1 Introduction

How should we understand the psychology of political cognition? Specifically, how do citizens think and reason about those aspects of the world—the economy, the behaviour of politicians, the state of society, and so on—that are relevant to democratic decision making? According to one influential theory, citizens are frequently biased in how they think about politics by partisanship. That is, support for political parties motivates people to embrace biased beliefs—typically beliefs that reflect favourably on such parties—and so powerfully distorts the processes by which they seek out and process political information (Achen & Bartels, 2016; Brennan, 2016; Campbell et al. 1960; Kahan, 2017; Mason, 2018). Such motivated irrationality is then thought to underlie or at least exacerbate numerous social and political problems, including various forms of political polarisation (Finkel, et al., 2020; Marietta & Barker, 2019; Mason, 2018) and costly group-based misinformation (Ditto, Liu, et al., 2019; Kahan, 2017; Williams, 2021b). I will henceforth refer to this view as the Partisan Bias Hypothesis.

The Partisan Bias Hypothesis is extremely influential (Achen & Bartels, 2016; Bisgaard, 2019; Ditto, et al., 2019a; Federico & Malka, 4.2 Experimental findings

There is a large experimental literature that purports to demonstrate that what people are motivated to believe based on their partisan identities influences how they interpret, evaluate, and process information (Bisgaard, 2019; Ditto, et al., 2019a; Kahan, et al., 2017). For example, in a recent meta-analysis including fifty-one experiments with over eighteen thousand participants, Ditto et al. (2019, p.274) argue that there is strong evidence for partisan bias in how both Democrats and Republicans in the USA handle information, where they define “partisan bias” as the “general tendency for people to think or act in ways that unwittingly favor their own political group or cast their own ideologically based beliefs in a favorable light.” To probe for this bias, their meta-analysis includes studies that ask experimental subjects to evaluate what they call “matched information…that is as identical as possible in every way except that in one case it favors the participant’s political affinities… and in the other it challenges those affinities” (Ditto, Liu, et al., 2019, p.274). Given this, I will refer to such studies as employing matched-information designs.

Here are some examples of the kinds of experimental results that Ditto et al. (2019) include in their meta-analysis:

  • Influential “party cue” designs (Tappin, et al., 2020) demonstrate that participants are more likely to endorse a policy if told that their party supports it, in some cases endorsing policies that they would otherwise oppose because of their association with party endorsement (Cohen, 2003; see Finkel, et al., 2020).

  • Kahan et al. (2012) show that liberals and conservatives form different judgements about identical protest behaviour—for example, about whether the protestors obstruct and threaten pedestrians—depending on whether the protest is described as being pro-life or pro-gay rights.

  • Several experiments demonstrate that people evaluate the quality of otherwise identical studies more favourably when they support their strongly-held political views (e.g., Lord, et al., 1979; Taber and Lodge, 2006).

  • Kahan et al. (2017) find that partisans who can correctly interpret quantitative information when framed in terms of the effects of skin cream are less capable of interpreting identical quantitative information when it is framed as challenging their attitudes about gun control.

Such experimental findings thus appear to show that how people interpret, evaluate, and process information depends on the congeniality of that information to their partisan allegiances. That is, they seem to provide strong evidence for the existence of partisan motivated cognition (Ditto, Liu, et al., 2019; Kahan, et al., 2017).

Once again, critics have argued that this interpretation is misguided. The basic problem is that the conclusions that partisans are assumed to be motivated to reach in such experiments correlate with what they in fact believe (Stanovich, 2021; Tappin, et al., 2020). Thus, when you vary the implications of otherwise identical information for such desired conclusions, you also vary the background beliefs that participants bring to bear in interpreting that information. However, what one believes influences how one processes information for reasons independent of motivated cognition. Thus, the results of matched-information designs appear to be consistent with purely cognitive explanations in which beliefs are influenced by other beliefs in the absence of any motivational biases.

Consider some of the experimental results just described, for example. First, as has long been noted (Sniderman, et al., 1993), the results of party cue designs might simply reflect people’s divergent beliefs about the trustworthiness of different parties. That is, if you trust your political party’s judgement, you will defer to this judgement in the absence of motivated cognition.

Second, Kahan et al.’s (2012) finding about how partisans interpret protest behaviour might simply reflect the influence of strong prior beliefs that co-partisans are more virtuous than opposing partisans (Baron & Jost, 2019; Tappin, et al., 2020). As noted above (§ 3), it is a basic tenet of Bayesian inference that how one responds to evidence should factor in one’s prior beliefs, and strong beliefs about the relative virtues and characteristics of different kinds of protestors will inevitably colour how one perceives their behaviour.

Third, the fact that people evaluate the quality of otherwise identical studies more favourably when they support their political views is consistent with non-motivated cognition (Stanovich, 2021; Tappin, et al., 2020). Given uncertainty about the reliability of the source of information, even optimally rational agents will draw upon their prior beliefs to evaluate the reliability of that information. There are formal arguments demonstrating this (Coppock, 2023; Koehler, 1993), but the underlying idea is intuitive: scientists, for example, should be more sceptical of studies purporting to undermine well-established scientific claims. In fact, divergent prior beliefs alone can even give rise to the phenomenon of belief polarisation, which occurs when individuals with opposed beliefs increase their confidence in such beliefs—and thus polarise even further—when exposed to the same body of mixed evidence (Lord, et al., 1979; Taber & Lodge, 2006). Although this phenomenon is often taken as extremely strong evidence of motivated irrationality (e.g., Lord, et al., 1979, p.2106), Benoît and Dubra (2019) demonstrate that this intuition is mistaken: belief polarisation can occur under conditions in which people are solely motivated by accuracy, and experimental research into the phenomenon provides no reason to believe that such conditions do not obtain.

Finally, the fact that people’s strong beliefs about a topic such as gun control interfere with their ability to rationally evaluate arguments and data in tension with such beliefs need not reflect any motivational bias. Human cognition is susceptible to belief bias, a purely cognitive tendency in which people find it difficult to rationally evaluate arguments whose conclusions challenge their beliefs (Tappin, et al., 2020). Thus, people are more likely to judge otherwise identical arguments as valid (or invalid) when they agree (or disagree) with their conclusions, including when it comes to neutral conclusions such as “Roses are living things” and “Mice are insects” (Stanovich, 2021).

Given this, experimental results widely believed to demonstrate partisan motivated cognition appear to be consistent with non-motivational explanations in which, as Baron and Jost (2019, p.296) put it, “the true source of the alleged bias… [is] purely cognitive, with no motivation involved—that is, purely a case of beliefs affecting beliefs rather than desires affecting beliefs.”

In response to this argument, some proponents of the existence of partisan motivated cognition concede that the experiments are consistent with purely belief-based (i.e., cognitive) explanations but argue that the prior beliefs themselves result from motivated cognition. For example, Stanovich (2021) suggests that the results of matched-information designs often reflect motivated cognition (what he calls “myside bias”) not because of the cognitive processing within the experiments but because the divergent priors that underlie such processing reflect what the participants are motivated to believe.

The basic problem with this response is that the experiments provide no insight into the origins of the beliefs that people bring to them. Thus, Stanovich’s suggestion that such beliefs result from motivated cognition might be true, but it is not something that can be determined by the experiments. Given that the experiments are often cited as the chief source of evidence for partisan motivated cognition, it is difficult to see what is supposed to justify the suggestion. As we have seen (§ 4.1), the mere fact that partisans hold divergent and sometimes inaccurate beliefs is insufficient: even optimally rational truth-seeking agents can end up with divergent and inaccurate beliefs if exposed to selective and misleading information from trusted sources, and critics of the Partisan Bias Hypothesis need not think that people are optimally rational in this sense anyway.

4.3 Summary

In summary, even though the Partisan Bias Hypothesis seems to be vindicated by a wide range of observational and experimental findings, such findings appear to be consistent with non-motivational explanations. Importantly, this basic situation is not new. It reflects one of the oldest methodological problems in social psychology: namely, that even though there is a broad consensus among psychologists that motivated cognition is widespread, it has proven remarkably difficult to experimentally vindicate its existence in ways that clearly rule out alternative non-motivational explanations (see Kunda, 1990; Stanovich, 2021; Tetlock and Levi, 1982). In this specific case, however, non-motivational explanations are bolstered by a wealth of findings showing that citizens are not the dogmatic partisans that some earlier research seemed to suggest: far from simply clinging to partisan beliefs—or even doubling down (Nyhan & Reifler, 2010)—when such beliefs are challenged, people are capable of changing their minds in response to contrary evidence and persuasive arguments, even when such evidence and arguments reflect unfavourably on their side (Bisgaard, 2019; Coppock, 2023; Nyhan, 2021; Tappin, et al., 2023).

Step** back, we therefore find ourselves in the following situation. On the one hand, a large body of research in political science and the social sciences more broadly assumes that partisan cognition involves motivated cognition and draws on this assumption to generate a wide range of conclusions about the nature of democracy and the limitations of democratic accountability (Brennan, 2016; Hannon, 2022; Kahan, et al., 2017; Somin, 2006; Williams, 2021c). As we have seen, however, the evidence base for this assumption appears to be weak.

5 Making progress

How should we make progress in this debate? Most obviously, if the problem with existing experimental research is that prior beliefs are confounded with conclusions that partisans are motivated to reach, scientists could develop better experimental designs that can discriminate between their effects. For example, this might include running experiments in which participants lack prior beliefs about the topic altogether, extracting people’s prior beliefs in such a way as to rule out their influence in driving observed outcomes, or intervening directly on directional goals (Tappin, et al., 2020).

Such experimental innovations are important, and recent experiments that attempt to disentangle the effects of prior beliefs from directional goals appear to find compelling evidence in favour of the existence of partisan motivated cognition (Thaler, 2022); that partisans will hypocritically deploy abstract principles and arguments in ways that vary with their persuasive utility (Pinsof, et al., 2023); and that partisan cognition will feature the emotional signature of motivated cognition, in which encountering evidence against motivated beliefs is experienced as threatening and aversive (Westen, et al., 2006). Moreover, I have focused mostly on the individual psychology of partisan motivated cognition, but motivated cognition also has an important social dimension: when groups are motivated to form certain beliefs, the task of rationalising such beliefs is often outsourced to a minority of information producers. The Coalitional Press Secretary Theory thus predicts that much of partisan media should be organised around the production of intellectual ammunition specialised for rationalising party-favourable conclusions (Williams, 2022). Future research should more systematically explore such predictions and evaluate their empirical support.

Second, most research on partisan motivated cognition explores its role in driving misperceptions (i.e., confidently-held false beliefs) and in making people susceptible to partisan misinformation (Nyhan, 2020; Pennycook & Rand, 2019). If the theory developed here is correct, such phenomena are only one possible consequence of partisan motivated cognition. As I have tried to demonstrate, partisan cognition can be extremely biased—that is, motivated by a goal (i.e., partisan advocacy) distinct from truth—even if partisans are responsive to evidence and arguments. The most successful press secretaries can present largely accurate information if they are skilled at combining, framing, and filtering that information in ways that support pre-determined conclusions, and the same lesson generalises to partisan cognition: citizens can function as savvy partisan press secretaries—and so can be highly biased and polarised in their interpretations of reality—even if the number of demonstrably inaccurate beliefs that they hold is small, and even if they are highly responsive to evidence. Given this, political scientists should move beyond the focus on misperceptions and misinformation as the sole consequences of partisan motivated cognition and explore the many subtler ways in which party allegiances can bias political thought (see Bisgaard, 2019; Gaines, et al., 2007; Malka and Adelman, 2022).

Third, although I have focused on partisan identity, the Coalitional Press Secretary Theory applies in all cases in which individuals are motivated to promote the interests of coalitions, and political parties are not the only important political coalitions. Depending on the context, for example, there are also racial and ethnic groups, nation states, social movements, factions organised around particular leaders, and groups organised around specific policy goals (e.g., Remainers and Leavers in the UK Brexit debate). Moreover, insofar as parties function as coalitions of coalitions, the Coalitional Press Secretary Theory predicts that partisan bias will involve group-serving biases on behalf of the various coalitions that make up each party. In an insightful analysis of the connection between group alliances and political ideology, Pinsof and colleagues (2023) find strong support for this prediction, demonstrating that citizens deploy what they call “propagandistic biases” (i.e., group-serving biases) in defence of the various groups that support their favoured party.

Fourth, the Coalitional Press Secretary Theory identifies one important source of politically motivated cognition, but it is not necessarily the only source. For example, citizens support political parties at least in part because such parties promote their independent values and interests, and such values and interests might constitute an independent source of motivated cognition (D. M. Kahan, et al., 2017). Further, some research explores the ways in which group allegiances drive motivated cognition via motivations to signal group identity and loyalty (Funkhouser, 2022; Williams, 2021a). Although such motivations might interact with the motivation to advocate for party interests described here—affirming beliefs that promote and justify party interests might be especially well-suited to signalling party loyalty, for example—ingroup signalling is a distinct motivation. Future research should therefore systematically explore their connections and the distinctive psychological and social predictions that this alternative theory of partisan motivated cognition generates.

Finally, I have said nothing about individual differences or environmental conditions that might influence partisan motivated cognition. However, it could be that individuals vary in their susceptibility to motivated cognition and that different environmental contexts and affordances influence the strength of motivational biases (Groenendyk & Krupnikov, 2021), and both of these topics deserve further exploration. Further, although I have framed the disagreement as if there are only two positions—partisan cognition either involves motivated cognition or does not—this was solely for expository convenience. In general, human beings are complicated: even if we care about promoting the interests of our ingroup, we also care about accuracy and appearing reasonable, and such goals are likely balanced in subtle ways that vary across individuals and contexts. Future research should therefore move beyond the question of whether partisan motivated cognition exists and more carefully quantify its strength and the various individual and environmental factors that moderate its influence.