Selective Exposure and the Authoritarian Dynamic: Evidence From Canada and the United States

This study explores to what extent selective exposure to political messages can produce political (in)tolerance among authoritarians and non-authoritarians. Drawing on a selection-exposure experiment embedded within an online survey conducted in the United States (N = 1978) and Canada (N = 1673), we explore how authoritarians and non-authoritarians react to framing around civil liberties controversies. Participants were randomly assigned to receive a message about a controversial group. In the forced-choice condition, participants were randomly assigned a political or non-political message. In a second condition, participants were given a choice of which message to read more about. The results show that authoritarians who are politically knowledgeable generally avoid messages that promote free speech by consuming non-political information. While messages about the dangers of free speech have the potential to produce more intolerance among authoritarians, we found that this effect was limited to those who are the least likely to consume them when given a choice. By contrast, we found that messages about the risk posed by free speech produced intolerance among non-authoritarians for whom threat-related cognitions were already chronically accessible. The effects of pro-civil liberties messages were limited to unthreatened non-authoritarians. Hence, we conclude that in the contemporary information environment selective exposure can increase polarization around a civil liberties controversy by producing attitude change but this occurs mainly among non-authoritarians.

in the comparative literature (Van Aelst et al., 2017), several important laboratory studies conducted in the United States link selective exposure to authoritarianism (Lavine, Lodge, & Freitas, 2005;Lavine, Lodge, Polichak, & Taber, 2002). The results of these experiments appear to confirm prior claims about the importance of "closedmindedness" in the authoritarian worldview (Altemeyer, 1981(Altemeyer, , 1988. Nonetheless, several key questions about selective exposure and authoritarianism remain: Do authoritarians seek out and react more strongly to messages that focus on the "risks" of tolerating groups that would appear to violate conventional social norms and disrupt their communities? Do they avoid messages that argue in favor of "putting up with" non-conformist groups? Are there limits to the effects of selective exposure? How reactive are non-authoritarians to such messages under conditions of choice? The aim of the present study is to examine when and to what extent selective exposure to political messages conditions the relationship of authoritarianism to political tolerance (for an earlier version see Hinckley, 2015). We test 1) the selective exposure hypothesis that authoritarians will be more likely to consume messages about the threat posed by disliked outgroups and 2) the competing hypotheses suggested by the political communications literature that authoritarians will react more (or less) strongly than non-authoritarians after exposure.
Our experimental design is the first to combine the least-liked approach to measuring political tolerance (Sullivan, Piereson, & Marcus, 1982) with a between-subjects, selective exposure design (Arceneaux & Johnson, 2013;Gaines & Kuklinski, 2011) that allows for comparisons between a choice group and a forced exposure group. A selective exposure design provides an ideal test because it mimics contemporary information environments by allowing participants to select among competing messages and an irrelevant entertainment option. We also incorporate elements of the participant preference design (Arceneaux & Johnson, 2013) that requires asking participants in a forced exposure group about their message preference before they read and react to any treatment, thereby allowing us to estimate the effects of the political messages among those who would normally prefer to read or to avoid such content. Furthermore, we extend the existing experimental literature by using online national samples from the United States and Canada. In the next section, we develop our theoretical expectations by examining recent research on the authoritarian dynamic as well as scholarship in the field of political communication.

Models of the Authoritarian Dynamic
Scholarly conceptions of authoritarianism -as a property of the individual rather than a feature of political systems -have evolved significantly over time (e.g., Adorno et al., 1950;Altemeyer, 1988). More recently, Feldman (2003) has argued that authoritarianism is best understood as a core value dimension defined by preferences for social conformity over individual autonomy.
Building on this conception, Stenner articulated an activation model of authoritarianism (Feldman & Stenner, 1997;Stenner, 2005). The key to this model is that exposure to a normative threat activates the authoritarian's need for "oneness and sameness" in society (Stenner, 2005, p. 17) that would otherwise lie dormant. When a normative threat is absent, authoritarians and non-authoritarians are expected to have similar attitudes. As the activation of the authoritarian disposition is an episodic feature of society, authoritarian intolerance is expected to rise and fall with changing levels of threat to social cohesion. Hetherington and Weiler (2009) posit an alternative dynamic to the activation model. While acknowledging the importance of threat to the authoritarian worldview, they argue that authoritarians persistently feel threatened (but also see Feldman et al., 2010). Hence, on issues relevant to their worldview (e.g., gay rights), authoritarians are chronically prone to intolerance. Moreover, and in direct contrast to Stenner (2005), Hetherington and Weiler (2009) argue it is the non-authoritarians that become more like authoritarians under conditions of high threat. In this view, authoritarian intolerance is relatively static, while a rising tide of intolerance in society results from a new threat perceived by non-authoritarians.
We believe that the distinction suggested by this literature between situational threats that are largely exogenous to the individual, as opposed to the chronic accessibility of threat-related cognitions, is worthy of further investigation.
If the information environment contains messages about the public activities of a disliked group, then authoritarians should exhibit an attitude shift that fits the activation model, especially when such a group is framed as posing a threat to social order. However, to the extent that threat is already a chronically accessible consideration (Zaller, 1992), exogenous messages may have little additional impact on their level of political tolerance. In the next section we consider how the contemporary information environment may shape opportunities for consuming and reacting to messages about disliked political groups.

Authoritarianism and Selective Exposure
Research on media fragmentation and public opinion suggests that authoritarians have many new opportunities to activate their latent intolerance. In the era when newspapers, radio and TV dominated, mass audiences received a relatively balanced political perspective that varied little across news sources (Prior, 2007). New media technologies have made consuming political information easier, while also tracking an increase in opinionated news on cable television and the internet (see Prior, 2013 for a review). Such a degree of choice also raises the possibility that selective exposure to information, that is, seeking out and consuming messages that fit our relevant dispositions, is on the rise (Bimber & Davis, 2003;Garrett, 2009;Garrett, Carnahan, & Lynch, 2013;Iyengar & Hahn, 2009;Stroud, 2008Stroud, , 2011Valentino et al., 2009).
Nonetheless, some evidence suggests that selective exposure involves consuming agreeable viewpoints but not necessarily avoiding balanced messages that pair agreeable and disagreeable views (Garrett, Carnahan, & Lynch, 2013). Hence, in this new information environment, authoritarians may be moving towards chronic intolerance because they can readily consume messages in line with their predisposition about the "risks" of disliked political groups or they may avoid pro-civil liberties views (or they may do both), to the extent that they possess the relevant political knowledge to make such choices (Zaller, 1992). At the same time, non-authoritarians have more opportunities to seek out and consume pro-civil liberties viewpoints.
The new information environment also provides for a second type of selective exposure. An increasingly fragmented information environment offers more opportunities to "tune out" by consuming non-political or entertainment news (Arceneaux & Johnson, 2013;Bennett & Iyengar, 2008Prior, 2007). Indeed, viewership of ideologically slanted news has declined in the USA as non-political options have become increasingly available (Prior, 2007).
It is therefore plausible that authoritarians will select entertainment news as a means of avoiding disagreeable viewpoints when given the choice. Moreover, it raises the possibility that those who choose to consume political messages will be affected differently than those who tune out (Gaines & Kuklinski, 2011). Selective exposure can increase the impact of political messages on attitudes because individuals are especially open to accepting those messages; however, it is also possible that those who select and consume well known viewpoints may be affected less than those who would normally avoid such messages (Arceneaux & Johnson, 2013).
When it comes to authoritarians, this means that we expect them to be more likely to seek out messages about the "risk" posed by disliked outgroups because this reflects their worldview about the dangers of social non-conformity and to avoid consuming pro-civil liberties messages. In sum, a fragmented information environment may serve to provoke authoritarian intolerance while also giving authoritarians the opportunity to avoid pro-civil liberties messages.
Nonetheless, we must also emphasize that the effects of selective exposure among authoritarians may itself be contingent. Zaller's (1992) work suggests that those for whom some considerations are chronically accessible, such as for those who know a lot about and regularly consume news, should be less influenced by new political information or messages. That means that for authoritarians for whom threat-related cognitions are already accessible, they should be less reactive to messages about the threats posed by disliked groups. And this should be particularly true for those who regularly consume news and understand politics because they are more regularly exposed to political messages. To the extent, then, that authoritarians in our studies have been previously and extensively exposed to messages of intolerance in the real world, particularly in the context of the 2016 US presidential campaign (Jamieson & Taussig, 2017;Lamont, Park, & Ayala-Hurtado, 2017), they may also be less reactive to the stimuli presented in our experiment.

Country Experiment Participants
The data for the study were collected from an online survey-based experiment conducted in the United States (N = 1978) and Canada (N = 1673). i The two countries share similar histories as former British colonies that have significant levels of ethnic diversity due to immigration as well as the presence of historic minorities. Free speech norms are enshrined in both countries, though unlike the US, Canada has both civil and criminal penalties for hate speech similar to many European countries. We have no reason to expect that the relationship between authoritarianism and free speech attitudes will differ. Nonetheless, as we have already suggested, prior exposure to intolerant messages may be higher in the USA than in Canada.
The study was in the field during a four-week period in January and February 2017. In Canada, the questionnaire was available in both English and French. Qualtrics, an online survey research firm, administered the data collection.
Respondents were selected from among those who had registered to participate in online surveys through several different organizations. The sample providers generally offered the equivalent of $1 US to participate. The average time to complete the survey was 22 minutes.
A quota system based on age, gender, and education was used to screen potential respondents, which resulted in samples that largely reflect these population parameters in each country as reported by the most recent census.
In addition, a language quota was applied in Canada. The final US sample was 52% female, 75% white, with a median educational level of "associates degree" and a median age of "30-39". The Canadian sample was 52% female, 81% white, and had the same median levels of education and age as the US sample. Among Canadians, about 65% reported English as their primary language, 30% selected French, and 5% indicated "other."

Procedure and Measures
Participants first completed a variety of trait measures including the widely used "child rearing scale" to measure authoritarianism (e.g., Hetherington & Weiler, 2009) and then selected a least-liked group (Sullivan et al., 1982).
To measure authoritarianism, we generally followed the American National Election Study in asking about child rearing values. Authoritarianism was signaled by agreeing that children should be obedient or that children need strict discipline and disagreeing with the idea that children should think for themselves or follow their own feelings rather than the rules. Responses to each statement were indicated on a four-point scale that ran from agree strongly to disagree strongly. An additive index was created and then fully standardized (M = 0, SD = 1). The authoritarianism scale showed some evidence of having two dimensions, with the resulting low value of Cronbach's alpha (.40), but which is similar to that reported using ANES data (e.g., Pérez & Hetherington, 2014). Nonetheless, the scale works as predicted in our regression models.
ii To reproduce the least-liked approach to measuring political tolerance (Sullivan et al., 1982), participants were asked to rate six controversial groups (radical Muslims, feminists, neo-Nazis, gay rights activists, Christian fundamentalists, and far-right activists) on an 11-point scale that ran from, "0=you dislike the group very much," to "10=you like the group very much" and then select their most disliked group among the six. iii In both Canada and the US, radical Muslims and neo-Nazis were the most common least-liked group, each capturing between around 40% of the sample in each country.
Next, participants were randomly assigned to either a forced exposure condition or a choice condition. In the forced exposure group, participants were assigned to read one of three political messages. The forced control group read a non-political news story. In the choice group, participants could select to read one of the three political messages or the control article. In the choice group, the list of political messages was randomized across participants.
In order to examine why selective exposure may condition the effects of political messages on tolerance attitudes, we also incorporated elements of the "participant preference design" described by Arceneaux and Johnson (2013, p. 62) This design requires asking participants in the forced exposure group about their message preference before they read and react to any treatment. By separating message preference and selection, we avoid confounding the effects of the message with selection based on prior attitudes. Moreover, it allows us to estimate the effects of the political messages among those who would normally prefer to read such content and those who would typically avoid it. If message effects are limited to those with a preference to consume them, then we can conclude that selection increased the effect. Conversely, if effects are limited to those who would prefer to read something else, this would indicate the opposite: those who prefer a particular message probably have, through prior exposure, already internalized the relevant ideas (Arceneaux & Johnson, 2013). To measure message preference in our study, all participants were asked, "Here is a title of a commentary that appeared recently online. How interested are you in reading it?" In the forced choice group, only the title for the treatment message was displayed. In the choice version, all four titles were displayed in random order.
Three political messages were embedded in "online news commentaries" that were initially identified by their titles, and which referred to a hypothetical group called the "United Front." (See Section 1 of the Appendix in the Supplementary Materials for the full wording.) The United Front was described as a new local organization with the same composition (e.g., neo-Nazis, feminists, etc.) as their least-liked group. The message that promoted intolerance was titled, "Local United Front Group: Consider the Risks for Our Community," and emphasized that such extremist groups present a risk to the community's safety and social cohesion. The pro-tolerance message was called, "United Front Group in Our Community: They Deserve Free Speech Rights, Too," and argued that failure to respect the free speech rights of controversial groups will compromise the liberties of all. The balanced message contained both arguments and was titled, "Local United Front Group: Weighing the Pros and Cons for Our Community." The title of the control article was, "The Most Inspirational Moments from the 2016 Rio Olympics." After exposure to the treatment, all participants rated their support for the civil liberties of that same fictitious group (the "United Front") purportedly comprised of people who were members of their least-liked group. Participants were asked to rate their agreement or disagreement with each of the following statements on a 4-point scale, where the fictional United Front was described as their least-liked group: "The United Front should be banned from holding meetings in my community," "Members of the United Front should be able to make public speeches in my community," and "The United Front should be allowed to hold rallies and demonstrations in my community." The responses to these three items were averaged to create an index of political tolerance that ranged from 1 to 4 (M = 2.4, SD = .82, α = .79).
Later in the survey all participants completed a word completion task designed to measure the chronic accessibility of threat and also answered two questions designed to measure political knowledge. Our chronic threat accessibility measure reproduced the death thought accessibility (DTA) protocol used in the terror management theory literature (see Hayes, Schimel, Arndt, & Faucher, 2010 for a review). Respondents viewed six word fragments (e.g., Sk _ _ l) and were asked to fill in the blank spaces based on the first word that came to mind (e.g., Skill or Skull). iv Each word fragment that was completed with a death-related word was scored as a 1, while any other word was scored a 0, creating a DTA scale (0-6, M = 1.8, SD = 1.2; α = .41) that was then fully standardized (M = 0, SD = 1) to help with the interpretation of the interaction terms v . Since the DTA measure occurred after the experiment to avoid the possibility of a priming effect, we also confirmed that the treatments did not have an effect on the level of DTA, F(7, 3303) = 1.46, p = .178.
We measured political knowledge using a scale that reflected if respondents could correctly identify Beverley McLachlin as the Chief Justice of Canada and knowing that the Conservative Party is the official opposition. Respondents in the US sample had to identify Ruth Bader Ginsburg as a Supreme Court Justice and the Republican Party as holding a majority in the House of Representatives. The sum of the number of correct answers was used in our regression models vi .

Information Exposure Preferences
Among the participants in the choice condition, the non-political news article was the most commonly selected at 42.7%, while 24.5% selected the risk message, 22.9% selected the balanced message, and 9.9% selected the free speech message. This overall preference for non-political news (which varied little between the USA and Canada), especially over a balanced viewpoint, reinforces the importance of creating selective exposure designs with a non-political information option when estimating media effects (see Feldman, Stroud, Bimber & Wojcieszak, 2013, for a discussion of design choices).
Next, we tested the possibility that authoritarians are more likely to select and read the "risk" message than nonauthoritarians. Given the conditional relationship between political knowledge and individual dispositions posited in public opinion (Zaller, 1992) and selective exposure literatures (e.g., Arceneaux & Johnson, 2013), we also examine if political knowledge moderates the effect of authoritarianism by including an interaction term. The reference category is selection of the control article. Addition of the predictors to a model that contained only the intercept significantly The results show that authoritarians did not select the risk message at higher rates than non-authoritarians but, instead, avoided the free speech message. In the US, the model coefficients indicate that authoritarians were less likely than non-authoritarians to select the free speech message than the control, and politically knowledgeable authoritarians were less likely than others to select the balanced message over the control. In Canada, politically knowledgeable authoritarians are less likely to select the free speech message than the control article, while authoritarians are less likely to select the balanced message than the control article regardless of political knowledge. Importantly, authoritarianism is not related to the probability of selecting the risk message at any level of political knowledge in either sample.
To aid in the interpretation of the parameters, Figure 1 illustrates the predicted probabilities of selecting each message and the control article by level of authoritarianism. To simplify the presentation, Figure 1 is limited to those with high political knowledge. In both countries the results are quite clear: politically knowledgeable authoritarians were about twice as likely as knowledgeable non-authoritarians to select the control article. In sum, this finding suggests that authoritarians are more motivated to avoid disagreeable viewpoints than in seeking out agreeable (intolerant) views.

Effects of Message Exposure on Political Tolerance
The results of regressing the political tolerance scale on the various experimental conditions, authoritarianism, DTA, and country are reported in Table 2. These results are also illustrated in Figures 2, 3 and 4. We follow past practice in estimating effects by comparing participants who were forced to read one of the political messages, or were able to select one message, against the forced control group (Arceneaux & Johnson, 2013;Gaines & Kuklinski, 2011). This approach has the advantage of measuring the difference between those selecting a message or those in the forced treatment group against the control condition while holding baseline traits constant across the three groups through random assignment (Gaines & Kuklinski, 2011). The second model includes continuous by categorical interaction terms to capture the different effects of information exposure at different levels of authoritarianism, DTA and by country. All models also include controls for affective response ratings of the least-liked group, as some selected groups were more disliked on average than others.
The regression models and illustrations presented below reveal several important findings. First, we tested the possibility that high DTA authoritarians would be less politically tolerant than non-authoritarians (and low DTA authoritarians) in the absence of exposure to a political message. Figure 1 illustrates the conditional effect of authoritarianism on political tolerance at different levels of DTA within the forced control group in each sample. In Canada there is clear evidence that the negative effect of authoritarianism on political tolerance increases as DTA reaches one standard deviation above the mean. In the USA, the conditional effect of authoritarianism on political tolerance is weaker. However, the difference in conditional effects between the samples was not statistically significant (p = .40). This finding offers one way to address the apparent contradiction between Stenner's (exogenous) activation model and Hetherington's (endogenous) persistence model. The difference can be partly resolved by measuring how readily individuals can already access threat-related ideas without exposure to any new information.
We return to this idea in the discussion.    Second, as suggested by the authoritarianism literature and Zaller's (1992) model of opinion formation, we tested if attitudinal differences between the control group and those exposed to the political messages were conditioned by the accessibility of threat-related cognitions (DTA). Model 2 supports this claim, as six of the seven coefficients for the three-way interaction of experimental group by authoritarianism by DTA are statistically significant. The only exception was the balanced message in the choice condition. The balanced message was associated with higher levels of political tolerance (relative to the control group) in the choice and forced exposure conditions in both samples.
To illustrate the effect of each message across the full range of authoritarianism, and at one standard deviation above and below the average level of DTA, marginal effects were computed separately for the Canadian ( Figure   3) and US (Figure 4) samples. The free speech message was associated with higher tolerance among low DTA non-authoritarians in both country samples. Interestingly, to a lesser extent, high DTA authoritarians selecting the free speech message in Canada were more tolerant than the control group. The balanced message was also associated with more tolerance in the US sample and, to a lesser degree, in the Canadian sample among low DTA, non-authoritarians. The risk message is associated with less tolerance among low DTA authoritarians and among high DTA non-authoritarians in the Canadian sample. Third, we examined the possibility that, given the widespread messages of intolerance available to Americans during the 2016 presidential election cycle, exposure the risk message would be associated with less political tolerance among Canadians but not Americans. This was supported as the slope for the conditional effect of the risk article in the choice group, which shows that on average the effect was -.625 (p < .05) less in the USA than Canada. The coefficients for effects of the other messages are also all negative (i.e., the effect was less in the US sample than the Canadian sample) but do not reach the conventional level of statistical significance.  at or above the mean level of authoritarianism and who selected the risk message were less tolerant than those who read the risk message in the forced exposure group, F(1,50) = 4.12, p < .05. For instance, those who were one standard deviation above the mean level of authoritarianism were -.40, 95% CI [-.77, -.02]; Cohen's d = .53, less tolerant in the choice group than in the forced exposure group. While this difference was larger among those with low levels of DTA, the difference was not statistically significant.
Why did the effect of the free speech message in both samples and, in Canada, the risk message appear larger under conditions of selective exposure? Levels of political tolerance may differ in the choice and forced exposure groups due to the differing effects of the messages when choices are available or because message selection was based on pre-existing attitudes. We examined these possibilities by taking advantage of the participant preference (Arceneaux & Johnson, 2013) element of our design. Table 3 reports the results of comparing message effects between those who preferred and did not prefer the message they were assigned in the forced exposure Selective Exposure and the Authoritarian Dynamic group. A significant interaction between preference and message would indicate that the message effects are larger among those who wanted to view the message rather than some other reason, including self-selection based on prior attitudes. In fact, this was the case for the free speech message (b = .399, p < .05). By contrast, the coefficients for the risk message did not reach the conventional level of statistical significance.
To explore once again the possibility of a three-way interaction between authoritarianism, DTA and message preference, we produced a second analysis (see Section 2 of the Appendix in the Supplementary Materials). As illustrated in Panel A of Figure 5, the risk message only had an effect among those low DTA authoritarians who did not prefer to read it. And, by contrast, Panel B shows that the effect of the risk message held among all high DTA non-authoritarians.
In sum, we can conclude that the causal effect of selecting the risk message was limited to chronically threatened non-authoritarians.

Discussion
A growing body of research shows that threat can produce authoritarian attitudes in the general public (Hetherington & Suhay, 2011;Hetherington & Weiler, 2009;Stenner, 2005) as well as potentially put democracy at risk (Merolla & Zechmeister, 2009). Nonetheless, the nature of the authoritarian dynamic in society has been the subject of much debate. This study has shed light on the informational aspect of the authoritarian dynamic by examining information-seeking behavior and the subsequent effects of consuming political messages on political tolerance.
The average citizen may not engage in "defensive avoidance" by actively screening out attitude-challenging information (Garrett, Carnahan, & Lynch, 2013). However, our findings clearly show that authoritarianism is associated with avoidance in both Canada and the USA. When given the choice, authoritarians (compared to non-authoritarians) avoided free speech and balanced (pro/con) messages and, instead, consumed non-political news that had no relevance to the civil liberties controversy they were given the opportunity to consider.
As expected by the activation model (Stenner, 2005), threatened authoritarians were the most politically intolerant group in our samples. Nonetheless, our results clearly point to the limited effects of risk messages on authoritarian intolerance in contemporary US and Canadian societies. While risk messages have the potential to foster intolerance among those who do not normally receive them, the opportunity for selective exposure did not increase the effect of the risk message among authoritarians who reported that they normally prefer such content. Instead, our results suggest that the difference between the choice and the control group that was observed in the Canadian sample was due to self-selection based on prior attitudes; that is, authoritarians who selected and read the risk message were already relatively intolerant.
Importantly, however, we also discovered that threatened non-authoritarians became more politically intolerant after exposure to the risk-focused message in a manner consistent with the work of Hetherington and his colleagues (Hetherington & Suhay, 2011;Hetherington & Weiler, 2009). Moreover, our results showed a significant increase in political tolerance after exposure to the free speech message was limited to unthreatened non-authoritarians.
And the opportunity for selective exposure increased this effect. Hence, these results clarify how political messages may produce polarization around a civil liberties controversy: opportunities for selective exposure can increase tolerance among those "libertarians" who attend to pro-tolerance commentary, while risk messages also reduce tolerance among threatened non-authoritarians.
A minor but noteworthy finding is that seven percent of threatened authoritarians in Canada did select and read the message in favor of free speech. This group was more tolerant than the control group after reading the free speech message. One possible explanation for this finding is that the effect was limited to low-knowledge authoritarians who did not understand the poor fit between the message and their disposition. We tested this possibility by adding the political knowledge scale score to the interaction terms in Model 2. However, the results showed just the opposite effect; that is, high knowledge, threatened authoritarians who selected and read the free speech message were more tolerant than those with low knowledge. This may indicate a "reappraisal" of the threat similar in manner to that found by Houston and Holmes (1974) or simply selection based on prior attitudes. Nonetheless, the group was too small for a definitive analysis of how they might differ from other authoritarians.
Taken together, our findings are important for broader debates about the potential threat authoritarian attitudes can play in a democracy. Past research has convincingly demonstrated authoritarians' willingness to deny basic rights to groups they find objectionable. Yet, recent attempts to place this finding in the real, mediated world have provided contrary expectations. Our work clearly shows that the risk of authoritarian intolerance is partly tempered by their avoidance of content (or issues) they find distasteful. At the same time, the strength of non-authoritarians' commitment to democratic principles can be weakened (or strengthened) based on dominant discourses about the risks or need for free speech.
Future research should further explore the complex connection between authoritarian dispositions and the ways these interact within mediated social contexts. Several avenues seem particularly fruitful. First, while our study showed very few substantive differences across Canada and the US, future research could dig more deeply into the ways in which countries vary in the signals they send. National media contexts certainly vary in terms of the diversity of media content available as well as the presence and market share of publicly-funded media. We might expect that journalistic norms, especially among public broadcasters, will lead to coverage that is more neutral and less inflammatory, which activates authoritarians and non-authoritarians differently. Countries in which media choices are limited, or online penetration rates are particularly low, may cushion the opportunities for selective exposure. Second, our study has focused on a singular exposure to a message. Real world media consumption is usually not characterized by a single exposure to a news article, but an inundation of information over time.
Exploring how repeated exposure to a frame changes the results versus repeated exposure to contradictory frames could prove a fruitful direction for future studies.

Conclusion
This study explored when and to what extent selective exposure to political messages can produce political (in)tolerance among authoritarians and non-authoritarians. With online samples from the United States and Canada, we examined how authoritarians and non-authoritarians reacted to framing around civil liberties controversies. The results show that authoritarians who are politically knowledgeable generally avoid messages that promote free speech by consuming non-political information. While messages about the dangers of free speech have the potential to produce more intolerance among authoritarians, we found that this effect was limited to those who are the least likely to consume them when given a choice. By contrast, we found that messages about the risk posed by free speech produced intolerance among non-authoritarians for whom threat-related cognitions, as measured by the death-thought accessibility protocol, were already chronically accessible. The effects of procivil liberties messages were limited to unthreatened non-authoritarians. Hence, we conclude that in the contemporary information environment selective exposure can increase attitudinal polarization around a civil liberties controversy by producing attitude change among non-authoritarians.

Notes
i) The content and design of the selective exposure experiment was previously tested in a pilot study among students at the State University of New York at Potsdam in 2014.
ii) The literature on authoritarianism continues to include a variety of measures. Given the range of choices, we opted to pattern our measure on the child-rearing battery to ensure that our results are comparable to those authors we cite in this paper, especially Stenner and Hetherington and his colleagues. Moreover, use of the much longer RWA inventory would have been costly given that we collected data from national samples. (The shorter RWA validated by Bizumic and Duckitt was not published until 2018.) Use of the Likert-type items produced a coefficient of reliability was very similar to those reported in other studies.
For an example of another project that used a very similar set of response options, see the documentation of the United States Citizenship, Involvement, Democracy Survey from 2006, at https://www.icpsr.umich.edu/icpsrweb/civicleads/studies/4607 iii) Unlike traditional least-liked scales, we had respondents rate their level of dislike for all groups before selecting the group they disliked "the most". By doing so, we are able to understand better how our selected groups varied across participants, not just how the least-liked group was received. Based on work by Gibson (2013), we believe the least-liked measure here captures a different attitude from general support for civil liberties.
iv) The French version of the protocal was provided by Bastien Trémolière, and was used in past research with French-speaking samples. See, for example, Trémolière (2013).
v) We acknowledge that the alpha coefficient appears low for the DTA scale. Interestingly, recently published works using the word fragment approach have not reported a coefficient of scale reliability. Hence, we have no basis to judge the value in this case. Nonetheless, the literature reviewed by Hayes, Schimel, Arndt, and Faucher (2010) clearly shows the validity of such an approach to measuring the accessibility of death-related thoughts.
vi) The associations between each dichotomous knowledge item in the US (Φ = .227, p < .01) and Canadian sample (Φ = .204, p < .01) was moderate. The additive scale of correct knowledge items also had the expected correlated with educational attainment in both countries (US r = .300, p < .01; Canada r = .241, p < .01).
vii) In addition to standard demographic controls, and also country sample, we add affective rating of the least-liked group because some groups are more disliked than others.

Funding
This project was made possible by the generous support of the Fonds québécois de recherche-société et culture (FQRSC), grant number 192786.