Still Not Important Enough? COVID-19 Policy Views and Vote Choice

Scholars have long been skeptical of citizens’ ability to vote on the basis of their policy views. Voters lack incentives to pay attention to politics and so are often unaware of the policy stances adopted by presidential candidates and parties. However, some scholars have suggested that voter attention may increase when policy issues become important to them, such as when a crisis disrupts their lives. The coronavirus pandemic provides an opportunity to test this proposition. It is one of the most severe crises the United States has faced. It has disrupted almost everyone’s lives, and many people know someone who has tested positive or died from the virus. It is thus salient and important to many—if not most—voters. Despite this context, we find that many voters remain unaware of the 2020 US Presidential candidates’ stances on coronavirus policies. Their levels of knowledge are about typical for other policies, which is middling. In the absence of knowledge, voters cannot connect their policy views on the virus with their presidential voting decisions.

I n one view of democracy, citizens hold politicians accountable by voting for those whose policy stances are closest to their own (Downs 1957). However, voters' attitudes about public policy have long been known to be inconsistent and unstable (Converse 1964;Freeder, Lenz, and Turney 2019). The attitudes they report on surveys often reflect the considerations that are at "the top of their head" (Zaller 1992). Voters also tend to adopt the positions of their preferred candidate (Brody and Page 1972;Lenz 2012). Moreover, voters frequently lack knowledge about candidates' positions on major issues in American politics, even in presidential elections (Layman and Carsey 2002). Consequently, some scholars have long been skeptical that policy issues have much impact on the way people vote (Campbell et al. 1960). Instead, most voters simply vote for the presidential candidate from the party they identify with (Bartels 2000).
In part, this may simply reflect rational ignorance on the part of voters (Downs 1957). Given the incredibly small likelihood of a voter casting a decisive vote, voters have little incentive to learn about politics. As Dahl (1966, 305) wrote, "politics is a sideshow in the great circus of life." Citizens' lack of knowledge of candidates' positions may also reflect the lack of easily digestible issue coverage in the media (Zaller 2003 andBennett 2003;Boczkowski, Mitchelstein, and Walter 2012). It might also reflect the clarity of the positions candidates present to voters (Tomz and Van Houweling 2009). The problem of how voters hold politicians accountable despite the collective action problem at the heart of democracy is sometimes called the "democratic dilemma" (Lupia and McCubbins 1998).
However, scholars of voting behavior have long recognized that voters may be able to vote on the basis of issues if they are personally important to them (Converse 1964;Krosnick 1988). According to this perspective, voters are motivated to search for information on important issues and to think more deeply about them (Boninger et al. 1995). Voters may have more of an incentive to learn about important issues because the benefit they derive from their preferred policy on such issues is larger. While A list of permanent links to Supplemental Materials provided by the authors precedes the References section.
there is evidence that voters seek out information about issues that are important to them (Bolsen and Leeper 2013), the evidence that importance increases issue-based voting is mixed at best (Bartle and Laycock 2012;Grynaviski and Corrigan 2006;Leeper and Robison 2020).
The 2020 presidential election fell amid one of the most severe crises in U.S. history-the coronavirus pandemic. This coincidence provides the opportunity to study whether severe crises can motivate voters to learn about politics, especially to learn what they need to know to hold candidates accountable. In this paper, we examine whether we see this learning among all U.S. residents-since the virus has affected everyone-but also among individuals who it affected more deeply.
One key obstacle to studying issue importance is measurement. Most scholars rely on subjective measures of personal or national importance (Boninger et al. 1995;Krosnick 1988). However, many voters' responses to these questions seem to reflect the salience of issues in campaigns and not necessarily their personal importance (Bartle and Laycock 2012;Johns 2010). One recent approach to measuring salience relies on analyzing decision-making by citizens (Hanretty, Lauderdale, and Vivyan 2020).
In this paper, we take a different approach. Since the virus affects all Americans, it should be personally important to virtually everyone. Fear of the virus, lockdowns, social distancing restrictions, school closures, office closures, and travel restrictions have disrupted lives in a way Americans have not experienced since at least the world wars. The virus has also affected some individuals even more deeply, including those who have caught the virus, have close friends or family members who have tested positive or who have even died. We assess whether these experiences increase the accuracy of voters' knowledge of parties' positions on the virus and whether they, in turn, increase the apparent weight of the virus on Americans' voting decisions in the November 2020 presidential election.
Prior studies have found that crises increase citizens' political engagement (Alaminos and Penalva 2012;Kern, Marien, and Hooghe 2015). There is also evidence that voters responded to deaths in the Iraq War and to the 9/11 terrorist attacks (Gartner 2008). None of these studies, however, have measured the critical piece of knowledge that voters need to hold politicians accountable: the candidates' policy stances. Does a major crisis induce citizens to learn what they need to know to vote for the candidate who best reflects their policy views?
While the argument about issue importance suggests that a major crisis should increase issue voting, there are reasons to expect it would not. Prior research shows that when candidates adopt ambivalent positions, voters project their own position onto their preferred candidate instead of learning the candidates' stances (Tomz and Van Houweling 2009). In the current crisis, the candidates have adopted contrasting stances on coronavirus issues. Biden has clearly supported masks, 1 for example, while Trump ridicules them. 2 However, Trump has been ambiguous enough that some voters may perceive him as being in favor of masks. For example, he once called masks "patriotic." 3 For a summary of news articles and Trump quotes on masks, refer to online appendix A.
To investigate candidate policy knowledge and voting during the pandemic, we ran a large online survey of 3,704 Americans during the 2020 presidential election campaign. 4 We first consider Americans' attitudes on coronavirus policies. We find that they are strongly supportive of measures to fight the virus. Even Trump supporters are broadly supportive of such policies. We then assess knowledge of the candidates' stances on these policies. Contrary to the importance argument, we find low levels of knowledge on virus-related issues, particularly among Trump voters. Moreover, levels of knowledge are lower among voters who have had the virus or who have a close friend or family member who has had it. Finally, we consider the relationship between coronavirus attitudes and vote choice. We find that coronavirus attitudes are only associated with vote choice among high-knowledge voters.
These findings cast doubt on the argument that voters tune in when issues deeply impact their daily lives. Even during a severe crisis, with an incumbent president out of line with voters, voter knowledge of this issue is not above levels we typically see for other issues.

Issue Importance and Voting Behavior
Ever since the classic works by Campbell et al. (1960) and Converse (1964), scholars have been aware of two major limitations to voters' ability to vote for the candidate who shares their policy views. First, voters' policy views tend to be unstable (Converse 1964) and to reflect the considerations that become salient when respondents are asked about them (Zaller 1992). Second, voters lack knowledge of the positions adopted by competing candidates and parties (Campbell et al. 1960;Freeder, Lenz and Turney 2019). Converse (1964) did suggest though that individual issues might have more importance to segments of the electorate he called "issue publics," for which they are particularly important. More recently, Krosnick (1988) has taken up this argument and argued that voters who subjectively believe an issue is important will put more effort into learning about the issue and thinking about it. On most issues, the benefit from particular policies is too small to justify the costs of learning about the issue for most voters (Downs 1957). On important issues though, the benefit from a desired policy may be larger, so voters may have more knowledge about them. In turn, they may be better able to connect their policy preferences to their vote choice (Boninger et al. 1995). People like to have consistent attitudes (Festinger 1957;Heider 1958) so they should 548 Perspectives on Politics put more effort into voting in a way that reflects their policy preferences on issues that are important to them. Selfinterest may also play a role in fostering issue importance (Boninger et al. 1995;Miller, Krosnick, and Fabrigar 2017).
Despite its intuitiveness, the evidence that issue importance increases issues-based voting is limited (Krosnick 1988). There is evidence that importance leads people to pay more attention to information about an issue (Bolsen and Leeper 2013). There is also evidence that importance reduces elite influence on opinions (Carsey and Layman 2006;Lecheler, de Vreese, and Slothuus 2009;Mullinix 2016). However, it is not clear that issue importance is associated with greater issue knowledge (Price et al. 2006). Moreover, several studies in recent years find that issue importance has no effect on issue voting (Bartle and Laycock 2012;Grynaviski and Corrigan 2006;Leeper and Robison 2020). The impact of self-interest on subjective issue importance may also be countered by other influences like group identification (Boninger et al. 1995). Voters for whom an issue like the coronavirus is objectively important due to their self-interest may subjectively perceive it as less important because their party dismisses it. Therefore, objective importance may not lead to enhanced learning or issue voting.
Issue importance may have effects on voters other than stimulating issue voting (Brody and Page 1972;Krosnick 1990a). In particular, voters may feel a heightened need for consistency in their attitudes. Voters may achieve consistency by changing their own policy views, but presumably if an issue is important to them they may not adjust their policy views to reflect their candidate preferences. Voters seem more likely to achieve consistency by projecting their own position onto their preferred candidate and possibly project a less preferred position onto the other candidate (Conover and Feldman 1982;Martinez 1988). New issues, less familiar candidates, candidate ambiguity, and the resulting low levels of knowledge would seem to facilitate such projection. When a candidate adopts ambiguous stances on an issue (Brody and Page 1972;Martinez 1988), they may make it easier for voters to project their own views onto their preferred candidate. Indeed, previous research has shown that voters respond to candidate ambiguity by projecting their positions onto the candidate they like (Tomz and Van Houweling 2009). Thus, on a new issue on which one or more candidates adopt ambiguous positions, projection may be a likely strategy to reduce inconsistency between the policy preferences of candidates and voters.
We therefore test two contrasting expectations. If issue importance improves citizens' ability to vote based on their policy views, we should find high knowledge of the candidates' coronavirus stances and a strong association between coronavirus attitudes and vote choice. Moreover, both knowledge of candidate policy stances and the link between coronavirus attitudes and vote choice should be stronger among voters who were more personally affected by the virus. Conversely, if issue importance leads voters to project their positions onto their preferred candidate, we should find a weak relationship between coronavirus views and vote choice. Both knowledge and the relationship between preferences and vote choice should weaken among voters who are more personally affected.

Data
We ran a pre-electoral survey using the opt-in online survey provider Lucid with 3,704 voting-age American respondents between August 19 and 22, 2020. 5 We created survey weights using iterative proportional fitting (raking) by matching the survey distributions on age, gender, race, Hispanic identity, and education to Census Bureau benchmarks. Given the problem with inattentiveness on opt-in panels (Aronow et al. 2020), we only accepted responses from individuals who correctly reported the calendar date and we also report on the robustness of our findings to an additional stringent attention check.
We asked respondents to provide their positions on thirty-five policy issues. We also asked them to provide their perceptions of the candidates' positions on fourteen of those issues. Table 1 shows the fourteen issues for which we have both respondents' positions and their perceptions of the candidates' positions. We asked respondents whether they "agree," "disagree," or "neither agree nor disagree" with each statement. Three of the issues relate to the coronavirus (Masks in Public, WHO, and Close Businesses). In addition to these fourteen issues, we asked respondents how much effort they want from government on six other issues. Three of these effort questions concern the virus: coronavirus testing, making sure people with the virus stay at home, and protecting Americans from the coronavirus. For question wording, refer to online appendix B. We also asked questions about respondents' own agreement with ten policies and five questions asking respondents how much they think the government should spend on various policy areas.
For analyses of issue voting, following Ansolabehere, Rodden, and Snyder (2008), we create a coronavirus attitudes scale using the six items about coronavirus policies, which varies from 0 to 1, where 0 reflects the lowest possible support for policies to protect against the coronavirus and 1 reflects the highest possible support for such policies (Cronbach's alpha is 0.84).
We have two indicators that the coronavirus is particularly important to some voters. The first is a question asking respondents whether they, a close friend, or a family member has tested positive for COVID-19 this year. Twenty-seven percent responded "yes" to this question. The second asks whether a close friend or family member has died from COVID-19. Fifteen percent responded "yes" to this question. Though uncertainty exists about actual infection and death rates, these percentages are broadly in line with estimates from COVID tracking models. 6

Public Support for Coronavirus Policies
What did Americans think about policies to protect them from the coronavirus? We asked respondents questions about this in two formats. First, we asked whether they "agree," "disagree," or "neither agree nor disagree" with mask requirements in public, closing businesses to slow the spread of the virus, and WHO membership. Second, we asked respondents how much effort they want the federal government to put into coronavirus testing, protecting Americans against the virus, and making sure people with the virus stay at home. On these questions, respondents could answer "much more than now," "somewhat more than now," "the same as now," "somewhat less than now," and "much less than now." Figure 1a shows the percentages of all respondents who agree with a mask requirement, closing non-essential businesses, and WHO membership. Figure 1b shows how much effort voters want from government to test for the coronavirus, protect Americans from the virus and to force Americans with the virus to stay at home.
Overall, Americans are supportive of policies to fight the virus. Sixty-nine percent agree with requiring masks in public (11% disagree), 49% agree with closing nonessential businesses (22% disagree), and 51% agree with WHO membership (20% disagree). Moreover, 70% want the government to do more to protect Americans from the virus (8% want it to do less), 70% want the government to do more to force infected people to stay at home (7% want the government to do less), and 65% want the government to do more to test people for the virus (10% want it to do less). Figures 1c and 1d show responses among Trump voters. With the exception of closing non-essential businesses and WHO membership, majorities of Trump voters support these policies to fight the coronavirus.

Placement Knowledge
Americans were strongly supportive of policies to fight the coronavirus in 2020. For voters to hold politicians accountable to these views, voters must have some sense of the candidates' stances (Campbell et al. 1960). How do these compare to knowledge of stances on other policies? As explained earlier, we asked respondents their perceptions of the candidates' positions on fourteen policy issues in addition to their own attitudes. These include three coronavirus issues: requiring masks in public, closing nonessential businesses, and WHO membership.
We consider respondents as knowing the candidates' stances if they placed Biden as more supportive than Trump on masks, closing non-essential businesses, and WHO membership. Recall that these questions had three response categories: "agree," "disagree," and "neither agree nor disagree." To be coded as correct, respondents had to place Biden at "agree" and Trump at either "neither agree nor disagree" or "disagree", on the one hand, or Biden at "neither agree nor disagree" and Trump at "disagree," on the other hand. To encourage people to volunteer knowledge, we did not include a "don't know" option, following Mondak and Davis (2001). We code the very small percentage of respondents who skipped these questions as incorrect (0-2 percent across all fourteen policy items).
According to this measure, only 39% of respondents correctly placed the candidates on the mask issue, only 41% did so on closing non-essential businesses, and only 47% correctly placed the candidates on the WHO. Given Reducing the size of the income gap between rich and poor Americans Close Businesses Closing non-essential businesses to slow the spread of the coronavirus Note: These are the issues for which we have both respondents' own positions as well as their perceptions of the candidates' positions.
Respondents were asked whether they and the candidates agree, disagree or neither agree nor disagree with each item.
550 Perspectives on Politics the enormity of the crisis, these levels of knowledge seem low. 7 To put these in context, figure 2 shows the percentage who place Trump on the conservative side of Biden on all fourteen policy questions we asked about. These questions use the same response format as the coronavirus agree-disagree policy questions and we code correct responses in the same way. This figure reveals a limitation to Americans' policy voting behavior. Even on longstanding issues in American politics like abortion, only  Note: This figure shows the percent placing Trump on the conservative side of Biden on each issue. It shows that, despite a once-ina-century crisis, the public hasn't learned more about presidential candidate stances on coronavirus policies than typical for other issues.
552 Perspectives on Politics about 40%-50% of the public correctly answer these questions. Given that 19% would get this right by chance alone, this is a low number. 8 The two issues on which respondents' perceptions seem most accurate are building a wall on the Mexican border and repealing the ACA, both of which are highly salient issues that Republican and Democratic elites have emphasized over the last few years. Voters have thus been exposed to the parties' positions on these issues for longer and so have had more time to learn about them. Because it takes time for voters to learn parties' positions (Adams, Ezrow, and, Somer-Topcu 2011) and because voters project their own positions onto their preferred candidates and parties, the length of exposure and the consequent better sorting on these issues should increase apparent knowledge. Low levels of knowledge about presidential candidate positions on the other issues are consistent with longstanding research findings (Layman and Carsey 2002;Freeder, Lenz, and Turney 2019). 9 Given the severity of the coronavirus, we might expect voter knowledge to be substantially higher on pandemic questions than on other questions. Instead, however, voter knowledge on pandemic questions seems typical of other policy items. The three pandemic questions fall in the middle of the distribution, below the items about the border wall and the ACA but above a handful of issues that voters find difficult, such as free trade and lowering taxes.
It is important to note that these low levels of knowledge do not arise primarily because candidates took highly ambiguous positions on these issues, which even highly knowledgeable respondents would get wrong. Among respondents who are generally knowledgeable about policies-measured by their percent correct on other policy questions-the percent correct across all fourteen policies, including the coronavirus policies, is typically above 90%.
These low levels of policy knowledge are also not about attention, though attention may contribute. As we noted earlier, we screened out respondents who could not provide the date of the month. We also included a very difficult attention check with an obvious but wrong answer, which only 36% of respondents passed (refer to online appendix B3). Among these 36% we do see somewhat higher levels of knowledge, but they are still low. The percent placing Biden as more supportive than Trump of masks in public rises from 39% in the full sample to only 46% in the group that passed this challenging attention check. Across all fourteen issues, those passing it place Trump as more conservative than Biden just 50% of the time on average.
Nevertheless, correctly placing the candidates may be unusually hard for Trump voters. They are supportive of aggressive measures on the virus and, therefore, out of line with Trump. As already mentioned, Trump has also sent conflicting signals about masks and other coronavirus policies. How much has that reduced accuracy among Trump supporters?
Before we examine this question, it is important to note that even in groups that do not face this conflict, such as Biden voters on the mask issue, we still do not see high levels of knowledge. Figure 3 shows the percent correct across all fourteen policies separately for Trump voters, Biden voters, and individuals who are unsure of their vote intentions, plan to vote for another candidate, or who do not plan to vote. It shows that even Biden supporters, who strongly support wearing masks and have a candidate who strongly supports wearing masks, still only correctly place the candidates about 60% of the time.
To examine how much the conflict Trump voters face reduces their knowledge, figure 4 shows perceptions of each candidate's stances by the candidate's own voters. Among Trump voters, we see signs that his lack of alignment with many of them depresses knowledge. A majority (56%) perceived that Trump supports masks and large minorities perceived that he supports closing nonessential businesses (35%) and the WHO (31%). The figure also shows that large majorities of Biden voters perceived Biden as supporting closing non-essential businesses, a mask requirement, and WHO membership. As we can see, few people skipped the questions asking about the candidates' coronavirus stances. Refer to online appendix D for voters' perceptions of the candidates on all the issues.
Trump voters' perceptions of Trump in figure 4 seem consistent with projection, and Biden voters' perceptions of Biden could be as well. Figure 5 provides more direct evidence of projection. It shows the percentage of each candidate's voters who "agree," "neither agree nor disagree," or "disagree" with each policy who believe their preferred candidate agrees with the policy. Overwhelming majorities of voters who agree with each policy also think their preferred candidate agrees with it as well. Very few voters who disagree with policies believe their preferred candidate agrees with those policies. Thus, voters may be projecting their attitudes onto their preferred candidate.
It is possible that voters project only in the absence of knowledge, but it is also possible that voters deceive themselves through motivated reasoning (Kunda 1990) or deceive the researcher through cheerleading (Bullock et al. 2015). 10 To gain a sense for these latter two mechanisms, we examine perceptions among voters with high knowledge of candidates' stances in general. We create a general knowledge measure based on the eleven candidate position questions not about the virus. We gave respondents one point for each issue on which they placed Trump on the conservative side of Biden. Among the top 20% of respondents on this measure, those voting for Biden saw Biden as more pro-mask than Trump 96% of the time. In contrast, those voting for Trump did so only 63% of the time. People who did not support either   Perceptions that Trump agrees with each policy by Trump voters' own preference candidate but had high knowledge did so 86% of the time. These findings are consistent with knowledgeable Trump voters engaging in motivated reasoning, projecting, or cheerleading. Clearly, lack of knowledge is not the only problem. Do voters for whom the coronavirus is particularly important have better knowledge of the candidates' stances? We consider two measures of issue importance: whether the respondent has a close friend or family member who tested positive for COVID-19 and whether they have a close friend or family member who died from COVID-19. On all three virus policies and using either measure of importance, we find that importance actually decreases the percentage of respondents who place the candidates on the correct sides of each other. For example, 40% of respondents without a close friend or family member who died correctly place the candidates on the mask issue, but only 34% of those who lost a close friend correctly placed the candidates on that issue. While we should not assume importance caused lower knowledge because being more personally affected by the virus is far from random, this finding suggests that importance fails to improve knowledge of the candidates' stances. 11 In sum, we find that voter knowledge of Trump's and Biden's stances on coronavirus policy is surprisingly low. The pandemic has disrupted voters' lives and yet they have not learned more about this issue than other issues. These findings thus provide more support for the projection expectation than for the importance expectation. To some extent, voters' tendency to engage in motivated reasoning and cheerleading may lower the level of knowledge among Trump voters.

Voting on an Important Issue
The previous section showed that voters lack a key precondition for voting on the basis of a policy issue: they lack knowledge of the candidates' stances on coronavirus policies, despite the virus's importance. Does this prevent them from voting in a way that reflects their attitudes on this issue?
We examine this question by considering the relationship between coronavirus attitudes and presidential vote choice. Before presenting this relationship, it is worth emphasizing that we cannot draw causal inferences from any association we find-attitudes and vote choice are measured at the same time from the same brain, and so the possible causal processes are numerous. In particular, we would expect a great deal of reverse causation: Trump voters who learn about Trump's opposition to masks and lockdowns, even if this opposition was not always consistent, may "follow the leader" and adopt his positions (Lenz 2012). If so, this could make it appear as if virus policy views were influencing Trump vote, that is, could bias the association upwards. In light of these complexities, we keep the analysis as simple and transparent as possible.
With this upward bias in mind, we nevertheless find a weaker relationship than we might expect between virus policy attitudes and vote choice. As a reminder, we measure virus policy attitudes with a six-item scale that varies from zero to one, where higher values indicate support for aggressive policies against the virus. Respondents at the midpoint of the coronavirus attitudes scale voted for Trump around 70% of the time. Trump vote percent does not fall below 50% until respondents reach around 0.8 on the 0-1 scale. Overall, 76% of voters supported aggressive policies (were above the midpoint of the scale), and yet the same voters nevertheless voted for Trump at a surprisingly high rate: 40%. Figure 6 shows these patterns. It presents the relationship between Trump vote choice (versus Biden) on the y-axis and the six-item coronavirus attitude scale on the x-axis. It summarizes the relationship using a loess line. Although the figure undoubtedly reveals a strong association, that association is shifted far to the right (and possibly bolstered by reverse causation). Thus, while there is a relationship between COVID-19 attitudes and vote choice, the categories of voters who are most numerous (i.e., who want strong policies to protect against COVID-19) remain relatively Among voters for whom the coronavirus is personally important, does this virus-vote relationship strengthen? In other words, are those who caught the virus or who have a close friend or family member who tested positive or died from the virus more likely to punish Trump if they support aggressive policies? Figure 7 shows the relationship between coronavirus attitudes and vote choice separately for people for whom the issue is especially important. The left panel uses a positive COVID-19 test for the respondent or a close friend or family member as a measure of importance. The right panel uses the COVID-19 death of a close friend or family member as a measure of importance. In both panels, we see some signs that respondents who want the government to do more to fight the virus defect from Trump at higher rates, especially among those above 0.75 on the scale. The panels also show that the relationship between coronavirus attitudes and vote choice is weaker for people who were personally affected. This holds true for both measures. Using the deaths measure, the relationship between COVID attitudes and vote choice is essentially non-existent among people with a friend or family member who died. These descriptive plots therefore provide no evidence that importance increases the relationship between coronavirus attitudes and vote choice.
Would voters hold Trump accountable if they learned his and Biden's stances? This is a difficult question that we also can't answer with certainty. To shed some light on it, figure 8 again presents Trump vote and the coronavirus attitude scale by voter knowledge about their stances on virus policy. It distinguishes voters with high knowledge (i.e., above median, meaning more than one of the three placements right) from those with low knowledge (i.e., below median, meaning none or one placement right). It shows an even stronger relationship between coronavirus attitudes and vote choice among voters with high virus stance knowledge: the probability of voting for Trump falls from about 1 all the way down to 0 as highknowledge voters become more supportive of measures to fight the virus. Among voters with low knowledge, however, that relationship is much weaker. Among them, the probability of intending to vote for Trump varies from just over 0.9 to about 0.6 as voters become more supportive of policies to fight the virus. More strikingly, even low knowledge voters who most strongly support policies to fight the virus have a high probability of supporting Trump (just above 0.60). Conversely, regardless of political knowledge, voters who are least supportive of policies to fight the virus very strongly support Trump. Thus, Trump seems to stay relatively popular by getting votes from voters who oppose policies to fight the virus regardless of their knowledge of candidates' stances and from voters who support such policies but do not realize that Trump is less supportive of them.

Conclusion
Given the infinitesimally small chance voters will ever cast a decisive vote, voters lack an incentive to pay attention to politics and thus to vote in a way that reflects their policy preferences. As a result, voters know surprisingly little about presidential candidates' policy stances. However, there is a long-running argument that, even though voters may not be able to vote in a way that is consistent with their policy attitudes on most issues, on some issues that are particularly salient or important to them, voters should be able to develop vote preferences that are consistent with their policy preferences (Converse 1964;Krosnick 1990b). We also considered the alternative possibility that voters project their stances onto the candidates on important issues (Krosnick 1990a;Martinez 1988;Tomz and Van Houweling 2009). We tested these possibilities on an issue that is undeniably important to voters: the coronavirus crisis. Despite facing the greatest crisis in the United States since World War II, voter knowledge about Trump's and Biden's stances on coronavirus policies is seemingly low. It is typical of the low levels found across most policy issues. In the absence of knowledge, voters appear to project their positions onto their preferred candidate. Motivated reasoning and cheerleading may contribute to the appearance of low levels of knowledge. Even among respondents who lack incentives to engage in motivated reasoning, however, we still find surprisingly low levels of knowledge. We thus find evidence that is inconsistent with the importance expectation and suggestive of projection: importance does not appear to help voters learn about the candidates' stances and vote for the candidate who shares their views.
Why have many voters failed to learn Trump's and Biden's stances? One explanation is that, even though people's lives are disrupted, they still face the core collective action problem at the heart of democracy-they lack an incentive to learn about politics since their votes will never matter. Another possibility is that they desire to learn but the pandemic has left them with even less time and other resources to engage (Verba, Schlozman, and Note: This figure shows the bivariate relationship between the six-item coronavirus policy scale and voting for Trump among voters with high (above median) and low (at or below median) knowledge of the candidates' coronavirus stances. It shows that the relationship between coronavirus attitudes and vote choice is much stronger among high-knowledge voters. Moreover, among low-knowledge voters, the probability of supporting Trump is over 0.6 even among voters who support the strongest measures to fight the virus. Points are jittered and the coronavirus scale is coded so that higher values indicate stronger support for policies to fight the virus. Loess line.

Perspectives on Politics
Brady 1995). Another possibility is that interest in politics may be too stable over the lifetime, reflecting a longstanding predisposition (Prior 2018), one not responsive to even severe crises.
These findings help us interpret the 2020 presidential election results. The coronavirus crisis likely had a smaller effect than its importance would indicate. More fundamentally, our findings have implications for our understanding of voting behavior. Issue importance does not seem to help voters connect their policy preferences to their vote choice. Instead, voters seem to project their positions onto their preferred candidate. Only for issues on which parties have acquired strong reputations over the past few years do majorities of voters have knowledge of the candidates' positions. Consequently, a candidate may be able to win an election despite holding unpopular stances on an important issue.

Supplementary Materials
To view supplementary material for this article, please visit http://doi.org/10.1017/S1537592721001997. McClellan 2019. 6 E.g., https://covid19-projections.com/us. For example, given death estimates by mid-August 2020, we would find that about 15% of people know someone who died if the typical person knew 300 individuals. Estimates of people's average network size range widely but are typically in the 300-700 range, e.g., Gelman 2013. 7 The nature of incorrect responses seems consistent with low levels of knowledge. On the mask question, about half of respondents we classify as incorrect say that both Trump and Biden agree that masks should be required in public. The next most common mistake, made by almost one-quarter of incorrect respondents, is to say that both candidates "neither agree nor disagree." The third most common mistake is to say that Trump agrees more than Biden, made by 10% of respondents. We classify only 3% as incorrect because they skipped the Biden or Trump placement questions. 8 Respondents can respond "agree," "disagree," or "neither agree or disagree," or skip the placement question for each candidate, yielding 4 Â 4 = 16 possible responses. They could place Biden as more supportive of masks in three ways, yielding a chance correct rate of 3 ÷ 16 = 0.1875. 9 Online appendix C presents the percent correct for all similar questions in ANES presidential election years, 1972-2016. The percentages in the Lucid survey are broadly consistent with the percentages presented there, though somewhat lower when compared with recent ANES surveys, especially 2016. We discuss possible explanations for the difference in the online appendix. 10 We often only think about cheerleading on valence issues, but public support for masks is sufficiently strong that some Trump supporters may want to downplay or deny his frequent (but not consistent) opposition to masks. 11 To examine whether confounders suppressed the relationship between these importance measures and knowledge, we adjusted statistically for demographics with least-squares regression and matching, but these revealed no signs of suppression.