What are the chances you’re right about everything? An epistemic challenge for modern partisanship

Introduction

Consider the following propositions:

Abortion: Abortion is morally wrong in most circumstances.

On the face of it, these propositions seem to be orthogonal. ¹ That is, your position on one of them doesn’t commit you to any particular position on the others.

Yet, in the context of the current political climate in the US, particular answers to these questions are strongly correlated. For instance, if you find that a randomly selected individual agrees with Abortion and disagrees with Policing, what would you expect them to say about Guns? You’d probably expect them to disagree with that claim. Likewise, if a person agrees with Wages and disagrees with Immigration, you’d expect them to disagree with Affirmative Action. Of course, not every individual will answer in this way, as there are people who do not fall neatly within either cluster of opinion. However, there are two noticeable clusters of opinion in the American context, and the partisan divide has been getting stronger. ²

The fact that doxastic attitudes toward orthogonal propositions cluster together among the population raises an epistemological problem. Suppose a person finds herself having the beliefs that are typical of one of the clusters of political opinion. Such a person, I shall argue, should moderate her political beliefs. For, the orthogonality of the issues involved makes it highly likely that something is going awry epistemically – either her beliefs have been subject to problematic irrelevant influences or she is in possession only of a biased subset of relevant evidence. ³

This article argues that strategies for vindicating the partisan’s epistemic standing face significant challenges. The plan going forward is this. In the second section, I lay out the polarized nature of political opinion in the US. Then, in the third section, I present the core argument. The fourth section examines possible responses that can be made on behalf of the partisan and argues that they fail. In the fifth section, I reexamine and defend in detail the prima facie plausible claim that the argument of this article rests on – namely, that several of the issues on which partisans disagree, for example, those laid out earlier on in this section, are really orthogonal. The sixth section argues that while the problem brought out in this article involves higher order evidence, it is importantly distinct from the well-studied epistemological issue of peer disagreement. How should a partisan moderate her political beliefs? This question is addressed in the seventh section. The eighth section concludes.

Polarization in the US

The American political landscape has become more polarized, particularly over the last two decades. A recent Pew Research Center (2014) report has documented trends in ideological consistency within the US from 1994 to 2014. What the study shows is a dramatic increase in ideological consistency over the two decades, not only among individuals identifying with a particular party, but in the general population as well. This polarization has a significant impact on the political process. The Pew report finds that ideologically consistent individuals are much more likely to vote and to make campaign contributions.

Using a five-point Likert-type scale, Cullen (2018) has gathered data on the nine political claims listed in the ‘Introduction’ section. The sample consists of 412 Amazon Mechanical Turk participants. ⁴ The correlations among the answers are striking. Individuals’ self-reported political identification was a strong predictor of their responses on the issues. The absolute values of the Pearson correlations between political ID (measured using a seven-point Likert-type scale) and the responses to the issues listed above ranged from 0.469 to 0.575.

There was significant correlation as well between attitudes toward seemingly orthogonal issues. For example, Immigration correlates with Guns with r = −0.390. The negative number indicates that people who agreed with Immigration tended to disagree with Guns. The correlation between Wages and Policing was 0.430; for Climate and Marriage, it was 0.389. The strongest relationship was observed between Abortion and Marriage, with r = −0.529, which is less surprising. These correlation coefficients indicate a moderate relationship, in statistical parlance, but they are quite striking results in a social scientific context. To put them into perspective, the correlation coefficient between the height and weight of male college students tends to be around 0.4. ⁵

Even more striking is the fact that principal component analysis on the data set revealed only one component, which explained 43.9 percent of the variation. This means that one underlying factor, that is, a linear combination of the variables, intuitively the ‘left–right’ axis, significantly explains people’s attitudes toward the nine issues.

However, there is some debate about whether a single left–right dimension meaningfully captures the ideological variation within the American population. For instance, Jost et al. (2009) lay out the case for a parsimonious unidimensional model, when aiming to meaningfully capture the distribution of political opinion. Feldman and Johnston (2014), on the other hand, argue that at least two factors are needed to model the distribution accurately.

Typically, the two dimensions are taken to be economic and social – see, for example, Carmines et al. (2012) and Feldman and Johnston (2014). Economic conservatives generally favor less government spending, fewer business regulations and trade barriers, and so on, as compared to economic liberals. Social conservatives are those who espouse traditional social values, and are usually marked by a pro-life stance on abortion, a traditional view of marriage as consisting of heterosexual unions, a restrictive position on drugs like marijuana, and so on. Of course, within the social sphere, what counts as conservative can vary greatly over time. Social liberals are those who oppose the kinds of positions listed above.

A two-factor model is necessary if it turns out that there is a good deal of ideological consistency within the economic and social camps but not across these camps. This would be true, if, for example, the people who think that the government should spend less also think that minimum wages are a bad idea, and environmental regulations do more harm than good, and so on – but, there is relatively little correlation between these beliefs and socially conservative beliefs on, say, abortion. In this way, the two domains would be independent, and a single right–left axis may not suffice to capture the distribution of political beliefs.

Weeden and Kurzban (2016) have recently argued that the ‘social’ category is too vague. The chief problem is that of locating various racial issues – relating to affirmative action, government spending and assistance programs for African Americans, immigration policy, racial discrimination, and so on. The authors find that opinions on these issues correlate strongly with opinions on economic issues, but not with opinions on issues typically characterized as ‘social’ – issues related to abortion, marijuana legalization, and homosexuality. The authors thus propose that the two ideological dimensions can be most fruitfully characterized as ‘racial/economic’ and ‘religious’.

Weeden and Kurzban find that there is significant correlation between racial and economic opinions. However, there is relatively less correlation between racial and religious opinions, as well as religious and economic opinions. This is the case with regard to the population as a whole. Nonetheless, the results are dramatically different in the case of non-Hispanic whites with high levels of human capital (top 20 percent). Weeden and Kurzban write,

This group has long used liberal-conservative labels as useful summaries of their issue opinions and shown a meaningful degree of left-right correspondence between religious and racial/economic issue domains. Further, this group has recently experienced a substantial rise in this cross-issue correspondence. (2016:54)

However, my argument in this article doesn’t rely on there being one ideological dimension, the left–right axis. Even if there are two dimensions, as Weeden and Kurzban claim, the skeptical problem arises if issues within a particular dimension are orthogonal. This seems to me very plausible. Consider, for example, Immigration, Wages, and Affirmative Action. These would fall within the ‘economic/racial’ category. But it seems that they are orthogonal. The problem gets more pressing with greater numbers of orthogonal issues, opinions on which cluster together. Of course, if Weeden and Kurzban are right, then in the American context, whites with high levels of human capital are particularly susceptible to the epistemic challenge presented in this article. This is because their opinions on a larger range of orthogonal issues show partisan polarization. Given that my argument doesn’t rely on whether there are two genuine ideological dimensions or one, I will proceed to talk, for ease of exposition, as if there is just a single left–right axis.

Another set of methodological questions that might be raised here have to do with whether the responses elicited in polls are an accurate representation of people’s beliefs. Perhaps when people answer such questions they could be signaling affiliation rather than reporting belief. To reduce this possibility, political identity was not made salient at the beginning of the questionnaire – rather, participants were asked to report their political ideology at the end. Relatedly, one might wonder whether survey research measures attitudes that are stable across time. ⁶ While this is important to think about, survey research like this is often the best we can do in order to elicit data from a wide range of participants. Moreover, surveys are ubiquitously used to uncover beliefs in a wide range of studies within political science, sociology, social psychology, and related fields. The methodology of the empirical results presented here, then, is on similar footing with this range of work. ⁷

The core argument

Why is polarized disagreement on orthogonal issues an epistemological problem for the partisan? In general, mere recognized disagreement need not put any rational pressure on members of an epistemic group to reduce their confidence in the propositions regarding which the disagreement exists. Often, we are justified in believing p even when we know that there are people who believe not-p. The fact that there are some people who believe that the earth is flat need not put pressure on your belief that the earth is round.

The same lesson can apply to a case involving multiple, independent issues. Suppose, for example, the general population is quizzed on a set of orthogonal and difficult multiple-choice physics questions, testing knowledge of Lagrangian mechanics, solid-state physics, general relativity, and so on. We would expect the physicists to get most of these questions right – and the mere fact that the laypeople put down different answers need not put any rational pressure on the physicists to reduce confidence in their answers. For, physicists are experts with respect to physics. Similarly, can a partisan claim that her side is composed of the experts, as it were, in matters of political issues, and thus justified in dismissing the disagreement?

There is, however, a key difference between the physics case described above and the case of political polarization, which makes this response challenging to sustain. The difference is that in the physics case we expect to see one cluster of opinion, not two. The physicists’ answers would cluster around the truth, whereas the laypeople’s answers would be all over the place. The laypeople’s answers, we would expect, amount to guesswork, given that the quiz is difficult. Another way to put it is that whereas the physicists are reliable with regard to questions of physics, the laypeople are unreliable with respect to difficult questions of physics. If the quiz featured only two options for each question, you’d expect the statistically average layperson to get about 50 percent of them right. Such a person would just be doing guesswork.

However, in the political case, things are different. There, if one side is reliable, then the other side must be anti-reliable, as opposed to merely unreliable. Since the two sides disagree with respect to a host of political issues, one side’s getting it consistently right entails that the other side is getting things consistently wrong. The other side reliably tracks the false with respect to political issues on which there is partisan disagreement. The point here is not that they’re anti-reliable with respect to a particular constrained domain, say the ethics of abortion. Rather, they succeed in consistently getting the wrong answer with respect to a large domain of rationally separable political questions!

It is easy to think of a situation where someone is anti-reliable with respect to some questions because of a core error. Thus, suppose an engineer is designing a spaceship to land on the moon. Suppose she mistakenly thinks that the gravitational acceleration at the moon is 2.62 m/s², but actually the value is 1.62 m/s². Here, her calculations regarding the optimal trajectory and so on will be systematically wrong because of a core false belief.

Analogously, in the realm of politics, suppose Bob, who self-describes as a libertarian, believes that a just society is one where the state is a ‘night watchman’. In other words, Bob thinks that the state is only justified in taxing individuals to provide for a police force, a court system, and some very limited number of public goods but not for various social welfare programs. Let us suppose that as a matter of fact, this kind of libertarian view is false. That is, the state is indeed justified in spending money on social welfare. We can then expect Bob to be anti-reliable with respect to the following kinds of questions, in the sense that he will consistently get them wrong. (a) How much money should the state allocate for health care? (b) Should there be government-backed unemployment insurance programs? (c) Ought the state provide subsidized housing for low-income individuals? Bob’s mistaken belief in libertarianism leads him systematically astray on questions having to do with the proper amounts of government spending. However, note that these issues are not orthogonal – they are related.

But what kind of belief forming method(s) would lead a group to get things consistently wrong on a set of orthogonal issues? What method(s) of forming beliefs would lead someone to get the wrong answer on each of the issues listed in the ‘Introduction’ section? I want to suggest that it is implausible that a psychologically realistic method of forming beliefs can be reliably falsehood tracking with respect to such a set of orthogonal issues. Therefore, the partisan with respect to such orthogonal issues on either of the two prominent sides ought to moderate her political beliefs.

It is important to stress here that the extent to which there is epistemic pressure on the partisan depends on how anti-reliable she is committed to saying the other side is. If she is committed only to thinking that the other side is slightly anti-reliable, the pressure to moderate will be very minimal. This is, however, not the case in modern American politics. Extensive research shows that people update their beliefs in ways that assume that their party is extremely reliable on policy matters. Moreover, work in this area suggests that the influence of party positions exhibits a strong influence over and above the individuals’ independent take on policy matters. ⁸

I want to grant in this article that this is not epistemically problematic if one’s party is in fact reliable and one has the required access to this reliability. ⁹ The thought here would be that the group’s being reliable in this way transmits the required positive epistemic properties to the political beliefs held by an individual within the group, so long as the individual has adequate epistemic access to the fact that the group is reliable. The group might be treated as an epistemic oracle, whose testimony can be properly relied upon. But even this way of justifying partisan’s beliefs will fail absent a story for why the other side is anti-reliable.

I claim that the epistemic challenge presented here applies to partisans located in one or the other cluster of opinion on polarizing issues. But, it might be thought, perhaps the epistemic pressure to moderate only applies to those who appreciate the force of the argument presented in this article. If that’s right, then the challenge will be much more modest and circumscribed than advertised earlier.

However, this interpretation underestimates the force of the epistemic challenge. Consider the following analogy. Suppose a friend of yours believes his 2-year-old is the cutest in the world. You point out to him that such beliefs are common among parents, and given his own general epistemic commitments, there is no reason to think he is perceptually special in this regard. Thus, from a purely epistemic perspective, he shouldn’t think that his 2-year-old is the cutest in the world. Now despite the fact that you ran through a bit of epistemological reasoning with him, it isn’t as if he faces the epistemic pressure to moderate only after you had this discussion. If he knew beforehand that this is a common sentiment among parents, he faced the epistemic pressure all along. (And he would know this unless he had been living under a rock.) You did not present him with new empirical evidence; rather you helped him think through what his broader epistemological presuppositions commit him to. And if he isn’t convinced by your reasoning, no matter – he faces the epistemic pressure regardless. Similarly, a partisan who knows which political issues form the subject of partisan disagreement faces epistemic pressure to moderate her positions regardless of whether she reads this article or is convinced by its reasoning, if the argument presented here is broadly correct. ¹⁰

Now, a case with a similar structure to the physics test example above seems to arise with respect to economics. As Caplan (2008) explains, the general public systematically disagrees with economists on a range of issues involving trade, profits, corporations, and employment. So, why think group anti-reliability on orthogonal issues raises some special epistemic problem? There are two possible dis-analogies with this case and the political case, however. First, there tends to be a high variance among the answers given by the public, so it is less of a cluster of opinion. ¹¹ Second, there can be plausible evolutionary explanations of what the biases are and how the general public would have come to possess them. Of particular relevance here will be the fact that our economic intuitions would have been shaped by the small group settings in which our ancestors lived. ¹²

Nonetheless, I do not want to claim that such cases are not possible. Rather, my more modest claim is that when there is purported anti-reliability on orthogonal matters from a large group of individuals, some further explanation is called for. The reasoning behind this is that while it is easy to understand how a person or group could be anti-reliable about some particular domain if they make a core mistake within that domain, it is hard to see how they could be anti-reliable with respect to a variety of orthogonal claims about different subject matters. This is of course not to say that such a thing cannot happen – indeed, there may be possible scenarios in which, owing to a variety of mistaken assumptions, one side gets it wrong whenever there is disagreement. One side might just turn out to be unlucky or may have a number of biases that lead them astray. The point here is just that the alternative explanation – according to which neither side has a monopoly on truth and partisans hold the views they do owing to non-truth tracking mechanisms is more plausible. The burden is shifted onto the defender of the partisan, I claim, to show how one side can be anti-reliable with respect to diverse orthogonal claims. The next section looks at possible strategies that may be used to discharge this burden.

Now, it might be thought that the problem only arises because I have assumed that political positions of the sort that elicit partisan disagreement are truth-apt. It only makes sense to talk of ‘reliability’ or ‘truth-tracking’ if this is right. But consider for instance the sentence ‘Homosexual couples should be permitted to marry’. Is this really the sort of sentence that can be true or false? Perhaps what’s really going on when we agree or disagree with a sentence like this is that we’re expressing some pro- or con-attitude, much like the way we cheer or boo for sports teams. We can’t track the truth with regard to the contents of such utterances, because there’s no truth to be had.

This view of moral utterances, however, has few contemporary defenders. The view was defended by Ayer (1936), among others, but famously ran into the Frege–Geach problem. Since then, philosophical descendants of the view, namely, the various kinds of expressivism, have taken on board the idea that moral statements can be true or false. ¹³ So, even if expressivism is the correct theory of metaethics, this will not blunt the force of the argument presented here. Furthermore, partisans themselves who think there is a truth of the matter on these issues are thereby committed to taking reliability and truth-tracking seriously. And, it is plausible to think that many people’s political views are like this – that is, they believe their opponents are wrong as opposed to merely in possession of different but rationally non-criticizable preferences.

Possible responses

In responding to the argument presented above, it is not going to be plausible to say that one side tracks moral falsehood tout court. ¹⁴ Moreover, neither side appears to be more selfish or less altruistic in general. Margolis and Sances (2013), for example, find that when the appropriate factors are controlled for, liberals and conservatives are equally charitable in terms of donations.

Rather, the more restricted and viable claim that is worth defending on behalf of the partisan is that the opposing side is anti-reliable with respect to political propositions on which there is partisan disagreement. One way in which such anti-reliability can arise is if there is a general cognitive or intellectual shortcoming that can be shown to be more prevalent on one side. One possibility is to see if an outlook of tolerance is more prevalent on one side as opposed to the other. Recent empirical literature, however, casts significant doubt on this claim. ¹⁵ A second possible proposal might be to see whether the mean individual on one side possesses more by way of intelligence. Even if this were true though, lack of intelligence should at best predict unreliability rather than anti-reliability. At any rate, empirical research does not bear out a significant intelligence advantage for either side. Rather, in general, it turns out that greater cognitive ability predicts greater social liberalism but greater economic conservatism. ¹⁶ What if it is not general intelligence but rather rational inference about political issues that separates the two camps? Recent work by Gampa et al. (2019: 1075) has tried to test this hypothesis. They find an equal tendency among liberal and conservatives to evaluate the logical validity of arguments in a way that supports their antecedently held beliefs, ‘leading to predictable patterns of logical errors’.

Another possibility is to see whether there is a systematic bias against expertise that afflicts one side rather than the other. If identifiable experts are reliable sources of information about a variety of policy topics, and if one side exhibits distrust of expertise per se then that side will be led systematically astray. Owing to their distrust of reliable sources of information, they become anti-reliable not just unreliable. Given rising skepticism among conservatives about the value of universities, and about expert knowledge on issues such as climate change, one might argue that this renders them anti-reliable about policy matters on which there is partisan disagreement. ¹⁷

In order for such a strategy to succeed, however, several challenges must be met. First, it is not empirically borne out that conservatives are skeptical of scientific expertise on empirical matters tout court. With regard to economic policy, for example, which is the source of many partisan disagreements, conservatives’ views often align with economists more so than those of people on the left side of the spectrum. ¹⁸ Relatedly, there are some pseudoscientific enterprises, such as astrology, that are seen as legitimate by greater percentages of self-identified liberals. ¹⁹ Second, formulating policy proposals based on the results of academic research requires interpretation of those results. Recent work described in Kahan et al. (2013) finds that both conservatives and liberals are more or less equally prone to interpret data in a way that supports their antecedently held political beliefs even when it does not. Interestingly, the authors find that people use their numeracy selectively; they seem to rely upon biased heuristics to incorrectly interpret data when doing so supports their antecedently held political views but use more careful ‘System 2’ reasoning to draw accurate conclusions in other contexts. In this way, polarization tends to increase with greater numeracy, the authors find. Third, in areas of discussion that are politically charged, research groups may have their own biases if they are sufficiently ideologically homogenous. This can undermine the quality of such research via publication bias and the acceptance of certain conclusions without sufficient evidence, among other things. ²⁰ Lack of trust of supposed experts in such areas, then, need not be epistemically bad.

Even if the core relevant difference is not a generalized lack of trust in experts, perhaps there are other, deeper and more fundamental psychological differences between the two camps. That is, members of a particular side may tend to possess more by way of certain psychological propensities on average. Moreover, it may be argued the possession of those propensities can explain reliable moral error when it comes to politics. For instance, if it turns out that confirmation bias affects one side more than the other, the case for that side being anti-reliable with respect to politics can be made. ²¹ There is a large literature examining the questions of psychological differences between conservatives and liberals. The question that arises in the light of this research is whether the psychological differences are of the right kind to ground a response to the worry sketched in this article.

One popular characterization of the difference is in terms of negativity bias. Hibbing et al. (2014), for instance, argue that liberals and conservatives show different levels of negativity bias. The authors note that

humans generally tend to respond more strongly, to be more attentive, and to give more weight to negative elements of their environment…Good evolutionary reasons exist for negativity bias given that negative events can be much costlier in fitness terms than positive events are beneficial; to state the obvious, infection, injury, and death curtail reproductive opportunities. (Hibbing et al., 2014: 303)

Yet, even if the authors’ hypothesis is true, it does not obviously entail that one side is systematically in error. The epistemic import of negativity bias is different from confirmation bias in important ways. In general, it seems that different levels of negativity bias will be optimal in different contexts, for both individuals and societies. To the extent that the world is a dangerous place, negativity bias can be beneficial because it will mean that individuals and societies pay more attention to the dangers that in fact threaten them. On the other hand, to the extent that the world is not a dangerous place, excessive negativity bias may lead to individuals and societies taking unnecessary precautions and will thus be costly in that way.

So, supposing that differences in negativity bias are in fact the core explanatory difference between liberals and conservatives, we still do not have any reason to think that one side is getting things reliably right. To make this claim, we would have to establish what level of negativity bias is appropriate in the current social and political context, and then argue that one side’s adherents have the right amount of negativity bias on average. Alternatively, the partisan could claim that the average amount of negativity bias manifested among individuals on her side is closer to the optimal level than the average amount manifested on the other side. Perhaps this task can be accomplished, and the partisan of a particular side vindicated. Nonetheless, this project is bound to be more difficult than the analogous project built upon confirmation bias.

Along similar lines, one might use the Moral Foundations Theory, as set out in the work of Haidt (2012), to defuse the epistemic challenge set out earlier. Haidt argues that there are five core moral foundations, and that liberals and conservatives possess different sets of these foundations. Think of moral foundations as moral taste buds; different moral foundations deal with different sorts of considerations, and they are activated by different sorts of inputs.

The five foundations, according to Haidt (2012: 178–179), are as follows:

The Care/harm foundation, which ‘makes us sensitive to signs of suffering and need; it makes us despise cruelty and want to care for those who are suffering’.

Haidt argues, based on survey data, that liberals (or more accurately, people who self-identify as liberals) are fairly responsive to triggers involving (a) and (b), but much less so with respect to (c), (d), and (e). Conservatives, on the other hand, are grounded in all five foundations to roughly the same extent. Further, they are (very) slightly less responsive to triggers involving (a) and (b) as compared to liberals.

These findings, if true, can form the basis of a promising strategy to show that how the other side can be anti-reliable, and not just unreliable with respect to the issues on which there is partisan disagreement. For instance, the liberal side could respond to the epistemic challenge by claiming that the Loyalty, Authority, and Sanctity foundations are sources of moral error. Sure, they played an important role in our evolutionary history, perhaps through group selection, but they lead people to make moral judgments in non-truth-tracking ways.

Importantly, the argument goes through even if it turns out that these three foundations tend to make individuals unreliable with respect to various moral truths. For, if liberals as a group are reliable with respect to political truths whereas conservatives as a group are unreliable, then it will turn out that conservatives are anti-reliable with respect to those issues with respect to which there is partisan disagreement.

Notice how this picture is importantly different from the physics exam example discussed earlier. In the physics exam case, we’d expect the physicists to produce answers that cluster together, whereas the laypeople’s answers will be all over the place. Hence, we’d see one cluster rather than two. If the argument above is right, however, the conservatives are not like the laypeople in the physics exam case. They’re not picking their political views at random, as if throwing darts at a board. Rather, the Loyalty, Authority, and Sanctity foundations lead them to converge on particular kinds of views with respect to politics. For instance, the Sanctity foundation may underlie conservative views on abortion or gay marriage – fetuses and the institution of marriage being treated as ‘sacred’ in this case. The Loyalty foundation may underpin nationalistic sentiment, which in turn can drive people toward policies of tougher border control as well as more restrictive views on immigration.

The presence of these foundations, then, can make it the case that a group of people is systematically attracted to a particular set of political views or attitudes. Nonetheless, overall, it would still be true that conservatives are unreliable with respect to morality, rather than anti-reliable. This is because they also have the Care and Fairness foundations. These two foundations will typically lead them to believe moral truths. For instance, they might rightly think that there are moral reasons to reduce poverty where possible. In this way, what we end up with is a model that differs from the Manichean angels versus the devils model described at the beginning of this section. According to the present explanatory strategy, the opposing side is not morally mistaken tout court – they do get some things right, as is overwhelmingly plausible. However, on the issues on which there is partisan disagreement, liberals have the upper hand. This is because they do not have high levels of the Loyalty, Authority, and Sanctity foundations, which are distorting influences on individuals’ capacities to form the correct moral and political beliefs. Hence, on those issues on which there is partisan disagreement, liberals are reliable, whereas conservatives are anti-reliable.

This is how the argument might look from the liberal side. The conservative side can claim that the Loyalty, Authority, and Sanctity foundations, far from leading people astray, help them to latch on to important moral truths. Liberals, because they lack these moral taste buds, so to speak, are unreliable with respect to issues that trigger these foundations. Therefore, with respect to issues on which there is partisan disagreement, conservatives are reliable, whereas liberals are anti-reliable. This is not to say the situation would be symmetric from an epistemic point of view; rather, one side’s beliefs could be vindicated with a substantive defense of a particular set of moral foundations as the best guide to moral truth. Engaging in this project, however, is well beyond the scope of this article.

While I think this is the most promising way to rebut the epistemic challenge presented here, I want to raise here some worries about Haidt’s methodology, which might undermine the style of response sketched above. The main worry is that the Moral Foundations Questionnaire (MFQ) might be triggering the Loyalty, Authority, and Sanctity foundations of conservatives in particular. ²³ If that’s right, then the research doesn’t so much show that liberals lack these foundations. Rather it leaves open the possibility that liberals also have these three foundations, but that their triggers are different.

For instance, one of the prompts in the MFQ, to which respondents are asked whether they agree or disagree on a seven-point scale, is: ‘Chastity is an important and valuable virtue’. A respondent’s agreeing with this claim is taken to show that she is motivated by the Sanctity foundation. It is then shown that people with strong Sanctity foundations tend to self-identify as conservative.

But there is a worry of circularity here. Since chastity is a value that is prominent among social conservatives, it is no surprise that people who identify as conservative will tend to agree with the prompt. Yet this doesn’t establish that liberals do not have a robust Sanctity foundation. For, the foundation may be triggered by other things. Haidt (personal communication) has acknowledged that there may be triggers for the liberal Sanctity foundation – for instance, having to do with the environment, or with wasting food. Similar worries arise with the methods by which Loyalty and Authority foundations are assessed. ²⁴

Perhaps all these worries can be adequately addressed. It may ultimately turn out that differences between the moral psychology of the average conservative and the average liberal can explain how they will systematically disagree on certain types of issues, thus allowing either side to explain why the other is anti-reliable. However, I still take the argument of this article to have established an interesting claim. The claim is that whether partisans can avoid the epistemic challenge depends on whether the kind of strategy described above can succeed. That is, the question depends on whether we can identify psychological differences between conservatives and liberals that can plausibly ground an explanation of why one side is anti-reliable with respect to the issues of partisan disagreement.

A separate style of response by the partisan might go like this. Naturally, powerful special interests have an incentive to build a broad coalition of voters. My political opponents thus have an incentive to manipulate voters by artificially linking together a variety of false political positions, in order to serve their real social or economic interests. This explains why they could be wrong on a host of rationally disconnected, but contingently politically connected, issues.

The problem with this response is that the situation is symmetrical – in a two-party system, both sides will have an interest in building a broad coalition. ²⁵ And if the issues up for grabs are orthogonal, then the coalitions will disagree on a variety of orthogonal issues. In this case, the symmetry is problematic for the partisan. For, what is her evidence that the other side is systematically wrong? Simply stating her arguments for her first order-views would be dialectically inappropriate. It would be like your friend saying, ‘But just look at my two-year-old. Of course he’s the cutest in the world!’

Are the issues really orthogonal?

The argument of this article is practically irrelevant if it turns out that the issues with regard to which partisans disagree are not really orthogonal. It would then only present an epistemic worry for partisans in hypothetical, non-actualized scenarios. But can we think of the claims laid out in the ‘Introduction’ section as being anything but orthogonal?

Let us first focus on the ‘racial/economic’ set of issues. Consider, for example, Immigration, Wages, and Affirmative Action. One possible answer that is worth considering is the following. Perhaps liberals and conservatives assign a different importance to helping those who are the worst off. Liberals assign a greater importance than conservatives to helping the neediest. Immigration, Wages, and Affirmative Action are thus not orthogonal – they all flow from a basic principle according to which it is especially morally important to help the worst off. So, for instance, the Priority View, defended in the work of Parfit (1997) is an example of a moral theory that could fill this role. According to the Priority View, benefits to individuals matter more morally the less well-off they are. A conservative partisan could enlist this putative observation as a way to explain the anti-reliability of the liberal side; she could claim that since the Priority View is false, and that the views which are to be found within the liberal cluster of opinion are applications of the Priority View given the empirical facts, liberals are prone to systematic error with respect to political issues. And, conversely, the liberal may contend that the conservatives’ anti-reliability is to be explained by their nonacceptance of the Priority View.

It is not obvious, however, that this strategy can succeed. The extent to which the minimum wage helps the poorest is a contentious issue in empirical economics. According to a long-standing view, increases in minimum wage also increase unemployment (Stigler, 1946). Some economists think that other positive features of the minimum wage can offset this drawback. Other economists disagree. ²⁶ Given this disagreement, it is not obvious that someone who holds the Priority View is committed to supporting the minimum wage. Similarly, there is disagreement about the extent to which race-based affirmative action policies help the worst-off members of society. Rachels (1978), for example, gives a philosophical account of how race-based affirmative action can help disadvantaged groups. Recently, Sander and Taylor (2012) have argued that such admissions preferences actually harm those students they are intended to help.

The broader point here is that many of the issues with respect to which the political camps disagree rest on contentious empirical claims. Hence, a high-level moral principle or theory – like the Priority View or utilitarianism – is not going to by itself yield Immigration, Wages, and Affirmative Action as corollaries. You can be a prioritarian and take either the liberal or conservative line on the three issues. For instance, you can take the conservative line if you think something like the following. Minimum wage laws make it harder for people without specialized skills to get jobs since employers respond by trying to automate the relevant tasks or shut down altogether, depending on the location and industry. Hence, in order to help the worst-off members of society, we should avoid having minimum wage laws. Second, as Sander and Taylor (2012) claim, race-based affirmative action policies harm the disadvantaged students they are aimed to help. So, we should scrap these policies. Third, the US should prioritize immigration from the poorest countries because that would constitute helping the worst-off. If the US is going to have a limit on the numbers of immigrants it accepts, and if the demand for spots is greater than the supply, the US should prioritize immigration from, say, a country like Ethiopia, which is much poorer than Mexico (historically the major origin of undocumented immigrants in the US).

I’m not saying this line of reasoning is correct or that the average conservative believes these things. I am also not denying that as a matter of fact, the Priority View may be more common among liberals. All I want to show is that the Priority View, supplemented with particular empirical assumptions that aren’t unreasonable and are up for debate, can lead to the conservative line on racial/economic issues. Moreover, since the different empirical debates are presumably orthogonal, the issues themselves are themselves orthogonal – a particular answer on Immigration does not commit you to a particular answer on Wages or Affirmative Action. Therefore, the issues remain rationally orthogonal even if one adopts the Priority View, and even if as a matter of fact, most people who adopt that view fall within one cluster with regard to the issues under dispute.

On the conservative side, one guiding principle that is often said to underlie opinions on a variety of issues is the principle of ‘small government’. One way to frame the idea is that government regulations and spending ought to be kept at the minimum possible level necessary to achieve law and order, as well as important public goods. Framing the issue this way, however, isn’t quite precise enough for our purposes, for a lot hangs on what the important public goods are – and presumably conservatives and liberals will disagree to some extent as to what these goods are. Indeed, there seem to be public goods that conservatives seem to want to spend more on – for example, border security. In addition, the principle of small government seems neutral or antithetical (depending on who you ask) to social issues like Abortion and Marriage.

The social issues themselves also seem prima facie orthogonal. A particular position on Abortion doesn’t commit you to any position on Marriage. However, the plausible and commonly held view is that religiosity in part explains the adoption of conservative social views with respect to issues like these. Thus, perhaps Judeo-Christian principles unite the conservative side on social issues? If that’s right, then those issues may not be orthogonal after all.

This suggests one possible way for secular liberals to proceed. Since Abortion and Marriage, and related issues, are not orthogonal, there is a way to explain how people might reliably come to wrong conclusions in this area – namely, by possessing and applying a particular interpretation of Abrahamic religion. Hence, these issues can be bracketed off in understanding the epistemic implications of political polarization. A left-liberal partisan then, might only face epistemic pressure to moderate her confidence on the non-religiously driven issues about which there is partisan disagreement.

The religious conservative can do the exact opposite – moderate his views on those same issues, while holding steadfast to his conservative views on Abortion and Marriage. From his perspective, the other side is anti-reliable with respect to such issues because they fail to correctly interpret and believe certain religious teachings. Nonetheless, given the nature of polarization on the other issues, he still faces pressure to moderate on the other propositions within the set, on which he has a conservative view. Like before, the claim is not that there is an epistemic symmetry or standoff here. Perhaps the conservative views are reliably wrong due to an unjustified set of religious beliefs. Rather the claim is just that the epistemic problem presented in this article is avoided for social issues in this way.

However, it is important to note the limitations of such an approach for defending a general strategy of partisan vindication. Indeed, it might be tempting to say the following: ‘We found a principle behind Abortion and Marriage and were able to bracket those issues off. Why not repeat the procedure for the other issues?’ Perhaps the procedure may have to be repeated several times – at maximum seven times for each remaining issue – but in the end, the partisan will have bracketed off all the issues and thus will face no epistemic pressure to moderate. The problem, however, is that this simply kicks the problem up one level. Suppose it turns out that five distinct principles underlie the nine issues at hand. The question then will be, what explains why the other side is reliably wrong about those five principles?

Nonetheless, such an exercise is not pointless. For, the fewer orthogonal principles that are necessary to capture partisan disagreement, the less bite the epistemic problem presented in this article will have. The limiting case is one – if one principle underlies all the positions on which there is partisan disagreement, then there is no particular epistemic problem raised by polarization over and above general worries about disagreement and irrelevant influences on belief formation. ²⁷

Higher order evidence

If the argument presented in this article is correct, then the partisan has higher order evidence that some of her first-order political beliefs are mistaken (or alternatively, some of her credences are inaccurate). One possible view, however, is that higher order evidence counts for nothing. ²⁸ If the partisan has in fact properly assessed the relevant first-order evidence, then she is justified in sticking to her guns. What matters is whether her political beliefs are actually a good response to her relevant evidence.

However, it is implausible that higher order evidence never matters. Suppose you realize that you were brainwashed to believe that New York City is the capital of New York. Learning this should make you reduce your confidence in that proposition. Or suppose that you participate in an experiment where you are given some statistical evidence about the effectiveness of pet therapy, regarding which all the participants, including you, had no opinion of beforehand. After the experiment, you are told that most of the people who were primed in a certain way ended up believing that pet therapy is effective, while most of the people who weren’t primed in that way ended up believing that pet therapy is not effective. They don’t say whether you were primed. Suppose you came out believing that pet therapy is effective. It behooves you to reduce your confidence in this – it is no good to simply say, ‘But I’ve seen the evidence for pet therapy, and it’s just so compelling!’ ²⁹ Importantly, this is so even if, as a matter of fact, the effectiveness of pet therapy actually is actually supported by the statistical evidence that was presented. Thus, there are cases where higher order evidence puts rational pressure on agents to reduce confidence in their beliefs. Of course, not all cases of higher order evidence against P ought to budge an agent’s belief in P. For, such evidence may be defeated by other pieces of evidence. You might learn that your brainwashers were incompetent or that the pet therapy study used a small N, for example. But what I have tried to show in the previous section is that such defeater(s) are hard to come by in the partisanship case.

While my argument relies upon the import of higher order evidence, it is not merely a special instance of the well-known peer disagreement debate in epistemology. This literature explores the question of what the appropriate response is when an epistemic peer disagrees with you. ³⁰ According to Richard Feldman, an epistemic peer is: ‘another person, every bit as sensible, serious, and careful as oneself, has reviewed the same information as oneself and has come to a contrary conclusion to one’s own’ (2006: 235). The question then is: how should one modify one’s beliefs/credences when faced with such a disagreeing peer?

Is the skeptical argument I have been sketching in this article just a special case of peer disagreement in general? No, for two main reasons. First, within the context of the peer disagreement debate, the epistemic peer is one who is stipulated to have the same epistemic qualities as you, as brought out in the Feldman quote. (The more we depart, in the negative direction, from Feldman’s stipulation, the weaker is the pressure to conciliate. If you know I’m terrible at arithmetic, then you face little pressure to conciliate upon our discovering that we came up with different totals.) But I rely on no such assumptions. Moreover, the problem a partisan faces with respect to political polarization is not merely that there may be some individual (who has the relevant epistemic gifts) who disagrees with her. Rather, the problem she faces has to do with two groups clustering around particular answers to orthogonal political claims – the challenge is to give an explanation of how such a pattern can arise, without undermining one’s own political beliefs.

In fact, if we restrict the problem to disagreement between individuals, it is not obvious that the partisan faces much (if any) pressure to reduce credence in her political beliefs – even from the perspective of a defender of the ‘equal weight’ view. This is because from the partisan’s point of view, a person who disagrees sufficiently with her political views may not count as an epistemic peer in the first place. For, as Elga (2007) has discussed in detail, someone is an epistemic peer for you with respect to p if, setting aside your reasoning about p that person is just as likely to get things right. But if p is some political issue on which there is partisan disagreement, then the partisan’s setting aside her reasoning about p will not involve setting aside her reasoning about a host of other issues on which, by her lights, the opposing partisan counterpart is wrong.

The second key point is that an epistemic peer, within the context of the peer disagreement literature, is stipulated to have reviewed the same information as you, in Feldman’s words. If I consider you my epistemic equal, in the sense of possessing the same epistemic virtues, but realize that you have some additional relevant information, there is no reason why I would not rationally defer to you with respect to a particular case. Thus, suppose you have seen the day’s weather report but I haven’t – it is obvious that in such a scenario I should rely upon your testimony as to whether it will rain later today. There is no special problem here.

However, I have not stipulated that representative members of the two political camps possess similar sets of information, in the context of my argument. In fact, a natural and overwhelmingly plausible partial explanation of why the two camps are as polarized as they are would be that the two sides are exposed to differing sets of information. Liberals and conservatives tend to have different social circles and consume different kinds of political news. ³¹ As a result, the kind of information that the conservative partisan is exposed to (in the typical case) is bound to be different from the sorts of evidence that her liberal counterpart is exposed to.

The core problem that the phenomenon of polarization on orthogonal issues raises is that of accounting for a pattern of belief (or credence) distribution. The problem for the partisan is not simply that someone disagrees with her. The problem is that political beliefs are distributed across the population in such a way as to make it highly unlikely that the partisan’s beliefs are all or mostly true.

Why and how to moderate

The epistemic challenge presented here applies to an individual depending on the extent to which her views fall within one of the clusters of opinion. To the extent that an individual’s political beliefs do not fall within a cluster, she is immune to the worry presented in this article. Hence, someone who has liberal views on the first five, but conservative views on last four issues presented in the first section, for example, avoids the problem presented here, though she may well have irrational political beliefs for other reasons.

The extent to which partisans face epistemic pressure to moderate their political beliefs, after noticing that polarization exists across orthogonal issues, depends on a variety of factors. In general, the stronger the clustering effects and the farther apart the clusters, the greater the rational pressure is to moderate. The larger the number of orthogonal issues, the greater is the rational pressure as well.

The epistemic concern described in this article is merely one among many that can apply to individuals’ belief formation. Avoiding the problem raised here doesn’t mean one is thereby rational in forming political beliefs – one may be irrational for altogether different reasons. Suppose Bill decides to form his political opinions based upon reading tea leaves. Consequently, he ends up having four beliefs typical of the conservative side and five typical of the liberal side. He thus evades the epistemic worry presented in this article. But of course, this doesn’t mean he has formed his political beliefs in a rational way. Further, the challenged raised in this article doesn’t apply to libertarians or Marxists or others whose views on political issues might plausibly be derived from a core principle. Nonetheless, of course, one can hold these views irrationally – by believing them simply because a charismatic orator professes them, for example.

If the argument presented in this article is correct, how should a partisan modify her beliefs? To answer this question, it is helpful to note who the epistemic worry applies to. It applies, paradigmatically, to people who have all the opinions of one side or the other and are very confident in them. The challenge applies less and less the more one moves away from this paradigm case. Thus, someone who has all the views on the conservative side but does not place high credence in them is more immune to the worry. In addition, someone who has views like Bill above – a combination of both sides to roughly equal proportion – is immune to the worry.

This suggests two main ways for the partisan to moderate. One way is to recognize the higher order evidence, and thereby become less confident about her political views. The other way is to come to acquire some of the views typical of the opposing side. How to do this latter thing? It would seem irrational for someone to acknowledge the higher order evidence presented in this article and then simply choose four or five of her beliefs to switch to the other side. Indeed, doing this doesn’t seem too different from what Bill is doing above when he reads tea leaves!

However, there is another option. As mentioned earlier, a key reason for partisan polarization is being exposed to different sets of evidence. If one only reads perspectives and sources that are friendly to one’s side, then likely, and naturally, one’s views will match the positions of that side. Furthermore, this can be a perfectly rational response to one’s first-order evidence. ³² Indeed, I can be rational in believing orthogonal propositions P₁, P₂…P _n and you can be rational in believing ¬P₁, ¬P₂…¬P _n if we’re each exposed to different sets of evidence. Partisan polarization can (but need not) be a rational response to the first-order evidence individuals have. Yet despite this, it can be true that the ways in which we acquire that first-order evidence is biased. It is well-known that people tend to seek out evidence that confirms their antecedent beliefs while avoiding disconfirming evidence. ³³

Thus, one way to proceed is to seek out the kinds of evidence that partisans of the opposing side typically possess, or alternatively, the kinds of evidence that would support the policy positions of the opposing side. Thus, for example, if someone has only read papers and news articles showing the adverse effects of minimum wage, they might seek out research or news articles challenging this view. This may well lead them to eventually become friendlier to the idea of a minimum wage. In this way, they can change their opinion as a result of acquiring relevant first-order evidence. ³⁴ What the higher order evidence suggests is that the first-order evidence held by the partisan is likely skewed. Remedying this will involve seeking out and engaging with the best arguments for the political convictions of the other side.

The two ways of moderating described above seem to be only part of the story, however. This is because it seems that another dimension along which one can moderate one’s political positions is to hold less extreme views. For example, believing that the minimum wage ought to be US$12 per hour is a less extreme view than believing that it ought to be US$15 per hour. Similarly, believing that abortion is seriously immoral after 20 weeks of pregnancy is less extreme than believing that abortion is seriously immoral at any point in the pregnancy.

Finally, it should be noted that the pressure brought out here is merely epistemic. I do not mean to be arguing against partisanship or political parties per se. Partisanship can have practical benefits including, but not limited to, promoting a sense of solidarity and community, facilitating engagement in long-term political projects and commitments, and helping to sustain motivation. ³⁵ Political partisanship may also have indirect epistemic benefits such as what White and Ypi (2016) call peer empowerment and hermeneutic resilience. In the same way, thinking that your 2-year-old is the cutest in the world may have practical benefits – it may help you to stay motivated to perform the challenging duties of parenthood. It may also have indirect epistemic benefits; you might pay more attention to your child if you have this belief, and this might in turn help you to better discover his needs. Yet the epistemic pressure remains nonetheless; given that many parents think this way and you have no reason to think you’re in a special perceptual position, you face pressure to reduce confidence in your belief that your 2-year-old is the cutest. Analogous points hold for the partisan.

Conclusion

The phenomenon of belief polarization across a population raises the prima facie worry that social and historical factors, as opposed to truth-tracking belief forming mechanisms, explain at least some of the predominant beliefs among the groups in question. The natural explanation for why a typical conservative or liberal holds the political views she does would appeal to facts about her social circles and, relatedly, the sorts of news outlets she pays attention to. Against the backdrop of this bogeyman, is there a possibility that a partisan can stand her epistemic ground?

The problem is not one of mere disagreement; it is not simply that there are people who disagree with the political views of the partisan. Mere disagreement of this kind is not sufficient to undermine justification for one’s beliefs. A biologist does not face rational pressure to reduce her confidence in the theory of evolution just because there are (many) people who disagree.

The case with political polarization, I have argued, is importantly different. If the partisan is to be vindicated, it must be the case that the opposing side is anti-reliable with respect to a host of orthogonal issues. I have argued here that an explanation of such anti-reliability is hard to come by. If the argument of this article is successful, then partisans – whose political opinions sufficiently toe the line with respect to the two main ideological clusters within American politics – face epistemic pressure to moderate their political beliefs. The conclusion holds even if there turn out to be more than two ideological clusters – so long as the number of issues on which there is partisan disagreement is sufficiently large. Moreover, while I have focused on the American context, the problem is perfectly generalizable. The same arguments apply to different political settings so long as there are orthogonal issues on which there is disagreement, and the disagreeing camps exhibit clustering effects.