The Cultivation of Consumer Confidence: a Longitudinal Analysis of News Media Influence on Consumer Sentiment

Albert R. Tims, University of Minnesota
David P. Fan, University of Minnesota
John R. Freeman, University of Minnesota
ABSTRACT - This study explores the ability of a new model of the dynamic influence of information on the distribution of sentiment in large populations (Fan, 1988) to predict the movement in the University of Michigan's Index of Consumer Sentiment between 1978 and 1988. Based on a detailed content analysis of 2,000 news stories about the economy, mathematical predictions of the influence of the news environment on the distribution of consumer sentiment were calculated for a period of 127 months. The results showed that the model yielded an excellent prediction of the secular trends in the Index. The findings are discussed in terms of contemporary research on the role of the media in shaping public perceptions of social conditions.
[ to cite ]:
Albert R. Tims, David P. Fan, and John R. Freeman (1989) ,"The Cultivation of Consumer Confidence: a Longitudinal Analysis of News Media Influence on Consumer Sentiment", in NA - Advances in Consumer Research Volume 16, eds. Thomas K. Srull, Provo, UT : Association for Consumer Research, Pages: 758-770.

Advances in Consumer Research Volume 16, 1989      Pages 758-770

THE CULTIVATION OF CONSUMER CONFIDENCE: A LONGITUDINAL ANALYSIS OF NEWS MEDIA INFLUENCE ON CONSUMER SENTIMENT

Albert R. Tims, University of Minnesota

David P. Fan, University of Minnesota

John R. Freeman, University of Minnesota

[This research was funded in part by U.S. Public Health Service Research Grant MH-39610. The authors wish to thank T. Kjellstrand for his service as research assistant for this project.]

ABSTRACT -

This study explores the ability of a new model of the dynamic influence of information on the distribution of sentiment in large populations (Fan, 1988) to predict the movement in the University of Michigan's Index of Consumer Sentiment between 1978 and 1988. Based on a detailed content analysis of 2,000 news stories about the economy, mathematical predictions of the influence of the news environment on the distribution of consumer sentiment were calculated for a period of 127 months. The results showed that the model yielded an excellent prediction of the secular trends in the Index. The findings are discussed in terms of contemporary research on the role of the media in shaping public perceptions of social conditions.

INTRODUCTION

The emergence of empirical research addressing questions about the cumulative effects of media message environments marks a major departure from the well-worn and sometimes rambling path of media effects studies: largely charted by theoretical and empirical emphasis on rather specific interactions between source, message, channel, and receiver characteristics (Bryant & Zillmann, 1986). The two areas of inquiry best representing this new effects perspective are research on the agenda setting influence of the news media on public perceptions of issue importance (McCombs & Gilbert, 1986) and "cultivation" research on the distortions in public perceptions of social reality resulting from the effects of cumulative exposure to entertainment television's depictions of society (Gerbner, et al., 1986).

Agenda setting research has essentially asked the question: does the amount and prominence of coverage given to issues in a community by the news media influence public perceptions of the importance of issues? Although agenda setting research emphasizes the ability of the news media to transmit cues about what's important and what's not so important -- rather than the ability of the news media to determine where an individual, or a community, stands on a given issue -- it does not explicitly deny the possibility that the news media may have a role to play in shaping public opinion.

While research on the agenda setting function of the press and research on entertainment television's influence in cultivating perceptions of social reality are quite distinct areas of inquiry, they have far more in common than one might expect. Like agenda setting research, cultivation research raises questions about the cumulative impact of media messages. Also like agenda setting research, it has set about the task of identifying cultivation effects by using content analysis to document dominant themes in media messages in combination with surveys of public perceptions of social conditions (Gerbner, et al., 1986). Most important of all, cultivation research also recognizes the central role of the media in society as the conduit through which much of what we know, or think we know, about the world outside our own direct experience is transmitted.

Because the original research questions about cultivation effects were so closely linked to entertainment television's ability to influence perceptions of social reality, cultivation effects have rarely been examined in association with other forms of mass communication. There is nothing inherent in the nature of the concept, however, that limits its application to questions about the effects resulting from the depictions of society in entertainment television. It becomes a far more powerful concept if it is applied more broadly to represent the influence of the media message environment -- not just commercial television -- on perceptions and evaluations of social conditions beyond the scope of personal experience.

The messages reaching the public from the news media, like those of entertainment television, are shaped by a multitude of institutional, political, and economic constraints. While the public generally accepts the notion-that the news media provide a veridical reflection of events, this faith is not generally shared by those who understand the news business. Heavy reliance on "official" news sources, economic incentives to maximize audiences, journalistic norms emphasizing the news value of dramatic and conflictual events, the tyranny of deadlines and time/space limitations, and a host of other factors all come into play in determining what reaches the public as the news of the day.

The present study of the cultivation of consumer sentiment is based on the premise that perceptions of economic conditions in society are largely determined by the information and interpretations provided by the news media. Our position is that what many would perceive to be public reactions to conditions and events in the world are, in a more direct sense, reactions to the representation and interpretation of events presented to the public by the media as news. The meaning of a change in interest rates, for example, is not something any of us outside the confines of large financial institutions instinctively understand. We are highly dependent upon the news media to generate the messages that alert us to the change; that provide the contextual cues and interpretations we need to understand the significance of the event for the economy as a whole; and that might prove useful to us in helping guide personal decisions we may want to make about savings, investments, consumer loans, and the like. As DeFleur and Ball-Rokeach (1982) point out, this dependency is a natural outgrowth of our modern day existence in highly complex social systems.

An adequate test of this model requires that we have (1) indicators of changes in consumer sentiment over time, (2) a means of identifying the volume and nature of media messages about economic conditions introduced in the social system over time, and (3) a model for predicting the impact changes in the message environment will have on consumer sentiment about economic conditions. Fan's (1984, 1985, 1988) recent work on a mathematical theory of the persuasive influence of information on large populations provides a methodological and theoretical framework well-suited for the task.

In a series of recent empirical investigations, Fan's ideodynamic theory has shown that changes in public opinion about a wide range of social issues such as defense spending, nuclear energy, presidential approval, Contra aid, and the spread of AIDS can be predicted with a high degree of precision over extended periods based on predictions generated from an analysis of characteristics of the messages about these issues reaching the public at large via the news media (Fan, 1988; Fan, McAvoy & Freeman, 1988; Fan & Tims, 1988).

In the present study we have applied Fan's model to explore the linkage between changes in the economic news environment and changes in the distribution of consumer sentiment. If the messages reaching the public about economic conditions can indeed be used to predict changes in consumer sentiment then we will have taken the first step in establishing a broader foundation for understanding the importance of the news media as a major player in helping to shape patterns of consumer sentiment and consumer behavior.

METHOD

Index of Consumer Sentiment

The Index of Consumer Sentiment was originally formulated by George Katona, a pioneer in the study of behavioral economics, as "a summary measure of trends in consumers' attitudes and expectations" (Curtin, 1983, p. 498). The measure is composed of five items asking for evaluations of changes in personal finances, short- and long-range expectations for business conditions, and buying conditions for large household goods (see Table 1).

The history of the University of Michigan's Surveys of Consumers dates back to the mid-1940s (Curtin, 1982). The sample sizes for the monthly surveys since 1978 vary between 500 and 1300+. Since 1985 the monthly samples have ranged between 500 and 700. The values for the Index of Consumer Sentiment used in this study were taken from the "Surveys of Consumers: Historical Data" and the July, 1988 update of the "Surveys of Consumers" (Institute for Social Research, University of Michigan, 1988).

Content Analysis of Economic News

A search of the Nexis electronic data base identified a total of 38,279 news stories from the Associated Press wire on the subject of economy in combination with the key root words: condition, health, prospect, future, forecast, outlook, projection, and variants of these roots during the period between January 1, 1977 and August 25, 1988. The Associated Press wire is the nation's largest and most extensive news gathering and reporting organization. The AP is received by over 1,400 daily newspapers and over 6,000 broadcast stations in the United States; it serves 84 percent of the daily newspapers in the United States and reaches 96 percent of the newspaper readers through these newspapers (Shaw, 1988).

A random sample of 2,000 full text stories (5% of the population), containing approximately 500,000 words, was retrieved from the data base and organized into 100 text fries. Each of the 100 files contained 20 stories ordered in random sequence.

The retrieved text was first filtered to remove stories originating in other countries or about economic issues in regions and countries other than the United States (see Table 2 for a list of the most frequently encountered character clusters used to identify news stories not about economic conditions in the United States). Stories containing character clusters indicative of a foreign news theme were deleted if the character clusters appeared either in, or within 60 characters, of the headline, dateline, or keyword index for the AP wire dispatch.

The stories remaining after the foreign news filtration (1,252) were subjected to a second filtration to identify paragraphs within stories containing character clusters representing economic terms. Table 2 contains examples of the character clusters most frequently encountered during this second filtration. All paragraphs containing words or phrases containing these character clusters were retained for content scoring. Paragraphs not containing at least one instance of any economic term character cluster were eliminated from the analysis. The first and second filtrations resulted in the elimination of 41 percent of the retrieved text.

Be text remaining after the two successive filtrations was analyzed using a complex set of relational rules and key words. Table 2 lists a partial dictionary of the most commonly encountered character clusters and terms used for the content scoring. The terms in the dictionary were grouped into broad concept categories. For example, terms indicating increases (boost, growth, gain, climb, hike, etc.) were grouped under one concept and terms indicating decreases (crash, cut, decrease, drop, tight, etc.) were grouped under another. Additional categories were formed for words indicating unhealthy economic conditions, healthy economic conditions, etc. In addition, words that changed the meaning of a statement (not, unlikely, few, little, etc.) were also used in the development of relation scoring rules. For example, the statement "under current economic conditions an increase in the rate of inflation is not expected in the future" was coded as favorable news about economic conditions since the scoring system was designed to consider the significance of the word not in the statement. Relational operators in the scoring system specified how any of the terms in one cluster would act on any of the terms in another cluster in evaluating how each paragraph would be scored.

TABLE 1

ITEMS IN THE INDEX OF CONSUMER SENTIMENT

TABLE 2

EXAMPLES OF CHARACTER CLUSTERS AND TERMS USED IN TEXT FILTRATION AND SCORING

The computer text analysis, using the scoring rules and concept categories described above, was applied to several of the randomly sequenced text files. The computer scoring was then compared to the scoring decisions of independent human coders. This process was repeated until the computer scoring rules yielded coding decisions comparable to those of the human coders.

Each paragraph was scored for favorable and unfavorable information about the economy. If a paragraph contained a passage identified as having either favorable or unfavorable information about the economy it was scored as containing an "infon" pressure in that direction. If a paragraph contained both favorable and unfavorable information, the paragraph was scored for both positions. A total of 4,282 paragraphs contained in 1,252 news stories were coded as containing either favorable or unfavorable infons. Eighty-two percent (1,027) of the stories had at least one paragraph scored as containing favorable economic news, and 84 percent (1,047) of the stories had at least one paragraph scored as containing unfavorable economic news. The total score for all favorable infons was 2,082 paragraphs. The comparable figure for unfavorable infons was 2,183 paragraphs.

Ideodynamic Modeling of Consumer Sentiment

The infons resulting from the content analysis described above are used as the basis for the mathematical modeling of the persuasive influence of the information environment (Fan, 1988). The infon pressures in both directions entering the system during every 24-hour period are added together to represent the messages pressure on public sentiment about the economy. The predicted movement of public sentiment resulting from the message pressures is operationalized as a function of the opposing pressures (favorable messages versus unfavorable messages) operating over time. The influence of each AP paragraph was hypothesized to decay at a rate of about 50 percent per day, so that after about one week a specific message is no longer capable of exerting influence on the subsequent movement of public sentiment. This average decay rate has been empirically validated in a series of studies reported by Fan (1988).

The mathematical model further specifies that the message pressures only influence opposing sentiments. The rationale for this assumption is quite simple. If a segment of the population is already categorized as having a favorable outlook about the economy, the introduction of favorable message pressures will not push them out of this category. On the other hand, the introduction of favorable message pressures will operate on the segment of the population with an unfavorable outlook about the economy. The general form of the model is:

Ft = Ft-1 + k'(GFt * Ut-1 - GUt * Ft-1)

where: F is favorable sentiment, k' is a constant, GF is message pressure on favorable sentiment, U is unfavorable sentiment, GU is message pressure on unfavorable sentiment, and t is time.

The constant in the equation is introduced to control the volatility of the predictions generated by the model. The actual mathematical formula used in the opinion predictions is somewhat more complicated since the GF and GU parameters take into account residual half-life influences of messages (see Fan, 1988, for a full discussion of the mathematical model).

RESULTS AND DISCUSSION

Figure 1 depicts the total number of messages analyze-d, aggregated by month, for the period beginning January, 1977 through August, 1988. Although the initial sample of 2,000 news stories should have been adequate to provide a good base for representing economic news content during the entire period, it seems clear from Figure 1 that AP wire coverage of the economic issues captured by this analysis increased substantially after 1981. As a consequence, fewer than 10 messages were captured during most of the months prior to the second quarter of 1979. While this is likely to be an accurate reflection of actual news densities -- given that stories were sampled at random rather than systematically -- it does suggest that the sampling fraction (2,000/38,_79 or 5%) may not have been large enough to provide desired reliability for predictions of media impact on consumer sentiment for periods during which the economic news density was extremely low.

Figure 2 shows the initial coding of the paragraphs as either favorable news about the economy or unfavorable news about the economy. This simple analysis is an excellent depiction of the conflictual nature of news reporting. While it is possible to detect periods during which good news was more prevalent than bad news, and vice versa, the most clearly observed pattern is the tendency for increases in news coverage to yield increases in both good and bad news. This finding adds considerable insight into the findings reported in agenda setting studies (McCombs & Gilbert, 1986). As an issue becomes more salient in the news media by virtue of increased coverage, so does the salience of the conflict about the issue. Indeed, subsequent studies may well find that it is not simple news salience that leads to agenda setting effects but issue conflict made salient by the news coverage.

The ideodynamic model starts with the raw count of paragraphs scored for favorable and unfavorable infons as a basis for the calculation of "infon pressures." If a particular piece of news is not repeated or given additional coverage its impact on the message environment is reduced to zero after several days. Figure 3 contains the transformations of the raw infon scores aggregated for each month into infon pressures. The results indicate that message pressures in one direction tend to be countered by message pressures in the opposite direction. The zero-order correlation between the two infon pressures is .71 (p < .001).

FIGURE 1

TOTAL MESSAGES ABOUT THE ECONOMY

FIGURE 2

FAVORABLE AND UNFAVORABLE MESSAGES

The net pressure on public sentiment from the media environment at any given point in time can be seen by simply examining the difference between the two infon pressures (see Figure 4). The analysis of net message pressures shown in Figure 4 suggests that the message pressures on public sentiment about the economy varies substantially during the period under investigation. Although this analysis provides a very useful summary of the characteristics of the economic news environment, it does not directly translate into changes in public sentiment resulting from the news environment.

Using the value of the Index of Consumer Sentiment during the first quarter of 1977 as a starting point, predictions for the Index were calculated using the cumulative impact of the infon pressures shown in Figure 4. The calculations were based on successive evaluations for each day during the period. Figure 5 shows the results of these predictions averaged by each month between January, 1978 and August, 1988. The figure does not show the calculations for 1977 since the Index of Consumer Sentiment was not gathered on a monthly basis prior to January, 1978.

Based only on the influence of news coverage of economic issues by the Associated Press the resulting infon pressures were used to predict the Index of Consumer Sentiment over a 127-month period. Since the ideodynamic model uses no information about prior values of the actual Consumer Sentiment Index beyond an initial starting point, it is entirely possible for the ideodynamic predictions to move off in directions unrelated to the movement of the Consumer Sentiment Index. Figure 5 shows that the Consumer Sentiment Index was predicted to show a strong downturn between 1978 and 1980. It was predicted to show a brief improvement followed by another decline during 1981 and a period of recovery between 1982 and 1984. After 1984, the prediction was for the Index to remain relatively high until mid-1986, after which a gradual downturn was predicted.

Figure 6 shows the actual values of the Index of Consumer Sentiment reported by the Institute for Social Research at the University of Michigan from January, 1978 through July, 1988. A direct comparison of the Index and the ideodynamic predictions for the Index is shown in Figure 7. The correlation between the consumer surveys (n = 127) and the ideodynamic predictions is .70 to < .001). These findings support the prediction that changes in the distribution of favorable and unfavorable news about the economy can be used to predict secular changes in consumer sentiment.

The Index of Consumer Sentiment is often broken into an Index of Sentiment about Current Conditions and an Index of Sentiment about Expected Conditions. Figure 8 shows the values of the Current Index, Expected Index, and the ideodynamic predictions. When the Index of Consumer Sentiment is broken out in this fashion it becomes clear that the variations in the ideodynamic predictions typically fall within the boundaries of the variation between the Current and Expected components of the composite Index. Overall, the ideodynamic predictions are a somewhat more accurate model of the Index of Consumer Sentiment about Current Conditions (r = .76, p < .001) than the Index of Consumer Sentiment about Expected Conditions (r = .62, p < .001).

DeFleur and Ball-Rokeach (1982) note:

... when media messages help to create an affective state of fear about one's own and the nation's economic future or the belief that a depression is unavoidable, people may not buy stocks, new cars, certain foods, or a multitude of other products that they would have otherwise bought (p. 275).

What we have shown is that the news media are doing more than simply communicating news about economic events; they are defining the meaning of the events. In turn, the media definitions appear to serve as excellent guides to the distribution of public sentiment about the economy.

In terms of a broadly defined cultivation hypothesis, these findings are equally significant. We have defined cultivation as the ability of the media message environment to have systematic influences on perceptions of social reality. Unlike traditional cultivation studies, which have been limited to attempts to document "distortion" of social perceptions, our approach is to demonstrate that changes in the distribution of message pressures can be used to model changes in the social perceptions of large populations.

The ideodynamic model recognizes that the media environment is not static -- the pressures it exerts are neither uniform nor unidirectional. It leaves open the possibility that the message environment contributes to the distortions so elegantly outlined by Gerbner and his associates (Gerbner, et al., 1986). It seems likely that, when organizational and economic forces operate to constrain the news gathering process and/or the content attributes of entertainment programming, systematic distortions may result. (See Bennett (1988) for a detailed discussion of this subject.)

Cultivation theory acknowledges the cumulative impact of messages on perceptions of social reality. It acknowledges the ability of media representations to have influence beyond heightened awareness. In these respects the theory is quite similar to ideodynamic theory. Ideodynamic theory differs in its emphasis on the dynamic influence of message pressures. In reporting news about the economy the news media do more than simply tell us that unemployment has increased; they tell us if it's okay because it is a temporary change resulting from success in dealing with some other economic concern or if it signals a serious economic problem. In reality, as we have shown, changes in economic conditions will result in conflicting interpretations. What ultimately matters is if the prevailing message environment favors one side or the other of the conflict. From our perspective, it is the news media's playing out of the issue conflict that influences the public's perception of social reality.

FIGURE 3

FAVORABLE AND UNFAVORABLE INFON PRESSURE

FIGURE 4

INFON PRESSURE ON CONSUMER SENTIMENT

FIGURE 5

INFON PREDICTIONS OF CONSUMER SENTIMENT

FIGURE 6

INDEX OF CONSUMER SENTIMENT (ISR)

FIGURE 7

IDEODYNAMIC PREDICTION AND CSI

FIGURE 8

CURRENT, EXPECTED, AND IDEODYNAMIC INDEX

Perceptions of the Economic News Environment

The Surveys of Consumers Series conducted by Michigan's Institute for Social Research includes several measures beyond those used for the Index of Consumer Sentiment. One of these is a measure asking respondents: "during the last few months, have you heard of any favorable or unfavorable news about business conditions?" The question format permitted respondents to mention hearing both favorable and unfavorable news. Although this measure does not entirely capture the scope of the Index of Consumer Sentiment nor fully represent the domain of the content analysis used for the ideodynamic predictions, it does provide a useful means for determining if changes in the perceived news environment covary with the Index of Consumer Sentiment and with the content analysis of economic news used for the ideodynamic predictions.

For the present study this item was converted to an index using the same procedure used in constructing the Index of Consumer Sentiment. Mentions of unfavorable news were subtracted from mentions of favorable news and added to 100. Thus when favorable news is exactly equal to unfavorable news the index has a value of 100. When unfavorable news is more often mentioned than favorable news the index has a value less than 100. Figure 9 presents the scores for this News Index. The perceived news environment represented by this index shows considerable variation, ranging from a low score of 10 in early 1980 to a high score of over 150 in early 1983. A sharp change in the perceived business news environment is evident during the second half of 1982.

Figure 10 shows the standardized values of the Index of Consumer Sentiment compared to the standardized values of the News Index (r = .78, p < .001). These findings show that the respondents in the consumer surveys were indeed aware of changes in the news environment and that substantial covariation exists between the perceived economic news climate and the level of consumer sentiment. This represents confirming support for our position that the movement of public sentiment is closely linked to the salience of favorable versus unfavorable economic news reaching the public.

A comparison of the perceived News Index and the ideodynamic predictions of consumer sentiment are shown in Figure 11 (r = .54, p < .001). The findings reported in Figure 11 are quite impressive keeping in mind: that (1) the ideodynamic predictions for each month are based on fewer than 10 news stories (on average); (2) the predictions were designed to model the Consumer Sentiment Index rather than favorable versus unfavorable business news; and (3) the news heard measure asks a very global question about hearing favorable or unfavorable news over the past few months.

The secular trends in perceptions of the news environment generally correspond to the ideodynamic model's predicted pressures of the news environment on consumer sentiment. While the fit is not perfect, it does tend to provide further validation of the ideodynamic model and the methodology used in the content scoring system.

CONCLUSIONS

The findings reported in this study are representative of an emerging trend in media effects research toward systematic investigation of the influence of the media message environment on social change in large social systems. The demonstrated ability of the ideodynamic model to predict changes in the aggregate distribution of consumer sentiment over a span of more than 10 years lends support to our hypothesis that the message environment plays a major role in cultivating perceptions of social conditions. The success of the predictive model also contributes to the further validation of Fan's mathematical representation of the influence of information in social systems (Fan, 1988).

This research is neither a throwback to early conceptions of powerful direct effects of the mass media on individuals nor a challenge to more recent theories describing the active role of the individual in categorizing, filtering, and integrating information. We recognize: (1) that specific messages are not universally available; (2) that individual differences abound in the interpretation, personal relevance, and processing of messages; and (3) that personal circumstances and experiences are likely to play a significant mediating role in the process of message assimilation. Nonetheless, we also recognize: (1) that much of what we know about the larger social environment in which we live comes to us via mediated channels; (2) that the changes in the representation of social conditions in these channels provide a unique common source of "experience" for members of the social system; and (3) that changes in the distribution of sentiment in large social systems is not random. One need not deny the uniqueness of the individual to accept these notions or the position that a primary source of influence in the movement of public sentiment in large populations is the information environment. In this study we have shown that, when the persuasive characteristics of the message environment are rendered systematically observable, changes in that environment provide a powerful foundation for predicting changes in the distribution of public sentiment.

A number of methodological issues will need to be addressed before we can make definitive statements about causality. Obviously, we need to examine changes in leading economic indicators along with the ideodynamic predictions and the measures of consumer sentiment using autoregressive integrated moving-average (ARIMA) models before we can rule out many of the alternative explanations that might account for the observed relationship. An extensive program of research in this area is now underway.

FIGURE 9

NEWS HEARD ABOUT THE ECONOMY (ISR)

FIGURE 10

NEWS HEARD AND CONSUMER INDEX

FIGURE 11

NEWS HEARD AND INFON PREDICTIONS

REFERENCES

Bennett, W. (1988). News: the politics of illusion (2nd ed.). White Plains, NY: Longman, Inc.

Bryant, J., & Zillmann, D. (1986). Perspectives on media effects. (Eds.). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Cohen, B. (1963). The press and foreign policy. Princeton, NJ: Princeton University Press.

Curtin, R. (1983). Curtin on Katona. In H. Spiegel & W. Samuels (Eds.), Contemporary Economists in Perspective (Vol. 1). Greenwich, CT: JAI Press, Inc.

Curtin, R. (1982). Indicators of consumer behavior: the University of Michigan Surveys of Consumers. Public Opinion Quarterly, 46, 340-352.

DeFleur, M., & Ball-Rokeach, S. (1982). Theories of mass communication (4th ed.). New York: Longman, Inc.

Fan, D. (1984). Mathematical models for the impact of information on society. Political Methodology, 10, 479-494.

Fan, D. (1985). Ideodynamics: the kinetics of the evolution of ideas. Journal of Mathematical Sociology, 11, 1-24.

Fan, D. (1988). Predictions of public opinion from the mass media. Westport, CT: The Greenwood Press.

Fan, D., McAvoy, G., & Freeman, J. (1989). The effect of the media on public opinion: a new mathematical model and computer content analysis applied to the Bush-Dukakis race, etc. Paper presented to the American Political Science Association Annual Meeting, September, Washington, D.C.

Fan, D., & Tims, A. (1989). The impact of the news media on public opinion: American presidential elections 1987-1988. International Journal of Public Opinion Research, 1, 151-163.

Gerbner, G., Gross, L., Morgan, M., & Signorielli, N. (1986). Living with television: the dynamics of the cultivation process. In J. Bryant & D. Zillmann (Eds.), Perspectives on Media Effects (pp. 17 - 40). New York: Lawrence Erlbaum Associates, Inc.

Institute for Social Research. (1988). Surveys of Consumers- July, 1988 (Monitoring Economic Change Program). University of Michigan, Ann Arbor, MI.

Institute for Social Research. (1988). Surveys of Consumers - Historical Data (Monitoring Economic Change Program). University of Michigan, Ann Arbor. MI.

McCombs, M., & Gilbert S. (1986). News influence on our pictures of the world. In J. Bryant & D. Zillmann (Eds.), Perspectives on Media Effects (pp. 1 - 15). New York: Lawrence Erlbaum Associates, Inc.

Shaw, D. (1988, April 3). The AP: it's everywhere and powerful. Los Angeles Times (Reprint).

----------------------------------------