Robinson and Sheehan, 1983). The more popular recent view is that media influence is significant, but only in shaping the problems the public considers most important—their agendas (McCombs and Shaw, 1972). In some respects, agenda research challenges the minimal consequences view, but both approaches share a core assumption. Both assume audiences enjoy substantial autonomy in developing their political preferences. Research contradicting the notion that media have minimal consequences or only influence agendas has emerged during the 1980s (see, e. g. the pioneering yet disparate work of such authors as Bartels, 1985; Patterson, 1980; Iyengar and Kinder, 1987; and Page, Shapiro, and Dempsey, 1987; cf. Rob- The author gratefully acknowledgesfinancialsupport from the John and Mary R. Markle Foundation and the Institute for Research on Poverty, University of Wisconsin, and thanks this journal’s referees and editors for useful suggestions. JOURNAL OF POLITICS, Vol. 51, No. 2, May 1989 Portions of this article appear in DEMOCRACY WITHOUT CITIZENS: THE MEDIA AND THE DECAY OF AMERICAN POLITICS by Robert M.
Entman. © 1989 by Robert M. Entman. Used by arrangement with Oxford University Press, Inc. 348 Robert M. Entman inson and Levy, 1986). 1 But this burgeoning research has not yet generated a theory that explicitly refutes the assumption of audience autonomy and explains more fully the media’s impact on public opinion. This article probes the theoretical underpinnings of the autonomy assumption and provides empirical evidence that media messages significantly influence what the public thinks by shaping what they think about. THE RESEARCH TRADITION
The audience autonomy assumption provides the foundation for the minimal consequences position. The assumption is that audiences form their political opinions in relative independence from the media. There are two somewhat distinct variants of this position. The first emphasizes that audiences think about communications selectively, screening out information they do not like (Klapper, I960; cf. McGuire, 1985). The second holds that audiences pay so little attention and understand so little that the news cannot influence them (Neuman, 1986; cf.
MacKuen, 1984). 2 In practice, both the selectivity hypothesis and the hypothesis of inattention and incomprehension (hereafter just “inattention”) hold that media messages tend only to reinforce existing preferences rather than helping to form new attitudes or change old ones. Thus the media have little net impact on politics. The central assumption of the more recent agenda setting research has been that media do exert significant influence, but only in a narrow sphere.
In this view, the public’s autonomy is not complete, but its susceptibility to media influence is limited to agendas. Agenda research almost always includes a sentence like this: “Although a ‘minimal effects’ model most accurately describes the media’s ability to change opinions, recent research has shown that the media can play a much larger role in telling us what to think about, if not what to think” (Lau and Erber, 1985, p. 60; almost identical assertions appear throughout the literature, e. g. , McCombs and Shaw, 1972; MacKuen, 1984, pp. 72, 386; and even radical critiques such as Parenti, 1985, p. 23; also see MacKuen and Combs, 1981; Behr and Iyengar, 1985; Miller, Erbring, and Goldenberg, 1979). 3 Agenda scholarship does not provide a comprehensive theory that explains why media influence is confined to agendas, but selecDeFleur and Ball-Rokeach’s “dependency theory” (1982) describes an important theoretical alternative to the autonomy assumption, but that work predates most of the recent surge in empirical evidence. 2 Neuman (1986, chap. ) grounds his argument in the lack of evidence that media can teach specific information or enhance political sophistication. The concern in this paper is with political evaluations and preferences, which do not require much information—often a simple emotional response will do (cf. Abelson et al. , 1982). A related argument cites the public’s inability to recall specific stories. But the influence of a single news story or show is rarely of interest. The primary concern is the effect of repeated news messages over time (cf. Graber, 1984). But compare Iyengar and Kinder, 1987, and Protess et al. , 1987, for agenda setting research showing that media influence of agendas also shapes, respectively, the mass public’s criteria of political judgment and public officials’ behavior. 1 How the Media Affect What People Think 349 tivity and inattention again seem to be key. In the agenda setting view, the media can overcome these barriers in determining the issues people think about but not in shaping how they evaluate issues or candidates (the most explicit discussion is MacKuen, 1984).
The problem with the agenda setting position is that the distinction between “what to think” and “what to think about” is misleading. Nobody, no force, can ever successfully “tell people what to think. ” Short of sophisticated physical torture (“brainwashing”), no form of communication can compel anything more than feigned obeisance. The way to control attitudes is to provide a partial selection of information for a person to think about, or process. The only way to influence what people think is precisely to shape what they think about.
No matter what the message, whether conveyed through media or in person, control over others’ thinking can never be complete. Influence can be exerted through selection of information, but conclusions cannot be dictated. If the media (or anyone) can affect what people think about—the information they process—the media can affect their attitudes. This perspective yields an assumption of interdependence: public opinion grows out of an interaction between media messages and what audiences make of them. I will call this the “interdependence model. The competing positions, the minimal consequences and the agenda perspectives, both endorse the assumption that audiences form preferences autonomously. I will call this the “autonomy model. ” INFORMATION PROCESSING AND MEDIA IMPACTS Combining a recognition of the interdependence of audiences and media with information-processing models developed by cognitive psychologists may offer the best foundation for a new understanding (cf. Graber, 1984; Kraus and Perloff, 1985). There is no consensus among those who study information processing.
But a number of generalizations pertinent to the mass media’s impacts can be gleaned from their work. Information-processing research shows that people have cognitive structures, called “schemas,”4 which organize their thinking. A person’s system of schemas stores substantive beliefs, attitudes, values, and preferences (cf. Rokeach, 1973) along with rules for linking different ideas. The schemas “direct attention to relevant information, guide its interpretation and evaluation, provide inferences when information is missing or ambiguous, and facilitate its retention” (Fiske and Kinder, 1981, p. 73). Schemas are not filters used to select out all unfamiliar or uncomfortable information. As Bennett writes, “[I]nformation processing constructs [i. e. schemas] like party identification and ideological categories should not be reScholars have used many other terms, including “scripts,” “inferential sets,” “frames,’ and “prototypes. ” While there are subtle differences among them, they need not concern us here. The term schema is as good as any, and for clarity’s sake I use the English plural “schemas” instead of the awkward “schemata. 4 350 Robert M. Entman garded as rigid cognitive frameworks that work infixedways to screen out unfamiliar information” (Bennett, 1981, p. 91). Certainly people fail to think about much of the news, but not necessarily because they choose only congruent messages, or because they inevitably misunderstand or deliberately ignore media reports. Selectivity and inattention are stressed by the autonomy model, but that model fails to explain why many citizens do think about a great deal of the new information they encounter.
Information-processing theory recognizes and helps explain how attitudes emerge from a dynamic interaction of new information with peoples’ existing beliefs. In Bennett’s (1981, p. 92) words, political thought is “data-driven” by external information and “conceptually-driven” by internal schemas. Information-processing theory suggests that whether people ignore or pay attention to new information depends more on its salience, on whether it meshes with their interests, than on whether it conflicts with their existing beliefs (Markus and Zajonc, 1985, pp. 162 and passim; Kinder and Sears, 1985, pp. 710-12).
While people may resist knowledge that challenges their fundamental values (Axelrod, 1973), most can accommodate new information and even hold a set of specific beliefs that may appear dissonant, contradictory, or illogical to an outsider (cf. Lane, 1962). The explicit model of thinking that cognitive psychologists have been putting together thus contradicts the implicit model in much of media research. Rather than resisting or ignoring most new or dissonant media reports, as the autonomy model assumes, the information-processing view predicts that people are susceptible to significant media effects.
In the information-processing perspective, a person first assesses a media report for salience. If salient, the person processes the news according to routines established in the schema system. Processing may lead the person either to store the information or discard it; if stored, the information may stimulate new beliefs or change old beliefs. So selectivity and inattention are not the whole story. Often people may screen out information that contradicts their current views; but other times they think about disturbing reports they find relevant.
The notion of an audience that actively resists all potentially conflicting information rests upon an assumption of a deeply involved and knowledgeable citizenry, a vision that does not apply to most people (e. g. , Converse and Markus, 1979; Kinder and Sears, 1985). Common sense suggests it takes more information and time to change the minds of strong adherents than weak ones, but sometimes even loyalists do change. When the implications are not obvious—for example when the information is contained in the form of a subtle slant to the news (see Entman, 1989, chap. )—the probability increases that even activists will store conflicting data without experiencing any immediate dissonance. And while it may take many repetitions of a media message to pierce the public’s indubitable haze of neglect and distraction, this very same political indifference may enhance the likelihood that messages which do penetrate How the Media Affect What People Think 351 will have an impact. Just because on most matters Americans have so little knowledge and such weakly-anchored beliefs, information provided by the media can significantly shape their attitudes.
Not only do the majority of audience members lack detailed, expert knowledge or strong opinions (cf. Fiske, Kinder, and Larter, 1983); sometimes there are no old attitudes to defend. Many of the most significant political contests are played out over emerging issues or leaders; audiences do not have set attitudes toward them. That clears the path for significant media influence. TESTING MEDIA INFLUENCE Identification as liberal, moderate, or conservative is a key component of the political schema system that much of the public applies to political information. Ideological leanings affect responses to specific media eports; different identifiers may read the same message differently. This is why the media, in common with all other sources of information, cannot dictate public views and why an interdependence model seems appropriate. The interdependence model predicts that media influence varies according to the way each person processes specific news messages. Instead of treating ideology as a tool people use to screen out reports that conflict with their liberalism or conservatism, the model sees ideology as a schema that influences the use people make of media messages in more complicated ways.
The interaction between the attributes of the message and the schemas of the audience shapes the impact of the news. One element of this interdependence is message salience, which may vary among the ideological groups. Stories that interest liberals may bore conservatives; items that intrigue ideologues on either side may not interest moderates, who have few strong beliefs. Another aspect of interdependence involves whether the message is relevant to peripheral or central attitudes.
The centrality of a message may vary for different groups, since liberals and conservatives appear to structure their ideas distinctively. Central to liberalism is attachment to ideals of change and equality; central to conservatism is attraction to capitalism (Conover andj^eldman, 1981). The two groups probably process some media messages^differently. This decidedly does not mean liberals, for example, screen out all material that challenges liberalism. Consider an editorial praising the ideal of capitalist markets and proposing to make the post office a private enterprise.
While the message conflicts with liberal ideology, it does so peripherally, since government ownership of public utilities is not fundamental to American liberalism. The message may not only bolster conservatism among conservatives, but weaken liberals’ commitment to liberalism, if only at the margin. Another point of interdependence involves whether the message comes from an editorial, with its overtly persuasive intent, or from a news story that is ostensibly designed merely to inform. Conservatives may be more likely 352 Robert M. Entman o screen out editorial than news items that favor the left, since the slant of news may not be obvious. Afinalaspect of interdependence lies in how new or unfamiliar the reported topic is. All else being equal, the less familiar the object of the news, the less likely a person will respond by fitting the report into an established category and maintaining a set attitude. Where the subject of the news is unfamiliar to all sets of ideological identifiers, all will be susceptible to media influence. Four hypotheses emerge from this use of information processing theory to develop an interdependence model of media influence.
They are not all the hypotheses that merit exploration, but they are the ones that can be tested with the data available, and they should provide support for the superiority of the interdependence over the autonomy model. Hypothesis #1: Editorials affect ideological identifiers more than moderates. Those identifying as liberals or conservatives are likely to find ideologically-charged editorial messages salient. Those with less-focused commitments, the moderates, may not find ideological editorials relevant.
Hypothesis #2: Liberal editorials should exert a leftward push on those attitudes of conservatives not central to their ideology. Hypothesis #3: Editorial content has stronger effects on new subjects of news coverage than on long-familiar ones. Hypothesis #4: News affects beliefs among liberals, moderates, and conservatives alike. People will tend to screen out news messages less than editorials. Shaped by objectivity rules, news stories are designed to appear neutral to audiences (e. g. , Schudson, 1978; Tuchman, 1978; Molotch and Boden, 1985). The appearance of neutrality may soften the audience’s defenses.
DATA The dataset combines a national survey on Americans’ political attitudes from 1974 and 1976 with information on the political content of the newspapers read by respondents. The 1974 Michigan Content Analysis Study provides extensive information on the front page news and editorial page content of ninety-two newspapers throughout the country. The total number of news and editorial items employed here is nearly 18,000. 5 The content information (Institute for Social Research, 1978) is matched to data from a representative national survey, the University of Michigan Center for Political Studies poll of 1974.
The sample analyzed consists of those who were surveyed and read ° The study included ninety-six newspapers, of which four had incomplete data; readers of those four were excluded from the analysis. How the Media Affect What People Think 353 one of the ninety-two newspapers included in the Content Analysis Study, a total weighted sample of 1,292 persons. 6 Excluded were those who did not read a paper (approximately 30% of those surveyed) or who read papers for which no data were collected. 7 The content data were gathered for ten days during October and November, 1974.
Even though the data were obtained over a short time period, a check suggests they accurately reflect the typical stands of the papers. For example, among the ninety-two newspapers, the Washington Post scores higher in editorial liberalism than the (defunct) Washington Star; the New York Daily News scores to the right of the New York Times, and so forth. 8 In any case, while far from perfect, the dataset is the most comprehensive collection linking media content to peoples’ attitudes. One measure of newspaper content taps diversity in news stories, the other liberalism in editorials. I expect both aspects of the newspaper’s message to encourage opinions to move toward more sympathy with liberal politicians, 6 The actual number of people interviewed was 1,575. The answers of some members of the sample were counted three times to make a weighted sample of 2,523. This was done in order to ensure adequate representation in the sample of sparsely populated areas of the country. Thus, the weighted sample is the most representative. 7 The demographics of the final reader subsample closely parallel those of the 1974 national cross section as a whole.
The mean education of the entire original sample, including non-readers (n = 2,523), is 11. 5 years, the mean of the sample analyzed (n = 1,292) is 12. 2; the mean income, about $11,000 versus $12,000. On other demographic and political characteristics, the two groups are virtually identical. 8 Further enhancing confidence in the validity of the content measures is their use in such important studies as Erbring, Goldenberg, and Miller, 1980. 9 Each editorial item was coded for zero, one, or two assertions favoring or opposing liberal and conservative policy stands.
The editorial liberalism index is a percentage formed by first counting the number of times a paper endorsed a liberal position or opposed a conservative position, then subtracting assertions favoring conservative or derogating liberal stands. The result was divided by twice the number of editorial items, since each item was coded for up to two liberal or conservative assertions. The higher the score, the more liberal the editorial page. This index uses variables 21 and 28 in the CPS Media Content Analysis Study 1974. A second measure employed data on news (variables 27 and 34 in the CPS study).
The news diversity measure taps a dimension of news slant that audiences are less likely to screen than editorial liberalism. Like most aspects of news slant, it is a subtle trait of reporting that few audience members would notice. The front page news items were coded for mention of zero, one, or two problems. For each problem mention, coders noted whether two different actors overtly disagreed with each other. Each news item was coded as having zero, one, or two instances of two actors asserting different points of view.
The diversity index is the number of times two actors expressed different positions divided by twice the number of stories. The higher the score, the more diversity of news. Examples of the actors coded in this variable include Gerald Ford, Richard Nixon, Nelson Rockefeller, Democratic Party, Republican candidates, and business leaders. Thus, a story might concern inflation and unions, and might contain opposing assertions by Gerald Ford and a Democratic Senate candidate on both the causes of inflation and the value of unions.
The story would be coded 2 for one disagreement on each of the two problems. If the two actors agreed (or voiced no opinions) on unions but disagreed on inflation, the code would be 1. If they agreed on both or neither agreed nor disagreed, the code would be 0. 354 Robert M. Entman groups, and ideas. The basis for predicting that news diversity moves audiences leftward is that the majority of local newspapers appear to promote a generally Republican and conservative perspective (cf. Bagdikian, 1974; Radolf, 1984).
Their editorial and perhaps news inclinations do not favor liberalism. All else being equal, I believe those papers with higher diversity probably provide more information that challenges the conservative editorial baseline. In addition, the mere presence of conflicting views in the news may convey an awareness of the diversity of the country, including its variety of races, economic classes, and viewpoints. Such consciousness may promote tolerance of change, and empathy for positions or groups that challenge the status quo. 0 Diversity may also undermine authority by conveying the impression that a range of ideas is plausible, that the existing distribution of power, wealth, and status is not immutable. As for the other content measure, while many readers no doubt skip editorial pages, Bagdikian (1974) shows that the editorial perspective tends to be mirrored in news slant. The editorial liberalism index may indirectly reflect the political tendency of news coverage. The survey included “feeling thermometer” questions. Interviewers asked respondents to express their feelings toward several well-known groups and politicians.
Respondents chose numbers ranging from “0” for the coldest feelings, through “100” for the warmest, with “50” meaning neutral or mixed feelings. I constructed five attitude indexes using factor analysis. 11 The Liberal Feelings Jndex combined ratings of Edward Kennedy, Hubert Humphrey, liberals, Democrats, and unions. The Radical Feelings Index consisted of thermometer ratings of radical students, black militants, civil rights leaders, and policemen. The Poor Feelings Index tapped thermometers of poor people, blacks, and George Wallace.
The Republican Feelings Index was created from ratings of Gerald Ford, Richard Nixon, and Republicans. Finally, the Conservative Feelings Index rated big business, the military, and conservatives. 12 The Michigan survey also asked respondents for their stands on government guaranteed jobs; dealing with urban unrest by solving the problems of unemployment and poverty; protecting legal rights of those accused of crimes; A competing hypothesis might be that diversity challenges initial viewpoints, so that it would promote conservatism among liberals and vice versa.
That idea is not borne out by the data. Diversity is consistently associated with more liberal views. 1 ‘ Surveys are described in Institute for Social Research, 1979. All feeling thermometers were classified on their face for relevance to the liberal-conservative continuum. Pertinent items received varimax factor analysis. Five factors had eigenvalues greater than 1. 0. Indexes added together scores on all feeling thermometer responses loading above . 40 on a factor. In two cases, items loaded more than . 40 on two factors; these were included on their highest loaded index.
All dependent variable attitude indexes used in this paper have Cronbach Alpha reliability scores greater than . 80. 12 Policemen and Wallace loaded negatively on their respective factors. The feeling thermometer responses to each were subtracted from the sum of the other items in forming the indexes. 10 How the Media Affect What People Think 355 busing to achieve racial balance; the Equal Rights Amendment; integration of schools; government aid to minorities; and self-placement on the liberalconservative spectrum. 3 Using factor analysis again, all but one of the responses (to the ERA) were associated together and became the Policy Preferences Index. Twofinalvariables come from readers of sampled papers who participated in surveys during both 1974 and 1976. Their responses in 1976 provide an opportunity to check for media impacts on feelings toward a previously unknown presidential candidate, Jimmy Carter (Carter Index), and on presidential vote (Vote76). FINDINGS Testing the four predicted media effects requires probing for impacts of editorial liberalism and news diversity on the seven attitudes and on presidential vote.
Regression analysis enables us to see whether, with all else equal, readers of more liberal or diverse papers exhibit more liberal attitudes and voting behavior. Editorial liberalism taps the persuasive element of the newspaper, or, in agenda-setting terms, the aspect of the paper that attempts to “tell people what to think. ” News diversity taps the putatively informational element that only “tells people what to think about. ” The interdependence model holds that both editorials and news provide information to think about and thereby influence attitudes, whether intentionally or not.
If selectivity or inattention precludes media influence, or if the effect is limited to agendas, the regressions should reveal no significant associations between attitudes and newspaper content. 14 Table 1 summarizes regression results for the impacts of newspaper content on the beliefs of the entire sample of readers. The feeling thermometers are coded from 0 to 100 so that higher scores are warmer (more favorable). The higher the policy preferences score, the more conservative the responses. Vote76 is 1 for Carter, 0 for Ford, so higher scores indicate voting for Carter.
The regressions include the following additional variables to control for forces that might also influence attitudes: urban-rural place of residence; age; years of education; family income; race; region; party identification; and ideological self-identification. 15 The impacts of these non-media variables follow expecVariables 2265, 2273, 2281, 2288, 2296, 2302, and 2305 in the 1974 NES Codebook. Although partisanship and ideology are not truly interval variables, the results of the regressions suggest that it is quite reasonable to treat them as such. 15 These variables are coded as follows.
Age: coded in years; non-South: 1 = North or West, 0 = South; income: coded in thousands; party i. d. : 7-point scale, 0 = strong Democrat, 3 = independent, 6 = strong Republican; urbanized: 1 = urban, suburban, 0 = rural; white race: 1 = white, 0 = nonwhite; education: coded in years; policy preferences index: adding six 7-point scales, so range is 6 = most liberal, 42 = most conservative; and ideology identification: 1 = most liberal, 4 = middle of the road or don’t know, 7 = most conservative. On the latter, note 14 13 356 Robert M. Entman tations, which bolsters confidence in the validity of the attitude measures. For a full display of coefficients for all independent variables, see Entman, 1987). Multicollinearity among the independent variables is not a problem. Of the forty-five intercorrelations, only three exceed . 20. The strongest was between education and income (r = . 357). Table 1 shows that the more editorially liberal the paper, the more warmly their readers respond on the Liberal Feelings Index. This relationship suggests that editorial liberalism influences the public’s evaluations of key leaders and groups associated with the liberal coalition: in this case, Hubert Humphrey, Edward Kennedy, Democrats, unions, and liberals.