THESIS (MW) MIICH IGAN STAT TEU I I IZIIIII II II II IIIIIIIIIIIIIIIIIIIIIIIII 076090 II This is to certify that the dissertation entitled THE RELATIVE IMPACI| OF MASS MEDIA ON AMERICAN LIBERALISVI: 1945-1990 presented by Dave D' Alessio has been accepted towards fulfillment of the requirements for Ph. D. degree in Camrunication fl / 1 / Ma jor/professor April 27, 1997 MS U is an Affirmative Action/ Equal Opportunity Institution 0- 12771 LIBRARY Michigan State University PLACE ll RETURN BOX to romovo this checkout from your rooord. TO AVOID FINES Mum on or bdoro date duo. DATE DUE DATE DUE DATE DUE It; =- 3 - r—l—S—W I-I I- II I- lI—I—T MSU loAn Afflrmutlvo Action/Equal Opportunity Institution Who-m .. . o‘c-«O —v V” "‘" THE RELATIVE IMPACT OF MASS MEDIA ON AMERICAN LIBERALISM: 1945- 1990 By Dave D’Alessio A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Communication 1996 ABSTRACT THE RELATIVE IMPACT OF MASS MEDIA ON AMERICAN LIBERALISM: 1945-1990 By Dave D’Alessio Based on a model holding that changes in public opinion are a consequence of a multi-stage process in which interested groups, termed “issue networks”, first act to set the public’s agenda, then prime and subsequently frame issues to the public through the media, the extensive agenda-setting literature is reviewed. From it five basic expectations about how public opinion changes were developed: That the media, and that policy makers would have direct influence on public opinion; that the media would also directly influence policy makers; that changes in liberal opinion would lead in turn to increases in liberal activity; and that the exact form of the model would differ on an issue-by—issue basis. Public opinion data was supplied by the National Opinion Research Center, while a sample of media opinion was drawn from the periodical literature and content analyzed for statements overtly biased in a liberal of conservative manner. Congressional and administrative actions were gathered and coded, as were a variety of external socioeconomic factors. Following data reduction, path models involving lagged variables of lags up to eight years were developed based on correlational techniques. This was done once for an aggregate model, and then twice more for single issue models. Results indicated that while the policy agenda directly influences public opinion in all three models, the media’s impact in two of the three models was indirect, action via its influence on the policy agenda. Changes in public opinion did not appear to be connect with subsequent overt behavior, but the single issue models indicated that the exact interrelationship of media, public and policy varies from issue to issue. The presence of a large number of even numbered lags suggests that public and policy debate, and subsequent opinion change, is tied to congressional election campaigning. ACKNOWLEDGEMENTS First, this research was possible in part to the assistance of the undergraduate and graduate students who did the content analysis: Eric Peltier, 'D'acey Birkenhauer, Chris Atwater, Lisa Noel, Kim Niemiec, Kathleen Starman, Romy Barrett, Sheryl Wnek,Paula Johnson, and especially Tim Rick of Michigan State; Patty Doran of Calvin College; and Cindy Pancurak, Matt Spicer, Elizabeth Goodwin, Lydia J avins and Holly Kingsley of Stockton College. Second, Dr. Tom Smith of the National Opinion Research Center graciously supplied all of the public opinion data for the cost of photocopying. This study reflects the inspiration of much of Dr. Smith’s research as well. Third, as with every dissertation, this could not be possible without the advice and support supplied by the committee: Jim Dearing (Chair), Chuck Atkin, Frank Boster, and Todd Simon. Fourth, as also with every dissertation, no dissertation would ever be possible without the support of the people who believed I could do this even when I wasn’t sure myself: Drs. Mike Allen and Nancy Burrell of the University of Wisconsin - Milwaukee, Dr. Paul Mineo of the University of Connecticut - Hartford, and Mr. and Mrs. John J. D’Alessio of Glen Head, N .Y. iv TABLE OF CONTENTS LIST OF TABLES vii LIST OF FIGURES x CHAPTER ONE LITERATURE REVIEW AND EXPECTATIONS Agenda Setting Studies Priming and Framing Media and Public Opinion Other Agenda-Opinion Relationships Theoretical Background Access to Media Issue Networks and Social Structures Operationalization Liberal Issues and Liberal 'h'ends Liberalism Bias in the Media Potential Drawbacks in the Use of Public Opinion Data Expectations 8&SEEBEEEGEKAH CHAPTER TWO MEASUREMENT AND ANALYSIS Measurement of Public Opinion Determining the Media Agenda Measuring the Policy Agenda Socioeconomic Factors Method of Analysis BEBEEE CHAPTER THREE RESULTS Public Agenda Media Agenda Policy Agenda §sss External Socioeconomic Factors Path Analysis - Overall Model Impact on Output Variables Sub—Analysis: Foreign Affairs Univariate Analysis Path Analysis - Foreign Affairs Model Sub-Analysis: Race and Ethnicity Univariate Analysis Path Analysis: Race and Ethnicity Model Comparison of Path Models Overall Findings CHAPTER FOUR DISCUSSION Public Opinion Change Causation Basic Findings Collaborative Influence External Socioeconomic Factors Role of Nation’s Media Media Bias Communication Generally Caveats Theoretical Considerations Power Concluding Note 165 165 167 168 169 172 175 177 186 188 LISTOFTABLI‘B Table 1 — Measurement of the Factors of Liberalism Table 2 - Gross Changes in Public Agenda and Number of Active Trends by Year and Content Area Table 3 - Topic Headings Used in Reader’s Guide Table 4 - Mean Media Bias by Year and Content Area Table 5 - Congressional Activity by Year and Content Area Table 6 - Administration Agenda by Year and Content Area Table 7 - Socioeconomic Inputs by Year and Input Area Table 8 - Socioeconomic Outputs by Year and Content Area Table 9 - Overall Analysis Matrix of Correlations for Public Opinion by Content Area Table 10 - Overall Analysis Autocorrelations and Partial Autocorrelations Public Opinion Table 11 — Intercoder Reliabilities on Media Agenda by Content Area Table 12 - Overall Analysis Matrix of Correlations for Media Agenda by Content Area Table 13 - Overall Analysis Autocorrelations and Partial Autocorrelations Media Agenda Table 14 - Overall Analysis Matrix of Correlations for Congressional Agenda by Content Area Table 15 - Overall Analysis Matrix of Correlations for Administration Agenda by Content Area Table 16 - Overall Analysis Autocorrelations and Partial Autocorrelations Congressional Agenda Table 17 - Overall Analysis Autocorrelations and Partial Autocorrelations Administration Agenda Table 18 - Overall Analysis Matrix of Correlations for Socioeconomic Inputs by Area Table 19 - Overall Analysis Autocorrelations and Partial Autocorrelations Socioeconomic Inputs - Economy Table 20 - Overall Analysis Autocorrelations and Partial Autocorrelations Socioeconomic Inputs - Propinquity Table 21 - Overall Analysis Matrix of Correlations for Socioeconomic Outputs by Area Table 22 - Overall Analysis Autocorrelations and Partial Autocorrelations Socioeconomic Outputs 28 EB 88? EB §§38833383 83 101 Tables (cont’d) Table 23 — Overall Analysis Lagged Correlations with Public Opinion Table 24 - Overall Analysis Original Matrix of Correlations for Lagged Predictors Table 25 - Overall Analysis Matrix of Corrected Correlations for Lagged Predictors Table 26 - Overall Analysis Matrix of Reconstructed Correlations Table 27 — Overall Analysis Matrix of Residuals Table 28 - Overall Analysis: Output Model Lagged Correlations with Socioeconomic Outputs Table 29 — Overall Analysis: Output Model Matrix of Correlations for Lagged Predictors Table 30 - Overall Analysis: Output Model Matrix of Reconstructed Correlations Table 31 — Overall Analysis: Output Model Matrix of Residuals Table 32 — Sub-Analysis: Foreign Affairs Autocorrelations and Partial Autocorrelations Public Opinion Table 33 - Sub-Analysis: Foreign Affairs Autocorrelations and Partial Autocorrelations Media Agenda Table 34 — Sub-Analysis: Foreign Affairs Autocorrelations and Partial Autocorrelations Administration Agenda Table 35 — Sub-Analysis: Foreign Affairs Autocorrelations and Partial Autocorrelations Congressional Agenda Table 36 - Sub-Analysis: Foreign Affairs Lagged Correlations with Public Opinion Table 37 - Sub-Analysis: Foreign Affairs Matrix of Correlations for Lagged Predictors Table 38 - Sub-Analysis: Foreign Affairs Matrix of Reconstructed Correlations Table 39 — Sub-Analysis: Foreign Affairs Matrix of Residuals Table 40 - Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Public Opinion Table 41 - Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Media Agenda Table 42 — Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Administration Agenda Table 43 - Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Congressional Agenda Table 44 — Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Socioeconomic Inputs: Occupational Segregation by Race Table 45 - Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Socioeconomic Outputs: Election of Black Congressmen Table 46 — Sub-Analysis: Race and Ethnicity Lagged Correlations with Public Opinion viii EEEEQQEEEQEEEEE EE 147 148 151 152 Tables (cont’d) Table 47 — Sub-Analysis: Race and Ethnicity Matrix of Correlations for Lagged Predictors Table 48 - Sub-Analysis: Race and Ethnicity Matrix of Reconstructed Correlations Table 49 - Sub-Analysis: Race and Ethnicity Matrix of Residuals Table 50 - Residual Matrix Comparison of Path Models ix 156 15) 161 LISI‘ OF FIGURES Figure 1 - McCombs’ Four Part Typology of Agenda Setting Studies Figure 2 - The Interrelationship of Media, Policy, and Public Agendas, and Socioeconomic Factors Figure 3 — Net 'h‘ends in Liberalism in the United States, 1945-1990 Figure 4 - Year-by-Year Changes in Net Liberalism, 1945-1990 Figure 5 -- Liberal Trends Demonstrating a “Liberal Plateau” Figure 6 — Liberal Trends Demonstrating a Perturbed Linear Trend Figure 7 — Operational Definition Overall Model Figure 8 - Cumulative Annual 'h‘end, 1940-1994 Figure 9 -— Average Net Trend, 1940-1994 Figure 10 — Overall Analysis - Final Path Model Figure 11 — Overall Analysis: Output Model Final Path Model Figure 12 - Foreign Affairs Sub-Analysis Final Path Model Figure 13 — Race and Ethnicity Sub-Analysis Final Path Model Figure 14 — Frequency Distribution of Net Liberal-Conservative Statements per Article fiééfissacasse q 53 H Chapter One Literature Review and Expectations The history of the study of mass communication effects in the US. is largely a history of a search for impacts on individuals. Whether conducted in the field or in the laboratory, the emphasis is typically on change in the individual subjects; error terms are generally assumed to be made up of the individual differences between the participants. The advantages of research about individuals are many. Small subject pools are readily available, manipulations can be made as strong as necessary (or allowable; Hovland, 1959), experiments or field studies can be constructed in ways which permit strong claims of internal and/or external validity (Campbell and Stanley, 1966), and thus results can be obtained which permit strong causal claims (Hume, 1748/1955). Research techniques appropriate to the study of individuals are widely taught and statistical procedures can be straightforward. Moreover, the micro-level focus on individuals reflects the widespread concern in democratic societies of undue influence on individuals and their rights. ~ However, the emphasis on media effects on the individual may be inappropriate, or at least too restrictive. For instance, the notion of the media as a social force, above and beyond being literally a "medium" or “mirror” (conveyer of a specific story, for example, the Kennedy assassination, Greenberg and Parker, 1965; or type of content such as violence, Milavsky, et al., 1982), with an important role to play in the governing and administration of a nation, is common in academic 1 2 disciplines such as political science (Linsky, 1986), sociology (Lang and Lang, 1968), and journalism (White, 1962, 1965). Communication, often studied through these diverse perspectives, may be especially appropriate to broad, as opposed to restrictive, emphases. One might also argue that to emphasize impacts on the individual is inherently reductionist, with the pitfalls and problems associated with the use of reductionist rather than holistic (or "systems") approaches. Von Bertalanffy (1968) points out that reductionist “covering law” approaches fail to permit higher order, complex interactions, and that the necessity of experimental control requires examination of closed systems; Monge (1973) draws the distinction between systems models, which recognize patterns in behavior, and “covering law” models, which impose patterns on behavior; together, these ideas suggest the possibility of destroying the phenomenon under observation by the act of observing it. Another problem unique to the study of mass communication and mass societies is the problem of effect sizes: If a television program were to be created of such power and impact that it would specifically and unquestionably cause 2,500 of its 25,000,000 viewers to run amok in the streets, scientists would be forced to tease from amongst the gamut of individual differences the very small effect size of .0001. 'II‘eated, however, at the level of the society, 2,500 rioters would create a substantial and discernible impact on the crime statistics of that historical epoch. Finally, there remains the recently abused notion of "mass" media. It is possible for commercially viable media to exist with audiences of only hundreds; it is unquestionably the fashion, for sound economic reasons, for media sources to engage in "narrowcasting", or the search for smaller but demographically and attitudinally more closely defined audiences for 3 which premium rates can be charged (DeFleur and Dennis, 1988). But the media institutions whose "power" is the subject of discussion and debate, protest and pr0posed legislation, receive that "power" directly from the fact that they reach millions of citizens quickly and efficiently, with content over which mass media decision-makers exert editorial control. Approaches to mass media as a monolithic social structure (i.e. “the media” in the same sense as “the government” or “the courts”) have been few, and, for the most part, limited to theoretical positions whose empirical testability is obscured by the broadness of the scope of the position (McLuhan, 1964; Meyrowitz, 1985). Within American culture, for instance, television signals reach working television sets in 98% of households (Biitner, 1985); and many of the 2% who would form a comparison group are so different from the majority in so many other ways as to make conclusions drawn from comparison problematic. Studies on the Innuit (Canadian Eskimos) do not generalize well to the US. population at large. One area where mass communication research has started to move in the direction of considering the media as one social structure (that is, as a coherent body with needs and desires which manifest themselves in the functioning of the society) and the people as a second social structure has been the area of agenda setting research. While many of the studies continue in the reductionist vein, the vast accumulation of public opinion research has permitted the examination of the populace's response as a whole, within the limits of sampling error. And that same accumulation permits analysis of trends, directions, and changes in public opinion, some of which may be explicable in terms of the dynamics of the populace, and some of which may be explicable in terms of the impacts of other structures in the society, including the educational process, the decisions of 4 government, and the media, on that populace. AgendaSetfingStudies Iyengar and colleagues (Behr and Iyengar, 1985; Iyengar and Kinder, 1987; Iyengar and Simon, 1993) increasingly conceptualize public opinion change as a multi-step process in which mediated messages increase the salience of issues (agenda setting), then communicate evaluative frameworks (priming), as well as the “evidence” on which evaluation is based (framing). This suggests that Opinion changes created by media consumption is a multi-stage, contingent process1 whose overall efliciency is limited by the efficiencies of the separate, intervening processes (Kenny, 1979). Each of the various stages has been the subject of some degree of empirical research. Agenda setting as a demonstrable communications phenomenon is generally traced to Maxwell McCombs’ and Donald Shaw's (1972) classic correlation (r=.967) between the number of news articles on a set of issues and the public's ranking of those issues in terms of importance during the 1968 election campaign. Since then, the ability of the media to set the public's agenda (that is, as Bernard Cohen, 1963, paraphrased "(The Press) is stunningly successful in telling its readers what to think about. ") has been demonstrated in a number of ways under a variety of conditions. Recent reviews (Weaver, 1984; Rogers and Dearing, 1988) leave little doubt of the generalizability of the McCombs and Shaw data, at least at the ' In short, the process can being conceptualized as being analogous to a grade school math class: the students are told what problem to solve (agenda setting), given the appropriate equation (priming). and told what numbers to put into the equation (framing). The result is an opinion. 5 level of aggregate data. Studies using individuals rather than publics as the objects of study (consider Erbring, Goldenberg, and Miller, 1980; Watt and van den Berg, 1981) show substantially smaller effect sizes, but are still able to demonstrate agenda setting effects, at least for the short term (Watt and van den Berg) or on certain topics (Erbring, et al.). Funkhouser (1973) was among the first to take advantage of the vast body of data represented by the Gallup polls and media content indexes. He discovered a Spearman's rank order correlation of .78 between the number of articles about a given topic and its selection as the "most important problem facing the nation" by Gallup respondents for the entire decade of the 1960's. This correlation suggests a strong relationship but not necessarily agenda setting per se, however, since the time ordering of content vs. opinion was not established. This is of import in that the relation discovered also can be construed as evidence of the so-called "mirror" hypothesis, that the media simply reflect the wants, desires, and attitudes of its consuming public (Bogart, 1969). In response, a wide variety of studies have attempted to not only demonstrate a causal priority of media over public opinion, but additionally, attempted to determine the length of time needed for the impact to occur. Watt and van den Berg (1981) were able to predict the number of resident complaints received by the Federal Aviation Administration concerning Concorde overflights using, among other factors, the amount and net "direction" of media coverage of the problem on the previous day. (The impact of media on this question was negligible six months after onset - people had learned to cope with the noise.) Stone and McCombs (1981) isolated impact lags of from two to six months for new topics to impinge on the public consciousness. Eaton (1989), with access to biweekly data, found 6 that for many subjects the largest impacts were immediate, but that several (fear of war, criticism of the government) showed lags of 4 to 10 weeks. This, combined with Shaw and McCombs' original finding that the largest correlations between media and public Opinion occurred when ALL media sources, and not just those known to be used by the subjects in question, were taken into account, suggests that media content has to "build up", or hold interest long enough to penetrate the sheer density of media content surrounding each member of the public (obviously, the Concorde noise issue failed to build up sufficiently to show long term impact). Both McCombs (1982) and Weaver (1984) have attempted to lay some sort of structure onto the numerous agenda setting studies conducted during the '70's and early '80's by proliferating McCombs' four part classification scheme through the literature. This classification scheme holds that the literature can be organized as a two by two array, shown as Figure 1, with the relevant dimensions being level of data (individual vs. group) and type of dependent measure (single measure or unrelated set vs. interlocked set). Weaver points out that studies using aggregate data (Types 1, interlocked, and III, single measure) show substantially larger agenda setting efi'ects than those based on individual data; and that the studies of Type II (Individual data on an interlocked set of measures) show the weakest effects of all. The study proposed herein will be of Type III, using group data on a single issue. This will permit the advantages of using aggregate data (as outlined above) without requiring the mathematical complexity of an interlocked set of dependent measures. The ability of the media to set the public’ 3 agenda has also been Level of Data Group Individual Interlocked Type I Type II Type of Dependent Measure Single measure Type III Type IV From McCombs (1982). Figure 1 McCombs’ Four Part Typology of Agenda Setting Studies 8 demonstrated experimentally by Iyengar and Kinder (1987). Iyengar and Kinder presented audiences with taped newscasts edited to either contain stories on topics under study, or else substitute,“control” stories. While their method of reporting their results is problematic,2 Iyengar and Kinder demonstrate agenda setting effects consistently across 14 studies, covering a wide variety of topics, and involving a total of 1151 participants. Iyengar and Kinder also used two-stage least squares in a separate analysis to demonstrate a contemporaneous impact of television news on the public’s selection of a “most important problem” in public opinion polls. Rogers and Dearing (1988) traced the development of agenda setting and tied the communications school tradition of examining the impact of the media on public opinion to the political science tradition of considering the interrelationship of media and policy makers (such as government officials). They in turn produced the integrated model reproduced as Figure 2, and also the terminology "media agenda", "public agenda", and "policy agenda", which are operationalized respectively as media content proportions, public opinion, and governmental action. This model is of consequence for two reasons. First, it integrates both research traditions and suggests hypotheses about the influential nature of the components. Second, it provides a mechanism for the intervention of the "real world”, permitting the possibility of measuring and controlling potential spurious variables and therefore correction of spurious relationships. However, it also raises the question of whether these ’ In particular, Iyengar and Kinder are somewhat cavalier with the common practice of indicating statistically significant results with an asterisk. They regularly star items with p<.20, and on one table have starred p<.60! However, their results are significant (by any standard) far more frequently than is explicable by chance, and non-significant results are consistently in the predicted direction, suggesting that the use of cumulative analysis techniques (Hunter, Schmidt, and Jackson; 1982) would support their claims. .0 s o a g :3 '2? (I) a G a < s .D a s i 0 an 5.0 < < ,. e s s 2 o. ociocconomic In - uts Figure2 The Interrelationship of Media, Policy and Pub- lic Agendas, and Socioeconomic Factors Source: Rogers and Dearing (1988) 10 research traditions are related apart from the identity of the dependent variable, Of the interrelationship Of the various agendas. Rogers, Dearing, and Chang (1991) attempted to provide a full test Of all aspects Of the Rogers and Dearing model for the issue of Acquired ImmunO-Deficiency Syndrome (AIDS), as well as adding agendas for two other elite groups, scientists and public opinion pollsters. Unfortunately, the analytic method chosen, time series analysis by the ARIMA method (Box and Jenkins, 1976) permitted neither the ability to test competing causes nor the ability to include public Opinion (which, due to the irregular periods between tests, did not form a regular time series). Given these caveats, Rogers, Dearing and Chang were able to discern four different periods in the life Of this developing story. In the first, the media agenda is driven by the science agenda and the number of AIDS cases reported (it is worth noting that the media agenda in turn drives the reporting Of AIDS cases and not the actual number Of cases). During the second era, the media influence the pollsters while in the third era there is reciprocal feedback between media and pollsters; and finally the media and the scientists drive policy makers to action. It can easily be seen how much could be gained by the inclusion Of public Opinion in this analysis. (The public agenda was stated to have been driven by the media agenda as a result Of the inspection Of a plot Of number of stories and public Opinion against time.) Subsequently, Trumbo (1995) extended Rogers, Dearing, and Chang's techniques to include both a measurement of the public’s agenda (which he termed “extreme concern index” or E01) and path analysis, and applied them to the topic Of global warming. Using a semi-monthly time frame, Trumbo showed a complex interrelationship between the media and policy 11 agendas -- news papers and the science press influence policy, which later influences television content, which itself influences policy about two months later. Both television content and the policy agenda affect the public agenda, the former within the same semi-monthly time period and the latter five months subsequently. Of interest in Trumbo’s results is that the policy agenda’s impact on the public agenda is negative in sign. This is inconsistent with the literature generally, although the length Of the lag suggests that process resulting in this effect may differ from that present in the short-term studies which make up most of the body Of the literature. (The interrelationship, positive in sign, which 'humbo discovers between the policy and media agendas indicates that in the short-term -- one month or less -- there will be a measurable, but spurious, positive relationship between the policy agenda and the public agenda.) Priming and Framing Priming and framing have a much shorter history in the examination Of media effects. Extending the works Of Herbert Simon (1979), Iyengar and Kinder (1979) posited the existence Of priming effects created by news casts and provided substantial experimental evidence in support of that effect across topics including energy, defense, inflation, civil rights, and so on (vide Iyengar and Kinder, 1979, pp. 65-70). Similarly, Krosnick and Brannon (1993) showed that media coverage of the Gulf war influenced the manner in which survey respondents evaluated the Presidential performance Of George Bush. Subsequently, Iyengar ( 1991) published the results Of his extensive studies into the ability Of the media to frame issues. He found framing 12 efl'ects in experimental settings which covered a plethora Of content areas, including crime, poverty, racial inequity, and terrorism. Further, Iyengar and Simon (1993) found evidence of both framing and priming effects in their analysis Of public Opinion data from the Gulf war time period. Pan and Kosicki (1996) seem to have bridged the gap between issue framing by the media and subsequent changes in public Opinion on the topic Of race. Although hypothesized impacts expected to be evoked by “non- information-oriented” media failed to be detected, the use Of information- oriented media by white survey respondents is associated with the belief that the condition Of blacks is due to “not trying hard enough” (frame) and the subsequent rejection of Affirmative Action and spending programs (Opinion). Outside the parameters Of the agenda setting/priming/framing approach, there have been a variety Of longitudinal studies Of the impact Of media, or, as Martin (1976) would point out, the impact Of the "informational role" Of the media. These studies share the general characteristic Of assuming the media are simply carriers Of "real-world" information tO the public, despite the knowledge that the media can, do, and must discriminate from among many stories (c.f. White, 1950; Bagdikian, 1971; Berkowitz, 1990). The most common Of these studies consider the dynamics Of presidential popularity, as a wide variety of pollsters have attempted to gauge the public's evaluation Of a sitting president (Hibbs, 1982; Brody and Page, 1975; Mueller, 1970; MacKuen, 1983); indeed, the Gallup organization has measured the President’s popularity on a monthly basis (with the 13 exception of election months, for fear Of potentially creating a bandwagon , effect) since the Truman administration, presenting a 43-year time series of data. Most Of these studies find a general decline in presidential popularity over time, which Mueller attributes to the alienation Of various minority groups across time. Economic factors such as inflation and unemployment also have predictive power, and it is difiicult to attribute these effects to the media per se; while the populace might have learned of these economic phenomena through the media, the fact remains that they are a part of the make-up of individual life -- increased bills at the checkout line, for instance, or the dificulties Of cousin Charlene in locating a job. Two other, salient, points emerge from this group of studies. One is that the media (or mediated information) do impact the public's Opinion of the president's performance (although not necessarily in a predictable way, as Brody and Page, 1975, point out). The second is that there appears to be two classes Of mediated event: Common, everyday news; and major, dramatic events which Mueller describes as having a “rally-around-the- flag” effect and which MacKuen describes as showing “presidential authority”. The impact Of the former class of event on the public dissipates relatively rapidly (on the order of several months -- see below) while the impact of the "rally" events is large and relatively permanent; MacKuen hold that such events "(touch citizens) in a new way, changing their fundamental view Of each president's character (pg. 184)." In other word, these major events create actual opinion change. (A mechanism for this will be suggested below.) Sufficient evidence exists to suggest that these findings are generalizable tO a degree. In Davis' (1952) seminal study of media on belief 14 change, the amount Of coverage accorded violent crime impacted beliefs about how frequently violent crimes occurred more strongly than the actual crime rate (although it should be noted that this finding was not as strong, or even was reversed, for other forms of crime). More or less simultaneously Davies (1952) isolated short-term reactions in public Opinion to economic and foreign affairs news that were fairly consistent across social and economic class; unfortunately, he did not consider longer term impacts. Again within the short term, Page, Shapiro, and Dempsey (1987) were able to isolate media impacts on a variety of public policy preferences covering "many different kinds Of foreign and defense (n=32) and domestic (n=48) policies (pg. 26)". By restricting their study to short term effects (to merit inclusion, policy time series had to cover no more than three months between data points), Page, et al., (1987 ) uncovered impacts due to statements of the Opposition party, interest groups, media commentary, experts, and the courts; the largest effect was due to media commentary, while that created by interest groups and the courts was actually negative! Mueller (1979) used news content (in this case, the belligerent posturings Of the US. and the USSR.) to predict the public's perceptions Of the likelihood Of the outbreak Of the next world war. The magnitude Of impact was strictly time ordered: belligerence within the last 10 days predicted public perceptions more strongly than news coverage across the previous four months, which in turn impacted public perceptions more strongly than news coverage across the previous year, although all time periods contributed (statistically) significantly to the model. Finally, Smith (1987) attempted an ambitious study of mutual causation between public Opinion on a variety Of topics (seven in all, 15 including crime, the economy, health care and public recreation) and newspaper mentions of those issues. He discovered that newspaper coverage led public opinion on economic development issues whereas public concern about the environment led newspaper coverage; however, problems with the use of Ordinary Least Squares techniques on autoregressive data (specifically, that the assumption Of uncorrelated error terms is violated; see Cook, Dintzer, and Mark, 1980) render Smith's finding problematic. As a summation of these studies, it is reasonable to state that media content can influence public opinion strongly in the short term and weakly in the long term. Data from the presidential evaluation studies suggests that the information creating long term impact is the vivid”, dramatic news while the impact of the day-tO-day content is transitory. Exogenous (that is, external) variables such as economic factors can also have impact, and, indeed, can have larger impact than information. And, finally, Smith's (1987) study suggests that public opinion has the potential to impact media coverage along with possible media impact on the public. The consequences Of these findings for the proposed study will be discussed below. Other agendaopinion relationships The media agenda is expected to affect not only the public agenda but also the policy agenda (Rogers and Dearing, 1988); additionally, the public agenda should influence the policy agenda (which is the function of classical democracy); and the policy agenda and the media agenda should demonstrate reciprocal impacts. 'Although not necessarily vivid in terms Of media depiction. As Iyengar and Kinder (1987) discovered, the public appears willing to “blame the victim” if possible, regardless of the TV production techniques used to communicate the person’s plight. 16 There is almost certainly an impact Of the public agenda on the media, specifically, an economic impact. Because the mass media are profit-making organizations dependent to some degree on mass audiences, and because it has been demonstrated that people will outright reject information which is too divergent from their existing viewpoints (see for instance Hovland, 1959), a pressure is created which forces the media to respond to audience preference, which are, by definition, middle of the road. Gitlin (1980), for instance, points out that the media regularly engage in the trivialization of extremist groups as a result of the media's essentially conservative (in the sense Of "supporting the status quo" rather than "supporting the right wing") nature. However, this “consumer feedback” is generally small, slow to take effect, and difficult to discern; gatekeepers simply keep their perceptions of the desires of their audience in the back Of their mind as they select material, so the resulting content is generally close to the audience’s values. In a theoretically perfect democracy, the impact Of the public agenda on the policy agenda should be absolute: the majority should rule. In the U.S., however, only the legislative branch, the chief executive, and lower levels of the judiciary are directly accessible to the voter, and then only at specified time intervals. As we have seen, however, the public's evaluation of the president occurs continuously and not just once every four years. Thus, while in theory the policy agenda is perfectly and immediately responsive to the public agenda and in practice the executive (as the creator of the policy agenda) is responsive only in quadrennial cycles, the reality is somewhere in between -- policymakers are neither perfectly responsive nor perfectly unresponsive within an election term. Erikson (1976), going back to Gallup data from 1936-1937, discovered 17 strong relationships (point biserial correlations of .97, .64, .63, and .17) between public Opinion and legislative action on capital punishment, the proposed child labor law constitutional amendment, and the use of female jurors. Although a correlation is not proof Of causality, the mechanism for voters influencing policy (by voting) is stronger and surer than the reverse, leading to the conclusion that public opinion had some impact on policy for these issues. Similarly, Page and Shapiro (1983) were able to find reasonably substantial impacts Of public Opinion on public policy on a wide variety of issues, including taxation, military action, and trade. They discovered that during the 1935-1979 epoch, on 231 issues in which public Opinion made a substantial change, public policy made a congruent change 43% Of the time, remained unchanged 33%, and made a non-congruent change only 22% Of the time (during the final 2% of cases a change whose impact was indeterminable was made); in other words, public Opinion preceded policy change in over 3/4 Of cases, these alterations leading to a congruent change almost four times in nine. Finally, Gaziano (1978) showed that there was a relationship between public Opinion and voting on the Supreme Court on First Amendment cases during the late '30s through the 19608, finding that the Court was more likely to uphold free speech rights for radical groups when public Opinion poll results favoring such rights were at or above 40%, and, especially, when public support was rising. While the Supreme Court is appointed for life (and, ergo, not subject to being turned out Of Office by irate constituents), as physical members Of the Washington establishment justices are subject to lobbying and other forms of public pressure. That the media agenda affects the policy agenda also seems well 18 established, although by fewer field studies and a greater degree of anecdotal data. Linsky (1986) reports that 96% Of government Oflicials stated that the media impacted federal policy (whether this is a case of the “third person” effect —"Yes, it sure impacts those other people even though it doesn't matter to me" — has yet to be seen). Cohen (1983) points out that although policymakers desire to some degree to be responsive to public opinion, they prefer idiosyncratic sources, such as an intuitive “feel” for the needs of voters, and including contacts with the media, to Objective reporting Of mass Opinion by polling organizations. In this sense the media function as an elite subsector of the public-at-large. Cook, Tyler, Goetz, Gordon, Protess, Leff, and Molotch (1983) conducted a field study examining the impact Of television news reports (about fi'aud and abuse in the health care/nursing home industry) on the public, on interest group elites, and on local government policy elites (i.e. government Oflicials with responsibilities concerned with health care and the elderly). The government policy elites exposed to the reports perceived significantly (statistical) more fraud and abuse in the health care industry, were more likely to believe that the public perceived the same, and were more likely to see the need for government action, than those who were not. (There was also a significant impact on the public, although not on the interest group elites, whose Opinion can be summarized with the phrases "We knew that already" and "There are other issues involved as well".) In their follow-up, Cook, et al., (1983) point out that the call for legislation on the health care fraud issue made by the Senate Permanent Investigations Subcommittee was the result of an interaction between the government policy makers and journalists directly, sidestepping the Opinions Of the 19 public (Cook, et al., 1983). . The influence of policy makers on the media is less well studied and based to a greater degree on anecdotal data, although Wanta and Foote (1994) recently demonstrated that the President, at least, has the ability to influence the media’s agenda in topic areas on which he is either an important source or has a personal interest. Wanta and Foote accomplished this by using time series analysis of Presidential statements recorded in the fleekGCnmnilatinnnfEmidentialermentsas a predictor ofTV news content as recorded in the Vanderbilt Amhiygs News Abstracts. Significant effects were found for both positive and negative lags, indicating that the President influences media coverage and that stories in the media influenced Presidential statements, respectively. More typical of the examinations of the impact Of policy makers on the media is that of Linsky (1986), who points out that relations between press and president were extremely close at the founding of the republic: both Thomas Jefferson and Alexander Hamilton started newspapers and among Andrew Jackson’s closest advisors were three newspaper editors. Contrariwise, according to Linsky, modern press-government relations are built on the principle that the press and government have an adversarial relationship. Gitlin (1980) reports instances of Lyndon Johnson calling editors Of national newspapers and magazines to discuss (loudly and obscenely) the interrelationship Of their reportage and his administration. Thompson (1973) documents Richard Nixon's disdain for, and attempted manipulations of, the national reporters covering his 1972 reelection campaign, and also points out the dependency Of reporters on government figures using the Eagleton affair as an example. Several reporters, most noticeably Donaldson (1987), remarked on Ronald Reagan's Z) willingness to influence the media agenda by the use of pseudo-news "exclusive" reports to individual members of the media. Weaver and Elliott (1985) extend this by examining the interrelationship between activity of the Bloomington, IN, City Council and local press reports Of that activity. While they were able to demonstrate a distinct gatekeeping effect by the press (238 of 405 total agenda items were eventually reported), they also found that items pertaining to the local economy (and its development — of course an issue of primary concern at the local level) were reported almost entirely and exactly as desired by the council. Weaver and Elliott attributed this to the necessity Of relying on the council members as information sources on those issues; thus, the government policy makers were able to transform the media agenda to their favor. Rogers and Dearing (1988) suggested that while public Opinion impacts policy, the reverse is probably not true. Little published data appeared to treat this issue, although a mechanism by which this reverse link could Operate can be envisioned. As Schlesinger (1939) had pointed out that policy in the US. tends to run in conservative/liberal cycles lasting on the order Of 14-16 years. These cycles tend to push the country in an overall liberal direction due to their nature: during liberal cycles legislation is passed which is ignored rather than repealed during conservative cycles. Thus, at the advent of a new liberal cycle, the new generation sees the previous liberal policy position as "the law Of the land", and the new liberal position pushes toward the left from there.‘ ‘ Although in excess of 50 years Old, this work reasonably predicted rightward turns in the 1948 and 1978 eras as well as a liberal trend circa 1963. 'l\vo further comments follow: the next liberal swing should occur shortly (ca. 1993); and the Reagan administration may have interrupted this process by its wholesale program of deregulation during the early 1980's. 21 Subsequently Sharp (1992) analyzed the impact Of administration and congressional policy on the public agenda and found that it was possible that Presidents Nixon, Carter, Reagan and Bush had been able to influence the salience Of the drug war to the public by their activities. In particular, she shows that President Nixon was able to mobilize public opinion in favor Of his policy of providing methadone to heroin addicts by creating fears of a heroin addiction epidemic. Similarly, a steady increase in congressional hearings on drugs starting in 1981 was followed by an upturn in the percentage of maple naming drugs as the most important problem facing the nation in 1985. Whackgmund Blumer (1946, 1964), via Price (1992), views public Opinion formation as a coalescence process: people identify a problem, interest groups form around it and propose a series of solutions to the problem, the proposals are debated, and policymakers execute one or more Of the solutions and then evaluate the results (see also Foote and Hart, 1953). The stages Of the process (and progress through them) is driven by a variety of social processes, the most important of which are communication - which allows interested parties to mobilize social groups and individuals to their causes — and legislation. Within this process, many writers divide the citizenry into two main groups, the public, which is engaged in the issue, and the mass, which is not.“ It is clear that the composition of the public varies to an extent on an issue-by-issue basis (with a set Of major exceptions which will be discussed ‘ Lippmann adds a third, the spectator, which observes the debate but does not enter into it. below.) When one considers the consequence of this at the level Of an issue of national consequence, it becomes clear is that - via communication through a wide variety Of media — members of the public come together into ad hoc groupings, formal or informal, characterized by some common stand on the issue in question. The stand may be in favor of or Opposed to any or all Of the proposed solutions to the problem; it may also be devoted to raising or lowering the national salience of the problem itself. We will call this ad hoc grouping an “issue networ ”.° Thus, the formation Of an issue is attendant with the formation Of a minimum Of two (pro and con) issue networks, and generally a substantially larger number. It seems likely that issue networks arise originally via interpersonal communication channels. Klapper (1960), Thibault and Kelley ( 1959) and others have remarked on the relative homogeneity of opinion within sociological groups. Within the group would develop, via interpersonal channels, common perceptions of the salience of an issue and the appropriate stand to take on the issue. Regardless of the stand, the salience Of the issue needs to be sufficiently high to motivate the group to action toward implementing social policy consistent with the stand. A leader or set Of leaders who were unable to persuade members Of their sociological groups of the importance Of an issue or the validity Of a solution would probably not be able to cause an issue network to coalesce. But the motivating force for the group can be 'The term issue network was chosen in order to avoid the pejorative connotations of the term "special interest group", which may be the most Obvious synonym. Formally defined, we can describe an issue network as a set of interconnected people who share a perception of the salience of an issue and a stance on that issue. The network generally has a formal structure which includes only a small number of the members, created as members elect each other to offices, award each other titles, and, when money is available, pay salaries to some of the officers. 23 stated simply as "power": the development of sufficient political force to be able to impose the group's solution on the problem. As Baumgartner and Jones (1993) point out, this usually takes the form of attempting to achieve a “policy monopoly”, which would yield to the network the opportunity to establish policy in the absence of effective Opposition. National influence, however, requires some form of national access, and it is impossible in this day and age to form a functioning issue network on an issue of national concern without access to some form of mass mediated communication’, especially bearing in mind the wide variety of such channels available, including news reports (free but uncontrolled by the source), advertisements (under source control but expensive), letters to the editor (free but with no guarantee of publication), and so on. Gaining access to mass media of various kinds moves the issue network from an interpersonal unit to a mediated unit. This in turn requires an organizational structure designed to communicate with the members of the network and to communicate for the network, including a (frequently professional) hierarchy and leadership, and a means of identification for the members (frequently ID cards are issued by issue networks; see Appendix A). As this membership broadens and becomes more diffuse, the pursuit Of power as a goal is joined by the perceived desire for prestige and money, since these will attract members and facilitate the 7 Note that in the issue network it is not necessary for the members to share social space or perception, even though the original group from which the issue network evolved may have. application of power.’3 The media serve the function Of enabling the social identification Of their consumers (Price, 1992). Members Of the mass come to identify with, and frame issues in the terms of mediated models with which they are presented. The ultimate target Of the various issue networks is not the public (which, being composed Of members Of issue networks, has already taken sides), but the mass. It is within the mass that the money lies, and within the mass that the votes lie. So public debate on an issue can be considered as the attempt to frame the issue for the mass audience in the terms of the position advocated by the advocating issue network.“ For our purposes, we will describe framing as the implantation into the minds Of an audience Of an image which is evoked with the evocation Of an issue. Entman (1993) points out that the fimctions of a flame is to define problems, diagnose causes, make moral judgments and suggest remedies; so, a well-selected frame causes the mass audience to associate the moral judgments and/or suggested remedies Of the framing issue network with 'This was very recently - 6 November 1996 - articulated by Libertarian Party Presidential candidate Harry Browne (1996) in a press release to the party (and educators interested in political communication). “...Although we can experiment with short-cuts, the only sure way to [communicate our message] is through the steady building of party membership, which will lead to the money necessary to attract attention, which will lead to the media coverage that will make us part of the national discussion.” Browne’s release, which was distributed electronically, is available as Appendix B. 'This is consistent with Baumgartner and Jones’ (1993) contention that political entrepreneurs attempt to influence policy by means of imposing their own definition on an issue. Clearly this refers to a framing process. % the issue at hand.” In short, agenda setting via framing can be not only a process affecting salience but also one which effects attitudinal responses. Thus, leadership in the issue network can be seen to require a minimum of two roles (although it is possible for the same person or set Of people to occupy both roles): there must be someone to conceptualize the frame appropriate to the network, and someone to articulate it to the public. This can be seen quite clearly in the arena of marketing issues, where an agency may be specifically hired to do the conceptual work, and the agency may in turn specify a particular spokesperson.’L1 This framing takes place largely through the media12 (as the most convenient means of reaching the mass audience), and is another of functions Of communication within the process. The media convey messages from the various issue networks to the audience. (While member units of the media...consortia Of journalists or newspapers...may also participate in the debate and either be members of issue networks or form their own, the message conveyed is no different than that of any other issue network. The media may serve as source, but they do serve as channel.) Access to Media Issue networks can gain access to national media via a number of mechanisms. They can, of course, buy media time if their financial ‘° Entman's definition of framing -- "'lb frame is to select some aspects of a perceived reality and make them more salient in a communication text, in such a way as to promote problem definition, causal interpretation, moral evaluation, and/or treatment recommendation (pg. 52)" -- is not inconsistent with ours but lacks the persuasive element implied by defining framing as a process designed to create an association between image and issue. " This thinking was in part elucidated by the Libertarian Party's Presidential nominating convention, July 4-6, 1996. In the Libertarian Party, the conceptual role is played in large part by founder David Nolan, who is alive and very much on the scene. The public articulation role is left to the Presidential nominee, however. " At this point, this theory overlaps that of Iyengar. % resources are sufficient (and their cause “mainstream” enough), or they can gain media access by staging “pseudoevents” and inviting media coverage. Each of these is problematic: ads are poorly attended (Bechtel, Achelpohl, and Akers, 1972), and the media event supplies the opportunity for the media to reframe issues in their own terms (c.f. Gitlin, 1980). Issue networks can also gain media access by recruiting or developing members of suficient prominence as to be “newsworthy”. (The latter is typified by, for instance, the establishment Of “think tanks” who provide “experts” to comment on news events; c.f. Cooper and Soley, 1990.) Next, we arrive at a notion of mass Opinion. The mass can be conceived as the playing field on which the issue networks vie for supremacy, again, because it is with the mass that the human resource (new members), the financial resource (donations, taxes and fees), and the political resource (votes) reside. (Although Fan, 1988, holds that mass opinion can be accurately reproduced by sophisticated content analysis of media messages, it is the mass, not the media, which possesses the numbers Of people and the financial resources.) Issue networks play their game by attempting to publicly frame the issue in their own terms”, generally in the terms most suited to the national media which act as their channels, i.e. the sound bite, PSA, and infomercial. The issue networks which have the ability to exert greatest influence on the process are those which exert the greatest influence on policy making bodies (via recruitment, financial inducement, or perception "A good example which is otherwise outside the scope of this paper is the abortion issue, debate on which revolves implicitly, although rarely explicitly, around the point at which human life arises. By claiming to be "Pro (human) Life" and by declaring abortion to be murder, it is clear that one side frames the debate in terms of human life starting at the point of conception. Pro-choice advocates have been heard to use the phrase "Potential human life" to describe the fetus, clearly implying that "human" life commences at some point after conception. 27 Of the ability to deliver votes) and those with greatest access to the national media, typically those with direct access to the media because they are members of the media (i.e. in this case serving in their role as source as well as channel). The simple act of framing an issue, Of course, is not necessarily suflicient to causing changes in public opinion; instead, some mechanism must exist which permits this process. Both Zaller (1992) and Iyengar (1991) suggest (or imply) such mechanisms, Zaller taking an approach which is derived from, and expands on, Anderson's (1974) Information Aggregation model, and Iyengar essentially a cultivation-based approach. Zaller modifies the Information Aggregation model using issue involvement; specifically, he feel that a highly involved person seeks out more messages and is more selective in receiving consonant information than a less involved person. Hence, a highly involved person in a mixed message environment differentially seeks out large numbers of consonant messages, thereby (via aggregation) developing a more extreme Opinion (i.e., becomes "polarized"). A less involved person seeks fewer messages and does so less selectively, and thus is more likely to be "mainstreamed" to the prevailing position created by the net flow Of messages. (It should be noted that the competition between polarization and mainstreaming suggests a way of operationally distinguishing between the public and the mass. Further, given that Zaller conceptualizes involvement as continuous rather than dichotomous, he does not make the public/mass distinction common elsewhere.) Iyengar does not specifically refer to the media effects he examines as occurring via a cultivation process, but it is clear that he is stipulating that sort of mechanism. Tapper's (1995) recent reconstruction Of cultivation $ (in order to account for otherwise anomalous research results) states that mediated messages are filtered by consumers via a variety Of processes, including selectivity, chronic affective state, and motives for viewing. Whether cultivation or integration based, each approach has the capability to explain mass media effects as we know them, which can be quickly summarized with the catchphrase “the less you know, the greater the effect.” In cultivation, information-based psychological defenses (e.g. selectivity) go into effect with the attempt to move (or change) closely held attitudes; these defenses act to resist movement and/or to restore a shifted attitude to its previous location. On the other hand, weakly held attitudes not well integrated into the cognitive structure and not held close to the belief “core” are shifted (and shiftable) to a much greater degree. In information integration theories, beliefs build an informational “mass” as messages accumulate, reducing the impact of subsequent messages. In both cases, when considering the audience member who is not strongly involved, message repetition plays a large role in impact. Additionally, it seems reasonable that a message's ability to stand out as dramatic or attention getting (in radio terms, to "punch") maximizes its effectiveness by maximizing its likelihood of being consumed rather than ignored. An effective frame, almost by definition, is dramatic and attracts attention because Of its innate ability to evoke powerful imagery; and it is more likely that a dramatic, evocative image than a mundane image will be repeated simply because of its dramatic nature (Patterson, 1993, points out the concern of the media with "the new, the unusual, and the sensational;” pg. 37). Hence, an efl'ective frame is by its nature better able to impact the members Of the mass via either the mechanism proposed by Zaller or that Of Iyengar. Q This model of cognitive structure also explains the distinction between the short-term impacts of everyday media information as opposed to the long-term effect of the "rally" event. The everyday information impinges only on the outer, regions of the attitude “sphere”; a single attitude is perturbed briefly and then the holder’s psychological defenses cause him or her to draw in new information which draws the perturbed attitude back to its "proper" location. But "rally" events, as pointed out above, penetrate to the central core Of the belief structure and perturb it, which in turn causes a reconfiguration Of the attitude structure around it. Since responses to Opinion questionnaires are derived in part from this attitude structure, the person in question will evince attitude change. Please note that "rally" events are by their nature large, compelling, and dramatic; also note that this large, compelling, and dramatic nature in turn impacts the media coverage they receive. While they may not specifically engender or communicate a new frame, they are covered more thoroughly, with greater amounts Of both visual material and also analytical material than the mundane story that disappears after a day or two on the fiont page. This gives greater opportunity for access to the underlying belief structure, and, under either paradigm, leads to greater Opinion change. Issue Networks and Social Structures Congress is a particularly interesting entity when considering the impact Of the issue network on public opinion, as it should be Obvious that the Congress is not a monolithic entity but instead can be considered a microcosm of the nation in that issue networks arise within it and debate ED publicly on the nature Of a solution.14 Unlike the mass, however, Congress is a small and relatively homogeneous body which by its definition is expected to be engaged in many problems and by its nature is responsive to the types Of inducements offered by an issue network. In short, while influence on Congress by issue networks is similar to that on the mass audience, it proceeds more quickly, and in ways which are more immediately Obvious. Recruitment of members Of Congress by an issue network greatly assists the network's attempt to flame an issue, as the Congress members can use their access to the media to further transmit the frame. This can happen in an Administration or on the Supreme Court as well; in fact, the existence of issue networks on the Court has been formalized in the form Of the issuing Of dissenting opinions. But the relatively small size of the Court (and the general delay required for it to exercise its powers), and the relative homogeneity Of thought within an Administration (which, ultimately, answers to the President) make the activity of the issue network less Obvious. It is worth noting, however, that issue networks can appear, or even be started, in an Administration: Wanta and Foote (1994) show that a President is able to influence the media’s agenda on issues “on which he has a pet interest” (pg. 437), and Sharp (1992) discusses the impact Of President Nixon’s interest in the drug problem on the public’s perception of “This is consistent with Lippmann's (1914/1962) classification of Congress into "routineers" and "inventors." (Note the analogy to "actors" and "spectators".) One can easily point to Senators who were clearly members of, or leaders of, an issue network throughout history: Taft on labor in the '40's, Kefauver on organized crime in the '50's, Hatfield on the Vietnam War in the '60's, Moynahan on welfare and social reform in the '70's, Nunn on defense in the '90's. But these are clearly exceptions to the standard model of a senator, and it is worth noting that none of them is associated with leadership in another issue network. 31 drugs as an important problem. It would seem that the issue network system remains fairly stable across time. Apart from the organizational benefits of the status quo (money, power, and prestige, as discussed above), the fact Of the competing issue network structure suggests that no win is either absolute or permanent. Ins become outs and outs ins, but the networks are in the game for the long run. In addition, to extend the sport metaphor, it is reasonable to assume that an issue network prefers not to be defeated, and thus is frequently willing to redefining its framing to avoid final defeat, in the same manner that strategies shift in team sports under the threat of loss. And defeat need not be permanent nor entail the end of the issue network, as the same people may still pursue the same end while using a different network structure.” Similarly, in an effort to maintain the level of relevance necessary to retain money, prestige, and power, issue networks can also reframe issues in order to designate them as relevant to the issue network.” ' This is reasoning consistent with Baumgartner and Jones’ (1993) contention that there are no equilibria in American politics. As they define the goal Of issue networks (which they call “political entrepreneurs”) as the development of a policy monopoly favoring themselves; a monopoly by definition requires that only one side prevail. Presuming rationality, each side strives to gain the “win” side of the “win-lose” dichotomy, and will act in such a way as to avoid the “lose” side. “As the Christian Coalition grew out of the Moral Majority, or the Reform Party from United We Stand. "Again drawing from the Libertarian convention, Libertarians define the issues of gun control, abortion, and gay marriage as civil liberties issues of the type central to libertarian philosophy (and in fact have similar positions on each of them). Opsrationalization For the purpose Of examining the impact of the issue network on the framing of public Opinion, measurement problems arise immediately. Activities of the issue network’s members who are also members of key social bodies, i.e. policy makers and the media, are a matter of public record. Congress, for instance, is specifically mandated by the Constitution to keep a journal Of its activities, and does so, in the form of the Congressional Record. More importantly, as decisionmakers, the actions Of Congress are considered inherently newsworthy. Similarly, the content- based articles in the periodical literature and major newspapers, and even the evening network news are extensively archived and indexed. But the activities of issue networks outside this social structure are at best — at best — incompletely recorded. Rallies, direct mail campaigns, infomercial and PSA play...all these vary by market and are rarely recorded or indexed; they are in essence only partially recoverable data. Thus, the only places where the issue network's impact can be measured are in the policy and media agendas. At the gross level, then, this again suggests a conceptual model resembling that presented by Rogers and Dearing (1988) and presented in Figure 2. The addition Of external socioeconomic factors, especially the input factors, is based on the assumption that the issues and grievances which engender issue networks have an Objective component, i.e. if there was no unemployment, there would be no unemployment “problem” around which an issue network could coalesce (note that the first step of the process involves identifying the problem; Blumer, 1964). Similarly, as noted above, the prime function Of the issue network is the accumulation and deployment Of political power, the application of which should be 33 demonstrated in subsequent events. Liberal Issues and Liberal Trends Apart from the purely pragmatic question Of the existence of a large body Of already prepared data on the issue, the construct “liberalism”, which overarches a wide variety of specific beliefs and Opinions (Smith, 1990), is of import in two manners. 'Ib return to the nature Of the governing of the United States, it is worth reiterating that many policy actions are in principle based on the notion Of vox populi, vox dei .17 Plainly, if the people speak with a liberal or conservative voice, a consequence Of this speech will exist in the law of the land, specifically, in the shape Of the law of the land. As Vedlitz (1988) points out, that shape is at least partly determined by the underlying tenets Of ideology. He identifies several beliefs basic to conservative political thought, including that the free market is the best progenitor of social and economic development, that the cultural of the United States is essentially a meritocracy, and that federal programs and regulation are at best ineffective and at times outright harmful. Similarly, Smith (1990) isolated liberal beliefs, including preferences for federal regulation and intervention, reform, civil libertarianism, and multinationalism, among others. The consequences Of an electorate choosing a conservative viewpoint espousing, for instance, decreased government regulation and taxation against a liberal one preferring the use of regulation and tax monies as a support to civil liberties are Obvious. The question then follows: 'Ib what extent is this underlying popular belief set manipulated and manipulable? For the extent of manipulation -_'-’The voice of the people is the voice of God. 34 represents control of the nation by elite groups of whatever political flavor. Liberalism The question of liberalism possesses academic import as well in its role as one of the overarching constructs Of American public Opinion in that its action (or lack thereof) constitutes evidence supporting or disconfirming entire classifications of theories of social change. Appelbaum (1970) classified theories Of social change into a four part taxonomy: evolutionary theories, equilibrium theories, conflict theories and “rise and fall” or cyclic theories. Although differing in details, these various classifications make predictions about the trend in liberalism in the United States: an evolutionary theory would predict a drift on one direction or the other; a change in the direction of the drift, from liberal to conservative or vice versa, is not predicted by evolutionary theories and thus suggests that they lack validity. Despite recent electoral results and the discussion revolving around the development Of a vocal religious right in the United States, the best Objective evidence is that the nation is substantially more liberal than it was during the 19503. Thornton, Alwin, and Camburn (1983), for instance, were able to use data from an 18 year panel study Of mothers and daughters to show a growing attitude towards more egalitarian gender roles in homemaking not just for the daughters (born 1962) but for the mothers as well. Mueller (1988) points out that the nation has grown more tolerant of "leftists", i.e. Communists, Socialists, the Russians in general, and so on, based on his analysis of General Social Survey (GSS) data gathered between 1940 and 1985 on items such as "Should a Communist teacher he fired?" 35 (the proportion stating "Yes" decreasing from 89% in 1954 to 51% in 1985). In a similar manner Page and Shapiro (1982) were able to track the policy preferences Of Americans, finding, for instance, that Americans preferred more civil liberties, greater availability of birth control information, and expanded civil rights in the 1970s than they did in the 19308, 19408, and 19508. Unquestionably the most thorough of the studies based on public Opinion data, however, is that Of Smith (1990), which is based on 455 time series collected between 1945 and 1987 which can reasonably be related to an over-arching construct Of "liberalism" subdivided into 17 subcategories (such as civil liberties, labor, foreign afi‘airs, and sex; see Smith, 19903) possessing at least face validity (this was derived using a content-analytic approach to published data; the formal definitional structure appears in Chapter Two). Smith analyzed the 455 time series using Taylor's (1980) hierarchical approach (to determine which of (a) constant, (b) linear and (c) non-linear trends best fits the data, proceeding in that order, so if the assumption of constancy is not rejected, further tests are not performed), and then used the signed average of all trends across all items to produce the data presented graphically in Figure 3. (Although Smith provides points pre-1945, he warns that they are "inadequate"; thus, they have not been reproduced herein.) As can be seen in Figure 3, public Opinion in the US was trending in the conservative direction (as indicated by the signed average less than 0.00) until 1959. As an example, in 1949, for every 1% rise in in a liberally trending item, there was a 1.53% rise in a conservatively trending item, yielding a signed average of -0.53% (liberalism having arbitrarily assigned to the positive end of the axis). Since 1959, public Opinion has trended in the u L PM L + L A 4 121 L 1. 3. 5. E 41 3 at L e I I .21 P I > .1. , 5' ' - , - . ..... , - . - , 1 1 L 1 1 Yfll' Figure 3 Net trends in Liberalism in the United States, 1945-1990 Source: Smith (1990) liberal direction at varying rates. Figure 4 shows the year-by-year changes in the overall curve, or what might be termed the "acceleration". As can be seen, public Opinion has been accelerating in the liberal direction since 1952-3, with only a brief conservative acceleration from 1974-1977. This leads to the question Of whether this construction is "real", or simply an artifact Of the measurement process. Smith concedes that the average conservatively trending single item gains responses about 1.31 percentage points more conservative yearly, while the average liberal item gains 1.32 points annually; the total liberal gain is due to the preponderance Of the number of liberal trends versus conservative trends. In one sense the veridicality Of the liberal trend is unprovable. It is impossible to determine whether a given set of new items conceived of by Gallup or Roper researchers is random or biased toward one end of the spectrum or the other. But at the moment of conception it is not known in what direction the item will trend; it is not until later that it develops that two-thirds Of the trends are towards the liberal end. Further, trends on at least two Of the subcategories are conservative (crime and social welfare) while others are essentially constant (spending and taxes, labor); one would expect a consistent methodological bias to impact these subcategories as well. It is not impossible to conceive of an inconsistent methodological bias - for instance, demand effects on the part Of the polls takers might account for the results — but it seems more likely that the polls are measuring a true change in public Opinion which has indeed been reflected in American society. To run the risk Of using instance confirmation, a recent controversy was raised in west Michigan aroun 1992 about a public Figure 4 Year-By-Year Changes in Net liberalism, 1945-1990 Source: Derived from Smith (1990) $ appearance to be made by former Freedom Rider James Meredith, then a staff member to North Carolina Senator Jesse Helms. Consider these differences versus 1955: In 1955, there are no former Freedom Riders, and there are certainly no black men on the stafl's of conservative southern senators. In fact, as White (1966) points out, a conservative southemer senator such as Helms would have been a member of the Democratic rather than the Republican party. A second viewpoint (Sullivan, Piereson, and Marcus, 1979) holds that such demonstrated increases in liberalism, and particularly intolerance for leftist groups (for instance, see Nunn, Crockett, and Williams, 1978), is an artifact of the measurement on a set of consistent items across time; that while the pollsters continue to ask about the same groups over time, Americans find new groups against which to demonstrate intolerance. They go on to demonstrate that when faced with a menu of extremist groups (including not only communists and socialists, but also the Klu Klux Klan, the John Birch Society, and the Symbionese Liberation Army), the public shows substantial abhorrence; thus the argument follows that Americans are as intolerant as ever, but that the same Old intolerance is directed toward a different set Of groups. Mueller (1988) has demonstrated several methodological flaws with Sullivan, et al.'s, (1979) work, one Of which, the inclusion of a "no Opinion" category in more recent data, provides the capability of simultaneously demonstrating both increased tolerance and intolerance, as both figures have declined since the 19508, when no such "middle ground" response category was accepted. Also related to the question of the veridicality of an increase in liberalism over the last four and a half decades could be Iyengar and Kinder’s (1987) priming effect. Although Iyengar and Kinder apply 40 priming to the evaluation Of presidential popularity, the notion of changing the standards against which activity is judged (priming) could apply equally well to judgments Of political appropriateness as well as political performance. That is, an Observed increase in the performance of liberal activities might occur not because the nation was becoming more self- consciously liberal, but because these activities are no longer perceived as liberal. In one sense this is a distinction without a difference. If the behaviors advocated by classic liberal thought (the basis of Smith’s content analysis) are performed and the positions advocated by classic liberal thought are supported, the result is the same whether these actions and positions are identified as liberal or not! Providing a test Of the role of priming in the process, i.e. whether the Observed changes are a priming efi'ect or not, would be fairly straightforward; participants would be presented with a menu of activities and positions, and asked to identify which of them were liberal. Substantial deviation from the positions and actions advocated by classic liberal thought would be evidence Of priming. Assuming the validity Of Smith's increase in liberalism, another question about the trend curve in Figure 3 is what it has to say about trends in the same items in the recent past, i.e. 1987-1990. Smith (1990) claims that the overall curve demonstrates what he terms a liberal plateau, or a decrease in the acceleration toward liberalism. This interpretation is demonstrated graphically in Figure 5; Smith also selects four items from his set that clearly show this effect. It is not unreasonable, however, to interpret the same data as a perturbed linear trend, illustrated as Figure 6 (actually, a very gentle curve with a large perturbation fits slightly better than a straight line). The Nam 41 the; E A v A in.» 12-- P wi- - v ‘ v ‘ v j V f f '1 v t v t v ' v ' V Figure 5 Liberal 'ITends Demonstrating a “liberal Plateau” New IDNh-bt p - V r VVTr ' r v I ‘ v ' v ' T ' I ' v v Figure 6 Liberal Ttends Demonstrating a Perturbed Linear Trend 43 distinction between trends will be Of consequence in the prediction of the causes of the liberal trend in public Opinion, as it is the locations of points of inflection which make possible stronger predictions from exogenous data; data expected to come on-line from the 1987-1993 polls in a few months (Smith, 1991) should enable the accurate selection of one Of those two models. A final question is whether these changes in liberalism can be simply explained as either artifacts in the shift Of the demographic composition Of the nation or are simply the result of exogenous causes outside the scope of the study (for instance, if all the variance in liberalism can be explained by changes in the education level of the nation, there is little reason to consider the role Of mass media in the process). Prima facie (“first face”) arguments Oppose this possibility; in general, the constituency of the nation has been aging since the 19508 and conventional rhetoric holds that the older audience is generally more conservative. Exter and Barber (1986) demonstrated that while older cohorts have grown more conservative in such areas as attitudes about capital punishment and pornography, they have moved toward the left on government spending, civil rights, and education; indeed, the Oldest cohort Of their study was substantially more liberal than the 30-40 year Old (baby boomer) cohort. Danigelis and Cutler's (1991) basic conclusion was that period effects influence the Older and younger cohorts in similar ways, and that age is an inadequate measure Of essential conservativism (and, by analogy, liberalism); Holsti and Rosenau (1980) showed that occupational status was a fair better predictor of conservatism than age (or, more precisely, historical epoch Of young adulthood, their division of the constituency being based on where the respondent's young adult years stood 44 with regard to World War H, the Korean War, the between-war years, and the Vietnam conflict). Davis (1975) provided a thorough analysis of the public Opinion data demonstrating a increase in tolerance in America, and concluded that 5 percent of the change in tolerance can be attributed to age cohort to which participants belonged, 4 percent to educational advancement between age cohorts, and 13 percent to increasing tolerance that crosses age cohort and educational group boundaries, i.e. for the purposes of this proposed study, constitutes a true liberal shift. Following Smith’s (1990) summary, Davis (1992) returned to the demographic analysis of the liberal shift. Although be located conservative shifts within cohorts for the late 1970’s followed by a 1980’s rebound, he felt that the shift itself was created by cohort replacement. Thus, it can be accepted that the United States has grown more liberal since the 19508. Such demographic analyses leave open the “why?” question. That Older age cohorts should be more conservative is suggested by information aggregation theories (Anderson, 1974): Older people have received more messages and thus have greater inertia to overcome. Why more recent cohorts should be more liberal than the cohorts they replace is unknown. In short, demographic analysis simply moves the question Of liberal shift to a difl'erent but functionally identical level. Bias in the media Any discussion of the interrelationship between the mass media and the liberalism Of the United States must include consideration Of the aft- 8tated position that the mass media possess a liberal bias (c.f. Efron, 1971; Keely, 1971, Agnew, 1971); ergo, if the nation has come to move toward the 45 political left, it is as a result of the deliberate machinations of a left-leaning industry. Fears in this regard are not completely unfounded; journalism as a whole is populated by people who are substantially more liberal than average (Schneider and Lewis, 1985) who are forced by the nature Of their industry tO "gatekeep" or select specific news items from the universe of news items (for instance, see White, 1950; Bagdikian, 1971). In fact, in the classic White (1950) study that led to the creation Of the gatekeeper metaphor, the editor in question specifically did reject certain stories based on ideology — terming an important human rights story of the day as "propaganda" and rejecting another as "too red". Similarly, liberal critics (such as Gitlin, 1980; Cooper and Soley, 1990) find a distinct conservative bias to the media. James (1988), faced with similar inconsistencies, suggests that this contradiction lies in the method of journalistic (rather than scientific) argumentation: a journalist states a thesis and then presents examples which support the thesis. This "reasoning from example" seems to be a part Of the "natural logic" of the human mind (Mineo, 1991), but must be suspect given the demonstrated tendency Of human beings to prefer material supportive Of their viewpoints. The genesis of this dichotomy may also lie in the nature of the mass media industry: While the bulk of media employees are liberal, those at the top Of the pyramid of “power” are conservative, being members of society’s elite. This may explain the Observed contradiction that while every day media content — prepared by those lower on the social rungs — is somewhat liberal, editorials - over which media employers exert more direct control, tend to more Often support conservative positions. Stevenson and Greene (1980) presented students with positive and 46 negative articles about the 1976 presidential candidates and counted the number of times the students stopped to counterargue (see Carter, Ruggels, Jackson, and Heffner, 1973). The students stopped more substantially more frequently in material they perceived as biased, implying that bias is itself a subjective phenomenon. To cope with the problem Of subjectivity in estimates of media bias, researchers have moved to the "objective" techniques of content analysis (thereby attempting to replace subjectivity with intersubjectivity). A8 a general rule, these studies demonstrate an extremely small but fairly consistent liberal bias. For instance, Russo (197 2) concluded that NBC's and CBS's coverage Of the conflict in Vietnam during the 1969- 1970 period was biased against the administration by an average Of about one-half of one point on a ten point scale. Stevenson, Eisinger, Feinberg, and Kotok (197 3) concluded that the favorability of coverage Of Nixon was roughly equal to that accorded Humphrey, although Humphrey did receive about 10 percent more coverage. Lawry's (1974) major findings included the discovery of a statistically significant number Of anti-Republican to anti-Democratic statements in the networks during the 1972 campaign — a difference Of 5 percent - and that ABC used more film footage Of McGovern. Robinson (1985) showed that Mondale and Ferraro received about 26,000 seconds of network time versus 25,000 for Reagan and Bush, and that only 2.5 percent of the statements made contained any issue-based bias; Of these 21 biased statements, 11 were liberal and 10 conservative. And finally, Friedman, Mertz, and DiMatteO (1980) had students code the expressions on the faces Of five prominent newscasters as they said "Gerald Ford" or "Jimmy Carter" (with the audio track out so that the students would not 47 know what was being said). They found a statistically significant difference of 0.21 points favoring Carter, measured on a 21 point scale. Shoemaker and Reese (1991) cite instances both of journalists who permit their biases to appear in their reporting, and those who work so assiduously to prevent such fiom happening that they actually overcompensate and lend support to viewpoints Opposite their own. Shoemaker and Reese found the results so ambiguous that they were unable to draw firm conclusions from them, except to say that some journalists show some bias at some times. A second approach to the measurement Of liberal bias in the media would be to take advantage of the notion that it is a subjective phenomenon and create intersubjectivity by summing across a representative sample of subjective responses in the manner similar to that of the Delphi technique (c.f. Jones and Twiss, 1978). For instance, ABC Survey #0121 (1984) found that 17 percent of those polled felt that TV news favored the Democrats and 11 percent felt it favored the Republicans; similarly, Schneider and Lewis (1985) report that 30 percent felt that the news media generally possessed a liberal bias whereas 13 percent felt they were conservative. Converting these figures to z-scores and then taking their midpoints yields the estimate that the media are in the range of .15 to .30 standard deviations more liberal than the population average. Both methods yield the same result: the news media have a small but measurable liberal bias. Because of this, and because it is necessary to at least consider the prevailing attitude of media liberalism, in the present study it will be necessary to measure this bias across time and control for and measure its impact. 48 PotmtialDrawbacksinflreUssofPublicOpinionData Public Opinion polling has drawn criticism concerning its scientific value from two directions. The first, summarized by Blumer (1948), is that the measurement Of public opinion by the random selection Of individuals and application of inferential statistics, provides an inadequate measure Of true public opinion because it proceeds from an atheoretic position. The second, derived from Katz and Lazarsfeld (1955), is that public Opinion polling obscures the interaction of power groups and Opinion leaders that lead to the formation Of that opinion; in other words, social structure is ignored. The purpose Of the present study is, in part, to address this second concern, by identifying three Of the roots Of public opinion, the policy agenda, the media agenda, and conditions in the outside “real world” and measuring their influence; and while public Opinion continues to be gathered atheoretically, the measurement process will permit post hoc theoretical grouping by the process of content analysis (Krippendorf, 1980). Another criticism of public Opinion as it is currently measured is that it acts to reduce the participation of the individual in the political process. As Herbst (1991) has pointed out, earlier in the history of the nation, in the 18008, for instance, public Opinion was frequently expressed in unstructured but public ways, i.e. strikes, petitions”, riots, and even the Whiskey Rebellion and the Fenian invasion Of Canada! Public Opinion polling, on the other hand, is both structured, i.e. the respondent is limited in the choices available both in terms Of questions and responses, and private, meaning that the respondent has no Opportunity for contact (or discussion) with others in making his or her response. Herbst argues that " The right to petition the government for redress of grievance was specifically reserved to the people in the First Amendment to the Constitution. 49 the shift to private and structured measurements of public Opinion weakens the democratic process by discouraging political discussion and by encouraging “bandwagon” effects, whereby people give responses which they they perceive as popular rather than “true” responses. The salience of this to the proposed study is that it suggests that changes in public Opinion . might be a result of changes in fashion rather than true changes in public Opinion.” Kaplowitz, Fink, D’Alessio, and Armstrong ( 1983) showed that bandwagon effects did occur on topics with which the respondents were not personally involved, but did not with subjects they were involved in. Further, as Luttbeg and Gant (1983) point out, while the electorate describes itself as being slightly conservative, few members of the populace are able to Offer a coherent definition of what that actually means. Thus, the liberal trend discussed above could be the result Of measurements made on topics the populace was not involved with combined with respondents picking response categories that sound popular (or “politically correct”, such as, “Would you vote for a __ (any minority group) for president?” Since no minorities have been major party candidates for president, the topic is uninvolving, and it is a simple thing to give the “popular”, and liberal, response “Yes”). The solution to this is to consider changes in behavior as well. Without getting into the lengthy literature on the attitude-behavior "The validity of Herbst’s conclusion regarding the general trends in the expression of public opinion may be inadequate. While there was no public opinion polling in the 18008, the fact is that public, unstructured expressions Of public opinion exist to this day. In the week this was originally written, in 1992, UAW members sponsored a “car bash” where people paid for the privilege of hitting a 'Ibyota with a sledge hammer; unidentified men were printing “Buy America” stickers and surreptitiously plastering them on imported automobiles in 'lbxas, and anti-abortionists were praying outside the house of a Kansas- based abortionist. It may simply be that more avenues are open for the expression of public oprnron. 50 relationship, it is reasonable to expect that a “real” change in public Opinion will be followed by a change in behavior of the same type, i.e. the expressed willingness to vote for minorities will be demonstrated by the increased election of minorities. If the change in Opinion is merely the result Of “bandwagon” effects, behavioral changes will not follow. Expectations Examining the role Of the mass media concerning the demonstrated liberal swing within the United States requires long term analysis Of four major areas: the media agenda, as potential causative agent; public opinion; the policy agenda, acting as both a potential change agent on its own and also as a mediating body between the media and public Opinion; and a variety Of external socioeconomic factors such as inflation or unemployment, some Of which might confound results and thus must be controlled, and some of which represent the consequences Of opinion change and thus must be measured to establish the veridicality of that change. As a result of the research reviewed above, expectations of the results Of the present study are: 1) The media will show a direct impact on public liberalism; 2) The policy agenda will show a direct impact on public liberalism; 3) The media will show a direct impact on the policy 51 agenda; 4) The increase in public liberalism will be followed by an increase in liberal activity in the nation. 5) Separate models Of the formulation Of public opinion will vary fi-0m issue to issue. All of these expectations were derived from the preceding discussion. The first four are explicitly mentioned above and hopefully need not be restated here. The fifth expectation is derived from the nature Of the expected issue network structure: because Of the nature of the coalescence process, issue networks are formed Of different people who necessarily make different “tactical” decisions in attempting to communicate their frame to the mass audience. Some issues, as we have seen (vide Sharp’s, 1992, discussion of the drug war), are driven from the “top” down, by policymakers. Others are driven from the bottom up, by the concerns of citizens who belong to issue networks (as the Americans with Disabilities Act seems to have been). Still others are driven overtly by the media’s agenda, as was the case with Watergate. In short, different issue networks with different members and different access to media and policymakers coalesce around different issues should yield different models of the interrelationship of public Opinion, policy agenda, media agenda, and external factors. Chapter 2 Measurement and. Analysis As seen in Figure 2 (Theoretical Model), this study examines five major variable “clusters” (or conceptual constructs): public Opinion; the agenda Of the media; the policy agenda; and two sets Of socioeconomic factors, the inputs and the outputs. Each of these variable clusters must be defined and measured, as discussed below; in addition, an analytical strategy is defined. Measurement of Public Opinion As discussed in the previous chapter, Smith (1990) assembled public Opinion data on a total of 455 items, each of which had been asked of the public at least twice, making possible the demonstration Of a net liberal trend which was presented in Figure 1. Table 1 contains the complete list of 16 content-based sub-categories constructed by Smith, along with the number Of trends which compose that sub-category (The final column, Interrupted?, will be discussed later in this section). In classifying responses as “liberal”, Smith reports using the following content rules: “. . . Contemporary liberalism is (1) reformist, Opting for change and generally Opposed to the status quo; (2) democratic, favoring a full extension of electoral rights; (3) libertarian, supporting civil liberties such as free speech and the right to protest; (4) regulatory and interventionist, backing the management of business and the economy by the government; 52 Factor Name 1 2 3 4 5 6 7 8 9 10 11 12 13 14 ESE-I 53 Table 1 Measurement Of the Factors of Liberalism Abortion Civil Liberties (CL) Crime (CR) Economic Regulations (ER) Family Feminism (FE) Foreign Affairs and Military (FA) Labor (LA) Lifestyles Miscellaneous Politics Race and Ethnicity (RE) Religion Sex Social Welfare (SW) Spending and Taxes (ST) Number of 'Itends 10 85398 03 H 8838888353428 Interrupted? yes yes yes yes no yes . yes yes 54 (5) centralist, using the federal government to set and enforce national standards and regulate state and local governments; (6) humanitarian, favoring a social welfare system for the care and protection of society in general and the lower class in particular; (7) egalitarian, advocating equal treatment for all and perhaps equal conditions for all; and (8) permissive, tolerating and Often approving of non-traditional lifestyles and practices . . . Internationally, contemporary liberalism is (1) internationalist, supporting active U.S. involvement in the world; (2) multinational, backing the UN and other collective efforts; (3) nonmilitarist, preferring nonmilitary solutions to international disputes; and (4) prodétente, advocating good relations with Communist nations and not emphasizing anti- Communism as a cornerstone of foreign policy (pg. 481).”1 Using this list as an operational definition, Smith was able to content analyze responses to poll items which shared three characteristics: the item had been asked at least twice; the wording Of the question had remained roughly comparable across samplings; and the content of the item had fallen into at least one of the fourteen content areas. He then scaled the items to yield the data in Figure 1 by summing the annual proportional change in each item across all items, with changes in the liberal direction being (arbitrarily) coded as positive. Results for items which were measured more than once during a given year were averaged; if a period Of more than one and less than six years elapsed between repetitions of an item, the results were approximated by linear interpolation;: if the elapsed time between repetitions of the item was six ' This list is inserted here in toto as it will be referred to in other sections of this chapter. “ Smith (1995) states that assumptions of non-linearity between measurements were tested but did not make much of a difference. 55 years or more the item was excluded fiom the analysis of that year’s data. (Including these items leads to a long term “smoothing” of the dependent measure.) The present study used an updated list of responses to 476 opinion items, available from the NORC, which extended the original data set by including additional items and polling results through 1994. Items were sorted and scaled as above; however, rather than use the cumulative (summed) figure as Smith did, the annual average net trend derived from each year’s data was used instead. Mathematically speaking, using the cumulated scores leads to a situation where the magnitude of the kth score is dominated by the first through (k-1)th scores, obscuring the annual “shocks” created by other variables. Using annual scores reduces the impact of the previous items in the series to an “inertia” or “carry-over” quantity, if any, and makes the impact of other variables less obscure. This method of scaling “liberalism” was retained as it possesses two advantages. The first advantage of continuing to use Smith’s scale is that, based on this definitional structure, it both reflects previous literature and possesses face validity, which are, of course, two of the expectations of a useful Operational definition (Miller and Nicholson, 197 6). Additionally, the scale possesses the additional advantage of being horizontally dense, that is, continuously measuring the construct in question across time. As seen in Table 1, while certain Of the sub-categories are interrupted (that is, there are periods where none Of the items composing that sub-category are being presented to the public), as far as the overall scale at any given time is concerned a minimum of 50 different trend lines are being measured. For the purposes of this study, seven of the 16 sub-categories have 56 been eliminated, for a variety Of reasons. The sub-categories of Abortion, Lifestyles, and Sex possessed items referring to actions whose legal status changed during the course of the study (for example, abortion, homosexuality, certain types of drug use, and the wearing of shorts in public by women). The legal status Of such items has the clear potential to bias survey results. “Miscellaneous” does not constitute a coherent classification enabling interpretation. Finally, Family, Politics, and Religion were excluded on the basis of being less than central to the differences between liberal and conservative thought; that is, while modern conservative thought stresses religious or family values, this is a fairly recent stress; and the liberal is certainly not Opposed to the family or God. (The existence of these content categories indicates that the Liberal- Conservative dimension is not necessarily unidimensional; note that Smith’s definition does not include a “religiosity” factor, for instance.) Along with the use Of “liberalism” as a single dependent measure, two of the sub-categories will be selected for more in-depth examination. As noted above, while the attitude of the public at large has become more liberal generally, this is not true across all Of the content sub-categories; specifically, the public responds more conservatively to questions concerning Social Welfare, Crime, Spending and Taxation, and also to a sub-set of the Foreign Affairs and Military items, those dealing with Military Recruitment (Smith, 1990). This is, in a sense, counter-intuitive: if the public is becoming more liberal, then it should follow that the public should be becoming consistently more liberal rather than inconsistently more liberal; in other words, whatever is causing the majority of the scales to trend towards liberalism should in principle cause all of the scales to trend towards liberalism. Selecting a subset of the subscales for closer 57 examination might have the possibility Of revealing patterns of influence which might otherwise be obscured by the magnitude of the overall trend. It will also permit explication and testing Of the fifth expectation. In considering the selection of sub-scales, it is preferred that the sub- set meet certain criteria: they should possess a wide variety of internal, overall trends (in order to gain the advantage laid out in the previous paragraph); they should take advantage of as many individual trendlines as possible (to minimize the impact of aberrant data and sampling error); and they should be continuous, i.e. have greater than 0 active trends, in order to make eflicient use of all of the data on predictor variables. Satisfaction of these criteria leads to the selection of Race and Ethnicity and Foreign Affairs for specific examination (see, again, Table 1). Race and Ethnicity is the sub-scale with the largest number of items as well as the largest minimum number of active trends; it also has the second strongest net liberal trend (after Abortion, which, as above, was excluded f'rom the analysis due to the confounding effect of its legal status). Foreign Afl'airs is the only other content area in which data was continuously gathered, and is of particular interest for the reason brought out above: opinion has become more conservative on the question of military service. Thus, liberalism was measured, on a yearly basis, as the overall net trend for all indicators for that year. Additionally, sub-scales for Race and Ethnicity, and Foreign Affairs were measured and attempts made to predict their behavior as well as that of the whole. Table 2 contains the number Of active trends on a yearly basis broken out by content area, as well as the gross change in public opinion in that area during that year. (The Annual Net 'ltend was determined by summing the gross across the content areas and then the active trends; finally, the latter was divided into 58 Table 2 Gross Changes in Public Agenda and Number Of Active Trends By year and Content Area Year CL CR ER FA FE LA N Gross N Gross N Gross N Gross N Gross N Gross 1941 0 0 3 -.046 0 1 .026 0 1942 1 .047 0 3 -.046 0 1 .026 2 .028 1943 1 .047 0 4 -.096 1 .045 1 .026 2 .028 1944 5 .053 0 4 -.049 1 -.036 l .026 3 .151 1945 5 .053 0 4 -.015 1 .009 1 .026 3 -.025 1946 4 .028 0 5 .034 2 .095 1 .042 1 .008 1947 4 -. 159 0 6 . 114 4 -.239 1 .042 5 .073 1948 4 -.059 0 6 -. 102 3 .050 1 .042 6 -.204 1949 5 -.298 0 6 -.053 3 .077 1 .042 7 .016 1%0 5 -. 141 0 6 -. 117 3 -.013 0 2 .008 1%1 2 -.027 0 7 -. 150 5 . 119 0 2 .008 1%2 2 -.027 0 6 -.117 6 .320 0 2 .008 1%3 2 -.027 0 6 -. 117 6 .141 0 1 .046 1%4 5 -.03 1 1 .035 5 .069 6 .246 O 4 .071 1%5 3 .033 3 .043 5 .069 7 .256 0 4 .071 1%6 2 -.025 3 .043 0 7 .035 1 .007 3 .037 1%7 2 -.020 3 .067 1 .003 7 .070 1 .007 3 .037 1%8 0 2 -.054 l .003 7 .122 1 .007 3 -.096 1%9 0 1 -.017 1 -.020 7 .093 1 .032 3 .042 1%) 0 2 -.010 4 .035 7 -.026 l -.004 5 -.016 1%1 O 2 .022 3 .055 3 -.001 1 -.004 5 -.016 1962 0 2 .022 6 .012 3 -.092 1 -.004 6 .005 1%3 0 2 .022 6 -.027 4 -.027 1 -.004 5 .081 1%4 0 2 -.021 5 -. 107 5 -. 103 1 .004 5 .046 1%5 0 2 -.021 5 -.037 5 . 118 1 .004 6 .101 1%6 0 6 -.081 10 -.102 7 .096 2 .002 10 .040 1%? 0 5 -.131 10 .155 7 .136 2 .002 9 -.165 1%8 0 6 .009 9 .178 7 .098 2 -.017 8 - . 115 19a 3 -.043 16 -.170 12 .473 9 .211 2 -.017 6 -.042 1970 4 .025 16 -.050 11 .379 9 .066 3 . 111 6 -.005 1971 4 .129 17 .046 12 .037 9 .283 5 .124 7 .036 1972 4 .129 18 -.036 12 .306 7 .021 5 .107 7 .036 1973 14 .337 20 -. 157 12 .249 9 -.002 6 . 113 4 .043 1974 17 -.009 Z) -.078 12 .386 11 -. 154 9 . 110 3 .035 1975 12 -.028 a) -.153 16 .177 12 -.253 12 .044 5 -.052 1976 12 -.037 21 -.251 15 -.033 12 -.343 14 -.066 8 -.147 1977 17 -. 197 Z) -.082 17 -.173 11 .053 15 .001 7 -.131 1978 18 .380 Z) -.015 20 .056 12 -.009 17 .298 8 -.060 1979 18 .046 19 -.002 21 .184 11 -.091 18 -.062 9 .073 m 17 .035 19 -.074 18 -.250 10 -.280 17 .058 9 -.208 1%1 17 -. 118 18 .091 16 -.040 9 . 179 m .223 8 -. 165 1%2 17 -. 118 18 -.206 10 -. 105 8 .241 18 .499 4 .105 1%3 17 127 17 -. 114 9 -.045 9 065 18 252 3 .023 17 5555555555 CL N Gross .184 -.313 .188 .208 -.047 .477 .047 .265 .066 .066 .051 55555555555 CR N Gross .540 -. 181 -. 140 .010 -.061 .176 -.056 .628 -.207 -. 192 -.020 59 Table 2 (Cont’d) ER FA N Gross N Gross 4 -.092 9 .038 4 -. 112 8 .092 3 -. 187 7 .050 2 .190 6 .064 0 6 .122 0 5 .174 0 5 .150 0 5 .316 0 5 -.237 0 5 .120 0 4 -. 124 N Gross Ocoecococoti'5‘52i55 FE .016 .137 .152 .191 .246 .035 . 154 . 106 .146 .094 .076 N 3 3 1 0 0 0 0 0 0 0 0 LA Gross .008 6) Table 2 (Cont’d) Year RE ST SW N Gross N Gross N Gross 1941 2 -.027 0 0 1942 1 -.015 0 0 1943 2 -.050 0 0 1944 5 -.050 0 0 1945 3 .060 0 0 1946 5 .151 0 0 1947 4 .196 0 1 .010 1948 2 .106 1 -.002 1 .000 1949 3 .108 1 .151 0 1%0 7 .436 1 -. 112 0 1%1 7 .160 1 -.014 0 1952 7 .055 1 -.190 0 1%3 7 .142 1 . 114 0 1%4 6 .089 1 .015 0 1955 4 .020 1 .015 0 1956 4 .061 1 .015 1 .020 1957 5 -.173 1 -.060 4 .042 1%8 7 .136 1 .049 4 -.001 1%9 11 .302 1 .049 4 .003 1960 10 .085 1 .027 4 .003 1961 9 .093 1 .027 1 -.059 1962 7 .030 1 -.015 1 -.011 1963 7 .030 1 -.052 0 1%4 24 .333 1 -.035 0 1965 3) .594 1 .021 5 -.083 1966 33 .274 1 .021 5 -.023 1967 I5 -.172 1 -.061 5 -.023 1968 32 .551 1 -.055 5 -.002 1969 3'3 .758 1 -.055 3 .108 1970 33 .684 7 .013 4 .148 1971 34 .759 8 .023 5 .188 1972 34 .600 18 -.083 5 .163 1973 31 .235 E -.098 6 .051 1974 a) .292 % -.210 6 .132 1975 so -.045 % -.322 7 .006 1976 ii) .227 m -.214 7 -.078 1977 29 .181 14 -.341 7 -.014 1978 33 .338 14 -.014 7 -.199 1979 a .173 14 -.241 7 .031 1%) 5 .184 14 -. 110 7 -.082 131 fi .223 15 .380 6 -.072 1%2 5 .201 16 .454 4 -.027 1%3 5 .080 17 .149 4 -.025 134 23 .316 17 .410 6 -.019 61 Table 2 (Cont’d) Year RE ST S W N Gross N Gross N Gross 1%5 22 -.069 29 .122 7 .023 1% 21 .152 29 -.072 7 .023 187 22 .163 27 .309 7 -.050 1%8 2 . 167 27 . 199 7 .031 1%9 22 .300 w .385 7 .127 1990 23 .012 % .223 7 . 113 1991 22 .216 m -.685 6 -.008 1992 21 .078 m .319 6 -.079 19% 18 .109 25 .046 4 -.072 1994 15 -.013 % -.480 4 -. 155 the former.) Determining the Media Agenda The media agenda is perhaps the most difficult factor to reconstruct historically for two reasons. First, many media, especially the electronic media, are essentially ephemeral in nature; unlike the books in a library, the broadcast signal waits for no man and it is fi'equently neither easy nor desirable to store it. Unfortunately, people access broadcast media far more frequently than they do print media (Bower, 1985). Second, the sheer mass of data makes the measurement and evaluation of the content of any medium problematic. A one month archive of even a small daily paper creates a stack of newsprint in excess of a foot high, and a 45 year archive fills two or more microform drawers. Fortunately. the Beaden’e Guide in Periodical Literature has been demonstratedly effective to the researcher interested in approximating the historical media diet (c.f. Funkhouser, 1974; D’Alessio, 1992). It attempts to index and classify by content a wide variety of magazine articles. The scope of the contents of the Bender-Le Guide is deliberately exhaustive rather than restrictive; rather than representing a sample, the Reader-Le Guide attempts to represent a population. The potential problem with using the Readefie Guide as a source is that articles from all sources are counted equally and not weighted by circulation figures. However, some of the data presented in Chapter One suggest that the media agenda functions as a group or cluster and not as a series Of individual stories (for instance, in that media impacts build up over a period of months); thus, there is a degree of correspondence between the number of articles appearing on a topic and both its importance and potential impact - “bigger” stories attract 63 more attention, and thus a larger number of articles. Although the Beadefie Guide only directly represents the contents of periodicals and not those of newspapers or television, the time and space constraints of the news media (Friendly, 1975; Bagdikian, 1971) ensure a fairly large degree of overlap across the content of all three media. Testing the impact of the media involves the evaluation Of persuasive trends within the media. This was accomplished by coding a sample from the articles listed in the W Guide as relevant relevant to each of the content areas; a random selection of the articles was recovered from archival files Of the the medium in question. The exact number of articles in each yearly sample was calculated to yield results of 3:25 percent accuracy. The exact list of content areas articles were selected from appears as Table 3. Coding was done by at least two judges for each article. A coding instrument derived directly from the operational definition of “liberalism” presented earlier was constructed (it is attached as Appendix C); coders were asked to read the each article looking for statements which were overtly biased along any of the lines discussed by Smith and listed earlier, and to make a simple raw count of statements biased toward either side. The number of liberally-biased statements (L’s) and conservatively-biased statements (C’s) were separately summed and a liberalism score for the article was constructed by subtracting the number of C’s from the number of L’ s . Coder training consisted Of a lengthy (ca. one-half hour) discussion of each of the bias areas, including several historical examples of each, at least one from each side. Coders were then presented with an article containing a known number of biased statements on each side; they coded 64 Table 3 Topic Headings used in Readefie Guide Civil Liberties -- Civil rights -- Due process of law Crime -- Crime and criminals -- Crime and criminals, U.S. - Assault -- Sex crimes Economic Regulation - Government Regulation of Industry -- Prices - Regulation -- Prices - Regulation - US Feminism -- Feminism -- Woman - equal rights -- Woman - employment Foreign Affairs -- United States - Foreign Relations -- United Nations Labor -- Strikes -- Trade Unions -- Labor Unions Race and Ethnicity -- Race Prejudice -- Race Discrimination - Negroes (Blacks after 1971) -- Negroes in the US - Segregation -- Free speech -- Privacy -- Burglary and burglers -- Murder -- Rape -- Industry and State -- Wage-Price Policy -- Equal pay for equal work -- Women’s liberation movement -- Woman - Occupations -- Conscription - US -- Military Service, Compulsory -- Strikes - US -- Trade Unions - US -- Labor Unions - US -- Race Problems -- Civil Rights Demonstrations -- Race Riots -- Discrimination in Employment -- Negroes in the US - Resistance to Segregation -- Negroes in the US - Civil Rights Social Welfare -- Poor Relief -- Social Welfare -- Economic Assistance, Domestic Spending and Taxation -- Taxation -- Government Appropriations - Income Tax - US -- Public Welfare -- Public Welfare - US -- Taxation - US -- Budget - Income Tax -- United States - Appropriations and Expenditures 65 it, and the results were discussed. This process was repeated on a second sample article. Coders were cautioned to be careful not to associate either liberalism or conservatism a priori to either major political party, and also to be careful to stick to the historical criteria and not diverge into recent debates (about, for instance, media bias, which appears to be more a function Of the party in power and not necessarily a liberal-conservative issue; Page and Shapiro, 1992). The performance of the coders was reviewed periodically, and questions answered as they arose. Coders were paid $5 per hour at the beginning of the study; later this was raised to $6.50. Conflicts between scores assigned by coders were handled in one of two ways. If the difference between the scores was three points or less, the two scores were averaged. If the difference was greater than three points, the conflict was resolved by having the article coded by a third coder. (The “three points” rule was arrived at by the decision that it was acceptable for coders to disagree on the valence of one statement and for one Of the two to see one more statement as overtly biased.) Table 4 contains the mean net bias scores, broken down by year and content area. Measuring the Policy Agenda In everyday circumstances, the government exerts day-to-day influence on the population through the mechanism Of legislation. As Schlesinger (1939) noted, this can include legislation designed specifically to move the country in a specific societal direction; for instance, to move the country in a liberal direction by making, for instance, racial or sexual equality the “law of the land”. This law has essentially two sources: the Congress, and the President and sitting administration, which can publicly request or oppose Year CL 1.3 1.5 1.5 2.8 2.0 3.9 2.1 6.9 1.3 1.1 0.9 1.5 0.0 0.9 2.9 1.0 1.8 2.1 -0. 1 2.1 0.2 2.6 1.3 1.3 1.8 0.3 2.1 3.6 5.3 6.7 2.4 1.8 2.3 2.2 2.3 1.7 1.0 0.8 2.8 0.6 1.3 2.0 2.1 as Table 4 Mean Media Bias By year and Content Area CR ER FA FE LA 0.6 0.7 2.0 2.8 -0.3 -0.6 0.5 2.5 1.4 0.4 0.4 -1.4 3.1 1.6 0.2 1.5 0.5 1.2 2.4 2.0 -0.5 0.1 -0. 1 1.8 0.8 0.1 0.1 0.5 2.1 0.6 0.3 2.0 1.6 2.8 0.0 0.4 0.6 -7. 1 0.0 0.2 0.5 -1.0 0.7 -0.4 -1.2 0.1 0.0 -0.2 -1.0 0.6 0.3 0.1 -0.6 0.9 0.5 0.3 -l.l 0.0 1.8 0.1 -0.3 0.4 2.5 1.2 0.2 0.4 0.5 1.6 1.5 0.9 0.3 1.0 1.5 2.0 0.8 0.0 -2.3 0.4 -0.3 0.3 0.9 0.0 -0.9 1.0 0.2 0.3 0.6 -1.3 2.6 0.8 0.0 -1.8 0.2 0.6 0.4 0.1 0.8 0.3 2.1 0.3 -0. 1 -0.4 1.5 1.3 -0.2 0.2 0.9 1.6 1.9 0.8 0.1 -1.7 0.3 1.6 -0.5 -1. 1 -0.9 3.0 0.1 0.7 -1.3 -0.6 0.7 6.8 0.9 0.1 0.6 2.3 3.8 3.3 0.1 1.3 1.6 2.7 0.4 -1.5 2.1 1.4 -2.1 -0.3 -0.5 1.5 -1.8 2.6 3.1 0.1 -2.5 1.2 1.4 0.2 0.5 -0.4 1.8 1.9 1.5 0.5 -0.4 1.1 3.6 0.0 0.9 1.2 2.3 2.3 1.0 0.5 -2.3 1.2 3.0 0.8 -0.6 0.0 -1.2 1.2 -0.2 0.1 -0.4 3.0 1.7 -0.7 0.0 1.2 -0.3 1.3 0.8 1.6 -0.6 -0.3 1.9 0.1 -0.4 0.2 -0.8 2.8 0.1 -0. 1 1.1 -3.0 0.9 0.6 -0.4 -0.9 0.6 0.4 0.3 0.0 -3.0 1.3 0.3 0.6 -0.1 2.9 -1.2 1.8 -0.1 RE 7. 1 1.6 1.7 2.3 3.0 2.4 1.4 0.8 2.3 2.0 1.8 2.3 1.6 3.0 0.4 2.1 1.6 1.3 4.4 2.3 1.8 0.0 2.8 1.0 2.8 1.5 4.1 9.3 2.7 4.2 1.2 2.0 5.8 0.5 1.3 3.4 0.7 1.3 0.8 1.9 1.9 1.3 0.6 ST 0.4 -0.3 -0.2 1.2 -0.4 0.8 -0. 1 -0.3 -0.8 -0.5 0.1 0.5 0.4 0.2 -0.4 0.7 -0.4 0.3 0.0 -0.2 - 1.0 0.7 1.2 1.0 0.0 0.8 0.6 0. 1 0.3 1.9 -0.5 1.9 -0.3 -0.6 0.0 0.6 -0.7 1.2 0.0 -0.5 0.9 -0.6 0.5 S W 2.8 2.6 2.0 1.2 3.5 2.5 -0.6 2.1 3.3 0.0 3.5 1.2 1.3 3.3 -2.0 1.3 -0.6 0.3 2.3 1.3 -0.3 4.4 0.4 2.5 1.8 0.7 0.0 -0.6 -0.7 -0.4 -1.3 -0.6 0.4 -0.4 -1.3 1.4 0.4 0.6 1.1 2.6 0.9 1.6 1.7 5.9 -0.7 1.1 0.3 0.0 0.4 0.0 67 Table 4 (Cont’d) 1.0 3.4 1.2 3.6 0.5 2.5 1.3 2.0 1.1 -1.9 2.3 3.1 0.6 1.1 0.4 1. 1 1.4 1.9 2.1 1.7 -0.2 -1.1 1.1 -0.7 0.5 1.4 0.3 1.7 68 legislation and which possesses veto power. (The Supreme Court also contributes to this process, but its contributions tend to be delayed and diffuse). Both the administrations’ activities and those Of the Congress in the relative content areas as recorded in WW1; were examined and scored. In the absence of a prior literature in this area, a simple scoring system was devised. For congressional actions: * Action (hearings, passage of a bill, etc.) in either single house of Congress was scored as one point (valence was assigned accordingly, i.e. defeat of a liberal bill is scored as a conservative point); * Simultaneous action in both houses was scored as two points; * Three points were scored if the completed bill reached the President’s desk for signing or veto - it turns out that numerous bills “disappear” in the process of reconciling Senate-and House versions in conference; * Four points were scored if a Presidential veto was overridden to enact a bill. For administration activities, each action initiated by a sitting administration (calling for legislation; vetoing legislation; proposing legislation) was scored as one point in the appropriate direction. Note that signing a bill into law was not counted as an action, as the vast majority of bills are signed; if the administration originated the bill, it was counted as above, as having been proposed. This limits the score for the administration to those issues on which it has taken a public stance. Similarly, legislation passed at the request of the administration was counted only for the administration and not double counted in Congress’s score. Table 5 contains the gross yearly scores for congressional activity broken down by Table 5 Congressional Activity By year and Content Area 131226033216322265119033m0m8m6014011334n303656 . . . 1. 1 .. SW ST -7 0 -1 — -14 0 -1 1 -2 RE -4 LA 2 -6 -11 0 -1 FE FA M51300633£0010212301000002360536030340404313001 0 “000004430 03009.. -3 8 . -1 -2 43.01”.233223243323000wd—410w1ufioo3 as 4884208 -2 L14454 0102293181030107323133330605 C. .. . ... . -16 -1 -10 Year 1945 1946 1947 1948 1949 1950 1%1 1952 1%3 1954 1%5 1956 1%? 1958 1%9 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1% 1%1 1982 1%3 1%4 1%5 19% 1%? 1988 1989 1990 70 content area. Table 6 contains the same for the administration’s activities. Sodoeoonomic Factors As discussed previously, two sorts of external socioeconomic factors should be considered part of a complete predictive model concerning the impact of media and other variables on public opinion. The first of these factors is the Socioeconomic Inputs, variables whose actions are so fundamental as to essentially act as causes of shifts in the public’s agenda. Economic and social factors have the potential to impact the liberalism of the nation by virtue of their obtrusiveness (or pervasiveness). An example is inflation: when the prices go up, virtually everybody is aware of it. Poor economic conditions can be expected to act to decrease liberalism because the responses to adverse economic conditions, for instance, isolationism, denigration of foreign products and by extension foreigners, and/or strife based on regional and racial origins, are more closely associated with the opposite of liberal viewpoints. The Statistical Abstract gfthg United States contains yearly data on the Consumer Price Index (CPI); while the inflation rate is not given directly, it is readily calculated from the tables of the progress of the CPI throughout the time period in question (see Table 7). Similarly, the Unified Crime Rate (again, available in the Statistical Abstract), might function in the same way, although on a separate factor of liberalism. The overall liberalism scale includes items on capital punishment, suggested penalties, and the like; the higher the per capita crime rate, the more likely one would expect that the populace would respond more conservatively, at least to that subset of items. This reasoning led to the development of the following criteria for Table 6 Administration Activity By year and Content Area 0 -1 -1 0 0 1 0 nmm.w..u.01103111000......1102400030024310..."...“.J.....0 -1 2160 ... 0 1 0 -2 -3 - 1 -4 -2 E0116011021131030436521332010000400014040 -3 -1 -2 -1 -1 -1 R -1 -1 -1 M041111034311200120422200340440n19w30015604401 M1401...23.0-0.434..20420042441010423320143340443.016.4243 72 Table 7 Socioeconomic Inputs By year and Input Area Ind. Seg Ind. Seg Crime Inflation Unemp Major Per Capita Year Blacks Women Rate Rate Rate Strikes Inc. 'lhx 1944 .281 .NA 114 NA NA NA 18.3 1945 .275 .315 132 NA 1.9 X 19.0 1946 .268 .334 142 8.5 3.6 X 16.3 1947 .262 .352 140 14.5 3.9 249 14.4 1948 .309 .348 136 7.6 3.8 222 14.2 1949 .315 .344 138 -0.9 5.9 232 12.1 1%0 .320 .340 133 1.0 5.3 371 11.3 1%1 .304 .358 128 7.9 3.3 356 12.8 1%2 .338 .367 139 2.3 3.1 3% 15.9 1%3 .318 .346 145 0.8 2.9 361 16.7 1%4 .298 .360 147 0.3 5.6 216 16.1 1%5 .310 .351 136 -0.3 4.4 $0 15.4 1%6 .311 .357 137 1.5 4.2 2% 16.9 1%? .314 .361 141 3.4 4.3 215 17.6 1%8 .318 .371 148 2.8 6.8 $2 16.4 1%9 .313 .361 147 0.9 5.5 183 16.9 1960 .305 .351 159 1.1 5.4 163 18.2 1961 .317 .350 156 1.0 6.? 140 17.9 1962 .308 .348 161 1.0 5.6 149 19.3 1963 .297 .349 166 1.3 5.7 1% 19.6 1%4 .286 .346 188 1.3 5.2 1% 19.5 1965 .284 .350 1% 1.6 4.4 182 19.0 1966 .266 .348 217 2.8 5.5 216 20.8 196? .264 .347 $0 3. 1 4.5 $3 22. 1 1968 .252 .344 2% 4.2 3.6 $8 23.5 1969 .236 .342 329 5.5 3.5 268 27.9 1970 .226 .345 363 5.? 4.8 245 27.1 1971 .220 .338 396 4.4 5.9 1% 24.4 1972 .210 .349 402 3.2 5.6 157 $.7 1973 .197 .349 418 6.2 4.9 197 26.2 1974 .195 .348 461 11.0 5.6 an 26.9 1975 .174 .344 481 9.1 8.3 144 23.1 1976 .181 .342 460 5.8 7.6 140 25.2 1977 .174 .344 467 6.5 6.9 179 28.3 1978 .162 .345 4% 7.6 6.0 1% 29.7 1979 .177 .346 548 11.4 5.8 138 31.8 1% .185 .338 59? 13.5 7.0 1% 31.0 1%1 .178 .335 5% 10.3 7.5 33 32.7 1%2 .183 .330 571 6.2 9.5 54 31.6 1%3 .172 .325 538 3.2 9.5 45 29.3 1%4 .173 .333 539 4.3 7.4 :5 28.9 1%5 .165 .336 55? 3.5 7.1 :1) 31.0 1% .166 .336 618 1.9 6.9 (B 31.4 Blacks .161 .173 .160 .156 .153 .156 Ind. Seg Ind. Seg Women .333 .329 .326 .325 .325 .327 73 Table 7 (Cont’d) Crime Inflation Rate Rate 610 3.6 637 4. l 663 3.4 732 6.9 758 NA NA NA Unemp Rate 6. 1 5.5 5.2 5.4 6.6 7.3 9 Major Per Capita Strikes Inc. Tax $ 33.8 21 32.3 27 35.2 23 32.6 N A NA NA NA 74 adding a variable to the list of socioeconomic inputs: 1) The variable must be obtrusive; 2) The variable must be tied directly to one or more of the content areas; 3) Changes in the variable can be linked logically to either conservative or liberal ideology. There is in addition the practical requirement that data on the variable had to be recoverable, whether directly from the literature, via computation fiom raw data, or via extrapolation between points not removed chronologically by more than five years.8 Both the annual Inflation Rate and the Unified Crime Rate meet all the criteria, the Unified Crime Rate being available in annual issues of the Statistical Abstract and the annual Inflation Rate calculated from Consumer Price Index data, also published annually. Five other variables meeting the criteria were selected for inclusion in the socioeconomic inputs: * The Unemployment Rate, published annually in the Statistical Abstract, whose impact on opinion can be expected to be similar to that of the inflation rate. * The Number of Major Strikes‘ per annum. This should be tied to the content area of Labor; and one would expect members of the public at large to be less sympathetic to organized labor as they are affected by the strike in progress. While the obtrusiveness of any given major strike might be low, the time period encompassed by the study contains a wide assortment of strikes which had major impact on the lives of the public, including strikes ’ Decennial census data, therefore,would be deemed insufficient in resolving power for inclusion. ‘The Department of Labor defines a major strike as one in which 1000 person-days of work is lost. 75 in the rail, coal, newspaper and music industries, and the raw number of major strikes called annually (the range across the course of the study is from a low of 21 in 1988 to a high of 395 in 1952) ensures widespread impact. (There is a slight dislocation in the strike data due to the prohibition on strikes in certain industries which was in force during and just after World War II.) The Number of Strikes per annum was corrected for the growth of the population across the time period in question, on the reasoning that the growth of the population was directly tied to the growth in industrial units; thus, all other things being equal, a larger population should generate a greater number of strikes. * Amount of Individual Income Taxes paid. Again, the magnitude of this would be expected to directly influence opinion on Spending and Taxes; it is clearly obtrusive in that it affects every household and sometimes multiple members of a household. The amount of income to the government from individual tax returns is reported in the Statistical Abstract; and these figures converted to a per capita figure by dividing through by the population. Additionally, the figures were corrected for inflation (as inflation had already been entered into the equation); * Industrial Integration of Blacks: it is established that, all other things being equal, people are more likely to like (i.e. have positive opinions of) people with whom they have contact (c.f. Perry, 1990). This is most likely to occur, involuntarily, in what has been termed “second” (or work) spaces, more so than in “first” (home) or “third” (social) spaces (Oldenberg, 1989). Thus, as the work force becomes integrated, more of the population has direct contact with people from different cultures. Increases in industrial integration should lead via propinquity to impacts on Race and Ethnicity. 76 Measurement of the integration of the work force has been dealt with in the literature on labor and industrial relations, with King (1992) nominating Duncan’s dissimilarity measure (Duncan and Duncan, 1955) as the most common measure of segregation between ethnic groups across industrial categories. Data on the distribution of blacks and/or minorities across occupational category has been available in the Statistical Abstract annually since 1954, and a variety of other sources provide data for the years 1944, 1947, 1948 and 1950-1952. Dissimilarity statistics were calculated for each year for which data was available, and interpolated for those in which they were not. In order to conform to the direction of the other Socioeconomic Inputs, the dissimilarity measure was retained instead of being converted to a similarity measure, as, as with the other, a . greater degree of dissimilarity should be associated with less propinquity and thus lower opinion on the topic. * Industrial Integration of Women: the reasoning for inclusion of this variable is similar to that for the inclusion of Integration of Blacks, with the obvious difference that many households have regular contact with women in first and third places. This variable is tied directly to the content area of Feminism and was treated in the same was as Industrial Integration of Blacks, except that data on the distribution of females by occupational category was available for 1953, and not available for 1948 or 1961. As with inflation and the Unified Crime Rate, annual figures for the other socioeconomic inputs are provided in Table 7. Finally, the discussions in the first chapter lead to the notion that, in order to be “real” (rather than “lip-service” or a bandwagon efl'ect), changes in public opinion have to eventually manifest themselves in changes in 7'7 public behavior; specifically, a more liberal public opinion needs to be followed by more liberal behavior. Table 8, for instance, includes a year-by- year listing of the number of minority and female members of Congress, this being an indication of liberal behavior in the areas identified by Smith as “democratic” and “egalitarian”. Again, each of these variables had to meet the criteria of being directly tied to a content area, being recoverable, and being tied to an ideological position; other variables which proved impossible to include due to a lack of available data include such items as the number of handguns in the nation. Variables recovered, besides the election of minorities and women to Congress (tied to Race and Ethnicity and Feminism, respectively), eventually included: * Amount of Annual Donations to United Way: Coming out of Social Welfare, these figures were available for many years in the Statistical Abstract and later in (fixing USA, a publication of the AAFRC Trust (1981- 1990). The figures for annual dollar amounts of donation were corrected for both inflation (to provide a constant dollar basis for comparison) and the growth of the population (on the grounds that a larger population both requires more services and provides for more potential donors)“ * Union Enrollments: Tied to Labor, estimated total Union Enrollments were also available for many years in the Statistical Abstract. While these estimates were unavailable for much of the 1980’s, Chang and Sorrentino (1991) provide a different set of estimates for the period of 1975 through 1990. The period of overlap made possible the calculation of a correction factor designed to reconcile the two estimation procedures (in the ‘ It is worth noting that the United Way scandal falls outside the scope of this study and thus did not influence these data. 78 Table 8 Socioeconomic Outputs By Year and Output Area Election of Election of United Way Union Relative Pay Year Minorities Women Donations Enrollments women to men 1944 4 8 NA 14.6 70.8 1945 4 8 221.0 14.8 69.4 1946 4 8 171.5 14.1 68.1 1947 4 8 128.3 14.2 66.1 1948 4 8 120.0 13.6 ‘ 65.6 1949 4 10 125.9 13.3 65.4 1%0 4 10 126.5 13.1 65.4 1%1 4 11 126.5 14.4 64.5 1%2 4 11 137.4 14.1 63.6 1%3 4 15 147.5 14.8 63.2 1%4 4 15 157.3 14.7 63.6 1%5 5 1? 162.4 14.2 62.5 1%6 5 17 178.0 14.5 63.4 1%? 4 15 186.8 14.2 63.7 1%8 4 15 196.2 13.? 62.9 1%9 5 17 196.5 13.6 60.6 1960 5 17 207.8 13.2 59.6 1961 6 17 207.6 12.4 58.4 1962 6 1? 211.9 12.4 58.5 1963 7 13 216.2 12.3 59.4 1%4 7 13 219.0 12.4 59.5 1%5 10 12 227.5 12.6 59.4 1966 10 12 247.3 12.9 57.3 196? 11 11 240.0 13.1 57.7 1968 11 11 246.0 13.3 58.2 19% 15 11 245.3 13.3 60.3 1970 15 11 244.7 13.4 59.1 1971 19 13 237.9 13.1 59.3 1972 19 13 234.? 13.1 57.2 1973 22 14 232.0 13.3 56.0 1974 22 14 221.? 13.8 57.5 1975 24 19 210.0 13.9 58.1 1976 24 19 212.4 13.6 58.9 197? 22 18 216.2 13.7 57.9 1978 22 18 216.1 13.3 58.3 1979 21 17 208.0 13.0 58.5 1% 21 17 194.0 12.1 59.2 1%1 23 21 191.9 11.3 58.4 1%2 23 21 188.7 10.9 59.8 1%3 29 23 198.0 10.6 61.8 1%4 a; 23 207.9 10.3 62.6 79 Table 8 (Cont’d) Election of Election of United Way Union Relative Pay Year Minorities Women Donations Enrollments women to men 1%5 30 24 213.3 10.2 63.2 1% so 24 219.4 10.0 63.0 1%7 34 25 223.6 10.0 63.9 1%8 34 25 228.4 9.9 65.0 1% 34 27 235.5 9.8 66.8 1990 34 27 226.6 9.6 69.7 80 form of a corrective multiplier); the corrective multiplier was applied to the Chang and Sorrentino data to bring it into line with that from the Statistical Abstract. These figures were then corrected for population growth, on the assumption that a larger population created a larger pool of persons to be recruited from; * And finally, Women’s Pay (expressed as a fraction of Men’s Pay), was derived from the document “Earning Differences between Women and Men” (Department of Labor, 1993), which provides data on the average earnings for men and women annually for the 1951-1992 period; relative pay was arrived at by dividing the average annual earnings of women by that of men. The annual averages for the period 1945-1950 were arrived at by linear extrapolation of existing trends in the data. As with the number of female and minority Congressmen, annual figures for the other socio-economic output variables are tabulated in Table 8. The creation operational definitions for the theoretical variable clusters shown in Figure 2 leads to the expanded or “measurement” model laid out in Figure 7. Method of analysis As time series data is approached with relative infrequency in communication research (Bruneau, 1994), there seems to be little agreement on appropriate statistical analysis of it. The analysis of time series is generally approachable from one of three directions, that is, multiple regression approaches, autoregression approaches, and what might be termed composite approaches. Cook, Dintzer, and Mark (1980) outline four specific procedures appropriate to the analysis of time series data which subsume two of those directions. 81 HI 15 El ssé 8:08 I! El =ou~bfl=€€< 3:333:00 lg who: 8 hi @8803 we ovum HI I! ace—£68m :85 El 85% .8. o I! 2888 El < .2 m In! .3? USED E] l! 8:83 we 882m Ell. I! 385 no 882m a a 8:930 3: E oMEocoooomoom 883.. 233 2888028 pars? 38m a 8i 2:5 Bus: 8me can 3%:on E swag: 3553‘ 36:55 93 8am El 883‘ «602 82: 2 825 !| a . 88.5 8.52 Emegom Q We 835:2 m5 amaze 863m - noun» w. t< . m a .8389 3589: o a a o 2888 = u n m . m a 8an - 89% a -Bwomoa 3889: a 883: E6 2am 5d. 2:85 Figure7 Operational Definition Overall Model 82 The first method overviewed by Cook, et al., is the use of Ordinary Least Squares (OLS) estimation; that is, the dependent measure is regressed onto the independent measures in the well-known manner of multiple regression. In order to determine causal ordering, the dependent measure is regressed onto the independent measures at previous time periods, with the expectation that if the lagged measures possess substantive predictive power even when the synchronous measures are controlled for, then the independent measure possesses some form of causal priority over the dependent measure. Cook, et al.’s, assessment is that OLS is inadequate to the task of testing the mutual impact of time series for two mathematical reasons. The first is that inadequate de-trending of data can leading to the measurement of spurious relationships between otherwise unrelated measures trending coincidentally (or because of mutual causation by a third, also trended, variable); the second is the possibility of unmeasured but interrelated variables creating autocorrelated error terms (see Kenny, 1979). The former has the potential to artificially inflate beta weights, and while the latter does not affect the OLS estimators, it does deflate the standard error of those estimators, leading to overly liberal significance tests (Hibbs, 1974; the presence of an autocorrelated error term can be detected by the presence of a significant Durbin-Watson (1 statistic for the analysis, or by Durbin’s h under certain circumstances). The second procedure overviewed by Cook, Dintzer, and Mark is that of Generalized Least Squares (GLS; see Ostrom, 1978), which is the general case from which OLS estimation is derived. GLS estimation corrects for autoregressive error by estimating the coefficient of the autoregressive error, and using that estimate to correct the original series to remove the 83 autoregressive component of the error terms fiom the original time series. Cook, et al., comment that GLS is potentially flawed in that a) if the autoregressive correlation in the error terms is higher than first order it is unlikely to be completely corrected for, and b) the technique makes a priori assumptions about causal ordering. It should be noted that for the proposed study, however, that a) an improperly corrected autoregressive error will still yield a significant test of correlated error terms, and that b) the causal ordering of variables has been largely determined in the prior literature; what is at issue herein is their interrelationship, largely in terms of relative magnitude of impact. Thus it may not be useful to reject the use of GLS estimation techniques for this particular study at this time. Cook, Dintzer, and Mark continue with a discussion of Auto- Regressive Integrated Moving Average (ARIMA) models (Box and Jenkins, 1976), which are first designed to describe the behavior of a single time series across time, specifically, in terms of autoregressive and moving average components. Thus, if there is presumption that X causes Y, the ARIMA structure of X is determined and then both series are “prewhitened” or de-trended using this structure. Finally, a transfer function which estimates the impact of X on Y at given (or for all) time lags is calculated. Cook, et al., point out, however, that use of the ARIMA model for X in pre-whitening Y may be inadequate in two regards. First, it again assumes the causal priority of X over Y. Second, they point out that the use of X’s ARIMA model might inadequately pre-whiten, leading again to the dificulty of spurious relationships based on a common, but not causal, trend being calculated between X and Y. 84 In order to eradicate the possibility of common trends indicating spuriously large causal relationships between variables, and additionally to assist in determining the question of causal priority in systems where it is unknown, Cook, Dintzer, and Mark, following the lead of Granger (1969), recommend constructing an independent ARIMA model for each variable in question, separate pre-whitening based on each variable’s specific model, and then calculation of transfer functions. This procedure both completely de-trends each variable (at least to within the capability of the ARIMA method), and also obviates the necessity of identifying one variable as the cause of the other. Unfortunately, this procedure also carries with it the danger of obscuring true causal relationships. There do exist variables whose value at a given point in time are dependent directly on their value at a previous point in time; for instance, the early stages of the spread of a disease through a large, geographically homogeneous, and non-immune population. The number of cases which will occur is directly a function of the number of carriers, i.e., the number of cases which have occurred. In this case, a second variable which was similarly trended, for instance, the average daily temperature during a springtime outbreak of the disease, would yield spurious indications of causal priority if both series were not properly de-trended. However, consider the alternative case, of a variable whose value is solely dependent on the value of a second variable at a previous point in time; in continuing with the health care example presented above, let us say the number of dollars spent on the care of the victims of that hypothetical disease (the genesis of the time lag being in the time needed for patients to demonstrate symptoms, and also that needed by the health care 85 system to generate billings). The ARIMA models will not be a perfect match; specifically, there will be a small discontinuity at the very beginning of the models, where the first people are becoming ill but as yet no health care dollars are being spent on them; but for a long epidemic they should match quite closely. When the second (health care dollars) time series is pre-whitened, almost all of its variability will be accounted for by its own previous behavior, and thus it will be assumed, incorrectly, to be a nearly perfect auto-regressive series just like the first time series. Krull and Paulson (1977) advocate a third process, which forms, in essence, a synthesis of the first two: the dependent variable is in effect “pre- whitened” as part of its own prediction equation by including itself, lagged, in the prediction equation as part of a Least Squares analysis. (Although mute on the specifics, Krull and Paulson advocate careful observation of the Durbin-Watson statistic, as well as Durbin’s h, which measure the correlation between the error terms, implying that GLS techniques be used if this statistical “pre-whitening” is insufficient.) Appropriate (i.e. significant) lags of independent variables are derived from analysis of a cross correlation matrix of independent variables at various positive and negative lags with the dependent variables. Together with the results of the autoregression analysis of the independent variable, a predictive model is derived and tested. This is also consistent with the medthods employed by 'Irumbo (1995). A commonly accepted (Smith, 1980) test of time series models is provided by the logic of Granger’s (1969) comment that “a perfectly deterministic series, that is, a series which can be predicted exactly from its past terms such as a non-stochastic series, cannot be said to have any causal influences other than its own past (pg. 430).” Granger called for as tests of fit based on the variance of the error of the dependent variable regressed onto the universe of variables being less than the error created by the regression of the dependent measure onto that same universe minus the proposed causal variable. (In other words, that the proposed cause have predictive power after all other variables have been controlled.) This position is clearly based on the rationale of protecting the body of knowledge against Type I error, that is, the possibility that, due to sampling or measurement error, the researcher will reach the conclusion that a relationship exists where in reality none does. Like many such tests, however, it does so at the cost of increasing Type II error, the error made when the conclusion is drawn that no relationship exists when in reality it does, such as that which would result in analyzing the hypothetical health care situation as presented above. Despite the fact that one variable was completely caused by the other, the conclusion would be that it was autoregressive, which is, after all, simply another false positive, albeit of a difi'erent, arguably more parsimonious, sort. Historically, this position derives from the attempts of logical positivism to create a form of objective research by making decision making statistically deterministic; hence, the mathematical cutoff of “statistical significance”. However, as Hunter has pointed out on many occasions, the true object of science is to make true statements about the universe, and thus to minimize error of all sorts. Krull and Paulson’s procedure obviates this concern by entering both the lagged independent variables and the lags of the dependent variable into the same regression equation, where they in essence compete as predictors of the independent variable. Thus, the analysis of the data gathered proceeded in four stages: 1) Data for each each of the variable clusters was gathered and 87 correlation matrices within clusters calculated. Cluster analysis via the methods advocated by Hunter (1978) was used to reduce the data to those items best representing the cluster as a whole, and the items aggregated for form a single trendline.‘ 2) Autocorrelations and partial autocorrelations were calculated for lags up to 8 years for each of the aggregated trend lines. This allowed the description of the each variable in question, albeit across the relatively short term. 3) A matrix of cross-lagged correlations was in turn constructed fi'om the trend lines. The matrix of cross-lagged correlations was limited to lags of 8 years due to the relatively small number of observations available to draw from. 4) The matrix of cross-lagged correlations was used to select lags of all variables for entry into the regression model. Two separate models were formed, one predicting the public’s agenda and the second socioeconomic outputs, in order to provide a test of the model presented as Figure 2. Also included in the regression equations were values of the dependent measure in previous years, as shown necessary by the analysis of partial autocorrelations (years with significant partial autocorrelations were included in the analysis to “pre-whiten” the dependent measure, as above). Durbin’s h, a statistic equivalent to the Durbin-Watson statistic but designed for use in models containing lagged endogenous variables (Ostrom, 1978), was observed, and in the case of a significant Durbin- Watson statistic, GLS estimation by the method of Instrumental Variables (IV-GLS; Johnston, 1970) replaced OLS estimation. The same procedure was followed in the two case studies as well, “As will be seen, this led the analysis of the Policy Agenda to be split into separate Administration and Congressional clusters. 88 except that the analysis of specific content areas obviated the necessity of data reduction, and thus the cluster analysis stage was skipped for those analyses. Selection of the year as the unit of analysis is based largely on the fact that many of the data are measured on a yearly basis. Ideally, in order to permit numerous simultaneous estimations under GLS, time periods should be as numerous as possible. Data on the media agenda, the policy agenda, and liberalism itself are available accurate to at least plus or minus one month, in principle making it possible to use the month or at least the quarter the unit of analysis. Unfortunately, many of the other variables, especially Crime Rate, Inflation Rate, and other social statistics, are available no more fiequently than annually. In addition, the choice of the year as the unit of analysis permits removal of two potential problems with more accurate data. One of these potential problems is seasonality within the data; for instance, unemployment decreases annually during the winter quarter due to the hiring of temporary help for the Christmas season. Additionally, as we have seen in Chapter One, agenda issues take a period of months to build up. Using the year as a unit of analysis will reduce sampling error by reducing the random shocks on liberalism created by the short term movement of agenda items, those movements that would NOT be described as being caused by “rally” events. Chapter Three Results In following the multi-tier approach laid out in the previous chapter, discussion of the results obtained in the individual content areas will precede the overall analysis. The last part of this section will be the analysis of the specific content areas of Race and Ethnicity, Foreign Afi‘airs. Public Opinion Figure 8 shows the result of cumulative annual net trend as calculated for the purposes of this study, the function of this calculation being the replication and extension of Smith’s original results (Smith, 1980, p. 496). Comparison of the two figures shows a high degree of correspondence, the most prominent differences being that the data in Figure 1 starts earlier and ends earlier. Key relevant features, including the strong upturn in liberal opinion starting in the early 1950’s, the acceleration which takes place round 1970, and the plateau around 1980, are as clearly shown in Figure 8 as in Figure 1. The additional data made available since the publication of Smith’s paper also answers one of the key questions raised earlier: the early 1980’s plateau proves to be short-lived; starting about 1984 the public’s agenda resumes its liberal trend. Note the beginning of another plateau in the mid- 1990’s (although this is outside the scope of the present study). Figure 9 graphs the average annual net trend against year. 89 m amnesia-.- 2:9 3.24 53:26 Figure8 Cumulative Annual 'll'end 1940-1994 .02 ‘ .015 ‘ °°: H'.1i'45')'4€i g 3 0 til i b 24 D -.005- «01‘ l -.015 -.-r-.--1-4fiv 193019401950196019701$019mm Year Figure 9 Average Net Trend 1940-1994 92 Following the measurement procedure presented previously, positive scores indicate liberal trends and negative scores conservative trends. Note that the net is positive during 43 years and negative in only 10. The average annual net trend was .4 points; with a standard deviation of .59 and a standard error of .08, this differs significantly from zero. Analysis of the reduced set of trends within the nine content areas retained yielded the correlation matrix presented as Table 9. Note that the content area by content area n’s are somewhat inconsistent due to years during which no active trends were present in one or the other content area; these years were treated as missing data and excluded from the individual correlation coefficients. Examination of the correlation matrix yields several interesting findings. Opinion in the area of Civil Liberties is generally negatively related to opinion in the other content areas, with the exception of that in Foreign Affairs. This tendency was unexpected, and may indicate the willingness of the public to accept intrusion into their civil liberties in order to advance other worthy goals. (The positive relation between Civil Liberties and Foreign Affairs is probably due to changes in opinion fi'om the 1950’s and 1960’s, as reductions in Cold War tensions led to the lifting of Cold War era restrictions on the freedom of association.) Feminism is associated with four issues other than Civil Liberties, all positively. Opinions on Labor, Race and Ethnicity and Social Welfare all seem reasonable to be associated with Feminism, but the highest relationship, in fact, the highest on the entire table, is between Feminism and Spending and Taxes. This could be explicable as some form of advocacy for government intervention in Affirmative Action if there was a similar correlation between Race and Ethnicity and Spending and Taxes, but there Area CL CR ER FA LA RE ST S W 93 Table 9 Overall analysis Matrix of Correlations for Public Opinion by Content Area CL CR ER FA FE LA RE ST SW 1.000 .122 1.000 -.094 .446***1.000 .397" .290* .034 1.000 -.300**—.141 -.204 .131 1.000 -.065 .369" -.138 .040 .257”I 1.000 —.432**-.110 .085 .024 .309“ —.219 1.000 -.097 -.084 -.035 .015 .509‘".180 .127 1.000 -.035 .002 .341" .210 .265" .195 .207 -.166 1.000 *=p<.05 **=p<.01 ***=p<.001 is not. Cluster analysis of this correlation matrix led to the removal of Civil Liberties from the cluster, due to its largely negative associations with the other content areas, after it was verified that these relationships were not the result of miscoding. All the other areas are significantly and positively related to at least one other content area, and removal of Civil Liberties removes all the statistically significant, negative results. Average annual net trend data for the eight remaining content areas was calculated (by summing the gross annual changes in each content area and dividing by the number of trends active in that year) and subjected to AR modeling. The results of this process are presented as Table 10. (Note: In this and subsequent AR modeling of the other variable clusters the autocorrelations and partial autocorrelations are presented up to lag 8. This was to create the possibility of detecting trends associated with the service of two-term presidents.) Clearly the Annual Net 'h'end data form a simple, first-order process. Media Agenda Before aggregation, inter-coder reliabilities were calculated for codes assigned to articles within each content area and to the sample of articles as a whole. These reliabilities are presented in Table 11. Although there is no hard-and-fast rule concerning the evaluation of reliability, Krippendorf (1980) suggests that items of reliability less than .67 be omitted, those in the range of .67 to .80 be discussed only tentatively, and only those of .80 and higher be used in the ordinary way. As will be seen on Table 11, all the content areas show reliabilities in excess of .67, while Crime, Labor, and Spending and Taxes fail to meet the higher standard. r' m on mammawrcu % Table 10 Overall analysis Autocorrelations and Partial Autocorrelations Public Opinion Partial Autocorrelation Autocorrelation .365" .365“ .012 -.132 .072 .136 .012 -.074 -.084 -.054 -.057 -.046 -.136 -.123 -.196 -.084 *=p<.05 **=p<.01 ***=p<.001 % Table 11 Intercoder Reliabilities on Media Agenda By Content Area Content Area Reliability Civil Liberties .867 Crime .763 Economic Regulation .844 Foreign Affairs .807 Feminism .876 Labor .789 Race and Ethnicity .864 Spending and Taxes .734 Social Welfare .873 Overall .849 97 In most cases, however, the individual article is only one of multiple indicators of prevailing media bias on a given topic within a given year. The use of multiple (typically 5) indicators has the effect of increasing the efl‘ective reliability (Hunter, 1978), in this case to well above acceptable levels.1 The overall reliability of .849 for all articles indicates that the coding scheme as a whole was acceptable. The matrix of correlations for the media agenda by content area is presented as Table 12. Only two of the correlations are statistically significant, that between Civil Liberties and Foreign Affairs (-.349, p< .01) and that between Labor and Feminism (.370, p<.01). The magnitude of these two correlations is substantial enough to make it unlikely that they are the result of Type II error; in fact, note also that these two pairs of content areas were correlated within the public opinion data as well, although the relationship between Civil Liberties and Foreign Affairs was positive for public opinion. Probably of greater consequence is the fact that the average correlation on Table 12 is only .022, essentially (as well as effectively) zero. If there exists a monolithic media agenda which shows systematic and overt bias in one direction or another, substantial intercorrelation would have resulted. The lack of patterning in the correlation matrix made data reduction by cluster analysis impractical (although several attempts were made). As a result, the trend line for the media agenda was calculated by simply summing the average for each content area for each year (no scaling was necessary as they were measured on the same metric, average net ' In this case, using five items of reliability .734 (the lowest appearing on Table 11) yields an overall scale reliability of .932 using the Spearman-Browne formula, nrx x/[(n- 1)rxx + 1]. Area CL CR ER FA FE LA RE ST S W % Table 12 Overall analysis Matrix of Correlations for Media Agenda, by Content Area (N =47 ) CL CR 1.000 -.028 1.000 .217 -.061 -.349**-.020 .063 .089 .102 .030 .200 -. 172 .029 . 118 -. 139 -.073 ER 1.000 -. 140 .174 .088 -. 150 .042 FA FE LA RE ST SW 1.000 .150 1.000 .022 .370**1.000 .193 -.156 -.173 1.000 .125 .114 -.005 .134 1.000 .096 .122 -.073 -.034 -.199 1.000 *=p<.05 ** =p< .01 *** = p < .001 t“ m an mQQOIACONI-i 99 Table 13 Overall analysis Autocorrelations and Partial Autocorrelations Media Agenda Partial Autocorrelation Autocorrelation .266* 266* .173 .110 .278* .226 -.194 -.354* .074 .233 -.036 -.221 -.222 -.014 .066 125 *=p<.05 **=p<.01 ***=p<.001 $833885 EB 1(1) statements L or C). This trendline was autoregressed, yielding the autocorrelations and partial autocorrelations presented in Table 13. Along with a significant, positive lag one partial autocorrelation, there is also a significant and negative lag four partial. The genesis of this is obscure, although the four-year period suggests a correspondence to the presidential election cycle. PolicyAsenda Scores for the congressional and administration agendas were calculated and correlated on a content area by content area basis. The correlations for the congressional agenda are presented in Table 14, and those for the administration in Table 15. Six of the 36 correlations on Table 14 are statistically significant, three of them involving the content area of Labor (with Civil Liberties, Spending and Taxes, and Social Welfare). The first is probably due to the extensive debate surrounding the passage, and eventual repeal of the Taft- Hartley act, which limited the rights of certain classifications of workers to strike, and thus was seen as a Civil Liberties question by some. Feminism in Congress is also related positively to action on Civil Liberties, as well as to that on Social Welfare (the correlation between Feminism and Race and Ethnicity, .235, is at the .05 < p < .06 level), which may indicate some sort of confluence of Great Society issues. Finally, Civil Liberties and Spending and Taxes are significantly and negatively related. This may be the result of a historical coincidence between the Red Scare and balanced budget attempts of the 1950’s. Cluster analysis of the correlation matrix presented in Table 14 led to the elimination of three of the content areas from the congressional agenda. Area CL CR ER FA FE LA RE ST SW 101 Table 14 Overall analysis Matrix of Correlations for Congressional Agenda, by Content Area (N = 46) CL CR 1.000 .045 1.000 -.089 .093 .164 -.018 .243‘“ -.158 .362" .025 .140 -.032 -.290* -.212 .134 .199 ER 1.000 .211 -.040 . 195 -. 140 .023 .237 * FA FE LA RE ST SW 1.000 .146 1.000 .008 .197 1.000 -.182 .235 .162 1.000 -.033 -.005 .259” .055 1.000 .236 .327* .334'“ -.013 -.005 1.000 * = p < .05 ** = p < .01 **=p<.001 102 Table 15 Overall analysis Matrix of Correlations for Administration Agenda, by Content Area (N = 46) Area CL CR ER FA FE LA RE ST S W CL 1.000 CR .296” 1.000 ER .134 .063 1.000 FA .275” -.009 .026 1.000 FE -.034 .065 .004 -.175 1.000 LA .103 .000 .096 .087 -.128 1.000 RE .083 .089 .347**-.013 .128 .371**1.000 ST .189 .025 .204 .034 .058 .431" .442**1.000 SW .193 -.166 .418*’“.094 -.096 .597***.472**"‘.541***1.000 *=p<.05 ** = p < .01 ***=p<.001 103 The column sum for Spending and Taxes was negative, indicating either that it was reverse coded or not part of the major cluster; once its coding was verified it was eliminated. Similar, the column sums for Race and Ethnicity and Crime were essentially zero, indicating that they were not central to the the congressional agenda, at least along the Liberal- Conservative dimension. A trend line was constructed for the congressional agenda by taking the sum of the score for each content area weighted by its factor score. The results of the autoregressive analysis of this trend line is presented in Table 16. Along with the strong Lag 1 partial (indicating that the activities of the previous year are a good predictor for any given year), there is also a significant Lag 4 partial, indicating a four year cycle in the data. This is probably tied to the presidential election cycle; for instance, many presume that legislative activity tends to be low during the election year (due to time off for campaigning) and high in the year following the election (as new programs are implemented). Table 15 shows the correlations by content area for the administration agenda are substantially stronger than those for the congressional agenda, as might have been suspected given the nature of the two: the congressional agenda must of necessity represent a degree of compromise in order to create activity, while the administration’s agenda can be substantially more monolithic. As a result, the average correlation in the matrix for the administration agenda is almost twice that of the Congressional Agenda (mean r = .146 vs. .078). In the administration’s agenda Labor, Race and Ethnicity, Spending and Taxes, and Social Welfare are all strongly interrelated, which seems reasonable given the traditional appeal of the Democratic Party to groups t" m on mfldimuthi-A 104 Table 16 Overall analysis Autocorrelations and Partial Autocorrelations Congressional Agenda Partial Autocorrelation Autocorrelation .607*** .607*** .394" .039 .298“ .068 450*“ .364* .416M -.025 .238 -.126 -.020 -.259 .079 .160 *=p<05 ** = p < .01 *** = p < .001 {3638815 $88 105 such as organized labor and minorities and its attempts to aid these groups via spending, and particularly spending on welfare. A Republican administration would reasonably be expected to reverse all these areas. Other significant correlations include Economic Regulation with Race and Ethnicity and Social Welfare, which again follows fi'om the party-based reasoning presented above. Civil Liberties is once again related to Foreign Affairs, possibly for the reasons suggested above, and to Crime. This latter may again be an artifact of the restrictions placed on the right of association during the Red Scare years. Cluster analysis eliminated Feminism and Crime from the administration’s agenda. The former seems unlikely given the argument concerning traditional party priorities made above, but it should be recalled that there was only one Democratic administration within the scope of the study which followed the advent of the Women’s Liberation movement, and it was not strongly feminist in orientation due in large part to President Carter’s religious (anti-abortion) beliefs. As a result, the column sum for Feminism is negative, indicating that it is not part of the central cluster. That for Crime is essentially zero; many of the administrations, both Republican and Democratic, have attempted at various times to “get tough on crime”, leading to a lack of variability in that content area. As with the congressional agenda, the trend line for the administration agenda was constructed fi'om the sum of the individual content areas weighted by factor score for each year. This trend, too, was autoregressed, leading to the autocorrelations and partial autocorrelations presented in Table 17. As with the other trends discussed so far, there was a significant lag 1 partial autocorrelation, indicating a year-to-year stability of the agenda. t" m m mk‘lmo'lnhODNO-d 106 Table 17 Overall analysis Autocorrelations and Partial Autocorrelations Administration Agenda Partial Autocorrelation Autocorrelation .518*** .518*** .418" .204 .314* .083 .201 -.053 .400” 346* .316* .033 .208 -.106 .337* .241 *=p<fl5 **=p<.01 ***=p<.001 818836?” 88 107 More interesting is that the administration agenda does not show the lag 4 cycle which appears in several of the other trends, but does show a significant 5 year cycle. The lack of a four-year cycle may possibly be explained by the fact that there was only one President who served exactly one four year term during the course of the study (Carter; the end of Bush’s term fell outside the bounds of the study). A president reelected can be expected to continue doing whatever it was he was doing in the previous four years; thus, that aspect of the election cycle would be subsumed into the lag 1 figures. Additionally, several of the presidents covered by the study initiated their terms outside of the four year cycle ('II'uman, Johnson, and Ford). The existence of a lag 5 cycle is more problematic to explain. Possibilities include: incomplete partialing due to Type II error; or the activities of not the politicians at the head of the agenda but of the professional managers who serve in the middle echelons of an administration. Common management practice calls for medium term planning cycles, typically five years in duration. This Lag 5 cycle might be the manifestation of the results of these management planning cycles percolating “to the top”, as it were. Examining the pattern of correlations between Table 14 and 15 indicates that the two political agendas are not particularly strongly associated. Only five of the nine content areas appear in both of their associated trendlines. Further, the correlation between the two trendlines is insignificant (r = -.199). This makes it unreasonable to combine the two into a single policy agenda, and thus the administration agenda and the congressional agenda will be maintained as separate agendas. 108 External SodoeconomicFactors The matrix of correlations among the socioeconomic input variables is presented as Table 18. The number of large correlations is noteworthy, and suggests again that there is a substantial amount of trending in the data. An obvious explanation for the substantial amount of intercorrelation among these variables is simply that they are all chronologically trended to a fairly large degree. Occupational segregation has been declining for both blacks and women at various rates since the 1960’s; during that same time period a “typical” unemployment rate has expanded from under 2% during the war years to today’s 5-6% nationally.2 Income taxes, driven by government expansion, have also tended to increase across that time period, despite various short-term fluctuations. Inflation, on the other hand, tends to be less strongly trended because of its more dynamic nature, leading to correlations which are less in magnitude; but it is still somewhat trended as witnessed by a “typical” inflation rate of 1 1/2% during the Eisenhower years versus 3-4% during Reagan’s second term. More noteworthy, perhaps, is the pattern of signs in Table 18. The economic indicators (with the exception of strikes) are negatively related to the indicators of propinquity. Both conceptually and mathematically this is the clearest possible indication of a two cluster structure. For this reason, the economic variables were separated from the propinquity variables and the two clusters maintained separately in further analysis. The simplest explanation for the correlations between the number of strikes and the other variables being negative is miscoding. (Conceptually, of course, the number of strikes should remain with the economic variables even though as originally coded they are positively related to the propinquity variables.) Unlike the content-based variables discussed earlier, the coding ' National unemployment as announced on September 1, 1996: 5.6%. 109 Table 18 Overall analysis Matrix of Correlations for Socioeconomic Inputs, by Area Integ Integ Income Blacks Women Crime Infl Unemp Strikes Tax Integratn Blacks 1.000 Integratn Women .726*** 1.000 Crime -.961*** -.746*** 1.000 Inflation -.499*** -.207 .410*** 1.000 Unemplmt -.568*** -.434** 576*" .100 1.000 Strikes 682*” .700*** -.693*** -.159 -.592”* 1.000 Income Tax-326*" -.692*** 956"" .358" 507‘" -.687*" 1.000 *=p<.05 **=p<.01 ***=p<.001 110 of a larger number of strikes being evaluated as having a “liberal” valence was hypothetical. For the purposes of data reduction by cluster analysis, the number of strikes was reverse coded, as the magnitude of the correlations involved strongly suggests that the number of strikes is related to the other variables; in fact, the cluster analysis indicated that all the variables should remain in one cluster or the other. For the purposes of determining a trend-line for the economic variables, a composite variable was constructed, consisting of the sum of the standard scores for each of the economic input variable. This, too, is a deviation from the procedures followed previously, which was necessitated by the fact that unlike the various items making up the various agendas, the variables mahng up the economic inputs were measured using (sometimes grossly) different metrics. The use of mean—deviation removes the considerations of scaling from these data. The propinquity variables, being measured on the same metric (the 0 to 1 scale yielded by Duncan’s D), were simply added. Autoregression of the economic input variables leads to the autocorrelations and partial autocorrelations presented in Table 19. The pattern of substantial autocorrelations back to lag 8 (and clearly indicating that it may continue into still longer lags) and the very high lag 1 autocorrelation indicates that, as suggested above, the data are strongly trended across time. The same is true for the autocorrelations and partial autocorrelations of the propinquity cluster, which is presented as Table 20. Of particular interest is the presence of significant higher-order partials on both Tables 19 and 20, in particular, the significant lag 8 partial for each, negative in the case of the economic variables and positive in the case of the propinquity variables. The partials for all variables were 1" no on mummawmu 111 Table 19 Overall analysis Autocorrelations and Partial Autocorrelations Socioeconomic Inputs - Economy Partial Autocorrelation Autocorrelation 966*" 966*“ .933*** . -.017 .909*** .124 .882*** -.049 854*“ -.023 834*" .120 829*" .118 .786*** -.409* *=p<.05 **=p<.01 ***=p<.001 l8 888888 L" on m mammpccrcu 112 Table 20 Overall analysis Autocorrelations and Partial Autocorrelations Socioeconomic Inputs - Propinquity Partial Autocorrelation Autocorrelation .979*** .979*** .960*** .037 .938*** -.079 .921*** .087 893*“ -.297* .868*** -.043 .835*** -.276 .820*** .449“ *=p<.05 **=p<.01 ***=p<.001 $8383 8 113 calculated out to lag 8 in search of potential political effects tied to the presidential election cycle. It is important to recognize that although the average presidential administration during the course of this study has been substantially less than eight years, the possession of the White House by one party or the other has proceeded on an almost perfect eight year cycle until broken by the defeat of Carter in 1980, i.e. 1945-1952 Roosevelt/Truman, 1953-1960 Eisenhower, 1961-1968 Kennedy/Johnson, 1969-1976 Nixon/Ford. The difference in signs is consistent with this: all of the presidents named actively encouraged the civil rights movement’ to some degree at the beginning of their administrations (hence a positive effect on propinquity), but Democratic and Republican fiscal policies are diametrically opposed, at least as defined for the purposes of this study. The propinquity data also show a significant, negative five year lag. As was brought out in the discussion of the autocorrelations of the administration’s agenda, the genesis of a five year lag is somewhat obscure; and it is unlikely that the two are related in that the five year cycle in the administration’s agenda is positive and the five year lag in the propinquity data is negative. If the two are related, it would indicate that the five year programs hypothesized above intended to be liberal but actually had the effect of reducing integration. This is not likely, but not outside the realm of possibility. The correlations between the socioeconomic output variables are also presented, in Table 21, which poses much the same sort of analytical problem as Table 18: one variable (union enrollments) is negatively related to the other variables, and one (relative pay of women to men) shows little apparent relationship to the other variables. The other three ’ It seems likely that this effect would have been magnified if the data on female workers had been removed from the propinquity cluster. 114 Table 21 Overall analysis Matrix of Correlations for Socioeconomic Outputs, Congrssnl Blacks Congrssnl Women United Way Donations Union Enrollments Relative Pay (F/M) Cong Blacks 1.000 .795*** .523*** -.780*** -.095 by Area Cong United Way Women Donations 1.000 .319* 1.000 -.781*** -.384** .041 -.482*** *=p<fl5 ** = p < .01 *** = p < .001 Union Relative Erlmnts Pay 1.000 -.094 1.000 115 variables...election of blacks to Congress, election of women to Congress, and donations to the United Way...are significantly correlated, possibly again because they are trended, albeit not as strongly as the input variables are. (For instance, the number of women in Congress reaches a minimum in the 1960’s, as opposed to the number of blacks, which expands throughout the period of the study. Both reach maxima by 1990, however.) As with the number of strikes in the analysis of socioeconomic inputs, there is no compelling reason to maintain that union enrollments was correctly coded with regard to valence; and the substantial magnitude of the correlations between it and the other output measures suggest that it should simply be reverse coded for the sake of cluster analysis. This cluster analysis indicated that the relative pay variable was at best only weakly related to other members of the cluster (if at all), and thus it was eliminated from further analyses. As with the socioeconomic inputs, and again for the reason of removing influences created by the different metrics on which they were measured, a composite variable for the socioeconomic outputs was created by summing the standard scores of each of the remaining four variables on an annual basis. Autocorrelations and partial autocorrelations for this composite variable are presented Table 22. As with the socioeconomic inputs, the results listed on Table 22, particularly the large but systematically declining autocorrelation as lag increases, suggests strongly trended data, in fact, even more strongly trended than was true of the composite measure of the socioeconomic inputs. Of particular interest is the significant and negative lag 6 partial, which indicates a six year periodicity in the data. This is completely unanticipated by the data and discussion which preceded this analysis: it 1" so an QOU'IIFwNI-t 116 Table 22 Overall analysis Autocorrelations and Partial Autocorrelations Socioeconomic Outputs Partial Autocorrelation Autocorrelation .988*** .988*** .973**"‘ -.132 .961*** -.150 .948*** -.048 929*" -.113 .901*** -.346* 872*" -.253 .841*** .164 *=p<.05 **=p<.01 ***=p<.001 [883839 888 117 does not correspond to the four year election cycle, or the five year cycle found in the administration’s agenda. It is not even quite half of the fourteen-to-sixteen year cycle in ideological behavior observed by Schlesinger (1939) and discussed above. In the absence of a coherent explanation for this observation, the most that can be done is to account for it in subsequent analyses. Path analysis - overall model To proceed with the path analysis of the overall model, the correlations of the public agenda with the media, administration, and congressional agendas, and the socioeconomic inputs (propinquity and economic inputs) were calculated for lags of 1-8 years. These correlations are presented as Table 23. Although only the economic inputs are synchronously correlated with the public agenda, examination of Table 23 reveals that each of the variables is so correlated at one or more lag times. The media agenda is significantly correlated with public opinion five years later, the congressional agenda and propinquity at lags of one to seven years, economic inputs at lags of zero to six years, and the administration agenda at lags of four and five years. The spectral pattern Irwzl < lrfidl < lrul > lrtrml > Iru+2| is associated with the association of one variable with another at lag t-l (Ostrom, 1978, Krull and Paulson, 1977). This pattern appears in several places in the matrix, leading to the inclusion of the following variables in the path model: public opinion at t-l (from Table 10); media agenda at t-4; congressional agenda at t-2 and t-6; administration agenda at t—5; L" on on CDQODO'IhODNI-ic 118 Table 23 Overall analysis Legged Correlations with Public Opinion Variable Media Congressional Administration Propinquity Economic Agenda Agenda Agenda Inputs .130 -.108 .054 .227 -.288* .055 -.320* -.037 .279* -.312* .108 -.531*** .143 .342“ -.290* -.043 -.361** .114 .328* -.291"‘ -.252 -.294* .360" .297 * -.368** -.039 -.341** .448*** .278* -.325* -. 134 -.444*** .220 .256* -.262* -. 162 -.330* .211 264* -.100 -.002 -.233 .069 .178 -.028 * = p < .05 ** = p < .01 **"‘ =p < .001 119 pr0pinquity at t-2; and economic enputs at t-4. The intercorrelation matrix of these variables and public opinion at t-0 appears as Table 24. Using the methods of Kenny (1979) and Blalock (1971), a path model was constructed from the correlations in Table 24. This model showed that the main predictors of public opinion at t-O were the congressional agenda at t-2, the administration agenda at t-5, and public opinion at t-l. Residual analysis, however, indicated that there was a significant amount of correlation between error terms: because of the presence of lagged endogenous variables (public opinion at t-O and t-l), Durbin’s h was used (Durbin, 1970; per Ostrom, 1978), yielding h=2.08, p<.01. As pointed out in the previous chapter, this precludes use of OLS estimation. Instead, again per Ostrom (1978), the amount of correlation among errors was estimated by the Instrumental Variable-Generalized Least Squares (IV-GLS) technique as in Johnston (1970). Using the congressional agenda as an instrumental variable, the amount of correlation between error terms was estimated (to be p=.595), and this figure used to correct the original data for the correlation, as required by the GLS technique. The corrected data yielded a corrected correlation matrix, which appears as Table 25. One consequence of the correction procedure is that many of the correlations are smaller in magnitude in Table 25, when compared to Table 24. This is to be expected, as the uncorrected correlations include components of both “true” relationship plus additionally a component of the correlation among error terms. More importantly, the correction of the correlation matrix has the effect of removing public opinion at t-1 from the analysis: the corrected correlations between it and the other variables range from -.134 to .187 (and 12) Table 24 Overall analysis Original Matrix of Correlations for Lagged Predictors Public Public Congress Congress Propin- Economic Media Agenda Agenda Agenda Agenda quity Inputs Agenda w t-l t-2 w t—2 s4 t,4 Public Agenda t-1 .384“ Congress Agenda t-2 -.531*** -.320* Congress en t- -. -. . Ag da 6 444'" 341* 450" Propin- quitytrZ .342* 279* -.586*** -.626*** Economic. Inputst-4 -.368* -.291* .427" .606*** -.920*** Media en a t- -. -. . . -. -, Ag d 4 252* 043 292* 029 017 050 Administ. Agenda t-5 .448" .360“ -.445** -.329* .576*** -.630*** .098 *=p<.05 **=p<.01 ***=p<.001 Public Agenda t-l Congress Agenda t-2 Congress Agenda t-6 Propin- quity t-2 Economic. Inputs t-4 Media Agenda t-4 Administ. Agenda t-5 121 Table 25 Overall analysis Matrix of Corrected Correlations for Lagged Predictors, Public Public Congress Congress Pr0pin- Economic Media Agenda Agenda Agenda Agenda quity Inputs Agenda t—O t-l t-2 t-6 t-2 t-4 t-4 .029 -.406** -.078 -.283* -.065 .281* .193 .104 -.311* -.379* -.242 -.134 .146 .399* -.790*** -.235 .083 .273* -.008 .004 -.082 .339* .187 -.352* -. 193 .185 -.307* -. 130 * = p < .05 ** = p < .01 *"‘* = p < .001 m all are insignificant), and the sum of the correlations, .126, and average correlation, .012, with the other variables suggest that the relationship between public opinion at t- 1 and the other variables is largely due to sampling error alone. Thus, public opinion (at t—1) was eliminated from the analysis. Subsequent path analysis yielded the path model presented as Figure 10. The direct causes of public opinion at t-0 are the administration’s agenda at t-5 and the congress’ agenda at t-2. Under this model the other variables correlated with public opinion are so because of their relationship to these two variables. This model shows no significant correlation among error terms (Durbin-Watson d=2.217, p>.05). In accord with the methods of Hunter (1977), the path coefficients were used to “regenerate” a correlation matrix (Table 26). Comparing this regenerated matrix to the original matrix yields a residual matrix, which appears as Table 27. None of the residuals are of sufficient size to be statistically significant, indicating that there is no major misspecification in this model, i.e. the model provides a good fit to the data. The media agenda at t-4 in particular is both negatively and indirectly related to public opinion, because of its effect on the congressional agenda at t-2. This is particularly interesting in view of the facts that 1) members of the House are elected on a biannual basis, and 2) Arnold (1990) has shown that congressional candidates prepare themselves and run on the issues of the previous campaign. This model suggests that two years after the congress responds to the media, the people respond to the congress, including by voting, leading to the sequence 123 v.» aesagxaaxaz m5. at. o.“ :8» 89%... new 82»: < SN. 8 < oo 83..- nr_aasw<. ca =ou§m§< SEED 255 mam. pom. mam. Na v-3 .039—EBA 3:9: 8828 can..- Figure 10 Overall Analysis -- Final Path Model Public Agenda t-l Congress Agenda t-2 Congress Agenda t-6 Propin- quity t-2 Economic. Inputs t-4 Media Agenda t-4 Administ. Agenda t-5 124 Table 26 Overall analysis Matrix of Reconstructed Correlations Public Public Congress Congress Propin- Economic Media Agenda Agenda Agenda Agenda quity Inputs Agenda t-O t-l t-2 t-6 t-2 t-4 t-4 X -.406* X -.187 X .281* .118 X -.122 -.325 -.150 X .112 .399“ -.790* -.142 X .273“ .077 -.013 .017 .339* X -.138 -.172 .243 -.307* -.038 * = fixed by specification of model X = variable eliminated from model Public Agenda t-1 Congress Agenda t—2 Congress Agenda t-6 Propin- quity t-2 Economic. Inputs t-4 Media Agenda t-4 Administ. Agenda t-5 125 Table 27 Overall analysis Matrix of Residuals Public Public Congress Congress Propin- Economic Media Agenda Agenda Agenda Agenda quity Inputs Agenda t-O t-l t-2 t-6 t-2 t4 t-4 X .000* X .096 X .000'“ -.075 X .189 .054 .092 X -.034 .000“ .000'“ .093 X .000“ .085 -.017 .098 .000* X .203 .032 .058 .000* .092 * = fixed by specification of model X = variable eliminated from model 1% Campaign 1: Media raises issue Campaign 2 (two years later): Congressional candidates respond to issue I Campaign 3 (four years total): Public responds to candidates’ response. The finding that the administration agenda at t-5 impacts public opinion is also interesting in terms of the significant five year partial autocorrelation in the administration’s agenda. It, too, seems to imply the existence of a five-year planning cycle. The final noteworthy element of this model is that, despite being substantially correlated with public opinion, the model does not provide any evidence of a causal relationship between either of the external socioeconomic factors and public opinion. Rather, the model states that those correlations are strictly due to the fact that public opinion and the external factors are both caused by administration and congressional action. In short, the relationship between the external factors and public opinion “partials out.” Impact on output variables Analysis of the impact of the various agendas on socioeconomic outputs was based on the desire, elucidated in the previous chapter, to determine whether shifts in public opinion resolve themselves into overt behavior. Toward that end, the analytic procedures used above were repeated, this time in an attempt to create a causal model predicting impacts on the outputs rather than public opinion, using public opinion as one of the predictors. With the univariate analyses already complete, the lagged correlations of public opinion, the media, congressional, and administrative agendas, propinquity, and the socioeconomic inputs with 127 the socioeconomic outputs were calculated. They are presented in Table 28. Once again the spectral pattern described above is evident in several places, and so the media agenda at t-4, the congressional agenda at t-6 and the administration agenda at t-3 were used in the subsequent modeling process. Several of the variables show their strongest association with the output variables at the t-O (synchronous) time period (with correlations declining afterward), and so the congressional agenda, propinquity, and socioeconomic inputs at t-0 were included in the model. Although none of the lagged correlations with public opinion were statistically significant, they also show the characteristic spectral pattern centered around the t-2 lag. As the particular point of this analysis was to examine the role of public opinion in predicting the composite socioeconomic output variable, public opinion at lag t-2 was also included in the analysis. Finally, after examining Table 22, autocorrelations of socioeconomic outputs at t-1 and t-6 were included in the modeling process. Inclusion of these variables in the modeling process led to the intercorrelation matrix presented as Table 29. This correlation matrix was subjected to path analysis in the manner described above, yielding the path model attached as Figure 11. The socioeconomic outputs at t-O variable was regressed onto the four variables predicted by the models to have causal priority on it in order to test for correlated errors, and a non-significant test result was obtained (h = - 0.64, p > .05), suggesting that the assumptions of the model with regard to error terms were not in jeopardy. The predictions of this model lead to the regenerated correlation matrix presented as Table 30. Subtracting the elements of Table 29 from 1" so an QQGO'IAOONHO - 128 Table 28 Overall analysis: Output model Lagged Correlations with Socioeconomic Outputs Variable Media Congressional Administration Propinquity Agenda Agenda Agenda -.049 .564*** -.580*** 5331*" -124 503*“ ”594*” -.800*** -.164 456""II -.611*** -.779*** n287* .411** a629*** a772*** “.348" .407" -.629*** -.765*** -.340** .408” -.606*** -.762*** ‘5307* .440** a591*** .5759*** -.317* .391" -.599*** -.767*** -.247 .380“ -.600*** _.774*** * = p < .05 ** = p < .01 **"‘ = p < .001 Economic Inputs .826*** .817*** .814*** .816*** .813*** .808*** .807*** .822*** .830*** Public Agenda -.196 -.202 -.228 -.202 -.185 -.125 -.071 -.130 -.173 129 Table 29 Overall analysis: Output model Matrix of Correlations for Lagged Predictors Soc-econ Soc-econ Soc-econ Cong Cong Admin Econ Public Outputs Outputs Outputs Agenda Agenda Agenda Prop Inputs Agenda t-O t-1 t-6 t-O t-6 t-3 t-0 t-0 t-2 Soc-econ Outputs t—l .9ss* Soc-econ Outputst—G .970 .983'“ Cong Agenda t-0 .564* .450 .553 Cong Agenda t-6 .590 .568 .564* .238* Administ. Agenda t-3 -.629* -.621 -.610 -.286 .371 Propinq. t-0-.890 -.901 -.917" -.508 -.430 .560 Economic Inputst-O .838 .848 .863 .478 .405 -.527 -.942* Public Agenda t—2 -.334 -.348 .345“ -.l96 -.166 .217 .325 -.306 Media Agenda t-4 -.348* -.343 -.337 -.158 -.205 .219 .309 -.291 .117 * = fixed by specification of model 130 w.“ Ob cow< 829.6 \ mam. 8... mum; on we”; 8&5 N.“ 88.5 08.5 mam. mm”: fl -3 398 83. o-“ 3:9: 2:88 «3.- Sm: Ora b38395 8=om< mgcoo c.» 88? «an: mow. f 89.5 Figure 11 Overall Analysis: Output Model Final Path Model Soc-econ Soc-econ Soc-econ Cong 131 Table 30 Overall analysis: Output model Matrix of Reconstructed Correlations Cong Admin Econ Public Outputs Outputs Outputs Agenda Agenda Agenda Prop Inputs Agenda t-0 Soc-econ Outputst-l .988*** Soc-econ Outputst-6 .901*** Cong Agenda t-0 .564*** Cong Agenda t-6 .440*** Administ. Agenda t-3 -.629*** Propinq. t-0-.831*** Economic Inputst-O .826*** Public Agenda t—2 -.228 Media Agenda t-4 -.348** trl .983*** .572*** .408" -.611*** -.861*** .840*** -.203 -.287* W W w as so w .533*** .564*** .233 -.476***-.445*** .361" -.917***-.586***-.592*** .576*** .s7o*** .500*** .661*** -.57s***-.942*** .345" .064 -.294"I -.037 .284"' -.320* -.010 .073 .029 .213 .026 -. 112 * = p < .05 *"' = p < .01 ***=p<.001 t-2 . 108 132 those of Table 30 leads to the residual matrix of the analysis, presented as Table 31. Unlike the other path analyses presented in this chapter, Table 31 contains several statistically significant elements, indicating that the model is misspecified in one or more locations. However, as Kenny (1979) has pointed out, it is possible for a path model to be misspecified in certain elements and yet correctly specified in others; that is, that because certain parts of the model contain misspecifications does not necessarily invalidate the rest of the model.‘ Since the part of the model of particular interest is the impact of public opinion, if any, on the socioeconomic output variable, and since the misspecification does not include any of the combinations of those variables, we can have confidence that the relevant part of the model is correctly specified. The model presented in Figure 11 indicates that it is more likely that public opinion is affected by the variables designated as socioeconomic outputs than the reverse. Although that section of the model can be constructed with a link from public opinion to outputs at t-O, such a construction would not fit the data as well as the model as presently constructed. In addition, the link fi'om opinion to output would be statistically insignificant. Sub-analysis: Foreign Affairs Analysis of the causes of public opinion on the sub-area of Foreign Afi‘airs is simplified as compared to the overall analysis due to the lack of measurable socioeconomic inputs and outputs. Propinquity could be ‘Indeed, this is one of the primary advantages of using Least Squares techniques rather than the more efficient Simultaneous Equations techniques, which distribute the errors in parameter estimation created by local misspecification across all the parameter estimates. t-0 Soc-econ Outputs t-l .000* Soc-econ Outputs t-6 .006 Cong Agenda t-0 .000* Cong Agenda t-6 .149 Administ. Agenda t-3 .000“ Propinq. t-0-.058 Economic Inputs t-0 .011 Public Agenda t-2 -.116 Media 133 Table 31 Overall analysis: Output model Matrix of Residuals Soc-econ Soc-econ Soc-econ Cong Cong Admin Econ Public Outputs Outputs Outputs Agenda Agenda Agenda Prop Inputs Agenda t-1 t-6 t—0 ‘ t-6 t—3 t—0 t0 t2 .000* -.122 .021 .160 .000* .000* -.010 -.134 .159 -.010 -.040 .000“ .079 .162 -.017 .009 - 007 -.023 -.256T .052 .000* -.146 .000* -.133 .127 .2531' .042 .014 -.056 -.3271’ -.231 -.234 .006 .2841' -.180 .009 Agenda t-4 .000* * = fixed by specification of model 1' = statistically significant residual 134 estimated using tourism and immigration figures, but these are only estimates and are contaminated by many outside influences. Worldwide economic input conditions are also poorly measured.I5 Outputs are also problematic; indeed, it borders on illegal for private US citizens to engage in foreign policy initiatives, prominent examples (Armand Hammer, Ross Perot, Jane Fonda) nonwithstanding. In the absence of external socioeconomic data, the basic model (Figure 2) devolves to the four agenda areas: public, media, administration, and congressional. Univariate analyses Autocorrelations and partial autocorrelations for the aggregate of the public opinion items belonging to the content area of Foreign Affairs are presented in Table 32. Few of the results are significant; most notable is one of the results that is not: public opinion on Foreign Affairs issues is not a lag one process, indicating that it is fairly unstable from year to year, and possesses little “momentum” of its own. The significant lag 5 partial autocorrelation indicates that the public opinion on Foreign Affairs does possess a five year cycle. Unlike significant five year lags in the administration’s agenda, which are sensible in terms of rational planning processes, the reason for this occurrence is unclear. The autoregression analysis of the media agenda on Foreign Affairs is presented in Table 33. This table shows results similar to those of Table 32: the only significant partial autocorrelation is at lag 5 and negative, again indicating a five year cycle, this time in the media’s agenda. It is possible that these cycles have a similar genesis -- news reporters are also “And in Marxist-Leninist nations, are subject to falsification for the purpose of serving revolutionary ideology (c.f. Heinlein, 1960/1980) L“ no on QNIO>UIUFOONH 135 Table 32 Sub-Analysis: Foreign Affairs Autocorrelations and Partial Autocorrelations Public Opinion Partial Autocorrelation Autocorrelation -.029 -.029 .191 .190 .208 .227 -.134 -.168 -.212 -.356** 050 .044 -.321* -.134 - 244 -.249 * = p < .05 ** = p < .01 *** = p < .001 r‘ m on mQOECJ'IIhODNI-l 136 Table 33 Sub-Analysis: Foreign Affairs Autocorrelations and Partial Autocorrelations Media Agenda Partial Autocorrelation Autocorrelation .030 .030 .128 .127 .135 .130 -.199 -.228 -.252* -.277* -.129 -.100 -.336* -.254 .014 .084 *=p<05 ** = p < .01 *** =p < .001 8883885 8 137 generally citizens of the nation in which they reside, and so this may be a manifestation of the media’s opinions as citizens, so matching the public’s opinion. Table 34 contains the autoregressive analysis of the administration’s agenda. As with the previous, there is apparently little stability in this, although there is a significant eight year cycle in the data. As pointed out earlier, the terms of specific Presidents have rarely adhered to the four year election cycle, but it is also worthwhile noting, as was done previously, that party control of the White House has been following very closely to an eight year cycle, to wit: Roosevelt/Human, 1944-1952; Eisenhower, 1952-1960; Kennedy/Johnson, 1960-1968; Nixon/Ford, 1968-1976. The only exception to the eight-eight pattern is Carter, who failed of reelection in 1980; but if one assumes that Reagan serves one of Carter’s terms, the eight-eight pattern continues to hold when Bush fails of reelection in 1992. Since the two parties often pursue different foreign policies, and since they flip-flop in control, is is possible that this eight year cycle represents control of the White House. Finally, the autoregressive analysis of Congressional activity is presented in Table 35. It is shown to be a lag one process, with activity in one year predicted by the previous year’s activity.“ Path analysis - Foreign Amairs model As with the overall model, the lagged correlations up to t-8 of the media, administration and congressional agendas with public opinion were calculated, and the results presented as Table 36. Although fewer of the correlations are significant (even in those three arenas alone) in Table 36 than in Table 23, they show the same spectral pattern, and so in the ' The partial of -.268 at lag 7 fails to be significant due to the number of degrees of freedom consumed by a seventh order partial. r; m an 0040:0145me 138 Table 34 Sub-Analysis: Foreign Affairs Autocorrelations and Partial Autocorrelations Administration Agenda Partial Autocorrelation Autocorrelation .223 .223 .086 .038 .034 .007 .017 .006 .080 .077 .014 -.021 .026 .021 -.277* -.305* I"=p<.05 **=p<.01 ***=p<.001 88588 8 L" m on 00403015me 139 Table 35 Sub-Analysis: Foreign Affairs Autocorrelations and Partial Autocorrelations Congressional Agenda Partial Autocorrelation Autocorrelation .359” .359“ .148 .022 .073 .016 .067 .038 .171 .152 .103 -.011 -.175 -.268 -.164 -.045 *=p<.05 **=p<.01 *** =p < .001 {858385 t L" no on QOCfinhODNl-‘O 140 Table 36 Sub-Analysis: Foreign Affairs Lagged Correlations with Public Opinion Variable Media Congressional Administration Agenda Agenda Agenda -.087 -.042 -.086 -.170 -.188 -.207 .062 -.192 -.315* -.132 -.127 -.172 -.104 -.087 -.297* -.249* -.015 -.134 -.020 .037 .016 -.188 246* -.202 -.013 .251* -.024 * = p < .05 ** = p < .01 *** = p < .001 141 same way the following variables were selected for construction of a path model: media agenda at lag 5; congressional agenda at lag 8; administration agenda at lags 2 and 4, and (from Table 32) public opinion at lag 5. An intercorrelation matrix was calculated, and the results presented as Table 37. Path analysis of this correlation matrix started in the same way as that performed for the overall analysis, by the methods cited above. However, the test for correlated errors, in this case an estimate created by adding the lagged residuals into the regression analysis (Ostrom, 1978)", yielded insignificant results (t = .626, p > .05), indicating no significant correlation among error terms and obviating the need for IV-GLS estimation. Hence, OLS estimation was used. OLS estimation of the model led to the model shown as Figure 12. In this model, changes in public opinion are created by the administration’s activity both two and four years previously; in addition, the five year cycle in public opinion continues. As in the overall model, the media’s agenda (four years previously) contributes only indirectly, through its influence on the the administration’s agenda at lag 2. (Also, as with the overall model, note that the impact of the media’s agenda on public opinion is negative in sign.) Finally, the media agenda at t-4 is shown to be in part created by (or a response to) the congressional agenda four years previously. As with the overall model, the path coefficients from this model were used to create a “reconstructed” correlation matrix (Table 38) as a part of a ' test of the goodness of fit of the model. The observed correlations were subtracted from the regenerated correlations to create a matrix of residuals, presented in Table 39. Residuals range from -.205 to .128; as with ’ Durbin’s h was calculated first. Under certain conditions, however, the denominator of the Durbin h statistic can be negative, yielding an imaginary b (that is, h would contain the square root of -1). This was the case in this analysis, precluding the use of h. Public Agenda t-5 Administration Agenda t-2 Administration Agenda t-4 Media Agenda t-5 Congressional Agenda t-8 142 Table 37 Sub-Analysis: Foreign Affairs Matrix of Correlations for Legged Predictors Public Agenda t-0 -.231 a312* -.297* -.249* .251* Public Admin Agenda Agenda 65 t-2 .005 -.059 .086 -.087 .398" -.127 -.166 * = p < .05 ** = p < .01 *** =p < .001 Admin Media Agenda Agenda t-4 t5 . 112 .011 -.367** 143 o-“ 838 325 8a.- Nam..- «A 883 gou3n§< wmm. v-“ 38?. £32 uwmr wé 88m< 82w: . Figure 12 Foreign Affairs Sub-analysis Final Path Model 144 Table 38 Sub-Analysis: Foreign Affairs Matrix of Reconstructed Correlations Public Public Admin Admin Media Agenda Agenda Agenda Agenda Agenda no t-5 t-2 t4 t5 Public Agenda t-5 -.231* Administration Agenda t-2 -.312* .072 Administration Agenda t-4 -.297* .069 .093 Media Agenda t-5 -.124 .029 .398* .037 Congressional Agenda t-8 .046 -.011 -.146 -.014 -.367* * = fixed by specification of model 145 Table 39 Sub-Analysis: Foreign Affairs Matrix of Residuals Public Public Admin Agenda Agenda Agenda t-O t5 t-2 Public Agenda t-5 .000* Administration Agenda t—2 .000* .067 Administration Agenda t-4 .000* .128 .007 Media Agenda t-5 .125 .116 .000* Congressional Agenda t-8 -.205 .116 .020 * = fixed by specification of model Admin Media Agenda Agenda t-4 tr5 -.075 -.025 .000'“ 146 the overall model, none of the correlations is significant, suggesting that the model is not grossly misspecificed in any particular, and so provides a good “fit” to the data. Sub-analysis: Race and Ethnicity As with the overall analysis, the Race and Ethnicity sub-analysis included not only public opinion and the media, congressional and administrative agendas, but also external socioeconomic variables. Propinquity was specifically measured for black workers as an input variable; election of minority congressmen was used as output. As discussed previously, propinquity is the simple idea that the more contact people have, the more they like each other; this sort of contact can be readily measured in the workplace in particular. Univariate analysis Contained in Table 40 is the autoregression analysis of the behavior of aggregate public opinion on items classified in the content area of Race and Ethnicity. The partial autocorrelations show that the data are strongly trended, with a substantial Lag 1 autocorrelation, and smaller four and seven year cycles. Again, the four year cycle may be associated with the nation’s electoral cycle. Unlike the public agenda data, there appears to be little or no trend in the media agenda data, as shown in Table 41. None of the partial autocorrelations are statistically significant, although two (.262 at lag 5 and -.267 at lag 6) approach it closely (t=1.63, df=41 at lag 5; t=1.62, df=40 at lag 6). The administration’s agenda, like the public agenda, shows t" m an ooqozcnuhcorcs-a 147 Table 40 Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Public Opinion Partial Autocorrelation Autocorrelation 536*“ 536*“ .350" .102 .170 -.039 -.168 -.312* -.184 -.017 -.320* -.197 -.509*** -.348* -.291* .087 *=p<.05 **=p<.01 ***=p<.001 {-1 m N mfiwmtht-t 148 Table 41 Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Media Agenda Partial Autocorrelation Autocorrelation .053 .053 .060 .057 .094 .088 -.047 -.059 .260“ .262 -.206 -.267 -.142 -.128 .090 .107 *=p<.05 **=p<.01 ***=p<.001 [883388 8 149 substantial year to year consistency, as demonstrated by by the large lag 1 autocorrelation in it on Table 42; there is also a significant, positive lag 2 correlation as well. This is probably not unexpected; affirmative action as an issue has been a strong component of the agendas of Presidents 'Ituman, Eisenhower, Kennedy, Johnson, and Reagan; in general it was necessary for each of them to repeatedly ask for the legislation they wanted before it was passed by Congress. Table 43 indicates that the year-to-year stability of the administration’s agenda is not matched by congressional activity. The only significant partial autocorrelation is again the lag four partial, once again suggesting that congressional activity in this area is tied to the Presidential election cycle. The controversial nature of racial politics makes legislators leery of election year activity; it seems more likely that activity takes place in the second or third year of the cycle. 'Itends in occupational segregation (the overt measure used to represent contact between the races, or propinquity) are by their nature evolutionary in nature, with the consequence that it is strongly trended, and this is clearly shown by the lag one autocorrelation of .979 in Table 44. In addition, there is a substantial four year cycle in the data, and it is again within reason to suggest that this is political in nature and tied to the electoral cycle. There are significant partials at lags 6 through 8.8 Similarly, the socioeconomic output variable, the election of black and Hispanic congresspersons, as shown in Table 45, is also evolutionary in nature, and thus also expected to be strongly trended. Further, as as congresspeople serve multiyear terms, in 50% of the years (the election year), the data should be identical to that of the previous year (that ' It is within the realm of possibility that they are the result of rounding error on the higher order partials. Lag mQOQOIrFODNt-t 150 Table 42 Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Administration Agenda Partial Autocorrelation Autocorrelation .473*** .473*** .528*** .392" .383“ .069 .415” .131 .432" .192 .341* -.020 .226 -.175 .310* .129 *=p<.05 **=p<.01 ***=p<.001 8838': 88 r' m an mQO>UII§ODNH 151 Table 43 Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Congressional Agenda Partial Autocorrelation Autocorrelation .168 .168 .161 .137 .222 184 .361“ .309* .020 -.114 .243 .169 .090 -.072 .196 -.095 I"=p<.05 **=p<.01 ***=p<.001 5:88:38” 83 152 Table 44 Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Socioeconomic Inputs: Occupational Segregation by Race Partial Lag Autocorrelation Autocorrelation 1 .979*** .979”** 2 .963*** .110 3 .946*** 4.018 4 .907 *** -.545*** 5 .876*** .059 6 .851*** .301* 7 .822*** 296* 8 816*” .654*** *=p<.05 **=p<.01 ***=p<.001 £18 [88388 Lag QN‘IQUIIGBCON 153 Table 45 Sub-Analysis: Race and Ethnicity Autocorrelations and Partial Autocorrelations Socioeconomic Outputs: Election of Black Congressmen Partial Autocorrelation Autocorrelation .990*** .990*** .984*** .191 .969*** -.427** .960*** .203 .941*** -.371* 927*" .051 .912*** .440“ .904*** .069 *=p<.05 **=p<.01 ***=p<.001 88:88: 88 154 representing the first year of the term for a representative). The pattern of significant partial every two years (lags 1, 3, 5, and 7) is probably a consequence of the alternate year election sequence. Path Analysis - Race and Ethnicity model Again, the lagged correlations of the media, congressional, and administrative agenda and the socioeconomic input (propinquity) with public opinion on the topic of Race and Ethnicity were calculated. These are presented as Table 46. As before, the table was examined for the characteristic spectral pattern, which was revealed by several of the variables at a number of lags. Selected for inclusion in the modeling process were: media agenda (lag 6), administration agenda at lags 2 and 6, and the economic inputs at lags 0 and 6. None of the lags of the congressional agenda reached statistical significance. Again, the controversial nature of race relations tends to preclude legislative activity with regard to them. For instance, one reason why, as pointed out above, 'II'uman, Eisenhower, and Kennedy made repeated requests for legislation was that the reforms they sought were filibustered by the “Dixiecrats” — legislators from southern states. The scoring system for congressional activity, based on legislation voted on, does not reflect these activities. The variables selected above were then correlated. The results are presented as Table 47. Following this, in the same manner with the previous models, the Race and Ethnicity data was path analyzed, leading to the model presented as Figure 13. On the topic of Race and Ethnicity, public opinion is found to be formed as a consequence of the activities of the administration and the media. as well as the behavior of opinion in t“ m on mQQCfiIFODNt-‘O 155 Table 46 Sub-Analysis: Race and Ethnicity Lagged Correlations with Public Opinion Variable Media Congressional Administration Agenda Agenda Agenda .025 -.152 254* .098 -.120 .088 .107 -.146 .316* .143 -.185 .156 .060 -.014 .044 .032 -.006 .235 .301* .202 .225 .088 -.046 .247* .041 .186 .281* *=p<05 ** = p < .01 *** = p < .001 Economic Inputs .284* .191 .125 .092 .204 .281* .345“ .341" .319* Public Agenda t-1 Public Agenda t-4 Public Agenda t-7 Administ. Agenda t-2 Administ. Agenda t-8 Media Agenda t-6 Economic Inputs t-0 Economic Inputs t-6 156 Table 47 Sub-Analysis: Race and Ethnicity Matrix of Correlations for Lagged Predictors, Admin Admin Media Econ Agenda Agenda Agenda Agenda Agenda Agenda Agenda Inputs Public Public Public Public to tr]. t-4 tr7 .347M .157 .359" -.476*** -.271 . 110 .397“ .103 .039 -.039 .281* .247 * .044 .088 .301* .032 .107 -.020 .292* .260* .356" .136 .231 .228 .250* .273* * = p < .05 ** = p < .01 *** = p < .001 t—2 t-8 t—6 W .341** .170 .010 .601*** .180 -.008 .707*** .593*** -.024 .746*** 157 T. V.“ 83.5 one: man. 8an 38a Ora new. 36?. as»: o.» How. h.“ 8390 239m mam: sou—E0 0:25 gm. mum. 8mm. Na 881‘ w.“ 88m< V. cougm§< no». man. as or 935 mm was. 9&5 mm Engage. Figure 13 Race and Ethnicity Sub-analysis Final Path Model 158 previous time periods. Although propinquity was correlated with the public agenda at synchronous time periods, this correlation is shown to be spurious. Regressing the public agenda onto its immediate causes yields an insignificant Durbin’s h (h=-.667, p>.05), indicating that error terms are not significantly correlated. As with the previous models, the coefficients in the Race and Ethnicity model were used to create a regenerated correlation matrix (Table 48), from which was subtracted the original matrix, yielding the residual matrix (Table 49). Residuals range from -.189 to .232, but none of them is statistically significant, indicating that the model is not grossly misspecified in any of its particulars. Comparison of Path Models Prima facia, it seems clear that, apart from the constraint that they attempt to provide some form of causal analysis of how public opinion changes, the overall, foreign affairs, and race and ethnicity models differ substantially. As noted above in their respective sections, one model contained a problem with correlated error terms while the other two did not; in two of the models the media agenda impacted public opinion indirectly while in the third its impact was direct; the administration’s and Congress’ agendas interact differently in each model; and so on. To formally test the extent to which the models differ, Table 50 was constructed. It consists differences in the lagged correlations of the media agenda, administration agenda and congressional agenda9 with public opinion between each of the models, calculated on a pair-wise basis (that is, the overall model was compared to the foreign affairs model and then the ’ These three variables, of course, are the three which appear in all three of the models. 159 Table 48 Sub-Analysis: Race and Ethnicity Matrix of Reconstructed Correlations Public Public Public Public Admin Admin Media Econ Agenda Agenda Agenda Agenda Agenda Agenda Agenda Inputs t-O t-1 t-4 t-7 t-2 t-8 tr6 t-O Public Agenda t-l .347* Public Agenda t-4 .125 .359* Public Agenda t-7 -.476* -.186 -.067 Administ. Agenda t-2 .397* .138 .049 .193 Administ. Agenda t-8 .166 .058 .157 .162 .419 Media Agenda t-6 .301* .104 .037 -.161 .119 .050 Economic Inputs t-0 .254 .128 356* .204 .527 . 180* .013 Economic Inputs t—6 .281 .097 .266 .273“ .7 07* .593* .084 .746* * = fixed by specification of model 16) Table 49 Sub-Analysis: Race and Ethnicity Matrix of Residuals Public Public Public Public Admin Admin Media Econ Agenda Agenda Agenda Agenda Agenda Agenda Agenda Inputs t-O t-l t-4 t-7 t-2 t-8 t-6 t-0 Public Agenda t-l .000* Public Agenda t-4 -.032 .000* Public Agenda t-7 .000* .085 -.167 Administ. Agenda t-2 .000* .035 .010 .232 Administ. Agenda t-8 -.115 -.189 .113 .074 .078 Media Agenda t-6 .000* .072 -.070 -.141 -.051 .040 Economic Inputs t-O -.038 -. 132 .000* .068 -.074 .000* .021 Economic Inputs tr6 .050 -.131 .016 .000* .000* .000* .108 .000* * = fixed by specification of model Lag 161 Table 50 Residual matrix Comparison of path models Overall vs. Overall vs. Foreign Affairs vs. Foreign Affairs Race and Ethnicity Race and Ethnicity Media Cong Admin Media Cong Admin Media Cong Admin Agenda Agenda Agenda Agenda Agenda Agenda Agenda Agenda Agenda .217 -.066 .140 .105 -.050 -.200 -. 112 .116 -.340** .225 -.137 .170 -.043 -.200 -.125 -.268* -.063 -.295* .046 -.339** .458*** .001 -.385** -.173 -.045 -.046 -.631*** .089 -.234 .286* -.186 -.176 -.042 -.275* .058 -.328* -.148 -.207 .657*** -.312* -.280* .316* -.164 -.073 -.341** .210 -.236** .582*** -.071 -.335* .213 -.281* -.009 -.369** -.114 -.481*** .204 -.435***-.646***-.005 -.321** -.165 -.209 .026 -.576*** .413” -.250* -.376** -.036 -.276* .200 -.449*** .011 -.484*** .093 -.043 -.419*** -.212 -.054 .065 -.305* *=p<.05 **=p<.01 ***=p<.001 162 race and ethnicity model, and then the latter two models compared to each other). As Table 50 demonstrates, there are numerous significant differences between the data sets from which the models were constructed. There are 27 correlations (lags 0 through 8 by three variables) in each of the model-by-model comparisons: in each comparison at least 10 significant differences appear between data sets.” Only one significant difference per per comparison would be expected if the difference in the models was due to chance alone. Overall findings As discussed in the first chapter, the literature review preceding this study and subsequent analyses led to the formulation of a number of expectations, as enumerated on pages 52-53 . 1) The media will show a direct impact on public liberalism. This expectation is not borne out in two of the three relevant models constructed. Only in the Race and Ethnicity sub-analysis does the media agenda have a direct impact on public opinion. In the other two models, the media agenda instead demonstrates indirect...and negative in sign...impacts on the public opinion via the mechanism of impacting on one element or the other of the policy agenda. 2) The policy agenda will show a direct impact on public liberalism. This expectation is borne out with the caveat that the notion of a policy agenda has been demonstrated to be more complex than previously formulated. In each of the three models one of either the congressional ‘° Overall vs. Foreign Affairs: 10 significant differences; Overall vs. Race and Ethnicity: 10 significant differences; Foreign Affairs vs. Race and Ethnicity: 13 significant differences. 163 agenda or the administration agenda has a direct impact on public opinion; in the overall model both elements have such impacts. The only model in which both elements of the policy agenda do not appear is the Race and Ethnicity sub-analysis. It is worthwhile noting that in the two models of which both the administrative and the congressional agendas are a part, when the consequence of one of those variables is positive in sign, the consequence of the other is negative in sign. 3) The media will show a direct impact on the policy agenda. As mentioned above, the media agenda’s impact on public opinion works in two cases out of three through one element or the other of the policy agenda. In the overall model, the media agenda directly impacts the congressional agenda two years consequently (bearing out Arnold’s, 1990, assertion); in the Foreign Affairs model the two year impact lag remains, but the impact of the media is on the administration’s agenda. In the Race and Ethnicity model, the media and the administration’s agenda are only spuriously correlated. 4) The increase in public liberalism will be followed by an increase in liberal activity in the nation. This does not appear to be the case. In fact, the models suggest that the socioeconomic input and output variables, as well as propinquity, have little causal role, either as predictor or criterion, in the formation of public opinion. In the only model in which the public agenda is shown to be related to an external socioeconomic factor, it is the expected outputs which impact the public agenda (in the overall outputs model). This is clearly counterintuitive. 5) Separate models of the formation of public opinion will differ from 164 issue to issue. This does appear to be the case. As pointed out above, on an issue-by- issue basis substantially different models emerge from the analysis of the same set of variables. Chapter Four Discussion Evident in the nature and design of this study, although not explicit until now, is that it is primarily exploratory in nature. As Rogers, Dearing and Bregman (1993) imply, the question of “How...the media agenda, the policy agenda, and the public agenda collaboratively influence each other, if they do?” (pg. 79) has yet to be answered. This study was intended as a first step toward that end. Ultimately, we conducted this project in order to consider three questions, each of which is implied by considering the collaborative nature of the various agendas. The three questions are specifically: * How does public opinion change in the long run? * What is the role of the nation’s media in this process? * How does communication play a part in this process? The first two of these are addressed specifically by the design of the study and the selection of its variables. The third is subsumed by, and a key part of, the theoretical implications of this study. Public Opinion Change It seems fairly clear fi'om Chapter One (see especially pp. 12-21) that public opinion is fairly susceptible to influence in the short run, i.e. the time period ranging from several minutes to several months. Mueller’ 8 (1970) long-term study of Presidential popularity is one of the few which is 165 166 explicit in distinguishing effects which extinguish from those which do not (the “rally” events). The question of the long term is a very different question. There have been very few studies which attempted to consider changes in public opinion across periods of a decade or more. Again, Mueller pointed out a general decline of Presidential popularity within the length of the President’s term of office, which he attributed to the alienation of groups (an attribution he was unable to explicitly test); Davis (1975) was able to parse certain long term effects (on the issue of social tolerance) into those attributable to cohort changes in education and age and those which transcended cohort changes (i.e., those which represented change in public opinion within the population rather than across it), but made no attempt to distinguish the causes of those changes which were consistent across cohorts. For the most part, studies measuring media effects and opinion changes across time periods of more than a few months - a semester at most - are few and far between. Examination of such long term efl‘ects seems to call for the use of long-term panel studies, and indeed, in the area of media violence several have been conducted (e. g. Milavsky, Kessler, Stipp, and Rubens, 1982 ; Leflrowitz, Eron, Walder and Huesmann, 1972). The long term panel study, however, is subject to perils of its own, even apart from resources. These include mortality and the question of external validity. Long term causal modeling is a second possibility, but the ex post facto nature of causal modeling requires a thorough discussion of the consequences of causal...or pseudo-causal...statements made in that way. 167 Causation When considering ex post facto the analysis of the impact of social forces on a population, an immediate and vital consideration must be the strength of the causal claims, and therefore the validity of the discovered model, which can be made given the nature of this data. As Cook and Campbell (1979) point out, the gain in external validity received by working with “field” data is offset to some degree by the loss in experimental control, and thus internal validity. In terms of the “classic” canons of causation, while this design gives us the possibility of establishing both concomitant covariation and time ordering, it does not provide complete protection against non-spuriousness. As Granger (1969) mathematically points out, to have complete confidence in the causal impact of one of the variables measured as proposed on another, one would require complete knowledge of the universe of variables, although he goes on to point out that only all those of appreciable impact themselves are truly necessary. Either of these conditions is impossible to achieve. Blalock (1971) refers to the results of path analysis as “causal inferences,” pointing out that path coefficients can be potentially be biased by unmeasured (and therefore uncontrolled) variables. It is one of the functions of path analysis to distinguish relationships which are causal from those which are spurious; in the absence of perfect measurement of all relevant variables, path coefficients and path models must be taken with the proverbial grain of salt. In the discussion which follows, use of the term “causes” or imputations of “proof” of causality will be avoided; they are not warranted by the techniques in use. The path coefficients and models do, however, 168 provide evidence that causation may be occurring, and it is from this assumption that discussion will proceed. The term “causal priority” will be used in this regard, to suggest the evidential nature of the relationship. Basic Findings Given the caveat concerning causality discussed above, it is important to note that in all three of the models...the overall analysis and the two sub-analyses...long term changes in public Opinion are associated with the activities of policymakers and the media. To the extent that we can say it, the policy agenda and the media agenda have causal priority over changes in public opinion. It should be noticed that these changes are neither large in magnitude nor immediate in occurrence. But in terms of magnitude they are sufficient to be detected by a study of no more than moderate power (see below). Further, there seems to be little question about the nature of the time lag involved. As pointed out in Chapter Three, the substantial number of even numbered lags present in the three models resulting from the three analyses (vide Figures 10, 12, and 13) strongly suggests that the nature of the influence is tied in some way to the congressional electoral cycle, as suggested by Arnold (1990). It is in the very nature of the American political system that Americans focus most firmly on political issues during the time of the election campaign. More importantly, the nature of the debate during that campaign is generally retrospective: most of the members of House of Representatives choose to defend their seats, and when they do so they are forced to defend their records...and many times specifically to show that the 169 actions they have made during the course of their two-year term are directly and specifically related to the positions they advocated during the previous election cycle. Political debate not only occurs in two-year cycles but also reflects back across the two year period to the previous cycle. This is clearly demonstrated by the findings of this project. Collaborative Influence As Rogers and Dearing (1988) have pointed out, there is an extensive literature that supporting the impacts of the media agenda and the policy agenda on the public’s agenda, and, by extension, and to a lesser degree, public opinion, and each other. The results of this study, in particular, those associated with the first three expectations support this literature, extend it into the long term realm, and help to elucidate that interrelationship. One particular element of the role of the policy agenda seems to be recognition that the policy agenda does not appear to act as a unitary construct. In both the overall model and the foreign afl‘airs model, not only do the congressional agenda and the administration agenda separately impact public public opinion, they do so in different directions - when one is positive in sign, the other is negative (see Figures 10 and 12). Part of this is tied up, of course, in the nature of the measurement of the two agendas: in an attempt to parse actual leadership from acquiescence, events which were coded as part of one agenda were not coded as part of the other and “contrary” events such as vetoes and overrides were coded while “consonant” events such as signings and veto sustainings were not. But certain of the results indicate that in some cases the Congress speaks louder than the Presidency (in the overall model the path coefficient from Congress to public is larger than that of the 170 administration to public, although not significantly so) and in others the reverse is true (in the foreign affairs model the administration agenda intervenes between Congress and the public, while the congressional agenda does not appear in the Race and Ethnicity model at all.) Equally interesting is the fact that the exact relationship between the congressional and administrative agendas, and between those and the media agenda, appears to be topic dependent. This was expected to some extent due to the ad hoc nature of the issue network structure: difi‘erent issue networks use different paths to political influence (i.e. some enter via the House, some via the media, and so on). Preliminary speculation about the nature of the relationship between the congressional, administration and media agenda can be developed starting with the discussion (on page 154) of the lack of an influence due to Congress on the sub-analysis of the Race and Ethnicity issue: it was then presumed that that, while Administrations generally had strong views on the subject, the Congress itself was split, and thereby forced into a relatively subordinate position. On the whole, across the three models it is the administration’s agenda which most consistently shows direct effects on public opinion. This follows logically in two ways: as pointed out previously (page 30), administration policy tends to be somewhat more monolithic than Congress’; and the administration, of course, possesses the “bully pulpit” of the Presidency. There is no question that Congress can resist the will of a President, but it is also clear that in the long run the administration is more likely to have an efi'ect on public opinion. It is also fairly clear that, consistent with the findings of 'D'umbo (1995), the greatest impact that the media agenda has on public opinion is 171 the impact it has on one element or the other of the policy agenda. Only in the Race and Ethnicity analysis...again, that in which the congressional agenda was stalled...did the media show a direct effect on public opinion, and it is worth noting as well that it is only in the Race and Ethnicity sub- analysis that the media’s relationship to public opinion is positive in sign.’l (In that sub-analysis it is also at its largest in magnitude). This may suggest, in a fairly preliminary way, that the media’s impact on opinion is largest when working in a relative policy vacuum; that the policy agenda normally speaks louder than the media agenda. This is consistent with the model of biannual public discourse tied to the electoral cycle mentioned above...people talk about what the candidates are saying, and much less about what the media are saying. Although not explicitly identified and tested as an expectation of this study, the time series nature of this project, involving as many of 45 years worth of data and the analysis of lags of up to eight one-year data periods, suggests that long term impacts, reasonably measured in years, were to be expected. This proved to be the case. Impacts on public opinion by other agendas involved variables lagged a minimum of two years - none of the synchronous variables were part of a relationship held to have causal priority on public opinion. Unexpected was the number of variables having causal precedence on public opinion which affected it negatively. Most of the literature, save almost solely for the results obtained by Trumbo (1995) and Page, Shapiro and Dempsey (1987), suggests that a positive relationship exists, explicit in Cohen’s statement about “telling the peeple what to think about.” This again may be pointing up the difference between agenda setting and changing public opinion, in that the former can be conceptualized as one ‘This is consistent with Pan and Kosicki (1996). 172 step in a process leading to the latter. By far the vast majority of studies involve only agenda setting, and it seem perfectly reasonable that a positive ability to set the agenda does not necessarily translate into a positive impact on public opinion, if for no other reason than the mass audience may select a different fi'ame. Given the relative mistrust of both journalists and politicians in the nation in recent years (McAneny, 1995’ ), the audience may deliberately select a different frame. An issue network approach to the dynamics of public opinion suggests a competing message environment, and, by extension, a competing fi'ame environment. The successful frame need not be the one supplied by the media, the Congress, or the President. Exta'nal Socioeconomic Factors The failure of expectation four to be supported continues a recent trend. Rogers, Dearing and Bregman (1993) implicitly reach the conclusion that “real world” indicators do not play an important role in the agenda setting process when they suggest that discovery of the reason why this does not occur is one of the important questions for the firture of agenda setting ' research, despite evidence (c.f. Behr and Iyengar, 1985; Erbring, Goldenberg and Miller, 1980; D’Alessio, 1992) that for highly obtrusive issues “real world” indicators can be part of the process. (The study cited as evidence, Danielian and Reese, 1989, uses drug deaths as its “real world” indicator. In 1988, however, only 4865 deaths due to overdoses of both legal and illegal drugs were reported in the US, according to the US Department of Health and Human Services, 1991. Even if each of these had indirectly affected 100 people in a consequential manner, drug deaths still would have affected only a minute percentage of the population, and so cannot be ’ In reporting public Opinion of the honesty and ethical standards of 26 professions, McAneny found that journalists ranked 11th, TV reporters and commentators 14th, newspaper reporters 15th, Senators 22nd, and Congressmen 25th. 173 considered particularly obtrusive.) It is counterintuitive to simply assume that external socioeconomic factors play no role in the dynamics of public opinion. The absence of support for this position shown herein is explicable in a number of ways: * It is possible that it is true that external factors play an unmeasurably small role in the dynamics of public opinion. The studies cited above are all studies of the agenda setting process, and we have seen previously, it is distinctly possible that agenda setting is only the first of a series of contingent processes which eventually lead to changes in public opinion. It may be that external factors have ability to influence the public’s agenda but have no impact on opinion due to the intervention of the framing process. Framing, of course, permits the social construction of meaning for “real world” events. * Recent research by Kim and Hunter (1993a, 1993b) points out that attitudes have a strong influence on behavior if the correct attitudes and behaviors are measured; specifically, the relationship is strong if the attitude in question is the attitude directly related to the behavior in question.” When dealing with global measures of public opinion, it is sometimes true that the opinion is directly related to the behavior in question; for instance, one of the output variables was the number of minorities elected to Congress, and one of the opinion items was “If your party nominated a qualified (minority) for Congress, would you vote for him?” However, the vast majority of opinion items were at best peripherally related to the behaviors in question. * It may be that the agenda setting effects cited above were too short term to be of consequence in a study where the unit of analysis is the year. ‘ Price (1992) points out that opinions are derived from attitudes by means of making the attitude explicit. The relationship between attitudes and opinions should be very high. 174 Behr and Iyengar, and Erbring, Goldenberg and Miller, were dealing with effects which they expected immediately, while D’Alessio was dealing with lags of up to two quarters (six months) More importantly, it may be that many variables considered to be obtrusive at the day-to-day level may not be as much so at the annual level. Inflation, for instance, is held to be a highly obtrusive variable, because it essentially afi'ects the entire consuming population. In the absence of catastrophic, crippling inflations of the sort which occurred only twice in the course of the study (after World War 11 price controls were released, and during the 1970’s gas 'crisis), the consequences of inflation are almost invisible. The importance of inflation is what it does in the very long run, to savings, pension plans, and the like, but on a day-to-day or week-to-week basis its effects are subtle. In this case, the “long run” is a period of decades, not one year to the next. Demers, Craff, Choi, and Pessin (1989) have commented on the difficulty of establishing what is obtrusive and what is not. The difficulty in conceptualizing this variable can be seen when they define obtrusive in terms of personal experience and then claim that drug abuse is obtrusive. More advanced conceptualization, perhaps in terms of individual consequence instead of simply experience, may be needed. * It may be that the external socioeconomic variables were misspecified. Some evidence of this appears in Figure 11, in which lagged “output” variables are shown affecting opinion rather than vice versa. On the other hand, the output variables were selected as outputs because of their behavioral nature: voting, joining unions, donating money, and the like. * It may be that external socioeconomic factors do influence public 175 opinion, but do so at an effect size which too small to be detected in this study. (A discussion of the power of the techniques employed appears below.) Role of Nation’s Media As pointed out previously, the role of the media in changing public opinion is both consistent and inconsistent from issue to issue. The media’s agenda plays a significant - statistically and substantively - role in each of the three models which were evaluated. In two of them, the media’s contribution to change in public opinion is indirect, via one or the other elements of the policy agenda, and negative. In the third, the Race and Ethnicity model, the impact of the media is positive in sign and direct. The causes of this variation can only be speculated upon, as there is little about any of the relevant theoretical positions that would predict this sort of outcome.‘ Also interesting as the pattern of lags themselves. In each of the overall model and the two submodels, as mentioned in the previous chapter, the media play some role. The specific role varies from model to model, but the media agenda is a part of each. More importantly, the all of the lag times for media effects are even; in fact most of the significant lags involving the policy agenda items are even as well. As brought out in chapter 3, Arnold (1990) has suggested that the agenda for a congressional campaign is set by a large degree by the media ‘ One could speculate that Race-and-Ethnicity—oriented issue networks such as the NAACP, in the absence of access to media via policymakers, attempted to frame their issues directly through the media and were successful in doing so. Several of the key images of the Civil Rights movement were intended as political demonstrations of unity and power, but also possessed many of the characteristics of “pseudo-” or media events. Most notable: Life magazine photos of a little black girl being escorted to class by National Guardmen and Sheriff “Bull” Conner loosing the dogs and water cannon on marchers in Birmingham. 176 coverage of the previous campaign in that district. This can perhaps be seen most clearly in the overall model, where the media agenda at t—4 impacts directly on the congressional agenda at t-2; that is, two years later. The number of even numbered lags involving the media and policy agenda suggests that the influence of these on public opinion is tied to the two-year congressional election cycle. An intuitive feel suggests that in the first (non- election) year people receive information about the performance of their congresspeople but in the second that performance, as well as future plans, are subject to actual debate, with the attendant attempted transmission of priming and framing information. Finally, it is worth noting that the media’s influence occurs relatively far into the future. The media agenda appears in the various models at lags from public opinion of 4-6 years, suggesting that it takes some time for the media’s agenda to filter down to the public. Explicit in the earlier chapters of this work was the idea that, for the purposes of this work, the media’s agenda would be measured by, and limited to, overtly biased statements by authors of works for periodicals. Part of this limitation was due to the exigencies of measurement: as pointed out on page 62, periodical works were considered to be the best possible source through which to access the media’s agenda. And part of this limitation is based on the nature of the role of the media in the political process, which will be discussed in greater depth in the section about the role of communication in the changing of public opinion, which appears several pages subsequently. 177 Media Bias In essence, although there does not appear to be a major theorist of media bias, there seem to be three bodies of thought about its nature. The first of the approaches to media bias deals essentially with the issue of gatekeeping: that writers and editors select from a body of potential stories those which will be presented to the public and, by extension, also select those stories of which the mass audience will hear nothing. This is the level of analysis of White’s ( 1950) “Mr. Gates” study, in which the editor selected among stories provided to him by his wire service, sometimes on ideological grounds. As pointed out previously (pg. 45), it is the nature of these ideological grounds which are of some concern. Page and Shapiro (1982) provide an overview of the wide variety of selection biases in the media. They include, among others, a nationalistic bias, in which foreign news is “slanted” to provide news about US interests; an anti-Communist bias; a procapitalist bias; a minimal government bias; and a pro-incumbent bias in electoral news. This list is particularly interesting in that most of these positions are associated with conservative rather than liberal thought (through the long association of pro-business interests and conservative thought; the media are, after all, businesses, at least, in the US). In an important sense, however, selection biases by themselves may be essentially unknowable. If one considers the universe of all stories as a population and a list of those which are covered as a sample, the presumption is that, because the “sampling” procedure has the biases outlined above built into it, the selection will therefore be biased. This is only a presumption, however, as the population is not only unknowable but unidentifiable. What would be “all the news of the world”? And, in the 178 absence of population data, while it is safe to presume that selection bias occurs, it is impossible to know, or even estimate, its magnitude.‘5 There do exist areas in which population data is available, such as major crime events, but these are greatly outnumbered by those in which population data is unavailable. The second type of bias considered in the literature might be called “coverage” bias. There is an extensive literature which attempts to codify media bias by measuring the physical amount of coverage each side of some issue receives (c.f. Stempel, 1969; Stempel and Windhauser, 1989; Stovall, 1985). This is typically measured in column inches for newspapers and newsmagazines, although photographs (Batlin, 1954; Klein and Maccoby, 1954), and headlines (Stempel, 1965) have been counted (and measured) as well, while analysis of television includes the amount of time devoted to sides of the issue (c.f. Doll and Bradley, 1974). The vast majority of these studies deal with the political realm generally and the specific area of partisan politics. In a two-party electoral system, it is reasonable to assume that half the coverage should be accorded to one side and half to the other, and that therefore deviations from this pattern are consistent with a coverage bias of some kind. Reporting on issues on which there are more than two legitimate positions (as was the case for several of the issues that compose this study) starts to render the notion that there is an a priori “fair” distribution of coverage untenable. Similarly, while two political parties roughly balanced in appeal to the electorate can be considered reasonably as being an “issue” with two sides which can be equitably reported by having equal coverage of ”Year to year changes in selection bias might yield relative figures, but it would be unknown whether a year to year change would have been due to changes in the manner in which the sample was selected, that is, changes in bias, or changes in the underlying population. 179 both sides, many other social issues do not lend themselves to this sort of artificial balance. An example from outside the scope of this study is again the issue of abortion: while there are economic, social, and constitutional arguments in favor of the “pro-choice” position, the “pro-life” position boils down to the simple argument that abortion is immoral. What division of coverage space between these very different types of argument constitutes fair coverage? How far can the media deviate fi'om this division and still be “fair”? As with selection bias, it seems impossible to measure coverage bias outside the electoral realm. The third type of media bias to be considered herein is what might be termed “statement” bias, that is, that members of the media can interject their own opinions into coverage of an issue. This type of bias is of the sort which engendered the first concerns about media bias (c.f. Ephron, 1971), and is the sort of bias which was discussed on pages 44-47, earlier. Statement bias, unlike the other two forms, while subjective, is also measurable to a degree in that the unit of analysis is knowable: a given media event which contains equal numbers of statements biased in one direction as those biased in the opposite can reasonably called “neutral”, and one which contains no overtly biased statements can be called “unbiased”. The discussion of pages 45-47 suggests that on the whole there is a small but statistically significant and measurable liberal bias at this level. Although the question of bias is in certain senses peripheral (although not completely unconnected) to this study, given the perceived importance of media bias in influencing the mass audience, it is worthwhile to briefly consider the results of the content analysis from which we determined the media’s agenda. When the net score of biased 180 statements was aggregated across the entire database of articles, a mean of +0.93 was calculated; that is, on the whole, across the entire database, there was about one more liberal statement than conservative statement per article. The complete distribution appears in the form of a histogram as Figure 14. As can be seen in Figure 14, the mode (at 297 articles) of the distribution is zero; the difiering mode and mean (as well as the shape of the distribution itself) indicates a certain degree of skew. More importantly, the “spike”-like shape of the distribution suggests not a normal (or bell curve) distribution, but instead but instead a highly narrow distribution typical of skilled practitioners attempting to hit a goal.” The standard deviation of this distribution is 2.34, and the mean therefore differs significantly from zero (t=17.63, p<.001). Although the range is 37 points across the entire distribution, the standard deviation suggests that a nine point scale would be sufficient" to capture 99% of the variation in this distribution. The discussion concerning the nature of the Liberalism scale points out that there may be an element of measurement error in the estimates of liberalism, as the Conservative end of the scale was derived by essentially reversing the Liberal items. As pointed out at that time, this fails to capture the extent to which the scale is not essentially bipolar, and specifically fails to account for those issues and ideas which are central to conservatism but peripheral to liberalism. In short, the finding of a liberal bias across the entire sample may simply be the result of measurement error. It is also worth noting that the mean bias of +0.93...less than a single 'Kurtosis = 15.3, indicating that the distribution is substantially narrower than normal (the kurtosis of a normal distribution is 0; Nie, Hull, Jenkins, Steinbrenner, and Bent, 1975). 7 2.34 x 2 tails x 1.96 (the 99th percentile) = 9.17 181 w). my m1 200‘ ‘ 215°: . 100‘ O . .0 . 501 ' M“ w .5(; r g , T , 1 - .20 .15 -1o -5 o 5 10 Score Figure 14 Frequency Distribution of Net Liberal-Conservative Statement per Article 182 point...on a nine-point scale is roughly consistent with the other studies of the subject, as discussed in pages 44-47. That is, again, a small but statistically significant and measurable liberal bias has been detected. Also revealed is that the sample indicates that there is indeed a competing messages environment, as is necessary to the theoretical positions articulated by Zaller (1992) and Tapper (1995). Along with the 297 studies which showed no net bias, 492 were coded as having a net conservative bias. Further, it is important to recognize that these are net scores: the proportion of articles which were coded as possessing no biased statements at all was fairly small (9.9%), while 62.5% of articles contained statements which were coded as liberal simultaneously with statements coded as conservative. Direct evidence of a competing messages environment is not consistent with the popular view of a monolithic, systemic entity called “the media” which is unrelentingly liberal in orientation. Neither is the lack of issue-by-issue consistency of media messages, as was demonstrated by the lack of correlation between issues in Table 12. If a persistent, systematic liberal bias does occur, it must do so at a level higher level of abstraction than that of the statement. Finally, there is the question of “What influences the media?” Only in the Foreign Affairs sub-analysis does causal modeling indicate that other elements of other agendas have any causal priority on the media’s agenda. In Figures 10, 11, and 13, the media agenda acts as an exogenous variable. This is particularly interesting in that the various external socioeconomic factors are as unrelated to the media agenda as they are on public opinion or the other agendas. Although, as was made clear above, the media agenda as measured herein consisted of statements of an overtly 183 biased nature, one would presume that “biased” reporters would seize on events for the opportunity to make their points. Communication generalb' Finally, apart fi'om the overt contributions to the debate on issues provided by members of the media, there is the question of what other roles does communication play in the change of public opinion. This in turn can be subdivided into questions regarding two difi'erent types of communication: interpersonal and mediated. Interpersonal communication was not measured or considered as part of this project, yet it is presumed to play several parts. It was assumed that interpersonal communication is part of the process of the development of interest networks; specifically, as issue networks are assumed to start at the local level, it must be interpersonal communication which brings the original members together. Further, as McCombs and Shaw (1972) point out, evidence indicates that knowledge of the specific consumption of mediated messages does not predict agenda setting efi'ects as well as the assumption of a message “milieu”. It is known that mediated messages provide topics for subsequent conversation, and Zhu, Watt, Snyder, Yan, and Jiang’s (1993) results suggest that talk is as much a part of the milieu as television. More saliently, separating the media agenda and the policy agenda into separate components of the analysis begs the question of how the policy agenda is to be communicated. Legislation qua legislation can have no direct impact on public opinion; there has yet to be a law passed which mandates specific responses to opinion items. Legislation, whether engendered by the Congress or the administration, can have a direct effect 184 on external socioeconomic factors, and indeed has been designed on occasion to influence them. But considering the impact of the policy agenda on the public’s agenda or on public opinion presupposes that that activities of Congress and the administration are communicated to the people in some way, and our understanding of the mass audience that composes the “public” in “public opinion” requires that they be communicated in large part by mediated messages. The leads in turn to the next question: at what point do mediated messages become the media’s messages? Patterson (1993) provides an illuminating example: “...Mondale loses the New Hampshire primary to Gary Hart, 41 to 29 percent. A desperate Mondale borrows an advertising line fi'om Wendy’s hamburgers to ask of Hart: ‘Where’s the beef?’ Reporters pick up the line even though Hart has provided them with a score of detailed, innovative position papers. The press concludes that Mondale is back in the thick the race after primary wins in Georgia and Alabama, despite losses in six other states on the same day.” (pg. 40) In the terms used herein, Mondale successfully reframed’ the issue of who the Presidential candidate should be into his own terms. This reframing was conducted via the media. Patterson makes it clear that in his opinion at least this reframing was unwarranted. This situation makes a classic discussion item: specifically, what was the media to do? The frame was supplied by Mondale. To fail to report it, while continuing to report on Hart, would have constituted bias (selection ‘ More specifically, he set the agenda by making stances on the issues the most salient factor in selecting a Presidential candidate (rather than, say, personal magnetism), primed the issue by claiming that a candidate should be evaluated by the substance of their positions on the issues, and then framed it by claiming that Hart’s positions were insubstantial. 185 bias). 'lb report it with the additional information about the position papers juxtaposed would have been, in essence, to call Mondale a liar. J uxtaposition, that is, the placement of images in combinations which create meaning (Eisenstein, 1942/1975), gives visual media the ability to create, deliberately or inadvertently, and communicate frames (c.f. Lang and Lang, 1974). Similar phenomena attend the print media, and for much the same reason: no medium can exactly reproduce all the information contained in an event, and thus salient elements have to be selected for transmission...and the judgment of salience is necessarily given to the communication channel rather than to the source or receiver. The results of this study indicate that, despite the role of the media in conveying the policy agenda, there is a point at which the media agenda and the policy agenda can be separated. Clearly, writers or reporters who use their medium to make an overtly biased statement of opinion can be seen to be the source of that particular message. Similarly, policy makers (or would-be policy makers) who produce their own materials - their own newsletters, newspapers, TV and radio shows - and who exercise creative control over their own materials can be seen as the sources of those materials. But it remains to be determined where in the middle ground in between that public opinion impacts created by media coverage of policy issues are the responsibility of the policy makers and when they are the responsibility of mass communicators. Caveats In virtually every scientific study there is the necessity that the potential problems of the study be considered, at least in passing, as such consideration will aid the scientific community in evaluating the “weight” 186 to apply to the results. A study such as this which attempts to explore a topic at a broad, systemic level for a long period of years by necessity makes a number of assumptions about which elements are to be included and which to be excluded. In addition, there are a number of assumptions about the appropriate way to operationalize broad, systemic variables, and once the operationalizations have been made, to analyze the interrelationships between those variables. For the most part, the previous chapters have been explicit about what assumptions have been made and why those assumptions were made, and to reiterate those here would be both redundant and tiresome. There are two additional areas, however, to which additional consideration need be paid before this work can said to be complete. These are theoretical considerations and the calculation of statistical power. Theoretical consideration Two theoretical formulations have played a part in this study, the formulation of agenda setting as the first in a series of processes which together lead to changes in public opinion and the issue network approach. This study was not a formal test of either theory; rather, these theoretical positions are presented as attempts to organize and to make clear the connection between seeming disparate processes and bodies of scientific literature. A perspective which holds, as presented on pages 24-25, that agenda setting, in conjunction with priming and framing effects, has the potential to influence public opinion seems to be clearly where Iyengar (Iyengar and Simon, 1993) is headed. Further, it would view agenda setting and the dynamics of public opinion as related rather than separate phenomena, 187 drawing together the extensive body of literature devoted to each. Such an approach is consistent with Cohen’s (1963) summation of the two fields — “(The press) may not be successful much of the time in telling people what to think, but it is stunningly successful in telling its readers what to think about...” (pg. 13) - in that, as the first step in a multi-step process, the achievement of agenda setting effects would be substantially easier than changing opinion, the last step in that process. Further, the intervening steps, priming and framing, would seem to be fairly difficult, particularly in the face of selectivity processes and a competing messages environment. If this perspective is accurate, it is perfectly consistent for the media to have substantial short term effect on the public’s agenda and only modest effects on public opinion in the long run. The issue network approach was devised to provide an explanation (or, more specifically, a rationale) for including both the media agenda and the policy agenda in a study on the dynamics of public opinion. As Rogers and Dearing (1988) point out, these are two separate research traditions, and typically studied in isolation by different groups of scholars. In the absence of a reason to combine them, one runs the risk of comparing apples to oranges. Yet the issue network approach not only suggests that the media agenda and the policy agenda are in some ways alike (because of their ability to access the mass audience), it also points to yet a third sociological grouping whose effects should be similar: celebrities. It is possible to name a number of celebrities, who, by reason of their celebrity have access to media time, and are willing to use it to advance causes in which they believe; that is, to advance the causes of issue networks of which they are members or supporters. 188 Consideration of the notion that celebrities could also influence public opinion in a manner similar to that of media members and policy makers, that is, by using their access to the media in order to communicate preferred flames on an issue, has already been suggested by Mankiewicz (1989; although many of the celebrities Mankiewicz names are media members themselves, again leading to the question of where the line between the media agenda and the celebrity agenda will be drawn). Further, the substantial literature (and continued use of) celebrity endorsers in advertising suggests that there may be a connection between celebrity endorsement and public opinion, at least on the issue of product purchase. Thus, an issue network approach has the potential to unite these three different literatures, and is thus worthy of future consideration, at least on the grounds of parsimony. wau' The question of statistical power is actually one half of a broader question, and that is the question of testing error. Type I error, the error of the false positive, arises from the necessity of sampling, and can be minimized by the establishment of a stringent alpha value (Lin, 1976). Type II error, the error of the false negative, arises in part from positivist requirements of parsimony. Power is the logical inverse of the probability of Type 11 error; in short, is is the ability of a statistical test to detect true relationships in a set of data. Power is a function of the effect size of the true relationship, the alpha level, and the number of cases in the relationship being tested (Cohen, 1988). N’s for this study vary, as the tables in Chapter Three show. This 189 occurs for a variety of reasons: the availability of data, in terms of what epochs it is gathered and published; missing data within an epoch; and the number of data points within the epoch. A study using annual data of the time period from 1945-1990 starts with a set of 46 data points. In some cases, for instance, in the measurement of public opinion, cases could be added at the beginning, and thus those analyses which deal only with public opinion have more cases than others. On the other hand, there was an amount of missing data from some of the external socioeconomic factors, most notably pr0pinquity, and thus those analyses may have a small number fewer cases. From the overall standpoint, it needs to be pointed out that 46 cases is not a particularly large number for a time series analysis. Nothing in the literature review led to the suggestion of negatively signed impacts. Consequently, one-tailed tests of statistical significance were used. This is important in that, although neither Lin (1976) nor Cohen (1988) recommend this, establishment of alpha level was perhaps the only means available to control both Type I and Type II error. Cohen (1988) suggests that effect sizes be considered as falling into three basic categories: small (r = .10), medium (r = .30), and large (r = .50). There was no expectation that this design would be suitable for the detection of small effects: the critical r for 46 cases using the one-tailed test is .246; power would be a mere .16, that is, there would be basically five chances out of six that this relationship would not be detected. On the other hand, large effects are readily detectable using 46 cases, yielding a power of about .98 (leaving only 2% chance of Type II error for relationships of this magnitude). The question of the medium effect size is the key question in establishing alpha, from the point of view of this study. As Cohen (1988) 190 points out, .3 is a fairly typical correlation in the social sciences. Continuing to use the one-tailed test after the detection of negatively signed results — as was done — increases the possibility of Type I error from .05 to .10; however, it also increases the power of the test for relationship of medium effect size fiom .54 for the two-tailed test to .66 for the one-tailed. In other words, the .05 loss in protection against Type I error was offset by a .12 gain in protection against Type II error. Potential problems with Type I error are minimized in the path analyses, as an isolated spurious relationship would lead to increased specification error in the model (as each path refers not to a single correlation but instead a consistent pattern of correlations). However, the likelihood of Me 11 error is substantially higher than that of Type I error, and it is possible that part of the reason that the three models (Overall, Foreign Affairs, and Race and Ethnicity) differ is Type II error. Concludingnote It seems at times that each scientific study ends with a call for more research in the area of study, and this shall be no exception. The overarching justification for this research was that it would start to explicate the long term interrelationship between public opinion, the policy and media agendas, and external socioeconomic factors. Several of the expectations (based on the existing literature) were supported and others were not; in some cases the interrelationships were more complex than expected. One area in need of further explication is the nature of the relationship between agenda setting, priming and fi'aming. As mentioned previously, there is a modicum of evidence suggesting that the three 191 processes are both related and contingent stages in the overarching process of mass opinion change. This is consistent with recent theory (c.f. Chaiken, 1980; Petty and Cacioppo, 1984) on the nature of persuasion, which holds, in essence, that involvement with an argument is a necessary precursor to actual persuasion, as opposed to compliance. This phenomenon could be studied under laboratory conditions using a methodology similar to that employed by Iyengar and Kinder (1987), utilizing mock newscasts. Field experimentation would be more complex, but could be accomplished given inside information involving a “breaking” story, or as part of the quadrennial Presidential primary campaigns: the researcher would have to have pre-primary opinion data on prominent candidates, and then would need to analyze candidate statements searching for attempts to set the agenda, and prime and frame issues; would need to analyze media content to see which of these attempts were transmitted; and would need to gather running data on issue salience and opinion in the affected audience.9 A second area which suggests itself for further empirical investigation is that of the question of obtrusiveness. As mentioned previously, it may be that in the long term there are no truly obtrusive circumstances. But this depends in large part in the conceptualization of “obtrusive”. Certainly there are socioeconomic factors which affect everyone, but for the most part, and in the long run, the effect of these factors is fairly small. Examples include issues such as inflation, education policy, and the Cold War: each affects every US citizen, but across a period of years the total effect is fairly small. It may prove that the only ”The 2000 Republican campaign for the nomination might be highly suitable for analysis, and any primary campaign conducted by the party out of office should provide good grounds for data gathering. (The party in office generally has fewer problems arriving at a candidate.) 192 truly obtrusive issue, at least from the viewpoint of a period of years, is war. War, course, affects every member of a society eventually, and can do so drastically. During the era of modern public opinion polling there have been three major, long term wars, World War 11, Korea, and Vietnam, and it should be possible to examine the relationship between public opinion and the actual events of those wars using techniques such as employed herein, or by 'IYumbo (1995). Another approach to examining the question of obtrusiveness would be to approach the subject from the standpoint of meta-analysis (Hunter, Schmidt, and Jackson, 1982). Once a formal, consistent operational definition of obtrusiveness is developed, it could be applied to extant studies of the issue. Studies could be classified as “Low” and “High” in obtrusiveness, accumulated, and effect sizes between the groups compared, with the expectation that the effect size of the high obtrusiveness group would be higher than that in the low obtrusiveness group. A third area ripe for further investigation is that of the development and influence of the issue network. Numerous writers, including Baumgartner and Jones (1993) and Zaller (1992), stipulate the existence of groups whose actions influence policy and opinion, frequently terming them “policy elites”. But the actual development and motivation of the issue network is assumed, or undiscussed and unexplored. An in-depth examination of an issue network would call for the use of historical techniques. An issue, such as the recent passing and signing of the Communication Decency Act, is suited to such an investigation. The policy debate took place publicly on the floor of the Senate and in the full view of the public. The debaters, US Senators, are identifiable not only as to identity but also as to prior association — both ideological and financial — 193 with a variety of extant issue networks, whose stands on the legislation can be determined as well from their statements on the issues. In short, it should be possible to trace a “chain of influence” backward from the event to issue networks which invented it...if they did. The nature of the liberal shift itself is a matter of some concern to the communication researcher, in that different sources of the shift support different theoretical models of opinion change. Consistent shifts of the sort shown in Figures 1 and 8 can be seen as being the result of two basic processes: either a small (but consistent) number of people are undergoing substantial amounts of conversion, with the others remaining more or less stable; or fairly large numbers of people are undergoing incremental changes in opinion. The latter position is consistent with the aggregation/cultivation type approaches taken by Zaller (1992) and Iyengar (1991). The former, on the other hand, would necessitate a reevaluation of the exact nature of media impacts on viewers, to a position more consistent with Mueller’s idea of alienation of sub-groups. A quasi-experiment designed to evaluate which of these two possible models holds would almost inevitably require a long-term panel study of several years. Mortality among members of the panel could be reduced by making financial inducements substantial enough to encourage them to remain in the study. (A relatively low cost way of doing this would be to offer inducements in the form of savings bonds.) A final area of further research more clearly related to the specific findings of this study is into the nature of the long term relationship between policy action and public opinion. As pointed out earlier, in the overall model and one of the sub-analyses the relationship between these 194 two was negative in sign, a finding which would have been completely unexpected had it not been for 'II'umbo’s (1995) earlier, similar, finding. The genesis of this effect, however, is unknown. What is known is that the the effect seems to be tied to the two-year cycle of congressional elections, suggesting that it may be tied to the nature of the debate and discourse which surrounds the election. In theoretical terms, there are a number of potential explanations: * Distrust: As McAneny (1995) has again demonstrated, “political figure” is among the least trusted professions in the nation. It is possible that as campaigning politicians discuss their accomplishments, their listeners assume that those accomplishments must somehow, 3 mind, be wrong. * Dissonance: Festinger (1957) has pointed out the possibility of attitude change as a result of decision making. Voting, of course, is a decision making process; having made the decision in favor of one candidate (and his or her positions) in one election, in the next the voter’ s sense of dissonance may be reinforced by attacks on the actions of the elected official by the opponent. * Reactance: Brehm has pointed out that, under certain circumstances, people reject things which they would otherwise desire if the things are presented unexpectedly or in an unexpected manner. It could be that many voters ideologically desire certain things, but resent them when presented with them in the form of legislation. (This would be symptomatic of a deep-seated, underlying anti-Federalistic value in US citizens.) A reactance-like position is viable when one considers the nature of political debate in the US, at least as characterized by an issue network (or 195 similar) conceptualization: as an issue arises, issue networks develop, each advocating different solution. These issue networks may be equally balanced on either side of the issue, but when policy-makers choose one of the alternative solutions proposed and legislate it into action, issue networks advocating other solutions from both sides of the issue are alienated. Empirically exploring whether one or more of these or other explanations hold depends in part in the nature of the theoretical position being tested. However, it is clear that opinion polling is generally far too superficial in its approach to determining how the American public feels about and reacts to it political representatives and their actions. The question far too infrequently asked in opinion polls is “Why?” The nature of an exploratory study is that it raises questions, and again this study is no exception. The issue network approach was shown to have some organizing utility, albeit not necessarily validity as a theoretical position: it remains to be formally tested. As pointed out above, negative relations between the various agendas and public opinion, which were unexpected, might be the result of any of a variety of processes. While overall effect sizes were not calculated, it is clear that much of the variation in public opinion remains unaccounted for. The border between the the policy agenda and the media agenda needs to be explored more thoroughly, both theoretically and empirically. In short...in very short...there is a great deal yet to be done. More research is needed in this area. APPENDICES APPENDDC A 1% Appendix A E-mail message from Libertarian candidate Harry Browne Memo #315559 Received: by fred.stockton.edu; (5.65v3.2/1.3/10May95) id AA17863; Wed, 6 NOV 19 96 20:16:25 -0500 Received: from hustle.rahul.net by fred.stockton.edu (smtpxd); id XA14725 Received: by hustle.rahu1.net with UUCP id AA27197 (5.67b8/IDA-1.5 for dalles@cosi.stockton.edu); Wed, 6 Nov 1996 17:05:47 -0800 Received: by dehnbase.fidonet.org (mailoutl.26); Wed, 6 Nov 96 16:06:50 PST Date: Wed, 06 Nov 96 15:50:26 PST Message-Id: <79701.3281281A.ann@HarryBrowne96.org> From: CampaignGHarryBrowne96.org Subject: a message from Harry Browne Sender: browne-lelp.org Reply-To: campaign@HarryBrowne96.org To: announce@HarryBrowne96.org (Harry Browne for President announcements) X-Mailer: mailout v1.26 released with lsendfix 1.6a >>> To Libertarians Everywhere >>> By Harry Browne I want to express my deep gratitude to Libertarians all over America * For your hard work and dedication in taking our message as far and wide as possible. * For the many kindnesses you have shown Pamela and me over the past two years. * For the confidence you placed in me to carry the Libertarian message to the public. As I write this, we don't have final vote totals. But, even so, there's no question we have a great deal to be pleased about. Our total vote will far surpass the 1992 total -- close to doubling it. Based on the media coverage received, we far outpaced the other candidates. We apparently received about 1/18 of Ross Perot's vote total, while getting less than 1/100 of his media coverage. We got 197 roughly the same vote as Ralph Nader, while receiving less than a fifth of his media attention. Evaluating our performance this way, it was even better when measured against the coverage given Clinton and Dole. And we far outpolled all the other small parties. For a $3 million campaign and a party with 20,000 members, we achieved a great deal. But it's obvious that we will never break into the big leagues with a $3 million campaign. The task ahead of us is clear: we have to create a party so big, so strong, so well-financed that in the year 2000 no one can ignore us. We have to make an enormous splash before and throughout the next election year, so that the media will have to give us the same attention and respect they give to the two old parties. We have to be so well known to the public that the Republicans and Democrats can't hold a debate without us. And if they decide not to hold the debates at all, then we must have an army of people so large that we can carry our message door- to-door to every voter, and we must have the money to tell our story through advertising. Is this a realistic goal? Yes. It is a formidable goal, but far from impossible. We are already well along the road to that goal . . . * We have doubled the party's membership in just the past two years. And our new compatriots aren't going to go away. * We have already made inroads, small but significant, into the business and investment community -- to where the money is. * The Internet has become our bailiwick. And it will be more and more influential in politics in the coming years. Libertarians know how to use it, and we'll always be in a position to make the most of any new technology that develops. * We have established wonderful relationships in talk radio. Over 300 radio and TV talk-show hosts endorsed the idea of my being in the Presidential debates; 69 of them endorsed me for President. We have established excellent relationships with popular talk-show hosts such as Michael Reagan, Art Bell, Mary Matalin, Oliver North, Blanquita Cullum, Barry Farber, Randy Jackson, Alan Combes, and dozens of other syndicated and big-city personalities who are now glad to hear from us whenever we have something we want to say. * We have developed good relationships with many people in the print media -- editors at political magazines, columnists, and local daily newspaper editors. Over 75 publications or columnists endorsed my being in the debates, of which 21 endorsed me for President. We will continue cultivating these relationships. Thanks to this campaign, everyone in politics and the media knows who we are and what we stand for. 198 These may seem like small satisfactions. But, in fact, they are an important foundation for the work ahead. We need to stay in the public eye by continuing to have Libertarians on talk radio and television as often as possible. We need to continue churning out press releases -- showing the Libertarian alternatives to the self-evidently meaningless proposals of the Republicans and Democrats. And, more than anything else, we have to build the party membership -- starting right now. With the numbers of members will come the money. With the money will come the media attention. With the media attention will come the public awareness of what we offer. Three quarters of the American people think government is way too large. We are the only party offering those people what they want -- significantly smaller government. There is nothing wrong with our message or the way we're presenting it. We simply need to have it heard by more people more often between now and the next election. We are the only party offering proposals that will make a substantial difference in the average person's life. We're going to repeal the income tax so that every dollar you make is yours -- to spend, to save, to give away as you see fit, not as the politicians think best. We're going to get Social Security out of the clutches of the politicians so that your parents and grandparents know their retirement is safe and the rest of us are free forever from the 15% Social Security tax. We're going to make the cities safe by implementing the only proposal that would dramatically reduce crime -— end the insane War on Drugs before it destroys America. And we're going to restore harmony among all the ethnic, social, lifestyle, and generational groups in America by taking away from the politicians the power to inflict one group's values upon another. This is what Americans want. This is what we offer. And we are the only ones offering anything of the kind. What is missing is the ability to let all Americans know what we offer. Although we can experiment with short-cuts, the only sure way to acquire that ability is through the steady building of party membership, which will lead to the money necessary to attract attention, which will lead to the media coverage that will make us part of the national discussion. So let's get started now building the party to the magnitude necessary for every American to know what we can give them. Let's begin no -- while the others relax and take a year or two off. I intend to speak out for the party wherever possible -- appearing on talk radio, television, public forums, and in print -- letting people 199 know there is hope for America. I will counsel Americans not to despair -- not to feel they're alone in their wishes for a government a fraction of today's size. I will invite them to join us in this great movement to make America a free country again. Again, I want to thank you for the honor you have given me -~ allowing me to be your candidate. And for all the help and encouragement you have provided. It has been a wonderful two years. But we're just beginning APPENDIX B 200 Project Coding Form Key Code: Coding Date: On Domestic Issues Reformist Reformist Opts for change and opposes gums quo Prefers not to change; if desires change, prefers reversions to a previously held posrtron Democratic Democratic Favors full extension of electoral rights to all persons Libertarian Supports civil liberties such as free speech and the right to protest Regulatory and Interventionist Backsthemanagementofbusinessandtheecamybythe government Centralist Using the Federal government to set and enforce national standards and regulate state and local goverranents Humanitarian Favoring a social welfare system for the care and protection of society in general and thelowerclassinpartiailar Egalitarian Advocating equal treatment for all and perhaps equal conditions for all Permissive Tolerating and often approving of non-traditional lifestyles and practices Internationalist Supporting US. involvement in the world Multinational Backing the United Nations and other international efforts Non-Militarist Preferring non-military solutions to international disputes Pro-détente AdvocatinggoodrelationswithCommunistnatiarsandnotanphasizing anti-communismasacomerstoneof foreign policy Favors occasional restriction d electoral rights based on psychological, social, or economic grounds Libertarian Favorsrestrictiondcivil rightsin someindividualcasesinfavorofthe protection of society as a whole Regulatory and Interventionist Opposes intervention of government in bu-ess affairs Centralist Prefers minimizing power of federal government and leaving the establish- ment of standards to states and municipalities Humanitarian Favors protection of society by means of employment, and by private and secular institutions Egalitarian Prefers a meritocracy where positions and conditions are earned rather than set by Federal legislation Permissive Approving of traditional lifestyles and practices over nan-traditional ones On International/Foreign Policy Issues Internationalist Prefers US. to address domestic concerns; i.e. ”America first" Multinational Backs US. and allied opinions against world opinion Non-Militarist Rdusestomleoutmilitaryactimasapolicytool Pro-detente Calls for suspicion dcomrnmist motives and prefers action to counter communist hegemony Check LIBERAL points here Check CONSERVATIVE points here LIST OF REFERENCES 201 LIST OF REFERENCES AAFRC Trust (1981-1990). GjyingHSA. Vols. 2635. ABC Survey #0121 (1984) Agnew, ST. (1971). Address in Boston, MA, March 18. Anderson, N. H. (1974). Information integration theory: A brief survey. In D. Krantz, R. Atln'nson, and P. Suppes (Eds. ), Measmement. PsxehonhxsicaanchurellnformationProceseing. San Francisco. W..H Freeman. Appelbaum, RP. (1970). Theories ongm’el Change. Chicago: Markham. Arnold, RD. (1990). The Logic QanngzeseinnflAetien. New Haven, CT: Yale University Press. Bagdikian, B.H. (1971). The InformetinnMeehinee. New York: Harper and Row. Batlin, R. (1954). San Francisco newspapers’ campaign coverage: 1896, 1952 Journalism Quarterlx. 31(3). 297-303. Baumgartner, ER, and Jones, RD. (1993). Agendeeandlnstabilityin American Rulings. Chicago: University of Chicago Press. Bechtel, R.B., Achelpohl, C., and Akers, C. (1972). Correlates between observed behavior and questionnaire responses on television viewing. In E ..A Rubenstein, G. A. Comstock, and J. P. Murray (Eds.,) Ielefieignend Sociechhaidor Vol 4. TeledsioninDaLtodaxLife Washington. DC: Government Printing Office. Behr, R.L., and Iyengar, S. (1985). Television news, real-world cues, and changes in the public agenda. Pnhlie Qninien Quarterly, flu), 38-57. Berkowitz, D. (1990). Refining the gatekeeping metaphor for local television news lodmalofBroadcastingendElectronicMedia. 34(1), 55-68. 232 Biitner, J .R. (1985). Waring and Ieleeemmnnieetien, second edition. Englewood Cliffs, NJ: Prentice-Hall. Blalock, H.M. (1971). Causal inferences, closed populations, and measures of association. In H.M. Blalock (Ed). CausalModonintheSocielSciences. Chicago: Aldine. Blumer, H. (1948). Public opinion and public opinion polling. American Sociological Region 13(5) 542-554 Blumer, H. (1964). Collective behavior. In A.M. Lee (Ed.) Edneinlee ef Seeielngy. New York: Barnes and Noble. Bogart, L. (1969). How the mass media work in America. In Baker, R.K., and Ball, 8., (Eds.) MmMedie andXielenee, Washington D.C.: US. Government Printing Office. Bower. RT. (1985). IheChengingTeloddsionAudienceinAmerica New York: Columbia University Press. BOX. C.E.P.. and Jenkins, GM. (1976). WWW and Centre]. Md Editien. San Francisco: Holden-Day. Brody, R. A, and Page, B. I. (1975). The impact of events on presidential popularity: The Johnson and Nixon administrations. In Wildavsky, A., (Ed. )PorsoectigesontheProsidencg Boston: Little-Brown. 136-148 Browne, H. (1996). To Libertarians Everywhere. Press release distributed by electronic mail. Bruneau, T. (1994). Subjective time, social interaction, and personal identity. Paper presented to the annual convention of the International Communication Association. Campbell, D..,T and Stanley,J. C. (1966) WWW exoenmentalllesignsforResearch. Chicago: Rand McNally. Carter, R. F., Ruggels, W.,L Jackson, K. M, and Heffner, M. B. (1973). Application of signaled stopping technique to communication research. In Clarke. R (Ed )NonModolsforCommunicationResearch Beverly Hills Sage. Chaiken, S. (1980). Heiristic versus systematic information processing and the use of source versus message cues in persuasion. Jenrnnl ef PemonalitgandSocialPsgcholomL. 4. 752-766. Chang, 0., and Sorrentino, C. (1991). Union membership statistics in 12 countries. MenthlylahorBeyieyz, December. 203 Clancey, M., and Robinson, M.J. (1985). The media in campaign ‘84: General election coverage: Part I. hhlie Qninjen, fi(6), 4954. Cohen, B.C. (1963). IheEneeeandEereignEoliey. Princeton, NJ: Princeton University Press. Cohen, B.C. (1983). The Enblicjs Imnaet en Ferejgn Benny. Lanham, MD: University Press of America. Cohen. J. (1988). Stalistical Rene: for the Behagicral Sciences. 2nd. Ed. Hillsdale, NJ: Lawrence Erlbaum Associates. Cengressandtheflafien. Washington, DC: Congressional Quarterly Service. Congressienalfiecem. Washington, DC: US. Government Printing Office. Cook. TD and Campbell. DT (1979) QuasizEnerimentatichlesignand WMMW. Chicago: Rand McNally. Cook, T.B., Dintzer, L., and Mark, M.M. (1980). The causal analysis of concomitant time-series. In Bickman, L., (Ed.) AppliedSmialEsgehelegy Annnel, vol 1. Beverly Hills, CA: Sage. Cook, F..,L Tyler, T..R, Goetz, E G, Gordon, M. T. ,Protess, D., Lefi‘, D .R., and Molotch, H. L. (1983). Media and agenda setting: Efl'ects on the public, interest group leaders, policy makers, and policy. Enhlie Qm'nign Quarterlg. fl(1).16-35 Cooper, M., and Soley, L.C. (1990). All the right sources. MetherJgnes, 15(2), 20-27, 45-48. Cronbach, L.J. (1954). Edneafienalfisyehelegy. New York: Harcourt, Brace, and J ohanovich. Danielian, L, and Reese, S. ( 1989) A closer look at intermedia influences on agenda setting. In P. Shoemaker (Ed) CemmnnieatienCamnaigne abeutnmge. Hillsdale, NJ: Lawrence Erlbaum Associates. Danigelis, N. L, and Cutler, S. J. (1991). Cohort trends 1n attitudes about law and order: Who’s leading the conservative wave? Enhlie Qninien Quarterlg. 55(1). 24—49 Davies, J. C. (1952). Some relations between events and attitudes. The AmericanReliticalSciencochien. m4). 777-789. Davis, F.J. (1952). Crime news in Colorado newspapers. IheAmeriean Journal of Sociclcu. 58(3). 325-330. 204 Davis, J .A. (1975). Communism, conformity, cohorts, and categories: American tolerance in 1954 and 1972-73. American Jonrnal omeiologg, 81(3), 491-513. Davis, J .A. (1992). Changable weather in a cooling climate atop the liberal plateau: Conversion and replacement in forty-two General Social Survey item. 1972-1989. Euhlic Qnim'on Quarterlg. 56(3). 261-306- Defleur, M. L., and Dennis, E. E. (1988). IlnderstandingMasa Cemmnnicatien, Boston. Houghton Mifllin. Demers, D. P., Crafl', D., Choi, Y, and Pessin, B. M. (1989). Issue obtrusiveness and the agenda-setting effects of national network news. CommunicationResearch. 16(6). 796-812 Doll, H.D., and Bradley, BE. (1974). A study of the objectivity of television news reporting of the 1972 Presidential campaign. CentralStatee Speech Journal. Z04), 254-263. Donaldson, S. ( 1987). 1191de Mr; President! New York: Random House. Duncan, O.D., and Duncan, B. (1955). A methodological analysis of segregation indexes. American Sociological Review, m, 210—217. Durbin, J. (1970). Testing for serial correlation in least-squares regression when some of the regressors are lagged dependent variables. Econometrica. 38. 410-421. Eisenstein, S. (1942/1975). The EiimSenae. San Diego: Harvest/HBJ. Entman, RM. (1993). Framing: Toward clarification of a fractured paradigm. Journal of Communication, {1(4), 51-58. Ephron, E. (1971). IheNefleMstere. Los Angeles: Nash. Erbring, L., Goldenberg, E.N., and Miller, AH. (1980). Front page news and real-world cues: A new look at agenda-setting by the media. AmericanlournalofRohticalScience. 2&(1).1649. Erikson, R. S. (1986). The relationship between public opinion and state policy: A new look based on some forgotten data. American Jennal of Political Science. 21(1). 25-36 Exter, T., and Barber, F. (1986). The age of conservatism. American W 8(November), 30-37. Fan. D. (1988).BredictionsofRuhthn1mon£romMassMod1a Westport. Conn.: Greenwood Press. Z15 Festinger, L. (1957). A Theory of Cognitive Dissonance. New York: Row, Peterson, and Co. Foote, N .N., and Hart, CW. (1953). Public opinion and collective behavior. In M. Sherif and MD. Wilson (Eds.). CrounRelationsattheCrossroads. New York: Harper and Bros. Friedman, H. S., Mertz, T. I., and DiMatteo, M. R. (1980). Perceived bias in the facial expressions of television news broadcasters. Jonrnal of Communication. 11(4). 103-111 Friendly. F.W. (1975). ThoCoodGuththadCugaandtheEirst Amendment. New York: Random House. Funkhouser, GB. (1973). The issues of the sixties: An exploratory study of the dynamics of public opinion. RuhlichinionQuarterlg. 31(1). 62-75. Gaziano,C (1.978) Relationship between public opinion and Supreme Court decisions: Was Mr. Dooley right? Qcmmnnicaticnflesearch, 5(2), 131-149. Gitlin, T. (1980). The Whale flcrld is Watching. Berkeley, CA: University of California Press. Granger, C.W.J. (1969). Investigating causal relations by econometric models and cross-spectral methods. Eccncmetrica, 31(3), 424-438. Greenberg, B. S., and Parker, R. (1965). The Kennodg Assassination and theAmericanPublic: SocileommunicationinCrisis Stanford. CA: Stanford University Press. Heinlein, R.A. (1960/1980). “Pravda” means “truth.” In R.A. Heinlein, ExpandedHniyerse. New York: Ace. Herbst, S. (1991). Classical democracy, polls, and public opinion: Theoretical frameworks for studying the development of public sentiment. Communication Theorg. 1(3), 225-238. Hibbs, D A. (1974). Problems of statistical estimation and causal infertence in time-series regression models. In Costner, H. L. (Ed. )Sociclcgical WW. San Francisco. Jossey-Bass. Hibbs, D.A., with Rivers, RD, and Vasilatos, N. (1982). The dynamics of political support for American presidents among occupational and partisan groups. Americanlournalof Poht1ca1Sc1onco. 25(2). m6 Holsti, O.R., and Rosenau, J .N. (1980). Does where you stand depend on when you were born? The impact of generation on post-Vietnam foreign policy beliefs. Public Cpinion Quarterly. 44(1). 1-22. Hovland, C. I. (1959). Reconciling conflicting results derived from experimental and survey studies of attitude change. The American Psgdiologist.14(1),&17. Hume. D. (1748/1955). AnlnouirgConcerningHumanllnderstanding Indianapolis: Bobbs-Merrill. Hunter, J .E. (1977). Cluster analysis. Paper presented to US. Civil Service Commission workshop on advanced statistics. Hunter, J. E., Schmidt, F..,L and Jackson, GB. (1982). WE; CumulatingResoarchEindingsAcrossStudies Beverly Hills= Sage. Iyengar, S. (1991).1s Ammne Responsible? Chicago: University of Chicago Press. Iyengar, S., and Kinder, DR. (1987). NensthatMatters. Chicago: University of Chicago Press. Iyengar, S., and Simon, A. (1993). News coverage of the Gulf crisis and public opinion. Communication Research. 2161365383. James. B. (1988). IheBillJamesRasehallAhstractzm New York: Ballantine. Johnston, J. (1970). EconemetricMethgds. New York: McGraw-Hill. Jones. H. and Iniss.B..C (1978).Eomcasnng1echnolog11£orPlanning Decisicns. New York: Petrocelli. Kaplowitz, S.A., Fink, E. L., D'Alessio, D., and Armstrong, G. B. (1983). Anonymity, strength of attitude, and the influence of public opinion polls. HumanCommunicationResearchilml). Katz, E. and Lazarsfeld, RF. (1955). PersonallnfluencnlhePartPlagcd MPeopleintheElonofMassCommunications. Glenooe. IL: Free Press Keely, J. (1971). TheLeft;LeaningAntenna. New Rochelle, NY: Arlington House. Kenny. DA. (1979). CorrelationandCausalitg New York: Wiley and Sono- Kim, M. and Hunter, J. E. (1993a). Attitude-behavior relations. A meta- analysis of relevance and topic. Jpnrnal chpnnnnnicaticn, $3(1),101-142. 207 Kim, M. and Hunter, J. E. (1993b). Relationships among attitudes, behavioral intentions, and behavior. Qcmmnnicaticn Research, m(3), 331- 364. King, M.C. (1992). Occupational segregation by race and sex, 1940-1988. MonthlgLahorRoinon. April, 1992. Klapper. J. (1960). The Eflscts of Mass Communication New York: Free Press. Klein, M.W., and Maccoby, N. (1954). Newspaper objectivity in the 1952 campaign. Journalistuarterlg. 31(3). 285-296. Krippendorf, K. (1980). MW. Beverly Hills, CA: Sage. Krosnick, J .A., and Brannon, LA. (1993). The media and the foundations of Presidential support. ,Lcnrnal of Social Issnes, 49(4), 167-182. Krull, R., and Paulson, A. S. (1977). Time series analysis in communication research. In P.M. Hirsch, P.V. Miller, and EC. Kline (Eds.,) Strategiesfcr Communication Research Beverly Hills. CA: Sage Lang, K., and Lang, G. (1974). The unique perspective of television and its effect: A pilot study. In W. Schramm and D. F. Roberts (Eds. ), The Rrpcess and Efiects of Mass Communication. Urbana. IL: University of Illinois Press. Lang, K., and Lang, GE. (1968). EcliticsandTeiflisicn, Chicago: Quadrangle Books. Lefkowitz, M. Eron, L., Walder, L, and Huesmann, L. (1972). 'lblevision violence and child aggression: A follow-up study. In National Institjne cf Mental Health. Tologision and Social Behagior. Reports and Papers 'Ibchnical report to Surgeon General’s Scientific Advisory Committee on 'lblevision and Social Behavior, Vol. III. Linsky.M. (1986). ImpaanonthePressAtIectsEederalPolicgmaking New York: W.W. Norton. Lippmann, W. (191411965). A Ereface chlitics. Ann Arbor, MI: University of Michigan Press. Lippmann, W. (1922/1991). Enhlichinicn. New Brunswick, NJ: 'Ii'ansaction Press. Lowry, D.T. (1974). Measures of network news bias in the 1972 presidential campaign. Journal of Broadcasting. 18(4). 387-402. m8 Luttbeg, NR, and Gant, M.M. (1983). The failure of liberal/conservative ideology as a cognitive structure. Pnhlichinien Quarterlr, 49(1), 80-93. MacKuen, M. B. ( 1983) Political drama, economic conditions, and the dynamics of presidential popularity American Journal ofPolitical Science. 21(2),165-192 Mankiewicz, F. (1989). From Lippmann to Letterman: The 10 most powerful voices. CannottCentochurnal. 3(2). Martin, L.J. (1976). Recent theory on mass media potential in political campaigns. In Martin, L.J. (Ed.) Role of the Mass Media in American Politics AnnalsoftheAmericanAcademgofPoliticalandSocialSnence. 421 (September), 125-134. Marx, K. (1844/1976) “Economic and Philosophical Manuscripts” in Early Writings, New York: V'mtage Books. McAneny, L. (1995). Annual honesty and ethics poll. Gallup Roll Mcnthlg, No. 362., 28-32. McCombs, M. E. (1982). The agenda-setting approach. In Nimmo, D., and Sanders. K., (Eds. ) Handbook of PohticalCommumcation. Beverly Hills: Sage. McCombs, M. E. and Shaw, D. L. (1972). The agenda-setting function of the mass media. PublicCpinionQuarterho 36(2).176-187. McLuhan. H.M. (1964). UnderstandingMeditheExtensionsofMan. New York: Signet. MeyrowitZ. J. (1985). NoSenseofflacmTheImpactofElectronicMediaon Smcial Beharicr. New York: Oxford University Press. Milavsky, J. R., Kessler, R. C., Stipp, H. H., and Rubens, WS. (1982). TologisionandAggressiomA PanelStiidg. OrlandO. FL: Academic Press. Miller, GR, and Nicholson, HE. (1976). Ccmmnnicaticnlndnirng Eerspoctiveona Bracess. Reading, MA: Addison-Wesley. Mineo, RJ. (1991). Personal communication to author. Monge, P. (1973). Theory construction in the study of communication: The system paradigm Journal of Commumcation. 21(1), 5-16. Mueller, J. E. (1970). Presidential popularity from Man to Johnson. The AmoricanPoliticalScienceReiden. 54(1),18-34. 209 Mueller, J. E. (1979). Public expectations of war during the cold war. American Journal of PolitimlScionce. 21(2). 301-329. Mueller, J. (1988). 'Ii'ends in political tolerance. RnhlichinionQnaflerlg. 3(1): 1'25- NenXorkTimeslndex. New York: The New York Times Co. Nie, N.H., Hull, C.H., Jenkins, J .G., Steinbrenner, K., and Bent, DH. (1975). StatisticalPackageforthoSonalSciences. 2nd. Ed. New York: McGraw-Hill. Nunn, C.A., Crockett, H.J., and Williams, J .A. (1978). Toierancefor Nonoonformity. San Francisco: Jossey-Bass. Oldenburg, RC. (1989). Theflreatfiood Place. New York: Paragon House. Ostrom. CW. (1978). IimoSonosAnalysimRegressionSlbchnioues. Beverly Hills, CA: Sage. Page, 3.1., and Shapiro, R.Y. (1982). Changes in Americans’ policy preferences, 1935-1979. Enhlic Qpinion Quarterly, fl 1), 24-42. Page, B. I, and Shapiro, R. Y. (1983). Effects of public opinion on policy. The American PoliticalScionce Region. 11(1), 175-190. Page, 8.1., and Shapiro, R.Y. (1992). TheRationaanhlic. Chicago: University of Chicago Press. Page, B. I, Shapiro, R. Y., and Dempsey, G. R (1987). What moves public Opinion? American PoliticalScionce Region. 81(1), 2343. Pan, Z., and Kosicki, G. M. (1996). Assessing news media influences on the formation of whites’ racial policy preferences. Communication Research, 21(2), 147-178. Patterson, TE. (1993). Qnt onrder. New York: Alfred A. Knopf. Perry, D.K. (1990). News reading, knowledge about, and attitudes toward foreign countries. Journalistuartcrlg. 51(2). 353-358. Petty, RE. ,and Cacioppo, J .T. (1984). The effects of involvement on responses to argument quantity and quality. Journal ofEerspnality and Social Psychology. 46(1). 69-81- Price. V. (1992). CommunicationConcepts 4:Piiblic Opinion Newbury Park, CA: Sage. U 210 Readers’ Guide to Periodical literature. New York: H.H. Wilson. Robinson, M.J. (1985) The media in campaign ‘84: Part II: Wingless, 'Ibothless, and hopeless. Ruhlichinion, 2(1), 43—48. Rogers, E. M, and .Dearing, J. W. (1988). Agenda setting research: Where has it been, where is it going? In Anderson, J. A., (Ed. ) Communication Yearbook 11. N ewbury Park, CA: Sage. Rogers, E. M, Dearing, J. W., and Bregman, D. (1993). The anatomy of agenda-setting research. Journal ofCommunication. 43(2). 68-84 Rogers, E. M. Dearing, J. W., and Chang, S. (1991). AIDS in the 1980’s: The agenda-setting process for a public issue. Journalism Monographs, 1% (whole). Russo, ED. (197 2). A study of bias in TV coverage of the Vietnam war: 1969 and 1970. Public Opinion Quarterly. £4). 539-543. Schlesinger, A.M. (1939). Tides of American politics. TheXale Region, M2), 217-230. Schneider, W., and Lewis, LA. (1985). Views on the news. Ruhthpinion, 2(4), 6-11. Sharp, EB. (1992). Agenda setting and policy results: Lessons from three drug policy episodes. Policy Studies Journal, 21(4), 538-551. Shoemaker. F.J.. and Reese. SD. (1991). Mediating the Messaage: Theories of Influences on Mass Media Content. New York: Longman. Simon, HA. (1979). Models of Thought. New Haven, CT: Yale University Press. Smith, KA. (1987). Newspaper coverage and public concern about community issues: A time series analysis. JournalismMonographs, 101 (whole). Smith, T. W. (1990). Liberal and conservative trends in the United States since World War II. Ruhlic Qpinion Quarterly, $01), 479-507. Smith, T. W. (1990a). 'Ibchnical supplement to “Liberal and conservative trends in the United States since World War 11”. Chicago: National Opinion Research Center. Smith, T.W. (1991) Personal communication to author. 211 Stemple, G. H. (1965). The prestige press in two Presidential elections. Journalistuarterly. gm. 15-21 Stemple, G. H. (1969). The prestige press meets the third-party challenge. Journalism Quarterly. 4841699706. Stemple, G.H., and Windhauser, J .W. (1989). Coverage by the prestige press of the 1988 Presidential campaign. Journalism Quarterly, Q4), 894-896, 919. Stevenson, R. L., Eisinger, R. A., Feinberg, B. M., and Kotok, A. B. (1973). Untwisting The News Twisters: A replication of Efron’ s study. Journalism Quarterly. 59(2), 211-219. Stevenson, R. L., and Greene, M. T. (1980). A reconsideration of bias in the news JournalismCuarterlg. 51(1).1980. Stone, G.C., and McCombs, ME. (1981). 'II'acing the time lag in agenda setting. Journalistuarterly. 58(1). 51-55. Stovall, J .G. (1988). Coverage of 1984 Presidential campaign. Journalism Quarterly. EM). 443-449. 484. Sullivan, J. L., Piereson, J, and Marcus, G. E. (1979). An alternative conceptualization of political tolerance: Illusory increases 1950’s- 1970’s. TheAmoricanPoliticalScionceRonen. 13(4), 781-794. Tapper, J. (1995). The ecology of cultivation. CommunicationTheom, 5(1), 3657. Taylor, D. G. (1980). Procedures for evaluating trends in public opinion. Public Opinion Quarterly. 44(1), 86-100. Thibault, J .W., and Kelley, H.H. (1959). TheSocialeyohology offiroups. New York: Wiley and Sons. Thompson. KS. (1973). EearandLoathing: CntheCamnaignIIail‘12. New York: Popular Library. Thornton, A., Alwin, D. F., and Camburn, D. (1983). Causes and consequences of sex-role attitudes and attitude change. American Sociological Region. 58(April). 211-227. 'li'umbo,C ..(1995) Longitudinal modeling of public issues: An application of the agenda-setting process to the issue of global warming. Journalism W 152 (whole). 212 US. Bureau of Census. StatisticalAbstractofthoCnitedStatos. Washington, DC: US. Government Printing Office. US. Department of Health and Human Services (1991). yital Statistics of theflnitedfitates. 1%. Washington, DC: Government Printing Office. US. Department of Labor, Women’s Bureau (1993). Earning differences between women and men. Report No. 93-5. VedlitZ. A. (1988). ConsomaligeMnliologyandPublicPohminAmerica New York: Praeger. von Bertalanfi‘y. L. (1968). ConoraISystemThoom Foundations. DegdopmonLApplications New York: Braziller Wanta, W., and Foote, J. (1994). The President-news media relationship: A time series analysis. JournaloeroadcastingandElectronicModia. m4). 437448. Watt, J .H., and van den Berg, S. (1981). How time dependency influences media effects in a community controversy. Journalism Quarterly, 58(1), 43- 50. Weaver, D. (1984). Media agenda-setting and public Opinion: Is there a link? In Bostrom. R.N.. (Ed) CommunicationXearbookfi. Newbury Park. CA: Sage. Weaver, D., and Elliott, SN. (1985). Who sets the agenda for the media? A study of local agenda-building. Journalism Quarterly, m0), 87-94. White, D. M. (1950). The gate-keeper: A case study in the selection of news. Journalism Quarterly. 21(3), 383-390. White, TH. (1961). The Making of the President m. New York: New American Library. White, TH. (1965). The Making of the President 1%. New York: New American Library. Zaller. J .R. (1992). TheNature and CriginsofMasstinion. Cambridge. UK: Cambridge University Press. Zhu, J. Watt, J. H, Snyder, L. 3., Yan, J, and Jiang, Y. (1993). Public issue priority formation: Media agenda-setting and social interaction. Journal of Communication. 43(1), 8-29. iliiiliiillilll