RESEARCH DOSSIER
Political Reasoning, Selective Exposure, and Belief Updating
Abstract. Political reasoning links information selection, evidentiary appraisal, and changes in confidence about collective affairs. This fundamental critical narrative research monograph reconstructs classic studies of biased assimilation and motivated skepticism, compares directional and accuracy goals with prior-based inference and expressive responding, and evaluates the empirical relationship between selective exposure and political beliefs. Its synthesis spans experimental argument-evaluation tasks, accuracy incentives, blinded scientific-evidence studies, headline discernment, fact-checking experiments, correction meta-analyses, and platform field interventions. The evidence supports context-sensitive preferences for congenial material, multiple mechanisms of asymmetric appraisal, and generally effective factual correction; it also demonstrates that incentive effects differ across unfamiliar and familiar claims and that experimentally changing exposure need not change short-term polarization. Methodological chapters examine construct validity, causal identification, replication, longitudinal follow-up, cross-cultural transportability, and new forms of human–artificial information mediation. The work is a critical narrative synthesis of published research, not an original experiment, registered systematic review, or novel meta-analysis.
Keywords: political reasoning; political cognition; motivated reasoning; selective exposure; belief updating; partisan bias; motivated skepticism; directional goals; accuracy incentives; Bayesian inference; political source cues; information credibility; political misinformation; fact-checking; correction; cognitive reflection; political polarization; causal inference; human–artificial information mediation
Political Reasoning as an Empirical Research Object
Political reasoning comprises the operations by which people interpret claims about collective affairs, evaluate evidence, draw causal or normative inferences, and revise judgments about political actors, institutions, policies, and events. The object of this dossier is the sequence connecting informational opportunity, choice of evidence, interpretation of encountered material, and subsequent belief revision. Each stage has a distinct observable outcome. Choosing a congenial article is not equivalent to believing its contents; judging an argument as strong is not equivalent to voting or changing a policy preference. An adequate psychology of political inference specifies which outcome is measured and the circumstances that generate it Kunda (1990) Taber & Lodge (2006).
Political information has unusual properties as a test case for general theories of cognition. Evidence is often incomplete, the disputed claims bear on personally meaningful groups, and the standards for assessing relevance and credibility may themselves be contested. A person can reason coherently from an inaccurate premise, distrust a source for reasons grounded in previous experience, or defend a favored conclusion by imposing a selective burden of proof. These possibilities require simultaneous attention to subjective priors, informational provenance, accuracy goals, identity commitments, and the actual evidentiary value of the message Bullock (2009) Druckman & McGrath (2019).
The present monograph adopts a source-based critical narrative synthesis of foundational theoretical writings, experiments, field studies, systematic quantitative syntheses, and modern methodological critiques. The evidence is grouped by causal question and outcome rather than by an assumed universal bias. It evaluates directional and accuracy motives, selective exposure, probabilistic updating, partisan factual beliefs, correction, digital environments, and methods for identifying mechanisms. It reports no original participants, preregistered systematic review, or new quantitative meta-analysis; conclusions are conditional on the available studies, their populations, and their manipulations Hart et al. (2009) Ditto et al. (2025).
Origins of the Modern Research Program
Historical foundations span social-psychological work on cognitive consistency, judgment under uncertainty, persuasion, and the psychology of public opinion. A decisive experimental reference is Lord, Ross, and Lepper's 1979 investigation of reactions to mixed findings on capital punishment. Participants with opposing preexisting positions evaluated putative studies in ways consistent with their prior views. The design made biased assimilation and subsequent polarization central research problems, but its constructed materials and immediate attitude measures do not establish that every political disagreement follows the same trajectory Lord et al. (1979).
Kunda's 1990 integrative account distinguished accuracy-oriented from directional goals and argued that motivation influences reasoning through accessible cognitive processes and defensible justifications. The theory treats a preferred answer as shaping which strategies and memories are recruited while preserving constraints imposed by a person's available knowledge. This formulation explains why motivated interpretation may appear thoughtful and elaborated rather than impulsive. It also poses a challenge for empirical research: researchers must establish the relevant motive independently of the selective evaluation they seek to explain Kunda (1990).
Taber and Lodge's 2006 experiments brought these ideas into political judgment using evaluations of issue arguments and opportunities to select further information. Their influential account of motivated skepticism joins a preference for congenial information with differential scrutiny of counterattitudinal arguments and possible attitude polarization. Later work has investigated which stages are robust, whether experimental materials confound accuracy with ideological direction, and how apparent motivated reasoning can be distinguished from appropriate prior-based inference Taber & Lodge (2006) Bullock (2009) Celniker & Ditto (2024).
Directional Goals, Accuracy Goals, and Motivational Pluralism
A directional goal favors a particular conclusion, affiliation, or self-consistent interpretation; an accuracy goal favors a conclusion appropriately responsive to evidence. Neither is a permanent property of a political person. The same respondent may exhibit careful calibration on one question and selective standards on another. Directional motivation can operate through information search, memory retrieval, argument generation, source trust, and response expression. Accuracy motivation can alter those same operations by increasing attention to credibility and by making an uncertain answer acceptable Kunda (1990) Bullock et al. (2015).
One theoretical danger is to infer the existence of a hidden motive solely from the outcome. For example, stronger criticism of an opposing claim could arise because its evidence is genuinely weak, because its source has previously been unreliable, or because the participant wishes to defend an identity. Experiments should cross the content's veracity and evidentiary quality with its partisan implications, and ideally manipulate incentives or conceal the politically identifying feature. When these features are entangled, the mechanism remains underidentified Bullock (2009) Celniker & Ditto (2024).
Motivational pluralism is a more tractable account than a single characteristic of political reasoning. Individuals can care at the same time about the truth of an assertion, acceptance within a group, consistency with a prior worldview, and economical use of attention. These pressures can reinforce or counteract one another. A theoretically precise model therefore specifies which goal is made salient, the available information, the anticipated audience, and the dependent variable. Different experiments isolate different portions of this causal chain Ditto et al. (2025) Druckman & McGrath (2019).
Prior Beliefs, Bayesian Updating, and the Normative Benchmark
Belief updating describes a change in confidence after new evidence is encountered. Bayesian inference offers a formal benchmark: posterior odds equal prior odds multiplied by a likelihood ratio describing the diagnosticity of the observed evidence under competing hypotheses. This identity clarifies a foundational problem in interpreting polarized reactions. Two people who start with different priors may rationally retain different posteriors even when they receive the same message, and may attach different likelihoods to evidence from a disputed source Bullock (2009).
Bayesian coherence does not automatically certify the accuracy of a person's priors or of a source model. A participant may update consistently relative to a self-selected and inaccurate model of the world. Nor does the observation of disagreement establish that processing is motivationally biased. Strong tests therefore require eliciting prior probabilities, presenting evidence with known objective diagnosticity or explicitly measured credibility, and observing quantitative belief changes over time. Researchers should compare observed posterior movement with multiple models: unbiased updating, source-specific learning, anchoring, and directional processing Bullock (2009) Thaler (2024).
Political claims often combine descriptive and normative propositions. A claim about an unemployment trend can be empirically assessed under a defined measure and period; a policy judgment additionally involves distributional values, causal assumptions, and acceptable tradeoffs. A factual correction can therefore change an estimate without changing a policy attitude. The analytical separation of factual confidence, value preferences, source credibility, and political choice prevents scientific studies from treating any stable preference as a failure to learn Wood & Porter (2019) Porter & Wood (2021).
Belief, Confidence, Attitude, and Expressed Partisanship
A political belief is a proposition held with some degree of confidence; an attitude is an evaluative orientation; an identity is a socially meaningful category of self-location; and an expressed response is an observable act within a particular measurement environment. These variables correlate but cannot be substituted for one another. A survey response may reflect sincere confidence, a symbolic expression of group membership, knowledge uncertainty, or strategic impression management. Measuring one while theorizing another is a recurrent source of apparent contradiction in this literature Bullock et al. (2015) Ditto et al. (2025).
Accuracy incentives and explicit 'don't know' options are diagnostic because they can change reported political facts without supplying new information about the world. Bullock and colleagues documented that incentivized and uncertainty-permitting responses reduce some apparent partisan factual gaps. The result demonstrates that a response format matters; it does not establish that all remaining disagreement is ignorance or that respondents have no genuine partisan differences in underlying beliefs Bullock et al. (2015).
Belief updating must be measured at a declared time scale. An immediate response following a correction may represent short-term recall, changed source trust, or temporary acquiescence. Later responses test durability and competing information; real-world behavior poses a still harder criterion. Valid research records baseline confidence, exposure, immediate posterior response, delayed response, and uncertainty, and distinguishes changes in factual estimation from movements in candidate evaluation or political identity Walter & Murphy (2018) Ecker et al. (2022).
Selective Exposure: Search, Attention, and Choice
Selective exposure refers to systematic differences in the probability of encountering information as a function of its congruence with prior attitudes, identities, or preferences. Its operational meaning changes with the research design. A questionnaire about preferred articles measures stated choices; a laboratory menu of hyperlinks measures selection under experimentally restricted options; logged news consumption measures behavior in an environment shaped by availability, search, ranking, and prior habits. A respondent who never sees a dissonant article may have actively avoided it, lacked an opportunity, or received a feed in which it was absent Hart et al. (2009) Garrett (2009).
Hart and colleagues synthesized the selective-exposure literature and found a modest average preference for congenial information, accompanied by meaningful differences across task, motivation, and informational circumstance. The result supports a probabilistic tendency, rather than a categorical avoidance rule. Researchers should report the baseline opportunities to select each type, the costs of exposure, and the distinction between ideological agreement and actual informational reliability. Strong selective exposure may be adaptive when sources differ in quality, yet it can also diminish encounters with corrective evidence Hart et al. (2009).
A full causal sequence can be separated into supply, encounter, attention, choice, processing, memory, and later response. Selection into a news environment and its informational effects are different estimands. If researchers observe that partisans choose like-minded articles, they have identified a preference conditional on a choice set. Demonstrating that the preference causes persistent misbelief requires a subsequent manipulation or a credible causal design linking exposure to a defined updating outcome. This distinction is central to evaluating proposed echo-chamber mechanisms Garrett (2009) Nyhan et al. (2023).
Preference for Reinforcement and Avoidance of Challenge
Reinforcement seeking and challenge avoidance are logically different strategies. A reader may value agreeable material while also consuming adversarial content to monitor opponents or understand an issue. Garrett's observational research on online political news use challenged an overly simple picture of segregated consumption: preferential engagement with supportive content did not imply wholesale avoidance of contrary information. Its observational design limits inference about why a person selected each item, but the distinction between seeking and avoidance is conceptually indispensable Garrett (2009).
A laboratory menu should distinguish exposure to congruent and incongruent claims from exposure to objectively strong and weak evidence. Without that crossing, preferring supportive material could reflect either identity consistency or an expectation of higher accuracy. The selective-exposure literature further identifies defensive motives, perceived utility, commitment strength, and the informational goals of the task as potential moderators. These are hypotheses about variation, not automatic diagnoses of each individual click Hart et al. (2009) Kunda (1990).
Measurement itself can create artificial differences. Headlines may differ in emotional vividness, familiarity, credibility, reading difficulty, or novelty. A provocative opposing article may elicit more clicks because it promises a strong counterexample; a supportive source may be trusted because it has a record of accurate reporting. Strong designs pretest content, randomize positions, record the full choice set and navigation sequence, and analyze both first exposure and subsequent selections. They also distinguish selection from attention after an item has opened Hart et al. (2009) Garrett (2009).
Political Source Cues, Credibility, and Epistemic Trust
Source cues provide information about expertise, incentives, institutional reputation, political affiliation, and possible deception. Such cues affect both initial selection and the evidentiary weight assigned to a claim. Trusting a recognized scientific institution can be a rational shortcut if its track record is informative; treating a partisan endorsement as definitive evidence regardless of content is a different pattern. A credible experiment must make these cue properties separable from the claim's truth and diagnosticity Celniker & Ditto (2024) Thaler (2024).
Blinded evaluation offers a particularly useful diagnostic design. By concealing which partisan constituency benefits from a reported finding while preserving its substantive empirical description, a researcher can estimate changes in perceived study quality attributable to political meaning. Celniker and Ditto's experiments examined evaluations of scientific evidence and the difficulty of separating preexisting informational expectations from preferred conclusions. Their work illustrates why both subjective priors and partisan preferences must be independently measured Celniker & Ditto (2024).
Epistemic trust also varies over time. An audience can learn that a source is unreliable and appropriately discount later messages; alternatively, repeated identity signals may make credibility depend excessively on group congruence. Longitudinal studies should separate prior source assessments from retrospective judgments made after reading a favored claim. They should include item-level truth checks, measured uncertainty, and tests for whether participants apply standards symmetrically to otherwise equivalent claims Kunda (1990) Ditto et al. (2025).
Algorithms, Information Supply, and the Echo-Chamber Hypothesis
Digital exposure is the product of user choices, social networks, platform recommendation, publisher supply, and ranked presentation. These causal inputs should be measured independently. An apparent partisan concentration of encountered content may reflect an individual's deliberate subscription, algorithmic amplification, friends' sharing behavior, or all three. A theory in which selective exposure changes beliefs needs evidence about both the distribution of encounters and the psychological effects of those encounters Nyhan et al. (2023) Bail et al. (2018).
Using a large Facebook field experiment during the 2020 US presidential election, Nyhan and collaborators reduced exposure to like-minded sources for 23,377 consenting users by roughly one-third. The intervention changed the informational environment but did not measurably change eight preregistered attitudinal outcomes, including affective polarization and belief in false claims, during the studied period. This is evidence about the marginal effect of a specific feed intervention under particular historical and platform conditions. It does not demonstrate that all sustained exposure environments are irrelevant or that individual-level learning never occurs Nyhan et al. (2023).
Other interventions demonstrate that the same descriptive term 'cross-cutting exposure' encompasses meaningfully different treatments. Forced attention to elite opposing arguments, voluntarily chosen news, and feed-level rebalancing differ in credibility, tone, intensity, duration, and audience. Their outcomes cannot be combined as if they were identical doses of one substance. Scientific synthesis must track treatment versions, uptake, compliance, timing, baseline polarization, and the outcome's level of aggregation Bail et al. (2018) Nyhan et al. (2023).
Cross-Cutting Exposure, Reactance, and Real-World Tests
An informative intervention exposes individuals to arguments associated with an opposing position and measures belief or attitude change after that exposure. Bail and colleagues' randomized Twitter study is widely cited because receiving opposing partisan messages was followed by increased conservative identification among the Republican participants, while the observed shift among Democrats was smaller and statistically uncertain. The findings concern a specific intervention using political messages from accounts, with limited scope for universal generalization Bail et al. (2018).
The mechanism behind an exposure effect is not identified merely by observing a changed attitude. A reader may update on the arguments' contents, react negatively to the messenger, infer that the opposing group is extreme, or become more aware of threats to group identity. Strong follow-up designs independently manipulate factual content, civility, source labeling, and conversational context. They should also measure whether participants actually attended to the material, what they learned, and whether the effect persists Bail et al. (2018) Druckman & McGrath (2019).
At the population level, a change in information flow does not entail a proportional change in public opinion. Learning can coexist with stable group identity and policy evaluations. The findings from major field interventions make this separation experimentally concrete: a treatment can appreciably alter exposure without producing detectable short-term changes on selected polarization scales. Understanding null findings requires examination of confidence intervals, outcome sensitivity, treatment duration, adherence, and the background informational environment Nyhan et al. (2023).
Biased Assimilation and Argument Evaluation Paradigms
The classic biased-assimilation paradigm presents ostensibly comparable evidence bearing on an issue about which participants already hold strong views, then elicits perceived methodological quality and attitude change. Lord and colleagues' 1979 capital-punishment study was important because opposing groups interpreted mixed findings in line with prior beliefs, and expressed stronger polarization after consideration. Yet a single immediate attitude measure cannot establish durable conviction, and hypothetical studies may not evoke the same responses as independently verifiable research Lord et al. (1979).
Political argument experiments should measure the evidence's actual diagnosticity, the participant's initial position, evaluation criteria, and subsequent judgments separately. A participant may recognize a counterargument as coherent while finding its conclusion insufficiently supported; another may reject it because its asserted facts conflict with trusted background knowledge. Separating such possibilities often requires material whose methodological strength and partisan implication can be factorially manipulated, and independent judges who assess study quality without partisan labeling Taber & Lodge (2006) Celniker & Ditto (2024).
Motivated skepticism can be tested with process measures such as time spent on disagreeable claims, counterargument generation, changes in perceived credibility, and scrutiny of methodological details. None is conclusive alone. Spending longer reading a disfavored argument might indicate hostile scrutiny, genuine interest, comprehension difficulty, or greater novelty. Converging evidence from blinded quality assessments, prior-belief elicitation, and controlled information search is stronger than inferring a motive from a single postexposure preference Taber & Lodge (2006) Ditto et al. (2025).
Analytic Reasoning and the Discernment of False Claims
An influential alternative to partisan identity as the dominant explanation for inaccurate news judgments emphasizes analytic engagement. Pennycook and Rand found that analytic reasoning was associated with improved discrimination between accurate and inaccurate news headlines, including politically congenial misinformation. The central distinction is between a person's willingness or capacity to scrutinize a claim and the politically favorable direction of its message. The original studies concern bounded headline tasks and should not be generalized uncritically to every form of political cognition Pennycook & Rand (2019).
Correlation between a cognitive-reflection measure and news discernment is compatible with several explanations: analytical engagement, educational familiarity with the task, general knowledge, and shared method variance. Stronger causal tests interrupt or facilitate deliberation while holding headline veracity and partisan alignment constant. Bago, Rand, and Pennycook used a two-response framework in which participants provided an initial quick judgment and then a more considered response; their results indicated that further deliberation improved assessment of false headlines under the examined task conditions Bago et al. (2020).
These findings qualify any theory in which more cognition necessarily produces greater partisan rationalization. Deliberation can support correction when evidence is accessible and the accuracy problem is well specified. Conversely, elaborate reasoning may generate post hoc rationalizations when a desired outcome shapes which evidence is consulted. An explanatory theory should predict when additional deliberation improves truth discrimination, when it changes apparent confidence only, and when it facilitates selective argument production Pennycook & Rand (2019) Tappin et al. (2021).
Cognitive Sophistication and the Identification Problem
People who score highly on reasoning measures also differ from other respondents in factual knowledge, media consumption, education, prior beliefs, and political engagement. If these variables are correlated with ideological commitments, a relationship between cognition and apparent polarization cannot by itself identify an enhanced ability to rationalize. Tappin, Pennycook, and Rand directly examined the purported link between cognitive sophistication and politically motivated reasoning, highlighting the difficulty of ruling out prior-belief explanations Tappin et al. (2021).
One solution is to define a normative benchmark for each claim before measuring performance: specify the true state of the world and the informativeness of the evidence. Another is to manipulate source or partisan implication independently of informational content. A third is to elicit prior beliefs before disclosure and estimate the update expected from the observed likelihood information. When all these elements are considered, effects that first appeared motivational may diminish, persist, or change sign across tasks Bullock (2009) Tappin et al. (2021).
Researchers must also avoid a symmetrical interpretive trap: calling a response motivated merely because it is ideologically asymmetric, or calling it rational merely because it can be rationalized after the fact. Both hypotheses require out-of-sample predictions. The decisive design is one in which objective evidence and expected partisan advantage are crossed while participants' source beliefs are measured in advance. That design makes the explanatory competitors answer the same empirical question Thaler (2024) Celniker & Ditto (2024).
Partisan Factual Beliefs and Accuracy Incentives
Surveys of political facts often elicit large differences between partisan groups, but response formats can make symbolic agreement and knowledge uncertainty look like private factual conviction. Bullock, Gerber, Hill, and Huber experimentally varied incentives for accurate responses and allowed respondents to admit uncertainty. Some observed partisan differences diminished, indicating that expressive responding is an important component of particular survey outcomes. These findings help decompose a response, rather than resolving the full question of political belief formation Bullock et al. (2015).
Accuracy incentives introduce their own interpretive challenges. Payment may encourage more careful recall, reliance on accessible background knowledge, or a change in the participant's understanding of what counts as a sincerely correct answer. Without independently assessing the accuracy of the answer, greater confidence or reduced polarization under incentives does not prove genuine learning. Where possible, researchers should compare incentivized and unincentivized responses against objective factual standards and repeat measurement after the incentive is removed Bullock et al. (2015) Stein et al. (2024).
The magnitude of an incentive effect also depends on the epistemic novelty of the claim. Familiar political facts may be associated with strategic expressive habits; unfamiliar statements minimize prior factual knowledge while potentially heightening dependence on partisan signals. It is therefore essential to distinguish mechanisms observed in well-known historical claims from those observed in newly introduced propositions. A result from one item class cannot be treated as a universal property of party affiliation Stein et al. (2024).
Blinding, Preference–Prior Separation, and Scientific Claims
Experimental blinding provides a practical test of whether political congruence changes the appraisal of otherwise similar scientific evidence. A participant may rate a study as persuasive because its result is welcome, because it confirms a more specific prior prediction, or because the source possesses features associated with methodological rigor. Celniker and Ditto's 2024 research addresses the intersection of preferences and priors in judgments about scientific evidence and shows why their contributions must be separately investigated Celniker & Ditto (2024).
To distinguish the components, a study can elicit participants' expected empirical results before showing the new evidence, independently assess their preferred policy result, and vary whether the partisan implication is visible. An effect of a blinded study on perceived credibility supplies a different test from an effect generated by overt party labels. If an experiment manipulates ideological meaning but also changes the scientific question, method, or presentation quality, it cannot attribute the evaluation difference to preference alone Celniker & Ditto (2024) Druckman & McGrath (2019).
Blinding is especially informative when paired with repeated evaluation of multiple studies and recorded standards for evidence quality. It can reveal whether the participant applies the same statistical and methodological criteria across preferred and disfavored conclusions. The general research program extends beyond laboratory vignettes to real scholarly controversies, in which scientists and citizens face uncertainty about external validity and the trustworthiness of expert consensus. Such extension demands careful separation of scientific uncertainty from partisan motivation Ditto et al. (2025) Celniker & Ditto (2024).
Political Misperceptions and the Correction Problem
Correcting a political misperception requires specifying the fact to be corrected, the initial false belief, the evidence supplied, and the subsequent claim measured. Corrections can successfully change factual estimates while leaving a larger narrative or policy judgment intact. Nyhan and Reifler's influential 2010 experiments showed that corrections of particular political claims sometimes failed and, in certain conditions, were associated with stronger endorsement of the false claim. The finding initiated a substantial research program about the circumstances and frequency of such backfire patterns Nyhan & Reifler (2010).
Backfire must be defined narrowly. A factual backfire is an experimentally attributable increase in false factual belief following a correction relative to an appropriate control. Continued partisan attitudes, lingering uncertainty, anger toward the corrector, or incomplete return to baseline represent different outcomes. A weakly powered subgroup difference can mimic a backfire; ceiling effects and ambiguous survey wording can create further distortions. Researchers should specify the estimand and correction outcome in advance Wood & Porter (2019) Ecker et al. (2022).
Later research has frequently found that corrections work on factual beliefs, with residual misinformation and limited downstream changes. Wood and Porter investigated a wide range of claims and found factual backfire difficult to reproduce under their tested circumstances. The scientific synthesis must preserve both historical observations and the stronger evidentiary base provided by independent large-sample tests. Backfire is a possible empirical pattern that requires demonstration in context, not an expected universal consequence of confronting a political belief Wood & Porter (2019) Nyhan & Reifler (2010).
Meta-Analytic Evidence on Correction Effectiveness
Quantitative synthesis helps assess whether individual correction studies reflect a broader tendency or a task-specific anomaly. Walter and Murphy synthesized experimental research on misinformation correction and reported an overall beneficial effect, with variation across domains and intervention designs. Averaged effects should not be treated as the response expected from each political constituency or each type of inaccurate assertion. The same literature includes heterogeneous outcome measures, time horizons, target topics, and credibility cues Walter & Murphy (2018).
An evidence-quality appraisal should examine whether included studies used verified false statements, randomized corrections, suitable controls, and comparable posttreatment measures. Meta-analytic estimates can be distorted by selective publication, dependence among outcomes within experiments, and the inclusion of corrections delivered in artificially high-attention settings. More studies and larger samples reduce sampling uncertainty but do not by themselves repair poor construct validity. Summary effects need to be reported with a clear definition of the intervention, outcome, and measurement window Walter & Murphy (2018) Ecker et al. (2022).
The psychological mechanisms of correction can include updating a factual representation, building an alternative causal explanation, changing perceived source reliability, and increasing uncertainty. Some corrections discredit a specific claim without changing the broader schema that made it attractive. A complete theory therefore predicts which new knowledge replaces the inaccurate proposition and whether that replacement can be retrieved later. This explanation-based approach also distinguishes an accurate short-term answer from stable resistance to repeated misinformation Ecker et al. (2022).
Fact-Checking Across Political and National Contexts
Fact-checking is an intervention class rather than a single uniform treatment. A correction may be a succinct contradiction, a detailed explanation, a credibility assessment by journalists, or a comparison with official records. Its outcome may concern recognition of the corrected assertion, trust in the source, intention to share, or a downstream political choice. Studies must clearly identify which of these is measured. Claiming that a fact-check changed voting behavior when it only changed a quiz answer exceeds the evidence Porter & Wood (2021).
Porter and Wood's simultaneous experiments across Argentina, Nigeria, South Africa, and the United Kingdom investigated whether corrective information improves factual accuracy outside the setting of a single country's political debate. The cross-national design provides evidence relevant to transportability, while reminding researchers that populations, media institutions, source trust, and political climates still vary. A demonstrated correction effect is conditioned by the sampled participants, the selected claims, and the format of the intervention Porter & Wood (2021).
Cross-national findings make it possible to examine whether political motivation and correction work only through highly specific US partisan identities or reflect more general cognitive operations. Comparative research should assess translation equivalence, the meaning of party cues, the reliability of local fact sources, and the relation between political identity and information access. Invariance cannot be assumed from use of the same questionnaire wording; content and institutional credibility require local validation Porter & Wood (2021) Ecker et al. (2022).
New Misinformation, Incentives, and Directional Belief Formation
Experiments using previously unfamiliar claims create a powerful identification opportunity. Participants have less direct issue-specific evidence in memory and may rely on ideological fit, source cues, or generic credibility heuristics. Stein, Keuschnigg, and van de Rijt investigated belief in newly encountered misinformation under accuracy incentives and found that partisan belief differences could persist even when correct judgments were rewarded. Such evidence qualifies the generalization that financial incentives eliminate all apparent partisan belief distortion Stein et al. (2024).
Experimental unfamiliarity is also an ecological limitation. A new social-scientific claim may concern topics about which participants have broad ideological expectations despite lacking knowledge of the exact fact. The investigator must separate general issue priors from desire for a favorable answer. In addition, responses to a laboratory stimulus may have little social consequence, whereas public statements in a real political environment may signal loyalty or reputational commitment. Both environments are needed to identify the full mechanism Stein et al. (2024) Bullock et al. (2015).
Together, the incentive studies yield a conditional synthesis. For some familiar facts, explicit rewards and uncertainty options reduce apparent partisan disagreement; for novel politically meaningful propositions, biased judgments may survive rewards. These are complementary findings about different tasks, not evidence that one set of researchers invalidates the other. Future work can use the same experimental sample to compare familiar and unfamiliar claims, preregister how partisan congruence is coded, and measure beliefs privately and publicly Bullock et al. (2015) Stein et al. (2024).
Accuracy, Sharing, and the Behavior–Belief Gap
Believing a claim, intending to share it, actually sharing it, and repeatedly consuming it are distinct phenomena. A person may circulate politically appealing content that they regard as uncertain, while another may accept a claim without any inclination to broadcast it. Work on false-news dissemination documents highly uneven sharing across users, and headline discernment experiments examine a related but separate accuracy judgment. Explanations of political information behavior should state whether they target epistemic confidence, social signaling, or the observable diffusion of a message Guess et al. (2019) Pennycook & Rand (2019).
Platform traces have substantial ecological validity for the sharing act but provide limited access to the user's private conviction. Survey responses can assess belief but are sensitive to question framing and impression management. Combining behavioral records with survey measures, source characteristics, and temporally ordered exposure produces stronger inference. Ethical and privacy constraints require appropriate consent and aggregate reporting, especially when individual political affiliations or inferred beliefs are linked to personal digital records Guess et al. (2019) Nyhan et al. (2023).
Research syntheses frequently compress heterogeneous outcomes under the vague term misinformation susceptibility. A better measurement framework tracks accuracy discernment, confidence calibration, expressed endorsement, retransmission, and delayed belief separately. Interventions should be assessed for the specific response they demonstrably change. This disciplined separation also clarifies why substantial improvements in factual correction may coincide with stable affective polarization or political preference: those outcomes occupy different levels of psychological organization Porter & Wood (2021) Nyhan et al. (2023).
Operationalization and the Measurement of Political Inference
Construct validity requires a precise mapping between a theoretical claim and the variable recorded. Motivated reasoning is sometimes operationalized as a difference in belief acceptance across partisan labels, but that contrast may combine knowledge, source trust, preferences, expressive responding, and genuine updating. Selective exposure is often inferred from platform composition, although observed distributions also reflect delivery. Belief updating may be reduced to a posttest score without a prior measure, making the direction and magnitude of revision unknown Bullock (2009) Ditto et al. (2025).
One measurement framework records exposure opportunity, source identification, initial factual confidence, an independent estimate of evidentiary quality, active choice, argument evaluation, posterior confidence, and delayed response. Researchers can compare probability distributions and shifts in individual confidence rather than dichotomizing a spectrum of uncertainty into correct and incorrect. Repeated items and hierarchical models allow estimates of person- and item-specific variability, but interpretations still depend on item quality and the content validity of the chosen factual domain Thaler (2024) Celniker & Ditto (2024).
A credible experimental report also states which information can be objectively evaluated and which is intrinsically uncertain or normative. When a policy outcome depends on contested forecasts and moral preferences, the experiment may reveal changing judgments without furnishing a unique accuracy benchmark. Data collection should distinguish knowing a fact, predicting an event, endorsing a normative goal, and supporting an intervention. Precision at this boundary protects the validity of both psychology and political science Druckman & McGrath (2019) Bullock (2009).
Causal Identification in Laboratory and Field Designs
Random assignment identifies the causal effect of an assigned treatment on measured outcomes when assignment and follow-up are valid. However, it does not automatically identify the psychological mechanism. Assigning a person to a partisan source cue may change familiarity, perceived hostility, credibility, and motivation simultaneously. Mechanistic inference therefore requires a design that separately manipulates or measures those mediators, along with transparent assumptions about interactions and confounding Celniker & Ditto (2024) Bail et al. (2018).
Field experiments improve contact with natural political information environments but introduce noncompliance, spillovers, variable treatment uptake, changing exposure, and limited observability of attention. A randomized feed intervention estimates the consequence of that particular ranking or supply change, not the effect of consuming every individual item that the algorithm promoted or demoted. Selective attrition and outcome measurement after the event can further compromise interpretation. Reports should include both intention-to-treat effects and any carefully justified compliance analyses Nyhan et al. (2023).
Observational digital trace data offer scale and behavioral detail but raise difficult problems of self-selection and reverse causality. People who choose ideological outlets may already differ in confidence, trust, or political interest, and changes in attitudes can themselves motivate a change in media diet. Time series, panels, and quasi-experiments can improve causal inference when their identifying assumptions are plausible. Every causal claim needs an explicit account of the comparison group, potential confounders, treatment definition, and counterfactual estimand Garrett (2009) Nyhan et al. (2023).
Statistical Evidence, Replication, and Uncertainty
An empirical finding must be interpreted with respect to its estimated effect, uncertainty, design, and measurement reliability. A statistically significant interaction between political group and message direction may be small in practical terms, while an imprecise nonsignificant effect may be compatible with consequential differences. Confidence intervals, sample sizes, preregistered primary outcomes, and the number of analytic choices are central to evaluating research. Studies of politically salient topics especially benefit from transparent item selection and analysis plans Wood & Porter (2019) Nyhan et al. (2023).
Independent tests can revise a field's understanding without erasing the original historical finding. For example, early evidence of correction backfire motivated a robust debate, while subsequent larger studies found this effect elusive across many tested claims. This trajectory favors treating backfire as a context-dependent empirical possibility. In contrast, one null feed intervention cannot establish the irrelevance of all social media exposure because the intervention operates within a specific environment over a specified period Nyhan & Reifler (2010) Wood & Porter (2019) Nyhan et al. (2023).
Replication demands more than repeating an experiment's headline. It requires preserving the causal contrast that instantiated the theory while adapting historical content to remain comprehensible. A replication of a partisan cue intervention should measure contemporary recognition and credibility of the cue; a replication of a false-claim correction should establish that participants can encounter the original falsehood plausibly. Multisite tests reveal contextual heterogeneity only when measurement invariance and protocol adherence are documented Porter & Wood (2021) Ditto et al. (2025).
Systematic Review, Meta-Analysis, and Evidence Hierarchies
A narrative monograph integrates experiments, observations, methodological critiques, and synthesis papers to explain mechanisms and contradictions. It must state that its selection of studies is guided by conceptual relevance and source accessibility rather than an exhaustive preregistered search. Hart and colleagues' quantitative analysis of congenial exposure and Walter and Murphy's correction synthesis provide cross-study perspectives, yet they combine distinct designs and outcomes. Interpreting a meta-analytic average always requires examining what the constituent studies actually measure Hart et al. (2009) Walter & Murphy (2018).
Evidence hierarchy is question-specific. Randomized laboratory studies are strong for a narrowly defined immediate effect, while well-conducted large-scale field experiments are informative about interventions in an actual platform. Verified observational logs describe real selection patterns that artificial choice menus may miss. Cross-national designs test whether an effect travels across institutional contexts. No single design dominates on internal identification, construct validity, realism, duration, and population generalizability at once Garrett (2009) Porter & Wood (2021) Nyhan et al. (2023).
Research synthesis should distinguish replicated directional tendencies, fragile heterogeneous estimates, theoretical interpretations, and open causal questions. In the present literature the existence of selective congenial preference, partisan differences in evaluation, and some benefits of factual correction enjoy multiple independent evidentiary strands. The size and mechanism of those effects, their temporal stability, and their relationship to real voting behavior remain dependent on the specific design. The data support calibrated conclusions rather than a universal single-cause explanation Hart et al. (2009) Bullock et al. (2015) Ditto et al. (2025).
Population Differences and Cross-Cultural Transportability
Political psychological constructs are embedded in institutional environments. 'Liberal', 'conservative', party affiliation, public trust, and political knowledge have different meanings across electoral systems and historical periods. An instrument calibrated to US partisan identity may capture a different structure in a multiparty setting. Transporting results requires checking measurement invariance, local party cue recognition, and the comparability of public claims and correction sources Porter & Wood (2021) Druckman & McGrath (2019).
Even within a country, participants differ in political interest, access to information, education, cognitive reflection, media skills, and uncertainty tolerance. These characteristics influence which messages are encountered and how study tasks are understood. Convenience online samples may overrepresent politically engaged or digitally active respondents. Demographic controls cannot reconstruct absent populations or eliminate unmeasured differences in causal mechanisms Garrett (2009) Pennycook & Rand (2019).
An effective research program combines cross-cultural replication with context-specific hypotheses. It should seek broad psychological regularities while recording how institutions shape incentives for accuracy, social identities, and the credibility of public authorities. Studies across Argentina, Nigeria, South Africa, and the United Kingdom illustrate an important step toward broader evidence, although four settings cannot exhaust world political diversity. Generalization is an empirical question that must be investigated rather than declared Porter & Wood (2021).
Source Credibility, Institutional Expertise, and Trust Repair
Political knowledge depends on testimonial chains: citizens routinely rely on journalists, researchers, agencies, witnesses, and other intermediaries to learn facts they cannot personally observe. Reasonable source evaluation can therefore involve institutional independence, transparent evidence, correction history, and methodological competence. This epistemic structure matters for psychological inference because discounting an unfamiliar source may be an informed credibility judgment or an identity-driven shortcut; the same overt behavior is compatible with both Bullock (2009) Thaler (2024).
To study trust causally, researchers can provide verified information about a source's prior accuracy before presenting an ideologically meaningful claim. They can randomize neutral and partisan source labels while retaining the identical evidence, and elicit changes in source reliability estimates separately from changes in belief about the issue. This procedure can distinguish a person's willingness to learn from a source from a tendency to endorse a message because it comes from an allied institution Celniker & Ditto (2024) Ecker et al. (2022).
Trust repair presents a dynamic question: following exposure to a false claim and a subsequent correction, which component of the person's model changes? The person may revise the original fact, downgrade the messenger, or develop generalized distrust of fact-checkers. Immediate fact recognition may improve even as judgments about institutions remain stable. Longitudinal measures should track those multiple outcomes with validated scales and include enough time for competing sources to reenter the informational environment Porter & Wood (2021) Ecker et al. (2022).
Dynamic Models of Information Selection and Revision
A useful dynamic account describes a loop: initial belief informs source choice; source choice affects encountered evidence; interpretation updates confidence; updated confidence in turn shapes future searching. This structure accommodates positive feedback without assuming that every loop magnifies polarization. Exposure to accurate, incongruent information can reduce misbelief, while reinforcement seeking can strengthen familiarity or conviction. Feedback can become weaker when curiosity, accuracy rewards, or trusted correction introduce alternative information Hart et al. (2009) Walter & Murphy (2018).
Mathematical models can represent belief as a probability distribution and source selection as a policy that values anticipated information, identity consistency, social rewards, and attention costs. A distinct likelihood function can describe perceived credibility of a source. Such models yield empirically testable contrasts: changing a source label should alter inferred credibility, changing monetary incentives should alter response strategy, and imposing a verified likelihood ratio should constrain posterior updates. Model fitting must use held-out data and sufficient variation to identify parameters Bullock (2009) Thaler (2024).
Network exposure adds another level of complexity. Individual-level belief changes aggregate through communication, selective retransmission, and platform delivery, yet network outcomes can differ from the average effect on a randomly assigned individual. A feed experiment can have a measurable effect on content composition without a corresponding aggregate attitude shift during the observation period. Models should specify the duration required for learning, whether sources are repeatedly encountered, and how individual changes propagate through social relationships Nyhan et al. (2023) Guess et al. (2019).
Human–Artificial Information Mediation in the Artificial Era
Artificial systems increasingly mediate encounters with political evidence through search summaries, conversational interfaces, recommendation systems, and automated content classification. In the Artificial Era, these are suitable empirical environments in which to study source cues, cognitive offloading, trust calibration, and changes in the structure of human information selection. A conversational system's fluency or personalized relevance may influence credibility judgments, even when its underlying evidentiary reliability must be evaluated independently Ecker et al. (2022) Nyhan et al. (2023).
A research design for human–artificial mediation could randomize whether an otherwise identical sourced claim is delivered by a human journalist, an algorithmic summary, or a conversational assistant. Researchers should verify factual equivalence, disclose the treatment honestly, measure baseline source expectations, and compare confidence updating, error detection, retention, and downstream sharing. Such studies would test mediation effects without treating output quality as proof of artificial consciousness, intentional motivation, or a psychological mechanism inside the system Bullock (2009) Pennycook & Rand (2019).
Important research questions concern whether adaptive explanations improve calibration; whether personalization amplifies the selection of congenial evidence; how provenance and citations affect appropriate trust; and whether diverse but credible perspectives reduce false factual certainty without increasing antagonism. These are hypotheses requiring direct experimentation. Existing social media studies give a methodological starting point, but their causal estimates cannot be transported wholesale to conversational artificial systems with different interfaces, incentives, and user expectations Nyhan et al. (2023) Bail et al. (2018).
Applications to Scientific Communication and Public Knowledge
The practical aim of research on political reasoning is to improve understanding of informational judgment and the design of trustworthy communication. Evidence-based approaches begin with a correct factual claim, a source whose methods can be inspected, and explicit treatment of uncertainty. Correction experiments indicate that people often revise factual beliefs when provided with reliable counterevidence. However, a factual estimate and a political preference may move independently, so communicators should select outcomes that match the intervention's stated purpose Wood & Porter (2019) Porter & Wood (2021).
Information diversity alone is not a fully specified intervention. Its effects depend on whether the material is accurate, comprehensible, civil, credible, and relevant, and on whether the recipient has a meaningful opportunity to evaluate it. Studies of opposing exposure and large feed changes reveal heterogeneous outcomes. Practical recommendations about cross-cutting content should therefore be tested against actual knowledge gains, trust calibration, and social consequences rather than inferred from an assumed monotonic relationship between ideological diversity and public agreement Bail et al. (2018) Nyhan et al. (2023).
Researchers can contribute to institutional practice by reporting correction effectiveness for concrete claims, measuring uncertainty alongside accuracy, and distinguishing education from persuasion. Neutral presentation is not achieved by concealing factual disagreements or assigning equal credibility to evidence of unequal quality. It is achieved through transparent methods for determining the claim's evidentiary support, consistent evaluation across political implications, and publicly understandable explanations of uncertainty. Psychological research can study how these institutional practices affect comprehension and trust Celniker & Ditto (2024) Ditto et al. (2025).
Competing Mechanisms and Discriminating Predictions
The main explanatory families make partially overlapping predictions about political information behavior. Directional motivated reasoning predicts that evidence favorable to a preferred conclusion receives more generous evaluation when motivational stakes are salient. Prior-based inference predicts divergent reactions when individuals enter a task with different likelihood assumptions. Analytic-engagement accounts predict improved discrimination under conditions supporting careful evidence assessment, while expressive-response accounts predict changes in reported political facts when incentives and audience expectations change Kunda (1990) Bullock (2009) Pennycook & Rand (2019).
The predictions can be jointly tested through a factorial experiment. Before the treatment, investigators elicit priors and preferred conclusions. They then present randomized factual evidence with independently validated strength, vary disclosed partisan alignment and source credibility, and manipulate private accuracy incentives. Repeated probability judgments yield estimates of updating and calibration. Under this design, a partisan-label interaction conditional on identical priors differs substantively from a purely prior-based response, and an incentive-dependent response shift differs from learning induced by new external evidence Bullock et al. (2015) Celniker & Ditto (2024) Thaler (2024).
An additional test observes what happens when participants freely search after the initial update. If source choice depends on both the changed posterior and the perceived social meaning of a result, the experiment can estimate how inference feeds back into later information selection. Including delayed follow-up separates immediate task effects from sustained learning. The full causal mechanism will be supported only if the separate stages align with the predicted process, not simply because a final attitude score moved in the hypothesized direction Hart et al. (2009) Taber & Lodge (2006).
Convergent Findings and Robust Limits
Across multiple lines of research, individuals frequently prefer congenial material, may scrutinize politically unwelcome claims differently, and can revise factual beliefs following trustworthy corrections. These effects do not imply that citizens reject all contrary information, that political identity fully determines perceived truth, or that motivated reasoning dominates every task. The weight of evidence depends on well-defined outcomes and on the separation of choices, attitudes, confidence, and behavior Hart et al. (2009) Taber & Lodge (2006) Walter & Murphy (2018).
Two contrasts are particularly important. First, accuracy incentives reduce some partisan gaps in familiar survey responses, while controlled tasks with unfamiliar partisan claims have found persistent differences despite such incentives. Second, exposure interventions can substantially alter the informational environment while producing little change on selected short-term polarization indices. These results require explanation through measurement, task structure, belief priors, treatment versions, and the broader social environment rather than a decision to discard whichever findings conflict with a favored theory Bullock et al. (2015) Stein et al. (2024) Nyhan et al. (2023).
The most responsible inference is mechanism-specific. Some distortions arise during selection, some during appraisal, and some during the reporting of an answer. Corrections can improve one component while leaving others intact. In particular, changing a false factual belief does not guarantee a change in ideological identity; receiving messages from an ideological opponent does not guarantee either persuasion or backlash. The field benefits from abandoning single-factor summaries in favor of testable intermediate stages Wood & Porter (2019) Bail et al. (2018) Ditto et al. (2025).
Open Scientific Problems and Priorities
Several unresolved questions determine the next stage of the research program. The first is how much asymmetric evaluation remains after careful adjustment for prior factual expectations and source-quality beliefs. The second is whether experimentally observed changes in confidence persist across weeks and months when daily information flows resume. The third is whether changes in real information selection causally alter durable knowledge, civic trust, and decision behavior. These questions are connected but require different empirical designs Celniker & Ditto (2024) Nyhan et al. (2023).
Measurement development should focus on probability judgments with calibrated scoring, explicit uncertainty, traceable information histories, and multiple outcome types. Cross-sectional partisan gaps offer a useful descriptive starting point, but they cannot identify updating rates without temporal data. Researchers should preregister the dimensions of congruence, source credibility, veracity, and decision stakes. They should also document stimulus materials, code, exclusions, missing data, and all planned interactions so independent scholars can reproduce the inferential process Bullock (2009) Thaler (2024).
A further frontier concerns new informational intermediaries. Artificial search and conversation systems modify the architecture of evidence access, and personalization can shape both selection and framing. Their effects on political reasoning should be studied using consent-based, neutral, preregistered protocols that hold factual content constant while varying provenance, explanation style, and user control. These interventions can be assessed for appropriate trust, accuracy and confidence calibration, understanding of disagreements, and durability. Claims about the psychological capacities of artificial systems require a separate research object and separate evidence Ecker et al. (2022) Nyhan et al. (2023).
An Integrated Methodological Agenda
A multi-method research agenda should begin with experimentally discriminating tasks and proceed to ecologically valid follow-up. Laboratory experiments can estimate the effect of evidence quality, partisan labeling, and incentives under controlled conditions. Panel surveys can document belief trajectories, while consented behavioral data can reveal what sources participants actually encounter and share. Large field interventions can identify whether changing information supply influences outcomes of genuine social significance Bail et al. (2018) Nyhan et al. (2023) Porter & Wood (2021).
The agenda also requires explicit theory comparison. Investigators can predefine likelihood-based, identity-motivated, accuracy-directed, and expressive-response models and evaluate predictive performance on held-out cases. To distinguish rational confidence from rationalization, a good experiment measures both what a person expected to be true before encountering evidence and what they wanted to be true. Mechanism claims should be assessed with independent process data, including response time, comprehension, source judgments, and subsequent choices, without treating any single measure as diagnostic Kunda (1990) Celniker & Ditto (2024).
Scientific knowledge advances when studies report inconvenient findings, methodological limits, and unsuccessful manipulations. A preregistered null result after a large change in news exposure is informative about that intervention's short-term causal effect. An observed difference under a novel-claim accuracy incentive is informative about a separate configuration of cognition and social meaning. These results belong in the same evidence structure, with their distinct estimands preserved. That structure offers more reliable cumulative understanding than rhetorical claims that one study settles the entire debate Nyhan et al. (2023) Stein et al. (2024).
Conclusion
Political reasoning is best understood as a structured sequence of information selection, credibility appraisal, inferential interpretation, and belief revision within socially consequential environments. Directional motivations, accuracy incentives, prior beliefs, source cues, and informational opportunity shape different stages of the process. The field's classic experiments established influential questions about asymmetric assimilation and motivated skepticism; subsequent work has clarified the need to distinguish prior-based judgments, expression, and cognitive engagement from specifically motivational distortion Lord et al. (1979) Kunda (1990) Taber & Lodge (2006) Bullock (2009).
The accumulated evidence supports meaningful preferences for congenial information, context-dependent partisan asymmetry in scientific argument appraisal, and the frequent efficacy of factual correction. It also demonstrates limits to simple inference. Accuracy incentives can change reported survey beliefs, but not all novel-claim judgments; cross-cutting exposure can have heterogeneous effects; and large shifts in a platform's exposure mix need not produce detectable short-term changes in polarization. These findings are mutually informative when their distinct experimental treatments, outcome definitions, and inferential limits are retained Hart et al. (2009) Bullock et al. (2015) Wood & Porter (2019) Nyhan et al. (2023) Stein et al. (2024).
A mature scientific account must identify which stages of political cognition are affected, determine the quality and diagnosticity of the available evidence, and track changes in knowledge independently of political identity and public expression. Its most productive future lies in longitudinal, cross-cultural, incentive-compatible, and ecologically valid tests that combine direct belief elicitation with observed information behavior. As Artificial Era systems increasingly mediate access to public evidence, these established methodological standards also provide a rigorous foundation for studying new forms of human–artificial informational interaction. The enduring scientific objective is accurate explanation of how people come to know, doubt, defend, and revise claims about collective life Ditto et al. (2025) Ecker et al. (2022).
References
- Bago, B., Rand, D. G., & Pennycook, G. (2020). Fake news, fast and slow: Deliberation reduces belief in false (but not true) news headlines. Journal of Experimental Psychology: General, 149(8), 1608–1613. https://doi.org/10.1037/xge0000729
- Bail, C. A., Argyle, L. P., Brown, T. W., Bumpus, J. P., Chen, H., Hunzaker, M. B. F., Lee, J., Mann, M., Merhout, F., & Volfovsky, A. (2018). Exposure to opposing views on social media can increase political polarization. Proceedings of the National Academy of Sciences, 115(37), 9216–9221. https://doi.org/10.1073/pnas.1804840115
- Bullock, J. G. (2009). Partisan bias and the Bayesian ideal in the study of public opinion. The Journal of Politics, 71(3), 1109–1124. https://doi.org/10.1017/S0022381609090914
- Bullock, J. G., Gerber, A. S., Hill, S. J., & Huber, G. A. (2015). Partisan bias in factual beliefs about politics. Quarterly Journal of Political Science, 10(4), 519–578. https://doi.org/10.1561/100.00014074
- Celniker, J. B., & Ditto, P. H. (2024). Of preferences and priors: Motivated reasoning in partisans' evaluations of scientific evidence. Journal of Personality and Social Psychology, 127(5), 986–1011. https://doi.org/10.1037/pspa0000417
- Ditto, P. H., Celniker, J. B., Spitz Siddiqi, S., Güngör, M., & Relihan, D. P. (2025). Partisan bias in political judgment. Annual Review of Psychology, 76, 717–740. https://doi.org/10.1146/annurev-psych-030424-122723
- Druckman, J. N., & McGrath, M. C. (2019). The evidence for motivated reasoning in climate change preference formation. Nature Climate Change, 9, 111–119. https://doi.org/10.1038/s41558-018-0360-1
- Ecker, U. K. H., Lewandowsky, S., Cook, J., Schmid, P., Fazio, L. K., Brashier, N., Kendeou, P., Vraga, E. K., & Amazeen, M. A. (2022). The psychological drivers of misinformation belief and its resistance to correction. Nature Reviews Psychology, 1, 13–29. https://doi.org/10.1038/s44159-021-00006-y
- Garrett, R. K. (2009). Echo chambers online? Politically motivated selective exposure among Internet news users. Journal of Computer-Mediated Communication, 14(2), 265–285. https://doi.org/10.1111/j.1083-6101.2009.01440.x
- Guess, A., Nagler, J., & Tucker, J. (2019). Less than you think: Prevalence and predictors of fake news dissemination on Facebook. Science Advances, 5(1), eaau4586. https://doi.org/10.1126/sciadv.aau4586
- Hart, W., Albarracín, D., Eagly, A. H., Brechan, I., Lindberg, M. J., & Merrill, L. (2009). Feeling validated versus being correct: A meta-analysis of selective exposure to information. Psychological Bulletin, 135(4), 555–588. https://doi.org/10.1037/a0015701
- Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480–498. https://doi.org/10.1037/0033-2909.108.3.480
- Lord, C. G., Ross, L., & Lepper, M. R. (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37(11), 2098–2109. https://doi.org/10.1037/0022-3514.37.11.2098
- Nyhan, B., & Reifler, J. (2010). When corrections fail: The persistence of political misperceptions. Political Behavior, 32, 303–330. https://doi.org/10.1007/s11109-010-9112-2
- Nyhan, B., Settle, J., Thorson, E., Wojcieszak, M., Barberá, P., Chen, A. Y., Allcott, H., Brown, T., Crespo-Tenorio, A., Dimmery, D., Freelon, D., Gentzkow, M., González-Bailón, S., Guess, A. M., Kennedy, E., Kim, Y. M., Lazer, D., Malhotra, N., Moehler, D., ... Tucker, J. A. (2023). Like-minded sources on Facebook are prevalent but not polarizing. Nature, 620(7972), 137–144. https://doi.org/10.1038/s41586-023-06297-w
- Pennycook, G., & Rand, D. G. (2019). Lazy, not biased: Susceptibility to partisan fake news is better explained by lack of reasoning than by motivated reasoning. Cognition, 188, 39–50. https://doi.org/10.1016/j.cognition.2018.06.011
- Porter, E., & Wood, T. J. (2021). The global effectiveness of fact-checking: Evidence from simultaneous experiments in Argentina, Nigeria, South Africa, and the United Kingdom. Proceedings of the National Academy of Sciences, 118(37), e2104235118. https://doi.org/10.1073/pnas.2104235118
- Stein, J., Keuschnigg, M., & van de Rijt, A. (2024). Partisan belief in new misinformation is resistant to accuracy incentives. PNAS Nexus, 3(11), pgae506. https://doi.org/10.1093/pnasnexus/pgae506
- Taber, C. S., & Lodge, M. (2006). Motivated skepticism in the evaluation of political beliefs. American Journal of Political Science, 50(3), 755–769. https://doi.org/10.1111/j.1540-5907.2006.00214.x
- Tappin, B. M., Pennycook, G., & Rand, D. G. (2021). Rethinking the link between cognitive sophistication and politically motivated reasoning. Journal of Experimental Psychology: General, 150(6), 1095–1114. https://doi.org/10.1037/xge0000974
- Thaler, M. (2024). The fake news effect: Experimentally identifying motivated reasoning using trust in news. American Economic Journal: Microeconomics, 16(2), 1–38. https://doi.org/10.1257/mic.20220146
- Walter, N., & Murphy, S. T. (2018). How to unring the bell: A meta-analytic approach to correction of misinformation. Communication Monographs, 85(3), 423–441. https://doi.org/10.1080/03637751.2018.1467564
- Wood, T., & Porter, E. (2019). The elusive backfire effect: Mass attitudes' steadfast factual adherence. Political Behavior, 41(1), 135–163. https://doi.org/10.1007/s11109-018-9443-y