Slant, Extremity, and Diversity: How the Shape of News Use Explains Electoral Judgments and Confidence

Abstract

The 2020 election and its aftermath present an opportunity to understand how audiences’ changing news consumption patterns within an expanded, digitized, and polarized media environment shape electoral judgments. This paper introduces measures that capture individuals’ partisan slant, partisan extremity, and overall diversity of news media use to understand how people interact with the contemporary news ecology. The measures combine self-reported media use with the partisan slant of news outlets along the ideological spectrum. Using these measures, we analyze a two-wave panel survey conducted before and after the 2020 US election to investigate how slant, diversity, and extremity are related to post-election beliefs, including public confidence in the election and the acceptance of fraud claims. Our findings show that Republicans have more insulated news use behaviors in terms of slant and diversity. The analysis also reveals that the slant of people’s news use is associated with post-election fraud beliefs, with right-wing news consumers more likely to believe such claims. However, a diverse news consumption style can moderate misinformation beliefs. Panel analysis points to the role of extreme-right news use in decreasing confidence in the legitimacy of the election.

The news media ecology in the United States has expanded, digitized, and polarized, providing audiences with nearly unlimited choices (Chadwick 2017). Fierce competition between outlets, intertwined with rising ideological extremity among elites and elements of the public, has driven news media to take niche positions to maintain audience engagement (Stroud 2011). This competition for attention has pushed some purveyors of partisan news content to stake out extreme political stances, especially during contentious periods (Mullainathan and Shleifer 2005). In such an ecosystem, US news consumers can easily choose content that is slanted toward particular partisan perspectives, favor ideologically extreme outlets, or opt for more balanced and diverse consumption across the spectrum.

While communication scholarship documented the increasing polarization of the news media ecosystem, these changes provoke additional questions: (1) What do news consumption patterns look like in the contemporary news ecosystem? (2) What are the implications of these patterns for political judgments? To answer these questions, we developed a set of measures to capture the partisan slant (i.e., ideological tilt), partisan extremity (i.e., the proportion of far-right and far-left partisan news consumption), and diversity of an individual’s news diet (i.e., range of news across the ideological spectrum). We combine self-reported news source usage with a partisanship score of the news source using a network approach (Benkler et al. 2017; Faris et al. 2017; Faris et al. 2020). Using a two-wave panel survey conducted before and after the 2020 US election, we compared the slant, extremity, and diversity of news consumption between partisans to understand asymmetric polarization regarding news consumption. In the 2020 campaign cycle, defined by unprecedented misrepresentation and contestation over election administration, cloistered communication consumption patterns may have shaped election fraud beliefs, especially on the political right, with individuals’ slant, diversity, and extremity of news use jointly shaping electoral judgments and confidence.

Literature Review

Explicating Partisan News Consumption

While digital journalism affords audiences greater access to a diverse range of information, the networked public sphere between partisan news outlets and their audiences has witnessed increasing homophily. Partisan media seeks niche audiences and then is constrained by them to confirm their ingroup membership (Stroud 2011; Wells et al. 2016). This creates an increasingly partisan media ecosystem, where both news outlets and audiences pursue ideologically insulated facts and reality (Carey et al. 2016), leading many news outlets away from the norm of neutrality that constrained journalists from adjudicating the truth among opposing perspectives (Pingree et al. 2013). Under this ecosystem, ideologically extreme outlets actively construct the “truth” to align with their partisan leanings.

Both the increasingly partisan media ecosystem and the often-selective audience (Iyengar and Hahn 2009) call for a closer examination of individuals’ patterns of news use and their implications. The shift in news production has altered consumption patterns, shifting aspects of individuals’ news diet, resulting in individual differences in slant, extremity, and diversity of sources. The sizable literature advancing news consumption measures needs updating in relation to media proliferation and political polarization. For example, a substantial body of work examines media effects by investigating the political consequences of viewing a particular outlet. For example, Gil de Zúñiga, Jung, and Valenzuela (2012) found that selective exposure to Fox News amplified anti-immigration attitudes. Some studies aggregate the type of news outlets and examine how categories of consumption shape different political behaviors (Warner 2018). Yet, focusing on the influence of distinct news outlets is insufficient to gauge the complexity of the networked public sphere. The same levels of right-wing news use, for example, could have differing impacts on people entrenched in right-wing news use compared to those with a more balanced news diet.

To understand the overarching structure of an individual’s news consumption, media repertoire scholars have classified audiences’ consumption by patterns of use to find combinations/clusters of audiences’ media usage (Hasebrink and Popp 2006). Cross-media and cross-platform use are examined to capture how people navigate online news, television, and newspapers (Edgerly 2015; Kim and Kim 2018). Trilling and Schoenbach (2013) further develop the approach using cluster analysis for the “fragmented” and “polarized” media use types, hinting at the linkage between audience and political structure. While the repertoire approach addresses the complexity of news, it does not explicitly connect individuals’ news diets to structural or ideological factors or consider different dimensions of individuals’ usage patterns and their relation to political judgments.

A set of integrated yet multidimensioned measures that consider the range of news use across the partisan spectrum is needed to unpack usage patterns. To capture different dimensions of an individual’s partisan news diet, we consider news use in relation to a spectrum of consumption options. We look at the slant of usage to understand the ideological tilt of news flow. We examine extreme news use relative to the individual’s total news volume to capture the skewed nature of partisan news use beyond a simple assessment of slant. The diversity of news use is also studied to map the homogeneity of one’s partisan news flow. In this way, the shape of an individual’s partisan news use can be understood in relation to the fractured media landscape.

Asymmetric Polarization in News Use

Contemporary political and media structures in the United States have seen both heightened levels of ideological polarization (Kubin and von Sikorski 2021) and an asymmetric nature to said polarization (Faris et al. 2020). This is evident in networked patterns that define media interrelationships. Analyses of the 2016 election showed ideologically cloistered patterns of news sharing on the political right and asymmetries in the degree of polarization of partisan networks (Benkler et al. 2017, 2018). Building on this work, Tripodi (2022) highlights the role of the right-wing information ecosystem in which conservative voters exercise greater agency in their political choices, facilitating conservative leaders’ propaganda. Notably, by these same measures, asymmetry increased between 2016 and 2020 (Faris et al. 2020). Trump supporters’ sharing of right-wing news became more extreme in 2020, whereas Biden supporters moderated toward the center relative to Clinton supporters in 2016 (Benkler et al. 2018). These metrics serve as a proxy for how news attention is divided along partisan lines, yet empirical confirmation of such asymmetric polarization requires direct testing. Hypothesizing that the slant of news use will be asymmetric toward the right, we compare patterns of individual news use between partisan groups.

(H1a) Republicans’ news use shows a preference for conservative outlets over liberal outlets, whereas Democrats’ news use shows a preference for liberal outlets over conservative outlets.

Another aspect of the asymmetry in news consumption is extremity. Extremity reveals the extent of polarization that is not fully captured by slant. The asymmetric development of news media has witnessed a larger number of popular extreme-right media outlets than extreme-left ones, and the slant of these right-wing outlets tends to be more extreme (Schroeder 2019). A “distinct and insulated media system” has been developing on the right focused around Breitbart and Fox News (Benkler et al. 2017), coupled with a relative void of center-right news outlets, pushing conservatives to consume the extreme ideology these news outlets produce (Faris et al. 2020). Extreme-right news outlets actively share links within the extreme-right media ecosystem, making the network insular and distinct from others (Benkler et al. 2020). Therefore, conservative audiences drawn into this system are exposed to extreme-right views without crosscutting exposure to opposing or contrasting perspectives. Audiences also engage more with extreme outlets through social media, further amplifying the power of extreme news consumption on the political right (Figenschou and Ihlebæk 2019; Nadler 2022).

Compared to the balkanization of the extreme-right media ecosystem (e.g., Nadler 2022), extreme-left news outlets share their media ecosystem with center-left and centrist outlets (Faris et al. 2020), and audiences with far-left ideology have more opportunities to be exposed to moderating political ideas (Benkler et al. 2020). Studies also found that while extreme-right news outlets get focused attention from audiences on the right, a relatively even number of outlets on the extreme left, left, and center get attention from other audiences (Benkler et al. 2020; Faris et al. 2020). These disparities not only underscore the asymmetrical consumption patterns of extreme-right and -left news media, but they also suggest varying effects of extreme news usage across the ideological spectrum. To capture the distinct patterns of extreme news consumption on the opposing ends of the political spectrum, this study differentiates between right-wing and left-wing extremity in news use patterns. This distinction considers the asymmetrical news consumption behaviors of liberal and conservative audiences. Attending to the unique engagement patterns of partisan audiences with extreme-right and extreme-left news outlets, we hypothesize:

(H1b) Republicans’ news use preference for conservative outlets is stronger than Democrats’ news use preference for liberal outlets.

Although the US media structure has become more polarized and highly asymmetric, the information environment as a whole has become unprecedentedly rich. Some studies suggest that people are exposed to heterogeneous content, highlighting the positive effect of information explosion (Shapiro 2013). Indeed, many scholars counter claims of selective exposure and echo chambers with claims of more omnivorous news consumption that is less ideologically isolated and biased (Bakshy, Messing, and Adamic 2015; Möller et al. 2018). Any media audience theory in this high-choice environment must consider both the possibility of selective exposure to pro-attitudinal content, leading to the reinforcement of preexisting attitudes, and the crosscutting exposure to a more diverse array of content and sources, leading to the reconsideration of existing attitudes (Shah et al. 2017). In this sense, diversity refers to the number of outlets the individual uses among an array of options available. Based on prior evidence about the “propaganda feedback loop” (Tripodi 2022) on the right, we hypothesize:

(H1c) Republicans have less diversity in news use across different outlets than Democrats.

Understanding how Independents consume news in relation to partisans is crucial, given their often misunderstood role and significant political impact (Klar and Krupnikov 2016). Unsurprisingly, abstention from political affiliation can disrupt ideological segregation, facilitating crosscutting news consumption (Oliveros and Várdy 2015). To explore this question, we consider the slant, diversity, and extremity of Independents’ news use patterns.

(RQ1) Do Independents exhibit unique news consumption patterns compared to Democrats and Republicans in terms of slant, diversity, and extremity?

Patterns of News Use and Electoral Confidence

Individuals’ news consumption patterns, characterized by asymmetric polarization and amplification (Benkler et al. 2017), can significantly shape their political beliefs and judgments. This was particularly evident during the 2020 election. Narratives questioning the legitimacy of the vote emerged across the news spectrum, especially on the right, months before Election Day.

Benkler and colleagues (2020) conducted an analysis of online media stories, tweets, and posts on public Facebook pages, revealing that Donald Trump orchestrated an agenda suggesting the potential for widespread voter fraud through mail-in ballots. This agenda was propagated by Trump and the Republican National Committee, through a combination of social media posts, press conferences, and interviews with favorable outlets (p. 2).

A substantial body of literature has established that elite opinion can significantly influence public opinion (Druckman 2004; Gabel and Scheve 2007). Berlinski and colleagues (2023) demonstrated that elites can shape confidence in elections using unsubstantiated claims. Here, the elite-driven and partisan-media-supported disinformation campaign on voter fraud amplified concerns of election legitimacy, particularly among conservative audiences (Benkler et al. 2017).

This disinformation campaign on the right peaked just after Election Day, questioning the results, with centrist and left-leaning news correcting the election fraud conspiracy and reiterating electoral legitimacy. Conservative-leaning news outlets like Breitbart and Fox supported Trump’s position, questioning the legitimacy of the election and promoting Trump as the true winner. Left-leaning news media, such as MSNBC and CNBC, countered Trump’s claim, condemning it as devastating to the integrity of democracy (Benkler et al. 2020). Contemporaneous studies found that Trump supporters who closely followed campaign news believed false claims that voter fraud was denying Trump’s reelection (Calvillo, Rutchick, and Garcia 2021; Pennycook and Rand 2021). The slant of news use, in this context, can be a proxy for information flows that erode confidence in election integrity and legitimacy, bolstering conspiracy theories and illegitimate authority. Thus, we hypothesize, concerning the relationship between news use slant after the election and post-election beliefs,

(H2) Individuals whose post-election news use shows a preference for conservative news outlets have (H2a) lower electoral confidence, (H2b) stronger belief in a voter fraud conspiracy, and (H2c) higher approval for Trump than individuals whose post-election news use shows a preference for liberal news outlets.

While post-election news slant is important for post-election attitudes and beliefs, voter fraud conspiracies were propagated long before the election. To further tease out the role of news use slant during the election, we consider the relationship between the change in news use slant and the corresponding change in beliefs before and after the election. We hypothesize:

(H2d) Compared to individuals whose post-election news use shows a preference for liberal news outlets, electoral confidence decreased and (H2e) evaluations of Trump increased more among individuals whose post-election news use shows a preference for conservative news outlets.

The insularity of the extreme-right media ecosystem is particularly problematic because these outlets weaponize misinformation in support of ideological positions (Benkler et al. 2020; Bergmann 2020). While both extreme-left and extreme-right news media are vulnerable to the penetration of misinformation, the isolation of the extreme-right media ecosystem inhibits conservative audiences from “fact-checking” given misinformation, further falling into the “propaganda feedback loop” (Benkler et al. 2020, p. 49). Extreme-right news outlets were especially enthusiastic in endorsing voter fraud claims regarding mail-in ballots and outlining related conspiracies (Hsu 2020). Unchallenged exposure to extreme-right news could have led Republican audiences to more strongly endorse voter fraud claims and conspiratorial thinking. Thus, we hypothesize:

(H3) Individuals who consume a larger proportion of extreme right-wing news have (H3a) lower electoral confidence, (H3b) stronger belief in a voter fraud conspiracy, and (H3c) higher approval for Trump than individuals who consume a smaller proportion of extreme right-wing news.

(H3d) Electoral confidence decreased more and (H3e) evaluations of Trump increased more as an individual’s extreme right-wing news consumption accounted for a larger proportion of the total news consumption after the election.

Information diversity is another critical factor shaping political beliefs. Lack of information heterogeneity can potentially reinforce predispositions and political misinformation. Meanwhile, exposure to a balanced and diversified informational environment can facilitate political tolerance and understanding (Mutz and Mondak 2006), attenuating the relationship between ideological extremity and projection (Wojcieszak and Rojas 2011). Diverse news use should bolster attitudes toward democratic processes, dampen belief in voter fraud conspiracies, and lower approval of Trump, given the widespread condemnation of his election fraud claims. We offer the following hypotheses concerning the diversity of news use:

(H4) Individuals with higher news diversity have (H4a) higher electoral confidence, (H4b) lower belief in a voter fraud conspiracy, and (H4c) lower approval for Trump, the source and primary amplifier of these claims, than individuals with lower news diversity.

We further hypothesize that shifting electoral confidence and evaluations of Trump are linked to change in the diversity of news consumption:

(H4d) Electoral confidence decreases more and (H4e) evaluations of Trump increase more as an individual’s news use becomes more diverse after the election.

The structure of all the hypotheses we test in this paper is summarized in table 1.

Table 1.

Summary of hypotheses.

Explanatory variable	Outcome variable	Direction	#
Partisan group (Republicans compared with Democrats)	Slant of news use	More biased	H1a
	Right extremity	Higher	H1b
	Diversity of news use	Less diverse	H1c
Slant of news use (More right leaning)	Electoral confidence	Lower	H2a
	Belief in voter fraud conspiracy	Higher	H2b
	Approval for Trump	Higher	H2c
Slant of news use [change] (Increasingly right leaning)	Electoral confidence	Decreased more	H2d
	Approval for Trump	Increased more	H2e
Right extremity of news use (Higher)	Electoral confidence	Lower	H3a
	Belief in voter fraud conspiracy	Higher	H3b
	Approval for Trump	Higher	H3c
Right extremity of news use [change] (Increasingly extreme)	Electoral confidence	Decreased more	H3d
	Approval for Trump	Increased more	H3e
Diversity of news use (Higher)	Electoral confidence	Higher	H4a
	Belief in voter fraud conspiracy	Lower	H4b
	Approval for Trump	Lower	H4c
Diversity of news use [change] (Increasingly diverse)	Electoral confidence	Increased more	H4d
	Approval for Trump	Decreased more	H4e

Method

Data

We conducted a two-wave panel survey through Social Science Research Network (SSRN) before and after the 2020 US presidential election in two US swing states, Wisconsin and Pennsylvania. We implemented a hybrid design in our sampling strategy. The panel participants were initially recruited using a nationally representative Address-Based Sampling method, followed by an opt-in, nonprobability sampling approach to increase the sample size. Quota-based sampling matched the sample populations on age, gender, race, income, and region. We used data from these surveys to (1) construct the measures for individuals’ partisan slant, extremity, and diversity of news consumption; and (2) examine how these indices are related to electoral confidence, voter fraud conspiracy beliefs, and support for Trump. The first wave of data was collected from October 21 to November 1, 2020. The second wave was collected from December 7 to December 15, 2020, successfully completing 39.6 percent of respondents from the pre-election wave. Since the original sample was a nonprobability sample, we cannot provide an AAPOR response rate. Of those the survey firm contacted to participate in wave 1, the cumulative response rate was 9.89 percent. To test our hypotheses, we only included (1) participants from both waves and (2) those who self-reported to be either Republican, Democrat, or Independent, yielding a sample of 1,833. The specific wording of questions and their order can be found in Supplementary Material III, table S4.

We ran several one-way ANOVAs with the panel’s post-election data (Wave 2) to compare patterns of partisan news consumption among Democrats, Independents, and Republicans. To understand the relationship between partisan news consumption and post-election beliefs and attitudes, we ran a set of cross-sectional regression models using the post-election data. Then, we leveraged the panel format and ran concurrent residual models to test the changes in citizens’ electoral confidence and approval of Trump over the election period relative to news usage patterns. Missing data in the analysis were handled through pairwise deletion, where each model was run using all available data for the variables included in that specification.

Measures

The slant of news use. To examine the ideological skew in news consumption among partisans, we calculate the slant of news use (M = −0.34, SD = 3.43, range: −9.50–14.08 in W1; M = −0.39, SD = 3.40, range: −9.42–15.08 in W2) by considering an individual’s use of 16 different outlets as a function of partisan attention scores (Benkler et al. 2017; Faris et al. 2020) of these outlets. These scores were derived from tracking the frequency of sharing these news sources among users who retweeted messages from either of the two general election candidates (Benkler et al. 2017). Specifically, the proportion of retweets associated with either candidate for each media source was used as a measure of candidate-centric partisanship, expressed on a -1.0 to 1.0 scale. Media sources included only in the tweets of accounts that retweeted Biden received a score of -1.0, and media sources included only in the tweets of accounts that retweeted Trump received a score of 1.0. Media sources included in tweets from equal numbers of accounts in both groups scored 0. This networked approach offers a nuanced, comparative understanding of the news media ecology (Faris et al. 2020). This methodology of mapping the “right-wing media ecosystem” has been extensively referenced in studies exploring the current media environment (Guess et al. 2021), online propaganda (Guess and Lyons 2020), and disinformation campaigns (Freelon, Marwick, and Kreiss 2020; Freelon and Wells 2020).

In both waves of our panel survey, participants were asked to specify their frequency of news consumption in the previous week using a detailed five-point scale for 16 widely circulated national outlets in the United States, as shown in table 2. The scale includes “Never” (0), “Rarely” (1), “Occasionally” (2), “Fairly often” (3), and “Very often” (4). Based on the 2020 Faris Partisan Report, the partisan election scores of the selected outlets covered the ideological spectrum, with eight on the left (Faris ideology score < 0) and eight on the right (Faris ideology score > 0).

Table 2.

Selection of news outlets and partisan attention score.

Category	Item	Faris partisan score
Left-wing outlet (<−0.5)	MSNBC cable news programs	−0.711
	HuffPost	−0.569
Center-left outlet (−0.5 ∼ −0.25)	NPR	−0.402
	The New York Times	−0.269
	The Washington Post	−0.255
	CNN cable news programs	−0.217
Center outlet (−0.25 ∼ 0.25)	National nightly news on CBS, ABC, or NBC	−0.171
	Politico	−0.171
	USA Today	0.002
	The Hill	0.099
	The Wall Street Journal	0.210
Center-right outlet (0.25 ∼ 0.5)
Right-wing outlet (> 0.5)	Daily Caller	0.924
	Breitbart	0.924
	Rush Limbaugh talk radio	0.972
	One America News Network	0.980
	Fox cable news programs	0.981

Note: The Faris partisan score was based on Faris et al. (2020). A positive Faris partisan score means a right-leaning slant, and a negative score means a left-leaning slant. Cut points of −0.5, −0.25, 0.25, and 0.5 were taken to classify the slant type of the outlets.

To calculate an individual’s news use slant, we multiply the frequency of using each news outlet (0–4) with the partisan attention score of this outlet.

$${\mathit{News\hspace{0.17em}}\mathit{use\hspace{0.17em}}\mathit{slant}}_i={\sum }_{i=1}^n({\mathit{NewsUse}}_i\times {\mathit{NewsSlant}}_i)$$

(1)

where ${\mathit{NewsUse}}_i$is the using frequency of news outlet i, and ${\mathit{NewsSlant}}_i$ is the slant of news outlet i based on the Faris ideology score.

The diversity of news use. In contrast to the slant of news use, diversity of one’s news use refers to individuals’ breadth of consumption, reflecting their likelihood of regularly encountering crosscutting information. We measure an individual’s diversity of news use by judging the degree to which their consumption is distributed across the sixteen outlets. The normalized Shannon’s H Information Entropy formula is applied to measure diversity of news use. Shannon’s H was a variant of the entropy formula of physics to study the diffusion of heat, which was first applied by Shannon (1948) to understand the concentration and diffusion of information. Scholars have since used this formula to study institutional agenda setting (Baumgartner, Jones, and Macleod 2000), policy attention (Jennings et al. 2011), and attention diversity of news organizations (Boydstun, Bevan, and Thomas 2014). Because of Shannon’s H’s ability to capture both the abundance and evenness of news media use with high sensitivity, we apply it to measure news use diversity. In this context, Shannon’s H increases as the frequency of use across news outlets becomes more even (see Boydstun, Bevan, and Thomas 2014). In this study, we use the normalized form of Shannon’s H, which adjusts the measure to range from 0 to 1 for both diversity measures:

$$\text{Normalized Shannon’s H =}\frac{-{\sum }_{i=1}^n\left(p\right(x_i\left)\right)\mathrm{*}ln\left(p\right(x_i\left)\right)}{ln\left(N\right)}\mathrm{ }$$

(2)

where $x_i$ represents the attention an outlet receives, $p\left(x_i\right)$ = the proportion of total attention the outlet receives, $ln\left(p\right(x_i\left)\right)$= the natural log of the proportion of attention the outlet receives, and N = the total number of outlets (N = 16).

As our scale for a single outlet use ranges from 0 to 4, and the natural log of 0 is undefined, we adapt the approach of Boydstun, Bevan, and Thomas (2014) to replace 0 with a tiny proportion (i.e., 0.000001). This replacement should not change the results and general shape of the diversity measure compared with other replacement techniques (Boydstun, Bevan, and Thomas 2014). Diversity scores approaching 0 indicate concentrated use of few news outlets, and diversity scores approaching 1 indicate diverse use.

The extremity of news useis measured relatively by the proportion of news consumption of extreme news outlets out of news consumption options. We define “extreme news” as outlets with a partisan attention score higher than 0.5. In our survey, two left-wing outlets, MSNBC and HuffPost, fall in the highly biased left-wing category, and five right-wing outlets fall in the highly biased right-wing category, namely, Daily Caller, Breitbart, Rush Limbaugh, OAN, and Fox cable news programs. The two extremity measures are calculated: left extremity (M = 0.10, SD = 0.14, range: 0–1 in W1; M = 0.10, SD = 0.14, range: 0–1 in W2) and right extremity (M = 0.21, SD = 0.30, range: 0–1 in W1; M = 0.21, SD = 0.30, range: 0–1 in W2).

To examine how different news use dimensions related to people’s attitudes concerning the 2020 election, we measure people’s electoral confidence in vote counting, belief in a voter fraud conspiracy, and approval for Trump as outcome variables.

Electoral confidence was measured in both W1 and W2 by five items assessing “How confident are you in the following items related to voting in the general election this year? My vote will be (were) counted; All in-person votes will be (were) counted; All mail-in votes will be (were) counted; All in-person absentee votes will be (were) counted; All ballots dropped off at an official ballot box will be (were) counted.” Respondents could indicate “not at all confident” (1), “not very confident” (2), “somewhat confident” (3), or “completely confident” (4) (M = 2.99, SD = 0.77, Cronbach’s $\alpha$  = 0.90 in W1; M = 3.14, SD = 0.90, Cronbach’s $\alpha$  = 0.95 in W2). Belief in voter fraud conspiracy (M = 1.91, SD = 1.15) was measured in W2 using the same stem with a single item: “A conspiracy related to ballot counting in the 2020 election stole the election from President Trump.” Response choices include “not at all confident” (1), “not very confident” (2), “somewhat confident” (3), and “completely confident” (4).

Approval for Trump was measured in both W1 and W2 using two questions: “Overall, do you approve or disapprove of the way Donald Trump is handling his job as president?” (1 = strongly disapprove; 2 = disapprove; 3 = neither approve nor disapprove; 4 = approve; 5 = strongly approve; 8 = don’t know), and “You will now be shown a list of people. Please indicate if you have a favorable or unfavorable opinion of each of them or if you haven’t heard enough about them yet to have an opinion: [Donald Trump] (1 = very unfavorable; 2 = somewhat unfavorable; 3 = neither favorable nor unfavorable; 4 = somewhat favorable; 5 = very favorable; 8 = haven’t heard enough).” Respondents who selected “don’t know” in the approval question and “haven’t heard enough” in the favorability question were excluded from the analyses. The answers to these two questions were averaged to create an indicator for approval of Trump (W1: M = 2.49, SD = 1.65, r = .94; W2: M = 2.45, SD = 1.64, r = .94).

Controls

We included demographic variables such as gender, race, age, education, and income as control variables in our cross-sectional models. The distribution of the demographic variables is presented in Supplementary Material I, table S1.

Analytical Strategies

In our two-part analysis, we first examine news media use patterns by partisan groups. Specifically, we map the differences in news use in terms of slant, diversity, and extremity across three partisan groups: Democrats, including leaners (n = 853), Republicans, including leaners (n = 733), and Independents (n = 190). Whether the differences between Democrats, Independents, and Republicans are statistically significant is tested using ANOVAs with post hoc Tukey HSD.

Next, we examine the relationship between patterns of news use and political beliefs during the 2020 election using both cross-sectional and panel models. We rely on the post-election survey for cross-sectional analyses and use linear regressions to establish the association between news use after the election, beliefs about the election’s legitimacy, and the Trump evaluation. Using panel data, we then apply concurrent residual models to explore the association between aggregate changes in news use patterns and political beliefs over time, regressing each subject’s news use patterns and political outcome variables (i.e., electoral confidence and approval for Trump) in W2 with their counterpart in W1. We get the residualized variables by removing from the W2 value the portion that can be predicted using OLS regression by the W1 value. Then we predict the residualized dependent variables with the residualized news use patterns, controlling for demographics.

In order to construct models for hypothesis testing and mitigate potential multicollinearity, we examine the correlation between news use variables in both the pre-election and post-election surveys, as well as the residualized estimators. Table 3 shows that the correlation coefficient between news diversity and news slant is low (r = −0.07 post-election), indicating it is safe to include these two variables in the same model. However, right extremity is highly correlated with news use slant (r = 0.75 post-election), leading us to exclude right extremity from the cross-sectional models to address multicollinearity concerns.

Table 3.

Correlation matrix between news use variables.

	Slant	Diversity	Right extremity	Left extremity
Pre-election value
Slant	1
Diversity	−0.07	1
Right extremity	0.75	−0.07	1
Left extremity	−0.41	0.26	−0.29	1
Post-election value
Slant	1
Diversity	−0.07	1
Right extremity	0.75	−0.06	1
Left extremity	−0.4	0.25	−0.3	1
Residualized value
Slant	1
Diversity	0.11	1
Right extremity	0.48	−0.01	1
Left extremity	−0.21	0.2	−0.11	1

Note: This table summarizes the correlations between news use measures in the pre-election survey, post-election survey, and the residualized values calculated from the pre- and postsurvey data.

Consistent with literature discussing the asymmetric pattern of news use extremity, post-election left extremity has a low correlation with both news use slant (r = −0.40) and news use diversity (r = 0.25). Including left extremity in the cross-sectional model with slant and diversity provides insights into how left extremity may or may not influence the debunking of voter fraud claims. Furthermore, the panel models allow us to distinguish and compare the unique effects of news use slant, diversity, and right and left extremity. To do so, we include the residualized values of all these variables in the same panel model, given the low correlation between the residualized news use variables. All the regression analyses were unweighted.

Results

Asymmetric News Use Across Partisan Groups

To investigate whether news use patterns differ between partisan groups and pure Independents, we conducted a one-way ANOVA to compare the group means of news slant, diversity, and extremity between Democrats, Independents, and Republicans (table 4) with the distribution of each news use variable by partisan group illustrated in figure 1.

Figure 1.

Density plots of news use patterns by partisan group.

Table 4.

Descriptive statistics and ANOVA results of news use patterns by partisan groups.

	Republican		Independent		Democrat		P	Post hoc Tukey’s test
	M	SD	M	SD	M	SD
Slant	0.62	0.93	−0.01	0.69	−0.53	0.78	<0.001	Dem < Ind < Rep
Diversity	0.39	0.30	0.38	0.32	0.51	0.31	<0.001	Ind, Rep < Dem
Right extremity	0.40	0.35	0.19	0.29	0.06	0.11	<0.001	Dem < Ind < Rep
Left extremity	0.05	0.09	0.08	0.11	0.15	0.16	<0.001	Rep < Ind < Dem

Note: The slant measure was standardized. The mean values and standard deviations are shown for Republicans, Independents, and Democrats for each news use pattern variable, as well as the results of ANOVA tests and the posthoc Tukey’s test comparing the parameter estimates between the three groups.

The ANOVA results reveal significant differences in every dimension of news use tested among the three groups. Unsurprisingly, the post hoc Tukey HSD test shows that Republicans’ news slant leans toward the right (M = 0.62, SD = 0.93, standardized), while Democrats have a left-leaning news diet (M = −0.53, SD = 0.78, standardized; Mean Diff = −1.15, p < 0.001). Independents maintain a near-neutral news diet (M = −0.01, SD = 0.69, standardized), leaning more toward the right than Democrats (Mean Diff = 0.53, p < 0.001), and more toward the left compared with Republicans (Mean Diff = –0.63, p < 0.001).

To examine asymmetric news use polarization, we compared Republicans and Democrats on (1) the proportion of partisans whose news use slant is consistent with their partisanship; and (2) the absolute value of news use slant for the consistent partisan news users. A larger proportion of Democrats (82.5 percent) have a news slant more consistent with their partisan preference than Republicans (69.4 percent). However, among those whose news slant is consistent with their partisanship, the degree of “rightness” for Republicans (M = 0.98, SD = 0.90) is significantly higher than the degree of “leftness” for Democrats (M = 0.77, SD = 0.59, t = 4.95, p < 0.001). This finding partly supports H1a.

We also find significant differences in news use diversity among partisan groups. Democrats report significantly higher news diversity (M = 0.51, SD = 0.31) than both Republicans (M = 0.39, SD = 0.30, Mean Diff = 0.13, p < 0.001) and Independents (M = 0.38, SD = 0.32, Mean Diff = 0.13, p < 0.001). However, there are no significant differences between the mean news use diversity of Republicans and Independents (p = 0.996). This supports H1b, that Republicans’ news habits are more insulated than those of Democrats (figure 1b).

Differences are also found between the three groups for the right extremity and left extremity. The proportion of highly biased right-wing news use is significantly higher among Republicans (M = 0.40, SD = 0.35) than among Democrats (M = 0.06, SD = 0.11; Mean Diff = 0.34, p < 0.001). On the other hand, Democrats consume significantly more highly biased left-wing news (M = 0.15, SD = 0.16) than Republicans (M = 0.05, SD = 0.09; Mean Diff = 0.11, p < 0.001). Independents, however, consume a much lower proportion of highly biased, but ideologically consistent, news than the partisans (right extremity: M = 0.19, SD = 0.29; Mean Diff = −0.21, p < 0.001; left extremity: M = 0.08, SD = 0.11; Mean Diff = −0.07, p < 0.001), although slightly higher than the counter-partisans (right extremity: Mean Diff = 0.13, p < 0.001; left extremity: Mean Diff = 0.03, p = 0.008). The difference in right extremity between Republicans and Democrats is much larger than the left extremity, supporting H2c. As depicted in figures 1c and 1d, most Democrats and Republicans consume minimal extreme media content that counters their partisanship. However, when compared to Independents, both Democrats and Republicans consume a relatively higher amount of extreme news outlets that align with their partisanship. This trend is more pronounced on the right side, thus supporting Hypothesis H1c.¹

Association between News Use Patterns and Post-election Attitudes

The correlation between news use patterns and election-related attitudes was evident during the campaign (see Supplementary Material II, table S3) and persisted post-election. Table 5 presents cross-sectional multivariate regression results analyzing post-election beliefs and judgments in relation to news consumption patterns, accounting for demographics.²

Table 5.

Cross-sectional models of post-election news use patterns and post-election beliefs.

	Dependent variable:
	Electoral confidence w2		Approval for Trump w2		Voter fraud belief w2
	coefficient	p value	coefficient	p value	coefficient	p value
News use slant w2	−0.503	<0.001	0.602	<0.001	0.596	<0.001
	(−0.545, −0.460)		(0.564, 0.639)		(0.557, 0.635)
News use diversity w2	0.077	<0.001	−0.052	0.012	0.006	0.763
	(0.031, 0.122)		(−0.092, −0.011)		(−0.035, 0.048)
Left extremity w2	−0.036	0.115	−0.063	0.002	−0.025	0.235
	(−0.081, 0.009)		(−0.102, −0.023)		(−0.066, 0.016)
Age	0.004	0.014	0.002	0.104	−0.002	0.070
	(0.001, 0.006)		(−0.0004, 0.005)		(−0.005, 0.0002)
Female	−0.083	0.052	−0.045	0.232	−0.037	0.345
	(−0.167, 0.001)		(−0.119, 0.029)		(−0.113, 0.040)
Black	0.064	0.550	−0.215	0.023	−0.195	0.047
	(−0.146, 0.275)		(−0.401, −0.030)		(−0.387, −0.003)
Hispanic	0.211	0.147	−0.195	0.128	−0.092	0.490
	(−0.074, 0.497)		(−0.447, 0.056)		(−0.353, 0.169)
Other	0.096	0.315	−0.260	0.003	−0.118	0.178
	(−0.092, 0.285)		(−0.429, −0.092)		(−0.290, 0.054)
Education	0.054	0.001	−0.043	0.003	−0.051	<0.001
	(0.022, 0.086)		(−0.072, −0.015)		(−0.080, −0.022)
Income	0.027	0.062	0.021	0.097	−0.002	0.853
	(−0.001, 0.055)		(−0.004, 0.046)		(−0.028, 0.023)
Constant	−0.401	<0.001	−0.006	0.950	0.324	0.002
	(−0.625, −0.177)		(−0.206, 0.193)		(0.120, 0.529)
Observations	1,672		1,646		1,672
R2	0.306		0.469		0.429
Adjusted R2	0.302		0.466		0.426
Residual std. error	0.829 (df = 1661)		0.730 (df = 1635)		0.757 (df = 1661)

Note: The table summarizes the results from the cross-sectional multiple regression models using Wave 2 (post-election) data. The models examine the effects of news use after the election on beliefs and attitudes about the election and Trump. All variables are standardized to compare beta coefficients. Reference groups for gender and race are Male and White, respectively. Parentheses contain 95% confidence intervals.

The slant of people’s news diet, post-election, significantly correlates with confidence in the election’s legitimacy. Those with a more right-leaning news slant are less likely to believe votes were properly counted (b = −0.50, p < 0.001). These individuals are more likely to harbor stronger beliefs regarding voter fraud (b = 0.60, p < 0.001) and have a more positive post-election evaluation of Trump (b = 0.60, p < 0.001). This supports Hypotheses H2a–H2c. Left-leaning extremity, however, does not significantly correlate with beliefs about election results, despite its association with lower Trump support (b = 0.06, p = 0.002).

Notably, diversity of news use, accounting for news use slant, correlates with increased confidence in vote counting. Those engaging with more diverse news sources express stronger belief in vote count accuracy compared to those consuming a more monolithic assortment of news, irrespective of the ideological slant of news use (b = 0.08, p < 0.001). Moreover, news use diversity is significantly negatively correlated with approval for Trump (b = −0.05, p = 0.012). These findings support H4a and H4c, but not H4b. Diversity in news consumption was not related to voter fraud beliefs, suggesting limits to the buffering of news diversity.

This could be due to voter fraud beliefs being mainly spread within the right-leaning ecosystem. The impact of news use diversity may differ by ideological slant of one’s news diet, with average effects canceled out. To validate this, we performed interaction models between media use slant and diversity as a robustness check. As shown in table 6, a more diverse news use especially benefits right-leaning news consumers by moderating negative effects of right-leaning news use slant on undermining electoral confidence (b = 0.31, p < 0.001) and promoting voter fraud belief (b = −0.27, p < 0.001).

Table 6.

The interaction effects between news use slant and diversity.

	Dependent variable:
	Electoral confidence w2		Voter fraud belief w2
	coefficient	p value	coefficient	p value
News use slant w2	−0.720	<0.001	0.787	<0.001
	(−0.776, −0.664)		(0.735, 0.838)
News use diversity w2	0.071	0.002	0.012	0.562
	(0.027, 0.114)		(−0.028, 0.052)
Left extremity w2	−0.083	<0.001	0.016	0.432
	(−0.127, −0.039)		(−0.024, 0.057)
Age	0.006	<0.001	−0.005	<0.001
	(0.003, 0.009)		(−0.007, −0.002)
Female	−0.106	0.010	−0.016	0.665
	(−0.187, −0.025)		(−0.090, 0.058)
Black	0.017	0.869	−0.154	0.105
	(−0.186, 0.220)		(−0.340, 0.032)
Hispanic	0.189	0.180	−0.072	0.577
	(−0.087, 0.464)		(−0.324, 0.180)
Other	0.067	0.471	−0.092	0.278
	(−0.115, 0.248)		(−0.258, 0.074)
Education	0.055	<0.001	−0.052	<0.001
	(0.024, 0.086)		(−0.080, −0.023)
Income	0.027	0.052	−0.002	0.843
	(−0.0002, 0.054)		(−0.027, 0.022)
News use slant w2 x news use diversity w2	0.308	<0.001	−0.271	<0.001
	(0.254, 0.362)		(−0.320, −0.221)
Constant	−0.501	<0.001	0.413	<0.001
	(−0.719, −0.284)		(0.214, 0.612)
Observations	1,672		1,672
R2	0.354		0.466
Adjusted R2	0.350		0.462
Residual std. error	0.800 (df = 1660)		0.733 (df = 1660)

Note: The table summarizes the results from the cross-sectional multiple regression models using Wave 2 (post-election) data. The models examine the effects of news use after the election on beliefs and attitudes about the election and Trump. All variables are standardized to compare beta coefficients. Reference groups for gender and race are Male and White, respectively. Parentheses contain 95% confidence intervals.

Change of Evaluation on Trump and News Use Patterns

We conduct parallel testing of concurrent residual models, summarized in table 7, to examine how the shifts in news consumption patterns during the election related to the change in electoral confidence and the assessment of Trump observed between survey waves.

Table 7.

Panel models reflecting the effects of change of news use patterns on change of judgment about the election and Trump.

	Dependent variable:
	Approval for Trumpres		Electoral confidenceres		Electoral confidenceres
	coefficient	p value	coefficient	p value	coefficient	p value
News use slantres	0.054	0.056	−0.072	0.007	−0.093	<0.001
	(−0.001, 0.109)		(−0.125, −0.020)		(−0.147, −0.040)
News use diversityres	−0.024	0.361	0.002	0.924	−0.001	0.983
	(−0.077, 0.028)		(−0.047, 0.052)		(−0.050, 0.049)
Right extremityres	0.107	<0.001	−0.136	<0.001	−0.106	<0.001
	(0.050, 0.164)		(−0.190, −0.083)		(−0.161, −0.050)
Left extremityres	−0.034	0.191	0.084	<0.001	0.086	<0.001
	(−0.085, 0.017)		(0.035, 0.134)		(0.036, 0.135)
Right extremityres x News use diversityres					0.061	<0.001
					(0.030, 0.093)
Constant	0.187	0.094	1.096	0.368	0.095	0.370
	(−0.032, 0.405)		(−0.113, 0.306)		(−0.113, 0.304)
Observations	1,561		1,614		1,561
R2	0.027		0.055		0.030
Adjusted R2	0.021		0.050		0.024
Residual std. error	0.983 (df = 1551)		0.959 (df = 1604)		0.981 (df = 1550)

Note: The dependent and independent variables are residualized after regressing W2 variables on W1 counterparts. Independent and dependent variables are standardized in each model to compare beta coefficients. Demographics were controlled in the models. Parentheses contain 95% confidence intervals.

The panel models mirror the cross-sectional models. As predicted, electoral confidence declined more among those shifting rightward post-election (b = −0.07, p = 0.01), supporting H2d. Additionally, increased consumption of extreme-right-wing news outlets correlated with decreased confidence in vote count (b = −0.14, p < 0.001). Conversely, a rise in the consumption of extreme-left-wing news resulted in an improvement in post-election confidence (b = 0.08, p < 0.001), supporting H3d.

Approval for Trump, however, was influenced minimally by shifts in news consumption shifts. Only those intensifying extreme-right-wing news consumption showed increased Trump approval (b = 0.11, p < 0.001).

In conclusion, the anticipated asymmetric impact of deeper immersion in extreme-left- versus extreme-right-wing news outlets was corroborated by the panel analyses, with compelling evidence of shifts toward far-right news consumption correlating with declines in electoral confidence and escalated support for Trump. It is noteworthy that the changing news use diversity was not a significant predictor. However, the interaction model between residualized right extremity and residualized diversity reveals that an increase in news diversity can reduce the effects of extreme-right-wing news on undermining electoral confidence.

Discussion

This study examined the relationship between the evolving partisan media ecosystem, asymmetric polarization, and news use patterns in the United States, finding that Republicans’ news usage is more ideologically biased compared to that of Democrats. Republicans typically engage with a smaller set of right-leaning sources, while Democrats’ news consumption is characteristically more diverse, adding support to the claim of asymmetric polarization between the left and the right (Knobloch‐Westerwick and Meng 2011; Thorson and Wells 2016).

Caught in a media-audience reinforcement loop and bolstered by the networked public sphere, individuals often choose information that corroborates their biases (Southwell and Thorson 2015) and disregard news that is inconsistent with their political inclinations (Southwell and Thorson 2015). Our conception and operationalization of multiple means of understanding the “shape of news use” reveals that partisans forge distinct consumption patterns, differing in ideological slant, extremity, and diversity. Partisan news outlets within this ecosystem form unique yet intersecting networks, some of which perpetuate inaccurate political narratives (Faris et al. 2020), contributing to a more polarized information landscape.

As partisan media, particularly those on the ideological right, deviate from established journalistic standards of balance and factuality, individuals’ news use patterns become crucial for understanding who believes what about US democracy and elections. A biased slant of news use, greater extreme news consumption, and limited media diversity can foster misunderstandings about democratic procedures, especially on the political right.

Our findings show that Republicans (relative to Democrats) not only consume information aligned with their political beliefs in terms of greater slant, less diversity, and more extremity, but that their news slant was significantly associated with post-election attitudes questioning election administration, elevating the politician advocating these claims, Donald Trump, and acceptance of voter fraud beliefs. This was observed when accounting for the diversity of news use, which was linked with electoral confidence and negatively related to Trump approval. Extreme-left-wing news was associated with disapproval of Trump.

More importantly, the change models we tested affirmed the role of right-wing news use, especially the increasing use of extreme-right-wing news, in a corresponding decline in electoral confidence and rising approval for Trump. It is now widely known that leading voices in right-wing media orchestrated a campaign of misinformation in the wake of the 2020 election, knowing that claims of ballot fraud and a stolen election were untrue yet still repeating Trump’s lies (Peters 2022). This study also highlights the importance of crosscutting news consumption by showing the moderating effect of news use diversity.

Another approach to questions of news bias, slant, and extremity focuses on hostile media perceptions (Vallone, Ross, and Lepper 1985), which suggests that bias in news is a subjective matter (Lichter 2017). Our empirical results do not refute this argument, as the relative hostile media perception (Gunther et al. 2001) posits that partisans view biased news asymmetrically based on group membership. Further research is needed to understand how bias, slant, and extremity are embedded in the news ecology and rooted in media perceptions. Future research should also build on this study’s effort to gauge the “shape of news” consumption in a way that moves scholarship beyond basic measures of news polarization or media repertoires. Linking asymmetrically polarized news use patterns with post-election beliefs about the voter fraud conspiracy illustrates the value of untangling slant, diversity, and extremity of news use.

Footnotes

1

As a cross-validation, we also considered the slant of individuals’ most frequently used outlets as an alternative measure for the slant of news use, and the diversity across different ideological types of news media as an alternative for the diversity of news use. The ANOVA comparisons of news use patterns of these alternative variables by partisanship can be found in Supplementary Material I, table S2.

2

A brief discussion on demographic factors influencing electoral judgments and confidence can be found in Supplementary Material II, section 2.

Supplementary Material

Supplementary Material may be found in the online version of this article: https://doi.org/10.1093/poq/nfae031.

Funding

This work was supported by the John S. and James L. Knight Foundation [G-2019-58809 to M.W.].

Acknowledgements

We extend our deep thanks to Mingcong Pan for his constructive and inspiring insights and suggestions. His feedback has been greatly appreciated throughout the development and writing process of this paper. Furthermore, we are thankful to the reviewers of this paper, whose comments and suggestions have helped us improve the quality of our work.

Data Availability

Replication data and documentation are available at: https://doi.org/10.7910/DVN/51X6HW.