHAPTER8 , / E X P E R I M E N T AR L ESEARCH
P R AC TICAL CO NS I DE RA T ION S Every research technique has informal tricks of the trade. These are pragmatic, commonsense ideasthat account for the difference between the successfulresearchprojects of an experienced researcher and the difficulties a novice researcherfaces.Three are discussedhere. Planning and Pilot-Tests
219
of the situation affectedtheir behavior.Finalln he or shecan explainthe importanceof not revealing the true nature of the experiment to other potentialparticipants.
RESULTSOF EXPERIMENTAL RESEARCH:MAKING COMPARISONS
Comparisonis the key to all research.By carefully examiningthe resultsof experimentalresearch,a researcher canlearn a greatdealabout threats to internal validity, and whether the treatmenthasan impact on the dependentvariable. For example,in the Bond and Anderson (1987)experimenton deliveringbad news,discussedearlier,it took an averageof 89.6and73.I secondsto deliverfavorableversus72.5or 747.2 secondsto deliverunfavorabletestscoresin privateor public settings,respectively. A comparison showsthat deliveringbad newsin public takesthe longest,whereasgood newstakesa bit longerin private. A more complexillustration of such comparisonsis shownin Figure8.4on the resultsof a seriesof fiveweight-lossexperimentsusingthe classicalexperimentaldesign.In the example, Instructions to Subjects the 30 researchparticipantsin the experimental Most experimentsinvolvegivinginstructionsto group at Enrique'sSlim Clinic lost an averageof subjectsto set the stage.A researchershould 50 pounds,whereasthe 30 in the control group word instructions carefully and follow a predid not losea singlepound. Only one person pared script so that all subjectshear the same dropped out during the experiment.Susan's thing. This ensuresreliability. The instructions ScientificDiet Planhad equallydramaticresults, are also important in creatinga realisticcover but 11peoplein her experimentalgroupdropped storywhendeceptionis used. out. This suggests a problem with experimental mortality. Peoplein the experimentalgroup at Carl's Calorie Counterslost 8 pounds, comPostexperiment Interview paredto 2 poundsfor the control group,but the At the end of an experiment,the researcher control group and the experimentalgroup beshould interview subjects,for three reasons. gan with an averageof 31 pounds differencein First,ifdeceptionwasused,the researcher needs weight. This suggestsa problem with selection to debrieftheresearchparticipants,tellingthem bias.Natalie'sNutrition Centerhad no experithe true purposeof the experimentand answer- mentalmortality or selectionbiasproblems,but ing questions.Second,he or shecan learnwhat those in the experimentalgroup lost no more the subjectsthought and how their definitions weight than those in the control group. It apAII social research requires planning, and most quantitative researchersuse pilot-tests. During the planning phase of experimental research, a researcherthinks of alternative explanations or threats to internal validity and how to avoid them. The researcher also develops a neat and well-organized system for recording data. In addition, he or she devotes serious effort to pilottesting any apparatus (e.g., computers, video cameras,tape recorders, etc.) that will be used in the treatment situation, and he or she must train and pilot-test confederates.After the pilot-tests, the researchershould interview the pilot subjects to uncoyer aspectsof the experiment that need refinement. t'
220
PART TWO /
FI c u RE 8 .4
CO NDUCTI NC Q UANTI TATI V ER E S E A R C H
weight-Loss comparisonsof Results,classicalExperimentalDesign, Experiments
Experimental ControlgrouP
Enrique's SlimClinic Posttest Pretest 14o(2e) 1e0(30) 18e(30) lse (30)
Experimental Controlgroup
Susan'sScientific Diet Plan Posttest Pretest 141(1e) 1e0(30) 18e(28) 1se(30)
Experimental ControlgrouP
Carl'sCalorie Counters Posttest Pretest 1s2(2e) 150(30) 18e(2e) 1el (2e)
pears that the treatment was not effective' Fauhne'sPoundsOffalso avoidedselectionbias and experimentalmortality problems.Peoplein her experimentalgroup lost 32 pounds,but-so that did thosein the control group'This suggests
a "natural"experimentis possibledue Occasionally, changesor a governmentintervenpolicy public to are able to measure,partlcltion, and researchers pate,andlearnfromit andconducta fieldexperiment with high extemalvatidity.This occurredin New OrUntilthe mid-1990s, lawson selling leans,Llousiana. liquorto underagecustomerswerebarelyenforced in New Orleans.lf caught,the offendingliquor retailer met privatelywith the liquor commissionand paida smallfine.Enforcingliquorlawswaslow priorso only threeenity for stateand localgovernment, forcementofficersmonitored5,000 alcoholoutlets
N atal i e' s N utri ti on C enter Posttest Pretest Experimental ControlgrouP
Experimental ControlgrouP
1e0(30) 1e2(2e)
188(2e) 190(28)
Pauline's PoundsOff Posttest Pretest 1s8(30) leo (30)
1e1(2e)
lse (28)
the maturation, history, or diffirsion of treatment effectsmay haveoccurred.Thus,the treatment at Enrique'sSlim Ctinic appearsto be the mosteflectiveone.SeeBox 8.2for a practicalapplicationof comparingexperimentalresults'
in the New Orleansarea. When public officials and plannedto shiftenforcementpriorities,Scribne.r tohen (200.1) examinedits impact'They had several peoplewho clearlylookedunder 1 8 yearsold illegally attempt to purchasealcoholicbeverages age)at of years 2l least (the law requiredbeingat Nobetween outlets liquor i 43 randomlyselected perThe 0)' (Time 1 996 vember1 995 andJanuary centagewho couldbuy liquorillegallywasthefretest the rate of illegalsales,the ,rorin. After assessing the policeissuedcitationsto 5'l variable, dependent ofthe salesoutlets,the primaryindependentvaiab
CHAPTER8 , / E X P E R I M E N T AR L ESEARCH
or treatment.Aboutthe sametime,governmentofficialsinitiateda mediacampaignurgingbetter law compliance.Thereweretwoposttestmeasures, firstin
221
Marchto April1 996 (Time1) andagainin November 1996 to January1997 (Tine 2), duringwhich the experimenters checkedthe 143 outlets.
DEPENDENT VARIABLE: PERCENTAGE WHO OBEYTHELAW
ir..l.i.t t&,1li a:,,n,:: Experimental (citation) Control (no citation) Total
6.7%
5 1%
29%
45
13j% 11.1%
3s %
17% 21%
98 143
The resultsallow us to comparerates of illegal sellingactivitybeforeand after citationsplusmedia campaign(ltretestand posttestmeasures)and to compareoutletsthat receivedcitations(experimental group)with thosethat did not receivecitationsand only hadmediaexposure(control group).We seethat the citationsand campaigndid not stop the illegal activity, but it had some effect. The impact was greateron outlets that experienceddirect punishment.In addition,by addinga laterfollow-up(Time 2), we see how the law-enforcement impactslowly decayedover time.As frequentlyhappensin a naturalexperiment, internalvalidityis threatened:First, the pretest measureshowsa differencein the two setsof outlets,with outletsthat receivedthe treatmentshowinghigherratesof illegalbehavior;this is potentialselection bias.Second,the mediacampaign occurredfor alloutlets,so the treatmentis reallya crtation plusthe mediacampaign. The authorsnoted that they hadintendedto comparethe NewOrleans areawith anotherareawith neitherthe medianor
A WO R D ON E T HI CS Ethical considerations are a significant issue in experimental researchbecauseexperimental researchis intrusive (i.e., it interferes). Treatments
40%
the citationcampaign, but were unableto do so. Sinceoutletsthat did not receivethe treatment(i.e., a citationfor lawviolation)probablylearnedaboutit from othersin the samebusiness, a formof diffusion of the treatmertcould be operating.Third, the researchers report that they beganwith I 55 outlets, but studiedonly 143 becauseI 2 outletswent out of business duringthe study.Theauthorsnotedthat none of the outlets that stoppedsellingalcohol closeddueto newlawenforcement, but ifthose outlets that receivedcitationshad more problemsand were more likelyto go out of business,it suggests experimental mortality.The experimentersdid not mentionany externaleventsin New Orleansthat happenedduringthe time of the study (e.g.,a publicizedeventsuchasunderagedrinkerdyingof alcohol poisoningfrom overdrinking).Researchers needto be awareof potentialexternaleventswhena study continuesfor a long time and considerpossible historyfficts.
may involve placingpeoplein contrived social settings and manipulating their feelings or behaviors.Dependentvariablesmay be what subjectssayor do. The amount and type of intrusion is limited by ethical standards.Re-
222
PART TWO /
CO NDUCTI NC Q UANTI TAT I V ER E S E A R C H
searchersmust be very carefrrlif they placeresearchparticipantsin physicaldangeror in embarrassingor anxiety-inducingsituations'They must painstakinglymonitor eventsand control what occurs. Deception is common in social experiments, but it involves misleading or lying to subjects.Suchdishonestyis not condonedunconditionally and is acceptableonly as the meansto achievea goalthat cannotbe achieved otherwise.Evenfor a worthy goal,deceptioncan be usedonly with restrictions.The amount and qpe of deceptionshouldnot go beyondwhat is minimally necessaryand researchparticipants shouldbe debriefed.
CONCLUSION In this chapter,you learnedabout random assignmentand the methodsof experimentalresearch.Randomassignmentis an effectiveway to createtwo (or more) groups that can be treated as equivalentand hencecompared.In general,experimentalresearchprovidesprecise and relatively unambiguous evidence for a causalrelationship.It follows the positivistapproach,producesquantitativeresultsthat canbe analyzedwithstatistics,and is often usedin eval(seeBox 8.2). uationresearch alsoexaminedthe parts of an chapter This how they can be combinedto and experiment producedifferentexperimentaldesigns.In addition to the classicalexperimentaldesign' you and quasi-experlearnedaboutpreexperimental learned how to express You also imentaldesigns. notation. design them using You learnedthat internal validity-the internal logical rigor of an experiment-is a key ideain experimentalresearch.Threatsto internal validity arepossiblealternativeexplanations to the treatment.You alsolearnedabout external validity and how field experimentsmaximize externalvalidity. The realstrengthof experimentalresearchis its control and logical rigor in establishingevi-
dencefor causality.In general,experimentstend to be easierto replicate,lessexpensive,and less time consumingthan the other techniques.Experimental researchalso has limitations. First, somequestionscannot be addressedusing experimentalmethodsbecausecontrol and experimental manipulation are impossible.Another limitation is that experimentsusuallytestoneor a few hypothesesat a time. This fragments to qrnthesize knowledgeand makesit necessary External reports. resultsacrossmany research because problem potential validity is another nonrandom onsmall many experimentsrely samplesof collegestudents.b You learnedhow a carefulexaminationand comparisonof resultscan alertyou to potential problemsin researchdesign.Finally'you saw somepracticaland ethicalconsiderationsin experiments. In the next chapters,you will examineother researchtechniques.The logic ofthe nonexperimentalmethodsdiffersfrom that of,the experiment. Experimentersfocus narrowly on a few They usuallyhaveone or twoindehypotheses. pendentvariables,a singledependentvariable,a few small groups of subjects'and an independent variable that the researcherinduces. By contrast, other social researcherstest many hypothesesat once. For example' survey researchersmeasurealarge number of independent and dependentvariablesand use a larger numberof randomlysampledsubjects.Their independentvariablesareusuallypreexistingconditions in researchparticipants.
Key Ter m s classical experimental design control group debrief deception demand characteristics design notation diffrrsion of treatment double-blind experiment
C H A P TE R 8 ,/ E X P E R IME N TA R ELS E A R C H equivalent time series experimental design experimental group factorial design field e4periment Hawthorne effect historyeffects interaction effect interrupted time series laboratory e4periment Latin square design maturation mortality one-shot casestudy placebo posttest preexperimental designs pretest quasi-experimentaldesigns random assignment reactivity selection bias Solomon four-group design
223
static group comparison treatment
Endnotes l. For additionaldiscussionsofthreats to internal validiry seeCook and Campbell(1979:51*68), Kercher(1992),Smith and Glass(1987),Spector (1981:24-27), and SulsandRosnow(19S8). 2. Thisexampleis borrowedfrom Mitchelland lol_ ley(1988:97). 3. Experimenterexpectancy is discussed in Aronson and Carlsmith(1968:66_7 0), Dooley( I 984:I 5l_ 153),andMitchelland)olley(1988:327_32il. 4. The Hawthorneeffectis describedin Roethlis_ bergerand Dickenson(1939),Frankeand Kaul (1978),and Lang(1992).Alsoseethe discussion in Cook and Campbell(1979:123_125) andDoo_ ley (1984:155-156). Gillespie(1988,l99t) dis_ cussedthe politicalcontextof the experiments. 5. SeePiliavinandassociates ( I 969). 6. SeeGraham( 1992)andSears ( 19g6).
NonreactiveResearchand SecondaryAnalYsis
lntroduction Nonreactive Measurement The Logicof NonreactiveResearch Observation or Unobtrusive Varietiesof Nonreactive and Documentation Recording Content AnalYsis What ls ContentAnalYsis? TopicsAppropriatefor ContentAnalysis and Coding Measurement Coding,ValiditY,and ReliabilitY Howto ConductContentAnalysisResearch lnferences Existing Statistics/Documents and Secondary Analysis AppropriateToPics SocialIndicators LocatingData Limitations lssuesof Inference and Theory Testing lnferencesfrom NonreactiveData EthicalConcerns Conclusion
224
CHAPTER9 , / NO NREACT I V ER E S E A R C H A N D S E C O N D A R YA N A L Y S I S
INT RO DUCT I ON Experiments and survey research are both reactiye;that is, the people being studied are aware of that fact. The techniques in this chapter addressa limitation of reactive measures.You will learn about four researchtechnicues that are nonreactite;that is, the people being studied are not aware that they are part ofa researchproject. Nonreactive techniques are largelybased on positivist principles but are also used by interpretive and critical researchers. The first technique we will consider is lessa distinct technique than a loose collection of inventive nonreactive measures.It is followed by content analysis,which builds on the fundamentals of quantitative researchdesign and is a welldeveloped researchtechnique. Existing statistics and secondary analysis,the last two techniques, refer to the collection of already existing information from government documents or previous surveys. Researchersexamine the existing data in new ways to addressnew questions. A1though the data may have been reactive when first collected, a researcher can address new questions without reactive effects.
!{'{@
N O NRE A CT I V E ME AS U R EME N T
Z2S
during both daytime and nighttime. Obsen.ers noted whether the driver was male or femalel whether the driver was alone or with passengers; whether other trafific was present; and whether the car came to a complete stop, a slow stop, or no stop. Later, we will contrast this type of observation to a slightly different type used in field researcn. Varieties of Nonreactive or Unobtrusive Observation Nonreactive measures are varied, and researchershave been creative in inventing indirect ways to measure social behavior (see Box 9.1). Becausethe measureshave little in common except being nonreactive, they are best learned through examples. Some are erosion measuregwhere selectivewear is used as a measure, and some are accretionmeasuregwhere the measuresare depositsof somethingleft behind.I Researchershave examined family oortraits in different historical erasto seehow sender relations within the family are reflected in seating patterns. Urban anthropologists have examined the contents of garbagedumps to learn about life-styles from what is thrown away (e.g.,liquor bottles indicate level of alcohol consumpti;n). Based on garbage, people underreport their
The Logic of Nonreactive Research Nonreactive measurement begins when a researchernotices something that indicates a variable of interest. The critical thing about nonreactive or unobtrusiyemeasures(i.e., measures that are not obtrusive or intrusive) is that the people being studied are not aware of it but leave evidence oftheir social behavior or actions "naturally." The observant researcher infers from the evidenceto behavior or attitudes without disrupting the people being studied. Unnoticed observation is also a type of nonreactive measure. For example, McKelvie and Schamer (1988) unobtrusively observedwhether drivers stopped at stop signs. They made observations
Fosterand colleagues (l 99S) examined the tomb_ stonesin I 0 cemeteries in an areaof lllinoisfor the periodfrom 'l 830 to 'l 989. Thevretrieved dataon birthand deathdatesandgenderfrom over2,000 of the 2,028 burials.The researchers learnedthe areadifferedfrom somenationaltrends.Thevfound that conceptions hadtwo peaks(springandwinter), femalesaged'l 0 to 64 hada higherdeathratethan males,andyoungerpeopledied in latesummerbut olderpeoplein latewinter.
226
PART TWO , / CO NDUCTI NC Q UANTI TATI V ER E S E A R C H
Iiquor consumptionby 40 to 60 percent(Rathje havestudied andMurphy, 1992:7l). Researchers the listenlnghabitsof driversby checkingwhat stationstheir radiosare tuned to when carsare repaired.They havemeasuredinterestin different exhibitsby noting worn tiles on the floor in differentparts of a museum.They havestudied differencesin graffiti in maleversusfemalehigh schoolrestroomsto show genderdifferencesin themes.Somehaveexaminedhigh schoolyearbooks to comparethe high schoolactivitiesof
PhysicalTraces Erosion: Wear suggestsgreateruse. toys at a children's examines Example: A researcher day care that were purchasedat the sametime' Worn-outtoys suggestgreaterinterestby the children. of physicalevidencesugAccretion; Accumulation gestsbehavior. the brandsof aluexamines E arptet A researcher minumbeveragecans in trash or recyclingbins in male and femaledormitories.This indicatesthe brandsand typesofbeveragesfavoredby eachsex' Archives Running Records: Regularly produced public recordsmayrevealmuch. examinesmarriagerecords Example: A researcher for the brideand groom'sages.Regionaldifferences suggest that the preferencefor males marrying femalesis greater in certainareasof the y*g", country. OtherRecords: lrregularor privaterecordscan reveala lot.
thosewho had psychologicalproblemsin latter Iife versusthosJwho did not. (AlsoseeBox 9'2') Recording and Documentation Creatingnonreactivemeasuresfollowsthe logic of quaniitativemeasurement.A researcherfirst a construct,then links the conconceptualizes structlo nonreactiveempiricalevidence,which is its measure.The operationaldefinition of the variableincludeshow the researchersystematicallynotesand recordsobservations'
findsthe numberof reamsof Example: A researcher .l by a collegedean'sofficefor 0 years paperpurchased whenstudentenrollmentwasstable'A sizableincrease paperworkhasincreased' suggeststhat bureaucratic Observation Appearance: How peopleapPearmayindiExtemal cate socialfactors. Example: A researcherwatchesstudents to see whetherthey are more likelyto weartheir school's colorsandsymbolsafterthe schoolteamwonor lost' CountBehaviors: Countinghow many peopledo somethingcanbe informative. countsthe numberof men Example: A researcher to and womenwho come a full stop and those who cometo a rollingstop at a stop sign'This suggests genderdifferencein drivingbehavior' TimeDuration: How long peopletake to do things mayindicatetheir attention. how longmenand measures Exinple: A researcher of a nudeman painting the womenpausein front of TimemaY woman' a nude of painting andin front of a or crossin same interest or indicateembarrassment sex. each bY sexnudity
CHAPTE R9 , / NO NREACTI VERESEAR C H A N D S E C O N D A R YA N A L Y S I S
Becausenonreactivemeasuresindicate a constructindirectly,the researcher needsto rule out reasonsfor the observationother than the constructof interest.For example,a researcher wantsto measurecustomerwalking traffic in a store.The researcher's measureis dirt and wear on floor tiles. He or shefirst clarifieswhat the customertraffic means(e.g.,Is the floor a path to anotherdepartment?Doesit indicatea good locationfor a visualdisplay?)Next,he or shJsystematicallymeasuresdirt or wear on the tiles. comparesit to that in other locations,and recordsresultson a regularbasis(e.g.,every month). Finally,the researcherrules out other reasonsfor the observations(e.g.,the floor tile is of lower quality and wearsfaster,or the location is nearan outsideentrance).
CONTENT ANATYSIS What ls Content Analysis? Content analysisis a techniquefor gatheringand analyzingthecontentof text. The contentlefers to words, meanings,pictures,symbols,ideas, themes,or any message that can be communi_ cated. The text is anything written, visual, or spokenthat servesasa medium for communica_ tion. It includesbooks,newspaperandmagazine articles;advertisements, speeches, official documents,films andvideotapes, musicallyrics,pho_ tographs,articlesof clothing,and works of art. The content analysisresearcherusesobjective and systematiccountingand recordingproceduresto producea quantitativedescriptionof the symboliccontent in a text.2There are also qualitative or interpretive versionsof content analysis,but in this chapterthe emphasisis on quantitativedataabouta text'scontent. Contentanalysisis nonreactivebecausethe processof placingwords, messages, or syrnbols in a text to communicateto a readeror receiver occurswithout influencefrom the researcher who anallzesits content.For example,I, asau-
227
thor of this book, wrote words and drew dia_ gramsto communicateresearchmethodscon_ tent to you, the student.The way the book was written and the way you readit arewithout any knowledgeor intention of its everbeingcontent analyzed. Contentanalysisletsa researcher revealthe content (i.e., messages, meanings,etc.) in a sourceof communication (i.e., a book, article, movie, etc.). It lets him or her probe into and discoyercontentin a differentway from the or_ dinaryway of readinga book or watchinga tele_ vrsronprogram. With content analysis,a researchercan comparecontentacrossmany textsand analyze it with quantitativetechniques(e.g.,chartsand tables).In addition, he or shecanievealaspects of the text's contentthat aredifficult to see.l.or example,you might watch televisioncommer_ cialsand feel that non-Whites rurely appearin commercialsfor expensiv. .orrsrrm", goods (e.g.,luxury cars,furs, jewelry perfumefetc.). Content analysiscan document-in obiective. quantitativeterms-whether your vaguefeel_ ings basedon unsystematicobservaiionare true. It yields repeatable,preciseresultsabout the text. Content analysis involves random sam_ pling,precisemeasurement, andoperational -Coding de_ finitions for abstractconstructs. turns aspectsof content that representvariablis into numbers. After a content analysisresearcher gathersthe data,he or sheanalyzesthem with statisticsin the sameway that an experimenter or surveyresearcher would. Topics Appropriate for Content Analysis Researchers haveusedcontentanalysisfor many purposes:to studythemesin popular songsand religioussymbolsin hgnns, trendsin theiopics that newspaperscoverand the ideologicaltone of newspapereditorials,sex-rolestereltypesin textbooksor featurefilms, how often peopleof
2 28
pA RTT wo / c o N D u c rl N c QU A N T l rArl vER E S E A R cH
differentracesappearin televisioncommercials and programs,answersto open-endedsurvey questions,enemypropagandaduring wartime, the coversof popular magazines,personality from suicidenotes,themesin adcharacteristics genderdifferencesin conververtisingmessages, sations,and so on. make on that researchers Generalizations the basisof content analysisare limited to the cultural communicationitself.Content analysis cannot determinethe truthfulnessof an assertion or evaluatethe aestheticqualitiesof literature. It revealsthe content in text but cannot interpretthe content'ssignificance.Researchers shouldexaminethe text directly. Content analysisis usefulfor threetypesof researchproblems.First, it is helpfirl for problems involving a large volume of text. A researchercanmeasurelargeamountsof text (e.g., yearsof newspaperarticles)with samplingand multiple coders.Second,it is helpful when a topic must be studied"at a distance."For example,contentanalysiscanbe usedto studyhistorical documents,the writings of someonewho in a hostileforeigncounhasdied,or broadcasts try. Finally,contentanalysiscanrevealmessages in a text that are difficult to seewith casualobservation.The creatorof the text or thosewho readit may not be awareof all its themes,biases, or characteristics.For example,authors of preschoolpicture books may not consciously intend to portray childrenin traditional stereotyped sexroles,but a high degreeofsex stereotyping has been revealed through content analysis.3 Measurement and Coding General Issues. Carefirl measurementis crucial in content analysisbecausea researcher convertsdiffrrseand murky symboliccommunication into precise,objective,quantitativedata. He or shecarefullydesignsand documentsproceduresfor codingto makereplicationpossible. The researcheroperationalizesconstructsin content analysiswith a coding system.A coding
systemis a setof instructionsor ruleson how to observeand recordcontentfrom systematically tailorsit to the specifictype of text.A researcher text or communicationmedium being studied (e.g.,televisiondrama,novels,photosin magaetc.). The coding system zine advertisements, unit of analysis. alsodependson the researcher's For example,in the studyby Lauzenand Dozier in the most poprr(2005)on genderstereotFpes lar U.S.films in 2002 (discussedin Chapter4), the authorsdevelopeda codingsystembasedon prior studiesof prime-timetelevisionshowsand film. Units. The unit of analysiscanvarya greatdeal in contentanalysis.It canbe a word, a phrase,a theme,a plot, a newspaperarticle, a character, and so forth. In additionto units of analysis,reuseotherunits in contentanalysisthat searchers may or may not be the sameasunits of analysis: recordingunits, contextunits, and enumeration amongthem,and units.Therearefewdifferences they are easilyconfused,but eachhaga distinct role.In simpleprojects,all threearethe same.
What Is Measured? Measurementin content systematic, analysisusesstructuredobservation: carefirlobservationbasedon written rules.The rules explain how to categoize and classifyobservations.As with other measurement,categories should be mutually exclusive and exhaustive.Written rulesmakereplicationpossible and improve reliability. Although researchersbegin with preliminary coding rules, they often conducta pilot studyand refinecoding on the basisof it. Codingsystemsidentifr four characteristics of text content: frequency,direction; intensity, from oneto all measures and space.A researcher in a contentanalysisresearch four characteristics project. Frequency. Frequencysimply meanscounting whether or not somethingoccursand, if it occurs,how often.For example,how many elderly peopleappearon a televisionprogramwithin a
CHAPTER9 , / NO NREACTI VERESEARC H A N D S E C O N D A R YA N A L Y S I S
given week? What percentage of all characters are they, or in what percentage of programs do they appear? Direction. Direction is noting the direction of messagesin the content along some continuum (e.g., positive or negative, supporting or opposed). For example, a researcherdevisesa list of ways an elderly television character can act. Some are positive (e.g., friendly, wise, considerate) and some are negative (e.g., nasty, dull, selfish). Intensity. Intensity is the strength or power of a messagein a direction. For example, the characteristic of forgetfulness can be minor (e.g., not remembering to take your keys when leaving home, taking time to recall the name of someone you have not seen in years) or major (e.g., not remembering your name, not recognizing your children). Space. A researchercan record the sizeof a text messageor the amount of spaceor volume allocated to it. Spacein written text is measured by counting words, sentences,paragraphs, or space on a page (e.g.,squareinches).For video or audio text, space can be measured by the amount of time allocated. For example, a TV character may be present for a few secondsor continuously in every sceneof a two-hour program. Coding, Validity, and Reliability Manifest Coiling. Coding the visible, surface content in a text is calTedmanifestcoding. For example, a researchercounts the number of times a phrase or word (e.g., red) appears in written text, or whether,a specific action (e.g.,a kiss) appears in a photograph or yideo scene.The coaing system lists terms or actions that are then located in text. A researchercan use a computer program to search for words or phrases in text and have a computer do the counting work. To do this, he or shelearns about the computer program, develops a comprehensive list of relevant
229
words or phrases, and puts the text into a form that computers can read.4 Manifest coding is highly reliable because the phrase or word either is or is not present. Unfortunately, manifest coding does not take the connotations of words or phrases into account. The same word can take on different meanings depending on the context. The possibilitythat there are multiple meanings of aword limits the measurement validity of manifest coding. For example, I read a book with a red cover that is a real red herring. Unfortunately, its publisher drowned in red ink because the editor could not deal with the redtapethatoccurs when a book is redhot. The book has a story about a red fire truckthat stops at redlights only after the leavesturn red.Thereis also a group of Redswho carry red flags to the little red schoolhouse. They are opposed by red-blooded redneckswho eatred meat and honor the red, white, and blue. The main character is a red-nosed matador who fights redfoxes,not bulls, with his redcape. Red-lipped little Red Riding Hood is also in the book. She develops red eyesand becomes red-faced. after eating a lot of redpeppets in the redhghtdistrict. She is given a redbackside by her angry mother, aredhead. In the study of gender stereot)?es in films in 2002, Lauzen and Dozier (2005) largely used manifest coding. Coders coded eachcharacter in a film as male or female, the estimated age of each character in one of7 categories,the occupation ofeach character, and whether a character was formally appointed to provide guidance or direction in a group or informally emgered in su-cha function. Latent Coiling. A researcher asing latent coding (also called semantic analysis) looks for the underlying, implicit meaning in the content of a text. For example, a researcher reads an entire paragraph and decideswhether it contains erotic themes or a romantic mood. The researcher's coding system has general rules to guide his or her interpretation of the text and for determin-
2 3O
pA RTT wo / c o N D U c rtN G e u A N T trA TtvER E S E A R cH
ing whether particular themesor moods are present. Latent codingtendsto be lessreliablethan manifestcoding.It dependson a coder'sknowledgeof languageand socialmeaning.sTraining, practice,and written rules improve reliabiliry but still it is difficult to consistentlyidentify themes,moods,and the like. Yet, the validity of latentcodingcanexceedthat of manifestcoding becausepeoplecommunicatemeaningin many implicit waysthat dependon context,not just in specificwords. A researchercan useboth manifestand Iatent coding.Ifthe two approachesagree,the final result is strengthened;if they disagree,the researchermay want to reexaminethe operational and theoreticaldefinitions. IntercoderReliability. Content analysisoften involvescoding information from a very large number of units. A researchproject might involveobservingthe contentin dozensof books, hundredsof hours of televisionprogramming, or thousandsof newspaperarticles.In addition to coding the information personally,a researchermayhire assistants to helpwith the coding. He or sheteachescodersthe codingsystem and trains them to fill out a recordingsheet. Codersshouldunderstandthe variables,follow the codingsystem,and askabout ambiguities.A researcher recordsall decisionshe or shemakes about how to treat a new specificcoding situation after codingbeginsso that he or shecanbe consistent. A researcherwho usesseveralcodersmust alwayscheckfor consistencyacrosscoders.He or she doesthis by askingcodersto codethe sametext independentlyand then checkingfor consistenryacrosscoders.The researchermeasuresintercoderreliabilitywith a statisticalcoefficientthat tellsthe degreeof consistencyamong coders.The coefficientis alwaysreportedwith Thereare the resultsofcontent analysisresearch. severalintercoderreliabilitymeasures that range from 0 to 1, with 1.0 signifying perfect agreement among coders.An interreliability coeffi-
cent of.80 or better is generallyrequired,alfor exploratoryrethough.70maybe acceptable search.When the codingprocessstretchesovera considerabletime period (e.g.,more than three alsochecksreliabilityby months),the researcher having eachcoderindependentlycodesamples of text that were previouslycoded.He or she then checksto seelghetherthe codingis stableor changing.For example,six hours of television episodesare codedin April and codedagainin )uly without the coderslooking at their original codingdecisions.Largedeviationsin codingneretrainingand codingthe text a second cessitate time. In the studyof the 100most popularU.S. films of 2002byLauzenandDozier(2005),three graduatestudentsworked ascoders.During an initial training period they studied the coding systemandvariabledefinitions.Next,the coders practicedby codingindependentof.oneanother severalfilms that were not in the study then comparingand discussingresults.For codingof study films, 10 percentof all films were double codedto calculateintercoderreliability measures.Intercorderreliability measureswere calculatedfor eachvariable.For the genderof the major characterin the film it was.99,for occupation of the chactersit was.91,and for the age of charactersit was.88.
ContentAnnlysiswithVisual Material. Using content analysisto study visual "text," such as photographs, paintings, statues, buildings, clothing,andvideosand film, is difficult. It comor emotionalcontent indimunicatesmessages rectlythrough images,qrnbols, and metaphors Moreover,visual imagesoften contain mixed at multiple levelsof meaning. messages To conductcontent analysison visualtext, the researchermust "read" the meaning(s within visualtext. He or shemust interpretsigns and discoverthe meaningsattachedto syrnbolic images.Such"reading" is not mechanical(i.e., imageX alwaysmeansG); it dependsheavilyon the cultural contextbecausethe meaningof an imageis culturebound. For example,a red light
CHA P T ER 9 / N ON R EA C T IVREES EA R CAHN D S E C ON D A RAYN A LY S IS does not inevitablymean "stop"; it means "stop" onlyln cultures where people have given it that meaning. People construct cultural meanings that they attach to syrnbolic images, and the meanings can changeover time. Some meanings are clearer and more firmly attached to s).rnbols and imagesthan others. Most people share a common meaning for key symbols of the dominant culture, but some people mayread a qnnbol differently. For example, one group of people may "read" a national flag to mean patriotism, duty to nation, and honor of tradition. For others, the same flag evokes fear, and they read it to indicate government oppression, abuse of power, and military aggressio4.A researcherpursuing the content analysisof imagesneedsto be aware of divergent readings of symbols for people in different situations or who may have diverse beliefs and experiences. Sociopolitical groups may invent or construct new symbols with attached meanings (e.g., a pink triangle came to mean gay pride). They may wrestle for control of the meaning of major existing symbols. For example, some people want to assigna Christian religious meaning to the Christmas tree; others want it to represent a celebration of tradition and familyvalues without specific religious conten| others seeits origins as an anti-Christian pagan symbol; and still others want it to mean a festive holiday season for commercial reasons. Becauseimages have symbolic content with complex, multilayer meaning, researchersoften combine qualitative judgments about the images with quantitative data in content analysis. For example, Chavez (2001) conducted a content analysisof the coversof major U.S. magazines that dealt with the issue of immigration into the United States.Looking at the covers of 10 magazinesfrom the mid-1970s to the mid1990s,he classified the covers as having one of three major messages:affirmative, alarmist, or neutral or balanced. Beyond his classification and identifring trends in messages,he noted how the mix of people (i.e., race, gender, age,
23I
and dress) in the photographs and the recurrent use of major syrnbols, such as the Statute of Libefty or the U.S. flag, communicated messages. Chavez argued that magazine covers are a site, or location, where cultural meaning is created. Visual images on magazine covers have multiple levels of meaning, and viewers construct specific meanings as they read the image and use their cultural knowledge. Collectively, the covers convey a worldview and expressmessagesabout a nation and its people. For example, a magazine cover that displayed the icon of the Statute of Liberty as strong and full of compassion (message:welcome immigrants) was altered to have strong Asian facial features (message: Asian immigrants distorted the national culture and altered the nation's racial make-up), or holding a large stop sign (message:go away immigrants). Chavez (2001: a$ observedthat "images on magazines both refer to, and in the process,help to structure and construct contemporary'American' identity." (SeeBox 9.3 for another content analysisexample.)
How to Conduct Content Analysis Research QuestionFormulntion As in most research, content analysisresearchers begin with a researchquestion.When the questioninvolves variablesthat are messages or syrnbols,content analysismay be appropriate.For example,I want to study how newspapers covera political campaign.My construct"coverage"includesthe amountof coverage, the prominenceof the coverage,and whetherthe coveragefavorsone candidateoveranother.I could surveypeopleabout what theythink of the newspapercoverage, but a better strategyis to examinethe newspapers directlyusingcontentanalysis. Units of Analysis. A researcher decideson the units of analysis(i.e.,the amount of text that is assigneda code).For example,for a political campaign,eachissue(or day) of a newspaperis the unit of analysis.
232
PART TWO /
CO NDUCTI NG Q UANTI TA T I V ER E S E A R C H
Two studiesthat examinedrace-ethnicityand advertisingin the UnitedStatesillustratehow content analysisis conducted.Mastro and Stern (2003) wantedto seewhethertelevisionadvertisingrepresents major racial-ethnicgroups proportionateto their presencein U.S.society.Theyexamined a oneweek randomsampleof prime-timetelevisionprogrammingfor sixU.S.televisionnetworks(ABC,CBs, NBC,Fox,UPN,and WB) drawnfrom a three-week periodin February2001. Primetime was Monday .l throughSaturday8:00 p.v.to 1:00 p.v.ESTand Sunday7:OO-11:00 e.r'a. Fourundergraduate studentsweretrainedascoders.Theyusedtwo unitsof analysis: a commercial (excluding localcommercials, politicaladvertisements, and trailersfor upcoming programs)and the first three speakingcharactersin a commercial. Variables includedproducttype based on a 3O-productcodingscheme,setting(e.g.,work, outdoors),relationto product (e.g.,endorse,use, neitheror both), job authority,familystatus,social authority,sexualgazing,andaffectivestate(e.g.,cry, showanger,laugh).Other variables includedrespect shownfor a character,character's age,and affability (friendlyor hostile).The study coded 2,880 commercialswith 2,3 1 5 speakingcharacters, among whom2,290 had a race-ethnicityidentified.Data analysis foundthat AfricanAmericancharacters were most often shownadvertisingfinancialservices(1 9 percent)or food (.17 percent),Asianswereassociated with technologyproducts (30 percent),and Latinoswere shownsellingsoap (40 percent).In general,Whiteswereslightlyoverrepresented, Blacks equallyrepresented, but Asians,Latinos,and Native Americans underrepresented. For example, Latinos .l are l2 percentof the populationbut had percent of speakingparts, and were usuallyscantlyclad young peoplewith noticeableaccents.The authors saidthat AfricanAmericans in appearin commercials a way that approximates their proportionin the UnitedStates,but other racialminoritiesare underrepresented or limitedto specificproducts.
In anotherstudy,Mastroand Atkin (2002) examinedwhetheralcohol advertisingto promote brandsand makedrinkingappearglamorousinfluencedhigh schoolstudentswho are too young to drink legally.They lookedat alcoholsignsand billboardsin a Mexican-American Chicagoneighbor hood.Theyfirst photographed alloutdoorbillboard and signsconcerning alcoholin the neighborhoo overa two-dayperiodin March'l 999. After a period ofcodertraining,two femalegraduatestudentscontent-analyzed the photographs, codingthe following variables:product type, product name,numberof humanmodels,and the race,age,genderof each model.More subjective-latent aspectsof models codedincludedattractiveness, sexiness, stylishnes friendliness, andactivitylevel.In addition,placemen of productsand colorsin the billboardwerecoded. Codersalso classifiedan overalltheme of the billboard as romance,individuality,relaxation,sports, adventure, or tradition.Next,a questionnaire wasdevelopedfor studentsat a high schoolin the neighborhoodwhere89 percentof the studentswere MexicanAmerican.Studentsin grades1 0, 1 1, and .l 2 wereaskedto volunteerto completethe survey .l acrossa three-dayperiod and 23 completedit. Questionnaireitems askedabout attention, exposure,recall,and brandexposureto the outdoorsigns andbillboards aswellasdrinkingintention,approva of underagedrinking,and pro-drinkingbeliefs.Resultsshowedthat a student'srecallof billboardimagesdid not affect his or her drinkingattitudes However,brandexposureand acceptingthe themes in the billboardswere associatedwith greaterapprovalof underagedrinking.The generalimpacton the studentswas presentbut not strong.The arrthors suggestedthat the weakimpactwas because there werefew MexicanAmericanmodelsand the modelswere older.Also,surveymeasuresof family beliefssuggestedthat the influenceofthe student's familyandculturemayhaveweakened the billboard's impacton pro-drinkingattitudes.
CHAPTER9 /
A N D S E C O N D A R YA N A L Y S I S NO NREACTI VERESEARC H
oftenuserandomsamSampling. Researchers pling in content analysis.First, they define the population and the samplingelement.For example,the population might be all words, all or all articlesin certain sentences, all paragraphs, typesof documentsovera specifiedtime period. Likewise.it could includeeachconversation,situation,scene,or episodeofcertaintypesoftelevision programsover a specifiedtime period. For example,I want to know how women and minorities are portrayed in U.S. weekly newsmagazines. My unit of analysisis the article.My population includesall articlespublishedin Time, Newsweek,and U.S.Newsand World Reportbetween1985and 2005.I firstverifrthat the three magazineswere publishedin thoseyears and definepreciselywhat is meantby an "article." For instance,do film reviewscount asartifor cles?Is therea minimum size(two sentences) as one counted Is multipart article an article? a or two articles? Second,I examinethe threemagazinesand find that the averageissueof eachcontains45 articlesand that the magazinesarepublished52 weeksper year.With a 2}-yeartime frame,my population containsover 140,000articles(3 X My samplingframeis 45 x 52 X 20 = 140,400). I decideon the samNext, a list of all the articles. looking at my budget ple sizeand design.After sizeto limit the sample and time, I decideto is 1 perratio 1,400articles.Thus,the sampling I sysdesign. avoid cent.I alsochoosea sampling issues are rnagazrne tematic samplingbecause calendar to the published cyclicallyaccording (e.g.,an intervalofevery52ndissueresultsin the issuesfrom each sameweekeachyear).Because I use stratified sammagazineare important, I,40013= sampling pling. I stratif'by magazine, that arto ensure 467 articlesfrom each.I want years, so I also 20 ticlesrepresenteachof the 23 articles in about stratifyby year.This results per magazineper year. Finally, I draw the random sampleusing a random-numbertableto select23 numbersfor the 23 samplearticlesfor eachmagazineforeach year.I developa samplingframe worksheetto
233
keep track of my sampling procedure. SeeTable 9.1 for a sampling frame worksheet in which 1,398sample articles are randomlyselected from 140,40I articles. Variables and Constructing Coding Categories. In my example, I am interested in the construct of an African American or Hispanic American woman portrayed in a significant leadership role. I must define "significant leadership role" in operational terms and express it as written rules for classifring people named in an article. For example, if an article discussesthe achievements of someone who is now dead, does the dead person have a significant role? What is a significant role-a local Girl Scout leader or a corporate president? I must also determine the race and sex of people named in the articles. What if the race and sex are not evident in the text or accompanyingphotographs? Howdo I decide on the person's raceand sex? BecauseI am interested in positive leadership roles, my measure indicates whether the role was positive or negative. I can do this with either latent or manifest coding. With manifest coding, I create a list ofadjectives and phrases.If someone in a sampled article is referred to with one of the adjectives, then the direction is decided. For example, the terms brilliant and top performer arepositive, whereas drugkingpin and uninspired are negative. For latent coding, I create rules to guide judgments. For example, I classify stories about a diplomat resolving a difRcult world crisis, abusiness executiveunable to make a firm profitable, or a lawyer winning a caseinto positive or negative terms. (Relevant questions for coding each article are in Box 9.4.) In addition to written rules for coding decisions, a content analysis researcher creates a recordingsheef(also called a codingform or tally sheet)on which to record information (seeBox 9.5). Each unit should have a separaterecording sheet.The sheetsdo not have to be piecesofpaper; they can be 3" x 5'' or 4'' X 6" file cards, or lines in a computer record or file. When a lot
234
PART T W O , / CO NDUCTI NC Q UANTIT A T I V ER E S E A R C H
TABLE 9.1
Time Time Time
ExcerptfromSamplingFrameWorksheet
1-7,1985 January
pP.2-3 p. 4, bottom p. 4, top
000001 000002 000003
No No Yes- 1
p p .2-5 p. 5, right c o l u mn p . 6 ,l eft column p .7
002101 002102
Y es-l 0 No
002103
No
002't04
No
pp.4-5 p.5,bottom p. 5, top pP.1-2 p.3
002201 oo2202 oo2203 0 10 0 3 0 0 10 0 3 r
Yes-22 No Y es-23 No Yes- l
0467
p .6 2
1 40401
Y es-23
1389
0001
a a a
Time Time
March'f -7,zOOs
Time Tine
0454
o a a
Time Time Time Newsweek Newsweek
De c e mb e2r4 -3 1 , 2 0 0 5
1 -7 ,1 9 8 5 J a n u a ry
0468 0469
a a a
U.5. News
De c e mb e2r5 -3 1 , 2 0 0 5
*"Yes" meansthe numberwaschosenfrom a randomnumbertable.The numberafterthe dashis a countofthe numberof articlesselectedfor a year.
Mqgazine
of the article.What is the magaCharacteristics zine?What is the date of the article?How large is the article?What was its topic area?Where did it appearin the issue?Were photographs used? 2. Peoplein thearticle.How manypeopleare named in the article?Of these,howmanyaresignificant in the article?What is the raceand sexof each personnamed?
roles.For each significantpersonin 3. Leadership
roles? the article,whichoneshaveleadership What is the field of leadershipor professionof the person? 4. Positiveor negativeroles.For each leadershipor role, rate how positivelyor negaprofessional tivelylt is shown.Forexample,5 : highlypositive,4 = positive,3 : neutral,2 : negative,1 = highlynegative,0 = ambiguous.
CHAPTER9 , / NO NREACTI V ER E S E A R C H A N D S E C O N D A R YA N A L Y S I S
235
Blank Example ProfessorNeuman,SociologyDepartment
Coder:
Minority/MajorityCroup Representation project in Newsmagazines ARTICLE #_
MACAZTNE:
DATE:
SIZE:_
col.in.
Totalnumberof peoplenamed_
Numberof Photos
No. peoplewith significant roles:_
ArticleTopic:
Person_: Person_: Person_:
Race:_
Cender:_ Gender:_ Cender:_
Race:_ Race:_ Race:_
Person_: Pe r s on' : Person_: Person_: Person_:
Cender:_ C e n d e r:_
Rac e :_ Race:_ Race:_ Race:_
Cender:_ Gender:_ Cender:_
Leader?:_
Field?_
Leader?:_ Leader?:_ Leader?:_ Leader?:_
Field?_ Field?_ Field?_ Field?_
Rating:_ Rating:_ Rating:_ Rating:_ Rating:_
Leader?:
Field?_ Field?_ Field?_
Rating:_ Rating:_ Rating:_
Leader?:_ Leader?:_
Exampleof Completed RecordingSheetfor One Article ProfessorNeuman,SociologyDepartment
Coder:SusanJ.
Minority/Majoritycroup Representation project in Newsmagazines
ARTICLE # 0454
MA C A Z IN ET:i me
D A TEMarch : 1-2,2005
Totalnumberofpeoplenamed5
Numberof Photos0
No. peoplewith significantroles:4 Person | Person 2 Person 3 Person 4 Person_: Person _: Person_: Person_:
: : : :
Race:White Race:White Race:Black Race:White Race: _ Race: _ Race:_ Race:_
S IZE 14 : col .i n.
ArticleTopic:ForeignAffairs Cender: Cender: Cender: Cender: Cender:_ Cender: _ Cender:_ Cender:
M M F F
of information is recorded for each recording unit, more than one sheetof paper can be used. When planning a project, researcherscalculate the work required. For example, during my pilot-test, I find that it takes an averageof tS min_
Leader?: y Leader?: N Leader?:y Leader?:y Leader?: _ Leader?: _ Leader?: _ Leader?:
Field?Banking Field?Government Field?CivilRiehts Field?1Qgygryngql Field?_ Field?_ Field? Field?
Rating: 5 Rating:NA Rating: 2 Rating: 0 Rating: _ Rating: _ Rating:_ Rating:
utesto read and codean article.This doesnot include samplingor locating magazinearticles. With approximately1,400articles,that is 350 hours of coding,not countingtime to verifi. the accnracyof coding.Because350 hours is atout
236
P A Rr r wo
R E S E A R cH / c o N D U c rl N c QU A N T ITA TIvE
nineweeksof nonstopwork at 40 hours a week, ascoders. I shouldconsiderhiring assistants Eachrecordingsheethas a placeto record the identificationnumber of the unit and spaces for information about eachvariable.I alsoput identi$ringinformation aboutthe researchproject on the sheetin caseI misplaceit or it looks similar to other sheetsI have.Finally, if I use multiple coders,the sheetremindsthe coderto check intercoder reliability and, if necessary, makesit possibleto recodeinformation for inaccuratecoders.After completingall recording sheetsand checkingfor accuracy,I can begin dataanalysis. lnferences The inferencesa researchercan or cannotmake on the basisofresultsis criticalin contentanalywhat is in the text. sis.Contentanalysisdescribes of thosewho crethe intentions reveal It cannot in the that messages the effects or text atedthe examFor receive them. who those have on text ple,contentanalysisshowsthat children'sbooks That doesnot necessarcontainsexstereotFpes. beliefsor behaviorsare children's mean that ily suchan inference stereotypes; by the influenced project on how chilresearch a separate requires develop. perceptions dren's
EXISTING STATISTICS/ DOCUMENTS AND SECONDARY ANALYSIS Appropriate Topics Many tlpes of information about the social world havebeen collectedand are availableto Someinformation is in the form the researcher. (books,reports,etc') documents of statistical Other ininformation' numerical that contain compilapublished form of in the formation is computerized or on in a library tions available cansearch records.In eithercase,the researcher a rewith information of through collections
searchquestionand variablesin mind,.andthen the information in new waysto adreassemble question. dressthe research It is difficult to specifrtopicsthat areapprothey because priatefor existingstatisticsresearch information which are so varied.Any topic on hasbeen collectedand is publicly availablecan be studied.In fact, existingstatisticsprojects may not fit neatlyinto a deductivemodel of recreativelyr9searchdesign.Rather,researchers organize the existing information into the louiiubl"rfor a researchquestionafter first finding what dataareavailable. i " E*p.ri-ents arebestfor topicswherethe researchercontrolsa situationand manipulatesan independentvariable.Surveyresearchis bestfor topics where the researcherask questionsand learnsaboutreportedattitudesorbehavior.Content analysisis best for topics that involve the in culturalcommunication. contentof messages Existingstatisticsresearchis bestfor topics that involve information routinely collectedby Public or prilargebureaucraticorganizations. vate organizationssystematicallygather many typesof information. Suchinformation is gatheredfor policy decisionsor asa public service.It is rarelycollectedfor purposesdirectlyrelatedto a specificresearchquestion.Thus, existingstatisticsresearchis appropriatewhen a researcher wantsto test hypothesesinvolving variablesthat are also in official reports of social,economic, and political conditions.Theseinclude descriptions of organizationsor the peoplein them. Often, suchinformation is collectedoverlong time periods.For example,existingstatisticscan be usedby a researcherwho wantsto seewhether in unemploymentand crimeratesareassociated 150citiesacrossa 2}-yeatperiod' Downey (2005)conductedan existingstatistics study on racial inequality (BlackAVhite) and living near a toxic pollution sitein Detroit. He usedcensusdataon the population/housing and manufacturingdirectoriesof manufacturing facilities.He alsoidentifiedhighly polluting industriesand usedthe EnvironmentalProtection Agency'sinventory of toxic chemicals.His
CHAPTER9 , / NO NREACTI V ER E S E A R C H A N D S E C O N D A R YA N A L Y S I S
unit of analysiswas the censustract. Downey testedcompetingmodels of environmentalinequalitp (1) racistsiting poliry: toxic siteswere placedin Black residentialareas,(2) economic inequality:low-incomepeoplewho are disproportionatelyBlackmove into areasnear toxic sitesbecausethey find low-cost housingthere, and (3) residentialsegregation: Whites move into specificareasand keepout non-Whites.He found greatestsupport for the residentialsegregation model. Paradoxically,it meant that Blackswerelesslikely thanWhitesto live closeto a toxic pollution site. This was becauseWhites had obtainedhousing near the factorieswhere they worked and kept Blacksfrom moving in but those factorieswere the maior sourcei of toxic pqllution.
237
health and nutrition, public safety,education and training, worh income,cultureand leisure, socialmobilig, and public parricipation. A more specificexampleof a socialindica_ tor is the FBI'suniform crime index.It indicates the amount of crime in U.S.society.Socialindicatorscanmeasurenegativeaspectsof sociallife, suchasthe infant mortality rate (the deathrate of infants during the first year of life) or alcoholism, or they can indicatepositive aspects, such as job satisfactionor the percentae of housingunits with indoor plumbing. Socialin_ dicatorsoften inyolve implicit valuejudgments (e.g.,which crimes are seriousor what constitutesa good quality of life). Locating Data
LocatingExistingstatistics. The main sources of existingstatisticsare goyernmentor interna_ During the 1960s,somesocialscientists, dissatis- tional agenciesand private sources.An enorfied with the information availableto decision mous volume and varietyof information exists. makers,spawnedthe "social indicators,moveIf you plan to conduct existingstatisticsrerment"to developindicatorsof socialwell-being. search,it is wiseto discussyour interestswith an Many hopedthat information aboutsocialweliinformation professional-in this case,a refer_ beingcould be combinedwith widelyusedindiencelibrarian, who can point you in the direc_ cators of economicperformancei..g., gross tion of possiblesources. nationalproduct) to better inform government Many existingdocument5nrs ..frss,'_1fin1 and other policymaking officials. Thus, reis, publicly availableat libraries-but the time searchers wantedto measurethe qualityof social and effort it takesto searchfor specificinforma_ life so that such information could influence tion can be substantial.Researchers who conpublic poliry.6 duct existing statisticsresearchspend many Today,there are many books,articles,and hours in libraries or on the Internet. After the reportson socialindicators,and evena scholarly information is located,it is recorded on cards, journal, SocialIndicatorsResearch, devotedto graphs,or recordingsheetsfor lateranalysis.Ofthe creationand evaluationofsocial indicators. ten, it is alreadyavailablein a format for comThe U.S. CensusBureau produced a report, putersto read.For example,insteadof recording SocialIndicators, and the United Nations has voting datafrom books,a researcher could usea many measuresof socialwell-beingin different socialsciencedata archiveat the UniversiWof nations. Michigan(to be discussed). A socialindicator is any measureof social Thereareso many sourcesthat only a small well-beingusedin poliry. Thereare many spesampleofwhat is availableis discussed here.The cific indicators that are operationalizationJof single-mostvaluablesourceof statisticalinforwell-being.For example,socialindicatorshave mation about the United Statesis the Statistical beend€velopedfor the following areas:populaAbstractof the United States,which has been tion, family,housing,socialsecurityandwelfare, publishedannually(with a few exceptions) since Social Indicators
238
R E S E A R cH P A RTT wo / c o N D U c rl N c QU AN T IT ATIvE
1878.The StatisticalAbstractis availablein all public librariesand on the Internet and can be purchasedfrom the U.S. Superintendentof bocuments. It is a selectedcompilation of the many official reports and statisticaltablesproIt contains ducedby U.S.governmentagencies. of more hundreds from information statistical want to may You reports' government detailed documents. government specific more examine (The detail of what is availablein government documentsis mind boggling.For example,you canlearn that thereweretvvoAfrican American femalesover the ageof 75 in Tucumcari City, NewMexico,in 1980.) The Statistical Abstract has over 1,400 charts,tables,and statisticallists from over 200 governmentand private agencies.It is hard to graspall that it containsuntil you skim through ih" tabl.r. A two-volumesetsummarizessimilar information across many years; it is called of the U.S.:ColonialTimesto HistoricalStatistics 1970. Most governmentspublish similar statistical yearbooks.Australia's Bureau of Statistics producesYearbook Australia,StatisticsCanada New Zealand'sDeproducesCanadaYearbook, ZealandOfpublishesNew Statistics partmentof the Kingdom, United the andin ficialYearbook, AbAnnual publishes Office Central Statistics Many nationspublish books stractof Statistics.T aswell. statistics, with historical statisticaldocuments government Locating existsolelyto publications Some is an art in itself. American the example, For assistthe researcher. and Guide A Comprehensive StatisticsIndex: U.S. the of Publications Index to the Statistical Governmentand StatisticsSources:A Subject SocialEduGuideto Data on Industrial,Business, U.S'and the Topics and Other cation,Financial for for the guides helpful are two Internationqlly internaand Nations The United United States.s suchastheWorld Bankhavetheir tional agencies own publicationswith statisticalinformation for variouscountries(e.g.,literacyrates,percentage of the labor force working in agriculture,birth Yearbook, rates)-for example,theDemographic
UNESCOStatisticalYearbook,and United NationsStatisticalYearbook. In addition to governmentstatisticaldocuments,there are dozensof other publications' Many are producedfor businesspurposesand can be obtainedonly for a high cost.They include information on consumerspending,the location of high-incomeneighborhoods,trends in the economy,andthe like.e Over a dozenpublicationslist characteristics ofbusinessesor their executives.Theseare found in largerlibraries.Threesuchpublications are asfollows:
PrincipalIndustrialBusiDun qndBradstreet is a guide to approximately51'000 nesses in 135 countrieswith informabusinesses officers, tion on sales,numberof employees, andproducts. WhoOwnsWhomcomesinvolumesfor nations or regions(e.g.,North America,the United Kingdom,Ireland,andAustra'lia).It and aslistsparentcompanies,subsidiaries, sociatedcomPanies. of Corporations, StandardandPoor'sRegister lists Directorsand Executives about 37,000 U.S.and Canadiancompanies.It hasinformation on corporations,products,officers, industries,and salesfigures. Many biographicalsourceslist famouspeople and provide background information on wants them. Theseareusefulwhen a researcher or career, background, social the to learn about The individuals. famous of other characteristics publicationsare compiledby companiesthat sendout questionnairesto peopleidentified as "important" by some criteria. They are public ,o,rt."t of information' but they dependon the cooperationand accuraryof indMdualswho are selected. publications. Politicshasits own specialized biographical has One types. Therearetwo basic The politicians. information on contemporary enlaws voting, on other type has information
CHAPTER9 /
A N D S E C O N D A R YA N A L Y S I S NO NREACTI VERESEARCH
acted,and the like. Here are three examplesof politicalinformationpublicationsfor the United States: Almanacof AmericanPoliticsis a biannual publicationthat includesphotographsanda short biographyof U.S. governmentofficials. Committee appointments, voting records,and similar information are provided for membersof Congressand leaders in the executivebranch. America Votes:A Handbookof ContemporaryAmericanElectionStatisticscontainsdetailed.voting information by county for most statewideand national offices. Primary electionresultsare includeddown to the countylevel. Vital StatisticsonAmericanPoliticsprovides dozensof tableson political behavior,such as the campaignspendingof everycandidate for Congress,their primary and final votes,ideologicalratingsby variouspolitical ': organizations,and a summary of voter registrationregulationsby state. Another sourceof public information consistsof listsof organizations(e.g.,business,educational,etc.)producedfor generalinformation purposes.A researchercan sometimesobtain membershiplists of organizations.There are givenby faalsopublicationsof public speeches mouspeople. SurveyData. Secondaryanalysisis Second.ary a specialcaseof existing statistics;it is the reanalysisof previouslycollectedsurveyor other datathat were originally gatheredby others.As opposedto primary research(e.g.,experiments, surveys,and content analysis),the focus is on analryzingratherthan collecting data. Secondary It is analysisis increasinglyusedby researchers. relatively inexpensive;it permits comparisons acrossgroups,nations, or time; it facilitates replication; and it permits askingabout issues not thought ofby the original researchers.
239
Large-scaledata collection is expensive and difficult. The cost and time required for a major national surveythat usesrigorous techniques are prohibitive for most researchers. Fortunately, the organization, preservation, and dissemination of major survey data sets have improved. Today, there are archives ofpast surveysthat are open to researchers. The Inter-University Consortium for Political and Social Research(ICPSR) at the University of Michigan is the world's major archive of social sciencedata. Over 17,000 survey research and related sets of information are stored and made available to researchers at modest costs. Other centers hold survey data in the United Statesand other nations.lo A widely used source of survey data for the United Statesis the GeneralSocialSurvey (GSS), which has been conducted annually in most years by the National Opinion ResearchCenter at the University of Chicago. In recent years, it has covered other nations as well. The data are made publicly availablefor secondaryanalysisat a low cost (seeBox 9.6).
Limitations Despitethe growth and popularity of secondary data ana\sis and existing statistics research, therearelimitationsin their use.The useof such techniquesis not troublefreejust becausea government agenq/or researchorganizationgathered the data. One danger is that a researcher may usesecondarydataor existingstatisticsthat are inappropriatefor his or her researchquesneedsto tion. Beforeproceeding,a researcher considerunits in the data (e.g.,qpes of people, organizations),the time and placeof data collection,the samplingmethodsused,andthe specific issuesor topicscoveredin the data(seeBox wantingto ex9.7).For example,a researcher amine racial-ethnic tensionsbetweenLatinos and Anglosin the United Statesusessecondary data that includes only the Pacific Northwest and New Englandstatesshould reconsiderthe questionor the useofdata.
240
PART TWO /
CO NDUCTI NC Q UANTI TA T I V ER E S E A R C H
The GeneralSocialSurvey(GSS)is the best-known for secsetof surveydatausedby socialresearchers "to make is GSS The mission of the ondaryanalysis. avairdata quality, scientifically relevant high timely, able to the socialscienceresearchcommunity" in many (Davisand Smith,1 992:1).lt is available formatsand is widelyaccessible computer-readable for a low cost. Neitherdatasetsnor codebooksare them copyrighted.Usersmay copy or disseminate You can find results withoutobtainingpermission. articlesand usingthe CSSin over 2,000 research books. Center(NORC) The NationalOpinionResearch has conductedthe CSS almostevery year since 1972. A typicalyear'ssurveycontainsa random .l A team sampleof about ,500 adultU.S.residents. selectssomequestionsfor inclusion, of researchers questions. researchers canrecommend andindividual year, each questions and topics They repeatsome add and cycle, fourto six-year on a includesome .l in 998, other topicsin specificyears.Forexample, and religion, the specialtopic wasjob experiences
and in 2000, it wasintergrouprelationsand multiculturalism. collectthe data through face-toInterviewers The NORCstaffcarefullyselectsinfaceinterviews. terviewersand trains them in social science About 1 20 methodologyand surveyinterviewing. GSS eachyear. on the work to 1 40 interviewers aremiddle most and percent are women, About 90 aged.The NORCrecruitsbilingualand minorityinwith respondentsare'raceterviewers.Interviewers matchedwith respondents.Interviewsare typically 500 90 minuteslong and containapproximately questions.The responserate has been 71 to 79 is a repercent.The majorreasonfor nonresponse fusalto participate. SocialSurveyProgramconducts The International similarsurveysin other nations.Beginningwith the CermanALLBUSand BritishSocialAttitudesSurvey, participationhasgrownto include33 nations.The nagoal is to conducton a regularbasislarge-scale tionalgeneralsurveysin whichsomecommonquestions areaskedacrosscooperatingnations.
A seconddangeris that the researcherdoes not understandthe substantivetopic. Because researcherswho the data are easilyaccessible, know very little about a topic could makeerroor falseinterpretationsabout neousassumptions needs results.Beforeusinganydata,a researcher to be well informed about the topic. For example,ifa researcher usesdataon high schoolgraduation ratesin Germanywithoutunderstanding the Germanysecondaryeducationsystemwith its distinct academicandvocationaltracks,he or she may make seriouserrors in interpreting results. A third danger is that a researchermay quotestatisticsin greatdetailto givean impression of scientificrigor. This can lead to the which occurs fallacy of misplacedconcreteness, when someonegivesa falseimpressionof preci-
sion by quoting statisticsin greaterdetail than warrantedand "overloading"the details.For example,existingstatisticsreport that the populabut it is betterto tion of Australiais 19,169,083, 19 million. One might saythat it is a little over peopleas percentage of divorced calculatethe of the 2000 data analysis in a secondary L5.65495 report is to it better GeneralSocialSurvey,but I people divorced'r are thatabout15.7percentof
Units of Analysisand VariableAttributes, A commonproblemin existingstatisticsis finding the appropriateunits of analysis.Many statistics not the individual. arepublishedfor aggregates, For example,a tablein a governmentdocument has information (e.g., unemployment rate, crimerate,etc.)for a state,but the unit of analysis for the researchquestion is the individual
A N D S E C O N D A R YA N A L Y S I S CHAPTER9 , / NO NREACTI VERESEARC H
Almosteverycountryconductsa census,or a regularcountof its population.Forexample, Australiahas d o neso si nce'l 88. |, Canadas i n c eI8 7 .| , a n dth e UnitedStatessinceI 790. Most nationsconducta censusevery5 or I 0 years.ln additionto the number of people,censusofficialscollectinformationon topicssuchas housingconditions,ethnicity,religious affiliation.education.and so forth. The censusis a majorsourceof high-qualityexistingstatisticaldata,but it can be controversial. In Canada,an attempt to count the numberof samesex coupleslivingtogetherevokedpublicdebate aboutwhetherthe governmentshoulddocumentthe changesin society.In Creat Britain,the Muslimminority welcomedquestionsabout religionin the 200,| censusbecausethey felt that they had been officiallyignored.In the UnitedStates,the measurement of raceand ethnicitywashotly debated,so in in the 2000 census,peoplecould placethemselves multipleracial,/ethnic categories. The U.S.2000 censusalsogenerateda serious publiccontroversybecauseit missedthousandsof people,mostfrom low-incomeareaswith concentrations of recent immigrantsand racialminorities. Somedoublecountingalsooccurredof peoplein highincomeareaswheremanyownedsecondhomes. A contentiousdebatearoseamongpoliticians to end miscountsby usingscientificsamplingand adjusting the census.The politiciansprovedto be lessconcernedaboutimprovingthe scientifica-curacyofthe censusthan retainingtraditionalcensusmethods that wouldbenefittheirown politicalfortunesor help their constituencies, becausethe governmentuses censusdata to draw voting districtsand allocate publicfundsto areas.
(e.g.,"Are unemployedpeoplemore likely to commit property crimes?").The potentialfor committing the ecologicalfullu.y is very real in this situation.It is lessof a problem for secondary surveyanalysisbecauseresearchers can
247
obtain raw information on each respondent from archives. A relatedprobleminvolvesthe categories of variableattributesusedin existingdocumentsor surveyquestions.This is not a problemif the initial data were gatheredin many highly refined Theproblemariseswhenthe original categories. datawere collectedin broad categoriesor ones that do not matchthe needsof a researcher. For example,a researcheris interestedin peopleof Asian heritage.If the racial and ethnic heritage categoriesin a documentare "White," "Black," and "Other," the researcher hasa problem.The "Other" categoryincludespeopleof Asian and otherheritages.Sometimesinformationwascollectedin refinedcategories but is publishedonly in broad categories. It takesspecialefFortto discoverwhethermore refined information was collectedor is publicly available. Valiility. Validity problems occur when the researcher'stheoretical definition does not matchthat of the governmentagencyor organization that collectedthe information. Official policiesand proceduresspeci$'definitions for official statistics.For example,a researcherdefines a work injury as including minor cuts, bruises,and sprainsthat occur on the job, but the official definition in governmentreports only includesinjuries that require a visit to a physicianor hospital.Many work injuries,asdefined by the researcher, would not be in official statistics.Another exampleoccurswhen a researcherdefinespeople as unemployedif they would work if a goodjob wereavailable,if they haveto work part time when theywant full-time work, and if they have given up looking for work. The official definition, however,includes only thosewho are now activelyseekingwork (full or part time) as unemployed.The official statisticsexcludethosewho stoppedlooking, who work part time out of necessity, or who do not look becausethey believeno work is available.In both cases,the researcher's definition differsfrom that in official statistics(seeBox 9.8).
242
PART TWO , / CO NDUCTI NG Q UANTI TA T I V ER E S E A R C H
In most countries,the official unemploymentrate measures only the unemployed(seebelow)asa percent of all workingpeople.lt would be 50 percent higherif two othei categoriesof nonemployedpeople were added:involuntarypart-timeworkersand workers(seebelow).In somecountries discouraged (e.g.,Swedenand UnitedStates),it wouldbe nearly aorUt" if it includedthesepeople.Thisdoesnot considerother nonworkingpeople,transitionalself-em(seebelow).What a ployed,or the underemployed is a theoreticaiandconceptualdecountrymeasures finitionissue:What constructshouldan unemploymentrate measureandwhy measureit?
An economicpolicyor labor marketperspective saysthe rate shouldmeasurethose readyto enter lt definesnonworking the labormarketimmediately. peopleas a supplyof high-qualitylabor,an inputfor to employers.By conusein the economyavailable resourceperspective human policy or a social trast, thosewho are not cursaysthe rate shouldmeasure rently workingto their fullest potential.The rate shouldrepresentpeoplewho are not or cannotfully utilizetheir talents,skills,or time to the fullest.lt definesnonworkingpeopleas a socialproblemofindividuals unable to realizetheir capacity to be productive,contributingmembersof society.
Categoriesof Nonemployed/FullyUtilized Peoplewho meetthreeconditions:lacka payingjob outsidethe home,are Unemployedpeople if it is ofto find work,can beginwork immediately takingactivemeasures fered. Involuntarypart-timeworkers workers Discouraged Other nonworking
Peoplewith a job, but workirregularlyor fewerhoursthanthey areableand willing. Peopleableto workandwho activelysoughtit for sometime,but beingunableto find it, havegivenup looking. Thosenot workingbecausethey are retired,on vacation,temporarilylaid full-timestudents,or in the processof movhomemakers, off,semidisabled, ing.
Transitionalself-employed Underemployed
who are not workingfull time becausethey arejust starting Self-employed goingthroughbankruptcy. are or a business overqualwith a temporaryfull-timejob for whichthey areseriously Persons job in whichthey canfullyapplytheirskillsand ified.Theyseeka permanent experience.
AdaptedfromTheEconomisl, Soarce: July22, 199 5, p.7 4.
bery arrestsas a proxy. But the measureis not Another validity problem ariseswhen offientirelyvalid becausemanyrobberiesarenot recial statisticsarea surrogateor proxy for a construct in which a researcheris really interested' ported to the police,and reportedrobberiesdo not alwaysresultin an arrest. becausethe researchercannot This is necessary A third validity problem arisesbecausethe collectoriginaldata.For example,the researcher lackscontrol overhow information is researcher wants to know how many people have been collected.All information, eventhat in official robbed,sohe or sheusespolicestatisticson rob-
CHAPTER9 , / NO NREACTI VERESEARCH A N D S E C O N D A R YA N A L Y S I S
governmentreports,is originally gatheredby peoplein bureaucracies aspart oftheir jobs.A researcherdependson them for collecting,organizing,reporting, and publishing data accurately.Systematicerrors in collectingthe initial information (e.g.,censuspeoplewho avoidpoor neighborhoodsand make up information, or peoplewho put a falseageon a driver'slicense); errorsin organizingand reporting information (e.g.,a police departmentthat is sloppy about filing crime reportsand losessome);and errors in publishinginformation (e.g.,a gpographical error in a table)all reducemeasurement validity. This kind of problem happenedin U.S.statisticson the numberof peoplepermanentlylaid off from their jobs.A universityresearcher reexaminedthe methodsusedto gatherdataby the U.S.Bureauof Labor Statisticsand found an error. Dataon permanentjob lossescomefrom a surveyof 50,000people,but the government agencyfailedto adjustfor a much highersurvey nonresponserate.The correctedfiguresshowed that insteadof a 7 percentdeclinein the number of peoplelaid offbetween1993andl996,ashad beenfirst reported,therewasno change.l2 Rekability. Problemswith reliability canplague existingstatisticsresearch.Reliabilityproblems developwhen official definitionsor the method of collectinginformation changesovertime. Official definitions of work injury disability,unemployment,and the like changeperiodically. Evenifa researcher learnsofsuch changes,consistentmeasurementover time is impossible. For example,during the early1980s,the method for calculating the U.S. unemployment rate changed.Previously,the unemployment rate was calculatedas the number of unemployed personsdivided by the number in the civilian work force.The new method divided the number of unemployedby the civilian work force plus the number of peoplein the military. Likewise, when police departmentscomputerize their records,there is an apparentincreasein crimesreported,not because crimeincreases but due to improvedrecordkeeping.
243
Reliabilitycanbe a seriousproblem in official governmentstatistics.This goesbeyondrecognizedproblems,such as the police stopping poorly dressedpeoplemore than well-dressed people, hence poorly dressed,lower-income peopleappearmore often in arreststatistics.For example,the U.S. Bureau of Labor Statistics found a 0.6 percentincreasein the femaleunemploymentrate after it usedgender-neutral measurementprocedures.Until the mid-1990s, interviewersaskedwomen only whether they had been "keepinghouseor somethingelse?" The women who answered"keeping house" werecategorizedashousewives, and not unemployed.Because the womenwerenot asked,this occurredevenif the women had been seeking work. Oncewomen were askedthe samequestion as men, "Were you working or something else?"more women saidthey werenot working but doing "somethingelse"suchaslooking for work. This showsthe importanceof methodologicaldetailsin how governmentstatisticsget created. Researchers often use official statisticsfor internationalcomparisonsbut nationalgovernmentscollectdatadifferentlyand the quality of datacollectionvaries.For example,in 1994,the official unemployment rate reported for the United Stateswas7 percentlapan'swas2.9 percent,and France'swas 12percent.If the nations definedand gathereddatathe sameway,including discouragedworkersand involuntary parttime workersrates,the rateswould havebeen9.3 percentfor the United States,9.6 percentfor |apan,and I3.7 percentfor France.To evaluate the quality of official governmentstatistics,The Economistmagazineaskeda team of 20 leading statisticiansto evaluatethe statisticsof 13 nations basedon freedomfrom political interference, reliability, statistical methodology, and coverageoftopics. The top five nationsin order were Canada,Australia, Holland, France,and Sweden.The United Stateswas tied for sixth with Britain and Germany.The United States spentmore per persongatheringits statistics than all nationsexceptAustraliaand it released
2 44
P A RTT w o / c o N D U c rl N G QU A N Tl rA rl vER E S E A R cH
datathe fastest.The quality of U.S.statisticssuffered from being highly decentralized,having fewer statisticiansemployedthan any nation, and politically motivatedcutbackson the range of datacollected.l3 MissingData. One problem that plaguesresearcherswho use existingstatisticsand documents is that of missing data. Sometimes,the datawerecollectedbut havebeenlost.More frequently,the datawerenevercollected.The decision to collect official information is made The decisionto ask within governmentagencies. arelater made whose data questionson a survey group of rea is made by publicly available decidewhat who those In both cases, searchers. reanother collect what to collect may not research a in to address searcherneeds order startor stopcolquestion.Governmentagencies political, budgetary or for lecting information the early during other reasons.For example, federal U.S. the by 1980s,cost-cuttingmeasures inmuch of governmentstoppedthe collection found had researchers formation that social
first nameis one that canbe for eiAn androgynous ther a girl or boy without clearlymarkingthe child's gender.Somearguethat the feministmovementdecreasedgendermarkingin a child'snameas part of its broadersocietalinfluenceto reducegenderdistinctionsandinequality.Othersobservethat gender featureof nampredominant the single-most remains groupsor sowhen racial Even societies. most ing in cial classesinventdistinctivenew first names,the genderdistinctionsare retained. ex(2000) examined and colleagues Lieberson isting statisticaldata in the form of computerized of 1 1 millionbirths recordsfromthe birth certificates of Whitechildrenin the stateof lllinoisfrom 1 91 5 to first namesare 1 989. Theyfoundthat androgynous
valuable. Missing information is especiallya coverlong time periproblemwhen researchers ods.For instance,a researcherinterestedin the number of work stoppagesand strikes in the United Statescanobtain datafrom the 1890sto the present,exceptfor a five-yearperiod after 1911when the federalgovernmentdid not collect the data.(SeeBox 9.9 for an existingstatistics example.)
ISSUESOF INFERENCEAND THEORY TESTING
I
lnferencesfrom Nonreactive Data
ability to infer causalityor t€st a A researcher's theory on the basisof nonreactivedata is limto ited.It is difficult to useunobtrusivemeasures establishtemporal order and eliminatealternative explanations.In content analysis,a refrom the content to searchercannot generalize its effectson thosewho read the text' but can only usethe correlationlogic ofsurvey research
rare (about 3 percent)and that there has been a very slight historicaltrend toward androgyny,but onlyin veryrecentyears.ln addition,parentsgiveandrogynousnamesto girls more than to boys' and in namingis unstable(i.e.,a name gendersegregation tends to lose its androgynousmeaningover time). The authorsnoted that the way parentsnam€childrenmimicsa patternof collectivebehaviorfoundto operatein anotherresearcharea:theracialsegrega Changein residenceis untion of neighborhoods. less movementby the with races among equal dominantgroup;the lesspowerfulgroup movesto occupyareasthat the dominantgroup has aban doned;and integrationis unstable,with new segre aftersometime. gationreappearing
CHA P T ER 9 ,/ N ON R EA C T IVRESEARC E H A N D S E C O N D A R YA N A L Y S I S to show an association among variables. Unlike the easeofsurvey research,a researcherdoes not ask respondents direct questions to measure variables,but relies on the information available in the text.
EthicalConcerns Ethicalconcerns arenot attheforefrontof most nonreactiveresearchbecausethe peoplebeing studied are not directly involved. The primary ethicalconcernis the privacyand confidentiality of usinginformation gatheredby someoneelse. Another ethicalissueis that official statisticsare social and political products. Implicit theories andvalueassumptionsguidewhich information is collectedandthe categories usedwhengathering it. Measuresor statisticsthat are definedas official and collectedon a regularbasisare objectsof political conflict and guidethe direction of policy. By defining one measureas official, public policy is shapedtoward outcomesthat would be difFerentif an alternative,but equally valid, measurehad beenused.For example,the collectionof information on many socialconditions (e.g.,the number of patientswho died while in public mentalhospitals)wasstimulated by political activityduring the GreatDepression of the 1930s.Previously,the conditionswerenot defined as sufficiently important to warrant public attention. Likewise,information on the percentageof non-White studentsenrolled in U.S. schools at various agesis availableonly since 1953,and for specificnon-White races only sincethe 1970s.Earlier,such information wasnot salientfor public policy. Thecollectionof officialstatisticsstimulates new attentionto a problem,and public concern abouta problemstimulatesthe collectionof new officialstatistics.For example,drunk driving becamea biggerissueoncestatisticswerecollected on the number of automobileaccidentsand on whetheralcoholwasa factor in an accident. Political and socialvaluesinfluencedecisions about which existing statisticsto collect. Most officialstatisticsaredesignedfor top-down
245
bureaucraticor administrative planning purposes.They may not conform to a researcher's purposesor the purposesof peopleopposedto bureaucraticdecisionmakers.For example,a government agencymeasuresthe number of tons of steelproduced,miles of highwaypaved, and averagenumber of peoplein a household. Information on other conditionssuchasdrinking-waterqualiry time neededto commute to work, stressrelatedto a job, or number of children needingchild caremaynot be collectedbecauseofficials say it is unimportant. In many countries,the grossnational product (GNp) is treatedasa critical measureofsocietalprogress. But GNP ignoresnoneconomicaspectsof social life (e.g.,time spentplayingwith one'schildren) and qpes of work (e.g.,housework)that arenot paid. The information availablereflectsthe outcome of political debateand the valuesof officialswho decidewhich statisticsto collect.la
c oN c t u s t o N In this chapter, you have learned about several types of nonreactive research techniques. They are ways to measure or observe aspectsof social life without affecting those who are being studied. Theyresult in objective, numerical information that can be analyzed to address research questions. The techniques can be used in conjunction with other tlpes of quantitative or qualitative social research to address a large number of questions. As with any form of quantitative data, researchersneed to be concerned with measurement issues. It is easy to take available information from a past survey or government document, but what it measuresmay not be the construct of interest to the researcher. You should be aware of two potential probIems in nonreactive research.First, the availability of existing information restricts the questions that a researchercan address. Second, the nonreactive variables often have weaker validity because they do not measure the construct of
246
RESEARCH P A RTT wo / c o N D U c rl N G QU AN T l rA rl vE
secinterest. Although existing statisticsand tecnresearch low-cost are ondarydataanalysis and ;i.que;, the researcherlacks control over' collection a"o int oi, substantialknowledge This introducesa potentialsourceot erDrocess. 'rors about which researchers need to be especautious' ciallyvigilant and i" ire next chapter,we move from designing researchprojectsand collectingdata-toanatechniquesapply to the liins. datu.The'analysis aboutin the previlearned you Jrru"iituti". data seenhow to move have you far, o.r, .haptets.So measures' design-and research from a iopic, to a how to le-arn will you Next, to collectingdata. about you tell can they what look at dataand see question' a hypothesisor research
Ke y T e r m s accretion measures coding coding system content analYsis erosion measures fallacy of misplacedconcreteness General SocialSurveY(GSS) latent coding manifest coding nonreactive recording sheet StatisticalAbstract of the United States structured observation text unobtrusive measures
Endnotes (1981:7-11)' 1. SeeWebbandcolleagues po, a.n"itions of content analysis'seeHolsti i. -' (isas,sgz),Krippendorff (r980t2r-24)' -Yi:\"tr (1974:5-6),Stone and Weber and associates (1983,1984,1985:81' 1l' (1992),andWeber "9tt (1972)is a classicin this colleagues and W.iz-att e. typeofresearch.
(1984'1985) 4' StoneandWeber (L992)andWeber
techr"*-uti""a computerizedcontentanalysis niques. ofreli5' S". end."n (1981:58-66)for a discussion analysis content in uUifiy. Coai"g categorization in Holsti (1969:94-126)' is discussed in 6. A dir.or.io.t of socialindicatorscanbe found Duncan (1966)' Carley(1981).Also seeBauer (1981)'Land 098i:233-235),Justerand Land (1980)' Gilmartin itssz),and Rossiand yearbook are alsoproduced; 7' iufu"y "o"-n"glish itatistiches Iahrbuchfor the Federal .lu*pt", fo, de la n prrUti.'of Cermany'AnnuaireStatistique Aus-, for Australia Book Year Francefor France, Arsoversrgt' Ti Statiskisk Denmark's tralia, and of its yearbook iupunp.oa"..s an Englishversion of Handbook lapan' .uU.a tn. Statistical gov8. Guidesexistfor the publicationsofvarious example' the Guide to British ".rr-.ntr-for Australian OfficialPubCiuun*r"t publications, irotlonr, atd lrish Official Publications'Similar existfor mostnations' publications andStewart(1984) 9. 3eeChurchilt(1983:140-167) for listsof businessinformationsources' include 10. dift* *q"t u.S. archivesof surveydata University Center' Research ttr. Natio"a Opinion Univerof Chi.ugo, the SurveyResearchCenter' SciBehavioral. the C"alifornia-Berkeley; of sity Data Cincinnati; of encesLaboratory,University Wis-*d Ptogr"- tiLrary Service,University of of University Center' Roper the consin-iltadison; for ReInstitute the and Connecticut-Storrs; Universityof North Carsearchin SocialScience, Nathan olina-ChapelHill. Also seeKiecolt and (1992)' cel ( 1985)andPar oftheseissues,seeDaleand coll l . fo. a dlrcr'rsrion Maier(1991)'and Parcel i.*".. (1988:27-3t), givesa good-discussion 0;g2). Horn (1993:138) of misplacedconfallacy the of *ltt ."u-pt.s creteness. (1996). 12. SeeStevenson "The Good statisticsGuide" ;;. ;;; TheEconomisr, "The OverlookedHouselg93), lI, (September and "FewerDamned 5,1994), (February t..p*" Lies?"(March30' I996)' (1984)' 14. SeeBlock and Burns (1986)'Carr'Hill (1991)'and Maier (1993)' (1g73),Horn Hindess Van denBergandVanderVeer(1985)'
Analysisof Quantitative Data
Introduction Dealing with Data CodingData EnteringData CleaningData Resultswith One Variable Frequency Distributions Measures of CentralTendency Measures of Variation Results with Two Variables A BivariateRelationship Seeingthe Relationship: The Scattergram Tables Bivariate Measures of Association More Than Two Variables StatisticalControl The Elaboration Modelof Percentaged Tables MultipleRegression Analysis lnferential Statistics The Purposeof InferentialStatistics StatisticalSignificance Levelsof Significance Type I and Type ll Errors Conclusion
248
PART TW O /
CO NDUCTI NG Q UANT I T A T I V ER E S E A R C H
recordingsheets.However,it getscomplexwhen the dataarenot well organizedor not originally in developrules the form of numbers' Researchers Ifyou reada researchreport or articlebasedon attributes' variable to numbers certain to assign quantitative data,you will probably find charts' 1 and feas males codes researcher a For example, graphs,andtablesfirll ofnumbers.Do notbe inmissand variable of a 2.Eachcategory malesas ii-idut.d by them. A researcherprovides the is a codebook A a code. needs charts,graphs,andtablesto giveyou, the reader' ing information the describing pages) (i.e., more or one do=cument a condensedpicture ofthe data.The chartsand for of data location the and procedure tablesallow you to seethe evidencecollected' coding use' can computers that format a in When you collect your own quantitativedata, variables When you code data, it is essentialto creyoo *ill want to usesimilar techniquesto help ate a well-organized,detailedcodebookand you seewhat is insidethe data'You will needto make multiple copiesof it. If you do not write organizeand manipulatethe datasothey can redown the detailsof the coding procedure,or if veal things of interest.In this chapter,you will you misplacethe codebook,you havelostthe learn the fundamentalsof organizingand anakey to thi dataand may haveto recodethe data llzingqaantitative data. The analysisof quantiagain. tativedatais a complexfield of knowledge.This begin to think about a coding Researchers chaptercoversonly the basicstatisticalconcepts before they collect codebook and procedure to unand data-handlingtechniquesnecessary precodes researcher a survey example, iutu. Fot derstandsocialresearch. Precoding data. collecting before a questionnaire Data collectedusing the techniquesin the ttt."trr placingthe codecategories(e.g.,1 for past chaptersare in the form of numbers.The male,2 for female)on the questionnaire.'Some-' numberi representvalues of variables'which of subjects,respondents, times, to reducedependenceon a codebook, measurecharacteristics alsoplacethe location in the surveyresearchers or other cases.The numbersare in a raw form, on note pads,recordingsheets' computerformat the questionnaire'on questionnaires, doesnot precode,thefirst step ifa researcher reorganizethem into a or paper.Researchers is to createa codebook.He data after collecting form suitablefor computers,presentchartsor casean identification numgives each or shealso graphsto summarizetheir features,and intercases.Next, the researcher ofthe ber to keeptrack pret or givetheoreticalmeaningto the results. from eachquestioninformation transfersthe canread' computers that naireinto a format DEALING WITH DATA
ilffi--*
Coding Data
Entering Data
Beforea researcherexaminesquantitativedata to test hlpotheses,he or sheneedsto otganlze them in a different form' You encounteredthe ideaof codingdatain the lastchapter'Here,data codingmeanssystematicallyreotganizingraw numericaldatainto a format that is easyto anacreateand lyze using computers.Researchers infortransferring rules for apply consistently to another. form from one mation Codingcanbea simpleclericaltaskwhenthe dataarereiorded asnumberson well-organized
Most computer programsdesignedfor statistical analysisneedthe datain a grid format. In the grid' each row representsa respondent,subject,or specase.A columnor a setof columnsrepresents a column go from to possible is It cific variables. and row location(e.g.,row 7' column 5) backto the original sourceof dutu (e.g.,a questionnaire item on maritalstatusfor respondent8). codessurveydata For example,a researcher for computers format in a for threerespondents Peoplecannot 10.1. Figure in like that presented
CHAPT E R1 O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
F I G URE
24]9
Coded Data for Three Casesand Codebook
Exerpt from SurveyQuestionnaire Respondent lD
Interviewer Name
Notethe Respondent's Sex: _ Male _ Female .l The question first is aboutthe presidentof the UnitedStates.Do you StronglyAgree,Agree,Disagree, ' StronglyDisagree, or HaveNo Opinionaboutthe followingstatement: The President of the UnitedStatesis doinga greatjob. strongAgree Agree Disagree strong Disagree _ No opinion 2. Howold areyou? Excerpt ofCoded Data Column 000000000111111111122222222223333333333444... etc. (tens)
4s6?8e ry I I 6!!' glr34s6?8' 0113
01 212736302 18273827410239 18.8239+7+61 ... etc. 02 2133348211249881542124218.213984123... etc. 03 42012398211372726312345 17.361487645...etc. etc. Rawdatafor first threecases,columns1 through42. Excerpt from Codebook Column VariableName
1-2 3 4
ID BLANK Interviewer
Sex PresJob
Description Respondent identification number Interviewer who collectedthe data: I = Susan 2= X i a 3 : Juan 4 : Sophia 5 : Clarence Interviewer reportof respondent's sex ' l = Mal e,2= Femal e The presidentof the UnitedStatesis doinga greatjob. = 1 StronglyAgree 2 : Agree 3 : No Opinion 4 = Disagree 5 = StronglyDisagree Blank= missing information
250
PART TWO , / CO NDUCTI NG Q UANTI TA T I V ER E S E A R C H
easilyread it, and without the codebook,it ts worthless.It condensesanswersto 50 survey questionsfor three respondentsinto threelines or rows.The raw datafor manyresearchprojects look like this, except that there may be over 1,000rows, and the lines may be over 100 columns long. For example,a l5-minute telephonesurveyof250 studentsproducesa grid of datathat is 250rowsby 240columns. The codebookin Figure 10.1saysthat the first two numbers are identification numbers. Thus,the exampledataarefor the first (01)' secNotice ond (02), and third (03) respondents. usezeroesasplaceholdersto rethat researchers duceconfusionbetweenI and 01.The ls areaIwaysin column 2; the 10sarein column 1. The codebooksaysthat column 5 containsthe variable"sex":Cases1 and2 aremaleand Case3 is female.Column 4 tells us that Carlos interviewedCases1 and 2, and SophiaCase3. There are four waysto get raw quantitative datainto a computer:
then usea bar-codereaderto transferthe information into a comPuter. Cleaning Data
Accuracyis extremelyimportant when coding data.Errorsmadewhen codingor enteringdata into a computerthreatenthevalidityof measures and causemisleadingresults.A researcherwho has a perfectsample,perfectmeasures'and no errorsin gatheringdata,butwho makeserrorsin the coding processor in entering datainto a computer,canruin a wholeresearchproject. verAfter verycarefulcoding,the researcher "cleans" the data. or of coding, ifiesthe acctracy He or shemay codea 10 to 15 percentrandom sampleof the data a secondtime. If no coding errors appear,the researcherproceeds;ifhe or she finds errors' the researchbrrechecksall coding. When the data are in the computer' recode verify codingin two ways.Possible searchers checking involves checkireg) code cleaning(orwild the categoriesof all variables for impossible l. Codesheet.Gatherthe information, then codes.For example,respondentsexis coded1 = transferit from the original sourceonto a Male, 2 = Female.Finding a 4 for a casein the grid format (codesheet).Next, type what is field for the sexvariableindicatesa coding error. on the codesheetinto a computer,line by A second method, contingencycleaning(ot line. checking),involvescross-classifying consistency 2. Direct-entrymethod,includingCATL As intwo variablesand looking for logicallyimpossible formation is being collected,sit at a computer keyboardwhile listeningto/observing combinations.For example,educationis crossclassifiedby occupation. If a respondent is the information and enterthe information, recordedasneverhavingpassedthe eighthgrade enter the inor have a respondent/subject andalsois recordedasbeinga legitimatemedical formation himselfor herself.The computer checksfor a codingerror. doctor,the researcher must be preprogrammedto accept the dataafterthey are modifr A researchercan information. not usemore remay she in the computer.He or 3. Opticalscan.Gatherthe information, then when cotlecting were used fined categoriesthan enter it onto optical scansheets(or havea or group inmay combine the original data,but enter the information) respondent/subject may researcher the formation. For example, by filling in the correct"dots." Next, usean ordinal into five data group ratio-levelincome optical scanneror readerto transferthe inAlso,he or shecancombineinformacategories. formation into a comPuter. tion from severalindicators to create.a new 4. Bar code.Gathetthe information and convariable or add the responsesto severalquesvert it into different widths of bars that are associatedwith specificnumerical values, tionnaireitemsinto an index score.
CHAPTER1 O , / A N A L Y S I SO F Q U A N T I T A T I V E DATA
R ESU tTSWITH ONE V A RIA B IE FrequencyDistributions
251
Measuresof Central Tendency
Researchers often want to summarizethe infor_ mation aboutonevariableinto a singlenumber. The word statistics can mean a set of collected They usethreemeasuresof centraltlndencv, or (e.g.,numberstelling how many peo_ measures of the centerofthe frequencydistribu_ "."T!"ry ple live in a city) aswell as a 6ranch of apiUea tion: mean,median,and mode,which are often mathematicsusedto manipulate and summa_ calTed.averages (a lesspreciseand lessclearway rize the featuresof numbers.Socialresearchers of sayingthe same thing). Eachmeasureof cen_ useboth typesof statistics.Here,we focuson the tral.tendencygoeswith data having a specific secondt'?e-ways to manipulateand summa_ levelof measurement (seeTable10.i. rizenumbersthat representdatafrom a research The modeis the easiestto useand can be project. used with nominal, ordinal, interval, or ratio Descriptivestatisticsdescnbenumerical data. data. It is simply the most common or fre_ They can be categonzedby the number of vari_ quently occurring number. For example,the ablesinvolved: univariate,bivariate,or multi_ mode of the followinglist is 5: 6 5 7 t0 9-53 5. A variate (for one, two, and three or more distribution canhavemore than one mode.For variables).[Jnivariatestatisticsdescnbeone vari_ example,the modeof this list is b oth5 and,7:56 able (uni- refersto one;-variaterefersto vari_ I 2 5 7 4 7.If the list getslong, it is easyto spot able).The easiestway to describethe numerical the mode in a frequericy distributionij"" frof. dataof one variableis with a frequencydistribu_ for the most frequentscore.Therewill alwaysbe tion. It canbe usedwith nominJ-, ordinal_,in_ at leastone casewith a scorethat is equalto the terval-,or ratio-leveldataand takesmanyforms. mode. For example,I havedatafor 400 respondents.I The medianis the middle point. It is alsothe can summarizethe information on the gender 50th percentile, or the point at which half the of respondentsat a glancewith a raw co.rirto. u casesareaboveit andhalfbelowit. It canbe used frequencydistribution (seeFigure with ordinal-, interyal-, or ratio_leveldata (but l1-..."jug. 10.2).I can presentthe sameinformatioi in not nominal level).you can..eyeball', the mode, graphic form. Somecommon typesof graphic but computing a median requiresa little more representations arethe histogram,bar chart, and. work. The easiestway is firit to organizethe pie chart.Bar chartsor gtaphsare usedfor dis_ scoresfrom highest to lowest,then co-untto tire cretevariables.They canhavea verticalor hori_ middle. If thereis an odd number of scores,it is zontal orientation with a small spacebetween simple.Sevenpeople arewaiting for a bus;their the bars.The terminologyis not elact, but his_ agesare:121720 27 30 55 g0.Themedianageis togramsareusuallyupright bar graphsfor inter_ 27.Note that the mediandoesnot changeeaiily. val or ratio data. Ifthe 55-year-oldand the g0-year-old6oth got For interval-or ratio-leveldata,a researcher on one bus, and the remaining people wlre often grqups the information into categories. joined bytwo 31-year-olds, the rneaianremains The.groupedcategoriesshould be mutujly ex_ unchanged.Ifthere is an evennumber of scores, clusive.Interval- or ratio-level data are often thingsarea bit more complicated.For example, plotted tn afrequencypolygon.In it the number ut a bus stop hive the following aies: of casesor frequencyis along the vertical axis, :T^p^.9!t 1720 26 30 50 70.The medianis somewhe-re"be_ and the valuesofthe variableor scoresarealong tween 26 and 30. Compute the median by the horizontalaxis.A polryon appearswhen the adding the two middle scorestogetherand di_ dotsareconnected. viding by 2, or 26 + 30 = 5612=26.The median
252
PART TWO /
FIcURE 1 0.2
CO NDUCTI NC Q UANTI TA T I V ER E S E A R C H
of UnivariateStatistics Examples PercentageFrequencyDistribution Percentage Gender
RawCount FrequencyDistribution FrequencY Gender
2s% 7s% 100%
Male Female Total
100 300 400
Ma le Female Total
Bar Chart of SameInformation Males Females
Exampleof Grouped Data FrequencyDistribution N FirstJob Annual lncome
25 50 100 150 50 25 400
Under$5, 000 $5, 000t o $9, 999 $ 10, 000t o $. 15, 99 9 $ 15, 000t o $19, 99 9 $20,000to $29,999 $30.000 andover
Total Exampleof FrequencyPolygon Frequency 50 45 40 35 30 25 20 15 10
elc.
10
12
14
16
18
20
22
24
26
IndividualIncome(in Thousandsof Dollars)
28
etc
CHAP T E R1 O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
*:,
TABTE 1 0.1
MeasuresofCentral Tendencyand Levelsof Measurement
Nominal
Yes
Ordinal
Yes
Interval
Yes
253
mode the lowest. If most caseshave higher scoreswith a few extremelow scores,the mean will be thelowest,the medianin the middle,and the mode the highest.In general,the median is bestfor skeweddistributions,althoughthe mean is usedin mostotherstatistics(seeFigure10.3). Measuresof Variation
Measuresof centraltendenryarea one-number summary of a distribution; however,they give Yes Yes only its center.Anothercharacteristicof a distriRatio Yes Yes Yes bution is its spread,dispersion,or variability around the center.Two distributionscan have identicalmeasuresof centraltendencybutdif[er in their spreadabout the center.For example, sevenpeopleare at a bus stop in front ofa bar. ageis 28,eventhough no personis 28 yearsold. Theiragesare25 2627 3033 3435.BoththemeNote that thereis no mode in the list of six ages dian and the meanare30.At a bus stop in front becauseeachpersonhasa differentage. of an ice cream store,sevenpeoplehave the The mean,also calledthe arithmetic averidenticalmedianand mean,but their agesare 5 age,is the most widely usedmeasureof central 1020 30 40 50 55.Theagesofthe groupin front tendency.It can be ttsedonly with interval- or of the ice creamstorearespreadmore from the ratiolevel data.2Computethe meanby adding center,or the distribution hasmore variability. up all scores,then divide by the number of Variability has important socialimplicascores.For example,the mean agein the previtions. For example,in city X, the median and ous exampleis 17 + 20 + 26 + 30 + 50 + 70 = meanfamily income is $35,600per year,and it = 35.5.No onein thelist is 35.5years has zero variation. 213;21316 Zero yariatiorzmeansthat old, and the meandoesnot equalthe median. everyfamily has an income of exactly$35,600. The meanis stronglyaffectedby changesin CityYhasthesamemedianandmeanfamilyinextremevalues(verylargeor very small).For excome, but 95 percentof its familieshave inample,the 50- and 7}-year-oldleft and wererecomesof $12,000per yearand 5 percenthave placedwith two 31-year-olds.The distribution incomesof $300,000per year.CityX hasperfect now lookslike this: 1720 26 30 3t 31.The meincomeequality,whereasthereis greatiniqualdian is unchanged:28.The meanis IT + 20 + 26 ity in city Y. Aresearcherwho doesnot know the + 30 + 3l + 3I = 155;15516= 25.8.Thus,the variabilityof incomein thetwo citiesmissesvery meandroppeda greatdealwhen a few extreme important information. valueswereremoved. Researchers measurevariation in three If the frequenrydistribution forms a "norways:range,percentile,and standarddeviation. mal" or bell-shapedcurve,the threemeasuresof Rangeis the simplest.It consistsof the largest centraltendencyequaleachother.Ifthe distriband smallestscores.For example,the rangefor ution is a skeweddistribution(i.e.,more casesare the bus stop in front of the bar is from 25 to 35. in the upper or lower scores),then the threewill or 35 - 25 = I0 years.If the 35-year-oldgot not be equal.If most caseshavelower scores onto a bus and was replacedby a 60-year-old, with a few extremehigh scores,the meanwill be the rangewould changeto 60 - 25 = 45 years. the highest,the median in the middle, and the Rangehaslimitations.For example,herearetwo Yes
254
PART T W O , / CO NDUCTI NC Q UANTIT A T I V ER E S E A R C H
FIcURE 1 0.3
Measures ofCentralTendency Normal Distribution
Numberof Cases
Lowest
Valuesof Variables
Highest
SkewedDistributions ModeMedianMean
groupsof sixwith a rangeof 35years:303030 30 3065 and20 4546 48 50 55. Percentiles tell the scoreat a specificplace within the distribution. One percentileyou arreadylearnedis the median,the 50thpercentile. Sometimes the 25th and 75thpercentiles or the 10th and 90th percentilesareusedto describea distribution.For example,the 25th percentileis the scoreat which 25 percentof the distribution haveeitherthat scoreor a lower one.The computation of a percentilefollowsthe samelogic as the median.If I have100peopleandwantto find the 25th percentile.I rank the scoresand count up from the bottom until I reachnumber 25. If the total is not 100,I simply adjustthe distribution to a percentage basis. Standarddeviationis the most difficult to computemeasureof dispersion;it is also the mostcomprehensive andwidelyused.The range and percentilearefor ordinal-, interval-,and ratio-level data, but the standarddeviationrequiresan intervalor ratio levelof measurement.
MeanMedianMode
It is based on the mean and gives an "average distance" between all scoresand the mean. People rarely compute the standard deviation by hand for more than a handful of casesbecause computers and calculators can do it in seconds. Look at the calculation ofthe standard deviation in Figure 10.4.If you add up the absolute difference between eachscore and the mean (i.e., subtract each score from the mean), you get zero. This is becausethe mean is equally distant from all scores.Also notice that the scores that differ the most from the mean have the largest effect on the sum ofsquares and on the standard deviation. The standard deviation is used for comparison purposes. For example, the standard deviation for the schooling of parents of children in classA is 3.317years;for classB, it is 0.812;and for class C, it is 6.239. The standard deviation tells a researcherthat the parents ofchildren in classB are very similar, whereasthose for classC are very different. In fact, in classB, the schoo'-
CHAPT E R1 O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
FIGUREI 0.4
255
TheStandardDeviation
Stepsin Computing the StandardDeviation 1. Computethe mean. 2. Subtractthe meanfrom eachscore. 3. Squarethe resultingdifferencefor eachscore. 4. Totalup the squareddifferences to get the sumofsquares. 5. Dividethe sumof squaresby the numberof casesto get the variance. 5. Takethe squareroot ofthe variance, whichis the standarddeviation. Exampleof Computing the StandardDeviation variable= yearsof schooling] [8 respondents, Score
15 12 12 .t0 16 18 8 9
Score - Mean
Squared (Score - Mean)
15-12.5= 2.5 12-12.5=-0.5 12-12.5:-0.5 10-'12.5:-2.5 16-12.5: 3.5 .r8-12.5= 5.5 8-12.5= 4.5 9-12.5=-3.5
6.25 .25 .25 6.25 12.25 30.2s 20.25 12.25
Mea n='1 5+ 12 + 12+ l0+ 16+ 18+ 8+ 9 = 100,1OO / 8: 1 2 . 5 Sumof squares: 6.25 + .25 + .25 + 6.25 + 12.25+ 30.25 + 20.25 + 12.25= 88 .11 : Sumof squares/Number Variance of cases= 88/8 = = Standarddeviation Squareroot ofvariance= 11= 3.317years. Hereis the standarddeviationin the form of a formulawith svmbols. Svmbols: X = SCOREof case X = VEltt
I = Sigma(Greekletter)for sum,add together N= Numberofcases
Formulaj
= ttfx Standard deviation VN
-
-xf
aThereis a slightdifference in the formuladependingon whetherone is usingdatafor the population or a sampleto estimatethe populationparameter
ing of an "average" parent is less than a year above or below than the mean for all parents, so the parents are very homogeneous. In class C, however, the "average" parent is more than six years above or below the mean, so the parents are very heterogeneous.
The standard deviation and the mean are used to create z-scores.Z-scoreslet a researcher compare two or more distributions or groups. The z-score, also called a standardizedscore,expressespoints or scoreson a frequency distribution in terms of a number of standard deviations
pART Two
25 6
, / c oNDUc r lNc
Q UANTIT A T I v E R E S E A K L H
from the mean.Scoresarein termsof their relative position within a distribution, not as absolutevalues. For example,Katy, a salesmanagerin firm Mike in firm A, earns$50,000per year,whereas B earns$38,000per year. Despitethe absolute them,the managers incomedifferencesbetvveen are paid equally relativeto others in the same firm. Katy is paid more than two-thirds of other employeesin her firm, and Mike is also paid more than two-thirds of the employeesin his firm. Z-scores areeasyto calculatefrom the mean and standarddeviation(seeBox 10.1).For example, an employer interviews studentsfrom
I do not like the formulafor z-scores, Personally, is: which Deviation, Z-score= (Score- Mean)/Standard or in symbols: z=-
x-x 6
where:X= score,X= mean,E = standarddeviation I usuallyrelyon a simpleconceptualdiagramthat does the samething and that showswhat z-scores reallydo. Considerdata on the agesof schoolchildrenwith a meanof Z yearsand a standarddeviation of 2 years.Howdo I computethe z-scoreof 5-yearz-score old Miguel,or whatif I knowthat Yashohda's First,I in years? know her age to I need is a *2 and draw a little chart from -3 to *3 with zero in the middle.I will put the meanvalueat zero,becausea zscoreof zerois the meanand z-scoresmeasuredistanceaboveor belowit. I stop at 3 becausevirtually allcasesfallwithin3 standarddeviationsof the mean The chartlookslikethis: in mostsituations.
Kings Collegeand QueensCollege.Shelearns that the collegesare similar and that both grade on a 4.0scale.Yet,the meangrade-pointaverage at KingsCollegeis 2.62with a standarddeviation of .50,whereasthe meangrade-pointaverageat QueensCollegeis 3.24with a standarddeviation of .40. The employersuspectsthat gradesat QueensCollegeareinflated.Suzettefrom Kings Collegehas a grade-pointaverageof 3'62, and ]orgefrom QueensCollegehasa grade-pointaverageof 3.64. Both studentstook the same courses.The employerwantsto adjustthe grades for the gradingpracticesof the two colleges(i.e., createstandardizedscores).She calculateszscoresby subtractingeachstudent'sscorefrom
lllllll
-3
-2
-.1
0
+1 +2 +3
Now, I labelthe valuesof the meanand add or subtractstandarddeviationsfrom it. One standard deviationabovethe mean(+1) whenthe meanis 7 and standarddeviationis 2 yearsis just 7 -l 2, or 9 I put 3 years.Thisis because years.Fora -2 z-score, it is 2 standarddeviations,of 2 years each (or 4 years),lowerthan the Meanof 7. My diagramnow lookslikethis: 357 1 rl l l l l l
9
11
13
agei nY ears
-3 -2 -1 0 + 1 + 2 + 3 It is easyto seethat Miguel,who is 5 yearsold, z-scoreof hasa z-scoreof - 1 , whereasYashohda's 'l from zread *2 correspondsto 1 yearsold. I can suchqs For fractions, scoreto age,or ageto z-score. to fraction. just same the apply .5, I a z-scoreof 1 is a z-score of 1 2 age an Likewise, years. get 4 to age of *2.5.
CHAPTE R1 O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
the mean,then dividing by the standarddeviation. For example,Suzette'sz-scoreis 3.62 2.62 = 1.00/.50= 2, whereas|orge'sz-scoreis = 1. Thus,the employer 3.64 - 3.24.= .401.40 learnsthat Suzetteis two standarddeviations abovethe mean in her college,whereasforge rs only one standarddeviationabovethe meanfor his college.Although Suzette'sabsolutegradepoint averageis lowerthan forge's,relativeto the studentsin eachoftheir collegesSuzette's grades aremuch higherthan forge's.
R ESU LTS WITH TWO V A RIA B TE S A BivariateRelationship Univariate statistics describe a single variable in isolation. Bivariate statistics are much more valuable. Theylet a researcherconsider two variables together and describe the relationship between variables.Even simple hypothesesrequire two variables. Bivariate statistical analysisshows a relationship between variables-that is, things that appear together. Statistical relationships are based on two ideas: covariation and independence.Covariation means that things go together or are associated. To covary means to vary together; cases with certain values on one variable are likely to have certain values on the other one. For example, people with higher values on the income variable are likely to have higher values on the life expectancy variable. Likewise, those with lower incomes have lower life expectanry. This is usually stated in a shorthand way by saying that income and life expectancy are related to each other, or covary. We could also saythat knowing one's income tells us one's probable life expectancy, or that life expectancy depends on income. Independenceis the opposite of covariation. It means there is no association or no relationship between variables. If two variables are independent, cases with certain values on one variable do not have any particular value on the
257
other variable. For example, Rita wants to knorv whether number of siblings is related to life expectancy. If the variables are independent, then people with many brothers and sistershave the same life expectancy as those who are only children. In other words, knowing how many brothers or sisters someone has tells Rita nothine about the person's life expectancy. Most researchersstate hypotheses in terms ofa causal relationship or expected covariation; if they use the null hlpothesis, the hypothesis is that there is independence. It is used in formal hlpothesis testing and is frequently found in inferential statistics (to be discussed). Three techniques help researchersdecide whether a relationship exists between two variables:(1) a scattergram, or a graph or plot of the relationship; (2) cross-tabulation,or a percentaged table; and (3) measuresof associition, or statistical measures that expressthe amount of covariation by a single number (e.g.,correlation coefficient). Seeing the Relationship: The Scattergram What Is a Scattergram (or Scatterplot)? A scattergram is a graph on which a researcher plots each caseor observation, where each axis representsthe value ofone variable. It is used for variables measured at the interval or ratio level, rarely for ordinal variables, and never if either variable is nominal. There is no fixed rule frrr which variable (independent or dependent) to place on the horizontal or vertical axis, but usually the independent variable (syrnbolized by the letter X) goes on the horizontal axis and the dependent variable (syrnbol ized by I on the vertical axis. The lowest value for each should be the lower left corner and the highest value should be at the top or to the right. How to Construct a Scattergram. Begin with the range of the two variables. Draw an axis with the values of each variable marked and write numbers on each axis (graph paper is helpful).
pA RTT wo ,/ c o N D U c rN c e u A N rtrA Tl vE R E S E A R cH
258
Next, labeleachaxiswith the variablenameand put a title at the top. You are now ready for the data. For each case,find the valueof eachvariableand mark the graphat aplacecorrespondingto the two values. For example,a researchermakesa scattergram of yearsof schoolingby number of children.He or she looks at the first caseto seeyears of schooling(e.g.,12) and at the number of children (e.g.,3). Thenhe or shegoesto the placeon the graphwhere 12 for the "schooling"variable and 3 for the "number of children" variableintersectand puts a dot for the case. The scattergramin Figure 10.5is a plot of datafor 33 women.It showsa negativerelqtionshipbetweenthe yearsof educationthe woman completedand the number of childrenshegave birth to. Whnt Can You Learn from the Scattergram? A researchercan seethree aspectsof a bivariate relationshipin a scattergram:form, direction, and precision.
FIcURE 1 0.5
Form. Relationshipscan take three forms: independence,linear, and curvilinear. Independence or no relationshipis the easiestto see. It lookslike a randomscatterwith no pattern,or a straightline that is exactlyparallelto tlie horizontal or vertical axis. A linear "relationship meansthat a straightline canbe visualizedin the middle of a mazeof casesrunning from one corner to another.A curvilinearrelationshlpmeans would form a U that the centerof a mazeof cases curve, right side up or upside down, or an S curye. Direction. Linearrelationshipscanhavea positive or negativedirection.The plot of a positive relationshiplooks like a diagonalline from the lower left to the upper right. Higher valueson X tend to go with higher values on Y, and vice example versa.The income and life expectancF describeda positivelinear relationship. . lookslike aline from Anegativerelationship the upper left to the lower right. It meansthat highervalueson onevariablego with lowerval-
Exampleof a Scattergram: Yearsof Educationby Number of Natural Children Jor 33 Women
6.00
tr
E = E o
5.00
\
4.00
\
I
(! 5 t5
z
3.00
o o lt 2.OO
\ \
E J
z 1 . 00
\ I-
\
0.00 8.00
18.00 1 2 .0 0 1 6 .0 0 1 4 .0 0 Education Years of Formal
22.0O
CHAPTER1O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
259
Figure 10.6 is a raw count or frequency table.Its cellscontain a count of the cases.It is easyto make,but interpretinga raw count table is difficult becausethe rows or columnscanhave differenttotals.and what is of realinterestis the relativesizeof cellscomparedto others. Researchers convert raw count tablesinto percentaged tablesto seebivariaterelationships. There are three waysto percentagea table: by row, by column, and for the total. The first two areoften usedand showrelationships. Is it best to percentageby row or column? Eithercanbe appropriate.Let us first reviewthe mechanicsof percentaginga table.When calcucomputethe perlating column percentages, Precision. Bivariaterelationshipsdiffer in their centageeach cell is of the column total. This degreeof precision.Precisionis the amount of includesthe total column or marginal for the spreadin the pointson the graph.A high levelof column variable.For example,the first column precision occurswhen the points hug the line total is 26 (there are26 peopleunder age30), that summarizesthe relationship.Alowlevel ocand the first cell of that column is 20 (thereare curs when the points are widely spreadaround 20 peopleunder age30 who agree).The percan "eyeball"a highly prethe line. Researchers cise relationship. They can also use advanced centageis20126=0.769or76.9percent.Or, for thefirst numberin themarginal,37ll0l = 0.366 statisticsto measurethe precisionof a relationship in a way that is analogousto the standard = 36.6 percent (seeTable 10.2). Except for rounding,the total shouldequal100percent. deviationfor univariatestatistics. Computing row percentagesis similar. Compute the percentageof eachcell as a perBivariate Tables centageof the row total. For example,usingthe samecell with 20 in it, we now want to know What Is a Bivariate Table? Thebivariateconwhat percentageit is of the row total of 37, or tingency table is widely used. It presentsthe by 20137= 0.541= 54.1percent.Percentaging sameinformation as a scattergramin a more for a row or column givesdifferentpercentages condensedform. The data can be measuredat cellunlessthe marginalsarethe same. althoughintervaland anylwel of measurement, let a reThe row and column percentages ratio datamust be groupedif therearemanydifferentvalues.The tableis basedon cross-tabula- searcheraddressdifferent questions.The row percentagetable answersthe question.Among tion; that is, the casesare organizedin the table those who hold an attitude, what percentage on the basisof two variablesat the sametime. tableis formedby cross-tabu- come from eachagegroup?It saysof responA contingency dentswho agree,54.Ipercentare in the underlating two or more variables.It is contingentbe30 age group. The column percentagetable causethe casesin eachcategoryofa variableget addresses the question:Among thosein eachage distributed into each categoryof a second(or hold differentattitudes? additional)variable.The table distributescases group,what percentage It saysthat amongthosewho areunder 30,76.9 into the categoriesof multiple variablesat the a reby category percentagree.From the row percentages, sametime and showshow the cases, ofone variable,are"contingentupon" the cate- searcherlearnsthat a little overhalf ofthosewho agreeareunder 30 yearsold, whereasfrom colgoriesof other variables.
ueson the other.For example,peoplewith more educationarelesslikely to havebeenarrested.If we look at a scattergramofdata on a group of maleswhereyearsof schooling(X axis)areplotted by number of arrests(Y axis)'we seethat most cases(or men) with manyarrestsarein the lower right, becausemost of them completed few yearsof school.Most caseswith few arrests mosthavehad more arein the upperleft because schooling.The imaginaryline for the relationship can havea shallowor a steepslope.More advancedstatisticsprovide precisenumerical measuresof the line's slope.
260
PART TW O , / CO NDUCTI NG Q UANTIT A T I V ER E S E A R C H
Age Group by Attitude about Changingthe Drinking FlG U RE I 0.6 Age, RawCount Table Age Group (b)
RawCount Table (a) Attitude (b)
Under 30
Agree No opinion Disagree Total (c)
20 3 (d) 3 26
Missingcases(f ) = 8.
30-45
46-60
10 10 _l
4 10 2L
2sfrt
61 and Older
3 2 10 15
Total(c)
37 25 3g
101
(")
The Parts of a Table information' and providesbackground (a) Civeeachtablea fifle whichnamesvariables (b) Labelthe rowandcolumnvariableandgivea nameto eachof the variablecategories. (c) Includethe totalsof the columnsand rows.Thesearecalledthe marginab.rhey frequencydistributionfor the variable. equalthe univariate to the intersectionofa categoryfor each (d) Eachnumberor placethat corresponds variableis a cell of a table. (e) The numberswith the labeledvariablecategoriesand the totalsarecalledthe body of a table. (f ) lf there is missinginformation(casesin whicha respondentrefusedto answer, said"don't know,"etc.),report the numberof missingcasesnear endedinterview, the tableto accountfor all originalcases.
the researcherlearns that umn percentages, amongthe under-30people,overthree-quarters agree.One way of percentagingtells about people who have specificattitudes;the other telis aboutpeoplein specificagegroups. may imply lookh1'pothesis A researcher's or the column percenting at row percentages ages.When beginning,calculatepercentages eachway and practiceinterpreting,or figuring out, what eachsays.For example,myhypothesis is that ageaffectsattitude, so column percentagesare most helpful. However,if my interest wasin describingthe agemake-upof groupsof peoplewith different attitudes,then row perareappropriate. centages
Unfortunately, there is.no "industry standard" for putting independentand dependent tableasrow or column, variablesin a percentage by row and column.A majoror for percentage placethe independentvariable ity ofresearchers by column,but a asthe column and percentage large minority put the independentvariableas bYrow. the row and percentage Reailing a PercentagedTable. Once you uhderstandhow a tableis made,readingit and figuring out what it saysaremuch easier.To reada table, first look at the title, the variablelabels, and anybackgroundinformation. Next, look at the direction in which percentageshave been
CHAPTER1 O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
computed-in rows or columns. Notice that the percentaged tables in Table 10.2 have the same title. This is becausethe same variables are used. It would have helped to note how the data were percentaged in the title, but this is rarely done. Sometimes, researcherspresent abbreviated tables and omit the 100 percent total or the marginals, which adds to the confusion. It is best to include all the parts of a table and clear labels. Researchers read percentaged tables to make comparisons. Comparisons are made in the opposite direction from that in which per-
TA BLE 1 0.2
centagesare computed. A rule of thumb is to compare across rows if the table is percentaged down (i.e., by column) and to compare up and down in columns if the table is percentaged across(i.e.,by row). For example, in row-percentaged Table 10.2,compare columns or age groups. Most of those who agreeare in the youngest group, with the proportion declining as age increases.Most no-opinion people are in the middle-age groups, whereasthose who disagreeare older, especiJly in the 46-to-60 group. When reading column-
Age Groupby Attitude about Changingthe DrinkingAge, PercentagedTables
T
Column-PercentagedTable
€roup .,. ', 3'Ar:45
45-60
Agree
76.9%
407"
11.4%
20%
36.6%
No opinion
I 1.5
40
28.6
13.3
24.8
Disagr:ee
I 1.5
60
66.7
3 8.6
Total
99.9
20 .100
100
r0 0 (t s )-
(N)
(2 s)-
(26).
( 3s) -
Missingcases= 8
100 (10r)*
Row-PercentagedTable
,ii;.
Agree
54.1%
27%
10.8%
8.1% 100%
(37).
No opinion
12
40
40
8
( 2s) .
12.8
53.8
2s.6
99.9
2 4 .8
34.7
14.9
' r0 0 . r
Disagree Total
7.7 25.7
261
100
(39)-
(1 0 1 )-
Missingcases= 8 *Forpercentaged tables,providethe numberof casesor N on whichpercentages arecomputedin parenthesesnearthe total of I 00%. Thismakesit possibleto go backandforth from a percentaged tableto a rawcounttableandviceversa.
2 62
pA RTT wo / c o N D U c l N c e u AN T trATtvER E S E A R C H
percentagedTable 10.2, compare across rows. For example, a majority of the youngest group agree,and they are the only group in which most people agree.Only 11.5 percent disagree,compared to a majority in the two oldest groups. It takes practice to see a relationship in a percentagedtable. Ifthere is no relationship in a table, the cell percentages look approximately equal acrossrows or columns. A linear relationship looks like larger percentagesin the diagonal cells. If there is a curvilinear relationship, the largest percentages form a pattern across cells. For example, the largest cells might be the upper right, the bottom middle, and the upper left. It is easiestto seea relationship in a moderate-sized table (9 to t6 cells) where most cells have some cases(at least five casesare recommended) and the relationship is strong and precise. Principles ofreading a scattergram can help you see a relationship in a percentaged table. Imagine a scattergramthat has been divided into 12 equal-sizedsections.The casesin each section correspondto the number of casesin the cellsof a table that is superimposed onto the scattergram. The table is a condensed form of the scattergram. The bivariate relationship line in a scattergram corresponds to the diagonal cells in a percentagedtable. Thus, a simple way to see strong relationships is to circle the largest percentage in each row (for row-percentaged tables) or column (for column-percentagedtables) and seeifa line appears. The circle-the-largest-cell rule works-with one important caveat.The categoriesin the percentagestable mustbe ordinal or interval and in the same order as in a scattergram. In scattergrams the lowest variable categoriesbegin at the bottom left. If the categories in a table are not ordered the same way, the rule does not work. For example, Table 10.3a looks like a positive relationship and Table 10.3b like a negative relationship. Both usethe same data and are percentagedby row. The actual relationship is negative. Look closely-Table 10.3b has age categoriesordered as in a scattergram. When in doubt. return to the basic difference between
TA B TE 10.3a
30-45
s% 25 . r5 25
46-60
35
45
61 +
45
35
Under30
TA B LE 1 O.3b
40
100
20
100
12 .t5
8
100
5
100
A gebyS chool i ng
.t5
(
100
45
12
8
100
15
25
40
20
100
5
25
30
40
100
61 +
45% 35
46-60
35
30-45 Under30
30 40
positive and negativerelationships. A positive relationship means that as one variable increases, so does the other. A negative relationship means that asone variable increases,the other decreases. Bivariate Tables without Percentages. Researcherscondenseinformation in another kind of bivariate table with a measure of central tendency (usually the mean) instead of percentages. It is used when one variable is nominal or ordinal and another is measured at the interval or ratio level.The mean (or a similar measure)of the interval or ratio variable is presented for each categoryof the nominal or ordinal variable. All casesare divided into the ordinal or nominal variable categories;then the mean is calculated for the casesin each variable category from the raw data. Table 10.4showsthe mean ageof people in each ofthe attitude categories.The results sug-
c H A p rE R | 0 ,/ A N A LystsoF euA N TtrA TtvED A TA
TABLE 1 0.4
AttitudeaboutChanging the DrinkingAge by Mean Age of Respondent
Agree
26.2
No opinion
44.5
(s7) (2s)
263
dicating a negative relationship and positive numbers a positive relationship. A measure of 1.0 means a 100 percent reduction in errors, or perfect prediction.
MO RET HA N T WO V A RI A B L E S StatisticalControl
Showing an association or relationship between two variables is not sufficient to say that an independent variable causesa dependent variabie. Missingcases= 8 In addition to temporal order and association, a researcher must eliminate alternative explanations-explanations that can make the hypothegestthat the mean ageof thosewho disagreeis sized relationship spurious. Experimental much higher than for thosewho agreeor have researchersdo this by choosing a researchdesign no opinion. that physically controls potential alternative explanations for results (i.e., that threaten internal validity). Measuresof Association In nonexperimental research, a researcher A measureof association is a singlenumber that controls for alternative explanations with statisexpresses the strength,and often the direction, tics. He or she measurespossible alternative exof a relationship. It condensesinformation planations vmth control variables,then examines about a bivariate relationship into a single the control variableswith multivariate tablesand number. statistics that help him or her decide whether a There are many measuresof association. bivariate relationship is spurious. They also The correct one dependson the level of meashow the relative size of the effect of multiple insurement.Many measures arecalledby lettersof dependent variables on a dependent variable. the Greek alphabet.Lambda,gamma,tau, chi A researcher controls for alternative expla(squared),and rho are commonly usedmeanations in multivariate (more than two varisures.The emphasishere is on interpretingthe ables) analysis by introducing a third (or measures,not on their calculation.In order to sometimes a fourth or fifth) variable. For examunderstandeachmeasure,you will needto comple, a bivariate table shows that taller teenagers pletea beginningstatisticscourse. like sports more than shorter ones do. But-the If there is a strong associationor relationbivariate relationship between height and attiship,then few errorsaremadepredictinga sec- tude toward sports may be spurious because ond variableon the basisof knowledgeof the teenagemales are taller than females, and males first, or the proportion oferrors reducedis large. tend to like sports more than females. To test A largenumber ofcorrect guesses suggests that whether the relationship is actually due to sex, a the measureof associationis a nonzeronumber researcher must control for gen'der; in other if an associationexistsbetweenthe variables. words, effects of sex are statistically remoyed. Table10.5describesfive commonlyusedbivariOnce this is done, a researchercan seewhether ate measuresof association. Notice that most the bivariate relationship between height and alrangefrom - 1to +1, with negativenumbersintitude toward sports remains. Disagree
61.9
(se)
264
PART TW O , / CO NDUCTI NC Q UANTI T A T I V ER E S E A R C H
TABLE I 0.5
FiveMeasuresofAssociation
Lambda is usedfor nominal-level data.lt is based on a reductionin errorsbasedon the modeand rangesbetween0 (independence) and 1.0 (perfectpredictionor the strongestpossible relationship). 6ammais usedfor ordinal-level data.lt is basedon pairsof variablecategories comparing andseeing whethera casehasthe samerankon each.Camma rangesfrom - I .0 to *1 .0, with O meaningno association. Tauis alsousedfor ordinal-level data.lt is based on a differentapproachthan gammaand takes careof a few problemsthat canoccurwith gamma. Actually,there are severalstatisticsnamedtau (it is a popularCreekletter),andthe one hereis Kendall's tau. Kendall's tau rangesfrom - I .0 to *1 .0, with 0 meaningno association. productmoment Rhois alsocalledPearson's correlationcoefficient(namedafterthe famous statisticianKarlPearsonand basedon a oroduct procedure). momentstatistical lt is the most commonlyusedmeasure of correlation, the correlationstatisticpeoplemeanif they usethe termcorrelation without identifyingit further.lt can
be usedonly for data measuredat the intervalor ratio level.Rhois usedfor the meanand standard andtellshowfar cases deviationof the variables arefrom a relationship(or regression)linein a scatterplot.Rhorangesfrom - 1.0 to *1 .0, with 0 lf the valueof rho is meaningno association. it hasa calledR-squared, sometimes squared, Runiqueproportionreductionin errormeaning. in onevariable squaredtellshowthe percentage (e.g.,the dependent)is accountedfor, or by,the othervariable(e.g.,the explained linearrelationships Rhomeasures independent). nonlinear or curvilnear measure lt cannot only. Forexample, a rho of zerocan relationships. or a curvilinear indicateeitherno relationship relationship. hastwo differentuses.lt can be used Chi-squared as a measureof associationin descriptivestatistics likethe otherslistedhere,or in inferential statistics.Inferentialstatisticsare brieflydescribed chi-squared can of association, next.As a measure be usedfor nominalandordinaldata.lt hasan upperlimitof infinityanda lowerlimitof zero, meaning no association.
Summaryof Measuresof Association ',:';L;,;.:':;:"ltjg$',' , .' '' l::|i::*
Lambda
t.
Camma
I
Nominal
.l
,::;:',::
.0
Tau (Kendall's)
t
Ordinal
Rho
p
Interval,ratio
+ 1 . 0 , -1 . 0 +t .0,-.1.0 + 1 . 0-r. , 0
Chi-square
x2
Nominal,ordinal
Infinity
Ordinal
A researcher controlsfor a third variableby seeingwhetherthebivariaterelationshippersists within categories of the control variable.For example,a researcher controlsfor sex,and the relationshipbetweenheight and sportsattitude
0 0
0
0 0
persists.This meansthat tall malesand tall femalesboth like sportsmore than short males and short femalesdo. In other words, the control variablehasno effect.When this is so,thebivariaterelationshipis not spurious.
CHAPTE R1 O , i A N A L Y S I SO F Q U A N T I T A T I V E DATA
I il
If the bivariaterelationshipweakensor dis_ appearsafter the control variableis considered, it meansthat tall malesareno more likelv than shortmalesto like sports,and tall femalesur. rro more likely to like sportsthan short females.It indicatesthat the initial bivariaterelationshipis spuriousand suggests that thethird variable.sex. and not height,is the truecauseof differences in attitudestoward sports. Statisticalcontrol is a key ideain advanced statisticaltechniques. A meaiureof association like the correlationcoefficientonlysuggests a re_ lationship.Until a researcherconsiderscontrol variables,the bivariate relationship could be spurious.Researchers are cautiousin interpret_ ing bivariaterelationshipsuntil they havecon_ sideredcontrolvariables.
265
in thepartial.Thus,it is possibleto breakapart a bivariatetable to form partials,or combinethe partialsto restorethe initial bivariatetable. Trivariate tables have three limitations. First, they are difficult to interpret if a control variablehasmore than four catigories.Second, control variablescanbe at any levelof measure_ ment, but interval or ratio control variables must be grouped (i.e., convertedto an ordinal level),and how casesare groupedcan affectthe interpretationof effects.Finaliy,the total num_ ber of casesis a limiting factor tecausethe cases aredividedamongcellsin partials.The number of cellsin the partialsequalithe number of cells in the bivariate relationshipmultiplied by the numberof categories in the control variable.For example,a control variablehasthreecategories, and a bivariatetablehas 12 cells,so the p"artials have3 X 12 = 36 cells.An averageoffive cases The Elaboration Model of per cell is recommended,so the iesearcherwill PercentagedTables need5 X 36 = lg0 cases at minimum. Constructing Triyariate Tables. In order to For threevariables,threebivariatetablesare meetall the conditionsneededfor causality,re_ logicallypossible.In the example,the combina_ searcherswant to "control for" or seewhether tio3s (l)gender byattitudi e) agegroup ge by aRaltemativeexplanationexplainsawayacausal attitude,and (3) gender by ageg.orrp.tt. iur_ relationship. If an alternativeexplanationextialsaresetup on the basisofihe initiul birruriat" plainsa relationship,then the bivariaterelationrelationship.The independentvariablein each ship is spurious. Alternative explanationsare is "agegroup" andthe dependentvariableis ..at_ operationalizedas third variables,which are titude." "Gender" is the control variable.Thus, caTled controlyariablesbecausethey control for the trivariatetable consistsofa pair ofpartials, alternativeexplanation. eachshowingthe agelattitudereiationshipfor a One way to take such third variablesinto givengender. considerationandseewhethertheyinfluencethe A researcher's theory suggests the hypothe_ bivariaterelationshipis to statisticallyintroduce sisin the initial bivariaterelationship;it alio tells control variablesusing trivariate or three-vari- him or her which variablesprovide alternative abletables.Trivariatetablesdiffer slightly from explanations(i.e.,the control variables).Thus. bivariatetables;theyconsistof multiple bivariate the choiceof the control variableis based.on tables. theory. A trivariatetablehasa bivariatetableof the _Theelaborationparadigm is a systemfor independentand dependentvariable for each reading percentagedtrivariate tables.3It de_ categoryofthe control variable.Thesenew tascribesthe pattern that emergeswhen a control bles are calledpartials.The number of partials variableis introduced. Five terms describehow dependson the numberof categories in the conthe partial tablescompareto the initial bivariate trol variable.Partialtableslook like bivariatetatable,or how the original bivariaterelationship bles,but theyusea subsetof the cases. Onlycases changesafter the control variableis considered. with a specificvalueon the control variableare The examplesof patternspresentedhere show
266
pA RTTw o ,/ c o N D U c l N G e u AN TtrA Tl vER E S E A R cH
strong cases.More advanced statistics are neededwhen the differencesarenot asobvious. The replicationpattern is the easiestto understand.It is when the partialsreplicateor reproducethe samerelationshipthat existedin the bivariatetable beforeconsideringthe control variable.It meansthat the control variablehas no effect. pattern is the next easiest The specification pattern.It occurswhen onepartial replicatesthe initial bivariaterelationshipbut otherpartialsdo not. For example,you find a strong (negative) bivariaterelationshipbetweenautomobileaccidentsand collegegrades.You control for gender and discoverthat the relationshipholdsonly for males(i.e.,the strongnegativerelationshipwas in the partial for males,but not for females). This is specificationbecausea researchercan specifr the categoryof the control variable in which the initial relationshippersists. The control variablehas a large impact in both the interpretation and explanation patterns.In both, the bivariatetable showsa rela' tionship that disappearsin the partials.In other words, the relationshipappearsto be independencein the partials.Thetwo patternscannotbe distinguishedby looking at the tablesalone.The differencebetweenthem dependson the location ofthe control variablein the causalorder of variables.Theoretically,a controlvariablecanbe in one of two places,eitherbetweenthe original independentand dependentvariables(i.e., the control variableis intervening),or beforethe original independentvariable. patterndescribes the situThe interpretation ation in which the control variable intervenes betweenthe original independentand dependentvariables.For example,you examinea relationship between religious upbringing and abortion attitude.Politicalideologyis a control variable.You reasonthat religious upbringing affectscurrent political ideologyand abortion attitude.You theorizethat political ideologyis logicallyprior to an attitude about a specificissue,like abortion.Thus, religiousupbringing causespolitical ideology,which in turn has an
impact on abortion attitude. The control variableis an interveningvariable,which helpsyou interpretthe meaningof the completerelationship. the sameasinTheexplanationpatternlooks terpretation.Thedifferenceis the temporalorder ofthe control variable.In this pattern,a control variablecomesbeforethe independentvariable in the initial bivariaterelationship.For example, the originalrelationshipis betweenreligiousupbringing and abortion attitude,but now gender is the controlvariable.Gendercomesbeforereltgious upbringing becauseone's sex is fixed at birth. The explanationpatternchangeshow a researcherexplainsthe results.It implies that the initial bivariaterelationshipis spurious. variablepattern occurswhen The suppressor the bivariatetablessuggestindependencebut a relationshipappearsin one or both of the partials. For example,religious upbringing ,and abortion attitude areindependentin a bivariate table. Once the control variable"region of the country'' is introduced,religiousupbringing is with abortion attitude in the partial associated tables.The control variableis a suppressorvarithe true relationship. ablebecauseit suppressed The true relationshipappearsin the partials. (SeeTable 10.6for a summary of the elaboration paradigm.) Multiple RegressionAnalysis Multiple regressionis a statistical technique whose calculation is beyond the level in this book. Although it is quickly computedby the appropriatestatisticssoftware,a backgroundin statisticsis neededto preventmaking errors in its calculationand interpretation.It requiresinherefor terval-or ratio-leveldata.It is discussed two reasons.First, it controlsfor many alternative explanationsand variablessimultaneously (it is rarelypossibleto usemorethan onecontrol tables).Secvariableat a time usingpercentaged ond, it is widely usedin sociology,and you are likely to encounterit when readingresearchreports or articles.
CHA P T E RI O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
267
Summaryof the Elaborationparadigm
Replication Specification Interpretation
Samerelationship in both partialsas in bivariate table. Bivariate relationship is only seenin one of the partialtables. Bivariaterelationshipweakensgreatryor disappearsin the partiar tabres(control variableis intervening). Bivariaterelationshipweakensgreallyor disappearsin the partial tables(control variableis beforeindependent variable). No bivariaterelationship; relationship appearsin partialtablesonly.
Explanation Suppressor variable
EXAMPLES OF EI.ABOMTIONPATTERNS
Replication BivariateTable
Low High
Low
High
85% 15%
15% 85%
Control= Low Low Low High
84% 167
Partials High
Control= High Low High
15% 84%
86% 14%
14% 86%
Interpretation or Explanation BivariateTable
Partials Control= Low Low
High Low High
85% 1s%
15% 85%
Low High
4s%
ss%
Control= High Low High
Hish
55% s5% 45% 4s%
45%
s5%
Specification BivariateTable
Low High
Partials
Low
High
85% 15%
85% 15%
Low High
Control= Low Low
High
Control= High Low High
957" s%
s%
so%
95%
50%
s0% so%
SuppressorVariable BivariateTable
Low High
Partials
Low
Hish
54% 46%
46% s4%
Low High
Control= Low Low
High
Control= High Low High
84% 16%
16% 84%
14% 86%
86% 14%
268
pA RTTw o / c o N D U c l N c e u A N TtrA TtvER E S E A R C H
Multiple regressionresults tell the reader two things. First, the results have a measure calledR-squared(R2),which tellshow well a set of variables explains a dependent variable. Explain meansreducederrors when predicting the dependentvariablescoreson the basisofinformation about the independentvariables.A good model with severalindependentvariables might accountfor, or explain,a largepercentage of variation in a dependentvariable.For example, an R2 of .50 meansthat knowing the independent and control variablesimproves the accuraryofpredictingthe dependentvariableby 50 percent,or half as many errors are made as would be made without knowing about the variables. Second,the regressionresultsmeasurethe direction and sizeof the effectof eachvariable on a dependentvariable.The effectis measured preciselyandgivena numericalvalue.For example,a researcher canseehow five independentor control variablessimultaneouslyaffecta dependent variable,with all variablescontrolling for the effectsof oneanother.This is especiallyvaluablefor testingtheoriesthat statethat multiple independentvariablescauseone dependent variable. The effecton the dependentvariableis measuredby a standardizedregressioncoefficientor the Greekletter beta (B). It is similar to a correlation coefticient.In fact,the betacoefficientfor two variablesequalsthe r correlationcoefficient. Researchers use the beta regressioncoefficient to determine whether control variables havean effect.For example,the bivariatecorrelation betweenX and Y is .75. Next, the researcherstatisticallyconsidersfour control variables.If the betaremainsat .75,thenthe four control variableshaveno effect.However,if the betaforX and Ygetssmaller(e.g.,dropsto .20), it indicatesthat the control variableshave an effect. Consideran exampleof regressionanalysis with age,income,education,and regionasindependentvariables.The dependentvariableis a
TABTE 10.7
ExampleofMultiple RegressionResults
Dependent Variablels PoliticalldeoloryIndex (HighScoreMeansVeryLiberal)
Region: South
-.19
Ate
.0.1 -.44
lncome Yearsof education Religiousattendance
.23 -.39
R 2= .39
scoreon a political ideologyindex.The mdltiple regressionresults show that income and religious attendancehave large effects,education and region minor efFects, and ageno effect.All the independentvariablestogether have a 38 percentacc'lracyin predictinga person'spolitical ideology(seeTable 10.7).The examplesuggests that high income, frequent religious attendance,and a southernresidenceare positively associatedwith conservativeopinions, whereashaving more education is associated with liberal opinions. The impact of income is more than twice the sizeof the impact of living in a southernregion.We havebeen examining descriptivestatistics(seeTable 10.8);next, we look at a different,gpe:inferentialstatics.
I NFERENTIALSTATISTICS The Purpose of Inferential Statistics Researchers often want to do more than describe; they want to test hlpotheses, know whether sampleresultshold true in a popula-
C H A P T E R'I O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
? dex
hiple relittion r All a38 olitisugfous losiions, ated neis ving ning ;we
deDOW
ula-
269
Summaryof Major Typesof DescriptiveStatistics
Univariate
Frequency distribution, measure of centraltendency,standard deviation, z-score
Describe one variable.
Bivariate
Correlation,percentagetable, chi-square
Describe a relationship or the association betweentwo variables.
Multivariate
Elaboration paradigm, multiple regression
Describe relationships amongseveral variables, or seehow several independentvariableshavean effect on a dependentvariable.
tion, and decide whether differences in results (e.g., between the mean scoresof two groups) are big enough to indicate that a relationship really exists. Inferential statistics use probability theory to test hypothesesformally, permit inferencesfrom a sample to a population, and test whether descriptive results are likely to be due to random factors or to a real relationship. This section explains the basic ideas of inferential statisticsbut does not deal with inferential statistics in any detail. This area is more complex than descriptive statisticsand requires a background in statistics. Inferential statistics rely on principles from probability sampling, where a researcher usesa random process(e.g.,a random number table) to select casesfrom the entire population. Inferential statistics are a precise way to talk about how confident a researchercan be when inferring from the results in a sample to the population. You have already encountered inferential statistics ifyou have read or heard about "statistical significance" or results "significant at the .05 level." Researchersuse them to conduct various statisticaltests (e.g., a t-test or an F-test). Statistical significance is also used in formal hypothesis testing, which is a preciseway to decide whether to accept or to reject a null hlpothesis.a
Statistical Signifi cance Statistical significancemeans that results are not likely to be due to chance factors. It indicates the probability of finding a relationship in the sample when there is none in the population. Because probability samples involve a random process, it is always possible that sample results will differ from a population parameter. A researcherwants to estimate the odds that sample results are due to a true population parameter or to chance factors of random sampling. Statistical significance uses probability theory and specific statistical tests to tell a researcher whether the results(e.g.,an association,a difference between two means, a regression coefficient) are produced by random error in random sampling. Statistical significance only tells what rs likely. It cannot prove anlthing with absolute certainty. It states that particular outcomes are more or less probable. Statistical significance is notthe same aspractical, substantive,or theoretical significance. Results can be statistically significant but theoretically meaningless or trivial. For example,two variablescan have a statistica\ significant association due to coincidence, with no logical connection between them (e.g.,length of fingernails and ability to speakFrench).
27O
pA RTTw o / c o N D U c rtN c QU AN TITA TIvE R E S E A R cH
Levelsof Significance usually expressstatisticalsignifiResearchers cancein termsof levels(e.g.,a testis statistically significantat a specificlevel) rather than giving the specificprobability.Thelevelof statisticalsig(usually.05,.01,or .001)is a way of nificance talking about the likelihood that resultsare due to chancefactors-that is,that a relationshipappearsin the samplewhen there is none in the population. If a researchersaysthat resultsare significant at the .05 level, this means the following: r Resultslike theseare due to chancefactors only 5 in 100times. r Thereis a 95 percentchancethat the sample resultsare not due to chancefactorsalone, but reflectthe populationaccurately. r The odds of suchresultsbasedon chance aloneare.05,or 5 percent. r One canbe 95 percentconfidentthat the results are due to a real relationshipin the population,not chancefactors. Theseall say the same thing in different ways.This may soundlike the discussionof sampling distributionsandthe centrallimit theorem in the chapteron sarnpling.It is not an accident. Both arebasedon probability theory which researchersuse to link sampledata to a population. Probability theory lets us predict what happensin the long run overmany eventswhen a random processis used.In other words, it allows preciseprediction over many situationsin the long run, but not for a specificsituation. Sincewe haveone sampleand we want to infer to the population,probabilitytheoryhelpsus estimate the odds that our particularsamplerepresentsthe population.We cannot know for certainunlesswe havethewholepopulation,but probabilitytheoryletsus stateour confidencehow likely it is that the sampleshowsone thing while somethingelseis true in the population. For example,a sampleshowsthat collegemen
and women differ in how many hours they study. Is the result due to an unusual sample' and there is really no difference in the population, or does it reflect a true difference between the sexesin the population?
Type land Typell Errors is basedon Thelogicof statisticalsignificance statingwhether chancefactorsproduceresults. You may ask,Why usethe .05level?It meansa 5 percentchancethat randomnesscould causethe results.Why not usea more certainstandardfor example,a I in 1,000probabilityof random chance?This givesa smallerchancethat randomnessversusa true relationshipcausedthe results. Therearetwo answers.The simpleansweris that the scientific community has informally agreedto use .05 as a rule of thumb for most purposes.Being95percentconfidentofresultsis the acceptedstandardfor explainingthe social world. A secondanswerinvolvesa tradeoff betweenmakingtwo typesof logicalerrots.ATTpe saysthat a reI error occurswhen the researcher lationship existswhen in fact none exists.It meansfalselyrejectinga null hypothesis'AType saysthat a relaII error occurswhen a researcher tionship doesnot exist,but in reality it does'it meansfalselyacceptinga null hypothesis(see wantto avoid Table10.9).Of course,researchers both kinds of errors.Theywant to saythat there is a relationshipin the dataonlywhen it doesexist and that there is no relationshiponly when therereallyis none,but they facea dilemma:As the oddsof makingonetypeof error decline,the oddsof makingthe oppositeerror increase' The idea of Tlpe I and TlPe II errors may seemdifficult at first, but the samelogical dilemmaappearsin manyother settings.For example,a judge can err by decidingthat an accusedpersonis guilty when in fact he or sheis innocent.Or thejudgecanerr by decidingthat a person is innocent when in fact he or she is
CHAPTE R1 O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
271
Type I and Type ll Errors
No relationship
No error
Type ll error
Causalrelationship
Type I error
No error
guilty. The judge doesnot want to make either error.A judge doesnot want to jail the innocent or to free the guilty. The judge must render a judgmentbasedon limited information andbalancethe two tFpesof errors.Likewise,a physician has to decidewhether to prescribei ,rerv medicationfor a patient.The physiciancan err by thinking that the medicationwill be effective and hasno sideefFects when, in fact,it hasa serious side effect,such as causingblindness.Or the physiciancan err by holding back an effective medicationbecause of fearof serioussideeffectswhen in fact there are none.The physican doesnot want to makeeither error. By making the first error, the physicancausesgreatharm to the patientand may evenfacea lawsuit.By maxing the seconderror, the physicandoesnot help the patient get better.Again, a judgment must be made that balancestwo types of possible errors. We can put the ideasof statisticalsignificanceand the two types of error together.An overlycautiousresearcher setsa high levelofsignificance.For example,the researcher might uie the .0001level.He or sheattributesthe resultsto chanceunlessthey are so rare that they would occurby chanceonly I in 10,000times.Sucha high standardmeansthat the researcher is most likely to err by sayingresultsare due to chance when in fact they arenot. He or shemay falsely acceptthe null hypothesiswhen thereis a causal relationship(a TypeII error). Bycontrast,a risktakingresearchersetsa low levelofsignificance,
suchas.10.His or herresultsindicatea relation_ ship would occurby chanceI in 10times.He or sheis likely to err by sayrngthat a causalrela_ tionship exists,when in fact random factors (e.9.,random samplingerror) actuallycausethe results.The researcheris likely to faiselyreject the null hypothesis(Type I error). In sum, the .05 levelis a compromisebetweenType I and TypeII errors. The statisticaltechniquesof inferentialsta_ tisticsarepreciseandrely on the relationshipbe_ tween samplingerror, samplesize,and central limit theorem.Thepowerof inferentialstatistics i: ability to let a researcherstate,with spe_ F.y cific degreesof certainty,that specificsampleie_ sults are likely to be true in J population. por example,a researcherconductssiatisticaltests and finds that a relationshipis statisticallysig_ nificant at the .05level.He or shecan statethat tbe sample results are probably not due to chancefactors.Indeed,there is a 95 percent chancethat a true relationshipexistsin the social world. Testsfor inferentialstatisticsare limited. Thedatamust comefrom a randomsample,and testsonly take into accountsamplingerrors. Nonsampling errors (e.g., a poor sampling frame or a poorly designedmeasure)are not considered.Do not be fooled into thinking that suchtestsoffer easy,final answers.Many-com_ puter programsquickly do the calculationfor inferential and descriptivestatistics(seeBox 10.2).
272
PART TWO , / CO NDUCTI NG Q UANTI TATI V ER E S E A R C H
who needsto calcuAlmosteverysocialresearcher late manystatisticsdoesso with a computerprogram, often using a basicspreadsheetprogram, aredespreadsheets suchas Excel.Unfortunately, functions. signedfor accountingand bookkeeping They includestatistics,but are clumsyand limited for that purpose.Thereare manycomputerprogramsdesignedfor calculatinggeneralstatistics. can be confusingto a beginner, The marketplace computer for productsevolverapidlywith changing technology. In recentyears,the softwarehasbecomelessdemandingfor a user.The most popularprogramsin andSPSS areMinitab,Microcase, the socialsciences Others for the SocialSciences). Package (Statistical AnalysisSystem),STATISincludeSAS(Statistical TICAby StratSoft,andStrata.Manybeganassimple, low-costprogramsfor researchpurposes.
CONCLUS I O N You have learned about organizing quantitative data to prepare them for analysis and about analyzingthem (organizing data into charts or tables, or summarizing them with statistical measures).Researchersuse statistical analysisto test hypotheses and answer research questions' The chapter explained how data must first be coded and then analyzed using univariate or bivariate statistics.Bivariate relationships might be spurious, so control variables and multivariate analysis are often necessary.You also learned some basicsabout inferential statistics. Beginning researcherssometimes feel their results should support a hlpothesis. There is nothing wrong with rejecting a hypothesis. The goal of scientific research is to produce knowledge that truly reflects the social world, not to defend pet ideas or hlpotheses. Hlpotheses are theoretical guessesbased on limited knowledge; they need to be tested.Excellent-quality research
The mostwidelyusedprogramfor statisticsin the arethat social lts advantages in SPSS. socialsciences researchersused it extensivelyfor over three quanit includesmanywaysto manipulate decades, meastatistical most titativedata,and it contains time a long take it can is that sures.A disadvantage complex and options its many of because to learn to purchaseunlessthe statistics.Also,it is expensive "strippeddown"student usergets an inexpensive, versionincludedwith a textbookor workbook. As computertechnologymakesusinga statistics that somepeothe dangerincreases programeasier, statisnot understand but plewillusethe programs, easily They can are doing. programs tics or whatthe statistical by a required assumptions basic violate procedure,usethe statisticsimproperly,and produce resultsthat are pure nonsensebut that look verytechnicallysophisticated.
can find that a hypothesis is wrong, and poorquality researchcan suppoft a hlpothesis. Good researchdepends on high-quality methodology, not on supporting a specific hlpothesis. Good researchmeansguarding againstpossible errors or obstaclesto true inferences from data to the social world' Errors can enter into the research process and affect results at many places: research design, measurement, data collection, coding, calculating statistics and constructing tables, or interpreting results. Even if a researchercan design, measure' collect, code, and calculate without error, another step in the researchprocessremains. It is to interpret the tables, charts, and statistics, and to answer the question: What doesit all mean?The only way to assign meaning to facts, charts, tables, or statistics is to use theory. Data, tables,or computer output cannot answer research questions. The facts do not speak for themselves.As a researcher,you must return to your theory (i.e., concepts, relationships
CHAPTERI O , / A N A L Y S I SO F Q U A N T I T A T I V ED A T A
amongconcepts,assumptions, theoreticaldefinitions) andgivethe resultsmeaning.Do not lock yourself into the ideaswith which you began. Thereis room for creativity,and new ideasare generatedby trying to figureout what resultsreally say.It is important to be carefulin designing and conductingresearchsothat you canlook at the resultsasa reflectionof somethingin the socialworld and not worry aboutwhetherthevare due to an error or an artifact of the research processitself. Beforewe leavequantitativeresearch,there is onelastissue.lournalists,politicians,and others increasinglyusestatisticalresultsto make a point or bolsteran argument.This hasnot produced greater acarracy and information in public debate.More often,it hasincreasedconfusion and made it more important to know what statisticscanand cannotdo. Theclich6that you can prove anything with statisticsis false; however,people can and do misusestatistics. Through ignoranceor consciousdeceit,some peopleuse statisticsto manipulateothers.The wayto protectyourselffrom beingmisledbystatistics is not to ignore them or hide from the numbers.Rather,it is to understandthe research processand statistics,think about what you hear,and askquestions. We turn next to qualitative research.The logic and purposeof qualitativeresearchdiffer from those of the quantitative, positi\rist approach ofthe past chapters.It is lessconcerned with numbers, hypotheses,and causalityand more concernedwith words,norms and values. and meaning.
273
contingency table control variable covariation cross-tabulation curvilinear relationship descriptive statistics direct entrymethod elaboration paradigm explanation pattern frequency distribution frequencypolygon independence interpretation pattern level of statistical significance linear relationship marginals mean median mode normal distribution partials percentile pie chart possiblecode cleaning range replication pattern scattergram skewed distribution specification pattern standard deviation statistical sigrificance suppressor variable pattern Tlpe I error Tlpe II error univariate statistics z-score
Key Terms bar chart bivariate statistics body ofa table cell ofa table code sheets codebook contingenry cleaning
Endnotes l. Note that coding sexas I = Male, 2 = Female,or as 0 = Male, I = Female, or reversing the sex for numbers is arbitrary. The only reason numbers are used instead of letters {e.g. M and F) is be_ cause many computer programs work best with all numbers. Sometimes coding data as azero can
274
PART TWO /
CO NDUCTI NG Q UANTI T A T I V ER E S E A R C H
createconfusion, so the number 1 is usually the lowestvalue. 2. Thereareother statisticsto measurea specialkind of meanfor ordinal dataand for other specialsituations,which arebeyondthe level of discussion in this book. For a discussionofthe elaborationparadigmand and Rosenits history,seeBabbie( 1998:393-401) berg(1968).
4. In formal hypotlresistesting, researcherstest the Theyusuallywant to rejectthe null nullh.ypothesis. becauserejectionofthe null indirectly suPports the alternativehypothesisto the null, the one they deducedfrom theory asa tentativeexplanation'
1 FieldResearch Introduction Research QuestionsAppropriatefor FieldResearch The Logic of Field Research What ls FieldResearch? Project Stepsin a FieldResearch Choosing a Site and Gaining Access Selectinga Siteand Entering Strategyfor Entering Learning the Ropes BuildingRapport Relationsin the Field Rolesin the Field Maintaining Relations Observing and Collecting Data Watchingand Listening TakingNotes DataQuality Focusing and Sampling The Field Research lnterview The FieldInterview Typesof Questionsin FieldInterviews Informants InterviewContext Leaving the Field Focus Groups Ethical Dilemmas of Field Research Deception Confidentiality Involvement with Deviants Publishing FieldReports Conclusion
276
PART THREE /
CO NDUCTI NG Q UALI TA T I V ER E S E A R C H
in a relativelyfixed setting(e.g.,a streetcorner, church,bar, beautysalon,baseballfield, etc.). In order to useconsistentterminology,we This chapterand the two that follow shift from the quantitativesryleof the pastseveralchapters cancallthe peoplewho arestudiedin a field settingmembers.They areinsidersor nativesin the to the qualitative researchsryle.The qualitative field andbelongto a group,subculture,or social and the quantitative stylescandiffer a greatdeal. wants This chapterdescribesfield research,alsocalled settingthat the "outsider"field researcher research. to penetrateand learn about. ethnographyor participant-observation haveexploreda wide variFieldresearchers It is a qualitativestylein which a researcherdiety of socialsettings,subcultures,and aspectsof sorectlyobserves andparticipatesin small-scale sociallifel (seeFigure11.1).Placesmy students cial settings in the present time and in the short-term,smallhave conductedsuccessful home culture, researcher's Many studentsare excitedby field research scalefield researchstudiesinclude a beautysalon, day-carecenter, bakery, bingo parlor, becauseit involveshanging out with someexbowling alley,church,coffeeshop,laundromat, otic group of people.Thereare no cold mathematicsor complicatedstatistics,and no abstract policedispatchoffice,nursinghome,tattoo parlor, andweightroom. deductivehypotheses.Instead,there is direct, face-to-facesocialinteractionwith "real people" Ethnography and Ethnomethodologt. Two in a naturalsetting. In field research,the individual researcher modern extensionsof field research,ethnography and ethnomethodology,build on the social the peoplebeing directlytalkswith and observes constructionistperspective.Eachis redefining studied. Through interaction over months or how field researchis conducted.Theyarenot yet years,the researcher learnsaboutthem,their life the core offield research,so they are discussed histories,their hobbiesand interests,and their only briefly here. habits,hopes,fears,and dreams.Meeting new Ethnographycomesfrom cultural anthropeople,developingfriendships,and discovering pology.zEthnomeanspeopleor folk, andgraphy new socialworlds canbe fun. It is alsotime consuming, emotionally draining, and sometimes refersto describingsomething.Thas ethnograplry meansdescribinga culture and understanding physicallydangerous. anotherwayof life from the nativepoint ofview. Ethnographyassumesthat peoplemake inferResearchQuestions Appropriate for ences-that is, go beyond what is explicitly seen Field Research or saidto what is meantor implied. PeopledisField researchis appropriatewhen the research playtheir culture(whatpeoplethink, ponder,or questioninvolveslearning about, understand- believe)through behavior(e.g.,speechand acing, or describinga group of interactingpeople. tions) in specificsocialcontexts.Displaysof behavior do not give meaning;rather,meaningis It is usuallybestwhen the questionis: How do inferred,or someonefiguresout meaning.Movpeopledo Y in the socialworld? or What is the ing from what is heardor observedto what is acsocialworld of X like?It canbe usedwhen other tually meantis at the centerof ethnography.For methods (e.g., survey,experiments)are not example,when a studentis invitedto a "kegger," practical,asin studyingstreetgangs. the studentinfersthat it is an informalpartywith Field researchers study peoplein a location peopleat which beerwill be other student-aged or setting.It hasbeenusedto studyentirecomserved,basedon his or her cultural knowledge. munities. Beginning field researchersshould Cultural knowledgeincludessymbols,songs'saystartwith a relativelysmall group (30 or fewer) who interactwith eachother on a regularbasis ings, facts,waysof behaving,and objects(e.g., INTRODUCTION
CHAPTERll
FIG URE 1 I . 1
/
F I E L DR E S E A R C H
277
Examplesof FieldResearchSites/Topics
Small-ScaleSettings Passengers in an airplane Barsor taverns Batteredwomenl shelters Cameraclubs Laundromats ' Socialmovementorganizations Socialwelfareoffices Television stations Waitingrooms Community Settings Retirementcommunities Smalltowns Urbanethniccommunities Working-class neighborhoods Children'sActivities Children'splaygrounds LittleLeaguebaseball Youthin schools Juniorhighgirlgroups
Door-to-doorsalespersons Factoryworkers Gamblers Medicalstudents Femalestrippers Policeofficers Restaurant chefs Socialworkers Taxidrivers Devianceand Criminal Activity Body/genital piercingand bnnding Cults Drugdealersand addicts Hippies Nude beaches Occultgroups Prostitutes Streetgangs,motorcyclegangs Streetpeople,homelessshelters
Airlineattendants Artists Cocktailwaitresses Dog catchers
Medical Settings and Medical Events Death Emergencyrooms Intensivecareunits Pregnancy and abortion Supportgroupsfor Alzheimerlcaregivers
telephones,newspapers, etc.).We learn the culture by watchingtelevision,listeningto parents, observingothers,and the like. Cultural knowledgeincludesboth explicit knowledge,what we know and talk about, and tacit knowledge,what we rarely acknowledge. For example,explicitknowledgerncladesthe socialevent(e.g.,a "kegger").Most peoplecaneasily describe what happens at one. Tacit knowledge includesthe unspokencultural norm for the proper distanceto stand from others. Peopleare generallyunawarethat they usethis
norm. They feeluneaseor discomfortwhen the norm is violated,but it is difficult to pinpoint the source of discomfort. Ethnographeri describethe explicit and tacit cultural knowledge that membersuse. Their detaileddescriptions and carefirlanalysistakewhat is describedapart and put it backtogether. Ethnomethodology is adistinctapproachdevelopedin the 1960s,with its own uniqueterminology. It combinestheory, philosophy,and method. Some do not considerit a part of sociology.
Occupations
278
pA RTr HR E E / c o N D U c rl N c QU AL ITA TIvE R E S E A R cH
is A simple definition of ethnomethodology the study of commonsenseknowledge.Ethnomethodologistsstudy common senseby observingits creationand use in ongoingsocial interactionin natural settings.Ethnomethodologyis a radicalor extremeform of field research, basedon phenomenologicalphilosophy and a socialconstructionistapproach.It involvesthe highlydetailedanalysisof micro-sitspecialized, uations (e.g.,transcriptsof short conversations or videotapesof socialinteractions).Compared to other field research,it is more concerned aboutmethod and arguesthat researchfindings result as much from the method usedas from the sociallife studied. Ethnomethodologyassumesthat social meaningis fragileand fluid, not fixed, stable,or solid. Meaning is constantlybeing createdand re-createdin an ongoing process.For this reaanalyzelanguage,inson,ethnomethodologists cluding pausesand the contextof speech.They assumethat people"accomplish"commonsense understanding by using tacit social-cultural rules,and socialinteractionis a processofreality construction.Peopleinterpret everydayevents by using cultural knowledgeand cluesfrom the social context. Ethnomethodologistsexamine how ordinary people in everydaysettingsapply tacit rules to make senseof sociallife (e.g.,to know whetheror not someoneis joking). examineordinary soEthnomethodologists cial interactionin greatdetail to identi$'the rulesfor constructingsocialrealityand common sense,how theserulesareapplied,and how new rules are created.For example,they arguethat standardizedtestsor surveyinterviewsmeasure a person'sability to pick up implicit cluesand apply common sensemore than measuringobjectivefacts.
TH E LOG ICOF FIE LDRE S E A RCH What Is FieldResearch? It is difficultto pin downa specificdefinitionof it ismoreof anorientation fieldresearchbecause
towardresearchthan a fixed setoftechniquesto apply.3A field researcherusesvariousmethods is a reto obtain information. A field researcher sourcefirl,talentedindividualwho hasingenuity and an ability to think on her or his feet while in the field. Field researchis basedon naturalism,which is also used to study other phenomena(e.g.' oceans,animals,plants, etc.).Naturalisn involves observingordinary eventsin natural settings,not setin contrived,invented,or researcher-created tings. Researchoccursin the field and outside the safesettingsof an of,fice,laboratory,or classroom. goalis toexaminesocial A field researcher's meaningsand grasp multiple perspectivesin naturalsocialsettings.He or shewantsto getinsidethe meaningsystemof membersand then return to an outsideor researchviewpoint. To and switchesperspectives do this, the researcher looks at the settingfrom multiple points of view simultaneously. Fieldresearchis usuallyconductedby a single individual, although small teamshave been effective(seeBox 11.1).The researcheris directly involved in and part of the socialworld are studied,sohis or her personalcharacteristics direct inrelevantin research.The researcher's volvementin the field often has an emotional impact. Field researchcan be fun and exciting, but it canalsodisrupt one'spersonallife, physical security, or mental well-being. More than friendother typesofsocial research,it reshapes ships, family life, self-identiry and personal values. Steps in a Field ResearchProject Naturalism and direct involvement mean that field researchis lessstructured than quantitative research.This makesit essentialfor' a researcherto be well organizedand preparedfor the field. It alsomeansthat the stepsof a project arenot entirelypredeterminedbut serveas an approximateguide or road map (seeBox TI.2\.
CHAPTERll
A field researcher doesthe following: .l
. Observesordinaryeventsand everydayactivities as they happenin naturalsettings,in addition to any unusualoccurrences 2. Becomes directlyinvolvedwiththe peoplebeing studiedand personallyexperiences the process of dailysociallifein the fieldsetting 3. Acquiresan insider'spoint of viewwhilemaintainingthe analyticperspective or distanceof an outsider 4. Usesa varietyof techniquesand socialskillsin a flexiblemanneras the situationdemands 5. Producesdata in the form of extensivewritten notes,as wellas diagrams,maps,or picturesto provideverydetaileddescriptions 6. Seeseventsholistically (e.g.,asa wholeunit,not in pieces)and individually in their socialcontext Z. Understandsand developsempathyfor membersin a field setting,and does not just record "cold"objectivefacts 8. Notices both explicit (recognized,conscious, spoken)and tacit (lessrecognized, implicit,unspoken)aspectsof culture 9. Obseirves ongoingsocialprocesses without upsettin&disrupting,or imposingan outsidepoint of view
10. Copeswith highlevelsof personalstress,uncertainty,ethicaldilemmas, and ambiguity
Flexihility. Field researchersrarely follow fixed steps.In fact, flexibility is a key advantage offield research,which letsa researcher shift direction and follow leads.Good field researchers recognizeand seizeopportunities,"play it by ear,"and rapidly adjustto fluid socialsituations. A field researcher doesnot beginwith a set of methodsto apply or explicit hlpothesesto test.Rather,he or shechoosestechniqueson the
/
F I E L DR E S E A R C H
279
1. Prepare oneself, read the literature, and defocus. 2. Selecta field siteand gainaccessto it. 3. Enterthefieldandestablishsocialrelationswith members. 4. Adopt a socialrole, learnthe ropes,and get alongwith members. 5. Watch,listen,and collectqualitydata. 6. Beginto analyzedataandto generateand evaluateworkinghypotheses. 7. Focuson specificaspectsofthe settingand use theoreticalsampling. 8. Conduct field interviewswith memberinformants. 9. Disengage and physicallyleavethe setting. 1 0. Completethe analysesand write the research repon. Note:Thereis nofixedpercentage of timeneeded for each step.Fora roughapproximation, (l 960:12) Junker suggested that,onceinthefield,theresearcher should expect to spendapproximately one-sixth of hisor hertime observing, one-third recording data,one-third ofthetime anallzing data,andone-sixth reporting results. Alsosee Denzin (l989:176)foreightstepsof fieldresearch.
basisof their valuefor providinginformation.In the beginning,the researcherexpectslittle control over dataand little focus.Oncesocializedt