The Role of Media Distrust in Partisan Voting
- First Online:
- Cite this article as:
- Ladd, J.M. Polit Behav (2010) 32: 567. doi:10.1007/s11109-010-9123-z
- 772 Downloads
As an institution, the American news media have become highly unpopular in recent decades. Yet, we do not thoroughly understand the consequences of this unpopularity for mass political behavior. While several existing studies find that media trust moderates media effects, they do not examine the consequences of this for voting. This paper explores those consequences by analyzing voting behavior in the 2004 presidential election. It finds, consistent with most theories of persuasion and with studies of media effects in other contexts, that media distrust leads voters to discount campaign news and increasingly rely on their partisan predispositions as cues. This suggests that increasing aggregate levels of media distrust are an important source of greater partisan voting.
KeywordsNews media Trust Voting Party identification Media skepticism
The news media play a central role in modern democracies. Citizens rely on the media for information about politicians’ proposals and actions. As Lippmann (1997 , p. 53) puts it,
Each of us lives and works on a small part of the earth’s surface, moves in a small circle, and of these acquaintances knows only a few intimately. Of any public event that has wide effects we see at best only a phase and an aspect…Inevitably our opinions cover a bigger space, a longer reach of time, a greater number of things, than we can directly observe. They have, therefore, to be pieced together out of what others have reported and what we can imagine.
While earlier research was more skeptical, political science scholarship increasingly agrees that the news media play a central role on democratic governance by, among other things, shaping public opinion and electoral preferences (Cook 1998; Kinder 2003).
Yet at the same time, journalists and academics have become concerned about the intense hostility the public expresses toward the news media. In the 1970s and 1980s, commentators often noted the news media’s popularity, which frequently surpassed that of other societal institutions (Gronke and Cook 2007). However, by the 1990s and into the twenty-first century, many noticed that the press had become much less popular. For instance, Fallows (1996, p. 1) opens his book assessing the state of political journalism by saying:
Americans have never been truly fond of their press. Through the last decade, however, their disdain for the media establishment has reached new levels. Americans believe that the news media have become too arrogant, cynical, scandal-minded, and destructive. Public hostility shows up in opinion polls, through comments on talk shows, in waning support for news organizations in their showdowns with government officials, and in many other ways.
Sanford (1999, p. 11), a first amendment lawyer, expresses similar concerns, stating:
A canyon of disbelief and distrust has developed between the public and the news media. Deep, complex and so contradictory as to be airless at times, this gorge has widened at an accelerating rate during the last decade. Its darkness frightens the media. It threatens not just the communication industry’s enviable financial power but its special role in ordering American democracy. It is a canyon of terrifying proportions.
This change was widespread. While Republicans have modestly lower press confidence than Democrats in almost every GSS survey, that gap is dwarfed by the secular decline found in each party as well as among independents (Gronke and Cook 2002). Consequently, by the 2000s, confidence in the press was lower than confidence in most other institutions.2
These dramatic changes in opinions toward the news media have led some to wonder about the consequences. Do these opinions affect the way people use the media to acquire information and form electoral preferences? Yet while existing studies examine how media trust moderates several types of media effects, none look explicitly at the consequences for voter decision-making.
That is the focus of this paper. The next three sections review relevant scholarship on partisan voting, news media persuasion, and the role of source credibility. The subsequent section lays out my expectations for how attitudes toward the media influence voter decision-making, specifically focusing on the role of partisan predispositions. This is followed by sections outlining the empirical analysis, interpreting the results, and briefly concluding. The paper ends with a brief concluding section.
Party Identification and Vote Choice
Few political phenomena have been studied more extensively than vote choice. While a definitive model of the voting process is still elusive, one robust empirical regularity is that voters have strong psychological orientations toward the major parties. These orientations serve as powerful baselines for voting decisions (Bartels 2000; Campbell et al. 1980 ; Green et al. 2002; Lewis-Beck et al. 2008; Miller and Shanks 1996). One consequence is that, in circumstances where a voter faces a new or unfamiliar decision, she can use her party identification (or “standing decision”) as a shortcut (Conover and Feldman 1989; Key and Munger 1970, p. 253; Rahn 1993). A second consequence is that, while new information can change a voter’s party identification, the effect is small unless the message is very dramatic or long lasting. The short-term exogeneity of party identification allows researchers to be reasonably confident in their estimates of its effects (Campbell et al. 1980 , pp. 531–535; Cowden and McDermott 2000; Gerber and Green 1998; Green et al. 2002; Jennings and Niemi 1981; Johnston 2006; Miller 1999).
In contrast, attempts to estimate the effects of election-specific considerations on voting have been more difficult (Ansolabehere 2006; Bartels 1992; Holbrook 1994). The main obstacle is endogeneity. Reported perceptions of economic performance and candidates’ personal characteristics or issue positions are as likely to be rationalizations as causes of vote choice (Achen and Bartels 2006; Bartels 2002; Brody and Page 1972; Kramer 1983; Lenz 2009; Page and Brody 1972; Rahn et al. 1994). Even studies showing an increase in the correlation between specific survey responses and vote choice over the course of a campaign are vulnerable to the worry that the campaign merely increases the amount of rationalization (Lenz 2009).3 Yet despite these challenges, several studies with creative research designs have documented outside the laboratory powerful media effects on voting.
Media Influence on Public Opinion and Voting
While some previous scholars doubted their power (e.g. Klapper 1960), in recent decades a consensus has developed that media influence over public opinion is more “massive” than “minimal” (Kinder 1998, 2003; Zaller 1996). This literature documents several different types of media effects on attitudes. These include priming (e.g. Iyengar and Kinder 1987; Krosnick and Kinder 1990), framing (e.g. Nelson et al. 1997), providing politically relevant information about national conditions (e.g. Gilens 1999; Hetherington 1996), and direct persuasion (e.g. Bartels 1993; DellaVigna and Kaplan 2007; Gerber et al. forthcoming; Kahn and Kenney 2002; Ladd and Lenz 2009).
Of particular interest here are the latter two and their relationship to voting. For instance, Hetherington (1996) shows that those who follow the news during a presidential campaign have different perceptions of national economic performance and consequently different voting preferences. Bartels (1993) finds that those who consume more news are more likely to change their views of the candidates during a presidential campaign. Kahn and Kenney (2002) find that newspaper editorial endorsements are associated with evaluations of Senate candidates. Using quasi-experimental designs, DellaVigna and Kaplan (2007) and Ladd and Lenz (2009) find evidence that the Fox News Channel in the United States and several newspapers in Great Britain, respectively, influence voting decisions. In a field experiment, Gerber et al. (forthcoming) find evidence that newspaper exposure influences gubernatorial votes. In summary, a few recent studies outside the laboratory provide some evidence that media messages affect voting preferences through providing political information and direct persuasion, although creative research designs or unusual historical circumstances are necessary to attain that evidence.
The Role of Source Credibility in Attitude-Change
While one stream of research documents the effects of the news media on the public, the more general literature on attitude-change emphasizes the central role of source credibility (Druckman and Lupia 2000). This is consistent across several major research traditions. For instance, early work by Hovland and his colleagues on the persuasive effects of communication argued that attitude-change depends on perceptions of the sender’s expertise, trustworthiness and similarity to the recipient (Hovland et al. 1953; Hovland and Weiss 1951–1952). Also, in the receive–accept-sample model conceived by McGuire (1969) and adapted to political science by Zaller (1992), source credibility is relevant at the acceptance stage. Perceiving the message as coming from a source with a different predisposition induces “partisan resistance” (Zaller 1992, p. 121).
In psychology, “dual-process” theories such as the elaboration-likelihood model (Petty and Cacioppo 1986) and heuristic–systematic model (Chaiken 1980; Eagly and Chaiken 1993) see source credibility as a heuristic individuals use to decide whether to accept an argument when they lack the desire or ability to analyze the message’s content. Zaller (1992) argues persuasively that, in modern American politics, the vast majority of the population processes political information by the heuristic route, where source credibility is central. Most people are neither involved nor interested in politics by the standards of Chaiken and Petty and Cacioppo’s experiments (Converse 1964; Delli Carpini and Keeter 1996; Kinder 1998), leading them to depend on source cues.4
Game theoretic models of strategic communication (called in some incarnations “cheap talk” or “signaling” models), while varying in their specifications, almost all predict that source credibility will be a key factor in determining if people are influenced by informative messages (e.g. Crawford and Sobel 1982; Gilligan and Krehbiel 1987, 1989; Lupia and McCubbins 1998). The key source criteria in these models are whether the source is knowledgeable and has the same interests as the message recipient. Studies where people use cues from elites as information shortcuts when forming their opinions are of a similar intellectual lineage and posit similar source credibility criteria (e.g. Lupia 1994; Popkin 1991; Sniderman et al. 1991). In summary, while scholars in communication, political science, psychology and economics have active research interests in attitude-change and have developed several distinct models of the process, almost all agree that influence depends on the recipient’s perception of the messenger.
Several studies have specifically tested the role of source credibility in media effects, finding that it serves as an important moderator. Miller and Krosnick (2000) find, in a laboratory experiment, that newspaper priming does not occur among those who distrust the media. In another laboratory experiment, Druckman (2001) finds that newspapers that subjects distrust do not produce framing effects. In several observational studies, Tsfati (2002) finds that those who distrust the media are more resistant to agenda-setting. Elsewhere, Tsfati (2003) examines people’s beliefs about others’ opinions, finding that those who trust the media are more likely to accept media messages about the national division of public opinion. Ladd (2004) finds, looking at cross-sectional and panel data, that those who distrust the media are less responsive to objective statistics when forming their beliefs about national conditions in a variety of policy areas, instead basing their beliefs on partisanship. Finally, in a laboratory experiment, Ladd (2004) finds that those who distrust the media do not update their beliefs about foreign conflicts in response to news reports. Taken as a whole, there are both strong theoretical reasons to believe that attitudes toward the news media will moderate media effects and a growing number of empirical studies indicating that they do. This paper extends this work by looking at the consequences of media trust’s moderating role in media effects for voting behavior.
Some previous work on media source credibility has focused on the credibility of specific outlets (Druckman 2001), while some has focused on views toward the media in general (Ladd 2004; Miller and Krosnick 2000; Tsfati 2002, 2003). Attitudes toward the news media as an institution are of special relevance because they have changed so dramatically over the past 40 years. The attitude-change literature does argue that views toward entire institutions can be consequential. For instance, Petty and Cacioppo (1981, p. 61) state that, in the literature on source credibility,
[t]he originator or source of a persuasive communication may be a person (e.g., the president of the United States), a group (e.g., your family), an institution (e.g., Stanford University), and so forth. (italics in original)
Political communication scholars, such as Cook (1998) and Sparrow (1999), argue that the news media function as their own political institution. Still, one could worry that people may not have clearly held attitudes toward the media as an institution, but only toward specific outlets. A second (related) worry is that, even if respondents have clear opinions about the news media as a societal institution, the various wordings used by different survey organizations may tap different considerations and consequently produce notably different responses.
Fortunately, studies validating these questions tend to allay both concerns. Refusal and “don’t know” rates for questions about the media as an institution tend to be unusually low: consistently less than one percent and often nearly zero across different surveys (Tsfati 2002, p. 67). Also, responses to these questions tend to be very consistent across wordings (Kohring and Matthes 2007; Ladd 2006a, pp. 26–28), including prominent wordings like the American National Election Study’s (ANES) media “trust” and “thermometer” questions and the GSS’s press “confidence” question. These attitudes are also stable over time, even when question wordings change in different waves of panel surveys (Ladd 2006a, pp. 28–29; Tsfati 2002, pp. 62–66).
In addition, open-ended questions about the media as an institution tend both to prompt similar responses as closed-ended questions and to also produce consistent responses across wordings (Ladd 2006c; Tsfati 2002, pp. 46–49). Finally, tests of discriminate validity find that attitudes toward the institutional news media are distinct from general mistrust, political ideology, and ideological extremism (Tsfati 2002, pp. 50–55). All this suggests that most people have a clear attitude toward the news media as an institution (not a “nonattitude”; Converse 1964), and that most prominent question wordings tap into that attitude. Tsfati (2002, p. 38) makes this point by concluding that “people have some mental schema for what ‘the media’ are,” and that “[m]edia skepticism is targeted toward the mainstream media in general.”
Given that most people have relatively firmly held attitudes toward the media, and the increasingly well-documented role of source credibility in media effects, it is natural to suspect that these attitudes may have important consequences for voters’ decision-making. As noted above, in the absence of additional considerations, voters tend to fall back on their party identification as a baseline for making choices. With more resistance to information about national conditions and other influential campaign news media messages, we might expect those with negative attitudes toward the media to rely more on this baseline. Stated simply, I expect voters who distrust the media to rely more on their party identification to make voting decisions.5 The remainder of this paper tests this prediction and interprets the results.
Data and Methods
Antipathy toward the news media induces partisan voting
Logit model with all variables measured 2004
Logit models with vote choice measured in 2004 and all other variables measured in 2002
Instrumental variables model with all explanatory variables measured in 2004 and instrumented with their values in 2002
News media thermometer rating
Strength of partisanship
Frequency of following government and public affairs
Frequency of political discussion
Campaign television program viewing
Average trust (government and people)
Average efficacy (internal and external)
Preferences on government aid to the poor
Feminists thermometer rating
Blacks thermometer rating
Defense spending preferences
Network news exposure
Media thermometer × network news exposure
Media thermometer × newspaper exposure
Number of observations
To ensure that the observed relationship between media trust and partisan voting is not spurious, I attempt to hold constant other variables that may lead to more partisan voting. To this end, I control for strength of party identification and two other variables that the literature consistently finds to be associated with stronger partisanship: age (e.g. Campbell et al. 1980 , pp. 162–163; Converse 1969) and political knowledge (e.g. Zaller 1992).7 In addition, to guard against the possibility that differences in exposure to campaign information are driving the apparent effects of media evaluations, I control for respondents’ frequency of following government and public affairs, their frequency of political discussion, and whether they viewed campaign television programming. I also control specifically for network news exposure and newspaper exposure. However, these questions are excluded from some models because they were not asked in 2004.
I also include several variables to ensure that media distrust is not a proxy for general disillusionment with society or the political system. I control for respondents’ general level of trust, calculated by averaging the ANES’s trust in government and trust in people questions. To account for respondents’ efficacy, I control for the average of their responses to the ANES’s internal and external efficacy questions.
Finally, I attempt to ensure that any apparent effect of media evaluations is not driven by respondents’ ideology. One way to do this is to control for respondents’ self-placements on a seven-point ideology scale. However, as with newspaper and network news exposure, this variable cannot be included in every model because it was not asked in 2004. Another potential problem with self-reported ideology is that researchers often find it to be only loosely correlated with policy preferences (e.g. Erikson et al. 2002, pp. 205, 222–230; Stimson 2004, pp. 84–95). For this reason, I include several more specific ideological questions. To account for economic views, I control for respondents’ preferences on government aid to the poor. To account for social ideology, I include thermometer ratings of “feminists.” To account for racial ideology, I include thermometer ratings of “blacks.” Lastly, to account for defense-related ideology, I control for government defense spending preferences.
Column 1 of Table 1 presents parameter estimates from a logit model where all variables are measured in 2004. As expected, the coefficient on media evaluations is negative and statistically significant.8 Since the sizes of logit coefficients are not directly interpretable, I calculate the marginal effect of media evaluations when all other variables are at their means. In the model in column 1, moving from a media thermometer rating of zero to a rating of 100 decreases the probability of voting for the candidate of one’s own party by 0.13 (Std. Err. = 0.05, p = 0.01).
However, models with cross-sectional data face the worry that causation runs in the opposite direction. Of specific concern here is the possibility that those who have already decided to cast partisan votes before the campaign are more likely to react negatively to campaign coverage, leading to disproportionate declines in their media trust. The psychological literature on the causes of media distrust lends credence to this possibility. A prominent finding in this literature is the “hostile media phenomenon” (or “hostile media effect”): the tendency of people with divergent prior opinions, when consuming the exact same news report, all to view the report as biased against their views (Christen et al. 2002; Giner-Sorolla and Chaiken 1993; Vallone et al. 1985). Given that those most predisposed toward partisan voting before the election are likely to have more extreme political preferences, election coverage may reduce their media trust through the hostile media phenomenon, producing a spurious association. While not eliminating all concerns about causal direction, measuring media evaluations several years before the election can reduce worries that reactions to campaign coverage drive this association.
The model in column 2 measures all explanatory variables with the same questions as in column 1, but using responses from the 2002 wave of the panel.9 Measuring variables in 2002 allows me to also include ideological self-placement, network news exposure, and newspaper exposure, which were not asked in 2004. Column 3 presents a model that includes these as additional controls. Columns 2 and 3 indicate that measuring explanatory variables 2 years prior and including additional controls has little effect on the main finding. Column 2’s results imply that moving from a media thermometer rating of zero to a rating of 100 (with other variables set to their means) decreases the probability of voting for the candidate of one’s own party by 0.20 (Std. Err. = 0.08, p = 0.01). Column 3’s results also indicate that the same change decreases the probability of a partisan vote by 0.20 (Std. Err. = 0.08, p = 0.01), meaning that (after rounding) the inclusion of ideological self-placement and network news and newspaper exposure does not alter the estimated effect of media evaluations at all.
As another alternative specification, the model in column 5 of Table 1 measures explanatory variables in 2004 while instrumenting them with their values in 2002.10 This model produces a statistically significant, negative, yet less precise estimate of the effect. The results imply that moving from a 0 to a 100 media rating decreases the probability of a partisan vote by approximately 0.69, yet the 95% confidence interval on this effect ranges from −0.03 to an impossible −1.35. Thus, while the estimated effect is large and negative, we should be cautious in interpreting it as larger than the estimates from the column 1–3 equations.11
The results from columns 2, 3, and 5 provide some reassurance that the negative relationship between media trust and partisan voting is not a product of endogeneity driven by the hostile media phenomenon. Admittedly, it is still possible that, prior to 2002, the hostile media phenomenon caused those who would cast more partisan votes in 2004 to disproportionately distrust the media. Yet, at least in the 2 years prior to the election, this type of endogeneity is not driving this relationship.
Postestimation diagnostics on these models are generally encouraging. Despite the inclusion of numerous control variables, multicolinearity does not appear to be severe problem. For instance, the variance inflation factors associated with media thermometer ratings in the models in columns 1, 2, and 3 are only 1.25, 1.18, and 1.19, respectively. This suggests that the control variables only minimally increase the sampling variances of the media thermometer coefficients. In addition, while the pseudo R2 statistics for these models are low, as they tend to be for all models using political survey questions, the inclusion of media ratings modestly increases these statistics. It increases the pseudo R2 from 0.230 to 0.243 in column 1, from 0.132 to 0.146 in column 2, and from 0.133 to 0.147 in column 3.
Additional Robustness Checks
Another alternative explanation for these results is that negative media attitudes induce partisan voting, not by causing resistance to the messages people receive, but by changing those individuals’ patterns of media exposure. For example, those who distrust the media as an institution may utilize mainstream news sources (like network television and newspapers) less often, instead either consuming less overall news or relying more on partisan media sources. However, the available evidence suggests that this is not producing the entire effect observed here. Column 4 in Table 1 shows a model that is similar to column 3, except that it includes the interactions of media thermometer ratings with network news exposure and with newspaper exposure. If the effect of media ratings on partisan voting occurs entirely through changes in mainstream media exposure, these two interaction terms should have negative coefficients and the “main effect” of media ratings should be close to zero. This is not what I find. The coefficients on these interaction terms are small and statistically insignificant, while the “main effect” of media ratings is negative and almost the same size as in columns 1–3.12
However, we should draw this conclusion cautiously because adding two interaction terms substantially increases multicolinearity. In column 4, the variance inflation factor for media thermometer ratings is 4.49, indicating that the control variables substantially increase its coefficient’s sampling variance (i.e. standard error). This larger standard error drops the coefficient below standard levels of statistical significance.
To ensure that Table 1’s results are not unique to this dataset, I looked for the same patterns in other datasets containing similar questions. As noted above, every GSS survey since 1973 has probed confidence in the press. Those conducted after a presidential election also probed respondents’ vote choices. These cross-sectional surveys reveal the same pattern of statistically significantly greater partisan presidential voting among those with less press confidence. The ANES began asking about media trust in its regular cross-sectional surveys in 1996. My analysis of the 1996 and 2000 surveys also indicates that partisan presidential votes are significantly more likely among those who distrust the media. Several respondents in the 1996 ANES survey were also interviewed for the 1993 ANES pilot study, where they were asked whether “[m]edia coverage of politics often reflects the media’s own biases more than facts.” This allows for a partial replication of the panel analysis presented in Table 1, but with a sample size of less than 350. The results show that, when media evaluations are measured in 1993, even with the small sample size, those with negative attitudes toward the media are significantly more likely to cast a partisan presidential vote in 1996. To save space and avoid redundancy, I do not report these results here, but they are all available in Ladd (2006b).
It is also possible to test my expectations with an alternative statistical specification. One could estimate a model where vote choice (Republican vs. Democratic vote) is a function of party identification, trust in the media, the interaction between these two variables, and control variables. Results from such a model (using either the 2002–2004 ANES data or any of the additional datasets described in the previous paragraph) are consistent with the results presented here. In each dataset, the interaction term’s coefficient is negative, indicating party identification is more influential when respondents distrust the media. These results are also presented in Ladd (2006b).
Finally, given that Republicans have consistently more negative attitudes toward the media, I checked whether the effect differed between Democrats and Republicans. When the models in Table 1 are estimated separately among Democrats and Republicans, the effect of media ratings is consistently negative in both parties, but often not statistically significant at conventional levels in the separate partisan groups because of small sample sizes. The difference in the magnitude of the negative effect between the parties varies considerably across models and depending on when party identification is measured. For instance, if one estimates the logit model from column 3 of Table 1 separated by 2002 partisanship, the coefficient on media evaluations is −1.32 (standard error = 1.23, n = 251) among Democrats and −1.31 (standard error = 1.41, n = 284) among Republicans. However, if they are separated by partisanship in 2004, the coefficient is −1.09 (standard error = 1.14, n = 258) among Democrats and −2.27 (standard error = 1.22, n = 283) among Republicans, in each case an insignificant partisan difference. Similar variation in the difference between parties occurs when control variables are included or excluded. I conclude from this analysis that it is possible that the affect varies across parties to some degree, a difference that could be revealed in future studies using larger panel datasets. However, based on the evidence available here, I cannot be certain of any partisan difference.
In summary, consistent with expectations, those with negative attitudes toward the news media are more likely to rely on their partisanship to make presidential voting decisions. This effect is robust to measuring the explanatory variables several years prior to the campaign, at least partially easing concerns about endogeneity resulting from the hostile media phenomenon. The relationship is also consistent across different datasets and statistical specifications. Finally, the available evidence does not reveal any consistent difference in this effect across partisan groups.
While a growing body of research documents the role of news media evaluations in moderating the effect of media messages on the public’s beliefs and preferences (Druckman 2001; Ladd 2004; Miller and Krosnick 1996; Tsfati 2002, 2003), scholars have not yet examined the consequences of this phenomenon for voting. This paper investigates the effect of voters’ attitudes toward the news media as an institution on how they form electoral preferences. Given the previous literature, I predict that partisan voting will be greater among those who distrust the media. An analysis of 2004 presidential voting is consistent with this prediction, even when explanatory variables are measured several years prior to the election.
This article began by noting that, in modern democracies, voters are called upon to make important and (at times) complicated decisions without all possible information (Lippmann 1997 ). As Conover and Feldman (1989, p. 917) put it, “[W]hen faced with difficult conditions created by an ambiguous political world, voters use their existing store of knowledge to make inferences about candidates.” The results reported here indicate that the political world becomes considerably more “ambiguous” when a voter distrusts news media messages. Like those in other low information situations, voters who lack confidence in the media are forced to rely on “existing store[s] of knowledge,” especially their “standing decision” (Key and Munger 1970, p. 253) among the parties. As Rahn (1993) states, “In partisan elections, the most powerful cue provided by the political environment is the candidate’s membership in a political party.” When a voter distrusts the press, voting based on party identification becomes the most instrumental choice.
This suggests that a broader range of phenomena could fall under the label of media effects. Traditionally, media effects research has focused on the power of media messages to persuade the public. Now, the growing body of work on source credibility and the news media suggests that political behavior depends not just on the volume and content of media messages, but also on attitudes toward the press itself. Even holding messages constant, changes in the news media’s institutional reputation can produce important effects on beliefs, opinions and voting preferences. Given the dramatic changes in the American public’s views on the media in the last several decades, these indirect media effects are likely to be increasingly important.
Television is excluded from the average in order to provide a clear comparison between the press and other institutions that are unlikely to be considered part of the news media.
While not the focus of this paper, there is also a substantial literature examining the causes of these negative attitudes toward the news media (see for instance Christen et al. 2002; Fallows 1996; Giner-Sorolla and Chaiken 1993; Ladd 2010; Patterson 1993; Sabato 2000; Vallone et al. 1985).
Some studies have had more success demonstrating economic effects on vote choice by pooling survey data over many years and using objective measures of economic performance rather than survey responses (Markus 1992; Zaller 2004). Unfortunately, questions probing attitudes toward the news media have not been asked over a sufficiently long time period to incorporate them into this type of analysis.
In 2000, the ANES used a split-mode design, with approximately half of respondents sampled by random digit dialing and interviewed by phone, and the other half selected by probability area sampling and interviewed in person. Reinterviews in 2002 and 2004 were conducted entirely by telephone. The American Association for Public Opinion Research’s official “RR1” response rate for the 2000 ANES is 61% (1,807 interviews out of a sample of 2,984). Of those 1,807 respondents, 1,187 were interviewed again in 2002 (a reinterview rate of 66%), with 840 of those interviewed again in 2004 (a reinterview rate of 71%). As this paper focuses on responses from 2002 and 2004, the relevant response rate corresponds to those who completed the entire panel. The RR1 response rate for the entire panel is 28% (840 out of an initial sample of 2,984). Panel data like these have the disadvantage of possibly introducing biases resulting from panel conditioning or panel attrition. While not settling the matter, existing scholarship is reassuring on this point, finding panel effects in the ANES to be small (Bartels 1999).
Unlike many other ANES surveys, a short quiz of basic political facts is not included in the 2002 and 2004 waves of this panel survey. Instead, I measure knowledge with interviewer ratings of the respondents’ “general level of information about politics and public affairs.” Interviewer ratings are often used as substitutes when factual questions are not available (e.g. Bartels 1996) because they tend to be highly correlated (Zaller 1985).
The negative association between media evaluations and partisan voting is present in a simple bivariate analysis as well. For instance, party voting occurs among 83.6% of those who rate the media at 70 degrees or higher and among 88.5% of those who rate the media at 30 degrees or lower, a difference of approximately 5 percentage points. This difference without controls has a p-value of 0.086. The inclusion of control variables somewhat increases the magnitude of the relationship, reducing the p-value.
Measuring all explanatory variables several years prior to the election has the disadvantage of introducing more measurement error (i.e. random variation) into these variables, potentially reducing the model fit and increasing the standard errors. This is evident in the fact that these models fit the data more poorly, with pseudo R2s of 0.15 rather than 0.24. This becomes a more serious problem if one tries to measure explanatory variables in 2000, four years prior to the election. This, plus the smaller sample size (less than 450), makes it impractical to use this approach. When such a model is estimated, the pseudo R2 drops to 0.13, and the effect of media thermometer ratings is still negative but statistically insignificant.
As the model in column 5 uses instrumental variables regression, it should be interpreted as a linear probability model (Aldrich and Nelson 1984), whose coefficients are not directly comparable in size to logit coefficients.
For the models in columns 1, 2 and 3, the 95% confidence intervals on the marginal effects are 0.03–0.23, 0.05–0.36, and 0.05–0.35, respectively.
I also checked for interactions between media evaluations and several other variables, including political knowledge and strength of partisanship, but found no evidence of heterogeneity in the effect.
I thank Doug Arnold, Larry Bartels, Martin Gilens, Erika King, Gabriel Lenz, Skip Lupia, Tali Mendelberg and seminar participants at the University of Delaware, George Washington University, Georgetown University, Princeton University and Temple University for helpful comments on earlier versions of this paper. All remaining errors are my own.