PREDICTIONS OF PUBLIC OPINION FROM THE MASS MEDIA Computer Content A n a l y s i s and Mathematical Modeling
D A V I D ...
22 downloads
474 Views
111MB 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
PREDICTIONS OF PUBLIC OPINION FROM THE MASS MEDIA Computer Content A n a l y s i s and Mathematical Modeling
D A V I D P. F A N
C o n t r i b u t i o n s to the S t u d y of M a s s M e d i a a n d C o m m u n i c a t i o n s , N u m b e r 12
G R E E N W O O D PRESS NEW YORK • WESTPORT, CONNECTICUT •
LONDON
L i b r a r y of C o n g r e s s C a t a l o g i n g - i n - P u b l i c a t i o n
Data
F a n , D a v i d P. Predictions o i public o p i n i o n from the m a s s media : c o m p u t e r c o n t e n t a n a l y s i s a n d m a t h e m a t i c a l m o d e l i n g / D a v i d P. F a n . p.
c m . — ( C o n t r i b u t i o n s to t h e s t u d y of m a s s m e d i a a n d
c o m m u n i c a t i o n s , I S S N 0732-4456 ; n o . 12) Bibliography:
p.
Includes indexes. I S B N 0-313-26296-9 (lib. b d g . : a i k . p a p e r ) 1. M a s s m e d i a — U n i t e d S t a t e s — I n f l u e n c e — D a t a p r o c e s s i n g . media—United States—Influence—Mathematical models. opinion—United States—Data processing. States—Mathematical models. Forecasting. I
Title
2. M a s s
3. P u b l i c
4. P u b l i c o p i n i o n — U n i t e d
5. P u b l i c o p i n i o n — U n i t e d S t a t e s -
6. C o n t e n t a n a l y s i s ( C o m m u n i c a t i o n ) — D a t a p r o c e s s i n g ,
II. Series.
MN90.M3F36
1988
303.3 ' 8 - d c l 9
88-5683
B r i t i s h L i b r a r y C a t a l o g u i n g i n P u b l i c a t i o n D a t a is a v a i l a b l e . C o p y r i g h t © 1988 by D a v i d P. F a n A l l rights r e s e r v e d . N o p o r t i o n of t h i s b o o k m a y be reproduced, by a n y process or technique, without the e x p r e s s w r i t t e n c o n s e n t of t h e p u b l i s h e r . L i b r a r y of C o n g r e s s C a t a l o g C a r d N u m b e r : 88-5683 ISBN:0-313-26296-9
First p u b l i s h e d in 1988 G r e e n w o o d Press, Inc. 88 Post R o a d W e s t , W e s t p o r t , C o n n e c t i c u t 06881 P r i n t e d i n t h e U n i t e d S t a t e s of A m e r i c a
T h e paper used in this book complies with the P e r m a n e n t P a p e r S t a n d a r d i s s u e d by the N a t i o n a l Information S t a n d a r d s O r g a n i z a t i o n (Z39.48-1984). 10 9 8 7 6 5 4 3 2
1
DEDICATION To M ary se
Copyrighte
Contents
Kill
A CK NOW
I.
« V I* I*
3
Ouüine
il
Organization
7
9 1.1
Strategies Used in Formulating Ideodynamics
1.2
Nature of the Population
10
1 3
Nature of Persuasion
11
1.4
Nature o f Persuasive Messages
12
1.5
Relationships between ldeodynamic Structures
16
1.6
Overview of Opinion Calculations
16
1.7
Details of Opinion Calculations for the A wares
17
1.8
Details of Opinion Calculations for the Unaw ares
22
1.9
Time Scale of ldeodynamic Analyses
22
Figures 1.1-1.3
24
9
VIII
Contents
2.1
Significant Features of Ideodynamics
27
2.2
Model Comparisons
32
CHAPTER
3: D A T A F O R
CALCULATING
PUBLIC
OPINION
37
3.1
T i m e Series o f O p i n i o n Polls
37
3.2
Relevant Persuasive Messages in the Associated Press
39
Figure 3.1
43
CHAPTER
4; C O M P U T E R T E X T A N A L Y S I S RV OF SUCCESSIVE FILTRATIONS
METHOD 45
4.1
General Text Analysis Programs
45
4.2
Strategy for Content Analysis Using Successive Filtrations
46
4.3
Sketch of Filtration and Scoring Computer Program Runs
47
4.4
Text Analyses for Defense Spending
49
4.5
Text Analysis for Troops i n Lebanon
50
4.6
Democratic Primary
52
4.7
Text Analysis for the Economic Climate
53
4.8
Text Analysis for Unemployment versus Inflation
53
4.9
Text Analysis for Contra A i d
53
4.10
Summary Features of Text Analysis by Successive Filtrations
54
4.11
Extensions o f the Text Analysis Procedure
56
CHAPTER 5:
PROJECTIONS
OF PUBLIC OPINION
57
5.1
Opinion Predictions for Defense Spending
57
5.2
O p i n i o n Predictions for Troops in Lebanon
62
5.3
Opinion Predictions for the Democratic Primary
64
5.4
Opinion Predictions for the Economic Climate
65
5.5
Opinion Predictions for Unemployment versus Inflation
66
5.6
Opinion Predictions for Contra A i d
66
Contents
Lx,
5.7
Summary o f Constants Used i n Poll Projections
67
5.8
Summary of Statistics for Poll Projections
68
Tables 5.1-5.3
69
Figures 5.1-5,45
72
CHAPTER
6: M E T H O D O L O G I C A L OF W O R K
SIGNIFICANCE 117
6.1
Validation o f Ideodynamics
H8
6.2
Data and Issues for Successful Ideodynamic Calculations
119
6.3
Positions for W h i c h Persuasive Messages Are Scored
121
6.4
Computer Text Scoring
121
6.5
Ideodynamic Calculations o f Opinion Time Trends
123
6.6
Insensitivity o f Predictions to the Starting Opinion Values
127
6.7
Interpretations for A l l Ideodynamic Parameters
127
6.8
Significance of No Opinion Change
127
6.9
Analysis o f Persuasive Messages Acting on Public Opinion
128
C H A P T E R 7: S I G N I F I C A N C E O F W O R K T O OF OPINION FORMATION
THEORIES 129
7.1
Mass Media Messages and Opinion Leadership
129
7.2
Reinforcing Role of Persuasive Messages
131
7.3
Cumulative Effects of Information Rather than M i n i m a l Effects o f the Media
132
7.4
Caveats for Laboratory Experiments
134
7.5
L a w o f the 24-Hour Day
134
7.6
Interpretations o f Ideodynamic Parameters
7.7
Nature o f Effective Persuasive Messages i n the Mass Media
APPENDIX
A.2
A:
MATHEMATICS
Structure of the Population
OF
IDEODYNAMICS
T35
136
141
141
JC
Contents
A.3
Structure o f Messages
142
A.4
Nomenclature Simplification
144
A.S
Infon Properties
144
Ai)
I n f o n Persuasive Force
144
A.7
Information Influencing the Unawares
144
A.8
Information Influencing the A wares
145
A.9
Effect o f Information on the Population
147
A. 10 Modifications for AP Infons Assuming No Unawares
149
A.11
152
Comparison w i t h U n i f o r m Distribution
A. 12 Modifications for AP Infons Assuming Non-negligible Unawares
153
A. 13 Extensions to Very L o n g Times
154
A . 14 Models w i t h N o Dependence on Subpopulations
154
o PIN^ (\\* r H AN f IF
L A T I N F I
L55
R.l
Defense SpcndinR--1977-1984
155
B .2
Troops in Lebanon-1983-1984
156
B.3
Democratic Primary-1983-1984
[56
B.4
Economic CUmate-1980-1984
157
B.5
Unemployment versus Inflation-1977-1980
157
B .6
Contra Aid--1983-1986
157
T a h i t i * _1-tt.fi
159
APPENDIX
C:
SUMMARIES
OF
TEXT
ANALYSES
165
C. 1
Strategy for Content Analysis by Successive Filtrations
165
C.2
Text Analysis for Defense Spending-Including Detailed Example
165
C.3
Text Analysis for Troops i n Lebanon
172
C.4
Text Analysis for the Democratic Primary
173
C.5
Text Analysis for the Economic Climate
173
Contents
xi
C.6
Text Analysis for Unemployment versus Inflation
174
C.7
Text Analysis for Contra A i d
174 116
P R O J K C T I O N ^ ^
PIR' ir
OPINION
^
D.l
Computations o f Persuasive Forces
183
D.2
Population Conversion Models
183
D.3
O p i n i o n Projections
184
AUTHOR
1NDFX
193
SUBJECT
INDEX
122
Tables and Figures
Figure 1.1
Example o f persuasive forces o f infons.
24
Figure 1.2
Population conversion model for defense spending.
25
Figure 1.3
Illustration o f the impact o f a single persuasive infon favoring m o r e defense spending.
26
Figure 3.1
Poll data for defense spending.
43
Table 5.1
Statistical comparisons for opinion projections.
69
Table 5.2
Candidate name counts in dispatches retrieved for the Democratic primary.
70
Table 5.3
O p t i m a l constants for opinion projections.
71
Figure 5.1
Persuasive forces o f A P infons scored for favoring more, same, and less defense spending.
72
Persuasive forces o f A P infons scored for favoring more and less defense spending.
73
Opinion on defense spending from dispatches scored to favor more, same, and less spending.
74
Figure 5.4
Constant optimization curves for defense spending.
75
Figure 5.5
Opinion f r o m a subset o f A P dispatches scored to favor more, same, and less defense spending.
76
Opinion f r o m another subset o f A P dispatches scored to favor more, same, and less defense spending.
77
Opinion on defense spending assuming the entire population favored more spending at the time o f the first scored A P infon in January 1977.
78
Figure 5.2
Figure 5.3
Figure 5.6
Figure 5.7
Copyrighte u
111 d
i d idi
xiv
List of Tables and
Figures
Opinion on defense spending from dispatches scored to favor more and less defense spending only.
79
Persuasive forces o f A P infons f r o m 1977 to 1986 scored for favoring more, same, and less defense spending.
80
Figure 5.10
Opinion on defense spending from 1977 to 1986.
81
Figure 5.11
Constant optimization curve for the contributions from paragraphs favoring less defense spending.
82
Effect o f stories on waste and fraud on public opinion on defense spending.
83
Figure 5.13
Persuasive forces for troops i n Lebanon from A P infons only.
84
Figure 5.14
Persuasive forces for troops in Lebanon from A P infons w i t h and w i t h o u t a truck bombing infon favoring more troops
85
Population conversion model for actions o f infons favoring more, same, and less troops in Lebanon.
86
Optimizations for the modified persuasibility constant, the weight for paragraphs favoring less troops, and the value o f the truck bombing infon favoring more troops.
87
Optimization curves far the persistence half-life and the value o f the truck bombing infon favoring less troops.
88
Figure 5.18
O p i n i o n on troops i n Lebanon assuming o n l y A P infons.
89
Figure 5.19
Opinion on troops in Lebanon w i t h a truck bombing infon favoring more troops.
90
Comparison o f opinion projections w i t h (solid line) or without (dotted line) the truck bombing infon favoring more troops.
91
Persuasive forces from A P infons w i t h and without a truck bombing infon favoring less troops.
92
Opinion projections w i t h and w i t h o u t a truck bombing infon favoring less troops.
93
Persuasive forces favorable to Democratic presidential candidates from AP paragraphs scored using bandwagon words
94
Figure 5.8
Figure 5.9
Figure 5.12
Figure 5.15
Figure 5.16
Figure 5.17
Figure 5.20
Figure 5.21
Figure 5.22
Figure 5.23
Figure 5.24
Persuasive forces unfavorable to Democratic presidential candidates from A P paragraphs scored using bandwagon words
Figure 5.25 Figure 5.26
Persuasive forces o f A P infons scored by name count only. Population conversion model for actions o f infons scored using bandwagon words.
95 96
97
Copyrighte
List of Tables and Figures
Figure 5.27
Figure 5.28
Figure 5.29
xv
Population conversion model for actions o f infons scored by name count only.
98
Optimization curves for constants for the Democratic primary.
99
Opinion on Democratic candidates when infons were scored by the bandwagon analysis.
100
Opinion on Democratic candidates when infons were scored by name count only.
101
Persuasive forces from A P paragraphs favoring better, same, and worse economic conditions.
102
Population conversion model for actions o f infons favoring better, same, and worse economic conditions.
103
Figure 5.33
Optimization curves for constants for the economic climate.
104
Figure 5.34
Opinion on economic climate.
105
Figure 5.35
Persuasive forces o f A P infons favoring unemployment more important equal importance, and inflation more important.
106
Population conversion model for actions o f infons favoring unemployment more important, equal importance, and inflation more important.
107
Optimization curves for the modified persuasibility constant and the infon weighting constants for unemployment versus inflation.
108
Optimization curve for the persistence constant for unemployment versus inflation.
109
Opinion favoring unemployment more important, equal importance, or inflation more important.
110
Persuasive forces o f A P infons scored by the author as favoring and opposing Contra aid.
111
Persuasive forces o f A P infons scored by S w i m , Miene, and French as favoring and opposing Contra aid.
112
Population conversion model for actions o f infons favoring and opposing Contra aid.
113
Figure 5.43
Constant optimization curves for Contra aid.
114
Figure 5.44
Opinion favoring and opposing Contra aid using infon scores by the author.
115
Opinion favoring and opposing Contra aid using infon scores by S w i m , Miene, and French.
116
Polls on the desirability o f increasing defense spending.
159
Figure 5.30
Figure 5.31
Figure 5.32
Figure 5.36
Figure 5.37
Figure 5.38
Figure 5.39
Figure 5.40
Figure 5.41
Figure 5.42
Figure 5.45
Table B. 1
Copyrighte U
I I ICLlwl l u i
xvi
List of Tables and
Table B.2
Figures
A B C News Poll on the stationing o f American troops i n Lebanon.
160
Table B.3
A B C News Poll on the Democratic primary.
161
Table B.4
A B C News Poll o n the economic climate.
162
Table B.5
N B C News Poll on the importance o f unemployment versus inflation.
163
Table B.6
Polls on the desirability o f sending Contra aid.
164
Table C. 1
Summary o f text analysis for defense spending.
176
Table C.2
Summary o f text analysis for defense waste and fraud.
177
Table C.3
Summary o f text analysis for troops i n Lebanon: scoring for more, same, and less troops.
178
Summary o f text analysis for Democratic primary: scoring for bandwagon words.
179
Summary o f text analysis for economic climate: scoring for better, same, and worse.
180
Summary o f text analysis for unemployment versus inflation: scoring for unemployment more important, equal importance, and infladon more important.
181
Summary of text analysis for Contra aid: scoring for infons favoring and opposing aid.
182
Table C.4
Table C.5
Table C.6
Table C.7
Copyrighte
Acknowledgments
In acknowledging the principal influences o n this book, I w o u l d like to begin w i t h m y father, Professor Hsu Y u n Fan, a physicist w h o persuaded me to obtain a bachelor's o f science degree i n physics rather than pursue m y o w n interests i n the social sciences. His viewpoint was that physics was easy to study given its precision and concreteness w h i l e the social sciences were m u c h more d i f f i c u l t w i t h their uncertainties and ambiguities. This background i n physics is seen i n the mathematical m o d e l o f ideodynamics at the core o f this book. That model uses equations i n the tradition o f such areas in physics as dynamics and thermodynamics. Next, I w o u l d like to acknowledge Professor Cyrus L e v i n t h a l , the thesis advisor for m y P h . D . studies i n biology. H e showed me how t o formulate assumptions w h i c h c o u l d both reflect problems o f interest t o me and be translated i n t o mathematical equations. Indeed, the equations o f ideodynamics are very similar t o those underlying my Ph.D. thesis on the metabolism o f the messenger ribonucleic acids involved i n the expression o f genes. Throughout m y studies i n the physical sciences, 1 have been struck by the fact that important advances often depended on the discoveries o f elementary particles such as protons i n nuclei, nuclei i n atoms, and atoms i n molecules, t a c h o f these particles has a l i m i t e d number o f well-defined and quantifiable characteristics. S i m i l a r l y , the rapid progress i n biology i n recent years has been based on the concept o f the gene, w h i c h is a discrete unit o f inheritance w i t h a simple structure and a small number o f properties. These are but some examples o f how the understanding o f complex natural phenomena were advanced by analyses o f discrete elemental units. The direct consequence o f this viewpoint is m y postulate that persuasive messages can be coded as infons. M y studies in biology included the use o f both genetic and biochemical techniques. The thought patterns i n these areas f o r m the bases o f the n e w method o f content analysis i n this book. F r o m biochemistry, 1 learned that the study o f complicated materials frequently benefits f r o m a series o f purification steps, each one removing extraneous components to y i e l d progressively more homogeneous preparations enriched i n relevant materials. This logic led to the strategy o f successive "filtrations" during the text analyses. The detailed strategies f o r the text filtrations and final scoring are derived i n large part f r o m the principles o f gene expression i n genetics. I n more recent times, m y biological research has turned t o w a r d the study o f the k i l l e r T cells o f the immune system involved i n protection against viral infections and rejection o f organ transplants. For those studies, I realized that the analyses c o u l d be greatly aided by a mathematical model to describe the k i l l i n g activity and a computer program based o n the model t o process the data. This need l e d me t o return to mathematical modeling, w h i c h I had not performed since m y Ph.D. days. I also learned computer languages i n order to write the programs needed for the data analyses. These exercises i n mathematical modeling and computer p r o g r a m m i n g gave additional
Copyrights
xviii
Acknowledgments
impetus to m y desire to examine the social sciences f r o m a mathematical and computational perspective. I had maintained an active interest i n social phenomena from before m y university days. Remaining at the general level, I w o u l d l i k e to thank the Graduate School o f the University o f Minnesota for the funds used for the data gathering and the hiring o f the graduate assistants who scored the text for the case o f the Contras i n Nicaragua. I n addition, i t is necessary to note the crucial role o f the tenure system at A m e r i c a n universities w h i c h permitted me to explore the social sciences using the techniques and thought patterns from my home department in the biological sciences. It was also very useful for my university to permit a sabbatical leave, during which I performed much o f the research i n this book. A number o f h e l p f u l investigators have also contributed i m p o r t a n t l y i n ways directly useful for this book. M a n y o f these individuals were associated w i t h the University o f Minnesota. W i t h o u t being exhaustive, and i n order o f topic rather than importance, I w o u l d like to note i n particular: Professors M i c h a e l Simmons, James Curtsinger, and Frank Enfield o f m y o w n Department o f Genetics and Cell Biology, Dennis Cooke o f the Department o f A p p l i e d Statistics, and Hans Weinberger o f the Institute o f A p p l i e d Mathematics w i t h w h o m I discussed many o f the details o f the mathematical modeling and statistical concerns. Thanks also t o Professor Donald McTavish o f the Department of Sociology, who graciously permitted me to explore his M C C A computer content analysis program; the late Professor F. Gerald K l i n e o f the School o f Journalism and Mass Communications, w h o made very useful suggestions, such as the use o f the Associated Press to represent the A m e r i c a n mass media; Professors John Sullivan o f the Department o f Political Science, and Eugene Borgida o f the Department o f Psychology w i t h w h o m I discussed the relationships between my w o r k and those o f others i n the social sciences; and Professor John Freeman o f the Department of Political Science who gave this book a careful and critical reading. I w o u l d also l i k e to thank Professor Bruce Russett o f Y a l e U n i v e r s i t y and the anonymous reader for the Greenwood Press f o r the useful comments m a d e b e f o r e completion o f this book. A m o n g other colleagues not at Minnesota, I w o u l d like to give special thanks t o Professors B e n j a m i n Page and Robert Shapiro and their associates at the National O p i n i o n Research Center i n Chicago for extremely helpful discussions, and particularly for access to their many time series o f p o l l data f r o m which I chose several to analyze. Those p o l l data were absolutely indispensible for the studies. Obviously, none o f these acknowledgements i m p l y anyone else's responsibility for this book. Clearly, that responsibility is totally mine.
Copyrighte
Predictions of Public Opinion from the Mass Media
Copyrighte
Introduction
This book concerns the power o f information o n society. T h e central thesis is that public o p i n i o n can be swayed in a predictable fashion by messages acting o n the populace. W h e n the b u l k o f the relevant messages are i n the press, then the press becomes the p r i n c i p a l determinant o f society's attitudes and beliefs. Although previous w o r k has suggested that the press is able to set the agenda f o r public discussions, this book is unusual i n demonstrating that the press is also able to mold o p i n i o n w i t h i n agenda items. The importance o f the press o n opinion has long been recognized. This is seen i n the concept o f governmental press censorship, w h i c h was invented long ago. However, the assignment o f the preeminent role o f the press in opinion formation i n a free democracy is i n apparent conflict w i t h a sizable body o f literature describing the " m i n i m a l effects o f the media." W i t h this shield, journalists and editors could w o r k without feeling that every one o f their daily choices was affecting opinion. However, the conflict between press importance and its m i n i m a l effect is more apparent than real. As summarized i n Chapter 7, the impact o f a piece o f news is most appropriately assessed quantitatively. I n other words, messages i n the mass media should be given numerical strengths. A l t h o u g h any one news story, or restricted group o f media messages, can have effects ranging f r o m very small through very large, opinions can frequently be computed f r o m the cumulative effect o f all news stories, most o f w h i c h can indeed have relatively m i n i m a l effects individually. Therefore, i n general, the concept o f the " c u m u l a t i v e effects o f i n f o r m a t i o n " comprising mainly mass media information for many issues—is more useful than the law o f m i n i m a l effects. T h i s idea o f the c u m u l a t i v e impact o f i n f o r m a t i o n s t i l l permits w o r k i n g members o f the press t o proceed w i t h o u t constantly w o r r y i n g about the effects o f their every w o r d . I n d i v i d u a l news items are themselves still likely to have small impact. However, over the long term, all the effects accumulate and the totality o f press messages is capable o f being the major influence on o p i n i o n . Thus society should realize that i n d i v i d u a l messages can indeed have m i n i m a l effects, but w i t h long-term trends being o f great importance. As j u s t noted, this book does not propose that the press is always the dominant force i n o p i n i o n f o r m a t i o n . Rather, the hypothesis is that i t is the totality o f relevant information w h i c h w i l l shape opinion. Therefore, the press w i l l only be the primary influence i f other messages are o f minor importance. Obviously, the importance o f the press is related to its credibility. This trust has n o direct relationship to whether the public ranks the press as credible in opinion polls. I t is only essential that the public as a whole uses no alternate sources o f
Copyrighte U
4
Introduction
information for polled issues discussed i n the mass media. For example, the press i n closed societies is not l i k e l y to be the main determinant o f attitudes i f its reputation is so l o w that sizable portions o f the populace rely on rumor and the underground press. I n an open society like that i n the United States, trust in the press is likely t o be greater. It was to explore the domain o f media dominance i n opinion formation that studies i n this book were performed on a variety o f topics. Issues were chosen f r o m both the domestic and foreign policy arenas. The two issues w i t h the clearest foreign policy implications concerned whether more troops should be sent to Lebanon (1983¬ 1984) and whether U.S. aid should be sent to the Contra rebels in Nicaragua (1983¬ 1986). The domestic topics included those on governmental policy and economic issues. The policy question was whether or not more should be spent for national defense (1977-1986). The two economic issues focused on whether unemployment or inflation was the more important problem (1977-1980), and whether the economic climate was i m p r o v i n g (1981-1984). The remaining domestic issue was voter preference for the best candidate in the Democratic presidential primary (1983-1984). For all these cases, the mass media has the p r i n c i p a l role in influencing public opinion. For the Democratic primary, the press could not be expected to be the dominant influence i f there were important additional informational sources such as campaign advertising. It was to avoid the complication o f such alternative information sources that the Democratic primary was studied before the I o w a caucuses, a time when national media stories should have been the most important source o f persuasive messages for opinion nationwide. A t these early times, campaign advertising was negligible countrywide w h i l e o p i n i o n p o l l results were obtained f r o m this large population base. F r o m the discussion above, a quantitative analysis is able to reconcile the m i n i m a l effects o f the media w i t h the cumulative effects o f information. Obviously, such quantitative assessments imply a mathematical analysis, and, indeed, this book describes the new mathematical model of ideodynamics for calculating the impact o f information on the population. This model was constructed on the premise that time trends o f opinion percentages could be predicted from the relevant messages available to the public. This model also has the important feature that i t can unify many seemingly c o n f l i c t i n g results. A useful analogy is the story o f b l i n d men reporting on an elephant; the man studying the leg could report that the elephant was like a tree trunk w h i l e the man examining the tail could find that the elephant was most like a rope. The contradiction vanishes when the entire elephant is considered in overview w i t h both the leg and tail being special cases o f the more general model w h i c h is the elephant. In the same way, the cumulative effects o f information can encompass both individual groups o f mass media messages having m i n i m a l effects and the totality o f the media having major effects. The unifying power o f ideodynamics derives importantly f r o m its quantitative nature. By g i v i n g numerical values to the contributions o f different phenomena, there is no need to assert or i m p l y that certain phenomena are always more or less important than others. Instead, the question becomes the relative importance o f different phenomena under specific circumstances. For instance, this book demonstrates that o p i n i o n f o r m a t i o n is frequently affected rather l i t t l e by reinforcement o f previous opinion due to the resolution o f cognitive dissonance i n the direction of favorable information. This statement does not deny the existence o f opinion reinforcement and does not assert that such reinforcement is never important. In fact, such reinforcement is e x p l i c i t l y included i n ideodynamics. Instead, the statement is merely that such reinforcement is small relative to the forces in the mass media causing opinion change for cases like the six studied in this book. The elephant analogy can be extended to the emphasis i n this book on the global behavior o f the population. The concern is less with the behavior o f subpopulations and selected media messages than on the effect o f the totality o f messages on attitudes w i t h i n the entire population. In the analogy, the theory is less concerned w i t h the
Introduction
5
behavior o f the parts o f the elephant during locomotion than w i t h the path taken by the elephant as a whole. There is no i m p l i c a t i o n that the elephant's path is more important to study than the effects o f the legs, for example, on elephant m o v e m e n t The analogy is only pursued to highlight the fact that this book is mainly about the macro effects o f the totality o f information on overall attitudes without a systematic dissection o f a l l c o n t r i b u t i n g factors, even t h o u g h some such dissections are performed The theory is also formulated w i t h very few parameters so that i t can be tested e m p i r i c a l l y . E m p i r i c a l testability means that confidence i n the model c o u l d be derived f r o m f i n d i n g that stories i n the Associated Press c o u l d give good time trend predictions o f public opinion percentages over time spans ranging from three months to nine years. The success o f applications to real data is crucially important because i t can demonstrate that approximations and calculated population parameters are reasonable, even though some might seem heroic at first glance. A m o n g the parameters examined, the most interesting lead to the conclusions that there is no lag before the onset o f persuasion and that the impact o f a mass media message decreases exponentially w i t h a half-life o f only one day. This means that the effect is entirely dissipated within a week. These results argue that there is no two-step transfer o f i n f o r m a t i o n f r o m the press to the populace v i a o p i n i o n leaders. Rather, the people are influenced directly by the mass media. Examination o f the equations also shows w h y the big lie can be effective i n propaganda, w h y the causes o f fringe groups can be helped by terrorism, and w h y the political Left and Right can both accuse the press o f unfair bias. T o be consistent w i t h the previous discussion on the importance o f quantitative assessments, these parameters might have other values i n future studies, resulting i n different implications for other circumstances. The mathematical predictability o f opinion indicates a large public malleability in the hands o f the mass media. This malleability is l i k e l y to arise f r o m the law o f the 2 4 - h o u r day w h i c h is first i n t r o d u c e d i n Chapter 1. T h i s l a w s i m p l y acknowledges that the public is constantly bombarded by new information, w i t h so much being available that a person can only reflect carefully on a small fraction. As a result, most information is taken at face value. T h i s importance o f superficial information is at the very heart o f the words reputation and prejudice. These words both i m p l y d e c i s i o n m a k i n g based on observations or information from the past. B y b r i n g i n g such prior information to bear, an i n d i v i d u a l is spared the time and effort needed to make a careful detailed examination o f the current details o f the issue. I n fact, the time needed to make careful evaluations o f all current information simply may not be available. These considerations stress another o f the recurring themes i n this book, the importance o f real t i m e for examinations o f social issues. I t is not enough to describe pathways and sequences for social changes w i t h o u t an appreciation o f the time spent i n each step. For example, i t has already been mentioned that real-time constraints lead to superficiality i n decisionmaking for the population as a whole. Such superficiality is not likely to be observed when people are forced to ponder issues carefully i n laboratory studies, focus groups, and interviews where people are asked to reconstruct their states of mind. O b v i o u s l y , superficial thoughts are simpler to analyze mathematically than complex ones. Therefore, i t is reasonable t o use straightforward mathematical equations t o calculate public opinion f r o m persuasive messages. For a wide variety o f issues, like those mentioned above, persuasion is further due to i n f o r m a t i o n largely confined to the mass media. Throughout this book, the emphasis is on message impact w i t h little discussion o f message generation. This emphasis certainly does not mean that message senders w o r k i n a vacuum, oblivious to other factors, including actual or anticipated public o p i n i o n . I n fact, both Chapters 2 and 7 discuss how message generation i n the model can be dependent o n o p i n i o n . However, the interdependence o f o p i n i o n formation and message generation do not exclude these t w o phenomena f r o m being studied separately.
Cop*
6
Introduction
In the same way, even though a nuclear w a r can i n v o l v e an exchange o f missiles, it is s t i l l possible to study separately missile damage and missile launch. Indeed, a thorough understanding of missile impact w i l l aid in a complete analysis of nuclear war. By analogy, a complete description of the persuasive process can benefit f r o m a careful study of the impact o f communications on the populace. A n accurate description o f message effect can then be used as a f i r m base from which to continue the analysis o f message generation. The key to uncoupling missile launch f r o m missile impact is a valid description of the pertinent properties o f the missile, namely its trajectory and megatonnage. Once these properties are recognized, it is possible to model both missile launch and impact i n terms o f these parameters. Given the appropriate parameters, the analyst o f missile impact can predict the devastation w i t h o u t regard for the factors influencing the launch. I n the same manner, once persuasive messages are coded in terms o f the equivalents of trajectory and megatonnage, knowledge o f the sender's motives is not important i n considering message effect. A n important goal o f this book is t o develop and validate parameters which are sufficient to describe a persuasive message without regard for the message sender. A later analysis could then turn to message generation, w i t h the messages coded in the same terms. W h e n both message generation and impact are understood, then the trade o f messages in a persuasive process can be explored i n the same way that an exchange o f missiles can be examined for a nuclear war. The w o r k i n this book focused o n message i m p a c t rather than message generation because impact was likely to be more predictable. The law o f the 24-hour day argues that o p i n i o n w i l l usually reflect messages. I n contrast, o p i n i o n is more frequently o n l y one factor rather than the sole factor i n influencing the message sender. I n addition to o p i n i o n , new discoveries and facts can greatly affect the messages broadcast I f not, nothing new w o u l d ever be disclosed by the mass media since the very novelty o f a discovery must mean that very few people are aware o f it and hence that there is very little opinion favoring the dissemination o f this rare event. Therefore, a thorough analysis o f the dissemination o f mass media messages must include n o t o n l y an examination o f o p i n i o n but new events w h i c h are unpredictable by their very nature. The foreseeable response o f the populace to information is clearly important for understanding social behavior for issues as t r i v i a l as fads and as profound as w a r and peace. For instance, the predictability and consequent superficiality o f information absorption suggest that the average member of the public i n modem democracies may make no more carefully reasoned decisions for most issues than persons i n more p r i m i t i v e societies. Furthermore, since the model should apply to all societies, the predictions should be as v a l i d i n dictatorships as i n democracies so long as a l l the information available to the public can be coded.
OUTLINE A s discussed above, this book explores the new mathematical m o d e l o f ideodynamics describing social responses to information. A l t h o u g h the outlines have already been published, the model has been m o d i f i e d as a consequence o f its application to empirical data, the focus o f this book. Therefore, this book begins w i t h a presentation o f ideodynamics followed by an examination o f the ability o f the model to incorporate previous theories (Chapters 1 and 2). Then data applications are considered (Chapters 3 to 5). A t the end (Chapters 6 and 7), there is a discussion o f the conclusions to be drawn from the w o r k . The e m p i r i c a l testing o f the model involves its use to predict p u b l i c o p i n i o n f r o m i n f o r m a t i o n i n the mass media. The computed o p i n i o n is i n the f o r m o f percentage support o f polled positions w i t h the values appearing as continuous time trends calculated every six or twenty-four hours. Therefore, to the extent that p o l l s are like snapshots, the trends from these new computational methods are like moving
Copyrighte i
Introduction
7
pictures, capable o f filling in the gaps between actual p o l l points and extending opinion estimates t o times when polls have not yet been taken. The studies i n this book show (hat media exemplified by Associated Press ( A P ) s t o r i e s - c a n be used t o predict the o p i n i o n percentages published by reputable national polling organizations such as A B C News, one o f the major p o l l sources for this book. Successful projections were made for all six o f die issues analyzed. For each issue, the procedure consisted of: (1) gathering the texts o f A P dispatches relevant to the issue, (2) scoring each story for the extent to w h i c h i t supported different positions w i t h i n the issue, (3) using these scores i n the equations o f ideodynamics to compute o p i n i o n time trends, and (4) comparing the computed time trends w i t h published poll data. Each study used the new I n f o T r e n d ^ methods. The first step i n the InfoTrend procedures relies o n a previously unreported computer procedure for scoring A P stories. T h e second step employs computer solutions f o r the equations o f ideodynamics.
mass
messages-as
ORGANIZATION Chapter 1 describes the deduction o f ideodynamics f r o m know n phenomena i n the area o f persuasion. A n appreciation o f this chapter is essential for understanding the opinion computations. Ideodynamics considers persuasive messages to have structures similar to that o f M I R V e d missiles. The analogs o f the independent warheads ( M u l t i p l e Independent Reentry Vehicles) are message components, each one able to have an impact o n appropriate target subpopulations. These message components are called infons. For example, a persuasive message relevant to defense s p e n d i n g could have one i n f o n or component favoring more spending, another infon favoring same spending, and yet another favoring less spending. L i k e M I R V e d missiles, a l l the infons are bundled together i n the same persuasive message and launched at the p o p u l a t i o n . T h i s chapter models mathematically the effects o f infons on the population. Chapter 2 discusses the m a i n features o f ideodynamics i n the context o f previous models for the impact o f information on society, especially those in the area o f public o p i n i o n . Therefore, readers p r i m a r i l y interested i n the new methodology can skip this chapter. Chapter 3 describes the data used for the calculations and therefore should be read. Chapter 4 describes the new InfoTrend computer method for obtaining infon scores for the messages discussed i n Chapter 3. Readers less interested i n computer content analysis than opinion projections need not read Chapter 4. This chapter is free-standing, describing a general technique o f content analysis able to score any text for the extent to w h i c h different ideas are favored. The methodology is not restricted to generating i n f o n scores s u p p o r t i n g different positions. F o r example, i t is also possible to use this text analysis f o r other purposes such as assessing whether a letter o f recommendation comments favorably on specific traits for a person being discussed. Since the major function o f Chapter 4 is to produce infon scores, i t is possible to bypass this chapter, for o p i n i o n projection studies, by using alternate scoring procedures. The most straightforward way w o u l d be to ask human judges to score the persuasive messages. However, the computer methods do have distinct advantages: the critical features o f the persuasive text are explicitly defined; large amounts o f text can be scored; a l l scoring criteria are applied u n i f o r m l y to the entire body o f text examined. I n f o T r e n d is a registered trademark for i n f o r m a t i o n a l analysis b y I n f o T r e n d I n c .
8
introduction
describes h o w infon scores for persuasive messages are used to compute expected public opinion and is the heart o f this book f r o m the standpoint o f opinion projections. Each o f the six studies is considered i n detail. M o s t o f the results fall into four major categories: 1. A set o f graphs describing the t i m e trends o f persuasive i n f o r m a t i o n favoring different positions, 2. A set o f graphs comparing published o p i n i o n - p o l l results w i t h o p i n i o n calculated on the basis o f infon scores and the first set o f published opinion percentages, 3. A set o f graphs showing the optimizations o f the various constants i n the Chapter
5
ideodynamic equations, and
4.
A table showing the goodness o f fît based on the squares o f the differences between the poll projections and the opinion percentages i n published polls. Chapter 6 examines the implications o f the studies and further applications o f the method. Strictly speaking, this chapter need not be read by those interested only in the technical aspects o f the methodology. However, this chapter is useful even for methodological considerations since i t examines both the strengths and robustness o f the techniques as w e l l as their weaknesses and limitations. Chapter 7 discusses the broader significance o f the w o r k i n this book to theories o f effective persuasion and examines the procedures by w h i c h ideodynamics can be extended to include theories o f message generation. Appendices. This book is written so that the reader can f o l l o w the main thrust of the arguments without a detailed study o f either the mathematics of ideodynamics or the computer text analyses. However, both o f these technical areas are explained more f u l l y i n the appendices: Appendix A for the mathematics o f ideodynamics, A p p e n d i x C for the computer text analyses, and A p p e n d i x D for the computer calculations o f opinion based on ideodynamics. The primary data for the analyses are also presented i n Appendix B . References are made to the appendices throughout the text Further technical details o f the procedures and computer programs used for this book are given in a pending patent application.
Copyrightec
1
Formulation of Ideodynamics
The m a i n thesis o f this book is that information controls public o p i n i o n . For many issues i n a free democracy, the d r i v i n g force f o r o p i n i o n change is persuasive messages i n the mass media. Support for this thesis derives f r o m the ability t o use messages i n the press to calculate time trends o f public opinion. The calculations are performed by computer using the new InfoTrend methods, and are divided into t w o main areas, those for text analysis and those f o r assessing the impact o f information on o p i n i o n . The studies are grounded in a general model for information impact applicable to the adoption o f both behaviors and attitudes. I n this book, however, the discussion w i l l focus o n attitudes since the applications are restricted to p u b l i c o p i n i o n . Appendix A presents an improved version o f the mathematical model w h i c h has already been c a l l e d i d e o d y n a m i c s ( F a n , 1984, 1985a, 1985b). T h e name ideodynamics is d r a w n f r o m idea w h i c h refers t o ideas, and dynamics, which emphasizes changes w i t h time. I n order t o present the arguments w i t h o u t undue distractions, this chapter discusses the formulation o f ideodynamics w i t h m i n i m u m reference to alternative models. Relationships to other models are discussed i n Chapter 2. Ideodynamics was developed to explain a number o f k n o w n features concerning the formation o f public o p i n i o n . Therefore, the model is deduced f r o m phenomena which needed to be explained and shares the deductive approach used by other workers like Downs (1957) i n An Economic Theory of Democracy.
1.1 S T R A T E G I E S U S E D I N F O R M U L A T I N G
IDEODYNAMICS
One o f the essential considerations i n f o r m u l a t i n g ideodynamics was that the model should be testable using data f r o m observations. This condition is important since, as w i t h any mathematical model, simplifying approximations are needed. The p r e d i c t i v e powers o f a model p r o v i d e a good test o f its soundness. I f the approximations are v a l i d for a large number o f circumstances, the model should successfully predict measured values for many cases. I f the approximations are appropriate f o r o n l y a small number o f examples, then the predictions f r o m the model should frequently fail. Therefore, empirical testability provides a method f o r assessing the validities o f the approximations. A s w i t h a n y set o f s i m p l i f i c a t i o n s , i t is a l w a y s possible t o i m a g i n e complications w h i c h w i l l lead to failure o f the approximations. Nevertheless, the model can succeed i n a large number o f instances i f the complications usually make o n l y m i n o r contributions w i t h i n the total constellation o f relevant phenomena. The usefulness o f the simplifications w i l l depend on the extent to w h i c h the resulting
Copyright
10
Predictions
of Public
Opinion from
the Mass
Media
predictions are accurate. T o guard against the possibility that the accuracy is fortuitous, the model can be tried under a variety o f conditions. The calculations w i l l gain in robustness as acceptable predictions continue to be obtained. The important advantage of empirical testability is that a crucial criterion for a model's success can be the predictions obtained. I n this way, the v a l i d i t y need not rely solely o n the plausibility o f the argument. T e s t a b i l i t y , however, is clearly a two-edged s w o r d . A l t h o u g h accurate predictions can argue that imagined concerns are o f m i n o r importance, consistently inaccurate predictions would also force abandonment of the model. Once a model can be tested empirically, it is both possible and desirable to be bold in postulating s i m p l i f y i n g approximations. After a l l , i f the simplifications are too extreme, then the model w i l l fail to give useful predictions. Therefore, a reasonable strategy is to begin w i t h the m i n i m a l model i n v o l v i n g the smallest number o f parameters. More complicated approximations involving more parameters w o u l d only be added i f the m i n i m a l model d i d not eive good predictions. Another important advantage o f using simple approximations is that the mathematics and resulting computations are less complicated. Not only w o u l d the procedure be simpler to understand, but fewer errors w o u l d be made i n f o r m u l a t i n g the mathematical theory and i n p e r f o r m i n g the resulting calculations. W i t h these considerations, ideodynamics was developed using q u i t e simple approximations. For many public issues, the population was assumed to f o l l o w blindly the information in the mass media. Interestingly, this simple model d i d give reasonable opinion projections suggesting that the media is not o n l y responsible for setting the agenda (Cook et al., 1983; E r b r i n g , Goldenberg, and M i l l e r , 1980; Funkhouser, 1973a. 1973b; Funkhouser and McCombs, 1972; Iyengar, Peters, and K i n d e r , 1982; M c C o m b s and Shaw, 1972; M a c K u e n , 1 9 8 1 , 1984) but is also the key agent i n determining opinion. 1.2 N A T U R E O F T H E P O P U L A T I O N Since one of the major requirements was empirical testability, ideodynamics was structured so that tests could be applied using readily available data, namely those f r o m public opinion polls. The starting point for any o p i n i o n p o l l is a question relating to a particular issue. In ideodynamics, the issues are defined as they are in o p i n i o n surveys. In particular, issues are topics on w h i c h members o f the populace can each h o l d o n l y one o f t w o or more m u t u a l l y exclusive positions or ideas. For instance, the first issue i n this book concerns American public opinion on funding for military defense. This issue was defined as having only three ideas or positions, favoring m o r e , same, or less spending since these were the positions i n several published polls. Since public opinion polls divide people into subpopulations, each holding a different viewpoint, ideodynamics also divides the population into subpopulations along the same lines. However, the model makes a distinction between individuals unaware o f the issue and persons aware o f the topic and h o l d i n g an o p i n i o n . "Unawares" comprise a portion o f the N o O p i n i o n or Don't K n o w groups i n opinion polls. " A w a r e s " are subdivided into those holding each o f the permitted answers to polled questions. The Don't K n o w s might also include some w h o are aware but are undecided. The defense spending analysis i n this book ignored the N o Opinions, including both the unawares and the awares but undecided, because the N o Opinions were few i n number, usually comprising less than 10 percent o f the total population. That left three subpopulations of awares supporting more, same, or less spending. The differences in treatment between awares and unawares are discussed in Appendix A and later i n this chapter.
Copyrighte
1
58
Predictions
of Public
Opinion from
tlic Mass
Media
gradual declines i n Figure 1.1 were compressed into spikes due to the condensed time scale o f seven years. The separate sets o f persuasive force functions belonging t o the t w o sets o f infon scores were used independently for opinion calculations. For both sets o f functions, i t was assumed that A P dispatches c o u l d represent all the relevant information available to the public, that all scores had the same weights, that the score for a position only contributed to the persuasive force for that position, and that the persistence constant had a one day h a l f - l i f e . The major difference between the t w o sets o f plots was the deletion o f infons favoring same spending. M o s t o f those infons were partitioned between the upper and lower curves o f Figure 5.1. T o use the persuasive force functions, i t was necessary t o devise a population conversion model as was already done for the defense spending analysis (Figure 1.2). When used for infons scored for more, same, and less spending, all three persuasive forces were used i n the calculations. W h e n the same model was applied to content scores for infons favoring o n l y more and less spending, the persuasive force function favoring same spending was always zero. The ideodynamic equations corresponding to the population conversion model were then w r i t t e n (Chapter 1 and Appendix A ) . T o use these equations, i t was first necessary to set their parameters: the persistence half-life, the modified persuasibility constant, and any required r e f i n i n g weights. I n i n i t i a l trials, the r e f i n i n g weights were all set to 1.0 corresponding to the approximation that all i n f o n content scores had the same weight. This was a safe strategy because the refining weights usually differed very little f r o m 1.0. T o set the other constants, o p i n i o n time trends were calculated using arbitrary values for the persistence half-life and the modified persuasibility constant. A t each t i m e corresponding to that o f an actual p o l l , deviations w e r e computed between calculated opinions and the actual values starting w i t h the measured opinions at the time o f the first p o l l (Appendix A , Equation A . 2 6 ) . The squares o f these deviations were calculated for all p o l l values, and averaged to g i v e the Mean Squared D e v i a t i o n ( M S D ) . The chosen persistence and m o d i f i e d persuasibility constants were those g i v i n g the m i n i m u m M S D . R e f i n i n g weights different f r o m 1.0 were o n l y tested i f the predictions were systematically h i g h or l o w for one or more o f the o p i n i o n positions. I f a r e f i n i n g w e i g h t gave a s i g n i f i c a n t i m p r o v e m e n t i n the M S D , then the persistence and modified persuasibility constants were reoptimized for the new r e f i n i n g w e i g h t s ) . The f i n a l constants were those g i v i n g the least M S D for a l l constants, unless otherwise stated. Rather than s i m p l y c o m p u t i n g the M S D f o r every set o f trial constants, t i m e trends o f o p i n i o n projections resulting f r o m a number o f arbitrary values for the constants were t y p i c a l l y plotted to examine q u a l i t a t i v e l y the consequences o f decreasing particular constants. Based on these plots, i t was clear that persistence half-lives m u c h longer than a day usually meant that the p o p u l a t i o n w o u l d have responded to media information more s l o w l y than was actually f o u n d . A l s o , as the m o d i f i e d persuasibility constant increased, the fluctuation i n o p i n i o n calculations became larger and larger around a general time trend. T h i s result was expected since a larger persuasibility constant corresponds t o more v o l a t i l e issues, w i t h more people being persuaded for the same amount o f information. G i v e n these q u a l i t a t i v e observations, systematic trials for the persistence constant t y p i c a l l y started w i t h a one day h a l f - l i f e . T h e n , additional values were tested, increasing iteratively by factors o f t w o until M S D values five to ten times the M S D at the one day half-life were reached. Test values for the persuasibility constant usually began w i t h values very close t o zero (e.g. 0.001 per A P paragraph per day), corresponding to a population being impervious t o persuasion, and then increased i n t w o f o l d steps beyond the value for w h i c h the m i n i m u m M S D was reached. I n the region where different parameter values gave approximately the same M S D , values were tested on a finer linear scale between the t w o f o l d j u m p s . For all p a r a m e t e r s , the M S D was plotted against trial values o f the p a r a m e t e r s (e.g., Figures 5.4 and 5.11). For these plots, the values for the other parameters
increasing or
Copyrightec
Projections
of Public
Opinion
59
those g i v i n g the m i n i m u m M S D . I n this way, the reader can assess the sensitivity o f the M S D to changes i n the t r i a l parameters. Where the optimization curves are steep, in the range o f a m i n i m u m , the M S D is very sensitive to parameter changes. Where the curves are shallow, w i d e variations i n the parameters w i l l have relatively s m a l l effects o n the M S D . I t is possible to read f r o m the o p t i m i z a t i o n curves the amount o f permissible variation i n a parameter before the M S D increased beyond the tolerance limits set by the analyst. The values discussed i n the remainder o f this chapter w i l l refer t o the values o f the parameters at the m i n i m a o f the optimization curves. T u r n i n g away f r o m general strategy and t o w a r d the specifics o f the projections using infons scored for more, same, and less spending, i t was found that the optimal modified persuasibility constant was 0.6 per A P paragraph per day. Expected opinion throughout the time period o f message collection was calculated using this optimized constant, the i n i t i a l p o l l values, and the persuasive force curves i n F i g u r e S.l computed using the best value for the persistence half-life, one day ( A p p e n d i x A , Equations A.2o and A.29). The expected o p i n i o n is plotted together w i t h actual poll data i n Figure 5.3. The one day value for the o p t i m a l persistence half-life was f o u n d by p l o t t i n g t r i a l values for the h a l f - l i f e versus the M S D (Figure 5.4, l o w e r frame). This o p t i m i z a t i o n p l o t shows that the M S D was s t i l l decreasing as the h a l f - l i f e was shortened to one day. I t was conceivable that an even shorter half-life w o u l d have been appropriate. However, i t seemed unreasonable to set the half-life much shorter than one day since there was at least that much ambiguity i n the t i m i n g o f the p o l l points and i n the t i m i n g o f the A P dispatches. The optimization curve f o r the persistence constant d i d , however, show a second m i n i m u m over 100 days before a rapid increase above that t i m e . The precise explanations for the steep rise at long half-lives are not clear. The upper frame o f Figure 5.4 shows the o p t i m i z a t i o n for the m o d i f i e d persuasibility constant. This constant clearly gives a single, well-behaved m i n i m u m M S D at 0.6 per A P paragraph per day. Comparison o f the projections i n Figure 5.3 w i t h actual p o l l points did not indicate that scores favoring any position were either systematically too high or too l o w . Therefore, all refining weights were left at the value o f 1.0 used for o p t i m i z i n g the persistence and m o d i f i e d persuasibility constants. These values meant that all infons were given the same weight. For the projections o f Figure 5.3, o n l y 692 o f the 9,314 identified dispatches were studied. It was conceivable that smaller samples c o u l d g i v e estimates w h i c h were j u s t as good. Therefore, the 692 stories were d i v i d e d into t w o approximately equal, random subgroups o f 325 and 383 dispatches each. O n l y sixteen dispatches i n one group were also present i n the other. U s i n g the m o d i f i e d persuasibility constant optimized above, opinions were recalculated using the t w o dispatch subsets (Figures 5.5 and 5.6). N o t surprisingly, there were greater deviations between the predictions and the p o l l results w i t h the smaller sample sizes. These differences c o u l d be seen quantitatively by the increase f r o m 7.2 p o l l percent for the total dispatch set to 9.4 percent and 10.3 percent for the t w o subsets (Table 5.1). Yet another projection was made to explore what w o u l d have happened to a subpopulation comprised only o f those favonng more defense spending. Therefore, the full set o f infons and the optimized modified persuasibility constant and u n i f o r m weights were used to remake the projections assuming that the i n i t i a l population only had people supporting more spending (Figure 5.7). The calculation snowed that after three years, the subpopulation should have behaved much as the population as a whole. T h i s result was significant technically because it meant that there was no need to account for the statistical errors inherent to the first p o l l point. The calculations u l t i m a t e l y h o m e d to the values dictated b y the i n f o r m a t i o n structure. The calculations for later times were not adversely affected even by extremely inaccurate i n i t i a l conditions. I f the errors i n the i n i t i a l p o l l p o i n t were not large, then the achievement o f the proper values w o u l d have occurred much more rapidly. were
Copyrighted material
60
Predictions
of Public
Opinion from
the Mass
Media
So fax, the discussion has concerned o p i n i o n calculations made f r o m infons scored for the three positions of more, same, and less defense spending. In addition, projections were also made for infons scored only for more and less spending (Figure 5.8). F o r these projections the data were f r o m all 692 dispatches. T h e same modified persuasibility constant o f 0.6 p o l l percent per A P paragraph per day was used for a l l infons so that the projections c o u l d be compared directly. A g a i n , all scores were given the same weight. The square root o f the M S D ( R M S D ) was 8.3 poll percent for the t w o - p o s i t i o n model i n contrast to 7.2 percent for the threeposition model. Therefore, the two-position scores gave slightly less accurate results than the three-position scores. I n the calculations f r o m both the t w o - and three-position scores, the projected time trends o f public o p i n i o n appeared to move i n steps since the time between infons was usually large relative to the week d u r i n g w h i c h an infon had its e f f e c t This was reasonable since o n l y 7 percent o f the total identified dispatches were studied. On an expanded scale, each step w o u l d have had the shapes in Figure 1.3. The steps were also much less tall because the modified persuasibility constant had a value over 3,000 times smaller i n Figures 5.3 and 5.5-5.8, than i n Figure 1.3. U s i n g infons scored for either two or three positions, the time courses for all three opinions favoring more, same, and less spending followed quite w e l l the main features o f the actual poll data. The change was most dramatic for people favoring more spending. There was a dramatic rise in o p i n i o n from 1979 to 1980 and an equally impressive drop f r o m 1981 to 1982. B o t h the timing and the magnitudes o f the actual changes were mirrored i n the calculated opinion. Comparison o f the opinion projections w i t h the i n f o n force curves (Figures 5.1 and 5.2) showed that the rise i n opinion favoring more defense spending i n 1979 was due to the great increase i n i n f o r m a t i o n favoring this position. D u r i n g this t i m e there was no d i m i n u t i o n i n messages arguing for less spending. T h e subsequent drop in support for more spending was not due to the disappearance o f messages favoring this idea. There was instead a significant augmentation in opposing messages. Besides permitting the calculation o f the best m o d i f i e d persuasibility constant, the optimization curve for this u n k n o w n (Figure 5.4, upper frame) also shows that the projected values are much better than w o u l d have been predicted by the model that o p i n i o n had stayed constant throughout the p o l l i n g period. The c o n d i t i o n o f n o o p i n i o n change is equivalent to a very small value for the modified persuasibility c o n s t a n t W h e n this constant is zero, the population is completely resistant to i n f o r m a t i o n and w i l l never undergo o p i n i o n change regardless o f the presence o f persuasive messages. Therefore, a very l o w modified persuasibility constant such as 0.001 per A P paragraph per day gave three almost unchanging o p i n i o n curves throughout the seven year p e r i o d - f l a t plots corresponding to the curves i n Figure 5.3. F r o m the optimization curve o f Figure 5.4, the M S D was over 250 poll percent squared for k'2 - 0 . 0 0 1 . The corresponding value o f around 50 p o l l percent squared for the projection using the best persuasibility constant was less than 1/5 as large. This decrease i n the M S D meant that the ideodynamic f i t was m u c h better than the model of no opinion change. For comparison, 1,000 simulations were made for predicted p o l l values drawn at random. For each simulation, the M S D was calculated for the differences between actual p o l l values and r a n d o m , predicted p o l l points. T h e p r o b a b i l i t y that the ideodynamic predictions were no better than chance was ascertained by counting the number o f simulations among the 1,000 where the M S D f r o m random p o l l results was smaller than that f r o m the ideodynamic predictions (right-hand c o l u m n o f Table 5.1). F o r completeness, R M S D values were computed b o t h for the r a n d o m and ideodynamic estimates. The average value among the 1,000 simulations is also given i n Table 5.1 (second column). Because there were 1,000 independent draws, it was also possible to compute a standard deviation for the M S D , and f r o m this standard deviation, the number o f standard deviations f r o m the random M S D to the ideodynamic estimate (Table 5 . 1 , third c o l u m n ) .
Copyrighted material
Projections
of Public
Opinion
61
The chances that the calculated fit was no better than that o b t a i n e d f r o m random p o l l points were less than 0.001 (Table 5.1). Therefore, the ideodynamic projection is statistically m u c h better than those obtained by either no o p i n i o n change or by a random choice o f p o l l values. I t s h o u l d be noted that both p o l l points and projected o p i n i o n values were correlated w i t h each other as time proceeded, since o p i n i o n at a later t i m e was dependent to some extent on opinion at earlier times. Therefore, i n the absence o f s i m p l i f y i n g a p p r o x i m a t i o n s w h i c h are d i f f i c u l t t o j u s t i f y r i g o r o u s l y , i t was inappropriate to calculate r2 regressions requiring time independence for the p o l l values. Ideodynatnics postulates that the parameters i n the opinion projection equations should be constant, changing very little over the time period o f the calculations. I f this is true and i f language usage remained the same, then i t should have been possible to extend the opinion projections to a later date simply by retrieving more A P dispatches and running the same programs under the conditions used for the studies f r o m 1977 to 1984. T o test this hypothesis, the Nexis data base was searched f r o m January 1, 1981 t o A p r i l 12, 1986. using the commands first e m p l o y e d for defense spending. F r o m the 10,451 dispatches identified, 1,067 were retrieved at random and analyzed using the same text analysis described earlier for the three positions favoring more, same, and less spending. The o n l y change was a single alteration i n the dictionary for the first filtration step. After 1984, the disease o f acquired immune deficiency syndrome became much more prominent. Therefore, its acronym A I D S was found i n a significant number o f dispatches describing spending for defense against A I D S . Consequently, i t was necessary to eliminate dispatches i f they contained the w o r d A I D S , and this w o r d was added t o the dictionary i n the first filtration together w i t h the words " f u n d " and " a i d . " When any o f these words appeared i n a dispatch, the story was eliminated as being irrelevant to American military spending. A f t e r the text analysis, 507 stories had non-zero scores f o r defense spending. This was about half the initial number o f dispatches and was not very different f r o m the 39 percent f o u n d i n the analysis for stories f r o m 1977 to 1984. Perhaps the 39 percent was a little l o w , since text ceased to be collected whenever the reader felt that a story was u n l i k e l y t o be about defense spending. Therefore, late mentions o f defense spending w o u l d have often resulted i n discarded articles i n the first set o f retrievals. T h i s manual interference i n the collection d i d not occur for the retrievals from 1981 to 1986. Since neither set o f retrievals included all possible stories, and since there was overlap between the t w o data sets, all paragraph scores were corrected to the expected value corresponding to ail dispatches being collected. For instance, i f o n l y 1/10 o f all dispatches were collected at random i n a time period, all paragraph scores i n that period w o u l d have been m u l t i p l i e d by ten. The persuasive force curves i n c l u d i n g these corrections are shown i n Figure 5.9. These data showed that there was very little change i n the i n f o r m a t i o n structure f r o m 1982 to 1986, w i t h the ratios o f information favoring the three positions staying relatively constant. Based on these results, i t was expected that o p i n i o n w o u l d also be quite stable. This was indeed the case both for projected and measured o p i n i o n (Figure 5.10). Stability i n the p o l l results c o u l d be seen even though there were fluctuations in the data f r o m different polling organizations after 1982. Despite this scatter, i t seemed plausible that o p i n i o n f a v o r i n g more spending might have been systematically overestimated after 1983. Therefore, a least squares optimization was performed over the entire time period f r o m 1977 to 1986 to see i f a better projection c o u l d be obtained by g i v i n g information favoring less spending a greater weight. F r o m the optimization curve i n Figure 5 . 1 1 , a weight o f about 1.2 gave a marginally better fit to the p o l l points. Given the small effect o f the weight increase (decrease i n M S D o f less than 1/20), i t was not used in the computations for Figure 5.10. One interesting result o f this optimization is that the small improvement i n the fit required a greater w e i g h t for data f a v o r i n g less spending. That meant that
Copyrighted material
62
Predictions
of Public
Opinion from the Mass
Media
information opposing more defense spending was more effective than i n f o r m a t i o n supporting more spending. Since many pro-spending infons came f r o m the Reagan administration, the suggestion is that a popular president and his administration were i n fact less effective than the opposition i n swaying public o p i n i o n , a result different f r o m that obtained by Page and Shapiro (1984) and Page, Shapiro, and Dempsey (1987). One possible explanation for the underestimate i n o p i n i o n favoring less defense spending was the presence o f additional indirect i n f o r m a t i o n not included i n the analyses described above. A good candidate seemed to be stories on waste, fraud, and corruption by defense contractors. Therefore, additional dispatches were retrieved focusing on these topics and all paragraphs discussing these issues were scored as favoring less defense spending. A l l these paragraphs were given the same weight as paragraphs direcUy supporting less defense spending and a persuasive force curve based on these waste and fraud infons alone was constructed using the usual one day half-life (Figure 5.12, top frame). I n c l u s i o n o f these extra infons i n die o p i n i o n computations meant that their persuasive force curve was added to the persuasive force curve for infons directly advocating less spending (Figure 5,9, b o t t o m frame). Q u i c k inspection shows that the waste and fraud infons (Figure 5.12, top frame) were negligible w h e n compared to the direct infons favoring less spending. Therefore, there was very litde difference between the persuasive force curves w i t h (Figure 5.12, center frame) and w ithout (Figure 5.9, bottom frame) the added infons. This sameness i n persuasive force meant that there was very little difference between the o p i n i o n projections f o r more spending w i t h and w i t h o u t considering waste and fraud ( t w o lines i n bottom frame, Figure 5.12). Therefore, b y taking into account all relevant i n f o r m a t i o n , an ideodynamic analysis was able to suggest that opinion on defense spending was not greatly influenced by information o n waste and fraud. The o n l y caveat to this interpretation is that the public might have weighted this information much more heavily than information directly speaking to the issue. Unfortunately, there was no direct method to test this possibility.
5.2 O P I N I O N P R E D I C T I O N S FOR T R O O P S I N L E B A N O N The issue o f troops in Lebanon was like defense spending in that o p i n i o n both increased and decreased significantly. As usual, opinion calculations began w i t h the construction o f infon persuasive force curves using A P dispatches scored for favoring more, same, or less troops. Then computations o f poll percentages were made w i t h the m o d i f i e d p e r s u a s i b i l i t y c o n s t a n t b e i n g the o n l y v a r i a b l e p a r a m e t e r . U n f o r t u n a t e l y , a good f i t was not o b t a i n e d . E x a m i n a t i o n o f the projections suggested that t w o modifications could improve the calculations. F i r s t o f a l l , o p i n i o n f a v o r i n g less troops seemed t o be s y s t e m a t i c a l l y underestimated so scores favoring this position were given a r e f i n i n g weight o f 1.6 by least squares o p t i m i z a t i o n . Persuasive force curves i n c l u d i n g this weight for scores opposing more troops and a weight o f 1.0 for other infons are plotted i n Figure 5.13. Furthermore, o n October 23, 1983, there was the unexpected explosion by a terrorist o f a truck laden w i t h explosives i n the headquarters o f the U n i t e d States Marines i n Beirut, k i l l i n g over 200 soldiers. I t seemed reasonable to suppose that the population reacted viscerally to this report, feeling that some action was required, either p u t t i n g more troops i n or p u l l i n g the ones there o u t . Therefore, a new persuasive force curve was computed i n c l u d i n g eighty paragraph equivalents f o r i n f o r m a t i o n f a v o r i n g more troops (Figure 5.14, lower frame) and no paragraphs f a v o r i n g t r o o p r e m o v a l . For reference, the persuasive force w i t h o u t this truck bombing infon is replotted f r o m Figure 5.13 (Figure 5.14, top frame). One difference between the plots o f Figures 5.13 and 5.14 is the beginning time for the curves. Figure 5.13 illustrates the fact that data retrievals always began at least six months before the date o f the first p o l l d a t e unless that w c a s b e f o r e J a n u a r y 1, 1977, the beginning date o f the Nexis data base for the AP. This was to assure
Copyright
Projections
of Public
Opinion
63
that the residual effects o f prior messages were included i n the o p i n i o n calculations. For the persuasive force curves described above, the universal persistence half-life o f one day was used. The o p i n i o n computations themselves then r e q u i r e d the f o r m u l a t i o n o f a population conversion model. T h e best one f o r troops i n Lebanon was a " d i r e c t conversion model" (Figure 5.15) where individuals could move directly f r o m any one subpopulation t o another w i t h o u t passing through any intermediate subgroups, however transient I y. As was standard practice, the persuasive force curves i n Figures 5.13 and 5.14 were calculated using constants o p t i m i z e d b y m i n i m u m M S D . These constants included the modified persuasibility constant and the weight f o r scores favoring less troops (Figure 5.16), and the persistence constant (Figure 5.17). The effect o f news o f the truck bombing o n October 2 3 , 1 9 8 3 , was modeled b y injecting a r t i f i c i a l l y , o n that date, t w o infons o f u n k n o w n m a g n i t u d e , a truck b o m b i n g i n f o n supporting more troops and a truck b o m b i n g i n f o n f a v o r i n g less troops. T h e persuasive force curves f o r these t w o infons were assumed t o be characterized by the same one-day persistence. For comparability, the content scores for these truck bombing infons were given i n A P paragraph equivalents. W h i l e the o p t i m i z a t i o n for the truck b o m b i n g i n f o n f a v o r i n g more troops showed a marked improvement at a value o f eighty A P paragraph equivalents (Figure 5.16, b o t t o m frame), there was n o need even t o i n v o k e a truck b o m b i n g i n f o n favoring troop w i t h d r a w a l since the optimization showed that the fit d i d not improve s i g n i f i c a n t l y as more paragraphs were added up to about f o r t y A P paragraph equivalents (Figure 5.17, lower frame). A s the number o f paragraphs increased above this number, the fit got appreciably worse. As f o r defense spending, the optimization for the persistence constant showed t w o m i n i m a , one w i t h a one day half-life and one w i t h a half-life o f fourteen days (Figure 5.17, upper frame). The criterion o f m i n i m u m M S D meant that the lower m i n i m u m corresponding to a one day half-life was chosen. A s argued for defense spending, i t seemed unreasonable to have an even shorter half-life. T h e p o l l data f r o m October 2 3 , 1983, were o m i t t e d f r o m the o p t i m i z a t i o n calculations since that was the date o f the truck b o m b i n g i n f o n . G i v e n the r a p i d changes i n the i n f o n and p o l l data, accurate projections o n this date w o u l d have required that the infons and p o l l values be assigned t o specific hours, an impossible task given the uncertainties i n the timing o f both the infons and p o l l itself. E x a m i n a t i o n o f the o p i n i o n projection patterns shows that one o f the largest effects was due to the introduction o f the October 23 truck b o m b i n g i n f o n . W i t h o u t this i n f o n , there w o u l d have been little o p i n i o n change d u r i n g the entire p o l l i n g p e r i o d , i n disagreement w i t h the p o l l data (Figure 5.18). Once the i n f o n was introduced, o p i n i o n fit quite w e l l (Figure 5.19). Comparison o f projections w i t h and w i t h o u t the truck b o m b i n g i n f o n favoring more troops (Figure 5.20) showed that this i n f o n d i d have a very large effect immediately after October 23. However, w i t h i n a couple o f months the effect was effectively dissipated, since n e w o p i n i o n reflected A P i n f o r m a t i o n at later times. This result is l i k e that i n Figure 5.7 f o r defense spending, where o p i n i o n m o v e d more gradually to conform w i t h the information structure. As discussed above, there was n o need to postulate any truck bombing infons f a v o r i n g fewer troops. T h a t was because there was already a large amount o f i n f o r m a t i o n favoring fewer troops (Figure 5 . 2 1 , t o p frame). T h e i n t r o d u c t i o n o f forty more paragraphs o n October 23 d i d not significantly change the shape o f the persuasive force curve favoring less troops (Figure 5.21, lower frame). Therefore, there was also litde effect o n the o p i n i o n calculations (Figure 5.22). This result is quite like that for stories o n waste and fraud for defense spending, where the additional news was insignificant compared w i t h other infons supporting less spending. In contrast, the truck b o m b i n g i n f o n f a v o r i n g more troops had a very large impact (Figure 5.20) because there was very l i t t l e other news i n favor o f that position (Figure 5.14).
CoDvriahted material
64
Predictions
of Public Opinion from the Muss
Media
The i m p r o v e d f i t by the addition o f the truck b o m b i n g i n f o n was manifested both by visual inspection o f the o p i n i o n projections (Figure 5.19) and by the R M S D (3.5 percent w i t h the i n f o n and 9.1 percent w i t h o u t Table 5.1). Table 5.1 (righthand c o l u m n ) also shows that the best projection had a probability less than 0.001 o f being found by chance.
5.3 O P I N I O N P R E D I C T I O N S FOR T H E D E M O C R A T I C
PRIMARY
Besides being able to m o d e l policy i s s u e s , ideodynamics i s a l s o applicable to electoral situations such as the Democratic primary o f 1984. The dispatches for this topic were scored both by the p r o x i m i t y o f bandwagon words to candidate names and by a count o f the candidate names (Chapter 4). Therefore, persuasive force curves were computed for both analyses. The bandwagon analysis yielded curves both favorable and unfavorable to M o n d a l e , G l e n n , or Others (Figures 5.23 and 5.24) while o n l y infons mentioning the three candidates were computed i n the name-count analysis ( F i g u r e 5.25). A l l scores were g i v e n the same w e i g h t o f 1.0 and the persistence constant had a one day half-life. The first study used the bandwagon content analysis. Since the persuasive force curves for this analysis were for positions both favorable and unfavorable to all three groups o f candidates, neither the sequential population conversion model (Figure 1.2) nor the direct conversion model (Figure 5.15) was appropriate. Instead, a m i x t u r e o f both models was used (Figure 5.26). T h i s model was unique among the models i n this book i n i n c l u d i n g persons w i t h N o O p i n i o n . I n other models, these persons were ignored, equivalent to the approximation that the majority o f o p i n i o n changes involved those w h o already had an o p i n i o n . However, for the Democratic primary, it seemed unreasonable that information unfavorable to a candidate w o u l d cause a supporter to favor any particular one o f the other candidates. Rather, i t s e e m e d more plausible that t h e conversion w o u l d be to Undecided or N o O p i n i o n . Therefore, this category was i n c l u d e d f o r the b a n d w a g o n analysis. Information actually favoring a candidate was presumed t o be able t o draw recruits f r o m any other subgroup. This model had features o f both the direct and sequential conversion models. A person favoring a particular candidate c o u l d be convinced to favor another, either direcUy or by first becoming disenchanted and moving temporarily into the undecided pool. W h e n content scores f r o m the name-count analysis were used, there was n o information unfavorable t o a candidate and hence favoring N o O p i n i o n . Therefore, the same approximation was made as for the other studies i n this b o o k . That is, the N o O p i n i o n subpopulation was assumed to consist o f people w h o were unconcerned about the and were not responsive t o i n f o r m a t i o n about the c a m p a i g n . Consequently, a l l persons w i t h an opinion were normalized to 100 percent. As noted earlier, the inclusion or exclusion o f the N o Opinions was not crucial t o the final curves since they only comprised a r o u n d 10 percent o f the total population at m o s t W i t h the exclusion o f the N o Opinions, the direct conversion model used f o r troops i n Lebanon was the most reasonable (Figure 5.27) since i t c o u l d not be argued that support for any candidate should be preceded by support for another. For the o p i n i o n projections for the bandwagon analysis, the persuasive force curves (Figures 5.23 and 5.24) were constructed using die persistence half-life o f one day and the same weight for all scores. The remaining u n k n o w n was the modified persuasibility constant. Both the persistence and modified persuasibility constants were fixed by least squares o p t i m i z a t i o n (Figure 5.28, top t w o frames) using the population conversion m o d e l o f F i g u r e 5.26. T h e best m o d i f i e d persuasibility constant was 1.5 per A P paragraph. The actual poll projections using the optimized constants showed a reasonable fit (Figure 5.29), w i t h the most dramatic change being the almost t w o f o l d decrease i n support for G lenn. F o r c o m p a r a b i l i t y , the same persuasibility constant was used to compute the poll projections using the name-count analysis. These calculations were based on the
issue
Copyrighted material
Projections
of Public
Opinion
65
persuasive force curves o f Figure 5.25 and the model of Figure 5.27. The projections (Figure 5.30) were o b v i o u s l y unsatisfactory. T h i s is seen in the R M S D o f 23.2 percent, over five times greater than the R M S D for the bandwagon (Table 5.1). In fact, when an o p t i m i z a t i o n was performed t o choose the best m o d i f i e d p e r s u a s i b i l i t y constant f o r the name c o u n t analysis, i t was f o u n d that n o i m p r o v e m e n t s were possible b e y o n d the s i t u a t i o n o f no o p i n i o n change corresponding to the m o d i f i e d persuasibility constant having a value close t o zero (Figure 5.28, bottom frame). The persuasive force curves o f Figure 5.25 give the reason for the inaccuracy. A t essentially all times, there was a large excess o f name mentions for the m i n o r candidates (bottom frame) as compared to the t w o front runners. The actual numbers for the name counts are g i v e n i n Table 5.2. For example, Jesse Jackson was discussed i n the news at a frequency (20 percent) between that o f Glenn (17 percent) and Mondale (27 percent) due i n large part to his efforts to free a naval flier downed in Lebanon. Cranston's name mention frequency o f 13 percent was not far behind that o f Glenn, due principally to his advocacy o f a nuclear freeze. I n fact, this excess i n the mention o f m i n o r candidates was seen throughout the p o l l i n g period. Therefore, the model predicted a net movement away f r o m both major candidates toward the m i n o r ones. T h i s was seen i n the projected d r o p in support for both Glenn and M o n d a l e w i t h an accompanying rise i n the calculated popularity o f the Others (Figure 5.30). Since the projected d r o p for G l e n n was accompanied b y the w r o n g movements i n the other t w o curves, this f i t was fortuitous. F r o m the optimizations for the bandwagon analysis, the proper choice for the modified persuasibility constant gave an M S D over two-fold better than the estimate f r o m a very small modified persuasibility constant equivalent to no change i n public o p i n i o n . T h i s improvement was less than that for defense spending (fivefold) and troops i n Lebanon (tenfold). H o w e v e r , i t was unreasonable to expect as m u c h improvement because the p o l l values changed much more for those other examples, so i t was less appropriate for them to be approximated by no o p i n i o n change d u r i n g the p o l l i n g period. Since the polls changed m u c h less for the Democratic p r i m a r y , an estimate o f no o p i n i o n change gave a much better fit. Nevertheless, this t w o - f o l d increase was substantial. The chances o f obtaining such a good fit by chance were less than 0.001 (Table 5.1). The projection using the name-count analysis was quite unsatisfactory. As mentioned above, the M S D o f 540 was substantially worse than the prediction o f no o p i n i o n change ( M S D o f 39). Indeed, the estimate was so bad that i t c o u l d be obtained 34 percent o f the time f r o m random p o l l points (Table 5.1).
analysis
5.4 O P I N I O N P R E D I C T I O N S FOR T H E E C O N O M I C
CLIMATE
The economic c l i m a t e was the first o f t w o economic issues studied. A s for defense spending and the Democratic p r i m a r y , there was no need to make any adjustments by weighting the i n f o n content scores. The persuasive force curves for this issue (Figure 5.31) were computed using the usual one day persistence half-life. For this topic both the direct and sequential population conversion models were tried. A m o n g these, the sequential conversion model was better (Figure 5.32). W i t h a l l infons w e i g h t e d the same, the persistence constant and c o m m o n modified persuasibility constant were both set by least squares optimization (Figure 5.33). This procedure led to an approximately tenfold improvement in the M S D over the estimate f r o m no o p i n i o n change, as w o u l d be expected f r o m a p o l l series where the o p i n i o n values d i d vary significantly f r o m the initial value. This is seen i n the projection curves (Figure 5.34). As w i t h all other satisfactory computations, the probability o f such a fit by chance was less than 0.001 (Table 5.1).
Copyrighted material
66
Predictions
of Public
Opinion from
the Mass
Media
5.5
O P I N I O N PREDICTIONS FOR U N E M P L O Y M E N T VERSUS I N F L A T I O N
The second economic topic was unemployment versus inflation. T h i s issue was like that for troops i n Lebanon in that i t was necessary to give different infon scores different weights. By least squares optimization it was found that scores favoring the importance o f inflation should have a weight o f 1.4 w h i l e those f a v o r i n g equal importance should have a weight o f 0.5. The remaining infon group supporting the importance o f u n e m p l o y m e n t had the reference w e i g h t o f 1.0. Least squares optimization was also used to give an optimal persistence constant w i t h a one day half-life. The persuasive force curves reflecting these constants are presented i n Figure 5.35. As for the other examples where the p o l l positions ranged f r o m one extreme through the middle to the other, it was possible to test both the direct and sequential population conversion models. O f these, the best was the direct conversion model (Figure 5.36). Least squares optimizations were used to determine the best values for all constants used i n the construction o f the persuasive force functions and the modified p e r s u a s i b i l i t y constant f o r the reference i n f o n s f a v o r i n g the i m p o r t a n c e o f unemployment (Figures 5.37 and 5.38). O p i n i o n projections were then made using the o p t i m i z e d constants (Figure 5.39). W i t h the large o p i n i o n changes d u r i n g the p o l l i n g period, the estimate o f no opinion change was so poor that the ideodynamic calculation could give an improvement i n the M S D o f four- to five-fold (Figure 5.37, top frame). For this analysis also, there was a probability o f less than 0.001 that the fit could have been obtained by random poll points (Table 5.1).
5.6
O P I N I O N P R E D I C T I O N S FOR C O N T R A A I D
Contra a i d was the o n l y study in this book where o p i n i o n was f a i r l y static d u r i n g the time period o f the study. T h e text analysis for Contra aid was unique, having been performed separately both by Fan and by three graduate research assistants ( S w i m et al. i n Chapter 4 ) . Therefore, there were two sets o f infon scores (Figures 5.40 and 5.41). A l t h o u g h S w i m et al. gave scores to about t w i c e as many paragraphs, the t w o sets o f scores revealed essentially the same overall i n f o r m a t i o n structure. For both sets o f scores, least squares optimizations showed that infons opposing aid needed to have a much greater weight than infons favoring a i d . The optimized weights were similar, being 2.0 for the Fan scores and 2.4 for the S w i m et al. scores. A l s o , o p t i m i z a t i o n o f the persistence constant resulted in the best halflives having values greater than one day for both i n f o n sets. However, the one day half-life was satisfactory, so that was the value used for calculating the persuasive force curves i n Figures 5.40 and 5.41. Since there w ere only t w o positions, the o n l y reasonable population conversion model was that o p i n i o n favoring one side c o u l d convert people favoring the other (Figure 5.42). This was obviously the degenerate case where the direct and sequential conversion models collapsed into the same model. As w i t h a l l other examples studied, the optimization for the persistence constant for Contra aid had t w o m i n i m a , one at a one day half-life and the other ranging f r o m seven to over 100 days (Figure 5.43, bottom frame). Since the one-day m i n i m u m was c o m m o n to a l l six issues, i t was used as the consensus value for a universal persistence constant for all issues. The one day half-life was certainly reasonable since this was also the lowest m i n i m u m for the five examples besides Contra aid. I t was also more reasonable to set the persistence constant for the other five issues, f o r w h i c h the i n f o r m a t i o n structure and o p i n i o n both changed significantly. For Contra aid, there was little change i n either o p i n i o n ( A p p e n d i x B, Table B.6) o r the ratio o f favorable to unfavorable information (Figures 5.40 and 5.41). Therefore, much o f any different fit in the persistence constant c o u l d have been t o errors in the opinion p o l l s .
Copyrightec
Projections
of Public Opinion
67
H a v i n g j u s t i f i e d the use o f a one-day universal persistence constant based on six independent optimizations for six different topics, the final opinion projections used this half-life. These projections also used refining weights o f 2.0 and 2.4 for the t w o sets o f infons favoring less aid, as described earlier. The final curves (Figures S.44 and 5.45) showed relatively little change in opinion during the entire three-year period once the m o d i f i e d persuasibility constant was also set by least squares optimization (Figure S.43, top frame). Nevertheless, there was a substantial improvement i n the fit over projections o f almost no change at a l l after the first p o l l point since the M S D was more than three-fold lower than that for 0.001 per A P paragraph per day for the m o d i f i e d persuasibility constant. The probability o f having the quality o f fit shown i n Figures S.44 and S.4S by chance alone were again less than 0.001 (Table 5.1). It was gratifying to find that t w o quite different text analyses c o u l d both g i v e essentially indistinguishable o p i n i o n projections (Figures S.44 and S.4S). T h i s result is not surprising since both analyses showed that there were the same approximate ratios o f infons f a v o r i n g and opposing C o n t r a aid t h r o u g h o u t the polling period (Figures 5.40 and 5.41).
5.7 S U M M A R Y O F C O N S T A N T S U S E D I N P O L L P R O J E C T I O N S This chapter has presented opinion projections for six quite disparate issues. The interpretations o f these results are g i v e n i n the next t w o chapters. For those discussions i t w i l l be useful to have a list o f all the parameters w h i c h were chosen by least squares o p t i m i z a t i o n for each o f the cases (Table 5.3). A l l constants were o p t i m i z e d under the best conditions for the other constants except for the issue o f Contra aid, where optimizations were performed using a one-day persistence half-life. The constants w h i c h were optimized fell into the f o l l o w i n g categories: 1. Persistence constant-this constant measured the ability o f an A P infon to continue to exert its effect after the date o f the dispatch. As noted i n the previous section, the optimal value for this constant was a one day half-life for five out o f six issues and this same value was also satisfactory for the sixth. Thus the optimizations for the six issues yielded a universal one day half-life for the persistence constant. 2. M o d i f i e d persuasibility constants- tor modified persuasibility constants, the least squares optimal value was calculated for an arbitrarly chosen reference set o f i n f o n scores. A n y needed variations i n these constants for infons favoring different positions were incorporated into the r e f i n i n g weights. W h e n all w e i g h t i n g values were the same, as was the case for defense spending, the Democratic primary, and the economic climate, any one o f the positions could have been chosen as the reference position. 3. Refining weight the refining weight for a position was the constant b y w h i c h all infon scores for a position were m u l t i p l i e d before construction o f the persuasive force functions. T h i s w e i g h t i n c l u d e d both differences between the p e r s u a s i b i l i t y o f the p u b l i c for d i f f e r e n t positions and imperfections in the infon scoring (Chapter 1 and Appendix A ) . Therefore, when the public was as easy to persuade for all positions and when all infons were scored correctly, a l l refining weights had the value o f 1.0. I n some cases the weights were different for different infon scores. A n example was the issue o f troops i n Lebanon, where the weight was 1.6 for scores favoring less troops and 1.0 for a l l other scores. This meant that a paragraph favoring less troops was 1.6 times as effective as a paragraph favoring same or more troops.
Cc
al
68
Predictions
of Public
Opinion from tlie Mass
Media
5.8 S U M M A R Y O F S T A T I S T I C S FOR P O L L P R O J E C T I O N S Statistically speaking, the estimate for a value is the mean, w i t h deviations f r o m that mean being characterized by the standard deviation. The standard deviation for a set o f values is computed by taking the square root o f the squares o f the deviations o f the i n d i v i d u a l points f r o m the mean. S i m i l a r l y , i f the estimate f o r an o p i n i o n percentage is given by the ideodynamic prediction, then statistical deviations can be represented by the R M S D computed by taking the square roots o f the squares o f the deviations between actual p o l l measurements a n d the ideodynamics predictions. Therefore, assuming that the differences between t w o values are statistical, the R M S D is like the standard deviation, so that differences between the predicted and measured p o l l values should be w i t h i n one R M S D about 68 percent o f the time and w i t h i n t w o R M S D about 95 percent o f the time. The R M S D values i n Table 5.1 are the m i n i m u m values corresponding to the best parameters chosen. These values were 3.5 percent f o r troops i n Lebanon, 4.3 percent for the Democratic primary, 4.7 percent for Contra aid, 6.6 percent f o r the economic climate, 7.2 percent for defense spending (1977-1984), and 7.7 percent for u n e m p l o y m e n t versus i n f l a t i o n . T h e errors increased w i t h the t i m e span o f the projections. F o r instance, the most accurate computations were for troops i n Lebanon and the Democratic primary, where the p o l l series o n l y covered four and seven months respectively. The analyses for the other four examples spanned periods f r o m three t o seven years. I t is quite possible that text and its interpretation changed w i t h time so that the text analyses should have been modified as time proceeded. Also, the m o d i f i e d persuasibility constants m i g h t have been dependent o n time, changing slowly over a period o f years. For comparison, national polls frequently have reported errors due t o finite sample size i n the range o f 4 percent at the 95 percent confidence level. T h i s is equivalent t o t w o standard deviations f r o m the reported values. Therefore, their equivalent to the R M S D w o u l d have been about 2 percent instead o f the 3.5 percent to 7.7 percent i n Table 5 . 1 . However, besides sample size errors, there are also systematic errors i n the polls due to such factors as question w o r d i n g and question sequence. Therefore, the 2 percent standard error is a m i n i m u m error i n the o p i n i o n measurements. The importance o f these systematic errors is seen i n the poll data o f Figure 5.10, where p o l l s were taken at close i n t e r v a l s b y d i f f e r e n t , respectable p o l l i n g organizations. There are fluctuations i n the range o f 5-10 percent w h i c h may have been r e a l , b u t w h i c h m a y also have been due t o differences i n t h e p o l l i n g instruments. I f actual measurements o f o p i n i o n can have errors i n the range o f 5-10 percent, then the ideodynamic errors are i n the same range and may be as good as those f r o m opinion polls. F r o m the summary data i n Table 5 . 1 , i t is also clear that the best ideodynamic o p i n i o n projections had very little probability (less than 0.001) o f being obtained b y chance alone. I n a d d i t i o n , a l l projections were also better than the model o f no o p i n i o n change after the first p o l l p o i n t This was seen i n the improvements i n the M S D as the modified persuasibility constant increased above values near zero.
Copyrighted material
A l l computations used the constants i n Table 5.3 except that the persistence half-life was one day f o r a l l calculations, including the t w o for Contra aid. Table 5.1. Statistical
comparisons
for opinion projections.
Ideodynamic Estimate
1000 M o n t e Carlo Estimates (Assuming Random Poll Values)
Issue
Ideodynamic RMSD (in Poll Percentage Points)
Mean Monte Carlo RMSD (in Poll Percentage Points)
Number o f Sutndard Deviations from Ideodynamic M S D to Monte Carlo Mean M S D
Probability of Obtaining Ideodynamic MSD by Chance
Defense Spending (1977-1984) Scores M o r e , Same, Less 692 retrievals 325 retrievals 383 retrievals Scores M o r e , Less 692 retrievals
7.2 9.4 10.3
24.3 24.3 24.3
5.2 4.9 4.7
6
86
CoDvriahted material
Figure 5.16. Optimizations for the modified persuasibility constant, the weight for paragraphs favoring less troops, and the value of the truck bombing infon favoring more troops. T h e best modified persuasibility constant (k'2) g i v i n g the lowest M S D
was 4.5 per A P paragraph per day (top frame); the optimal weight for infons favoring less troops was 1.6 when all other infons had weights o f 1.0 (center frame); and the most favorable value f o r the October 2 3 , 1983, i n f o n f a v o r i n g more troops was equivalent t o e i g h t y A P paragraphs ( b o t t o m frame). F o r a l l calculations, the persistence half-life was one day, and there was assumed to be no truck b o m b i n g infons favoring less troops. I n addition, all optimizations were performed at the best values f o r the other constants.
Mean S q u a r e d
Deviation ( i n Poll
k ' 2 f o r Infons Mean _S q u a r e d D e v i a t i o n
Percent
Squared)
F a v o r MORS. SAMK ( p e r P a r a - D a ^ J ( i n P o l l P e r c e n t Squared)
Weight f o r P a r a s F a v p r L E S S T r o u p a Mean S q u a r e d D e v i a t i o n ( i n P o l l P e r c e n t S q u a r e d ) 80
BO 40
20
0 0.001
0 . 0 1
1
10
Truck
Bomb News F a v o r MORE T r o o p s
100
(AP P a r a E q u i v
87
Copyrighte
Figure 5.17. Optimization curves for the persistence half-life and the value of the truck bombing infon favoring less troops. T h e lowest M S D g i v i n g the best half-life
was one day (top frame). There was essentially n o change i n the M S D f r o m almost zero to almost f o r t y A P paragraph equivalents f o r the October 2 3 , 1983, truck b o m b i n g i n f o n f a v o r i n g less troops ( b o t t o m frame). B o t h o p t i m i z a t i o n s were performed at the best values for the other constants as indicated i n Figures 5.16 and 5.17.
Mean S q u a r e d
Dcviation
( i n Poll
Mean S q u a r e d
Deviation
LLn_.PoiJLPfrcfy?JLSqviaredi
88
Percent Squared)
Figure 5.18. Opinion on troops in Lebanon assuming only AP infons. The projections beginning w i t h the first p o l l point (solid lines) used the three persuasive force curves o f Figure 5.13, the model o f Figure 5.15, and the o p t i m a l constants o f Figures 5.16 and 5.17, w i t h the exception that the truck b o m b i n g infons favoring more and less troops were both o m i t t e d . Calculations were at six-hour intervals. A c t u a l p o l l points are plotted as squares. October 23 is indicated by the tic between October and November.
% Favor
MORH T r o o p s
(no
Bomb
Paras)
%
Favor
SAME
Troops
(no
Bomb
Paras)
*_Favor
LESS
Troops
(no
Bomb
Paras_^
Oct Years
Nov 1983-1984
Dec
Jan
89
Copyrighte
Figure 5.19. Opinion on troops in Lebanon with a truck bombing infon favoring more troops. The computations (solid lines) were the same as for Figure 5.18 except that a truck b o m b i n g infon o n October 23 favoring more troops at the o p t i m a l value o f eighty A P paragraph equivalents (Figure 5.16, b o t t o m frame) was also included. Therefore, the persuasive force curve favoring more troops was that i n the l o w e r frame o f Figure 5.14. T h e other persuasive force curves are i n the b o t t o m t w o frames o f Figure 5.13. X
Favor
MORS
Troops
(AP
»
Bomb
Favor
More
Troops)
SAME
Troops
(AP
+
Bomb
Favor
More
Troops)
LESS
Troopa
(AP
+
Bomb
Favor
More
Troops1
70 60 50 40 30 20 10 0 X
Favor 70 60 50
40 30 20 10 0 %
Favor 70 60 50 40 30 20 10
0 O c t ^ e ! i
r
Nov 1983-U H 4
Dec
Jan
90
Copyrightec
Figure 520. Comparison of opinion projections with (solid line) or without (dotted tine) the truck bombing infon favoring more troops. T h e curves i n Figures 5.18 and
5.19 are plotted together.
Favor
MORE
Troopa
( + . - Bomb
Favor
More
Troops!
X Favor
SAME
Troops
( + , - Bomb
Favor
More
Troops)
Favor
LESS
Troopa
( + . - Bomb
Favor
More
Troopa)
X
X
Oct Years
Nov 1983-1984
Dec
J a n
91
Copyrighte
Figure 521. Persuasive forces from AP infons with and without a truck bombing infon favoring less troops. The A P dispatch scores w i t h o u t an extra truck bombing
infon (top frame) are the same as the b o t t o m frame i n Figure 5.13 except that the p l o t began w i t h the date o f the first available p o l l p o i n t on troops i n L e b a n o n ( A p p e n d i x B , T a b l e B.2). T h e b o t t o m frame includes the a d d i t i o n o f a truck bombing infon favoring less troops equivalent to forty AP paragraphs on October 23» 1983, the date o f the extra tic o n the horizontal axis. The truck b o m b i n g i n f o n also had a one-day persistence half-life.
Oct Nov Years_J 983-1984
92
Dec
Jan
Figure 5.22. Opinion projections with and without a truck bombing infon favoring less troops. Projections w e r e made for public o p i n i o n as i n Figure 5.19 except that
there was the addition o f a truck bombing infon o n October 23 favoring equivalent t o forty A P paragraphs (dotted lines). For this projection, the force curve favoring less troops was that i n the bottom frame o f Figure comparison, projections w i t h o u t this truck bombing i n f o n favoring less replotted f r o m Figure 5.19 (solid lines).
X F a v o r HORK Troo 70
less troops persuasive 5.21. F o r troops are
- Bomb F a v o r L e s s T r o o p s )
60 50 40
30 20 10 0 - Bomb F a v o r L e 3 3
Troops)
(•*•.- Bomb F a v o r L e s s
Troops)
X F a v o r SAME Troops 70 60 50 40 30 20 10 0 X F a v o r LESS Troopa 70 60 50 40
30 20 10 0 Oct Years
Nov
Dec
Jan
1983-1984
93
Copyrightc
•
Figure 5.23. Persuasive forces favorable AP paragraphs scored using bandwagon
to Democratic presidential candidates from words. Paragraphs were scored as favoring
Mondale, Glenn, and Others i f candidate names were close to favorable combinations o f bandwagon words (see Appendix C, Section C.4-1). T h e infon force curves were calculated using a one-day persistence half-life and a weight o f 1.0 f o r all infons.
Bandwagon
Infona
Jul Years
Favor
MONDALE (AP P a r a s )
Sep 1983-1984
Nov
Jan
94
Copyrightec
Figure 5.24. Persuasive forces unfavorable to Democratic presidential candidates from AP paragraphs scored using bandwagon words. Paragraphs were scored as
unfavorable t o M o n d a l e , G l e n n , and Others i f candidate names were close to combinations o f unfavorable bandwagon words (see A p p e n d i x C, Section C.4-1). The i n f o n force curves were calculated using a one-day persistence half-life and all infons w i t h a weight o f 1.0 as i n Figure 5.23. Bandwagon
Infons
D i s f a v o r
MONDALE
Infons
Disfavor
GLENN
(AP
Paras)
10 8 6 4 2 0 Bandwagon
(AP
Paras)
12 10 8 6 4 2 0 Bandwagon._Infons_Disfavor
OTHERS_
(AP_.Paras)
10
8 6 4 2 0
Jul
Sep
Nuv
Jan
VvjairB _ 1 «IH:» -1 Utf 4
95
CoDvriahted material
5.25. Persuasive farces of AP infons scored by name count only. Paragraphs mentioning Mondale, Glenn, and Others were scored as discussed i n A p p e n d i x C, Section C.4-2. The i n f o n force curves were calculated w i t h a one-day persistence halt life and all infons w i t h a weight o f 1.0.
Figure
I n f o n a Mention MONDALE (AP P a r a s )
Infons
Mention
Jul
OTHERS
(AP
Sep
Paras)
Nov
Jan
Vi*aj!LB__lfl83-19_84
96
Copyrights
Figure 5.26. Population conversion model for actions of infons scored using bandwagon words. The boxes denote the subpopulations under consideration. The
words i n the boxes begin w i t h "B" to refer t o those " believing" o r having an o p i n i o n favoring Mondale, Glenn, and Others or having N o O p i n i o n . T h e persuasive forces at any particular time were read f r o m the curves i n Figures 5.23 and 5.24 and were favorable o r unfavorable t o M o n d a l e , G l e n n , o r Others. W o r d s describing these forces begin w i t h " G , " w i t h favorable information denoted b y Pro and unfavorable b y Con. T h e candidate o r group o f candidates is indicated by the first three letters o f the name or group. T h e arrows indicate opinion conversions due t o the persuasive forces shown i n Figures 5.23 and 5.24.
***************** *
GProGle
*****************
*
* BMondale * * * *****************
> *
-*
I f S -S 8 ri
4* •4
44
i
—
44
**•
44 H
c
C
•4
2
44 44 •4 •M-
S 1'
4* 4* 44 44
4*
*4
44 44 +4 ••4
*#-
*•
44 44 44 44 •M-
S 5 CO a;
V
4* +4 +4 4444 44 44 •H
«44* 44 44
M
Q
•**
44 44 «• 44 4*
4*
44
4*
M
¡> u *- — JS > € £fi r •¬ § 2 2 2 S c -2 X *5 -
o
o y .
44 4* •M44
44 44
-
S
f
5
>
»e a « O C
4*
98
Figure 5.28. Optimization curves for constants for the Democratic primary. Bandwagon analysis: Using the population conversion model o f Figure 5.26 and the
persuasive forces o f Figures 5.23 and 5.24, the best modified persuasibility constant \k'2) g i v t o g the lowest M S D was 1.5 per A P paragraph per day (top frame), and the o p t i m a l persistence half-life f o r the same analysis was one day (center frame). T h e t w o o p t i m i z a t i o n s were each performed under the best c o n d i t i o n s f o r t h e other constant. Name-count analysis: The bottom frame shows that the best value f o r the modified persuasibility constant was very close to zero using a persistence half-life o f one day, the persuasive forces o f Figure 5.25, and the population conversion model o f Figure 5.27.
Mean S q u a r e d D e v i a t i o n
( i nPoll
Percent Squared)
Bandwagon k 2 f o r A L L I n f ona ( p e r P a r a - D ayj^ Mean S q u a r e d D e v i a t i o n ( i n P o l l P e r c e n t S q u a r e d ) 9
H a I f - l i f e f o r Bandwagon T e x t A n a l y s i s ( i n D a y a ) Mean S q u a r e d D e v i a t i o n ( i n P o l 1 P e r c e n t s q u a r e d )
Naae C o u n t
k ' 2 f o r A L L I n f ona
( p e rPara-1)a
99
Copyrighte
Figure 5.29. Opinion bandwagon analysis.
on Democratic
candidates
when infons were scored
by the
T h e projections (solid lines) f o r the three groups favoring Mondale, G l e n n , and Others began w i t h the o p i n i o n measurements o f the first p o l l o n June 19, 1983, and continued using the persuasive forces shown i n Figures 5.23 and 5.24, the m o d e l o f Figure 5.26, a n d the o p t i m i z e d m o d i f i e d persuasibility constant o f 1.5 per A P paragraph per day. Calculations were at six-hour intervals. Actual p o l l points are plotted as squares.
X F a v o r MONDALK_(Bandwagon T e x t A n a l y s i s )
% F a v o r GLENN (Bandwagon T e x t A n a l y s i s )
% Favor OTHERS (Bandwagon T e x t A n a l y s i s )
Jul Years
Sep 1983-1984
Nov
Jan
100
Copyrighted material
Figure 5,30. count only.
Opinion
on Democratic
candidates
when infons were scored
by
name
T h e projections (solid lines) for the three groups favoring Mondale, Glenn, and Others began w i t h the o p i n i o n measurements o f the First p o l l on June 19, 1983, and continued using the persuasive forces shown i n Figure 5.25, the model o f F i g u r e 5.27, and the same m o d i f i e d persuasibility constant o f 1.5 per A P paragraph per day used for Figure 5.29. Calculations were at six-hour intervals. Actual poll points are plotted as squares. %
Favor
MONDALE
%
Favor
OTHERS
J u l Years
(Name
(Naae
Count
Count
Sep
A n a l y s i s )
A n a l y s i s )
Nov
Jan
1983-1984
101
c
Figure 531. Persuasive forces from AP paragraphs favoring better, same, and worse economic conditions. Paragraphs favoring these three positions were scored (see
Appendix C, Section C.S) and used for the construction o f persuasive force curves employing a one-day half-life and equal weights o f 1.0 for a l l infons. Infona
Favor
Economic
C l i m a t e
B e t t e r
(AP
I n f o n s F a v o r Economic C l i m a t e Same (AP
Paras)
Paras)
8
Infona
Favor
Economic
C l i m a t e
Worse
(AP
Paras)
2
I
- 5 © - C O Si V3 l/S
0
>1
2 f i e *
** *-
V5 O .
•*
4>
** *
*•
-*
**-
§ 2 T § 4> . O •OcQ °
55
o
4> — £
O
o
o
+4—
4i
3 •2 i t/3
n i/1
5 CO :
y 2
1-2
= =
3 a,
S f— .2 '2
1
I
S S -&.S?
103
CoDvriahted material
Figure 5.33. Optimization curves for constants for the economic climate, Usinj: the population conversion model of Figure 5.32 and the persuasive forces of Figure 5.31, ihe best modified persuasLoiliry constant (¿'2) giving the least M S D bad a value of 0.09 per AP paragraph per day (top frame). Trie optimal persistence half-life was one day (bottom frame). Both optimizations were performed under the most favorable conditions for the other constant. Mpari
3quar_GDeviation
1—1
0 "1
Perccnt_Squarcd)
1—1
o.oi
0.001 Mean
( i n Poll
1—r
0.1
1
k ' Z F o r A L L Inform ( p e r Para-Day) Squared D e v i a t i o n ( i n Pol1 Percent. Squared)
40
0 T
!
2 Half-life
1
1 4
1—r—1 6
10
1
1 40
for Econoiic Cliiate
1—r—r 100 ( i n
Dftya)
101 MaTepnan, sauiMmeHHbiH ¿1« re |> - . " < : •• .' •
105 Figure 5.34. Opinion on economic climate. The projections (solid lines) for the three groups feeling that the climate was better, same, or worse began with the opinion measurements of the first poll on March 6, 1981, and continued using the persuasive forces of Figure 5.31, the population conversion model of Figure 5.32, and the optimal constants from Figure 5.33. The computations were performed every 24 hours. Actual poll points are plotted as squares.
%
Believe
Egonniic
Conditions
It
Jji'Hevt!
Ecofloinic
Conditions
i
1
Tear*
1
Rl^ 1981-IBS4
1
Ratter
Worse
1
B3
1
T
#4
Figure 5.35. Persuasive forces of AP infons favoring important, equal importance, and inflation more important.
unemployment
more
Paragraphs favoring these three positions were scored (see Appendix C, Section C.6) and used for the construction o f persuasive force curves employing a one-day persistence half-life and the f o l l o w i n g weights for infons supporting different positions: 1.0 for infons favoring unemployment more important, 0.5 for infons favoring equal importance, and 1.4 for infons favoring inflation more important. Infons
Favor
UNEMPLOYMENT
Infons
Favor
EQUAL
InIons
Favor INFLATION
Jul
78
Import,
(AP
Jul
106
(AP
Paras-Weight
1.0)
(AP P a r r i s - W e i j g h t
Paras-Weight
79
Jul
80
0.5)
1.4)
Jul
S £ £ * s ** ^ t*
í 3
ills i S-sgB h d
?
*
=
I
e^^
Ife" i •ill S
Q H
V
c
È5
_
.s s Işı
c
Figure 5 J7. Optimization curves for the modified persuasibility constant and the infon weighting constants for unemployment versus inflation. U s i n g the population
conversion model o f Figure 5.36 and the persuasive forces o f Figure 5.35, the best m o d i f i e d persuasibility constant g i v i n g the least M S I ) had a value o f 7.0 per A P paragraph per day (top frame). T h e o p t i m a l w e i g h t f o r infons f a v o r i n g equal importance was 0.5 (center frame), a n d that f o r infons f a v o r i n g i n f l a t i o n more important was 1.4 (bottom frame). A l l optimizations were performed under the most favorable conditions for the other constants and w i t h a one-day persistence half-life. Mean
Squared
Deviation ( i n Poll
Percent
Mean
Squared
Deviation ( i n Poll
Percent Squaredl
Weight Mean
f o r Paras Favor
Suuared D e v i a t i o n ( i n P o l l
Weight
f o r P a r a s Pavor
EQUAL
Squared)
Impor tanot?
Percent
INFLATION
Squared)
Iaportant
10K
Copyrighted material
Figure versus
5.38. Optimization curve for the persistence constant for unemployment inflation. U s i n g the same conditions and the o p t i m a l constants f r o m Figure
5.37, a persistence constant o f one-day gave the least M S D .
Mean S q u a r e d
Deviation { in P o l l
Percent
Squared)
H a l f - 1 i f e .for Uncmploy. va. I n f 1 a t . ( i n Days)
109
Copyrighted material
Figure 5.39. Opinion favoring unemployment more important, equal importance, or inflation more important The projections (solid lines) for the three opinions began
w i t h the measurements o f the first p o l l on M a r c h 22, 1977, and continued using the persuasive forces shown i n Figure 5.35, the population conversion model o f Figure 5.36, and the o p t i m a l constants f r o m Figures 5.37 and 5.38. The computations were performed every 24 hours. Actual poll points are plotted as squares. X_Bo1 i o v e
X
Be 1 i e v e
% Believe
UNEMPLOYMENT
EQUAL
Important
Impor tance
INFLATION
J u l Years
more
more
78 J u l 1977-1980
Important
79
J u l
80
J u l
110
Copyrighte
Figure 5.40. Persuasive opposing Contra aid.
forces
of AP infons scored
Paragraphs for
by the author as favoring
Fan
and
these t w o positions were scored by (see Appendix C, Section C.7) and used for the construction o f persuasive force curves employing a one-day persistence half-life, a weight o f 1.0 for infons favoring aid, and a weight o f 2.0 for infons opposing aid.
InfonH
FAVOR__Contra _ A i d
( AP
Paraa— Fan
Scores)
16 30 20
10
li
0
I n f o n s OPPOSF C o n t r a Aid (AP ParaB^-Fan
Scores)
40
30 2 0
E
10
0
83.5 84 Y«»H r-K
84.5
85
85.5
86
111
C
I
Figure 5.41. Persuasive forces of AP infons scored by Swim, Miene, and French as favoring and opposing Contra aid. Paragraphs far these two positions were scored (see Appendix C, Section C,7) and used for the construction of persuasive force curves employing a one-day persistence half-life, a weight of 1-0 for infons favoring aid* and a weight of 2A for infons opposing aid. Infojiri
FAVOIt A i d _[ AP _ P a r i i a _ - - S w i m n t
a [
SuoresJ
60
83.5 Years
HA
H4.5
85
85.5
Hi)
112 M i l T C P H a n . 3iiUJHLJJCHHblH ilR TC[I
L
" ' i
I
•
Figure 5.42. Population conversion model for actions of infons favoring and opposing Contra aid. The boxes denote the subpopulations under consideration. The
words i n the boxes begin w i t h " B " to refer to those " b e l i e v i n g " or having an o p i n i o n favoring o r opposing Contra aid. The arrows indicate o p i n i o n conversions due to the persuasive forces (beginning w i t h " G ) shown i n Figures 5.40 and 5.41. H
*************** *
GOpi*>se
*
* BFavor * * * ***************
*************** >
< GFavor
113
*
*
* BO| >pose * * * ***************
Figure 5 A3. Constant optimization curves for Contra aid. The M S D was calculated for o p i n i o n projections using both infon scores obtained by the author as plotted i n Figure 5.40 ( s o l i d lines) and b y S w i m , Miene, and French as plotted i n Figure 5.41 (dotted lines). For all calculations, the population conversion model was the one i n Figure 5.42. A one-day persistence half-life was used for determinations o f the best m o d i f i e d persuasibility constants (k'2) favoring Contra a i d (top frame)--1.2 p e r AP paragraph per day for Fan infons and 1.6 for S w i m et al. infons—and calculations for the optimal weights for infons opposing Contra aid (center frame)--2.0 for Fan infons and 2.4 f o r S w i m et al. infons. T h e o p t i m a l values f o r the other constants were used for c o m p u t i n g the best persistence half-lives (bottom frame)--forty-one days f o r Fan infons and seven days for S w i m et a l . infons.
Mean Squart?d D e v i a t i o n ( i n P o l 1 P r e c e n t 100
Squared)
80 60 40 20 0 0.01
0.001
0.1
1
10
k'2 f o r FAVOR C o n t r a A i d ( p e r P a r a - D a y ) Mean S q u a r e d D e v i a t i o n ( i n P o l l P r e c e n t S q u a r e d ) 350 300 250 200 150 100 50 0
J 1.5 2 2.5 3 Weight f o r P a r a a OPPOSE C o n t r a A i d Mean S q u a r e d D e v i a t i o n ( i n P o l l P r e c e n t S q u a r e d ) 120 100 80 60 40 20 0
1
2 4 6 8 10 20 40 Lt.-JL TÇ f o r C o n t r a A i d _ ( j n Dayn) 1
114
Copyrightec
115 Figure 5.44. Opinion favoring and opposing Contra aid using infon scores by the author. The projections (solid lines) for the two opinions began with the opinion measurements of the first poll on August 20, 1983, and continued using the persuasive forces shown in Figure 5.40, the population conversion model of Figure 5.42, a persistence half-life of one day, and optimal values for the modified persuasibility constant favoring Contra aid and the weight for infons opposing aid ( Figure 5.43, solid lines). Computations were performed every 24 hours. Actual poll points are plotted as squares.
Figure 5.45. Opinion favoring and opposing Contra aid using infon scores by Swim, Miene, and French, The projections (solid lines) for the two opinions began with the opioion measurements of the first poll on August 20, 1983. and continued using the persuasive forces shown in Figure 5A\, the population conversion model of Figure 5.42. a persistence half-life of one day, and Optimal values for the modified persuasibility constant, favoring Contra aid and the weight for infons opposing aid (Figure 5.43 doited lines). Computations were performed every 24 hours. Actual poll points are plotted as squares. % PA Villi q o n t . r i i
A i d _£Sw_i m_ rt_
n 1 __Soprriii_}
(it) •
20 T
0 % OPPOSE HO 60
Cn_pt.rn.
Aid
(Swim
el. a l
St-nros)
-
10 • 20
'
0 83.5
T
"I
P
1
84
H4.5
85
85.5
T86
M a TOP n a n , 3iiiiinLiicnHbin a B T o p c « w i * • i
:••
•
6
Methodological Significance of Work
The previous three chapters have applied computer lent analyses and ideodynamics equations to project expected public opinion. The present chapter examines the methodological implications of the InfoTrend procedures. The next chapter will consider the theoretical significance of the results. The studies in this book focused on public opinion where there was relatively little social or economic cost to persons changing their minds. A deliberate choice was made to avoid economic issues, including product purchase, because such activities required weighing such complicating factors as competing uses for financial resources. Messages due to these important factors were difficult to study directly. The introduction of innovations into society was also not studied because they frequently involved not only economic considerations but also social factors. While economics was obviously important for the classical studies of the adoption of hybrid com (Ryan and Gross, 1943), investigations on the acceptance of family planning (Berelson and Freedman, 1964) also had to contend with complex societid forces related to sexuality and reproduction. In contrast, people were unlikely lo have deep convictions for the issues studied in this book. Members of the population as a whole clearly have no good idea how much should be spent for defense. Troops were only briefly in Lebanon, with little time to form ingrained prejudices. None of the Democrats running for president had ever held that office, so none had the advantages of incumbency. The public was well aware that the economy could change, so there was no reason to feel that the economic climate should always be good or bad or that unemployment should always be more or less important than inflation. As with Lebanon, the Contras were in a distant land with which few Americans had personal contact and about which there was little inherent opinion. These theoretical arguments for opinion malleability and volatility were bolstered by actual findings that there were .substantial opinion movements in all examples except that of Contra aid. In fact, one of the reasons for studying these other ins lances was that opinion did Change, thereby providing the most critical tests of the methodology. However, it was also useful to demonstrate that the calculations could project unchanging opinion for Contra aid when constancy was actually observed The model would definitely have been weakened by data showing that there was a change in the ratio of favorable to unfavorable information while there was no simultaneous change in opinion. Although these studies showed the applicability of the calculations to issues where people hold shallow and changeable convictions, the same methodology should Theoretically succeed even for more firmly held beliefs i f all the relevant messages
M a T e p n a n . saLaumeHHbiti aBTopc«wi* • :
• •
/18
Predictions
of Public
Opinion from
the Mass
Media
available t o the p u b l i c can be coded. F o r f i r m l y held beliefs, the persuasibility constants w o u l d s i m p l y be decreased so that more i n f o r m a t i o n w o u l d be needed t o cause an o p i n i o n s h i f t
6.1 V A L I D A T I O N O F I D E O D Y N A M I C S Since n o applications o f ideodynamics have been described previously, i t was important to validate the model using not o n l y logical argument but also empirical tests. Such tests benefited f r o m one o f the unusual capabilities o f the m o d e l - i t s ability to consider the time dimension d i v i d e d i n t o infinitesimally small intervals. A s noted i n Chapter 2, this was possible because o p i n i o n calculations were not based solely o n the i n f o r m a t i o n available t o the p o p u l a t i o n , b u t also o n the o p i n i o n i n the previous t i m e i n t e r v a l . T h e use o f s m a l l t i m e intervals, l i k e the 6- o r 2 4 - h o u r periods i n this book, permits the tracking o f r a p i d changes l i k e that f o r troops i n Lebanon, where opinion favoring more troops changed f r o m under 7 percent to over 30 percent and back d o w n t o under 9 percent, all w i t h i n a few days. T h e use o f small time intervals is important t o e m p i r i c a l tests o f ideodynamics because the number o f o p i n i o n predictions increases and the t i m i n g o f the computed o p i n i o n values becomes more precise as the intervals f o r the o p i n i o n calculations decrease i n size. The result is more o p i n i o n projections at more closely defined times. These very precise predictions can then be tested against measured poll data. I n the actual cases studied, the number o f times o p i n i o n values were calculated for each time series ranged from approximately 400 f o r troops i n Lebanon to over 3,000 f o r defense spending. Therefore, the model c o u l d be tested b y i t s a b i l i t y t o m i m i c poll data for hundreds to thousands o f time points. For each o f the cases i n this book, it was necessary t o assign one to three i n d e p e n d e n t p a r a m e t e r s i n a d d i t i o n t o the persistence constant, w h i c h was set t o h a v e a one day h a l f - l i f e f o r a l l studies. Therefore, the m i n i m u m number o f p o l l points needed t o set the parameters s h o u l d be those s u f f i c i e n t t o g i v e one t o three independent opinion measurements. A t each p o l l time, there were t w o t o four polled positions. O p i n i o n f o r one o f the positions c o u l d be determined by subtraction f r o m 100 percent, so the number o f independent o p i n i o n measurements was one less than the polled positions. A s a result, a time series w i t h three p o l l measurements w o u l d y i e l d a m i n i m u m o f three independent p o l l percentage values (the Contra case, w i t h o n l y t w o positions, f o r or against aid). Therefore, three well-spaced polls w i t h t w o o r more positions per measurement should be sufficient to establish the three parameters i f the polls had n o errors. I n addition, the p o l l percentages at the earliest time i n the series were used as the starting point for the calculations, so o p i n i o n values at this t i m e were the boundary conditions also needed f o r setting the parameters. Therefore, i n the worst case, a p o l l series w i t h f o u r t i m e points should uniquely define the three parameters. O n e o f these poll measurements w o u l d correspond t o the i n i t i a l p o l l conditions, a n d the other three p o l l points w o u l d be used t o set three parameters. I n the best cases, w i t h o n l y one parameter aside f r o m the consensus half-life o f one day (defense spending. Democratic primary, and economic climate), o n l y t w o accurate and well-spaced poll points w o u l d be needed, in principle, to define the parameters. Once the parameters were set, any additional p o l l points c o u l d n o longer be f i t by adjusting the parameters and w o u l d c r i t i c a l l y test the model e m p i r i c a l l y . T h i s estimate o f needing p o l l series o f t w o t o f o u r points t o establish one t o three parameters is only approximate because the o p i n i o n values are not truly independent Instead, o p i n i o n at any time is dependent o n o p i n i o n at earlier times. Nevertheless, these arguments d o indicate that series w i t h eight (Democratic p r i m a r y ) t o s i x t y - t w o (defense spending) time points d i d indeed provide meaningful empirical tests o f the model.
Copyrightec
Methodological
Significance of Work
119
Since ideodynamics uses a number of apprarirnations, it will not always be clear which are faulty if the model does not give satisfactory predictions. In contrast, generally accurate calculations for a number of issues will render plausible the total constellation of approximations* The quality of die Tit shown in the tables and figures of Chapter 5 indicates that the methodology has good predictive powers. It was also significant that the methods in this book were successful for si* oui of s i l cases tested. In consequence, the collection of approximations used for the computations stands a reasonable chance of being valid. Since the empirical tests for ideodynamics involved comparisons between poll results and opinion computed from information available to the population, it was essential that scores for persuasive messages be obtained independently of opinion measurements. For this reason, four precautions were taken: 1. The conditions for the infon scoring of Chapter 4 were developed by examining AP stories in random order so that the analyst would not be tempted-consciously or unconsciously-to bias scores with the goal of fitring F * " results. 2, The same computer scoring was applied to all stories, so any changes in scoring rules could not be applied p referen liai I y to a chosen subset of Stories in order to achieve scores which would result Ln good opinion predictions, 3- The vocabulary and scoring rules had to be logically defensible. For instance, word combinations implying less spending could not be used to score for more spending. A. For the example of Contra aid, the Roper Center was first contacted to establish that there was likely to be enough polls to construct a reasonable time series for the model testing. However, no poll values were actually obtained until the text analysis was finished both by Fan and by Swim, Miene, and French. Only after completion of the text analysis were actual poll percentages obtained frnm the Roper Center. Therefore, for this example, there was absolutely no way for the analysts to adjust the scoring to match measured poll values. 0
6.Z
DATA AND ISSUES CALCULATIONS
FOR
SUCCESSFUL
1DEODYNAMIC
Since methodological development is one of the significant aspects of this book, it is useful to consider the appropriate conditions for applying the method"logicsOne important suggestion from the studies presented above is that ideodynamics is applicable to a wide variety of issues. After all, the model was successful for issues drawn from areas as diverse as foreign policy (troops in Lebanon and Contra aid), economic issues (economic conditions and the importance of unemployment versus inflation), domestic policy {defense spending), and political campaigns (Democratic primary). The accuracy of ideodynamics derives from the fact that the calculations include all relevant persuasive messages. All the issues just mentioned shared the condition that the mass media were likely to contain the majority of them. Other informational sources such as books or éducation in schools could not keep abreast of the pertinent news as it was being generated. The main other sources which had the potential to inform as rapidly were personal experience and underground means of communication such as ru mors. Personal experience was unimportant for all the issues studied. For the Democratic primary in the early stages, defense spending. Contra aid, and troops in Lebanon, it was plausible that the mass media were likely to be the primary sources of all pertinent communications. For the economic climate and the importance of inflation versus unemployment, on the other hand, personal experience and observation might have been expected to provide significant messages. However, the empirical resting showed that reasonable opinion time trends could be calculated
Ma
TOP Ha n , sawHiueHUbiH a B T o p c i w i * • :
•
•
¡20
Predictions of Public Opinion from ike Mass Media
without modifications to account for personal experiences. In other words, personal experience messages could be ignored. This result suggested that information in the media might color substantially an individual's interpretation of personal experiences. For instance, m both the best and worst of limes, people will be aware of others out Of work. When the economy is considered to be good, people might interpret a person's unemployment to be his own fault. However, with bad economic news, people might well shift the blame from the individual to general conditions. Word-of-mouth communication such as rumors would have been important if the mass media were not trusted by the population. Indeed, if a government controlled press is perceived by the public to be biased and untrustworthy, then alternative persuasive messages will become important. Examples would range from underground publications in totalitarian states to rumors during time of war. Reliance on rumor and an underground press is probably minimal in the major democracies at this time. Another implication of these calculations is that successful calculations can frequently be made using only AP messages. The tests in this book deliberately made the extreme simplifying approximation that AP stories could represent all relevant persuasive messages for the topics studied. If this approximation is valid for a large number of diverse topics, then the AP İs indeed likely to be representative of most of the news in the mass media in agreement with the finding of similarity in much of the mass media by Paletz and Entman (1981), Furthermore, opinion calculations will have been shown to be independent, generally speaking, of special considerations for special events. Indeed, among all six examples in this book, it was only necessary to add one non-AP infon once. That addition was the infon favoring more troops being sent to Lebanon after the truck bombing of American Marine headquarters and was needed to calculate the great increase in opinion just after the truck bombing. In general, the success of the empirical tests argues for the appropriateness of assuming ihat (he AP could represent both the print and electronic media except in very rare cases Like the truck bombing incident. Previous investigators have frequently used the New York Times or the Vanderhi 11 summaries of television news (e.g., MacKuen, 1981; Ostrom and Simon, 1935: Page, Shapiro, and Dempsey. 1985, 1987). There might have been minor differences between the news content İn the AP and the New York Times or television broadcasts. However, the variations were probably not large. Nevertheless, there was likely to be a significant difference between the news stories used for this book and those identified as relevant by human judges. With human judges, there is probably preferential identification of stories concentrating on the topic under study. Stories with oblique inclusion of pertinent messages are Likely to be ignored. The retrievals for this book did not have this bias since the full texts of all dispatches in the Nexis data base were searched using combinations of key words chosen by the investigator (Chapter 2). AH phrases relevant to the topic were identified even if they were minor components or a story mainly discussing some other topic. Since only text in the region of discussions of the relevant issue was collected, the number of retrieved words per dispatch gave a good idea of the fraction of the typical dispatch devoted to the question. The typical AP dispatch had 400-900 words. About 420 words were retrieved per average dispatch for troops in Lebanon, consistent with the observation that a large number of these dispatches were devoted mainly to this topic. For the Democratic primary and Contra aid, those numbers were 310 and 258, respectively, already significantly less than a full story, while the equivalent values for defense spending, the economic climate, and unemployment versus inflation were all under 200, meaning that less than half of the typical story was on these topics. This inclusion of articles mainly about other topics was appropriate since members of (he general public were frequently not looking for (ex( on particular polled topics and thoughts were probably absorbed from whatever was read regardless of whether the ideas were surrounded by similar or dissimilar information.
Ma
TOP n a n , sauiHmcHHbiH aBTopc«wi* • :
•
•
Methodological
Significance
121
of Work
The computer searches not o n l y selected articles in which other topics were the m a i n focus, but such a search also guaranteed consistency. U n l i k e a human j u d g e w h o m i g h t have been distracted when the major emphasis was on another subject, the computer always found the programmed w o r d combinations. The thoroughness o f the data base searches was seen i n the identifications ranging f r o m 1,156 dispatches for Contra aid to 12,393 for the economic climate. S i m p l y to read these numbers o f articles w o u l d have been a daunting task for any human investigator. F r o m a methodological standpoint, i t is very useful that the A P alone was able to represent all persuasive messages aside f r o m the truck b o m b i n g example. T h i s f i n d i n g indicates that calculations can usually succeed w i t h o u t r e q u i r i n g ad hoc adjustments to account for special events. The a p p l i c a b i l i t y o f the methods to six quite varied topics suggests that many future computations are also l i k e l y t o be generally v a l i d i f o n l y A P messages are analyzed.
6.3
P O S I T I O N S FOR W H I C H P E R S U A S I V E M E S S A G E S A R E S C O R E D
Once the issues have been defined and the relevant persuasive messages have been assembled, it is then necessary to obtain scores for the infons i n messages. As noted i n A p p e n d i x A, there w i l l be overlap between the positions that infons are favor and those that people are to support i n o p i n i o n polls. However, the overlap need not be complete. For instance, i n the bandwagon scoring for the Democratic p r i m a r y , there were people w i t h N o O p i n i o n and no infons f a v o r i n g this p o s i t i o n , w h i l e there were infons unfavorable to M o n d a l e and no persons polled to have this viewpoint. For the other cases, the overlap was m u c h greater. For Contra aid, both infons and subpopulations either favored or opposed aid. In the remaining cases, there were three p o l l positions, r a n g i n g f r o m one extreme t h r o u g h the center to the other extreme. A n example was positions o f more, same, and less spending for defense. Here, infons were also scored to favor either the same three positions o r the t w o positions o f more and less spending. Since all messages were considered to contain different infons, each w i t h a content score r a n g i n g f r o m zero t o some positive n u m b e r o f paragraphs, some A P dispatches were m i x e d messages w i t h positive scores for t w o or more infons. F o r issues w i t h t w o extremes and one central p o s i t i o n , a neutral message f a v o r i n g the central idea was d i s t i n g u i s h e d f r o m a message w i t h t w o equal components favoring the t w o extreme positions. In the First case, the dispatch w o u l d have had a positive score for the neutral position. In the second case, the story w o u l d have had t w o positive scores, one i n favor o f each o f the extremes. Both types o f scores were found and provided a more subtle means for extracting information f r o m messages than w o u l d be possible by scoring a message w i t h pro and con components as being equivalent to a neutral message. As a r g u e d i n Chapter 2, a m i x e d pro and con message should not have the same persuasive effects as a neutral message. Since dispatch scores were given i n paragraphs, stories w i t h more relevant paragraphs had higher scores. Therefore, the dispatch content scores were weighted i n proportion to message salience.
scored to
measured
6.4 C O M P U T E R T E X T S C O R I N G A l t h o u g h human j u d g i n g can be used to obtain position scores for A P stories, this book relied on a computer method to guarantee u n i f o r m i t y o f scoring. Since the techniques and ramifications o f the text analysis have already been presented i n Chapter 4 and are i n A p p e n d i x C, they w i l l not be repeated here. However, as an o v e r v i e w , there are several important novelties i n the I n f o T r e n d computer text analysis.
Copyrighte
122 Predictions
of Public Opinion from the Mass Media
One feature is the decision to return to the idea o f t a i l o r i n g the analyses to individual issues instead o f using a generalized computer content analysis program with a fixed dictionary and invariant scoring rules. The extra rime and effort spent i n making up specific dictionaries and rules were compensated for by the lack o f need to interpret the results, a situation opposite f r o m that with predefined dictionaries and rules. The method was also quite general, being applicable to any texts, i n c l u d i n g those not examined f o r o p i n i o n f o r m a t i o n . T h e possible analysis o f letters o f recommendation has already been mentioned i n Chapter 3. T h e generality o f the method was not at the level o f the dictionary or rules, b u t rather at the level o f strategy, including text nitrations as key elements. Precision i n the analysis was greatly aided by the use o f repeated text nitrations to r e m o v e irrelevant text. The r e m a i n i n g text was more homogeneous, thereby s i m p l i f y i n g subsequent steps and p e r m i t t i n g the use o f words w h i c h w o u l d be ambiguous i n a general setting. These f i l t r a t i o n steps were also useful for incorporating relationships between words i n the input text for f o l l o w i n g steps. The InfoTrend procedures were also surprisingly robust i n t h a i apparently important changes i n the dictionaries and rules d i d not affect the basic shape o f the curves fur predicted opinion. Thus, dispatches for defense spending could be senred either to favor more, same, and less spending or just more and less spending. Also, nuclear arms reduction c o u l d either be interpreted to favor Less defense spending or not. A number o f words in the scoring dictionaries c o u l d be changed. Yet the basic opinion projections stayed more or less the same. Therefore, there was no need to be overly concerned that words or rules were either o m i t t e d or misassigned d u r i n g construction o f the dictionaries and rules. A n equally compelling argument for the robustness Of the text analysis was the fact that Fan and S w i m el al. could independently arrive at different dictionaries and rules for the text analysis for Contra aid (Chapter 4), The dictionaries and rules were quite different for the t w o analyses, w i t h S w i m et aL using quite different strategies and assigning scores to twice as many paragraphs as Fan. Despite the differences, the resulting o p i n i o n projections were essentially the same, suggesting that the text presented the same thoughts in a number o f alternate ways. I n addition* strong conclusions c o u l d be made about the message components critical for persuading the public since the computer applied the dictionaries and rules b l i n d l y . This was an important advantage o f analyzing the text using a structure where all dictionary words and rules had to be j u s t i f i e d o n logical grounds. N o artificial considerations c o u l d be introduced for the sole purpose o f f i n i n g desired output results. The dictionaries and rules for the text analyses were a l l constructed o n l y b y l o o k i n g at die retrieved t e x t without the application o f expert knowledge specific to the topics under study. This strategy was reasonable since the goal was to study o p i n i o n i n the general public, comprised mainly o f nonexperts. The success o f the calculations for six out o f six issues, despite the absence o f expert knowledge, suggests that appropriate d i c t i o n a r i e s a n d rules can be made i n general by nonspccialists, Besides the omission o f possibly pertinent information outside o f the messages themselves, portions o f the text indirectly supporting a position were also ignored w i t h t w o exceptions. One was the interpretation o f waste and fraud infons to support less defense spending, and the other was the i n c l u s i o n o f indirect messages i n the scoring for Contra a i d by S w i m et al. Otherwise, i t was sufficient to include o n l y •hose phrases w i t h text directly arguing for a position, phrases analogous to the verv b r i e f extracts o f f i l m reviews i n advertisements (e.g., best o f the year..New York
Times'), I t is l i k e l y that there was usually a h i g h correlation between indirect and direct messages, since the same opinion calculations for Contra aid were obtained f r o m text analyses by S w i m et al* using many indirect messages and by Fan scoring o n l y those w o r d clusters directly taking a position on the desirability o f aid (Chapter 3), The
MaTepnan,
aauiMmeHHbiH
as rt p - . ydue t o recruitment f r o m other subpopulations Pr* w h i l e the second double s u m o v e r / and r gives the loss o f members f r o m due t o i n f o r m a t i o n favoring other positions. The single sum over / gives the g a m i n subgroup P. due t o r e c r u i t m e n t f r o m the unawares, and the last t e r m reflects loss f r o m P y t o unawareness due to forgetting. The detailed explanation of these terms is as follows;
Recruitment f r o m Other Subpopulations-First Double Sum in Equation A.14 Members o f a particular subpopulation P can be persuaded by persuasive force functions Hyr t o j o i n subgroup Py Resistance t o change due to reinforcing infons I irk and saturation w i t h infons I n't has already been incorporated into Hy (Equation A . 1 3 ) . T h e n u m b e r persuaded at t i m e t is p r o p o r t i o n a l t o the pool o f p o t e n t i a l converts P The size o f this pool f r o m Section A . 2 is A(t)*Br(t). T h e number converted is also p r o p o r t i o n a l to the persuasive force function Hy describing the effectiveness o f infons iij'k i n the face o f r e i n f o r c i n g infons I irk* T h e constant o f proportionality is *2yy, the "persuasibility" constant w i t h subscript 2 indicating that r
r
r
r
the c o n s t a n t i s a p e r s u a s i b i l i t y c o n s t a n t .
The summation is over a l l possible f and r w i t h constant ¿2/77 having values o f either * 2 texo. Constant * 2 has the same value f o r eacn *2/>y i n any one equation where persuasive force f u n c t i o n //y>can actually persuade members o f subpopulation P to change their opinions f r o m position Q and j o i n subpopulation / j f a v o r i n g idea Qy T h i s constancy for ¿2 values is consistent w i t h Chapter 1 postulating that the persuasibility constant measures the closeness o f the issue to the core beliefs o f the population. However, ¿2 is a constant w h i c h can change f r o m issue to issue. Constant * 2 / r / P° ' ^ he positive for any c o m b i n a t i o n o f indices. F o r instance, *2/V/ * 2 is possible for t w o d i f f e r e n t / but the same r a n d j . This w o u l d mean t h a t t w o types o f persuasive forces can persuade members o f the same target p o p u l a t i o n P t o j o i n the destination population Py A n example w o u l d be o
r
r
r
c a n
D e
s t u
a t e
t o
m
r
Copyrights
148
Appendix
A
i n f o r m a t i o n favoring both more and same defense spending persuading people f a v o r i n g less spending to support same spending. I n contrast, kij'rj - 0 if a transition is not permitted. For example, a persuasive force function favoring less defense spending should usually not persuade people favoring less spending t o favor m o r e spending. The details o f the * 2 / r y array correspond to the postulated " p o p u l a t i o n conversion models for the awares" (see Figures 5.2, 5.18, 5.29, 5.30, 5.35, 5.39, and 5.45 o f Chapter 5 for examples).
Loss o f Believers-Second Double S u m i n Equation A . 14 T h i s sum merely reflects the fact that the population stays constant i n size. I f people are persuaded b y persuasive force ///'/to leave p o p u l a t i o n Pj to j o i n population P w i t h *2/;> • *2« then there w i l l not o n l y be a gain i n population P but also a loss f r o m population Py T h i s loss results i n the second sum i n Equation A.14 h a v i n g a negative sign. The magnitude of the loss is the same as the gain by p o p u l a t i o n P , so the terms i n the first t w o sums o f the equation have the same form. r
r
r
Conversions o f Unawares to Awareness-Single
Sum i n Equation A.14
These conversions are based o n the argument that the unawares cannot remember i n f o r m a t i o n so they learn about the issue through infon persuasive force functions Fjit) (Equation A.6). The rate at w h i c h the unawares P\j move to hold position Qj due t o Fjit) w i l l be proportional both t o the number o f people (/ - A(t)) i n Py and to Fj% w i t h an "attentiveness" constant o f p r o p o r t i o n a l i t y kjjj. The subscript / denotes an attentiveness constant (Chapter 1). Constant kjjj has the same structure a s constant * 2 i W - However, since the o n l y t a r g e t population under consideration is the unawares Pfj, there is no need t o specify the target population i n the c o n s t a n t It is sufficient t o specify the index / f o r the persuasive force F / and index j for the destination population Pj. As w i t h * 2 / > / ¿7/7 either has a constant value denoted by kj or a value o f zero, depending on a chosen "population conversion model for the unawares." The sum i n Equation A . 1 4 is over all / . I f there is no contribution f r o m a function F/, the consequence w o u l d be kjjj * 0 for the corresponding attentiveness constant. For instance, i f the m o d e l proposes that a l l functions Fj' first m o v e the unawares into a population Pj o f aware but uncommitted, kjjj - kj o n l y for this one value o f / A l l other kjjj - d .
Forgetting o f a P o s i t i o n - L a s t T e r m o f Equation A.14 I n the reverse process, i t is assumed that any aware can forget the issue and become unaware. The unlearning o f the issue is characterized by constant u w i t h the rate o f conversion o f awares favoring position Qj to unawareness being u*A(t)*Bj(t). I n this expression, the chances o f forgetting are the same f o r all i n d i v i d u a l s so the t o t a l loss f r o m awares f a v o r i n g idea Qj is p r o p o r t i o n a l t o the size o f the corresponding subpopulation A(t) Bj(t). As argued for Equation A.7, the D o n ' t K n o w s i n this book were usually less than 10 percent, so that A(t)*4 T h e n Equation A.14 becomes m
m
dB/0
(A.15)
V di
lk2frrHj'r(t)*Br
for a l l k w i t h t < i¿ and for a l l o d d i. Since ku*j - kj o r zero (see discussion f o l l o w i n g Equation A . 1 4 ) , and since kjjj i n Equation A . 3 1 is always m u l t i p l i e d by kvAP kaAP> convenient to define a " m o d i f i e d attentiveness" constant a n d
li
(A.33)
k'j=kj*k kaAP vAr
T h e n , Equations A . 3 1 and A . 3 3 , together w i t h analogs to Equations A . 2 0 - A . 2 3 , yield (A.34)
A(t)
. A(t-At)
+ U(TIR>%
k']jj'F'jHHl-A(t-At))l
-
u*A(t-At))*At
jj'
for all j and f where (A.35)
F'jit)
-
£ ij\k
w jj--*sifk*e-P'' ' k) i
t t
for all odd i . all7", and k w i t h t < í¿. W i t h Equations A.34 and A.35 and postulated values o f k'jjj-m addition to the values o f Section A . 1 0 a b o v e - i t is possible to calculate A(t) at intervals o f At i f an i n i t i a l value o f A(t) corresponding to those w h o had not yet heard o f the issue, is available f r o m a p o l l . Then, w i t h the postulated values o f k'jjj and calculated A(t) it is possible to calculate the fraction o f the total population favoring position Qj by approximating the solution o f Equation A.14 by %
9
(A.36)
A(t)*Bj(t) + (TfR)*Af(
+ (TlR)*AfI
-
A(t-At)»Bj(t-At)
I
Gy(/>/*'2/r/*f^^^ k'lffF'jitHl-A(t~At))l
X
-
wA(t-At)*Bj(t-At).
J
Copyrighte
154
A.13 EXTENSIONS TO VERY LONG TIMES As noted in Section A.2 above, the assumption was that the population was constant during the time period of the calculation. In other words, birth, death, and migration into and out of the population were ignored. These assumptions are likely to be valid for the periods under a year used for the Lebanon and Democratic primary examples. However, as times increased to over nine years for the case of defense spending, this assumption may start to break down. To account for death, it is possible to introduce another term like that at the end of Equation A.14. To account for birth, it is necessary to add additional terms to Equation A.14 adding members to the population of unawares. Then their conversion to awareness can follow the discussion in Section A.12 above. Furthermore, in the absence of population constancy, all the equations of this chapter will also need to be modified so that all computations are in terms of absolute number and not percentages.
There is another place where the model may need to be changed. The constants in Equation A.14 and its derivative equations may slowly change with time as society changes. For instance, the reputation of the medium may slowly drift. Therefore, for truly long term studies, the constants may have to be converted into time dependent functions reflecting the accumulated experience of the population. A.14 MODELS WITH NO DEPENDENCE ON SUBPOPULATIONS Unlike ideodynamics, some models do not calculate opinion percentages based on a subdivided population. Instead, only forces on the population as a whole are taken into account. Ideodynamics can also be used to calculate opinion based solely on persuasive forces by making the assumption that opinion change is sufficiently slow so that dBj/dt ~ 0. In this case, Equation A.15 converts to (A.37) if
With dBj/dt ~ 0, all the Bj are constants and calculable given only the persuasive forces Hjr and the persuasibility constant kjr. Equation A.37 is actually a system of simultaneous equations, one for each position with subscript j . With j equations, there is a unique solution for each of the Bj. Given no rapid change in Bj, the H functions which drive the change must also be reasonably constant at calculation time t.
Appendix B
Data for Calculating Opinion Change
The data used for projecting public o p i n i o n were o f t w o types: (1) t i m e series o f public o p i n i o n polls f r o m published data and (2) A P dispatches relevant to the polled topics retrieved f r o m the Nexis electronic data base sold by Mead Data Central, 9393 Springboro Pike, P.O. B o x 9 3 3 , D a y t o n , O h i o , 4 S 4 0 1 . l i t i s data base contained a l l A P dispatches since January 1, 1977. Polls and A P dispatches were obtained f o r six issues. A l l retrievals were restricted to text w i t h i n f i f t y words both before and after one o f the k e y w o r d s used i n the o r i g i n a l search. T h e f i f t y - w o r d l i m i t e l i m i n a t e d irrelevant sections o f the dispatches a n d was chosen because the w o r d s at the beginnings and ends o f the retrieved regions t y p i c a l l y showed transitions t o other topics. Articles concentrating on an issue t y p i c a l l y had the key search words w i t h i n 100 words o f each other (fifty words after one search w o r d and fifty words before another). These articles were automatically retrieved i n their entirety. A l l A P searches began before the first p o l l p o i n t i n order t o account for the residual effects o f p r i o r messages. T h e search was made f o r a l l dispatches u p to six months before the first point i n the p o l l series unless the s i x - m o n t h period extended before the beginning o f the data base on January 1, 1977. I n that case, the search began w i t h this date. A l l searches stopped at the end o f the p o l l i n g period.
B.l
DEFENSE
SPENDING-1977-1984
Four variant p o l l series were found f r o m 1977 t o 1984 for the issue o f whether more, same, o r less s h o u l d be spent o n defense (Table B . l ) . A l t h o u g h earlier polls existed, they were n o t studied because the N e x i s data base o n l y contained A P dispatches back to 1977. I n a l l polls, the vast majority o f the population had definite opinions and were d i v i d e d i n t o three groups: those f a v o r i n g more, same, o r less defense spending. There was also a group o f Don't K n o w s o r N o t Sures, t y p i c a l l y i n the range o f 5-10 p e r c e n t T h i s subpopulation was subtracted f r o m the total and the p o l l data was renormalized among all those w i t h an o p i n i o n . F o r ideodynamics, this step effectively assumed that the s m a l l n u m b e r o f persons w i t h no o p i n i o n stayed i n that category. Even i f this assumption was not entirely v a l i d , the numbers were sufficiently small that the results w o u l d not have been much affected. Fortunately, the same t i m e trend was seen f o r a l l four polls after the D o n ' t K n o w s were removed (Figure 2.1). F o r this figure, n o adjustments were made
Copyrights
156
Appendix
B
beyond the r e m o v a l o f those w i t h no o p i n i o n . G i v e n the agreement between the different poll series, the data were pooled. Relevant i n f o r m a t i o n i n the Nexis data base was identified by searching the full texts o f a l l dispatches using combinations o f key words chosen by the investigator. For defense spending, the search was for ( D E F E N S E o r M I L I T A R Y o r A R M S ) w i t h i n five words o f ( B U D G E T ! or E X P E N D I T U R E o r S P E N D ! o r F U N D ! ) f r o m January 1, 1977, t o A p r i l 1, 1984. The T permitted the trailing characters to be a n y t h i n g , so that both budgeted and budgetary w o u l d have been f o u n d w i t h BUDGET!. The search command yielded 9,314 dispatches w i t h the data base numbering the dispatches i n reverse chronological order, number one being the most recent and number 9,314 being the earliest. F r o m a random 692 o f these dispatches, text was retrieved i f i t was w i t h i n f i f t y words o f one o f the seven key search words given i n the previous paragraph. I f t w o key words occurred w i t h i n 100 words o f each other, the entire intervening text was collected. The total retrieval was 820,000 characters of text A sizable number o f dispatches were not about American defense spending. As soon as this became clear, the retrievals were stopped. The analyses using the 692 A P dispatches f r o m 1977 t o 1984 and the poll data in Table B . l were extended in t w o ways. First, additional p o l l data were collected f r o m the Roper Center at the University o f Connecticut (see Table B . l below for more details) f r o m January 1977 to A p r i l 1986. A time series o f s i x t y - t w o separate polls c o u l d be obtained by p o o l i n g these additional polls w i t h those i n Table B . l . Besides polls f r o m the National O p i n i o n Research Center, N B C News, and the Roper organization, pooled polls also contained results f r o m A B C N e w s , C B S N e w s , and the G a l l u p and Harris organizations. The same commands used for i d e n t i f y i n g the 9,413 dispatches f r o m 1977 t o 1984 were used again to locate 10,451 stories f r o m January 1, 1 9 8 1 , to A p r i l 12, 1986. O f these, 1,067 w e r e retrieved randomly for extending the Study to 1986. T o determine the importance o f stories on waste and fraud on o p i n i o n o n defense spending, the Nexis data base was further searched for ( D E F E N S E or M I L I T A R Y o r A R M S ) w i t h i n five words o f ( W A S T E or F R A U D or C O R R U P T I O N ) from January I , 1977, to A p r i l 12, 1986, y i e l d i n g 878 dispatches o f w h i c h 512 were retrieved at random for text w i t h i n fifty words or one o f the search words.
B.2
TROOPS I N L E B A N O N - - 1 9 8 3 - 1 9 8 4
A single p o l l series p r o v i d e d o p i n i o n for whether more, same, or less troops should be sent to Lebanon i n 1983-1984 (Table B.2). As for defense spending, the N o Opinions were i n the 5-10 percent range and were subtracted f r o m the total. The other opinions were renormalized to 100 percent and the resulting values were used for the remainder o f the calculations. Pertinent A P dispatches were again retrieved f r o m the Nexis data base, searching for ( L E B A N ! and ( ( A M E R I C A ! or U.S. o r U N I T E D S T A T E S ) preceding b y t w o words or less the words ( T R O O P o r M A R I N E or F O R C E ) , f r o m M a r c h 26, 1983, to January 17, 1984. The search began six months before the first p o l l p o i n t and ended w i t h the last p o l l date. The search yielded 1,517 dispatches among w h i c h 467 were retrieved at random for 1,570,000 characters o f t e x t As for defense spending, the retrieval was for text w i t h i n fifty words of one o f the search words.
B.3
DEMOCRATIC
PRIMARY-1983-1984
The polls for candidate preference before the Iowa caucuses were f r o m A B C News and ran f r o m June 19, 1983, to February 15, 1984 (Table B.3). D u r i n g this t i m e , the major candidates were John G l e n n and W a l t e r M o n d a l e . There were other
Copyrighte
157 candidates, but none of their percentages ever exceeded 15 percent so those percentages were all pooled. AP dispatches were retrieved i f they contained the name of at least one of the candidates tallied in the polls. The search was for (REUBEN preceding ASKEW by two words or less) or (ALAN preceding CRANSTON by two words or less) or (JOHN preceding GLENN by two words or less) or (GARY preceding HART by two words or less) or (ERNEST preceding HOLLINGS by two words or less) or (JESSE preceding JACKSON by two words or less) or (GEORGE preceding MCGOVERN by two words or less) or (WALTER preceding MONDALE by two words or less) from December 19, 1982, to February 15, 1984. As for troops in Lebanon, the search began six months prior to the first poll date. Although the last names would probably have been sufficient for relatively rare names like Cranston and Mondale, it was necessary to include the first names due to more common names, of which Jackson would have been the most ambiguous. Therefore, the search used the condition that the first name of every presidential candidate must precede the last name by no more than two words. With this condition, a middle initial could be present in the names some of the time and missing in others. The search yielded 2,435 dispatches of which 425 (1,100,000 characters) were retrieved at random for text within fifty words of a search word. B.4 ECONOMIC CLIMATE--1980-1984 The polls for public opinion on the economic climate were taken by ABC News and covered a three-year period from March 1981 to January 1984 ( Table B.4 ). The No Opinion category was very low at all times, never exceeding 3 percent, so this fraction was subtracted and the other percentages were renormalized to 100 percent for the calculations. AP dispatches in the Nexis data base were searched for (ECONOM! within twenty-five words of (CONDITION! or HEALTH or PROSPECT! or FUTURE or FORECAST! or OUTLOOK! or PROJECT!)) from September 6, 1980, to January 17, 1984. This search also began six months before the first poll date. A total of 12,393 dispatches were identified of which 461 (730,000 characters) were retrieved at random for text within fifty words of a search word. B.5 UNEMPLOYMENT VERSUS INFLATION--1977-1980 Polls from NBC News asked about the relative importance of unemployment and inflation ( Table B.5 ). The 3 percent or less of the population who were not sure were subtracted and the remaining percentages were renormalized. AP dispatches on this topic were identified searching for (UNEMPLOY! within twentyfive words of INFLATION) from January 1, 1977, to August 23, 1980. The search began with the beginning of the data base in January 1977, about three months before the first poll date. The search identified 1,591 AP dispatches of which most (1,582 with 2,300,000 characters) were randomly retrieved for text within fifty words of a search word. B.6 CONTRA AID--1983-1986 Polls on the topic of whether aid should be sent to the Contras fighting the government of Nicaragua were obtained from the four organizations listed in Table B.6. Despite significant wording differences from poll to poll, there was very little change in opinion during the entire polling period, so all polls were pooled. The criteria for choosing the polls was that they ask about American opinion on either aid with no qualifiers or on both military and nonmilitary aid. No published polls found
158
Appendix
B
before the last p o l l i n Table B.6 were excluded i f they met these criteria and were i n the P O L L data base at the Roper Center. T h e CBS-New York Times p o l l was obtained independently f r o m C B S N e w s and was the o n l y a d d i t i o n a l p o l l found meeting the criteria given above. For a l l tables, the p o l l i n g date was assumed to be the m i d p o i n t between the beginning and the end o f the p o l l i n g period. Where no date was given, the midpoint o f the polling month was used.
Copyrights
The data were f r o m four different variants o f polls on defense spending compiled w i t h the aid o f B. I . Page and R. Y . Shapiro, and their colleagues at the N a t i o n a l O p i n i o n Research Center. The symbols are the ones used i n Figure 3 . 1 . W h e n o n l y the month was available, the poll date was assigned to be the middle of the month. P O L L V A R I A N T N B C 1 : Source: N B C N e w s , 30 Rockefeller Plaza, New Y o r k , N Y 10020. Question: D o y o u t h i n k that the defense budget for next year should be increased, decreased or should i t be kept the same as i t is now? Responses: (1) Increased, (2) Kept the same as now, (3) Decreased, (4) N o t sure. P O L L V A R I A N T N B C 2 : Source: N B C N e w s . Question: D o y o u t h i n k the federal government's spending next year on defense and the m i l i t a r y s h o u l d be increased, decreased, or kept about the same? Responses: (1) Increased, (2) Kept about the same, (3) Decreased, (4) N o t sure. P O L L V A R I A N T GSS: Source: General Social Survey, N a t i o n a l O p i n i o n Research Center, 6030 E l l i s Ave., Chicago, I L 60637. Question: W e are faced w i t h m a n y problems i n this c o u n t r y , none o f w h i c h can be s o l v e d easily o r inexpensively. I ' m going t o name some o f these problems, and for each one I ' d l i k e you to tell me whether y o u think we're spending too m u c h money o n i t , too little m o n e y , o r about the right a m o u n t . The m i l i t a r y , armaments a n d defense. Responses: (1) T o o little, (2) About right, (3) T o o much, (4) Don't k n o w . P O L L V A R I A N T ROPER: Source: Roper Center f o r Public O p i n i o n Research, P.O. Box 440, Storrs, C T 06268. Question and responses: Identical to those for variant GSS above. Table BJ.
Polls on the desirability
Symbol and Poll Source A V A a V • V O • • • A 0 0 0 V A 0 V A 0 A
GSS ROPER GSS NBC1 ROPER NBC1 ROPER NBC1 NBC1 NBC1 NBC1 GSS NBC2 NBC2 NBC2 ROPER GSS N BC2 ROPER GSS NBC2 GSS
of increasing
defense spending.
Percent Response Date
03/ /77 12/07/77 03/ /78 10/17/78 12/06/78 12/12/78 02/05/79 09/ /79 12/12/79 01/18/80 01/30/80 03/ /80 01/22/81 02/ /81 11/17/81 12/09/81 03/ /82 03/30/82 12/08/82 03/ /83 0 1 / /84 03/ /84
0)
a)
(3)
(4)
23.6 23 27.0 28 31 24 41 38 51 63 69 56.3 65 63 34 29 29.4 24 19 24.1 23 17.3
45.4 40 43.6 45 35 47 35 36 31 21 19 25.7 23 25 47 38 35.8 47 37 37.8 46 41.2
22.9 24 21.8 21 23 22 16 16 9 8 5 11.5 6 8 14 27 30.1 25 38 32.5 26 38.1
8.1 13 7.6 6 11 7 9 10 9 8 7 6.5 6 4 5 7 4.7 4 6 5.6 5 3.5
Copyrights
Table
B.2.
ABC
News
Poll
on the stationing
of American
troops in
Lebanon.
Results f r o m A B C News Poll, 7 West 66th Street, N e w Y o r k , N Y 10023; Report 95 i n 1984 compiled w i t h the aid o f B . I . Page and R. Y . Shapiro and their colleagues at the National O p i n i o n Research Center. The question was: W o u l d y o u say the U.S. should send more troops to Lebanon, leave the number about the same, o r remove the troops that are there now? The responses were; Send more troops; Leave number the same; Remove troops there now; N o o p i n i o n .
Date
09/26/83 10/23/83 10/25/83 10/27/83 11/07/83 12/13/83 01 /03/84 01 /04/84 01/17/84
Send More
7 21 31 17 13 9 5 8 7
Leave Same
48 21 26 36 41 38 30 29 31
Remove
40 48 39 42 39 48 59 57 58
No Opinion
5 10 5 5 7 5 6 6 4
160
Copyrights
Results f r o m A B C N e w s P o l l c o m p i l e d w i t h the aid o f B. I . Page and R. Y. Shapiro and (heir colleagues at the National O p i n i o n Research Center. This question was asked to registered voters w h o identify themselves either as Democrats o r as independents w h o lean t o w a r d the Democrats: Imagine your state holds a Democratic p r i m a r y and these are the candidates: Reuben Askew, A l a n Cranston, John Glenn, Gary Hart, Ernest H o l l i n g s , Jesse Jackson, George VtcGovem, and Walter Mondale. Whether you are a Democrat o r not, f o r w h o m w o u l d y o u vote: A s k e w , C r a n s t o n , G l e n n , H a r t , H o l l i n g s , Jackson, M c G o v e r n , or Mondale? (slight variation o f w o r d i n g starting September 2 6 , 1983). T h e responses for M o n d a l e , G l e n n , and N o O p i n i o n were tabulated separately. A l l other opinions were pooled and included volunteered responses for other m i n o r candidates and for those w h o said they w o u l d not vote. T h i s last category was 1-2 percent i n all polls. Table BJ.
Date
06/19/83 08/01/83 09/26/83 11/07/83 12/13/83 01/15/84 01/17/84 02/15/84
ABC
News
Poll on the Democratic
Mondale
42 43 36 47 44 51
45 55
Glenn
primary.
Others
28 28 26 21 23 11 22 13
24 23 26 19 24 19 27 21
No Opinion
6 6
11 9 7 18
5 9
161
Copyrighte
Table B.4. ABC News Poll on the economic climate. Results f r o m A B C N e w s P o l l c o m p i l e d w i t h the aid o f B . I . Page and R. Y . Shapiro and their colleagues at the N a t i o n a l O p i n i o n Research Center. The question was: D o y o u t h i n k the nation's economy is: Getting better; Getting worse; Staying the same; N o o p i n i o n .
Date
03/06/81 05/20/8 09/20/81 10/18/81 11/22781 12/12/82 01/30/82 03/08/82 04/25/82 08/17/82 09/13/82 10/11/82 12/18/82 01/22/83 03/02/83 04/12/83 05/15/83 06/19/83 08/01/83 09/26/83 11/07/83 12/13/83 01/17/84
Better
9 14 12 17 11 12 17 13 21 17 21 21 20 18 39 37 43 36 50 44 44 46 49
Same
Worse
36 36 44 41 22 32 31 27 30 31 33 28 26 36 39 40 39 42 30 35 36 31 31
54 49 42 40 55 54 50 59 47 50 45 48 52 46 21 21 17 20 19 20 20 20 19
No Opinion
2 1 2 2 1 2 2 1 2 2 1 3 1 1 1 2 1 2 0 1 1 2 1
162
Copyrighte
163
Table B.5. NBC News Poll on the importance of unemployment versus inflation. Results from NBC News Poll compiled with the aid of B. I . Page and R. Y. Shapiro and their colleagues at the National Opinion Research Center. The question was: In your opinion which is the more important problem facing the country today--finding jobs for people who are unemployed or holding down inflation? The responses were: Finding jobs; Both equal; Holding down inflation; Not sure. Date 03/22/77 04/26/77 08/03/77 03/22/78 05/02/78 06/28/78 08/08/78 09/02/78 11/14/78 12/12/78 03/20/79 09/11/79 05/29/80 07/09/80 08/06/80 08/23/80
Unemployment 43 41 50 39 32 33 28 27 22 22 23 21 30 30 5 26
Equal 18 14 11 10 9 10 11 9 8 9 11 10 15 15 14 20
Inflation 37 43 36 49 56 55 59 61 69 68 64 67 52 53 48 53
Not Sure 2 2 3 2 3 2 2 3 1 1 2 2 3 2 3 1
Results f r o m (he p o l l i n g organizations indicated. A l l data f r o m the Roper Center for Public O p i n i o n Research and C B S News, 524 W . 57th St., New Y o r k , N Y 10019. The question wordings by poll number are: 1. (President Reagan has taken a number o f steps i n Central A m e r i c a to meet what he says is the mounting supply o f arms f r o m Russia and Cuba going to left-wing rebel forces i n E l Salvador and to the Sand in is ta government in Nicaragua.) Let me ask y o u i f you favor o r oppose arming and supporting the rebels i n N i c a r a g u a w h o are t r y i n g to o v e r t h r o w the Sandinista government in that country? Favor; Oppose; N o t sure. 2. D o y o u favor or oppose the U.S. a r m i n g and supporting the rebels i n Nicaragua w h o are t r y i n g to overthrow the Sandinista government i n that country? Favor; Oppose; N o t sure. 3. ( N o w let me read you some statements about President Reagan's handling o f foreign affairs. For each, tell me i f you agree or disagree.) (InterviewerRotate Question Order)...It is w r o n g for the C I A (Central Intelligence A g e n c y ) to help finance the anti-Sandinista forces i n Nicaragua? Disagree; Agree; N o t sure. 4 . D o you favor or oppose...the U.S. ( U n i t e d States) a r m i n g and supporting the rebels i n Nicaragua, w h o are t r y i n g to o v e r t h r o w the Sandinista government i n that country? Favor; Oppose; N o t sure. 5. Should the U n i t e d States be g i v i n g assistance to the g u e r r i l l a forces n o w opposing the M a r x i s t government i n Nicaragua? Yes; N o ; D o n ' t k n o w . 6. President Reagan recently asked Congress to authorize $100 m i l l i o n i n U.S. a i d to the rebels seeking to o v e r t h r o w the c o m m u n i s t g o v e r n m e n t i n Nicaragua, including $70 m i l l i o n for m i l i t a r y purposes and $30 m i l l i o n for non-military purposes, such as food and medical supplies. D o y o u think the Congress should o r should not authorize this new a i d package? S h o u l d authorize (includes 2 percent volunteering should authorize n o n - m i l i t a r y o n l y ) ; Shouldn't authorize; N o o p i n i o n . 7. The House o f Representatives has refused Reagan's request for 100 m i l l i o n dollars i n m i l i t a r y and other aid to the contra rebels i n Nicaragua. D o y o u approve or disapprove o f that action by the House? Disapprove; A p p r o v e ; Don't know/No o p i n i o n . 8. D o you favor or oppose the U.S. sending $100 m i l l i o n i n m i l i t a r y and n o n military aid to the Contra rebels i n Nicaragua? Favor; Oppose; N o t sure. 9. D o you think the U.S. government should g i v e $100 m i l l i o n i n m i l i t a r y and other a i d to the Contras t r y i n g to o v e r t h r o w the g o v e r n m e n t i n Nicaragua? Yes, should; N o , shouldn't; N o o p i n i o n . Table B.6.
Polls on the desirability
Poll Number and Source
1 2 3 4 5 6 7 8 9
HARRIS HARRIS HARRIS HARRIS GALLUP GALLUP ABC HARRIS CBS
of sending Contra
aid.
Percent Response Dale Favor
08/20/83 09/12/83 07/10/84 03/04/85 08/28/85 03/07/86 03/22/86 04/07/86 04/08/86
66 60 55 53 58 52 60 62 62
Oppose
23 24 32 36 29 37 35 33 25
Don't K n o w
11 16 13 11 13 11 4 5 13
Copyrights
165 Appendix C
Summaries of Text Analyses C.1 STRATEGY FOR CONTENT ANALYSIS BY SUCCESSIVE FILTRATIONS Text was first processed through a series of "filter" program runs to remove irrelevant material. Finally the remaining, fairly homogeneous text was scored for its support of each of the polled positions. The outline of these steps is provided in Chapter 3. This appendix summarizes the dictionaries and rules used for the six analyses in this book. All dispatches were given infon content scores corresponding to the positions for which poll data were available (data in Appendix B). To illustrate, a single dispatch is followed in detail through all the analytic steps for the defense spending example. C.2 TEXT ANALYSIS FOR DEFENSE SPENDING--INCLUDING DETAILED EXAMPLE 1. Filtration to select for dispatches on American defense spending The first step was a filtration to discard all dispatches not directly relevant to American defense spending. The entire text was marked for words referring to America (denoted by { A } ) , defense (denoted by { D } ) , and spending (denoted by {S}). Articles with all three word classes were retained for further analysis unless they also had the word "aid" or "fund," which led to the story being rejected. The following actual dispatch, dated February 19, 1983, was kept since it had many America, defense, and spending words and neither of the prohibited words "aid" or "fund": SECTION: Washington Dateline LENGTH: 576 words BYLINE: By MAUREEN SANTINI, Associated Press Writer DATELINE: WASHINGTON KEYWORD: { A } Reagan-{D}Defense
166
Appendix
C
President { A J R e a g a n , i n v o k i n g the menace o f A d o l f H i t l e r , asked { A } C o n g r e s s on Saturday to suppress the urge to reduce his " m i n i m a l " 1984 {DJmilitary {S}budget I n his weekly radio broadcast f r o m the W h i t e { A } H o u s e , the president said his $238.6 b i l l i o n { D j d e f e n s e {SJspending proposal for fiscal year 1984, w h i c h begins O c t . 1, was necessary "unless we're w i l l i n g t o g a m b l e w i t h o u r immediate security and pass on to future generations the legacy o f neglect we inherited." "That k i n d o f neglect w o u l d o n l y weaken peace and stability i n the w o r l d , both now and i n the years ahead." { A J R e a g a n s a i d , " N o w , I k n o w this is a h a r d t i m e to c a l l f o r increased { D } d e f e n s e {SJspending. I t isn't easy t o ask { A } A m e r i c a n families w h o are already m a k i n g sacrifices in the recession ..." " O n the other hand, it's always very easy a n d very t e m p t i n g p o l i t i c a l l y to come up w i t h arguments f o r neglecting { D j d e f e n s e {SJspending i n t i m e o f peace," the president said. " O n e o f the great tragedies o f this century was that i t was o n l y after the balance o f power was allowed to erode and a ruthless adversary, A d o l f H i t l e r , deliberately w e i g h e d the risks and decided to strike that the importance o f a strong {Djdefense was realized too late." Though {AJReagan called for an overall freeze on domestic {SJspending i n his 1984 {SJbudget, the {D}defense portion increased by 14 percent A n d that was after the president cut $8 b i l l i o n f r o m the Pentagon { S Jrequest before submitting i t to {AJCongress. { A J R e a g a n said he and his administration had "agonized" over the current { D j d e f e n s e { S J b u d g e t by t r i m m i n g { S j r e q u e s t s and c u t t i n g non-essential programs. " T h e { D j d e f e n s e {SJbudget we f i n a l l y presented is a m i n i m a l {SJbudget t o protect our country's v i t a l interests and meet o u r commitments," he said. The president said i t was "far better to prevent a crisis than to have to face i t unprepared at the last m o m e n t That's w h y w e have an o v e r r i d i n g m o r a l obligation to invest n o w , this year, i n this {SJbudget, i n restoring { A J America's strength to keep the peace and preserve our freedom." H e said the Soviet U n i o n outspends the { A } U n i t e d States o n ... "... fits and starts," he said, " w e w i l l never convince the Soviets that it's i n their interests t o behave w i t h restraint and negotiate genuine { D J a r m s reductions. W e w i l l also burden the { A J A m e r i c a n taxpayer t i m e and again w i t h the h i g h {SJcost o f crash rearmament" "Sooner o r later, the bills f a l l due." { A J S e n a t e M i n o r i t y Leader Robert C. B y r d o f W e s t V i r g i n i a gave the Democratic response to Reagan's comments and took issue w i t h the president's contention that c u t t i n g the administration { D j d e f e n s e {SJbudget {SJrequest w o u l d expose the country t o danger. " F o r example, we d o not need t w o new manned { D } b o m b e r s - one o f w h i c h w i l l be obsolete almost immediately after i t is b u i l t , " B y r d said, referring t o the B - l { D J b o m b e r under construction and the advanced Stealth plane expected to emerge f r o m development late in this decade. A r g u i n g that the national { D j d e f e n s e depends on a strong economy, B y r d stressed the need for greater... Ellipses (...) i n the text indicate that the remainder o f a sentence was n o t retrieved due to the text being further than fifty words f r o m one o f the seven search
Copyrighte
Summaries
of Text Analysis
167
words: D E F E N S E , M I L I T A R Y , A R M S , S P E N D ! , E X P E N D I T U R E , F U N D ! , or B U D G E T ! (see Appendix B ) . After this filtration, the total number o f characters o f text dropped f r o m 820,000 to 600,000. The average number o f words per dispatch increased f r o m 148 t o 199. T h e dispatch n u m b e r dropped more dramatically ( f r o m 692 to 377) than the number o f characters o f text because very little was retrieved f r o m irrelevant dispatches. The c o l l e c t i o n was stopped as soon as a story was seen t o be not pertinent d u r i n g the retrieval f r o m the Nexis data base. T h e increase i n average w o r d c o u n t per dispatch was a natural consequence o f discarding dispatches f r o m w h i c h very few words were collected.
2.
Filtration to select for paragraphs on defense spending
The second text analysis step selected only paragraphs directly discussing defense spending. The c o n d i t i o n was that a defense w o r d (denoted by { D } ) be close t o a spending w o r d (denoted by { S } ) . The paragraphs f r o m the dispatch given above were scored using this rule. T h e decision for each paragraph is g i v e n directly below the paragraph: President Reagan, i n v o k i n g the menace o f A d o l f H i t l e r , asked Congress o n Saturday to suppress the urge to {SJreduce his " m i n i m a l " 1984 { D J m i l i t a r y {SJbudget. A B O V E P A R A G R A P H W A S KEPT. I n his w e e k l y radio broadcast f r o m the W h i t e House, the president said his $238.6 b i l l i o n { D j d e f e n s e {SJspending proposal for fiscal year 1984, w h i c h begins O c t l , was necessary "unless we're w i l l i n g to g a m b l e w i t h o u r immediate security and pass o n to future generations the legacy o f neglect we inherited." A B O V E P A R A G R A P H W A S KEPT. "That k i n d o f neglect w o u l d o n l y weaken peace and stability i n the w o r l d , both n o w and i n the years ahead." ABOVE PARAGRAPH WAS DISCARDED. Reagan said, " N o w , I k n o w this is a hard t i m e t o c a l l f o r {SJincreased { D j d e f e n s e {SJspending. I t isn't easy to ask American families w h o are already making sacrifices i n the recession ..." A B O V E P A R A G R A P H W A S KEPT. " O n the other hand, it's always very easy and very t e m p t i n g p o l i t i c a l l y to come up w i t h arguments for neglecting { D j d e f e n s e { S J s p e n d i n g i n t i m e o f peace," the president said. A B O V E P A R A G R A P H W A S KEPT. "One o f the great tragedies o f this century was that i t was o n l y after the {SJbalance o f power was allowed to erode and a ruthless adversary, A d o l f Hitler, deliberately weighed the risks and decided to strike that the importance o f a strong {Djdefense was realized too late." ABOVE PARAGRAPH WAS DISCARDED. T h o u g h Reagan called for an overall freeze on domestic {SJspending i n his 1984 {SJbudget, the {Djdefense portion {SJincreased b y 14 {SJpercent. A n d that was after the p r e s i d e n t s J cut $8 b i l l i o n f r o m the {DJPentagon { S Jrequest before submitting i t to Congress. A B O V E P A R A G R A P H W A S KEPT. Reagan said he a n d his a d m i n i s t r a t i o n had " a g o n i z e d " o v e r the current { D j d e f e n s e {SJbudget by t r i m m i n g {Sjrequests and { S J c u t t i n g non-essential programs.
Copyrights
168
Appendix
C
A B O V E P A R A G R A P H W A S KEPT. "The { D j d e f e n s e {SJbudget we f i n a l l y presented is a m i n i m a l ( S ) b u d g e t to protect our country's v i t a l interests and meet our commitments," he said. A B O V E P A R A G R A P H W A S KEPT. The president said i t was far better to prevent a crisis than to have to face i t unprepared at the last moment. T h a t ' s w h y we have an o v e r r i d i n g m o r a l o b l i g a t i o n to invest n o w , this year i n this {SJbudget, i n restoring America's {SJstrength to keep the peace and preserve our freedom," ABOVE PARAGRAPH WAS DISCARDED. He said the Soviet U n i o n outspends the United States on ... A B O V E PARAGRAPH W A S DISCARDED. "... fits and starts," he said, " w e w i l l never convince the Soviets that it's i n their interests t o behave w i t h restraint and negotiate g e n u i n e { D } arms {S J reductions. W e w i l l also burden the American taxpayer time and again w i t h the high { S } cost o f crash rearmament." A B O V E P A R A G R A P H W A S KEPT. "Sooner or later, the bills f a l l due." ABOVE PARAGRAPH WAS DISCARDED. Senate M i n o r i t y Leader Robert C. B y r d o f West V i r g i n i a gave the Democratic response to Reagan's comments and took issue w i t h the president's contention that { S J c u t t i n g the administration { D j d e f e n s e {SJbudget {SJrequest w o u l d expose the country to danger. A B O V E P A R A G R A P H W A S KEPT. " F o r example, we do not need t w o new manned { D J b o m b e r s , one o f w h i c h w i l l be obsolete almost immediately after i t is b u i l t , " B y r d said, referring to the B - l {DJbomber under construction and the advanced Stealth {SJplane expected to emerge f r o m development late i n this decade. A B O V E PARAGRAPH WAS DISCARDED. A r g u i n g that the national { D j d e f e n s e depends o n a strong economy, B y r d stressed the need for greater... ABOVE PARAGRAPH WAS DISCARDED. This sample dispatch was chosen because it illustrates most o f the features o f the text analysis. I n consequence, this story was one o f the most complex found and was somewhat atypical i n c o n t a i n i n g a substantial amount o f i n f o r m a t i o n i n d i r e c t l y relevant to the topic o f defense spending. M o r e frequendy, the discarded text was o n a topic other than defense spending. I n press conferences, for instance, the shifts i n topic c o u l d be quite a b r u p t Nevertheless, the relevant thoughts i n the discarded text, even i n the above example, were almost always also f o u n d i n the retained t e x t For example, the first discarded paragraph was an expansion on the point i n the previous, retained paragraph rather than being a new idea. Also, the next to the last discarded paragraph o n l y illustrated the point i n the previous, retained paragraph. A l t h o u g h a s m a l l a m o u n t o f relevant i n f o r m a t i o n may have been lost by discarding the paragraphs w i t h pertinent information w h i c h was i n d i r e c t the gain was the immense s i m p l i f i c a t i o n o f the subsequent analysis, w i t h the total text f r o m a l l dispatches being reduced from 600.000 characters to 220,000.
3.
Numerical scoring for three positions o n defense spending
The paragraphs retained f r o m the second filtration described above were then scored for favoring more, same, o r less defense spending. Since the second filtration
Copyrights
Summaries
of Text A nalysis
16V
had already guaranteed that a defense w o r d was close to a spending w o r d , the scoring o n l y depended on a defense w o r d (denoted by { D } ) being close to modifiers i m p l y i n g these three positions. T h e modifiers fell into the three classes favoring more (denoted by { M J ) , same (denoted b y { S } ) , and less (denoted b y { L } ) - - w i t h a "less" w o r d close to a " m o r e " w o r d being equivalent t o a "same" w o r d and w i t h a less w o r d close t o another less w o r d also being equivalent t o a same w o r d . I n some combinations, w o r d order and p r o x i m i t y were also i m p o r t a n t I n addition, the prefix " n o n " (denoted by { n } ) preceding a defense word meant that the defense w o r d was not considered t o be relevant t o the military. A l l paragraphs had a total score o f 1.0 w i t h each cluster o f modifier words close t o a defense w o r d contributing t o the final score. I f a paragraph o n l y had one such cluster, the entire paragraph score o f 1.0 was assigned t o the appropriate position. W h e n there was more than one cluster, the score o f 1.0 was d i v i d e d i n t o equal fractions w i t h each cluster receiving one part. This scoring procedure is illustrated using the retained paragraphs o f the dispatch considered above. T h e score for each paragraph is given immediately f o l l o w i n g die paragraph. The scores were for whether the paragraph favored more, same, and/or less defense spending. Comments f o l l o w i n g the scores explain w h y the computer arrived at the decisions. President Reagan, i n v o k i n g the menace o f A d o l f H i t l e r , asked Congress o n Saturday t o {LJsuppress the urge t o { L } r e d u c e his " m i n i m a l " 1984 { D J m i l i t a r y budget. S C O R E F A V O R I N G : More=0.00 Same-1.00 Less=0.00 The "suppression" o f a " r e d u c t i o n " i m p l i e d f a v o r i n g the same level f o r the " m i l i t a r y " b u d g e t The three words were treated as a cluster because they were close to each other. I n his w e e k l y radio broadcast f r o m the W h i t e House, the president said his $238.6 b i l l i o n { D j d e f e n s e spending proposal for fiscal year 1984, w h i c h begins Oct. 1, was {MJnecessary unless we're w i l l i n g t o gamble w i t h o u r immediate security a n d pass o n t o future generations the legacy o f { M J n e g l e c t w e inherited." S C O R E F A V O R I N G : More=0.00 Same=0.00 Less-0.00 N o score here. "Necessary" and n e g l e c t " w h i c h i m p l i e d more spending, were Kx> far away f r o m "defense." Reagan said, " N o w , I k n o w this is a hard time t o c a l l f o r { M } i n c r e a s e d { D j d e f e n s e spending. I t { L J i s n ' t easy to ask American families w h o are already m a k i n g sacrifices i n the recession ..." S C O R E F A V O R I N G : M o r e - 1 . 0 0 Same=0.00 Less=0.00 The operative w o r d c o m b i n a t i o n was "increased" "defense." T h e w o r d " i s n ' t " only changed the sense o f words like "increased" i f i t preceded them. " O n the other hand, it's always very easy a n d very t e m p t i n g p o l i t i c a l l y t o come { M J u p w i t h arguments f o r { M J n e g l e c t i n g { D j d e f e n s e spending i n time o f peace," the president said. S C O R E F A V O R I N G : M o r e - 1 . 0 0 Same=0.00 Less=0.00 T h e scored c o m b i n a t i o n was " u p " ... " n e g l e c t i n g " "defense". " U p " and " n e g l e c t i n g " were scored together as meaning more s h o u l d be spent. T h e inclusion o f words like "neglect" and "inadequate" i n the dictionary d i d permit the public to reason and thereby take indirect information into a c c o u n t These words were included because they were usually found i n the context o f arguments that defense spending should have been increased i f it was neglected or inadequate. T h o u g h Reagan called f o r an overall {SJfreeze o n domestic spending i n his 1984 budget, the {Djdefense portion {MJincreased b y 14 percent A n d that was
Copyrightec
770 Appendix
C
after the president { L } c u t $8 b i l l i o n f r o m the { D } P e n t a g o n request before s u b m i t t i n g i t to Congress. S C O R E F A V O R I N G : More=0.50 Same=0.00 Less=0.50 T h i s paragraph was scored as m a k i n g t w o different statements, one f a v o r i n g more defense spending ("defense" "increased") and one f a v o r i n g less ( " c u t " "Pentagon"). Therefore, the paragraph score o f 1.0 was d i v i d e d i n t w o . "Freeze" was too far away f r o m "defense" to have a connotation for defense, as was consistent w i t h the actual meaning o f the paragraph. Reagan said he and his a d m i n i s t r a t i o n had " a g o n i z e d " o v e r the c u r r e n t { D j d e f e n s e budget by { L J t r i m m i n g requests and { L J c u t t i n g { n } non-essential programs. S C O R E F A V O R I N G : M o r e - 0 . 0 0 Same-1.00 Less-0.00 The reasonable score favoring unchanged military spending was serendipitous. T h i s score was due to " t r i m m i n g ' ' and " c u t t i n g " being equivalent to the concept o f same spending. This combination close t o "defense" gave the score favoring same spending. T h e " n o n " d i d not have a function here. I f the w o r d "defense" occurred i n place of the w o r d "essential," then the concept o f defense w o u l d have been n u l l i f i e d , indicating that the topic was not about defense. "The { D j d e f e n s e budget we f i n a l l y presented is a m i n i m a l budget to protect o u r country's vital interests and meet our commitments," he said. S C O R E F A V O R I N G : M o r e - 0 . 0 0 Same=0.00 Less-0.00 T h i s paragraph had no score since "defense" was close to no modifier words. I n fact, when this paragraph was read by itself, i t was consistent w i t h any o f the p o s i t i o n s . T h e actual i n f o r m a t i o n f a v o r i n g one p o s i t i o n o r another was elsewhere i n the text. "... fits and starts," he said, " w e w i l l never convince the Soviets that it's i n their interests to behave w i t h restraint and negotiate genuine arms {LJreductions. W e w i l l also burden the A m e r i c a n taxpayer time and again w i t h the { M J h i g h cost o f crash rearmament." S C O R E F A V O R I N G : M o r e - 0 . 0 0 Same=0.00 Less-0.00 T h i s paragraph also had n o score since there were no words d i r e c t l y c o n n o t i n g defense. I t c o u l d be argued that i t favored more defense spending i n d i r e c t l y . However, the statement is probably weaker than those above speaking d i r e c d y t o the issue. Senate M i n o r i t y Leader Robert C. B y r d o f West V i r g i n i a gave the Democratic response to Reagan's comments and took issue w i t h the president's contention that { L J c u t t i n g the administration { D j d e f e n s e budget request w o u l d expose the country to {LJdanger. S C O R E F A V O R I N G : M o r e - 0 . 0 0 Same-0.00 Less-1.00 The score o f favoring less spending came f r o m " c u t t i n g " ... "defense." "Danger" was too far f r o m "defense" to be scored. The score was p r o b a b l y correct, although a sounder basis for the conclusion w o u l d have included: " t o o k issue"..."cutting"..."defense"., "danger." The final score for this dispatch was 2.5 paragraphs favoring more, 2.0 favoring same, and 1.5 favoring less spending. T h i s sample dispatch was one o f the most complex retrieved. Six o f the nine paragraphs were scored for supporting one o f the three positions. For comparison, the average number o f relevant paragraphs was only 1.7 among all dispatches w i t h at least one paragraph w i t h a positive score. Since the average A P paragraph had approximately thirty words, the final scoring came f r o m approximately fifty words per story although approximately eighty words w e r e e x a m i n e d in e a c h s c o r e d d i s p a t c h .
Copyrights
Sunwiaries
of Text Analysis
171
The difference between the fifty and eighty words meant that 30-40 percent o f the paragraphs had n o score. T h i s was true for three o f the nine paragraphs i n the dispatch just examined. The scoring for defense spending came f r o m paragraphs representing 5-10 percent o f the words (400-900) i n the average dispatch. Only 20 percent o f the dispatches had fewer words and 10 percent had more. Being f a i r l y l o n g this dispatch also illustrated most o f the scoring features. I n fact, the problems were more severe f o r this text than for most others. The more typical dispatches had smaller numbers o f relevant paragraphs and usually made their points about defense spending quite directly before proceeding to other topics. There tended to be fewer crosscurrents to complicate the scoring. The most appropriate base for considering the scoring is the 377 dispatches retained after the first filtration step. The others were not about defense spending or were about n o n - A m e r i c a n forces. O f these 377, 72 percent were used i n the final scoring. t
4.
Numerical scoring for t w o positions on defense spending
As an alternative to the evaluation j u s t described, the text was also scored to favor o n l y t w o positions—more or less defense spending. T o do so, the concept o f same spending was eliminated. As a result, some modifier words like " m a i n t a i n " and "keep" were omitted f r o m the dictionary. These words were previously interpreted to favor the concept o f same spending. Other words, like "freeze" and "frozen." were moved f r o m the same spending class to the modifier class connoting less spending. N o w , a "less" w o r d preceding a " m o r e " w o r d was assigned t o favor less instead o f same spending (e.g., "cut...increase"). S i m i l a r l y , t w o nearby less w o r d s (e.g., "cut.areduction") were also assigned to favor more instead of same. Other dictionary changes included the deletion o f a few words favoring more ("bolster") and less C alternate," "weaken," " w i t h o u t " ) spending. T h e words "nuclear" and "arms" were added t o the list o f words referring to defense. Thus "nuclear arms reduction talks" was interpreted to support less defense spending w h i l e this phrase was simply ignored in the previous scoring. Using this alternate dictionary and its associated rules, the text scored i n the preceding section was rescored. Those paragraphs with changed final scores are listed below w i t h comments: President Reagan, i n v o k i n g the menace o f A d o l f H i t l e r , asked Congress o n Saturday t o {L}suppress the urge to { L } r e d u c e his " m i n i m a l " 1984 { D J m i l i t a r y budget. SCORE F A V O R I N G : More»1.00 Less-0.00 The "suppress" ... "reduce" was interpreted previously to favor same spending instead of more spending.
Reagan said he and his a d m i n i s t r a t i o n had " a g o n i z e d " o v e r the c u r r e n t { D j d e f e n s e budget by { L J t r i m m i n g requests and { L J c u t t i n g { n j non-essential programs. SCORE F A V O R I N G : M a r e - 1 . 0 0 Less-0.00 The " t r i m m i n g " ... " c u t t i n g " was misscored previously to favor same spending and was misscored this time to favor more spending. A g a i n , the w r o n g score was not entirely inconsistent w i t h the sense o f the paragraph. "... fits and starts," he said, " w e w i l l never convince the Soviets that it's i n their interests t o behave w i t h restraint and negotiate g e n u i n e { D J arms
Copyrighte
/ 72
Appendix
C
{LJreductions. We w i l l also burden the American taxpayer t i m e and again w i t h the { M j h i g h cost o f crash rearmament'' S C O R E F A V O R I N G : M o r e - 0 . 0 0 Less-1.00 Previously, this paragraph had a zero score since " a r m s " was not i n the dictionary i n order to o m i t reference t o arms reduction. Here, ''arms" " r e d u c t i o n " was interpreted to suggest that defense spending, likewise, should be diminished. W i t h these m o d i f i c a t i o n s , the new score was 4.5 paragraphs f a v o r i n g more spending and 2.5 favoring less. The text recoveries d u r i n g the scoring are presented in Table C . l .
5.
Text analysis for defense waste and fraud a.
Filtration
to remove
dispatches
not on American
defense
spending.
The
first step discarded dispatches i f they were not about American waste and fraud. The U.S. was usually not the focus when there was a mention o f a n o n - A m e r i c a n region i n the heading portion prepared by the A P and listed before the body o f a dispatch. T h e heading t y p i c a l l y contained most o f these designators: dateline, headline, k e y w o r d , and section (see example at the beginning o f Section C.2). Therefore, the filtration c o m m a n d s i m p l y removed stories w i t h one o f these designators f o l l o w e d closely by a w o r d referring to a foreign part o f the w o r l d . Dispatches referring to defense against waste and fraud for non-defense topics were also not retained f o r further study i f the stories mentioned other key words such as "hazardous" and " t o x i c " referring to non-military waste. b. Filtration to select paragraphs on defense waste and fraud. This filtration was accomplished by l o o k i n g for w o r d combinations referring to both the defense industry and to waste. S o m e c o m b i n a t i o n s w e r e s i m p l e , s u c h as "overcharge"..."weapons." O t h e r c o m b i n a t i o n s were m o r e c o m p l e x , such as "defense"..."contractor"..."cut corners." C. Numerical scoring for stories on defense waste and fraud. Any word combination suggesting defense waste such as those i n the preceding subsection led to the A P paragraph containing the combination t o be scored as favoring less defense spending. The recovery data for this waste and fraud analysis are given i n Table C.2.
C3
T E X T A N A L Y S I S FOR T R O O P S I N 1.
LEBANON
Filtration to select for paragraphs o n American troops i n Lebanon
The first filtration selected paragraphs containing words referring to America, troops, and Lebanon. A t this step, a mention o f policy or a synonym was considered to be equivalent to troops since p o l i c y often referred to troops. Paragraphs were discarded i f they had w o r d s r e f e r r i n g to n o n - A m e r i c a n troops (e.g., " A r a b , " " C h r i s t i a n , " " D r u s e , " " S y r i a n , " " I s r a e l i " ) , a non-Lebanon region o f the w o r l d (e.g., "Grenada"), or non-military activities (e.g„ "economy"). I n this analysis, paragraphs were considered to be about A m e r i c a or Lebanon after a previous m e n t i o n o f words indicating these geographic areas unless there was a w o r d (e.g., " C h r i s t i a n , " " S y r i a n , " "France") indicating non-American troops or a non-Lebanese location (e.g., "Israel"). Also, pronouns such as " t h e y " and " t h e m " were taken to refer to troops i f there was a mention o f troops in the previous paragraph.
Copyrights
Summaries of Text Analysis 2.
Filtration to remove entertainment
paragraphs o n m i l i t a r y
¡73
action and Christmas
This step removed a l l text o n actual combat and a l l paragraphs o n entertainer B o b Hope's Christmas visit to Lebanon.
3.
Numerical scoring for dispatches o n troops in Lebanon
The final scoring step used the major criterion that a w o r d referring to troops o r p o l i c y should be near m o d i f i e r words favoring more, same, o r less, although some words—such as "stay" and " w i t h d r a w " - - w e r e able by themselves t o favor keeping o r removing troops. Therefore, the paragraphs had i n f o n content scores favoring more, same, o r less troops. The recovery data for the analyses are given i n Tabic C.3.
C,4
T E X T A N A L Y S I S FOR T H E D E M O C R A T I C 1,
PRIMARY
Analysis using bandwagon words
a. Filtration for paragraphs about candidates. First, paragraphs were selected and kept o n l y i f they had the name o f at least one o f the Democratic candidates appearing i n the A B C News Poll o f Table b. Scoring using bandwagon words. T h e n the text was scored f o r being either favorable o r detrimental t o John Glenn, Walter Mandate, o r Others (Reuben As t e w , A l a n Cranston, Gary H a r t , Ernest H o l l i n g s , Jesse Jackson, o r George M c G o v e r n ) . These scores depended o n modifier words i m p l y i n g success or failure being close to a candidate name. T h e scores belonged to six positions, three favorable t o Mondale, Glenn, and Others, and three unfavorable to these candidates. The recovery data are i n Table C-4.
2,
Analysis using name count
The paragraphs i n the o r i g i n a l retrievals were scored w i t h o u t any further filtration steps. Every paragraph was stilt given a total score o f 1.0. I f o n l y one candidate was mentioned i n the paragraph, then the score i n favor o f that candidate was 1,0. I f several candidates were discussed, then that score o f 1.0 was shared among the candidates. This type o f scoring o b v i o u s l y d i d not generate any scores unfavorable to a candidate- Therefore, o n l y three types o f paragraph scores w e i e obtained, those mentioning Mandate, Glenn, and Others (recovery data in Table CA).
C.5
T E X T A N A L Y S I S FOR T H E E C O N O M I C I.
CLIMATE
Filtration to eliminate dispatches o n non-American economies
T h e f i r s t s t e p discarded dispatches i f they were not about the U n i t e d States. T h e procedure was very s i m i l a r t o the one described earlier w h i c h looked f o r n o n - U , 5 . words in the dispatch heading region.
MaTcpnan, samuLUCHHbw aerope*MI* npasoM
¡74
Appendix 2.
C
Filtration to select paragraphs discussing the economy
The next filtration step selected paragraphs w i t h at least one w o r d referring t o some aspect o f the economy. The reference d i d not have to be to the economy as a whole but c o u l d include components such as " a g r i c u l t u r e . " A l s o p e r m i t t e d were words like " r a l l y " describing economic performance.
3.
Numerical scoring
I n the f i n a l step, i n f o n c o n t e n t scores were assigned f r o m single w o r d s suggesting better, same, or worse i n the context o f economic conditions. Therefore, the dictionary had qualifiers like "best," "confusion," and " b a d " d i v i d e d i n t o classes f a v o r i n g better, same, and worse. A d d i t i o n a l dictionary words included those (e.g., " n o t " and " d i f f i c u l t " ) w h i c h could alter the sense o f the q u a l i f i e r words. Since the previous filtration had already guaranteed that each paragraph had to make reference to an aspect o f the economy, the score c o u l d be determined by single words suggesting better, same, o r worse. T h e recovery data for all steps are given i n Table C.5.
C.6
T E X T A N A L Y S I S FOR U N E M P L O Y M E N T V E R S U S 1.
INFLATION
Filtration to eliminate dispatches on non-American economies
T h i s step to remove dispatches o n foreign countries was l i k e the first n i t r a t i o n s for dispatches on defense waste and fraud and the economic climate.
2.
Numerical scoring
T h e scoring was f o r : u n e m p l o y m e n t more i m p o r t a n t , equal i m p o r t a n c e , or inflation more important. The m a i n c r i t e r i o n was that inflation, unemployment, or their synonyms should be close t o m o d i f i e r words indicating that the p r o b l e m was i m p o r t a n t . A s i g n i f i c a n t n u m b e r o f paragraphs spoke o f b o t h problems being i m p o r t a n t . I n r e c o g n i t i o n o f this fact, the score was for equal i m p o r t a n c e i f a m o d i f i e r w o r d made i n f l a t i o n i m p o r t a n t and i f an u n e m p l o y m e n t w o r d f o l l o w e d shortly after, as i n the phrase " w e must combat both i n f l a t i o n and u n e m p l o y m e n t ' * S i m i l a r l y , i f a paragraph had one w o r d cluster supporting Ihe importance o f each o f the t w o topics, the problems were considered to be equally crucial (recovery data are i n Table C.6).
C.7
T E X T A N A L Y S I S FOR C O N T R A A I D 1.
Filtration to select paragraphs o n Contra aid
Each paragraph retained i n this filtration step was required t o contain words i m p l y i n g Nicaragua, the U n i t e d States, and funding. For this filtration step i t d i d not matter i f the reference to Nicaragua was to the Contras opposing the government or to the government side. A mention o f Nicaragua meant that the next paragraph was also about Nicaragua unless there was discussion o f another C e n t r a l A m e r i c a n country like El Salvador or Honduras.
Copyrights
Summaries of Text Analysis 1.
175
Numerical scoring by Fan
The paragraphs Surviving the first filtration were scored by the author by l o o k i n g for m o d i f i e r words close to w o r d combinations discussing both die Contras and funding. Combinations o f modifier words were examined for whether they favored or opposed Contra aid. A n example o f a w o r d cluster favoring Contra a i d w o u l d be "approve"..."Contra",,,"aid," I f there were c o n d i t i o n a l w o r d s l i k e " i f " i n the paragraph, the paragraph was considered to favor both positions equally. Therefore, the final scores were for paragraphs either supporting or opposing Contra aid.
3,
Numerical Scoring by Simone French, Peter Miene, and Janet S w i m
The paragraphs scored by Fan were scored independenUy by these three graduate research assistants w o r k i n g as a team. Their scoring method was quite different f r o m that o f Fan, l o o k i n g at w o r d c o m b i n a t i o n s f a v o r i n g o r o p p o s i n g a i d w i t h o u t requiring that these words be close to words discussing C o n t r a a i d . T h e y also included many indirect pieces o f i n f o r m a t i o n that i m p l i e d a position o n Contra aid. For instance, the w o r d cluster "administration",./propaganda" by itself was scored as opposing Contra aid. This could safely be done because the paragraphs were already scored as being relevant to Contra aid by the initial filtration. The recovery data for both scoring methods are given i n Table C.7
Marepnan, 3awnmcHHbin a e r o p c s H M npnsoM
Table C.I. Summary of text analysis for defense spending. The upper portion gives the recoveries o f the text and paragraphs at different stages o f the text analysis. The Nexis search identified 9,314 dispatches, o f w h i c h 692 were retrieved at random. For the calculation, words are assumed to be approximately eight characters long. The lower p o r t i o n gives data for each position scored after the final step i n the upper p o r t i o n o f the table. T h e data f o r " a n y p o s i t i o n " refer to a l l p o s i t i o n s c o m b i n e d .
Step i n Analysis
Characters of Text No.
Dispatches
% Orig.
No.
8-Char. Words
% Orig.
per Dispatch
Nexis Retrieval
820,000
100
692
100
148
First F i l t e r
600,000
73
377
54
199
Second Filter
220,000
27
340
49
81
272 280
39 40
Scoring Runs: Scored to Favor More, Same, Less Scored to Favor M o r e and Less O n l y
Position Favoring
Average Paragraphs In
Total Dispatches
Dispatches W i t h at Least One Paragraph Favoring
W i t h A t Least One Paragraph Favoring T h i s Position
This
Position
Scored to Favor More, Same, Less: More 1.3 Same 1.1 Less 1.2 A n y position 1.7
177 66 132 272
Scored to Favor M o r e and Less O n l y : More 1.3 Less 1.3 A n y position 1.7
197 167 280
I
176
Copyrights
ud. Scoring o f all
Table C.2. Summary mentions of waste an