White Media Attitudes in the Trump Era

Scholars and political commentators point to Trump’s war on the media since the 2016 election as an unprecedented attack on a vital check to Presidential power. However, little attention has been paid to the role that White audiences play in this critical debate. In this article, I examine the relationship between Trump, the media, and White audiences. Using data taken from the American Trends Panel, I show that affect for Trump is conditional on Whites’ selective partisan exposure to conservative news media. My analysis also shows that exposure to political and election news directly from Trump intensifies the relationship between Whites’ perceptions of media bias and their distrust of national news organizations. The findings provide a novel and unique contribution to the existing scholarship by demonstrating the causal effect of selective exposure to conservative media outlets on affect for Trump.


Introduction
Since his election in 2016, Trump and his network have declared war on the media (Grynbaum, 2017a;Kellam & Stine, 2017)-an institution regarded by many as the "Fourth Estate" of the American political system (Schultz, 1998). Popular commentary and scholarly research debating the relationship between Trump and the media have tended to focus on the ways in which Trump's attacks on the news media represent a significant threat to a vital check on Presidential power (Kalb, 2018).
Largely ignored in these critical debates about Trump and the media is the role of an additional contextual variable-Whiteness-and how Trump's relationship with the media is shaped by the consumption behavior and attitudes of White audiences. This gap is surprising, given that there is an intrinsic link between Whiteness and estimations of Trump. Whiteness plays an increasingly important role in shaping Whites' political behavior as it relates to their evaluations of political figures and their voting preferences in the Trump era (Jardina, 2019(Jardina, , 2020). An important theoretical justification for studying White Americans' attitudes towards the media in the Trump era is that, demographically, the audiences of outlets that cover Trump in a favorable light are overwhelmingly White,too. 1 In the era of fragmented media, outlets such as Fox provide a critical platform for Trump to appeal to White audiences. Indeed, reflecting on Clinton's poor showing among White voters in the 2016 Presidential election, President Obama pointed to the ubiquity of Fox News as one reason for Trump's surprising victory in 2016, noting that White voters "turned out in huge numbers for Trump [,] [a]nd I think that part of it is. . . Fox News in every bar and restaurant in big chunks of the country." (Wemple, 2016). At the same time, the data show that, even in the realm of digital media, Trump appeals to a largely White audience. For instance, a 2016 study found that 84 per cent of Trump's followers on Twitter were White (Wang et al., 2016). Understanding how White consumption behavior feeds into affect for Trump, and the ways in which Trump shapes Whites' attitudes towards the Fourth Estate is thus of critical importance.
There is reason to expect that there may be heterogenous effects of media consumption on affective evaluations of political figures by race. An important difference between Whites and non-Whites are their racial attitudes. Among Whites, racial attitudes are especially driven by prejudice or hostility towards non-White groups. Even among Whites there are significant variations in levels of racial resentment. Literature on racial priming demonstrates that racist attitudes or resentment may be the driving factor behind exposure to conservative media and affect towards Trump (Schaffner et al., 2018;Valentino, 1999). Consistent with these streams of literature, it is reasonable to expect, therefore, that the effect of selective exposure to conservative media on affect for Trump become more significant for Whites than they do for other racial groups.
In this article, I investigate the relationship between President Trump and White consumption behavior and media attitudes. I argue that the traditional framework by which political actors try to garner positive media coverage to gain visibility and appeal to their core voter constituencies only partly explains Trump's relationship with the media. Media audiences in America today are highly polarized. Many Americans form "echo chambers" in which they only consume media from outlets whose coverage reinforces their existing political attitudes (Jamieson & Cappella, 2008). When Trump gives televised interviews, he tends to insert himself into Republican-leaning and conservative echo chambers such as Fox. Given the nature of audience polarization, I argue that the purpose of such a strategy is to reinforce the estimations of those who already hold positive views towards Trump, rather than to convert those with lessfavorable estimations. Trump also has a fraught relationship with national news organizations. He disparages outlets such as the New York Times, CNN, and MSNBC by accusing them of being purveyors of "fake news." I argue that such a strategy is an effort by Trump to increase levels of audience distrust.
My empirical analyses draw on nationally representative survey data from the American Trends Panel. I show how affect for Trump is shaped by the polarization of media audiences, selective exposure to conservative media, and, importantly, how Whites' attitudes towards the media are being shaped by Trump himself. Results show that affect for Trump is strongest among Republican leaning-partisans who get their political and election news from outlets whose audiences also tend to be Republican and conservative leaning. My analyses also show that affect for Trump is conditional on Whites' partisan selective exposure to conservative media. Furthermore, exposure to political and information news from Trump intensifies the relationship between Whites' perceptions of media bias and their levels of distrust in national news organizations. This descriptive finding is important because it further highlights the unique relationship between Trump supporters and media outlets.

Americans' Attitudes Toward the Media
Attitudes towards the US news media have been the focus of a robust literature in recent decades (Jones, 2004;Ladd, 2012). Americans widely believe that bias is present in the news media, although the extent of this bias varies and may be difficult to quantify (D'Alessio & Allen, 2000;Groeling, 2013). Individuals who believe that bias is inherent in news coverage are more likely to believe that the direction of media bias is counter to their own political beliefs (Vallone et al., 1985). In their seminal article, Vallone et al. (1985) call the propensity for individuals to perceive news as antagonistic to their own beliefs the "hostile media phenomenon." Analyzing TV coverage of the 1982 Beirut massacre, the authors found that both pro-Lebanese and pro-Israeli subjects interpreted the same news reports covering the massacre as hostile to their own personal opinions on the story. Subsequent research has found that important social identity precursors to the "hostile media phenomenon" include individual partisan identification (Eveland Jr & Shah, 2003;Matheson et al., 2001). In the US, perceptions of media bias and media hostility are especially prevalent among strong partisan identifiers. The literature finds weak evidence for the belief that news bias skews in a conservative direction among Democratic partisans (Mayer, 2005). Republican partisans, however, are very likely to perceive that media bias skews in a liberal direction (Groseclose, 2011). Importantly, elite criticism of the news media-and especially that from conservative sources-also feeds into heightened public perceptions of media bias (Watts et al., 1999). However, numerous content analyses find little evidence to support such a hypothesis (Budak et al., 2016;Covert & Wasburn, 2007;Groeling, 2008).

Media Choice and Political Behavior
The hostile media phenomenon has been examined at great length in the literature (Feldman, 2011). A particularly important strand in this body of literature analyzes the consequences of the hostile media phenomenon on consumption behavior. Lower levels of efficacy and trust in mainstream outlets drive Americans to consume relatively newer forms of news media. Before today's more "fragmented" media landscape (Dilliplane et al., 2013), the pervasiveness and accessibility of traditional forms of news media meant that Americans were compelled to consume media from mainstream sources even if they believed that such outlets were biased (Tsfati & Cappella, 2003). However, new sources of media have led to a more heterogenous media environment. 2,3 This change has given Americans with high perceptions of media bias an alternative to mainstream media sources. Consequently, consumers are now able to avoid mainstream media sources and instead turn to sources that suit their political preferences and beliefs. 4 The behavior by which Americans can now easily select media outlets that share their political beliefs is called partisan selective exposure (Stroud, 2010).
The growth and availability of partisan media via increased fragmentation has increased audience polarization (Arceneaux & Johnson, 2010). Polarization is the tendency of outlet audiences to be composed of devotees and nonviewers. For instance, research indicates that the audiences of networks like Fox tend to be more conservative (Morris, 2005). In recent years, this increased polarization has led to the rise of the idea of the "echo chamber" in the communication literature. Jamieson and Cappella (2008) define the "echo chamber" as a "bounded enclosed media space that has the potential to both magnify the messages delivered within it and insulate them from rebuttal" (p. 76). In turn, this echo chamber creates "a common frame of reference and positive feedback loops for those who listen to, read, and watch. . . media outlets" (Jamieson & Cappella, 2008, p. 76).
Research shows that politically interested individuals with a preference for a diverse range of media sources tend to avoid echo chambers (Dubois & Blank, 2017). However, in a high choice media environment, it is also the case that individuals will only select information that conforms to their own ideological predispositions or individual partisan preferences (Del Vicario et al., 2016). Echo chambers can thus promote political polarization by increasing ideological and partisan distance been individuals who consume different media (Flaxman et al., 2016). The effect of echo chambers on political polarization is such that political polarization may increase when individuals are exposed to information from outside their media bubble (Bail et al., 2018).
One of the most important debates in the communication literature is whether partisan selective exposure feeds into political polarization. Mutz argues that partisan selective exposure should lead partisan identifiers to "polarize further in the direction of their political views" (2006, p. 227). The literature finds some evidence of a causal relationship (i.e., that partisan selective exposure feeds into political polarization). In her landmark study, Stroud (2010) found that consumption of liberal and conservative news media among respondents who identified as Democrats and Republicans lead them to express more polarized political views. However, Prior (2013) argues that such research does not constitute evidence that partisan media has contributed to greater polarization. Prior (2013, p. 119) makes the important qualification that individuals such as those analyzed in Stroud's (2010) study tend to be strong partisans who turned to more "ideologically congruous media" when such formats became available.
Additional research shows that those who watch partisan television form only a small part of the wider viewing public (Levendusky, 2013a). Individual consumption of partisan media has the effect of further polarizing individuals who are already relatively extreme. Slanted news may feed into attitude polarization (Martin & Yurukoglu, 2017), but the effect is not necessarily conditional on partisan media slant (Prior, 2013). As Levendusky puts it, partisan media contribute to polarization not by converting moderates, but by "further polarizing those who are already away from the political center" (Levendusky, 2013b, p. 2). Partisan media might only reach a small percentage of Americans, but this percentage of the American public already tend to be highly interested in news and politics (Davis & Dunaway, 2016). Thus, by affecting a more vocal and engaged section of the public, partisan media have an influence on US politics that begets its small audience size relative to the mass public.
Beyond polarization, media consumption also feeds into Americans' political behavior as it relates to evaluations of political figures. Consumption of social media has been shown to increase favorable attitudes towards Barack Obama (Houston et al., 2013). Indeed, media choice not only influences affective evaluations of political candidates, the literature indicates it also feeds into voter choice (Gerber et al., 2009). 5 For instance, DellaVigna and Kaplan (2007) analyzed the entry of Fox into local television markets and the impact the network had on voting in the 1996 and 2000 Presidential elections. The authors found a substantial effect of the introduction of Fox into local television markets and increased Republican vote share between 1996 and 2000. Depending on the audience measure, watching news on Fox had the effect of convincing between 3 and 28 per cent of its viewership to vote Republican. Elsewhere, the literature finds that ideologically distinct media sources reinforce the electoral loyalty of co-partisans without influencing others. For instance, Hopkins and Ladd (2012) assessed whether Fox influenced vote choice for the Republican candidate in the 2000 Presidential election. While the average treatment effect was indistinguishable from zero, the authors found that consumption of Fox influenced the vote choice of Republican partisans and independents.

Trump and the Media
A rich body of literature examines the relationship between the media and politicians. As part of the "mediatization" of the political realm (Mazzoleni & Schulz, 1999), politicians have increasingly adapted their behavior to the requirements of the news media in order to achieve coverage. The news and entertainment media are seen as an important tool for politicians to gain electoral success (Mutz, 1989). Voters are unlikely to support candidates they know little about, and so politicians recognize the importance of the media as a platform from which they might be able to convince the public (Just et al., 1996). However, Trump represents a departure from this model; since his time in office, Trump and his network have made numerous attacks on the media. For instance, Trump has repeatedly referred to the media "enemies of the people" (Grynbaum, 2017b;Rogers, 2018). And another tactic frequently deployed by Trump on Twitter is to claim that the mainstream media reports "fake news" (Coll, 2017).
Scholars observe that contentions of "fake news" are an attempt by Trump to generate distrust in the mainstream media whilst portraying himself as a purveyor of truthful information (Ross & Rivers, 2018). Krugman (2016) calls Trump's Twitter strategy the "big liar technique," wherein the factual truthfulness of many of Trump's claims has become increasingly irrelevant in dictating their effectiveness as discourse. While the literature has explored the pervasiveness of fake news and its influence on the 2016 election (Allcott & Gentzkow, 2017;Bakir & McStay, 2018), it is only recently that the relationship between elite discourse concerning fake news and Americans' evaluations of the validity of real news has been tested. Duyn and Collier (2018) find that exposure to elite discourse about fake news leads to lower levels of trust in real news. Therefore, Trump's designation of the news content as "fake news" also has a priming effect by which Americans are distrusting real news less.

Theoretical Expectations
To my knowledge, no previous articles have directly tested whether estimations of President Trump are shaped by media effects, nor have they tested the relative influence of Trump in shaping White Americans attitudes towards the media. Building on the streams of literature in the previous section, I now turn to developing a set of theoretical expectations.
Do individual partisanship and the ideological composition of outlet audiences shape affect for President Trump? In an era of increasing political polarization, partisanship and ideology are becoming increasingly difficult concepts to disentangle (Barber & Pope, 2019). This polarization also feeds into media consumption behavior, giving rise to ideological selectivity in media use (Iyengar & Hahn, 2009). One logical expectation is that, given the increasing intractability of partisan preferences and ideological placement across different spheres of American political behavior, individual partisanship should be highly correlated with the ideological composition of outlet audiences. Of course, affect for political figures is also highly correlated with individual partisan preferences and individual ideological placement (Rogowski & Sutherland, 2016). Thus, it is reasonable to expect that these factors shape evaluations of President Trump: H1: Individual partisan preferences and the ideological composition of outlet audiences are correlated with affect for President Trump.
What happens when individuals with positive estimations of President Trump watch news media with a conservative slant (for instance, if a White American who supports Trump watches Fox)? Here, White Americans' perceptions of President Trump are reinforced rather than challenged. If individuals are presented with information that runs counter to their beliefs, they are more likely to express skepticism (Ditto & Lopez, 1992;Taber & Lodge, 2006). Conversely, consumption of information from media outlets whose coverage of a political figure conforms to an individual's estimations of that figure will strengthen those estimations (Baumgartner & Morris, 2006). Partisan selective exposure to conservative media 6 will therefore increase affect for President Trump: H2: Partisan selective exposure to conservative media leads to greater affect for President Trump There is little research that examines the effect of political figures on individual attitudes towards the media. Of course, it is of critical importance to note that Trump did not sow the seeds of Americans' distrust in the news media-the data show that Americans' trust in the media has been on the wane for decades. 7 However, Trump has repeatedly claimed that the mainstream media are purveyors of "fake news" (Coll, 2017). Given the significant relationship between fake news and perceptions of media bias, White Americans' exposure to political and election news disseminated by Trump should have a significant effect in driving their attitudes towards the mainstream media: H3: Exposure to political and election news from Trump intensifies the relationship between perceptions of media bias and distrust of national news organizations.

Analytical Strategy
To test these hypotheses, I use data from Wave 57 of the Pew Research Center's American Trends Panel (ATP), conducted between October 29, 2019 and November 11, 2019. The ATP is a national, probability-based online panel of non-institutionalized American adults over the age of 18 who live in the US (including Alaska and Hawaii). A total of n = 12,043 panel members (composed of both English and Spanish speakers) completed Wave 57. The margin of error for the full sample based on weighted estimates using the sampling weights provided by Pew is ± 1.43 per cent. In the results that follow, I present estimates from both the full sample, as well as those that are parsed by race. Models estimated using the full sample include a dummy variable for Whites. I also provide estimates from a subsample of respondents who White are only (n = 8,501) to see if the effects of media consumption of affect for Trump are larger among this group in particular. All subsequent analyses also use the sample weight provided by Pew to make inferences that a generalizable to the US adult population.
To assess the relationship between individual partisanship/outlet audience ideology and affect for President Trump, I estimate a simple linear model. This model takes the form: Where β 1 is an indexed variable of an individual's leaned partisan identification and the audience ideology of a given outlet j that is the main source of an individual's political and election news.
While the simple linear model gives us an idea of the ways in which affect for Trump is shaped by the interaction between individual partisanship and outlet audience ideology, it does not tell us whether affect for Trump is conditional on consumption of a specific news source. Therefore, to assess whether affect for President Trump is conditional on partisan selective exposure to conservative media, I estimate the average treatment effect (ATE) of consumption of conservative media on White affect for Trump.
Media preferences are not randomly assigned; rather, individuals selectively expose based on their ideological and partisan preferences (Stroud, 2010). Nonetheless, I use a variety of strategies to address confounding where possible given the observable data. Treatment effects are estimated using the inverse probability weight (IPW) estimator. IPW attempts to remove confounding by creating a "pseudo" population in which the treatment is independent of the measured confounders. Overall, I find strong balance on the key explanatory variables and sociodemographic covariates that are used to estimate treatment models. I report balance analysis for treatment effects in Supplemental Section A1.
Another important factor to consider is the problem of omitted-variable bias that may arise from any unmeasured confounders. Of particular interest in the context of the current study is the role of racial attitudes and trust in government, since both of these variables are related to both media use and affect for Trump (Moy & Scheufele, 2000;Schaffner et al., 2018). In the case of racial attitudes especially, therefore, there is reason to expect that such attitudes are a mechanism that underlines the effects of news consumption on affective evaluations of Trump. Due to a lack of items concerning trust in government and White racial attitudes in the ATP dataset, I also perform sensitivity analysis using the sensemakr package in Stata. 8 The results of the sensitivity analysis suggest that the results are (largely) robust to unmeasured confounding; for all models with the exception of the Daily Caller, the sensitivity analysis indicates that the results are robust to an unobserved confounder that accounts for approximately 20 percent of the residual variance in both the treatment and the outcome. 9 Full results of the sensitivity analysis are presented in Supplemental Section A2.
Finally, I am also interested in testing the hypothesis that exposure to political and election news from Trump intensifies the relationship between perceptions of media bias and distrust of national media organizations. To assess this hypothesis, I estimate ordered probit models. These models begin with a two-way interaction term between the dichotomous variables for exposure to political/election news from Trump and perceptions of media bias. I then use postestimation to graph the marginal effect of exposure to political/election news from Trump on the probabilities that a White respondent will trust/distrust national news media organizations as their perceptions of media bias increase. 10

Data and Measures
Affect for Trump. Affect for President Trump is gauged on the standard thermometer scale. Respondents were asked to rate President Trump on a scale ranging between 0 and 100. A minimum score of 0 indicates "very cool or unfavorable feelings," whilst a maximum score of 100 indicates "very warm or favorable feelings."

Polarized Audiences
This is an indexed variable created by Pew and is constructed using an item that measures an individual's leaned partisan identification and a second item that groups each media outlet according to the ideological composition of its audience. This latter item is based on (1) where respondents reported that they got political and election news from and (2) how respondents described themselves ideologically. 11 The recoded variable is coded such that 1 = "Republican who only gets news from outlets with right-leaning audiences,"; 7 = "Democrat who only gets news from outlets with leftleaning audiences."

Partisan Selective Exposure
My key measures for partisan selective exposure include a number of items that ask respondents whether they have received political and election news from a variety of outlets in the past week. The items are all dichotomous variables coded 1 = "yes," and 0 = "no." In order to demonstrate that affect for Trump is conditional on consumption on conservative news media across a variety of media formats, I include shows across a range of formats including talk radio, cable television, and online news websites. For talk radio, I include The Rush Limbaugh Show and The Sean Hannity Show. For cable television, I include Fox. For online news websites, I include The Daily Caller and Breitbart. And for print, I include The New York Post. Another key measure of exposure is whether a respondent received political and election news directly from President Trump in the past week. This variable is also dichotomous and is coded 1 = "yes," and 0 = "no."

Media Hostility
My key measures for media hostility are items that gauge Whites' levels of distrust in national news media organizations, as well as their perceptions of bias. The item for distrust of national news media organizations is a four-point ordinal variable ranging from 1 = "trust a lot," to 4 = "does not trust at all." The item for perceived media bias is a dichotomous variable that asks respondents whether, in presenting the news and dealing with political and social issues, the media tend either 1 = "deal fairly with all sides," or 0 = "tend to favor one side."

Control Variables
In addition to the variables outlined above, my analysis includes a number of individual-level covariates that I employ as controls. News interest is a four-point ordinal variable that asks respondents how closely they follow political and election news. The variable ranges from 1 = "very closely," to 4 = "not at all closely." Political knowledge is coded 1 for respondents who correctly answered which party controlled the Senate at the time of the survey, and 0 for respondents who answered incorrectly. Party ID is a threepoint categorical variable where 1 = "Republican," 2 = "Independent," and 3 = "Democrat." 12 Ideology is a fivepoint ordinal variable that ranges from 1 = "very conservative," to 5 = "very liberal." White is a dichotomous variable where 1 = "White," 0 = "nonwhite." Age is a four-point ordinal variable ranging between 1 = "18-29 years old," 4 = "65+." Females are coded as 1, and men 0. A respondent's level of education if a six-point ordinal variable ranging from 1 = "less than high-school," to 6 = "postgraduate." Family income is a nine-point ordinal variable ranging from 1 = "less than $10,000," to 9 = "$150,000 or more." 13 Church attendance is a six-point ordinal variable that asks respondents their frequency of attendance at a religious service. The variable ranges from 1 = "more than one a week," to 6 = "never." South is a dichotomous variable coded 1 for respondents living in the South, and 0 for those residing elsewhere.

Individual Partisanship, Outlet Ideology, and Affect for President Trump
I first present results of a simple linear regression using the full sample from the ATP assessing the strength of the relationship between affect for Trump and individual partisan identification/outlet audience ideology. Figure 1 plots the strength of this relationship. Perhaps unsurprisingly, affect for Trump is strongest among Republican-leaning individuals who get political and election news from outlets whose audiences are likewise composed of Republican and conservative-leaning viewers. The overall trend of the regression plot is that, as the ideological composition of outlet audiences becomes more Democratic and left leaning, affect for Trump decreases (that is, individuals will give Trump a lower thermometer rating).
To get a better picture of the relative influence of media outlets, each outlet is also assigned a weighted score that corresponds to the number of individuals who reported that outlet as their main source of political and election news. As indicated by Figure 1, the outlet with the largest audience size for political and election news is Fox. Overall, 24 per cent of White respondents reported that Fox was their main source of political and election news. The results are consistent with observations made in popular commentary and contemporary scholarship concerning the wide viewership of Note. Each outlet is assigned an average affect score for Trump (based on the average thermometer rating for Trump when respondents from the full sample is nested by their principle source of political and election news) and a mean score for the partisan lean/ideological composition of an outlet's audience. The mean score for party ID/audience ideology is then regressed against the average outlet affect score for Trump. Circles are markers weighted by audience size. Red markers indicate that a given outlet's audience is more Republican and right leaning. Blue markers indicate that a given outlet's audience is more Democratic and left leaning. Adjusted R 2 = 872.
Fox and the influence of the network in shaping the political behavior of its audience (Martin & Yurukoglu, 2017). 15 Given the relative influence of Fox on individual media preferences, we begin to understand Trump's continued obsession with television ratings in the age of digital media (Adalian, 2017); he recognizes the importance of inserting himself into non-digital media formats that garner wide viewership.

Partisan Selective Exposure and Affect for Trump
To understand the causal effect of selective partisan exposure to conservative media on affect for Trump, and whether such effects may be larger for Whites, I estimate treatment effects. Here, the causal effect of interest is the impact that consuming political and election news from conservative media outlets (the treatment) has on individual estimations of Trump. How much higher do individuals who get their political and election news from conservative outlets rate Trump on the thermometer scale than those who do not? The average treatment effect (ATE) in this case is the difference in means of the treatment groups and control groups selective exposure to conservative media. Table 1 shows the ATE of consuming political and election news from each of the outlets in both the full sample and the White subsample. The size of each effect is the estimated change in Whites' thermometer rating for Trump when individuals consume conservative media. The first column in Table 1 presents the ATEs for the full sample with the White dummy variable. Conversely, the second column in Table 1 presents the results of the Whites only sample. Comparing the estimates, we see that the ATEs for the full sample (that includes nonwhites) are somewhat smaller than those for the White subsample. The results thus provide some evidence in support of the hypothesis that there are heterogenous effects by race in the effect of media consumption on affect for Trump; With the exception of selective exposure to the New York Post, the results indicate that selection exposure to political and election news from a variety of conservative outlets will increase affect for Trump, and that this effect is more salient among the White only sample.

Media Hostility
Does the effect of exposure to political and election news from Trump on Whites' estimations of the trustworthiness of national news organizations become more salient as perceptions of media bias increase? The results of two ordered probit models using both the full sample and the White subsample are presented in Table 2, which reports: (1) the main effect of getting political and election news from Trump on distrust in national news organizations, (2) the main effect of perceptions of media bias on distrust in national news organizations, and (3) the coefficients of the interaction between an individual's perceptions of media bias and their exposure to political and election news. Figure 2 also graphs the outcomes from the two samples. The graphs plot the marginal effect of exposure to political and election news from Trump at each level of the dichotomous variable for media bias across outcomes of the dependent variable. The top panel in Figure 2 graphs the results from the full sample. Here, the trajectory of the marginal effect is somewhat flat across outcomes of the dependent variable for trust in national news organizations. Given these trends, does limiting the sample to Whites only result in more significant changes in the probability of trusting news organizations? Using the Whites only sample, the bottom panel in Figure 2 shows that exposure to political and election news from Trump reinforces the views of Whites who are polarized by their levels of trust in national news media organizations. Among Whites who trust the national news media "a lot" and those who do not trust the national news media "at all," the trajectory of the marginal effect of exposure to political and election news from Trump is positive. The trend is such that, for higher levels of perceived bias, the effect of exposure to political and election news from Trump is estimated to be higher.
Conversely, the trajectory of the marginal effect of exposure to political and election news from Trump is negative Note. Each reported coefficient is from a different estimator. Robust standard errors given in parenthesis. Coefficient estimates from OLS regressions weighted by Pew sample weight. Dependent variable is Trump thermometer (rescaled to range between 0 and 1). OLS controls include source of news (dummy codes for radio, cable television, online news websites, and print), digital savviness (online news websites only), interest in political and election news, political knowledge, party ID, ideology, race (full sample only), age, gender, education, family income, frequency of church attendance, and region. Estimation method is inverse-probability weighting (IPW). ***p < .001. Data are weighted. Figure 2. The marginal effect of exposure to political and election news from trump on the probability of trusting National News Organizations by perceptions of media bias, by race.
Note. Dependent variable is level of trust in national news organizations. Ordered probit model contains an interaction term between dichotomous variables for exposure to political and election news from Trump and perceptions of media bias. Model also controls for source of news (dummy code for social media), digital savviness, interest in political and election news, political knowledge, party ID, ideology, race (Whites only) age, gender, education, income, frequency of church attendance, and region. All covariates are set to their respective means. Vertical lines are 95 per cent confidence intervals. Data are weighted. Note. Table entries are beta coefficients with robust standard errors given in parenthesis. Dependent variable is level of trust in national news media. Models also control for social media use, digital savviness, news interest, political knowledge, party ID, ideology, race (full sample only) age, gender, education, income, church attendance, and region. All variables coded to range between 0 and 1. Data are weighted. *p < .05. **p < .01. ***p < .001.
among Whites who are less polarized by their estimations of trust in national news media organizations. That is, as levels of perceived bias increase, exposure to political and election news from Trump decreases the probability of trusting national news organizations among Whites who trust such organizations either "some" or "not too much." The results of the interactive model thus provide some evidence that exposure to political and election news from Trump feeds into distrust of national news organizations as perceptions of bias become more salient.

Discussion
Trump is a unique case in the annals of American Presidents when it comes to his relationship with the media. The focus in commentary and scholarly debate has focused on the ways in which Trump's behavior represents a threat to a critical check to Presidential power. However, little attention has been paid to the important role of White audiences. I argue that this gap is surprising given that, demographically, White audiences compose an overwhelmingly large part of Trump's following on both cable and digital media formats (Price, 2018;Wang et al., 2016). Platforms such as Fox and Twitter offer Trump a vital link to his key constituency (i.e., White Americans). In this article, I have addressed this critical gap by analyzing the relationship between Trump, the media, and White audiences. Specifically, I have sought to better understand how affect for Trump is shaped by audience polarization and Whites' partisan selective exposure to conservative media outlets. I have also argued that Trump attempts to foster distrust in national news organizations. He does this through the usage and deployment of tactics that are commonplace in populist discourse, including accusing the media of being "enemies of the people," as well as questioning the veracity of mainstream media content through the epithet "fake news." My analyses show that Trump is an important moderator of the relationship between perceptions of media bias and distrust of national news organizations. When Whites who are less polarized by their views concerning levels of trust in the national news media are exposed to political and election news from Trump, they tend to trust the news less as their perceptions of media bias increase. This is a critically important finding because polarized audiences tend to only represent a small overall percentage of the wider viewing public (Levendusky, 2013). Those who are less polarized, conversely, compose a larger segment of audiences. If exposure to information from Trump is leading more moderate audiences to distrust national news media less therefore, then Trump's efforts to delegitimize the national news media are influencing a relatively large section of White audiences.
Nonetheless, it is important to note that these results offer a purely descriptive illustration of the unique relationship between Trump supporters and media outlets, rather than an attempted to establish causality. Though a more in-depth analysis is beyond the scope of the current article, there is likely to be a degree of reverse causality between exposure to messages for Trump and trust in media outlets. For instance, research shows that trust in institutional actors drives support for Trump among Republican identifiers (Dyck et al., 2018) Consequently, if individuals with low levels of trust seek out information from actors such as Trump as they do traditional media (Stroud, 2010) it is possible we may also observe a similar pattern of a negative correlation between bias and distrust Tellingly, this may occur even if Trump's messaging had not direct effect.
Media sources such as cable television remain a critically important link to reaching his audience. Conservative media outlets such as Fox are behemoths in the cable television market. Trump's continued links with Fox make sense given the correlation between the cable network's audience size and Trump's preoccupation with television ratings. Of course, how many people watch news matters. There is a tendency for survey respondents to overreport their levels of self-reported news exposure. Prior (2009) found that, compared to actual ratings from Nielsen, some demographics exaggerated exposure rates by a factor of three. However, these concerns are mitigated to an extent given the audience demographics of outlets that are favorable to Trump. Older audiences overreport significantly less than younger demographics (Prior, 2009, p. 136). This is an important observation because audiences of outlets such as Fox are generally likely to skew older. Further, much has been written about the exodus of young audiences from outlets such as Fox in recent years. Despite these concerns, the pervasiveness of Fox in reaching White audiences is acknowledged by many including President Obama.
My critical argument in this article concerning the role of Whites' partisan selective exposure to conservative media is that, by consuming conservative media, Whites' existing estimations of Trump are reinforced (and in some cases, heightened) rather than challenged. The current nature of audience polarization is that, today, conservative media essentially functions as a giant echo chamber (Jamieson & Cappella, 2008) in which Whites with favorable estimations of Trump consume information that conforms to their existing estimations of the President. In the Trump era, however, many have observed that coverage of Trump by conservative outlets goes beyond partisan coverage. Outlets such as Fox have been criticized by political commentators as "destructive propaganda machines," (Haag, 2018) "pro-Trump state TV" (Mayer, 2019) in their coverage of Trump. My analysis of the data here would seem to lend weight to the notion that sycophantic coverage of Trump in conservative media is evidence of such hyperpolarization. It is not simply the case that Whites who consume conservative media have favorable estimations of Trump. In many cases, the ATE of partisan selective exposure to conservative media is large enough that consumption of conservative media is able to cut through less favorable or cooler evaluations of President Trump.
In sum, these consistent set of results demonstrate that Trump plays an important role in shaping White American attitudes towards the media. I have argued that it is critical that we understand Trump's relationship with the media in the context of his audience. White Americans, who make up an overwhelming share of the Trump audience on both traditional and non-traditional media platforms, are an important contextual variable in furthering our understanding of this relationship. The recent success of radical populist parties and political actors around the world means that media institutions face an unprecedented crisis of legitimacy as populists seek to emphasize and drive distrust between "the people" and media elites. The "Fourth Estate" of American democracy has always faced challenges from executive actors with an axe to grind for negative coverage. However, the challenge posed by Trump seems to be more existential if-as my analysis in this article would indicate-he is able to influence levels of distrust in the media among his key audience while concurrently avoiding the scrutiny that politicians face in with traditional media contexts.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplemental Material
Supplemental material for this article is available online. Notes 1. A 2018 report citing viewership data from Nielsen showed that the Fox News audience was 94 per cent non-Hispanic White, 3 per cent Hispanic, 2 per cent Asian American, and 1 per cent African American (Price, 2018). 2. Cable outlets such as Fox dominate the new media landscape.
Political commentators point to the success of Fox because of the networks ability to appeal to conservatives who were disillusioned with traditional network television news and who perceived traditional venues as possessing a liberal bias (Collins, 2004). 3. Nationally syndicated radio shows such as The Rush Limbaugh Show and The Sean Hannity Show now dominate the airwaves. In recent years, radio stations have found themselves in an increasingly perilous position as stations with a music format have become more unprofitable. Many stations have adjusted by developing a successful model which involves converting to talk-formats that air nationally syndicated shows (Berry & Sobieraj, 2011). 4. Media fragmentation has also been accompanied by a plethora of new entertainment television, meaning that Americans can avoid political news altogether if they so choose (Bennett & Iyengar, 2008;Prior, 2007). In turn, people who abandon political news in favor of entertainment television become less likely to learn about politics (Prior, 2005). 5. Equally, scholars have found that some concerns regarding the effect of media choice on voting behavior are exaggerated (Hayes, 2009). An alternate explanation posits that effects such as distrust in the media and media skepticism are likely to influence vote choice (Ladd, 2010). 6. It is critically important to acknowledge that judgments regarding which outlets should be defined as "conservative media" in analyses concerning Americans media preferences can significantly influence results (Druckman et al., 2018). In this article, conservative media are defined according to the ideological composition of outlet audiences. Outlets whose audiences are more conservative and Republican leaning than the average American adult are defined as "conservative media" (see Jurkowitz et al., 2020, p. 19). I then cross-matched outlets that met this criterion of "conservative media" with available items in the Pew dataset. After searching the dataset using this strategy, relevant items were found for the following outlets: The Rush Limbaugh Show, The Sean Hannity Show, Fox, The Daily Caller, Breitbart News, and The New York Post. All subsequent analyses into the effect of "conservative media" on affect for Trump use items that ask respondents about these outlets 7. Trends in White Americans' attitudes towards the media indicate that a growing percentage of Whites are likely to report that they have little faith in the media. According to the General Social Survey (GSS), just 15 per cent of Whites reported that they had "hardly any" confidence in the press. By 2018, this figure had risen to 50 per cent (See Supplemental Figure A1). 8. Sensemakr is useful because the package is able to handle multiple potential confounders. Critically, the package generates sensitivity statistics such as robustness values, which provide an indication as to the minimum strength that unobserved confounders would need to have to overturn a research conclusion (Cinelli & Hazlett, 2020). 9. Though this article does quantify whether the approximate figure of 20 percent is an "acceptable" level of confounding given the use of inverse probability weighting (IPW) to estimate treatment effects, it is nonetheless imperative to note that this level of robustness is important to be aware of. As Cinelli and Hazlett (2020) note, this is so that researchers are able to "conduct an imperfect investigation, while transparently revealing how susceptible our results are to confounding." 10. Results of the full models are presented in Supplemental  Table B1. 11. For a comprehensive discussion of the construction of this variable see Pew Research Center (2020). 12. While desirable, the American Trends Panel dataset did not include the standard seven-point scale for partisan identification. 13. Respondents who refused to provide their incomes were omitted from the analysis.