Partisan Bias in Political Judgment

INTRODUCTION

Partisan (noun): an adherent or supporter of a person, group, party, or cause, especially a person who shows a biased, emotional allegiance.

-Dictionary.com Red and blue America responded differently to the COVID-19 pandemic, with deadly consequences. Compared to politically liberal Americans, politically conservative Americans perceived the risks of COVID-19 as being lower (Bruine de Bruin et al. 2020, Relihan et al. 2023) and were less likely to engage in behaviors to avoid or mitigate the effects of infection (Grossman et al. 2020, Ye 2023, Young et al. 2022). These polarized reactions began early in the pandemic and strengthened over time (Relihan et al. 2023), reinforced by left-leaning and right-leaning media sources spreading different information about the seriousness of COVID-19 and the effectiveness of measures to prevent and treat it (Pennycook et al. 2020, Rathje et al. 2022). Consequently, along with age and race, political partisanship was one of the three strongest predictors of COVID-19 fatalities in the United States (Sehgal et al. 2022, Wallace et al. 2023) and a major contributor to a US death toll of over one million people.

Political partisanship in the twenty-first century often resembles performance art, with political combatants engaging in exaggerated arguments on social media, often over issues of little material consequence, ideally while "owning" their political opponents in the process. However, recent years have seen two poignant examples-the COVID-19 pandemic and the January 6, 2021 attack on the US Capitol-in which different views of factual reality among political partisans had major personal and political consequences. These examples highlight a growing divergence Downloaded from www.annualreviews.org. Guest (guest) IP: 149.102.246.27 On: Wed, 05 Feb 2025 06:59:02 in factual beliefs along partisan lines, or what Kellyanne Conway, advisor to President Donald Trump, famously referred to as "alternative facts." Whether the topic is health, elections, crime, or climate, what people believe to be true aligns increasingly well with the political party they support, and vigorous disagreement about factual issues is central to a deeply contentious partisan battle in the United States characterized by escalating mutual animosity across the political aisle (Finkel et al. 2020).

And yet, disagreement is an inescapable aspect of human social life. Disagreement is also the raison d'être of politics, as without disagreement, there would be no need for political behaviors like advocacy, negotiation, compromise, or even voting. A key challenge, then, is to identify when political disagreement is somehow untoward, not just an example of everyday diversity in values and beliefs but the result of a "bug" in our thinking that results in nonoptimal decision-making that becomes an obstacle to functional governing.

In this article, we explore conceptual and empirical work on the role of partisanship in political judgment. We first review ample empirical data suggesting that political judgment and behavior are characterized by a core ingroup favoritism: a systematic tendency to seek out, believe, and remember information that supports existing political beliefs and affinities. However, the extent to which this preferential treatment of politically congenial information reflects some form of judgmental bias, rather than a more rationally justifiable variation in belief and interpretation, has been hotly debated in both psychology and political science. At the center of this debate is the phenomenon of motivated reasoning, a form of thinking that has intrigued researchers for decades but has been challenging to both define clearly and demonstrate definitively. We review the decadeslong debate over the existence of motivated reasoning and argue that, although most research on political judgment is poorly suited to the technical task of distinguishing bias from rationality, the data that exist are difficult to square with a fully rational account of political judgment. More importantly, we argue that focusing on the normative aspects of partisan favoritism, whether or not it can be defended using conventional standards of rationality, is less productive than approaching it from a more descriptive perspective focused on identifying the specific psychological mechanisms underlying those judgments. In particular, a normative orientation privileges an either/or view of partisan favoritism-either it is rational or it is not-rather than a less mutually exclusive view in which both prior beliefs and political motivations combine to produce a pattern of partisan favoritism that poses a substantial challenge to democratic governance.

Defining Partisan Bias

Political judgment is not immune from the various quirks found to characterize human judgment in other domains. Two of the most robust and empirically reliable of these quirks are ingroup favoritism-the tendency to give preferential treatment to people with whom we share a group identity (Tajfel 1979)-and a suite of judgmental tendencies often referred to collectively as myside bias (Baron 1995, Stanovich 2021)-the tendency for people to "evaluate evidence, generate evidence, and test hypotheses in a manner biased towards their own prior opinions or attitudes" (Stanovich et al. 2013, p. 259).

In the domain of political judgment, these tendencies are most aptly captured by the term "partisan bias," which can be defined as a 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" (Ditto et al. 2019b, p. 274). In the United States, the term "partisan" usually denotes support for one of the country's two major political parties, the liberal (left-wing, blue) Democratic party and the conservative (right-wing, red) Republican party. We rely primarily on this meaning throughout this article, both because of the reality and relevance of the left/right distinction in US politics and culture (Iyengar & Westwood 2015, Jost 2006) and because the majority of empirical research operationalizes partisanship either as support for the Democratic versus the Republican party or as self-description on the liberal-conservative dimension. As suggested in the opening definition, however, partisan feelings can be driven by many sources, including attachment to a political figure (e.g., Donald Trump), a political group or faction within or outside the left/right continuum (e.g., the progressive wing of the Democratic party), or a specific cause (e.g., antiabortion or pro-same sex marriage).

The latter part of the opening definition is also worth noting. In normal usage, partisanship is closely associated with biased judgment, particularly bias flowing from emotional commitment ("biased, emotional allegiance"). Definitions from other sources characterize partisanship similarly as support that is unresponsive to evidence ("blind, prejudiced, and unreasoning allegiance"; Merriam-Webster), impulsive ("without considering or judging the matter carefully"; Cambridge English Dictionary), or otherwise excessive ("a fervent, sometimes militant supporter"; The Free Dictionary). However, support for a party, person, or cause is not inherently biased or irrational. Political partisans naturally hold many beliefs that lead them to favor their chosen party and are attracted to different groups and causes precisely because they have differing beliefs about the importance and magnitude of social problems and the preferred approaches to address them. It is important to distinguish between these normative aspects of partisanship and the kind of overzealous emotional allegiance that is often implied in usage of the term.

The contours of this distinction will be the primary focus of this article, but let us first distinguish between two intuitive senses of the term "bias" that will prove useful as we proceed. Following Klayman (1995) and Stanovich (2021), one use of the term bias refers to a systematic response tendency, with no firm implications for the rationality or irrationality of that tendency. This definition of bias captures everything from simple preferences (favoring chocolate over vanilla ice cream) to broader response tendencies, such as always responding with the first answer that comes to mind on multiple-choice tests. In the latter case, one may suspect that always choosing one's first response may be a suboptimal strategy, but the response tendency can be described as a bias with or without drawing that inference.

The other use of the term "bias" goes beyond a simple response tendency to imply a flawed reasoning strategy of some sort, a problem of thinking that violates normative standards of rationality. It is often assumed that violations of rationality are particularly likely to lead to inaccurate judgments and counterproductive behaviors, and thus considerable energy has been expended adjudicating the rationality of particular judgment tendencies. As with simple response tendencies, however, violations of rationality may or may not have problematic behavioral consequences (Arkes et al. 2016), and thus a crucial question when evaluating any form of bias is the extent to which reliance on it has adverse implications for individuals or society.

PARTISAN BIAS: THREE DIFFERENT KINDS

If we operationalize partisan bias as a simple response tendency-a tendency to favor information that is politically congenial (i.e., reflects positively on one's political beliefs or affinities) over information that is politically uncongenial (i.e., reflects negatively on one's political beliefs or affinities)-the empirical evidence for it is voluminous. In the following sections, we review three research traditions in which politically congenial information is favored over politically uncongenial information.

Selective Information Seeking

When given a choice about what information to seek out, people prefer information that supports their beliefs and affinities over information that challenges those beliefs and affinities. This Downloaded from www.annualreviews.org. Guest (guest) IP: 149.102.246.27 On: Wed, 05 Feb 2025 06:59:02 tendency was famously theorized by Festinger (1957) as "selective exposure," a way to avoid the cognitive dissonance experienced via exposure to counterattitudinal information. Despite mixed evidence supporting specific predictions of the dissonance perspective (Frey 1986, Freedman & Sears 1965), the general preference to seek out congenial over uncongenial information is robust. In a meta-analysis involving over 300 field and laboratory studies, Hart et al. (2009) found that, when given a choice, participants were nearly twice as likely to select information that supported preexisting attitudes and beliefs over information that challenged them (Cohen's d = 0.36).

Numerous studies, many using large representative samples, confirm a preference to seek out congenial over uncongenial information in the political domain (Knobloch-Westerwick 2014, Stroud 2010). This preference has been found in both experimental (Iyengar & Hahn 2009, Knobloch-Westerwick & Meng 2009) and nonexperimental studies (Garrett 2009a,b;Wicks et al. 2014), examining the use of both traditional (Rodriguez et al. 2017) and online media (Bakshy et al. 2015, Dejean et al. 2022). A similar effect has been found in people's choices to interact with others who either share or do not share their political views, both in the laboratory (Frimer et al. 2017) and in the real-world phenomenon of residential sorting (Gimpel & Hui 2015, Motyl et al. 2014). There is also evidence that political selective exposure has been increasing over the last few decades (Rodriguez et al. 2017), mirroring the development of modern media environments that afford selectivity in information search (Dylko 2016). This is not to say that people exist in impermeable echo chambers of politically congenial information, as there are several moderators of partisan information seeking. For example, evidence suggests that the draw of politically congenial information outweighs the avoidance of politically uncongenial information (Garrett 2009b, Garrett et al. 2013), and politically engaged individuals, even those with strong party loyalties, often pursue uncongenial information, particularly to gain an edge over opponents (Knobloch-Westerwick & Kleinman 2012).

Selective Information Evaluation

When asked to evaluate the quality or validity of information, people judge that information more favorably (as higher quality or more likely to be true) when it supports their beliefs and affinities than when it challenges those beliefs and affinities. This effect can be traced back to studies from the 1940s ( Janis & Frick 1943, Morgan & Morton 1944) showing that syllogisms with attitudeconsistent conclusions were judged to be more logically sound than similar syllogisms where the conclusion conflicted with prior attitudes (see Gampa et al. 2019 for a modern replication). The famous case study by Hastorf & Cantril (1954) of a contentious football game between Princeton and Dartmouth illustrated that social allegiances influence information evaluation: Students from both universities interpreted how clean and fair the game was in ways that favored their team. Subsequent research has found pervasive evidence of self-and ingroup favoritism in information evaluation (Hewstone et al. 2002, Stanovich 2021).

Similarly, an extensive body of research on political misinformation reveals that politically congenial news headlines, whether fake or real, are identified as true more readily than politically uncongenial ones (Gawronski et al. 2023, Pennycook & Rand 2021). The best evidence for selective information evaluation, however, comes from matched information experimental designs in which groups of partisans are presented with information that is identical in every way other than its political implications (Ditto et al. 2019b). For instance, a seminal study by Lord et al. (1979) preselected participants who had strong positions for or against capital punishment and then presented them with scientific evidence that was identical except that the results either supported or refuted the deterrent efficacy of capital punishment. Despite holding methodology constant across conditions, participants rated the studies as better-quality research when the results supported Downloaded from www.annualreviews.org. Guest (guest) IP: 149.102.246.27 On: Wed, 05 Feb 2025 06:59:02 their views on capital punishment than when the results challenged those views. Lord et al. (1979) manipulated the content of otherwise identical information (i.e., an outcome switching design), but other matched information studies manipulated the source of otherwise identical information (i.e., a party cues design; Cohen 2003, Tappin et al. 2020). In a meta-analysis of 51 matched information experiments examining views regarding a diverse set of political topics, Ditto et al. (2019b) found evidence of a modest (r = 0.25) but consistent effect for political congeniality: Partisans evaluated identical forms of evidence more positively when that evidence reflected favorably on the political ingroup compared to the political outgroup. The levels of bias observed were similar between outcome switching and party cues designs and, notably, for partisans on both the left and the right (for debate about these results, see Baron & Jost 2019, Ditto et al. 2019a).

Like selective information seeking, selective information evaluation is moderated by other factors. Researchers have long noted that information evaluation is subject to plausibility constraints (Festinger 1957, Heider 1958). People may bend their interpretation of information in the direction of desired conclusions, but ultimately they are constrained by their ability to construct seemingly reasonable justifications for these updated beliefs. This has been referred to as a need to maintain an "illusion of objectivity" (Pyszczynski & Greenberg 1987, p. 302) such that a desired belief is only maintained if a justification can be constructed that would persuade a dispassionate observer (see also Kunda 1990). Consistent with this view, political beliefs are more responsive to politically uncongenial information that is clear and consistent, with selective processing being more likely when receiving mixed information (Anglin 2019).

Selective Memory for Information

When asked to remember information, people are more likely to recall information that supports their beliefs and affinities than information that challenges those beliefs and affinities. Evidence for congeniality effects in memory traces back to a study by Levine & Murphy (1943) in which participants with pro-Communist attitudes showed better memory than anti-Communists for pro-Soviet Union messages, whereas anti-Communists showed better memory for anti-Soviet Union messages. Similar congeniality effects have been documented for information that reflects positively and negatively on the self (Chew et al. 2020, Sanitioso et al. 1990).

A growing literature provides evidence of congeniality bias in politically relevant memories, with many studies using a false memory paradigm in which participants are exposed to both fabricated and genuine political events. In the first study of this kind, liberals were about three times more likely than conservatives to falsely remember President George W. Bush vacationing in the wake of Hurricane Katrina, while conservatives were about twice as likely as liberals to falsely remember President Barack Obama shaking hands with Iranian President Mahmoud Ahmadinejad (Frenda et al. 2013). Subsequent work has consistently found that partisans (both in the United States and in the United Kingdom) report more false memories for politically congenial than uncongenial events (Calvillo et al. 2023;Greene et al. 2022;Murphy et al. 2019Murphy et al. , 2021)).

Given this evidence, one might expect people to exhibit a widespread congeniality bias in memory. Eagly et al.'s (1999) review of the relevant literature found evidence for an overall effect of congeniality on memory (Cohen's d = 0.23) but also found substantial variation in the strength of the effect, with some studies showing no advantage for congenial information and others showing enhanced memory for uncongenial information. Eagly et al. (2001) found that better memory for counterattitudinal messages was associated with more counterarguing and less persuasion, and thus it may be a byproduct of active attempts to refute information that challenges one's beliefs.

INTERPRETING EVIDENCE OF PARTISAN BIAS

Our review of the literature reveals that political partisans show a consistent tendency to seek out politically congenial information over politically uncongenial information, to evaluate politically congenial information more favorably than politically uncongenial information, and to better remember politically congenial than politically uncongenial information. Each of these tendencies is found outside the realm of political judgment as well, yet each is moderated by other factors, many of which are explainable as a function of pragmatic pressures.

There seems to be little question, then, that political judgment is biased based on a definition of bias as a systematic response tendency (Stanovich 2021). The question that remains is whether the consistent patterns of favoritism toward politically congenial information found in the literature reflect a deeper irrationality in political information processing.

Motivated Political Reasoning

The interpretation of partisan favoritism that has received the most attention is that it reflects motivated reasoning (Kunda 1990). Following Kunda (1990), motivated reasoning is often defined as judgment that is directionally motivated rather than accuracy motivated (Druckman & McGrath 2019). The typical goal of information processing is thought to be uncovering accuracy or truth; the individual's desire is to find the correct answer for the judgment in question. Directionally motivated reasoning occurs when the individual also has a collateral or accuracy-independent goal (Kahan 2016), a preference not for the true answer, but for a particular kind of answer. Directionally motivated reasoning is often associated with affective influences on judgment (Ditto 2009). Whereas accuracy-motivated reasoning is typically seen as a cold, inferential process, directionally motivated reasoning is viewed as a "hot" process, with preferences for particular answers being grounded in emotional attachments to certain ideas or identities (Van Bavel & Pereira 2018) and driven by motivational forces like desires, wishes, needs, and fears (Ditto 2009, Kunda 1990).

The idea that desires influence judgment and belief has a long history in psychology, beginning with the New Look in Perception movement in the late 1940s and early 1950s (Bruner & Goodman 1947, Postman et al. 1948) and its marquee phenomenon of perceptual defense, in which participants responded more slowly during a word recognition task when stimulus words had anxiety-provoking (e.g., sexually related) content. Research on cognitive dissonance processes, which captivated many social psychologists during the 1960s and 1970s, was similarly driven by the idea that desires-in this case, the desire for consistency between thoughts and behaviors-were an important influence on both information seeking and judgment. Finally, motivated reasoning had another resurgence in the late twentieth century with the phenomenon of self-serving bias, in which judgments and memory were argued to be distorted by the desire to maintain positive selfesteem (Kunda 1990). Despite a host of methodological and conceptual differences between the three research programs, the core idea driving each was that cognitive processes like perception, inference, and memory can be affected by our desires: What we perceive, believe, and remember is partly a function of what we want to perceive, believe, and remember.

Motivated reasoning is intriguing, as, if it exists, it would seem the epitome of irrationality. Believing something because you want it to be true violates the fact-value distinction [Weber 1946[Weber (1917))]. It is obviously illogical for the value one places on a fact being true to affect one's belief that the fact is indeed true (Simon & Read 2023). The nature of the bias posited by motivated reasoning is also unique. The many errors and biases cataloged by psychologists and decision scientists over the last half century can mostly be seen as inadvertent errors that result from myopic or misdirected attentional focus, over-or underweighting of a decision factor, or some other nonmotivational cause (Nisbett & Ross 1980). Motivated reasoning, in contrast, implies a process Downloaded from www.annualreviews.org. Guest (guest) IP: 149.102.246.27 On: Wed, 05 Feb 2025 06:59:02 in which judgment is influenced by affective forces that lead individuals to prefer particular conclusions over others. Although motivated reasoning is thought to operate without awareness (Kunda 1990, Taber & Lodge 2016), individuals engaging in motivated reasoning seem more culpable for their faulty judgment than, for example, individuals falling prey to the representativeness heuristic (Tversky & Kahneman 1974). Motivated reasoning produces a pattern of judgment that violates fundamental principles of fairness and impartiality, and it resembles a kind of self-serving hypocrisy that people find troubling ( Jordan et al. 2017, Shklar 1984). An impartial judge should not be affected by what they prefer to be true, especially if that preference is self-serving. When experiments reveal that people are more charitable toward information that implicates themselves or their ingroup compared to the very same information about other individuals and groups (Ditto et al. 2019b), this seems precisely the type of self-serving pattern that is at the heart of accusations of political bias in the media and lay political discourse ("If a Democrat had done what that Republican did, the media would have ripped them apart!"; Helgason & Effron 2022).

Motivated reasoning, if it exists, is a form of bias that seems rationally unjustifiable, morally problematic, and likely to have important negative consequences such as resistance to valid but uncongenial information and a tendency to make unflattering and unfair judgments about outgroup members.

The Observational Equivalence Problem

Motivated reasoning is not only a particularly intriguing form of judgmental bias but also an intuitively plausible explanation for the pattern of partisan favoritism found in the literature. To most people, the idea that political partisans unconsciously favor their own political group seems reasonable, just as most people accept motivated phenomena like denial and wishful thinking as explanations for behavior they observe in other facets of everyday life (Ditto 2009). So it is striking that finding unambiguous empirical support for motivated reasoning has been one of the most recalcitrant problems in the history of psychology (Ditto 2009, Erdelyi 1974, Tappin et al. 2020).

Dubbed the "observational equivalence problem" by Druckman & McGrath (2019), the core issue is that empirical data purporting to show motivated reasoning can almost always be explained in ways that do not require motivation as an explanatory concept. This empirical ambiguity has dogged research on motivated reasoning phenomena. Research on perceptual defense was originally met with considerable excitement, but it quickly fell out of scientific favor as commentators noted that longer recognition times for sexually explicit words did not necessarily indicate emotional resistance to perceiving those words but could occur simply because participants did not expect to be shown dirty words in a scientific context, and thus required longer exposure times to be confident enough to offer such a seemingly unlikely (and potentially embarrassing) response (Erdelyi 1974). Cognitive dissonance theory's motivated explanation for behavior-induced attitude change was challenged by self-perception theory (Bem 1972), which could explain those same phenomena as a function of cold self-inference (Bem 1967). Compelling critiques of self-serving bias research argued that because people generally expect and intend to obtain positive outcomes, it is reasonable to make internal attributions when one's efforts produce those positive outcomes but look for external causes for unexpected and unintended negative outcomes (Miller & Ross 1975, Tetlock & Levi 1982).

In the political domain, versions of this critique have been directed toward research on both selective information seeking (Freedman & Sears 1965) and selective memory (Eagly et al. 1999), but the issue has been debated most vigorously in research on selective information evaluation (Baron & Jost 2019;Ditto et al. 2019a,b). In the classic study by Lord et al. (1979), for example, the motivated account is that the findings are driven by participants' desires: People evaluate politically congenial information more positively than politically uncongenial information because of their The observational equivalence problem in political judgment.

desire to believe the former and disbelieve the latter. A plausible alternative account, however, is that the effect is driven not by participants' desires but by their priors: People evaluate politically congenial information more positively than politically uncongenial information because the former is more consistent than the latter with their prior beliefs (Lord 1992). Similarly, Pennycook & Rand (2021) observe a consistent pattern across multiple studies such that politically congenial headlines, whether fake or real, are believed more readily than politically uncongenial ones. They dismiss the claim that these data are evidence of motivated political reasoning, however, on the basis that the data can be explained equally well by the consistency of politically congenial headlines with prior beliefs (Tappin et al. 2020).

The persistence of the observational equivalence problem over decades of research and across many judgment domains is a testament to the significant methodological challenges of disentangling the role of priors and desires in judgment. Political partisans have both beliefs (priors) and affinities (desires). First, partisans enter our studies with prior beliefs about political topics, derived in large part from past exposure to political information from family, friends, media, and the like. Of course, partisans from different political sides bring with them different priors (i.e., alternative facts), again partly based on exposure to different informational content that occurs both de facto (Freedman & Sears 1965) and through the selective information-seeking processes described in Section 2.1. Holding beliefs about the nature of social problems and how to address them is a key aspect of what it means to be partisan.

Political partisans also enter our studies with prior attachments to political beliefs, derived from current and past intellectual, moral, and social commitments. Partisans are invested in particular answers to political questions; they prefer certain beliefs to be true because support for those preferred beliefs reflects positively both on their own epistemological and moral competence and on the competence and righteousness of the political groups that make up important aspects of their identities. Again, because partisans are committed to different intellectual and moral positions and identify with different political groups, desired answers to political questions also differ across the political spectrum. Partisans do not just hold political beliefs, they become committed to them as they invest intellectual, behavioral, and social capital in advocating policies based on those beliefs being true.

When researchers present political partisans with information that either supports or challenges a belief that is part of their partisan identity, that information has two distinct but confounded effects. Politically congenial information is consistent with both priors and desires. When a typical liberal is presented with research supporting anthropogenic climate change, for example, that information is likely to both make sense (as it corroborates information about climate change they have heard in the liberal media they consume) and feel good (as it validates both their sense of personal competence and the wisdom of their identification with groups for whom that belief is central). Conversely, politically uncongenial information is inconsistent with both priors and desires. When a conservative is presented with that same pro-climate change information, the information neither makes sense (as it clashes with prior beliefs based on their conservative media diet) nor feels good (as it threatens their sense of competence and the beliefs central to the political groups they identify with). Of course, liberal and conservative partisans should have precisely opposite, but equally confounded, reactions when presented with information politically congenial to conservatives.

The deep entanglement between priors and desires is what leads to the interpretational challenges faced by all research on partisan bias, even studies using high-quality matched information designs. When partisans more readily believe politically congenial than politically uncongenial information, it could be because politically congenial information fits better with their prior beliefs or because it confirms their desired beliefs. It is this latter, desire-driven process that most researchers would identify as true partisan bias, and the goal of matched information experiments is to isolate this motivated effect by presenting partisans with information that is identical except for its consistency with their desired beliefs. The problem with this approach is that the consistency of information with priors is central to both prior-based and desire-based accounts. Even if every other detail of the information provided to participants is held constant, once details are introduced to make a message more or less consistent with participants' political beliefs (e.g., changing the results of an otherwise identical scientific study), the message inevitably becomes both more or less desirable to one side than the other (the intended effect) and more or less plausible to one side than the other (the confounding effect). This is the empirical catch-22 faced by research on motivated reasoning (Ditto et al. 2019b).

The observational equivalence problem is particularly acute in research on political reasoning because political motivations are almost always operationalized as participants' self-described political views; that is, they are measured or selected for rather than manipulated. Political research also involves plausibility constraints that make the manipulation and control of confounding variables challenging. Early motivated reasoning research used laboratory-constructed scenarios and fictitious information to manipulate and control for relevant factors (Ditto et al. 2003, Miller 1976). Because political judgment is enmeshed in real events and experiences, and is studied using ecologically realistic materials, partisan bias researchers have less methodological flexibility to address sticky interpretational problems. Despite the devotion of considerable effort and methodological ingenuity over many years, there is no current consensus about the strength of empirical support for motivated reasoning, and the issue remains particularly contentious in the realm of political judgment (Baron & Jost 2019;Ditto et al. 2019a,b;Druckman & McGrath 2019;Pennycook & Rand 2019).

THE NORMATIVE APPROACH: PARTISAN BIAS VERSUS RATIONALITY

One way to characterize the observational equivalence problem is as an empirical tie. Data showing partisan favoritism can be explained plausibly by either a motivational (desire-based) or a cognitive (prior-based) account. This tie is often treated, however, as a win for the cognitive side.

The fact that motivated biases are seen as different from other kinds of judgmental bias often leaves them with the burden of proof against what seem like simpler, more parsimonious explanations based on reasonable inferences participants might make given their prior beliefs (Nisbett & Ross 1980). This view flows naturally from a normative perspective deeply embedded in the judgment and decision-making tradition, whose core research strategy is to pit experimental findings against predictions based on rational principles to reveal judgmental bias (Tversky & Kahneman 1974). Through this normative lens, prior-based accounts for motivated reasoning effects are treated as a rational baseline rather than an alternative theoretical account, and the core research question is framed not as a competition between equally viable alternative explanations for the same patterns of data, but in terms of whether a particular research finding can be clearly distinguished from a rational baseline. If it cannot, the finding is chalked up as another mundane example of human rationality. In the normative view, ties go to the rational runner.

The Bayesian Account of Partisan Bias

Many studies on partisan bias adopt this normative framing and focus on whether empirical effects that putatively reflect motivated partisan bias can be explained instead as a function of purely rational inference (Baron & Jost 2019, Gerber & Green 1999). In early motivated reasoning research, this critique was usually guided by an intuitive analysis of rational judgment (Miller & Ross 1975). Contemporary research typically frames the issue in terms of a more formal but functionally equivalent analysis of rational belief updating based on Bayes' theorem.

Our goal here is to avoid the technical aspects of Bayesian analysis. Readers interested in formal modeling of partisan bias from a Bayesian perspective can find many examples available (Bartels 2002, Bullock 2009, Gerber & Green 1999). The essence of Bayesian belief updating, however, is straightforward. To update a belief based on new information, a rational (Bayesian) judge should combine their confidence in their prior belief about the topic with their confidence in the new information about the topic. The nature and degree of belief updating is a function of the content of the new information (its consistency with prior beliefs), weighted by the judge's subjective confidence in the validity of the new information relative to their subjective confidence in their prior beliefs.

Bayesian analysis is often taken to support the rationality of using prior beliefs to determine confidence in new information (Baron & Jost 2019, Druckman & McGrath 2019, Koehler 1993). According to this view, when new information confirms prior beliefs, it should be confidently accepted as true. When new information contradicts prior beliefs, however, it is rational to treat that information more skeptically. As such, the tendency to evaluate politically congenial information more positively than politically uncongenial information is said to be rational from a Bayesian perspective, particularly because people tend to have high confidence in the validity of their prior beliefs and some degree of uncertainty about the validity of new information (Baron & Jost 2019).

Limitations of the Bayesian Account of Partisan Bias

The precision of Bayesian models makes them a clear advance over intuitive analyses of rationality. Like all normative models, however, their application to real-world judgment is tricky. For example, compelling normative analyses have argued both that partisan bias is consistent with Bayesian rationality (Bullock 2009, Koehler 1993) and that it is not (Bartels 2002). There is also reason to argue that, consistent with the general tendency to privilege rational explanations over motivated accounts described earlier, Bayesian critiques of partisan bias research often stretch normative logic in three related ways.

First, the normative appropriateness of using prior beliefs to assess confidence in new information is questionable (Kahan 2016). Bayesian updating specifies a process for combining prior beliefs with new information to formulate a new (posterior) belief. It does not, however, specify a process for assigning a confidence value to either prior beliefs or new information (Ditto et al. 2019b, Kahan 2016). Bayesian models are coherence based (Arkes et al. 2016); they specify an internal logic for how elements of the model should fit together (based on the mathematics of conditional probabilities), but they are silent on how the specific values assigned to those elements should be determined. Moreover, using the value assigned to one component of a conditional probability equation (confidence in prior beliefs) to assign value to another component (confidence in new information) violates basic principles of independence that underlie coherencebased models (Simon et al. 2004). When evaluating evidence in a legal setting, for example, a juror's assessment of the validity of eyewitness testimony should not influence their assessment of the validity of fingerprint evidence, or vice versa. Using confidence in one piece of evidence to evaluate confidence in another before combining the two is a kind of judgmental double-dipping that is normatively suspect (Simon & Read 2023).

Second, even if we accept the rationality of using prior beliefs to evaluate new information, the normative appropriateness of consistently favoring prior beliefs over new information can be questioned. A feature of coherence-based models is that, because they specify no external constraints on how values in the model are determined, there are multiple ways for elements of the model to fit together coherently (e.g., different modes of dissonance reduction; Festinger 1957). Reducing confidence in the validity of new information is one rationally defensible strategy for resolving its inconsistency with prior beliefs. However, it is equally defensible (and equally consistent with Bayesian rationality) to respond to inconsistent information by acknowledging its validity and reducing confidence in prior beliefs. The fact that people consistently favor the former inference over the latter is usually attributed by Bayesian accounts to the strength of participants' prior beliefs (Baron & Jost 2019), but the strength of prior beliefs is almost always assumed rather than measured. Moreover, a comprehensive analysis of rationality would take into account the objective quality of the new information (Ditto et al. 2019a, Kahan 2016). If the quality of belief-inconsistent information is low, a rational judge should reject it and maintain confidence in their prior belief.

If the quality of that information is high, on the other hand, the rational response would be to accept its validity and update prior beliefs accordingly. Because Bayesian models base their analysis of rationality exclusively on coherence criteria (i.e., internal consistency), they cannot easily incorporate correspondence-based factors (i.e., external reality) and often make assumptions about unmeasured variables that favor rational interpretations of observed response tendencies.

Finally, even if we accept the rationality of derogating the validity of information that challenges prior beliefs, Bayesian counter-explanations for motivated reasoning effects often rely on overly expansive conceptions of what can be rationally inferred from prior beliefs. If an individual with strong and specific prior beliefs expresses skepticism about the accuracy of information that directly challenges those specific beliefs, we agree that skepticism is not clear evidence of motivated reasoning. Many motivated reasoning studies, however, involve very general prior beliefs Downloaded from www.annualreviews.org. Guest (guest) IP: 149.102.246.27 On: Wed, 05 Feb 2025 06:59:02 and judgments about new information that are not plausibly explained as products of those beliefs (Ditto et al. 2019a). Consider a study by Kopko et al. (2011). The study used a matched information design to show that the same flawed ballot was more likely to be judged as legitimate if cast for a candidate of the participant's own party than for a candidate from the rival party. The relevant prior belief was not measured but might be assumed to be something like "Candidates of my party are more likely to win elections than candidates from the other party." It is not clear how a Bayesian analysis could be stretched to infer from this prior that the same flawed ballot should rationally be judged as more legitimate if cast for a favored candidate than for an unfavored one. For this finding to conform with Bayesian logic, one would need to assume a prior such as "Voters of my party are more likely to cast flawed ballots." While partisans may invoke such beliefs as post-hoc rationalizations, it strains plausibility to suggest this belief was guiding their judgments from the outset.

Studies claiming to show motivated reasoning have been conducted in a number of different judgment domains and have used a variety of dependent measures. The results of many, perhaps most, of them are difficult or impossible to differentiate from a Bayesian account. For others, however, rational explanation is considerably more strained and less plausible (Ditto et al. 2019b). Bayesian models are often cast as blanket explanations for all partisan bias effects (Baron & Jost 2019, Druckman & McGrath 2019), but this requires many cases where Bayesian just-so stories must be given a substantial benefit of the doubt.

A DESCRIPTIVE APPROACH: IDENTIFYING DRIVERS OF PARTISAN BIAS

Normative analysis is a hallmark of a judgment and decision-making paradigm that has made enormous contributions to the fields of psychology and behavioral economics. It has played an important role in motivated reasoning research, highlighting key conceptual and empirical ambiguities (Ditto 2009, Miller & Ross 1975). Normative approaches to human judgment have also been sharply criticized, as scholars disagree about both the value of comparing judgment to a rational standard and the specifics of what constitutes that standard (Gigerenzer 1991, Krueger & Funder 2004).

The normative approach to partisan bias research suffers from the same strengths and limitations. Bayesian analyses have stimulated important progress in conceptual precision (Druckman & McGrath 2019, Tappin et al. 2020), but legitimate debates about what constitutes rational judgment diminish the utility of research that requires appeals to nebulous standards of logical coherence. The practical importance of logical incoherence has also been questioned (Arkes et al. 2016).

An alternate approach is to frame the core question in partisan bias research not as a normative question ("Is partisan bias rational?") but rather as a descriptive one ("What psychological factors contribute to partisan bias?"). A descriptive approach to partisan bias differs from a normative one in three important and related ways.

First, unlike a normative perspective, a descriptive approach treats prior beliefs as figure rather than ground. Because rational judgment is generally seen as normal, functional, and implicitly unproblematic, research from a normative perspective treats prior beliefs as a factor to be ruled out rather than one to be studied and explored. In most partisan bias studies, priors are assumed rather than measured, and their effects on judgment are seldom analyzed. A descriptive approach makes prior beliefs more central, where empirical and theoretical questions about their role in partisan bias can be examined.

Second, unlike a normative perspective, a descriptive approach does not equate prior-based processing with rationality. The Bayesian perspective, in particular, highlights the rationality of reliance on prior belief in judgment. However, a host of other research traditions in psychologywork on anchoring (e.g., Epley & Gilovich 2006), belief perseverance (Anderson et al. 1980), and expectancy confirmation (Darley & Fazio 1980), for example-identify prior beliefs as potential sources of bias, often leading to overly conservative patterns of belief-consistent information processing (Oeberst & Imhoff 2023, Simon & Read 2023). A descriptive approach avoids the conflation of priors with rationality, encouraging a view in which both priors and desires are potential contributors to partisan bias effects.

Finally, unlike a normative perspective, a descriptive approach does not imply that the effects of priors and desires on judgment are mutually exclusive. Most research on partisan bias has operated under the assumption that partisan bias effects are due either to motivated reasoning or to priorbased inference. Framed as a problem of differentiating motivated bias from a rational baseline, research can support one view or the other, but not both. A descriptive view does not necessitate this either/or framing. Instead, it suggests an approach in which the phenomenon of interest becomes the systematic response tendency to favor congenial over uncongenial information and the key empirical question is the extent to which priors and desires contribute independently to that response tendency.

Not all research on partisan bias embraces an explicitly normative approach (Gawronski et al. 2023, Pennycook & Rand 2021), but remnants of the normative agenda linger, such as the relative lack of empirical attention to priors and the tendency to treat priors and desires as alternative explanations for partisan bias rather than independent contributors to it. In the remainder of this section, we explore implications of this descriptive approach and review empirical research exploring the role of prior beliefs in information evaluation.

Controlling for Prior Beliefs

Ironically, one implication of a descriptive approach to partisan bias is that it affords an empirical strategy for examining the adequacy of normative accounts. As an empirical matter, the Bayesian account of motivated reasoning effects is that they are driven exclusively by the consistency of new information with prior beliefs. This claim can be examined empirically by incorporating prior beliefs into research on partisan bias and measuring reactions to congenial and uncongenial information that are equated (experimentally or statistically) for their perceived likelihood of occurrence. A motivated reasoning effect is supported if the differential evaluation of politically congenial and uncongenial information is obtained while simultaneously controlling for the effect of prior beliefs.

Examples of this approach come from research on motivated reasoning about health information. Using a procedure that presented college students with the results of a fictitious medical test (Ditto & Croyle 1995), Ditto and colleagues consistently found lower ratings of diagnostic test accuracy when a test result was unfavorable (suggesting elevated susceptibility to future disease) than when it was favorable (suggesting reduced susceptibility to future disease; Ditto & Lopez 1992;Ditto et al. 1998Ditto et al. , 2003)). Like partisan bias effects, relative resistance to unfavorable diagnoses is consistent with either a motivated or a prior-based account (i.e., test derogation could be driven by either the desire to be healthy or the relative unexpectedness of an unfavorable diagnosis). Because these studies used a fictitious health condition, it was possible to plausibly equate participants' prior beliefs across diagnosis conditions using prevalence information and to confirm with manipulation checks that participants perceived the favorable and unfavorable diagnoses as equally likely. Results indicated that even when favorable and unfavorable diagnoses were perceived as equally likely, participants still rated the unfavorable result as less accurate than the favorable one (Ditto & Lopez 1992, Ditto et al. 1998).

Statistical procedures can also be used to control for the effects of priors. As noted, participants in the studies described above were asked about the perceived likelihood of receiving a favorable Downloaded from www.annualreviews.org. Guest (guest) IP: 149.102.246.27 On: Wed, 05 Feb 2025 06:59:02 versus an unfavorable result prior to test administration, and other studies included post-result measures of self-reported surprise, on the logic that less expected outcomes should produce more surprise. The tendency to derogate the accuracy of unfavorable test results (relative to favorable ones) persisted after statistically controlling for the effects of subjective likelihood and/or selfreported surprise (Ditto et al. 1998(Ditto et al. , 2003)).

Similar findings have been observed in studies of political judgment. Munro & Ditto (1997) found that participants with anti-gay attitudes rated an identical study more positively when results confirmed (rather than challenged) prevalent stereotypes, while participants with pro-gay attitudes showed the opposite pattern. Most important for current purposes, the studies also found that the differential evaluation of stereotype confirming and disconfirming information was predicted by overall attitudes about homosexuality, even after controlling statistically for participants' endorsement of the specific stereotypical beliefs that the scientific studies confirmed or refuted (see Munro et al. 2002 for similar findings in judgments about a presidential debate).

These studies have methodological limitations characteristic of older research (e.g., measurement and power issues), and many examine judgments about health, not politics. More recently, Stagnaro et al. (2023) included measures of prior beliefs in a large preregistered study examining the controversial phenomenon of motivated numeracy. Motivated numeracy refers to the idea that system 2 processes (like numerical reasoning) can be co-opted to defend politically congenial beliefs (Kahan et al. 2017). In support of this contention, Kahan and colleagues found that while highly numerate individuals were more likely than less numerate ones to draw accurate conclusions from contingency table data when the topic was nonpolitical, greater numeracy was associated with more partisan bias (i.e., a higher likelihood of interpreting the data as supporting participants' own political views) when the data involved a contentious political issue. Seeking to replicate Kahan et al.'s (2017) results, Stagnaro et al. (2023) found a partisan bias effect but no evidence that this effect was enhanced in more numerate individuals (see also Persson et al. 2021). They also found that the partisan bias effect persisted even when the consistency of the information with priors was controlled, concluding that this finding supports the motivated nature of the effect.

Modeling the Influence of Priors and Desires

The studies described in Section 5.1 demonstrate one benefit of incorporating prior beliefs into studies exploring motivated reasoning effects. By controlling for priors (methodologically or statistically), these studies provided data consistent with an effect of information congeniality on judgment that is difficult to attribute to the consistency of information with prior beliefs. However, because the goal of the studies was simply to exclude priors as causal factors, little attention was devoted to examining their independent predictive role in reactions to congenial and uncongenial information.

There is good reason to explore the role of priors in political judgment. First, it is plausible to expect that people will be more skeptical of information that contradicts prior beliefs than of information that confirms them, independent of any motivational influence. The normative debate over partisan bias concerns whether consistency with expectations is its sole driver, but that a process akin to Bayesian belief updating accounts for some of the relative resistance to politically uncongenial information seems likely, and it is supported by a wealth of research showing that previously held beliefs are resistant to change even in the absence of strong motivational forces (Anderson et al. 1980, Epley & Gilovich 2006). It is reasonable to expect that at least some of the tendency to evaluate politically congenial information more favorably than politically uncongenial information is attributable to the greater consistency of politically congenial information with prior beliefs. Second, assessing the effect of prior beliefs on the evaluation of new political information is important because priors reflect the political information streams to which people have been exposed. As documented in Section 2.1, liberal and conservative partisans have different media diets, and these partisan media diets are likely a key contributor to the different prior beliefs the two sides bring to evaluations of new political information (Earle & Hodson 2022). Gauging the independent contributions of priors to the evaluation of politically congenial and uncongenial information helps to assess the extent to which the alternative media ecosystems inhabited by liberals and conservatives contribute to the selective evaluation of new information.

Unfortunately, few studies have examined the independent contributions of prior beliefs and political motivations on political information evaluation, and these studies have important limitations. MacCoun & Paletz (2009) conducted a large-scale exploration of partisan bias in assessments of scientific research on public policy controversies. As part of a random-digit dial survey of California residents, participants heard a description of a scientific study on one of four political topics, with results manipulated to be consistent with liberal or conservative viewpoints. Respondents supplied both their general political ideology and their prior attitudes about each specific topic (i.e., their support for policies to address each problem). As expected, California residents were more skeptical of identically described research when the findings contradicted rather than supported their political views. Most important for our purposes, study evaluations were independently predicted by their consistency with both general political attitudes and specific policy support. As in previous studies, participants' political ideology consistently predicted skepticism about study results independent of the effect of specific topical attitudes (the study's proxy for priors). Participants' topical attitudes also consistently predicted evaluations of study quality independent of the effect of political ideology: The more a study's results contradicted participants' attitudes about the specific topic examined, the less confidence participants expressed in its validity. A clear limitation of the study, however, is measurement imprecision: Political ideology is a questionable proxy for political motivation untainted by prior beliefs, and policy support is a questionable proxy for prior beliefs untainted by political motivation.

Using an innovative approach to disentangle the two factors, Tappin et al. (2017) leveraged natural variability in beliefs about the 2016 presidential election competition between Hillary Clinton and Donald Trump to estimate the independent contributions of priors and desires on belief updating. Participants provided both their preferences about who they wanted to win the election (desires) and their beliefs about who would win the election (priors). Participants were then randomly assigned to view polling data that suggested that either Clinton or Trump would win the election. Results showed a robust influence of desires on belief updating, such that partisans more readily updated their beliefs about who would win when presented with data favoring their desired candidate. In contrast, the independent effect of participants' beliefs about whom they expected to win was not robust across analyses. In sum, using an election as context allowed Tappin et al. (2017) to parse desired and expected outcomes more cleanly than previous research and yielded clear evidence for an effect of information congeniality independent of prior beliefs. The key limitation of the study is that its dependent measure was belief updating rather than information evaluation, which may also account for the surprising lack of effect for prior beliefs.

These studies provide templates for a descriptive approach to partisan bias research that treats prior beliefs and political motivations as independent contributors to partisan bias effects (see also Celniker & Ditto 2024). The challenges of empirically disentangling these effects from one another remain, however, and future research must develop better methods for measuring (and perhaps manipulating) key constructs and assessing their independent contributions to the preferential treatment of politically congenial information. Some general methodological

SOME METHODOLOGICAL RECOMMENDATIONS FOR PARTISAN BIAS RESEARCH

1) Supplement matched information designs with measurement of relevant priors and desires. Measures of priors should tap specific, realistic expectations about the likelihood of desired and/or undesired outcomes. Measures of desires should be purely evaluative (i.e., uncontaminated by priors). Researchers should consider results across modeling approaches to assess whether priors and desires predict unique variance in judgments (Westfall & Yarkoni 2016).

2) Use specific measures of information evaluation. General measures of belief updating (e.g., "How persuasive is the information?") are more susceptible to Bayesian counter-explanation than are granular measures of information quality (e.g., "How adequate is the study's sample size?").

3) Include neutral comparison conditions to better estimate effects. Most matched information experiments compare how opposing partisans respond to identical stimuli or how like-minded partisans respond to congenial versus uncongenial stimuli. Including control participants (e.g., Independents) and/or control conditions (e.g., nonpolitical stimuli, evaluations blinded to partisan implications; Celniker & Ditto 2024) allows for better estimation of effect sizes and the ability to isolate and compare effects attributable to type of partisan (e.g., liberals vs conservatives) or type of information (e.g., congenial vs uncongenial).

recommendations for future research are presented in the sidebar titled Some Methodological Recommendations for Partisan Bias Research.

THE FUTURE OF PARTISAN BIAS RESEARCH

Viewing partisan bias through a normative lens encourages methodological approaches that seek to rule out the rational role of prior beliefs in responses to politically congenial and uncongenial information. Similarly, framing the empirical challenges of motivated reasoning research as a problem of observational equivalence (even without an explicitly normative goal in mind) implies that prior beliefs and political motivations are mutually exclusive explanations. Viewing partisan bias more descriptively, on the other hand, affords a view in which reliance on prior belief is not equated with rationality, and beliefs and affinities are conceived of not as competing explanations for partisan bias effects, but as independent factors contributing to it. This simple change in orientation has a number of implications for future research.

Sharpening the Focus of Partisan Bias Research

Incorporating measures of prior beliefs into partisan bias studies helps break the observational equivalence logjam by converting normative accounts from background assumptions into testable hypotheses. Evidence of partisan bias is found even when the consistency of new information with prior beliefs is controlled either methodologically or statistically (Celniker & Ditto 2024, MacCoun & Paletz 2009, Munro & Ditto 1997, Stagnaro et al. 2023). Disentangling the effects of priors and desires remains a serious methodological challenge, and studies done to date have limitations that future research must address, but together with other research findings that are challenging to explain via Bayesian logic (e.g., Kopko et al. 2011), there is now persuasive evidence in the literature supporting the existence of motivated political reasoning.

A focus on prior beliefs can also provide insight into conceptual ambiguities. Alongside its empirical travails, partisan bias research has been hampered by a lack of specificity about psychological processes (Ditto 2009, Tappin et al. 2020). Like any other source of bias, desires must affect judgment via some effect on the information processing sequence (Ditto & Lopez 1992). Research on partisan bias, however, often treats motivational influence as essentially mechanism free. People are thought to believe politically congenial information simply because they want to believe it.

However, as researchers have recognized for decades, people do not believe whatever they want to believe (Festinger 1957, Heider 1958). In Kunda's (1990, p. 480) words, people's ability to reach desired conclusions is "constrained by their ability to construct seemingly reasonable justifications for these conclusions."

Stated another way, partisan bias is both sustained and constrained by prior beliefs. Prior beliefs are necessary grist for a motivated cognitive mill (Ditto & Lopez 1992). The tendency for partisans to believe politically congenial information should be strongest when desired beliefs are consistent with prior beliefs, that is, when information both feels good and makes sense (see Figure 1). Partisan differences in the interpretation of information should thus be largest when liberals and conservatives both have plausible reasons to believe their divergent interpretations (e.g., each side can access prior beliefs consistent with their desired explanation). Motivated reasoning should be weak in the absence of beliefs to support reaching a desired conclusion, and especially constrained when prior beliefs conflict sharply with desired ones, making desired beliefs highly implausible (e.g., the belief that one can fly).

This view conforms well with a view of priors and desires as independent contributors to (rather than competing explanations for) partisan bias effects. Beliefs and affinities work in conjunction to allow people to believe what they want to believe when they can do so without offending their illusion of objectivity (Kunda 1990), but they constrain motivated reasoning when desired beliefs are highly implausible (Ditto et al. 1998). This analysis fits well with theoretical accounts that subsume prior-based and desire-based processing under the same conceptual umbrella. Cognitive dissonance theory, for example, sees both expectancy violation and self-concept threat as factors that evoke the psychological discomfort of dissonance and cognitive strategies to reduce it (Festinger 1957, Thibodeau & Aronson 1992). Related literatures on correspondence-based reasoning (Simon & Read 2023) and processing fluency (Schwarz et al. 2020) similarly see both inconsistency with beliefs (logical incoherence) and inconsistency with desires (affective or emotional incoherence) as leading to dissonance-like discomfort. These literatures could provide an integrative theoretical framework for research on motivated reasoning.

Widening the Aperture of Partisan Bias Research

A key point of this review is that the preferential treatment of politically congenial information occurs at multiple levels-in information seeking, information evaluation, and memory-and processes at each level can contribute to the broader tendency for partisans to hold beliefs that support their political allegiances. For example, researchers have noted the important connection between media exposure and partisan biases in information evaluation (Derreumaux et al. 2022, Pennycook & Rand 2019), and prior beliefs are the psychological link between the two. Selective exposure to partisan media sources contributes to partisan differences in prior beliefs, and these prior beliefs, together with political motivations, in turn conspire to drive partisan differences in the interpretation of new political information. Preferential memory for politically congenial information (see Section 2.3) can also lead to prior beliefs that make politically desirable outcomes seem more likely.

At a practical level, this suggests that any attempt to reduce partisan evaluative bias must also address the biased information streams that feed and support it. Partisan differences in the interpretation of political information will be difficult to combat in the modern media environment in which partisans are exposed to different media sources offering alternative Downloaded from www.annualreviews.org. Guest (guest) IP: 149.102.246.27 On: Wed, 05 Feb 2025 06:59:02 factual accounts of political events. At a theoretical level, this wider lens on partisan bias further complicates the extent to which a particular judgment can be construed as rational or irrational. Are judgments based on prior beliefs rational if those beliefs are based on motivated information seeking or memory? An important focus for future research is how different forms of partisan bias might operate in mutually reinforcing ways to provide individuals with plausible reasons to hold politically desirable beliefs.

Getting Practical about Partisan Bias Research

www.annualreviews.org • Partisan Bias

Downloaded from www.annualreviews.org. Guest (guest) IP: 149.102.246.27 On: Wed, 05 Feb 2025 06:59:02