What Is Selective Exposure?
Selective exposure is the tendency to choose information, sources, conversations, or media that agree with an existing belief, attitude, preference, identity, or decision. When a person can select among competing explanations, agreeable material may seem more attractive, useful, reassuring, or credible. Researchers call such material congenial information; information that questions a person's position is often called uncongenial or counterattitudinal information. The scientific question concerns the difference between the information a person could encounter and the information that person chooses to inspect. The strongest general synthesis finds a moderate average preference for congenial information, with important changes in its strength according to motivation, confidence, information quality, and task goals. Hart et al. (2009).
Consider a person who believes a particular investment is sound. When offered two equally accessible, equally relevant research reports, one supportive and one critical, the person may read the supportive report first, spend longer with it, or choose to share it. A similar pattern can arise in health-information searches, political news, product comparisons, and professional forecasting. Each case requires attention to what alternatives were genuinely available, whether their quality was comparable, and what the person's objective was. Seeking information that directly answers one's question can be rational even when the answer is congenial; scientific evidence of selective exposure comes from a preference that survives relevant comparisons or manipulations.
Selective exposure is an empirical construct in psychology and communication research. It is neither a clinical diagnosis nor a universal explanation for disagreement. It describes observable selection under particular opportunity structures. It may be measured through clicks, requested articles, time spent reading, willingness to receive challenging evidence, or the composition of an information diet. It also concerns choices made before information is interpreted; what people subsequently believe requires a separate measurement.
The Scientific Boundary: Selective Exposure and Related Concepts
Confirmation bias, motivated reasoning, and belief perseverance
Confirmation bias refers broadly to asymmetries in testing, seeking, interpreting, and sometimes remembering evidence. Selective exposure names one component: which information is encountered voluntarily. A researcher studying confirmation bias might present everyone with identical evidence and compare their judgments; a researcher studying selective exposure would ask which evidence each participant chooses to see. The distinction matters because a person can willingly read opposing arguments yet interpret them defensively, or avoid an argument while being capable of evaluating it accurately when required. Nickerson (1998).
Motivated reasoning describes how goals influence which beliefs are favored and how arguments are assessed. Motivated information choice is one pathway through which those goals operate. Belief perseverance concerns maintaining a conclusion after its original support is discredited. These processes can reinforce one another, but their observable outcomes and experimental tests differ. The classic political-judgment experiments of Taber and Lodge (2006) help explain why exposure to disagreeing arguments need not produce neutral evaluation.
Heuristics, fallacies, preferences, and symptoms
A heuristic is a strategy for reducing search or computational demands. Reading a trusted source first is often a useful heuristic. Selective exposure becomes a candidate bias when source congruence systematically dominates relevant signals of evidential value and leads to poorer goal-directed information acquisition. A logical fallacy, by contrast, is an inferential problem in an argument. An inaccurate probability judgment is evaluated against a specified statistical standard. Exposure selection is a behavior that can cause, amplify, or sometimes prevent such errors. Its observation alone establishes no psychiatric symptom, neurodevelopmental condition, or disorder.
Choosing supporting information is not the same as rejecting opposing information
Preference for agreement may arise through active approach rather than avoidance. Someone may eagerly click on familiar viewpoints while still being open to unexpected evidence. Online behavior studies distinguish reinforcement seeking from challenge avoidance, and that difference affects whether forcing additional opposing material into a feed is likely to alter beliefs. Garrett (2009) found evidence consistent with the importance of reinforcement seeking in a behavioral study of political-news users. A choice for a congenial item should therefore be decomposed into attraction to agreement, aversion to disagreement, and other determinants such as quality or convenience.
History and Development of Selective Exposure Research
The research tradition grew from studies of attitude change, consistency, persuasion, and postdecision information seeking. Cognitive-dissonance theory proposed that inconsistency among important beliefs or actions can produce an uncomfortable motivational state. After making a difficult decision, a person might prefer descriptions that affirm the chosen alternative. This framework established a historically influential hypothesis, while later research added accuracy motives, issue involvement, expected usefulness, identity, and information quality.
Early laboratory paradigms often invited a decision and then offered a menu of positive and negative claims about the selected option. Later studies asked about controversial attitudes, information search prior to choice, political-news sources, and online browsing. Moving from laboratory menus to actual platforms revealed a crucial complication: people encounter messages through social connections, institutional distributions, rankings, headlines, and recommendations before they click anything. Modern research accordingly separates available content, displayed content, selected content, and the evidence actually processed.
The most influential quantitative synthesis, Hart and colleagues (2009), brought together experimental findings to evaluate both defense and accuracy motives. Subsequent research on online media and digital trace data extended selective-exposure analysis to large-scale networks and naturalistic behavior. Large samples increase statistical precision, although sample size cannot by itself identify psychological causation. Bakshy, Messing, and Adamic (2015) and González-Bailón and colleagues (2023) explicitly distinguish several stages of the digital exposure process.
Why Agreeable Information Feels Attractive
Defense motives and the protection of existing commitments
An established belief can organize decisions, relationships, and self-understanding. Information that supports it preserves coherence and may reduce the immediate cost of revising commitments. When a belief is central to someone's values or group membership, challenging evidence can signal a broader threat than disagreement about a narrow fact. Defensive motivation predicts stronger congenial information selection in some high-commitment contexts. The meta-analysis of Hart et al. (2009) found that the magnitude of congruence preferences varied with circumstances associated with defending prior positions.
A defensive explanation still requires evidence. Researchers cannot infer a motive merely because a selected headline agrees with the reader's view. The same choice could reflect better prior information, genuine differences in source expertise, or greater relevance to an ongoing task. Studies strengthen causal inference by manipulating motivation or holding source characteristics equal while varying agreement.
Accuracy motives and the value of being corrected
People also want answers that improve performance. A clinician evaluating a possible diagnosis, an engineer testing a design, or an investor checking a forecast can benefit from disconfirming evidence. If accurate decisions are rewarded, counterattitudinal material may have greater expected value than another reassuring repetition. The Hart et al. (2009) synthesis found that, when uncongenial information was relevant to accomplishing a current goal, preference could reverse toward the challenging evidence.
The implication is substantive: congruence and informational value are separate variables. A person can remain strongly attached to a view while deliberately seeking the observation most likely to distinguish among competing explanations. In well-designed studies the analyst measures both the initial attitude and the respondent's incentive to learn, rather than assuming one is a substitute for the other.
Anticipated emotion, uncertainty, and self-relevance
Selecting information can regulate anxiety, surprise, embarrassment, or uncertainty. Anticipated unpleasantness may make a challenging source less attractive; curiosity or concern about future regret may make it more attractive. Effects depend on whether the person expects reading to help with the relevant goal. Confidence introduces a further distinction. Someone with little confidence may seek reassurance, whereas highly secure beliefs can sometimes tolerate challenge. These are testable conditional hypotheses rather than fixed personality labels.
Social identity, belonging, and reputational incentives
Information has a social function as well as a factual one. A news source can signal membership; a shared article can invite affirmation or criticism from others. Choosing an in-group outlet may therefore maximize social rewards even if it supplies little novel evidence. In political and identity-salient domains, selective exposure can interact with social sorting, partisan evaluations, and motivated argument processing. Interpretations require measurements of identity, source labels, article content, and audience expectations. Taber and Lodge (2006) provide foundational evidence of how prior attitudes color evaluation of politically relevant arguments.
How Selective Exposure Is Measured
Forced choice, optional sampling, and information menus
A common experiment first measures a participant's position and then presents a balanced menu of supporting and challenging items. Researchers record the items selected and the order of selection. A simple preference score might subtract the proportion of uncongenial items selected from the proportion of congenial items, but such a score is interpretable only when the categories were comparably numerous and accessible. More informative designs manipulate expected quality, perceived credibility, argument strength, and the costs of opening material independently of congruence.
Optional sampling tasks allow a participant to stop searching. They measure how many pieces of evidence are requested before a judgment, whether information selection changes after initial feedback, and whether the participant samples both hypotheses. An experiment can reveal an exposure preference without showing that it caused the final judgment. To establish downstream effects, investigators need to measure belief change, prediction accuracy, or choice quality after the selection stage.
Reading time, clicks, recalls, and self-report
Digital experiments measure clicks, scroll depth, exposure duration, subscription behavior, and revisits. Each proxy has limitations. A click may reflect outrage, accidental curiosity, or a misleading headline. Long dwell time may indicate careful consideration, confusion, or disengagement. Recollection is affected by memory and interpretation after exposure. Survey questions about preferred news outlets may also differ from behavioral logs because people misremember their habits and respond to social expectations.
A strong design triangulates multiple indicators, makes the available menu explicit, and checks whether attitude-congruence judgments are reliable. Researchers should distinguish availability, impression, attention, engagement, comprehension, and actual changes in belief. A platform's impressions are not equivalent to information actually read.
Experimental identification and causal inference
Randomly assigning source cues or item positions can help isolate the influence of apparent agreement. Randomly changing the choice menu distinguishes selection from the supply of information. An especially informative study varies source agreement while holding factual quality and article length constant. Observational analyses can estimate associations between prior beliefs and consumed content; causal claims require assumptions about confounding, selection, measurement, and interventions.
The first scientific step is to specify the estimand. Is the researcher asking whether agreement increases click probability, whether a ranking change alters exposure, or whether additional exposure changes polarization? The three questions belong to different causal stages and can yield different answers without contradiction.
When Is Selective Exposure a Cognitive Bias?
Selective exposure becomes normatively problematic when it impairs a well-defined information goal. The criterion may be the expected accuracy of a forecast, the ability to discriminate among hypotheses, the calibration of probabilities, or the expected utility of a consequential choice. There is no universal rule requiring equal time for every position. Evidence quality and prior probability matter: spending half one's time on a demonstrably unreliable claim can reduce accuracy. The proper benchmark is relevance-weighted evidential value, not numerical symmetry between opinions.
In a Bayesian account, prior beliefs should be revised according to the likelihood of observing the new evidence under competing hypotheses. A good information search preferentially obtains observations with high expected discriminative value, after accounting for search costs. If two sources provide independent evidence, the less congenial one may be more diagnostic; if it merely repeats low-quality misinformation, it may be rational to disregard it. The normative judgment depends on the likelihood model and decision costs. The behavioral question is whether subjects choose the most informative option given what they reasonably know.
Sometimes the decision goal is psychological comfort, social connection, entertainment, or identity expression. Congenial media consumption may serve that chosen goal. A scientific article must distinguish this descriptive preference from the narrower conclusion that the choice worsens factual understanding or decision performance.
What the Meta-Analytic Evidence Actually Shows
Hart et al. (2009) reported a moderate overall preference for congenial over uncongenial information, with standardized mean difference d = 0.36. This number summarizes effects across heterogeneous experimental designs. It is not a claim that 36 percent of people avoid disagreement, or that 36 percent of all clicks are politically aligned. Standardized mean differences compare conditions or preferences relative to variability under an operationalization; individual tasks cannot simply inherit the pooled effect as their expected result.
The same synthesis found moderation consistent with defense and accuracy accounts. Preferences became weaker in several settings associated with less defensive pressure, and counterattitudinal preference emerged when such information was important for a current accuracy goal. These findings justify describing a systematic average tendency and meaningful boundary conditions. They do not establish that every participant, political group, culture, or platform behaves identically.
Selective exposure has also been studied in laboratory comparisons, behavioral logs, and field interventions. Because these designs use different denominators and outcomes, their effect sizes should be reported in their own units. A percentage of cross-cutting posts in a network is not commensurate with a randomized difference in attitudes, and neither can be substituted for a laboratory d estimate.
The Online Information Pipeline
What is available is different from what is displayed
Digital exposure emerges through several filters. First, a user's friends, subscriptions, and followed accounts determine a pool of potentially available content. Next, algorithms rank and display a subset. Then the user may click or scroll past it. Finally, the person may interpret, remember, or share the message. Networks, platforms, and individuals all influence what is ultimately encountered. Selective exposure is most directly reflected in voluntary selection; algorithmic ranking is a separate mechanism that interacts with it.
In an observational study of 10.1 million Facebook users, Bakshy et al. (2015) estimated the effects of social connections, algorithmic presentation, and users' own choices on contact with ideologically diverse news. The study found that the composition of social relationships and click behavior were important filters, with individual click choice playing a larger role than the ranking algorithm in reducing exposure to cross-cutting content under the study's definitions. The platform population, ideological classifications, and observational design limit generalization to all social media or later recommendation systems.
Large-scale segregation evidence
Using aggregated 2020 election data from roughly 208 million US Facebook users, González-Bailón et al. (2023) compared political news users could potentially encounter, information they were shown, and content with which they engaged. Ideological segregation was substantial and increased across those stages. The research also documented asymmetries in the observed political-news ecosystem. It does not establish a universal psychological preference in every user, and its measurement of political content does not represent all topics within a social network.
Echo chambers and filter bubbles are different claims
An echo chamber is typically used to describe an information environment in which similar views are repeatedly reinforced and external perspectives are discounted. A filter bubble often emphasizes personalization or algorithmic curation. Selective exposure is an individual-level selection mechanism that can contribute to either environment, but neither concept follows automatically from one congruent click. People can inhabit a socially homogeneous network while remaining curious about outsiders, or live in a diverse network and still open only agreeable links. Garrett (2009) illustrates why reinforcement and avoidance should be evaluated separately.
Does More Disagreeing Content Change Beliefs?
The downstream effect of exposure is an empirical question. Choosing more diverse material may increase awareness of arguments, but the relationship between exposure and trust, accuracy, and polarization depends on content, source legitimacy, emotional tone, social cues, and interpretation. Information can be challenged, ignored, misunderstood, or evaluated against an existing belief without changing that belief.
In a social-media field experiment, Bail et al. (2018) offered users an opportunity to follow automated accounts sharing messages from political opponents. Some measured responses moved toward greater rather than lower polarization, particularly in parts of the analyzed sample. The treatment combined disagreement, political source cues, repeated exposure, and a particular platform environment; it is evidence against assuming that cross-cutting content always depolarizes, not evidence that all intellectual disagreement is harmful.
A larger randomized field experiment reported by Nyhan et al. (2023) reduced exposure to like-minded sources by approximately one-third among 23,377 Facebook users during the 2020 US election. The manipulation increased exposure to cross-cutting sources but produced no measurable effects on eight preregistered attitude outcomes, including affective polarization and ideological extremity. The study places useful limits on claims that a single exposure change will rapidly transform political beliefs. It does not prove that all interventions, longer horizons, or other psychological outcomes have zero effects.
These findings support a causal separation between exposure diversity and belief change. A successful intervention should report whether it altered the intended exposure variable and whether that alteration improved the specified judgment or behavioral outcome. Altering the first is a meaningful technical result; claims about the second require their own evidence.
Everyday Contexts and Decision Consequences
Health, medical information, and risk assessment
Health information can become identity-relevant when people have invested in a diagnosis, treatment preference, or interpretation of symptoms. They may seek reassuring accounts rather than estimates of benefit and harm. In clinical decision-making, a stronger evidence strategy is to compare treatment outcomes, methodological quality, and applicability to the patient's circumstances before considering whether a conclusion feels congenial. The existence of selective exposure cannot diagnose a patient or determine whether any treatment is appropriate. Medical decisions should be made using qualified professional guidance and current clinical evidence.
Consumer judgment and financial decisions
A consumer may watch favorable product reviews after choosing a brand and skip reports of failure. An investor may concentrate on optimistic forecasts after buying an asset. Researchers would need to separate selective exposure from the fact that people legitimately prioritize sources with demonstrated expertise or relevance. For decisions with meaningful financial consequences, useful practices include recording falsification conditions, assessing base rates, and examining alternative outcomes before increasing commitment. These are decision procedures; specific improvements require context-sensitive evaluation.
Relationships, groups, and public discussion
People often choose interlocutors whose values and experiences are familiar, a preference with social and emotional benefits. The scientific risk arises when this pattern reduces access to diagnostic evidence or creates unwarranted certainty about outsiders. Group-level effects emerge from interpersonal networks as well as personal choices. One cannot infer the presence of selective exposure from disagreement itself, nor conclude that balanced source consumption guarantees tolerance or accurate public judgments.
Workplaces, expert panels, and legal reasoning
Teams evaluating a hypothesis may over-sample confirmatory reports and neglect failure cases. A well-designed investigation predefines competing explanations, asks which observation would discriminate among them, and keeps an auditable record of information considered. In high-stakes work, a second reviewer can be assigned the task of evaluating the strongest credible alternative rather than merely opposing the first analyst. The established cross-disciplinary research on cognitive bias in criminal investigations illustrates the importance of distinguishing general information selection from applied forensic safeguards.
Selective Exposure in Human–AI Interaction
Search engines, recommender systems, and generative AI assistants add another layer to information choice. A person can choose a tool because it tends to agree, phrase a prompt that invites confirmation, or prefer answers that affirm an earlier decision. A conversational system can also adapt to user preferences and repeatedly present similar perspectives. The relevant evidence questions are which sources are made available, which are surfaced, which are selected, and how the interaction changes eventual beliefs or decisions. Existing studies of social-media ranking establish useful methodological distinctions but cannot establish the magnitude of selective exposure in every modern AI assistant.
One design goal is contestable assistance: show evidential grounds, make relevant alternatives accessible, distinguish verified findings from uncertain hypotheses, and allow users to inspect sources directly. Another is to evaluate responses for answer quality independent of how pleasing users find them. These are proposed safeguards that require product-specific experimental validation. Research on AI as authority and automation bias addresses a neighboring but distinct phenomenon: trust in a system's outputs. The broader Artificial Era provides a philosophical horizon for these interactions; evidence about selective exposure remains an empirical matter of behavior and system design.
Individual Differences, Culture, and Context
Belief strength, confidence, issue relevance, motivation, and perceived expertise can change the probability of choosing congenial material. Some conditions favor defense, while others favor accuracy-driven exploration. The existence of a relationship in one sample does not warrant treating a person's ideology, nationality, gender, age, or personality category as a reliable diagnostic marker of selective exposure. Effects must be estimated at the level of particular tasks and populations.
Cross-cultural research requires careful measurement equivalence. The same source label may signal expertise in one setting and group identity in another; the available media ecology may change the cost of obtaining alternative views. National differences in trust, platforms, or political institutions also influence source availability. Researchers should test whether an apparent effect is explained by exposure opportunities and source cues before treating it as a universal cultural trait. Generalizability remains more limited than the wide use of the term sometimes suggests.
Methods for Improving Information Selection
Define an information goal before beginning a search
The first useful intervention is to decide what the information is for: predicting an outcome, comparing treatments, choosing a product, understanding a controversy, or maintaining social connection. When the task concerns accuracy, explicitly state the unknown quantity and what would count as persuasive evidence. This reduces the tendency to equate a reassuring headline with valuable information. The contrast between defense and accuracy motives has direct empirical support in the selective-exposure meta-analysis. Hart et al. (2009).
Search for diagnostic evidence, not merely an opposing opinion
Construct at least two plausible hypotheses and identify observations that would favor one over the other. An information source has value when its contents could meaningfully change a probability estimate or decision. A strong contradictory study is usually more informative than a low-quality contrarian essay. A systematic evidence review should preserve the distinction between evidence strength and ideological balance.
Compare source quality using common criteria
Before opening the material, assess methods, provenance, transparency, relevant expertise, and possible conflicts of interest according to a consistent rubric. Apply the same rubric to congenial and uncongenial sources. A process that demands rigorous evidence only from challenges has introduced a different form of asymmetry at the interpretation stage. The research on motivated reasoning explains why equal exposure alone cannot guarantee equal evaluation.
Use precommitment and structured review
For consequential decisions, document which evidence would cause a change of mind, specify thresholds before seeing results, and designate a review stage for credible failure cases. Teams can use independent assessment or a structured challenge to the prevailing plan. These techniques follow plausible decision-theoretic principles, but their effectiveness in reducing real-world selective exposure has to be tested in the actual workflow. Laboratory improvement does not automatically establish long-term transfer to news consumption, clinical practice, or organizational performance.
Evaluate interventions by downstream outcomes
The large Facebook field trial of Nyhan et al. (2023) shows why increasing exposure diversity is an intermediate outcome, not proof of reduced polarization. A rigorous evaluation should measure changes in which content is seen, accuracy on a specified task, calibrated confidence, robustness of choices, and persistence over time. A short intervention can be useful without transforming political identity; successful debiasing should be stated at the level actually demonstrated.
Limitations and Open Scientific Questions
A laboratory menu can make congenial and uncongenial options artificially symmetrical. Real-life sources differ in accuracy, cost, prior reliability, and relevance. Researchers sometimes classify congeniality using political categories that imperfectly approximate a person's beliefs. Other studies use recalled exposure or self-reported preference, neither of which measures actual information consumption perfectly. Observational platform data can include enormous samples while still leaving causal pathways uncertain. Experimental manipulations can identify causal effects within a setting while limiting external validity.
Publication practices and analytic flexibility also deserve scrutiny. Any claim that selective exposure is universal or that a particular intervention eliminates it requires preregistered comparisons, appropriate uncertainty intervals, and independent replication. The Hart et al. (2009) synthesis predates much of the present social-media ecosystem, while later large-scale research pertains to specific platforms and historical election conditions. Reading them together gives a more precise account than substituting one era's methods for another's.
The distinction among psychological mechanism, network structure, ranking algorithm, and individual click choice is an enduring research priority. Another is the relation between exposure and persuasion. A person can seek disagreement because they expect to rebut it, and a person can encounter disagreement without considering it seriously. Measuring attention, comprehension, trust, and subsequent decisions alongside exposure can reveal which pathway operates.
Frequently Asked Questions
Why do people seek information that agrees with them?
Supporting information can affirm an existing belief, protect an important commitment, feel more useful, or come from a familiar source. People also pursue accuracy and can prefer challenging information when it is clearly relevant to a goal. The relative influence of these motives depends on task, confidence, and context. Hart et al. (2009).
Is selective exposure the same as confirmation bias?
Selective exposure concerns the selection of information to encounter. Confirmation bias is a broader construct covering searching, testing, interpreting, and sometimes remembering evidence. Selecting contrary evidence and later dismissing it is an example of why these stages must be measured separately.
Does selective exposure mean someone is irrational?
Only when behavior is assessed against a defensible goal-specific benchmark can an inference about decision quality be made. Preference-consistent reading may be recreational, socially valuable, or based on superior information quality. It is problematic for an accuracy goal when it systematically displaces more diagnostic evidence without sufficient justification.
Do social-media algorithms cause selective exposure?
Algorithms affect what is displayed, while friends and subscriptions affect what is available and users choose which material to engage with. Observational data document contributions from these stages, but the causal effect of any particular algorithm must be identified experimentally or through credible natural experiments. Bakshy et al. (2015).
Is reading opposing views enough to reduce political polarization?
No universal effect has been demonstrated. Different field studies find heterogeneous results, including a randomized reduction in like-minded exposure with no detectable improvement in eight measured attitudes. The content, duration, audience, and outcome matter. Nyhan et al. (2023); Bail et al. (2018).
Can selective exposure be measured without asking people what they believe?
Behavioral logs can record what people click or consume, but a defensible measure of congeniality typically still requires some estimate of prior position and source content. Researchers can obtain that estimate experimentally, from premeasured attitudes, or through carefully validated indicators. None is free from classification uncertainty.
Is a preference for reliable sources selective exposure?
Choosing sources because they have demonstrated accuracy is an epistemically useful selection strategy. It becomes potentially biased if agreement with a preferred conclusion is substituted for source reliability or if reliability criteria are applied inconsistently.
How can AI assistance reduce one-sided information searches?
Systems can make high-quality contrary evidence discoverable, expose uncertainty, document the provenance of claims, and prompt the user to specify what would change a decision. These are design proposals requiring tool-specific evaluations. An agreeable answer alone cannot determine whether the user's evidence search is balanced or accurate.
Related Articles
Confirmation Bias: Why We Favor Supporting Evidence — The broader family of asymmetric evidence search, interpretation, and recall.
Motivated Reasoning: How Goals Shape What We Believe — How desired conclusions and identity-related goals influence evidence evaluation.
Cognitive Bias in Criminal Investigations: Tunnel Vision and Confirmation Bias — An applied domain in which narrow evidence search can affect high-stakes judgments.
AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making — Technology-mediated credibility judgments and the distinct question of overreliance on machine outputs.
References
Bail, C. A., et al. (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.
Bakshy, E., Messing, S., and Adamic, L. A. (2015). Exposure to ideologically diverse news and opinion on Facebook. Science, 348(6239), 1130–1132. https://doi.org/10.1126/science.aaa1160.
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.
González-Bailón, S., et al. (2023). Asymmetric ideological segregation in exposure to political news on Facebook. Science, 381(6656), 392–398. https://doi.org/10.1126/science.ade7138.
Hart, W., Albarracín, D., Eagly, A. H., Brechan, I., Lindberg, M. J., and 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.
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175–220. https://doi.org/10.1037/1089-2680.2.2.175.
Nyhan, B., et al. (2023). Like-minded sources on Facebook are prevalent but not polarizing. Nature, 620, 137–144. https://doi.org/10.1038/s41586-023-06297-w.
Taber, C. S., and 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.