RESEARCH DOSSIER
Electoral Judgment, Candidate Evaluation, and Voting
Abstract. Electoral judgment comprises the psychological processes through which citizens appraise candidates and parties, integrate policy and performance information, form electoral preferences, and decide whether and how to vote. This monograph critically synthesizes foundational electoral research, cognitive and social-identity models, candidate impression experiments, campaign-persuasion trials, turnout field experiments, comparative candidate-evaluation research, and recent randomized studies of digital and artificial-mediated political communication. It distinguishes candidate favorability, stated vote intention, reported vote, validated turnout, and observed electoral returns as different dependent variables. The evidence shows enduring roles for party identification, identity-motivated judgment, issue and performance information, rapid social inferences, civic norms, resources, and election context. Randomized evidence finds substantial heterogeneity: some campaign contacts have negligible average effects on vote choice, whereas particular informational messages and mobilization treatments can produce measurable effects. The analysis reconstructs historical theory, reviews identification and psychometric limitations, examines cultural and institutional transportability, and outlines unanswered questions about digital information, artificial interlocutors, and behavioral validation. This is a critical narrative synthesis of published scholarship, not a registered systematic review, original experiment, or new pooled meta-analysis.
Keywords: electoral judgment; candidate evaluation; voting behavior; vote choice; voter turnout; party identification; partisan identity; issue voting; retrospective voting; political heuristics; candidate competence; campaign persuasion; get-out-the-vote; behavioral field experiments; validated turnout; political cognition; political psychology; Artificial Era
Introduction: Elections as Psychological Decisions
Electoral judgment is the organized psychological activity through which a person interprets political alternatives, evaluates candidates, assigns significance to party and policy information, forms a preference, and acts—or abstains—when an election offers an opportunity to vote. Its empirical objects include the judgments individuals report, the information they encounter and remember, their evaluations of candidates, the decisions they state before voting, and their behavior as represented in validated turnout or election returns. An election is a temporally bounded institution embedded in a much longer history of identity, trust, social relations, and political experience. Explaining its results therefore requires models of both relatively stable predispositions and contextually altered information, motivation, and opportunity Campbell et al. (1960) Blais (2000) Van Bavel & Pereira (2018).
Voting research examines several distinct outcomes. Candidate favorability expresses an evaluation and may vary without a change in vote intention. Reported intention concerns a prospective choice and may be revised. Survey-reported voting is a retrospective account susceptible to imperfect memory and socially desirable answers. Administrative turnout indicates whether a recorded individual participated, usually without disclosing which candidate received the ballot. Official returns aggregate ballots cast under the jurisdiction's election rules. Each variable answers a different question, and the causal mechanisms that move one variable may leave another unchanged. Treating all five as interchangeable measures of influence obscures the field's most important experimental disagreements Gerber & Green (1999) Kalla & Broockman (2018) Allcott et al. (2024).
This research monograph is a critical narrative synthesis of original election studies, randomized interventions, observational research, theoretical books, and contemporary methodological developments. It analyzes the formation of electoral evaluations, the temporal dynamics of campaigning, the psychology of deciding to participate, and the interpretation of evidence across different methods. Its central questions concern what citizens know, how candidate impressions are acquired, when partisan and policy cues are influential, which interventions actually change behavior, how outcomes should be validated, and what contemporary artificial communication adds to the research agenda. The monograph reports no newly collected participants, new quantitative effect synthesis, or preregistered systematic-review procedure Lau & Redlawsk (1997) Broockman & Kalla (2023) Lin et al. (2025).
Constructs, Units of Analysis, and Measurement Boundaries
A candidate evaluation is an appraisal of a specific office seeker along one or more dimensions such as perceived competence, integrity, warmth, policy agreement, trustworthiness, or leadership. Party identification is an enduring or intermittently salient relationship with a political party, while expressive partisan identity denotes the self-relevance and emotional significance of belonging. Issue preference is a respondent's stance toward a policy, whereas issue importance concerns how much disagreement on that policy affects an electoral choice. Performance evaluation concerns perceived consequences of prior governing action. These constructs can covary strongly, but rigorous tests distinguish their measurement and model their contribution to a particular electoral decision Huddy et al. (2015) Alvarez & Morrier (2025) Healy & Malhotra (2009).
Turnout is participation in a specified election under a specified denominator: eligible residents, voting-age population, registered voters, or a research sample of registrants. Differences in denominators change turnout rates even if the number of ballots remains identical. Individual turnout can be measured through administrative voter files or reported in a survey; the administrative record may offer better behavioral validity while having incomplete matches, jurisdictional differences, or registration gaps. Candidate choice is ordinarily private and must be measured differently from individual turnout. Election-level vote share additionally reflects who was eligible, mobilized, registered, and present. An ecological association between precinct attributes and votes does not reveal the preference of every person in that precinct Verba et al. (1995) Gerber et al. (2008) Allcott et al. (2024).
A scientific account of electoral judgment requires explicit temporal order. Predispositions and candidate familiarity exist before treatment; campaign messages can change beliefs or activation; turnout requires subsequent capacity and opportunity; the observed election reflects candidate supply, institutions, and external events. Survey experiments estimate responses to controlled information among recruited respondents, field experiments estimate behavior under practical conditions, and panel studies trace within-person change while retaining threats from attrition and time-varying confounding. Election return regressions can model aggregate correlations but require additional assumptions for individual causal inference. The analysis consequently separates theoretical representations, measured predictors, experimental contrasts, and verified electoral behavior Arceneaux & Nickerson (2009) Broockman & Kalla (2023) Allcott et al. (2024).
The Historical Formation of Modern Electoral Research
Modern electoral behavior research emerged from the interplay of survey research, social psychology, economics, and formal political theory. The American Voter, first published in 1960 by Angus Campbell, Philip Converse, Warren Miller, and Donald Stokes, used the Michigan tradition of election studies to explain how stable party attachments, orientations to candidates, and political issues structure choice. Its major contribution was a framework in which particular campaign experiences act upon a preexisting organization of loyalties and perceptions. Historical findings about mid-twentieth-century U.S. electorates should be read in their institutional and media setting; the model's constructs remain useful, while their distributions and effects must be measured afresh in later populations Campbell et al. (1960).
An economic and formal-theoretic tradition asked whether voting can be understood through anticipated benefits, policy proximity, information costs, and the possibility that a ballot changes an outcome. The 1968 calculus-of-voting analysis by William Riker and Peter Ordeshook made explicit the difficulty of explaining widespread turnout when the chance of a decisive vote is extremely small. The extended account treated civic duty and related motivations as part of the decision, opening questions about norms, expressive rewards, and social expectations. André Blais later examined the empirical strengths and limits of rational-choice explanations across electoral settings. This tradition offered analytical clarity about incentives but also showed why the psychological meaning of political participation matters Riker & Ordeshook (1968) Blais (2000).
Research subsequently developed methods that directly change electoral environments and observe downstream behavior. In the 1990s, studies of information shortcuts and candidate choice explored how imperfectly informed citizens might nevertheless select a candidate consistent with their considered interests. By the end of that decade, randomized canvassing experiments measured turnout using voting records instead of relying exclusively on retrospective interview reports. Twenty-first-century studies added increasingly large field trials, experimentally manipulated candidate profiles, digital exposure, and naturalistic platform experiments. This methodological expansion replaced some broad claims about campaign effectiveness with more specific estimates tied to populations, outcomes, treatments, and time horizons Lau & Redlawsk (1997) Gerber & Green (1999) Kalla & Broockman (2018).
Party Identification, Loyalty, and the Michigan Framework
The Michigan framework treats party identification as a psychological orientation that can organize what political information means to an individual. A familiar party label may provide a usable summary of prior experiences, perceived political coalitions, and expectations about policy. When such identification is strong, information about candidates is rarely encountered as entirely new data: it arrives in a field of associations, affective expectations, and social meanings. This property helps explain why campaign messages often change immediate candidate impressions more readily than they change strongly consolidated electoral commitments. At the same time, party identification is a variable whose strength and meaning must be established empirically in each electoral system Campbell et al. (1960) Broockman & Kalla (2023).
The expressive account of partisanship emphasizes identification with the party as a social group. Huddy, Mason, and Aarøe compared partisan identity with policy and ideological measures using four studies and experiments, finding that strongly identified partisans reacted emotionally to threats and reassurances concerning their party's electoral standing. Those responses informed campaign involvement. Party identity can consequently shape participation through belonging and emotionally salient group outcomes, not only instrumental calculation about policies. An identity-based model of political belief further proposes that group commitments influence attention, interpretation, and selective endorsement of information. Such models need separate tests of identity salience, perceived accuracy, behavioral conformity, and candidate choice Huddy et al. (2015) Van Bavel & Pereira (2018).
The field must also distinguish a party heuristic from unconditional partisan rigidity. A shortcut can be informative when party labels reliably forecast governing positions; it can become misleading when candidates diverge from party expectations or when information is inaccurate. In four experiments conducted during the 2020 U.S. presidential campaign, Broockman and Kalla found that messages about less familiar candidates were more persuasive and that more informative messages were more influential. The pattern supports a role for voters' prior information and updating alongside their partisan allegiances. It does not establish that every voter always updates accurately; it shows that information content and baseline candidate familiarity help explain where persuasive effects arise Broockman & Kalla (2023).
Formal Choice, Civic Duty, and the Participation Decision
Electoral choice and electoral participation are analytically separate. A citizen may possess an unambiguous preference yet face barriers to casting a ballot; another may be strongly motivated to participate while remaining uncertain between candidates. Formal turnout theory highlights the small probability that one ordinary ballot is pivotal in a large election. This makes social obligation, expressive satisfaction, institutional confidence, practical opportunity, and the subjective value of participating central research concerns. Riker and Ordeshook's calculus provided a formal statement of this puzzle, while Blais tested the applicability of rational-choice and duty-centered explanations. A useful contemporary model treats perceived costs and benefits as psychological variables shaped by the institutional setting and social experience Riker & Ordeshook (1968) Blais (2000).
The civic voluntarism model expands the explanation by examining resources, psychological engagement, and recruitment. Verba, Schlozman, and Brady documented unequal opportunities for political participation associated with time, money, civic skills, organizational experience, and invitations to participate. These factors constrain whether a preference is enacted and whether a political voice receives institutional representation. The model also clarifies why turnout inequalities matter scientifically: the pool of observed voters is selected from a larger eligible population through processes that are themselves psychologically and socially patterned. Analyses confined to voters can therefore produce a distorted account of the political evaluations held by nonvoters Verba et al. (1995).
Research on mobilization confirms that a voting intention is an incomplete behavioral endpoint. Randomized field experiments have altered face-to-face contact, mailing language, and planning prompts, sometimes producing measurable shifts in validated turnout. The causal effect belongs to the particular treatment, electorate, and election conditions and cannot be generalized from the existence of a statistically significant intervention to a universal property of all campaign contact. Social pressure, contact quality, preexisting turnout propensity, and the salience of the election are relevant moderators. Mechanisms must be inferred from direct measures and comparison conditions, rather than read backward from a single observed percentage-point change Gerber & Green (1999) Arceneaux & Nickerson (2009) Gerber et al. (2008).
Retrospective Accountability and the Meaning of Performance
Retrospective voting models propose that citizens use perceptions of previous governing performance to evaluate incumbents, parties, and candidates. A record of economic conditions, public services, or crisis response may be interpreted as evidence of competence or responsibility, yet judgments can conflate policy decisions with events outside an officeholder's control. Achen and Bartels developed a wide-ranging critique of idealized democratic accountability using historical electoral patterns and the limitations of voters' causal attribution. Their argument makes electoral competence an empirical problem concerning the connection between perceived outcomes, actual responsibility, and feasible alternatives, rather than a presumption that voting always reflects informed policy comparison Achen & Bartels (2017).
Healy and Malhotra investigated disaster policy and electoral returns and found that voters rewarded relief expenditures more visibly than investments in disaster preparedness. The study's central inference concerns short-run responsiveness to easily attributed benefits and weaker reward for prospective risk reduction. It invites psychological questions about salience, temporal discounting, retrospective memory, and attribution of responsibility. Because the design links observational political outcomes with governmental choices, interpretation must consider how disasters, spending, incumbent characteristics, and constituent demands are related. Its contribution is a theoretically important challenge to simple performance-accountability assumptions, rather than an experimental measure of an isolated mental process Healy & Malhotra (2009).
Evidence of retrospective judgment can be interpreted through several competing mechanisms. Some voters may reasonably use coarse indicators when individual policy contributions are difficult to identify; others may apply partisan priors to interpret an ambiguous outcome; still others may react to immediately experienced hardship irrespective of policy responsibility. Longitudinal surveys can measure changes in attribution, while experiments can manipulate identical performance information and vary the assigned decision maker. Administrative and archival data help identify external shocks, but they cannot by themselves reveal private thought. A mature theory therefore integrates external performance, perceived responsibility, available alternatives, and political identity into an explicit account of candidate evaluation Campbell et al. (1960) Healy & Malhotra (2009) Taber & Lodge (2006).
Candidate Appraisal as Person Perception
Political candidates are evaluated simultaneously as prospective officeholders and as perceived social persons. Voters can form judgments about competence, honesty, warmth, leadership, familiarity, representativeness, and policy alignment, often without access to equivalent information about every candidate. A verbal description, face, accent, party label, or occupational history may activate expectations that affect later interpretation of more detailed evidence. This places candidate appraisal within general psychological processes of person perception and social categorization, while its institutional stakes and repeated campaign context make it a distinct research object. The scientific question concerns which specific cues shift which evaluation dimensions and whether those shifts affect an actual electoral decision Todorov et al. (2005) Bauer (2017) Kim & Kweon (2024).
Appraisal instruments frequently ask how favorably respondents feel toward a candidate, how competent or trustworthy a candidate seems, or how likely they are to vote for that person. These measures vary in scale construction, response format, information provided, and contextual comparators. A mean change in favorability may occur among individuals with stable vote intention, while vote intention can change without corresponding changes across every trait rating. Experimental manipulations should measure candidate recognition, prior party identity, policy knowledge, and the strength of preexisting preference. Otherwise, a short-term response to unfamiliar candidate profiles may be mistaken for an effect likely to occur among highly familiar national figures Broockman & Kalla (2023) Bauer (2017).
Evaluations may contain causal chains. A party cue can shape assumptions about issue positions; those assumptions can change perceived integrity; a global impression can then alter willingness to consider contrary information. Alternatively, a personal encounter or discrete record of public action may change the initial appraisal and later affect partisan choice. Testing such mediation requires separate measurement of the intervening variables, an intervention that plausibly changes them, and careful treatment of post-treatment confounding. The observation that two scales move together offers weaker evidence than a design that varies one cue while controlling the others Taber & Lodge (2006) Van Bavel & Pereira (2018) Broockman & Kalla (2023).
Rapid Facial Inference and the Limits of Visual Signals
An influential experiment by Todorov, Mandisodza, Goren, and Hall asked participants to infer candidate competence from photographs, with extremely brief exposure in some conditions. Competence judgments predicted outcomes of U.S. congressional races above chance; one reported analysis correctly anticipated the winner in 68.8% of the examined 2004 Senate contests. The psychological finding is that observers can form rapid trait impressions from facial appearance that correlate with a politically consequential outcome. It does not establish that the faces reveal actual governmental competence, nor that visual impressions caused the full electoral result. Candidate selection, prior publicity, and correlated physical characteristics require continued examination Todorov et al. (2005).
Later observational work considered how attractiveness and visible emotion interact. Masch, Gassner, and Rosar analyzed candidates in the 2017 German federal election and reported associations between physical attractiveness, facial expressions, and vote shares. This extends the methodological question beyond static facial structure to photographed social behavior. Yet facial expression is influenced by media selection, setting, photographer, and political role. The election-level association does not isolate voter exposure or determine whether appearance precedes support in a causal sequence. Controlled stimulus experiments, naturalistic exposure records, and actual behavioral outcomes are needed to connect these components Masch et al. (2021).
The correct interpretive distinction is between perceived trait, actual trait, and predictive cue. A face may evoke a competence impression without being a valid test of competence; a repeated television image may contribute to familiarity; an appealing photograph may correlate with professional campaign resources. When evaluating such studies, researchers should predefine stimuli and randomize display conditions, distinguish individual judgments from aggregate elections, test rival explanations, and report how generalization changes across age, gender, ethnicity, and political context. Visual heuristics are a real component of electoral judgment, while their weight relative to party labels, policy and experience is a contingent empirical question Todorov et al. (2005) Masch et al. (2021) Campbell et al. (1960).
Gender, Class, Stereotypes, and Candidate Representation
Candidate evaluations are partly organized by socially learned beliefs about the characteristics associated with leadership and group membership. Gender-linked expectations can attach to policy competence, authority, warmth, emotional expression, or interpersonal style. A consequential research design distinguishes reactions to the candidate's demographic category from reactions to explicitly presented qualifications, messages, and issue positions. Nichole Bauer used experiments to show that strategies emphasizing counterstereotypic leadership competence can improve evaluations of women candidates across trait dimensions. The result underscores that stereotype effects depend on message content and evaluative context rather than operating as a fixed penalty in every election Bauer (2017).
Class and gender can interact. In conjoint experiments conducted in the United States and the United Kingdom, Kim and Kweon found that candidate class background could have different evaluative consequences for women and men, with working-class backgrounds particularly disadvantaging women in their designs. Conjoint methods vary multiple characteristics and estimate average differences across displayed profiles; their strength lies in reducing confounding among randomized attributes. Their transfer to actual elections is limited by the supplied candidate pool, realism of the profiles, information that voters ordinarily encounter, and the difference between hypothetical selection and an official ballot Kim & Kweon (2024).
Interpretation also requires attention to who appears in the candidate pool before voters encounter them. Party nomination, financing, media visibility, social networks, and occupational inequalities condition which demographic and policy combinations are available. The estimate from an experimental profile is therefore a conditional statement about response to a specified candidate, not a complete explanation of underrepresentation. Evidence is more informative when research links impression formation with recruitment, primary elections, candidate experience, and institutional rules. Cross-country tests help reveal which biases follow broadly shared social categorizations and which depend on local party systems and political histories Verba et al. (1995) Bauer (2017) Kim & Kweon (2024).
Issue Voting, Policy Agreement, and Salience
An issue-based electoral judgment compares a candidate's stated or inferred position with a voter's policy preferences. Such comparison presupposes that the voter identifies the issue, understands relevant alternatives, and treats the difference as electorally important. Issue importance should not be inferred automatically from a high level of polarization or from the respondent's intensity of agreement. An issue can be deeply contested in public debate while carrying modest weight in an individual's actual choice; another, less visible policy can prove decisive. Candidate-policy experiments offer a method for estimating the contribution of agreement when candidate attributes and issue statements are systematically varied Alvarez & Morrier (2025) Campbell et al. (1960).
Alvarez and Morrier introduced a conjoint-based approach to issue importance in a study using a sample of registered U.S. voters following the 2022 congressional midterm elections. They examined how agreement or disagreement with candidate positions influenced hypothetical electoral choices across issues, reporting a stronger relationship between measured importance and partisan polarization than between importance and policy-position polarization. Their evidence helps separate individual choice weight from the broader ideological spread of an issue. The temporal setting, issue menu, candidate profiles, and self-selection into survey participation delimit the result's generality Alvarez & Morrier (2025).
Policy proximity is one model among several. A voter may choose the candidate nearest to their position, prefer credible competence over proximity, prioritize a single high-stakes issue, or rely on a party label as a proxy for a complex policy package. Some people emphasize anticipated outcomes; others evaluate expressive alignment with values and social groups. These models imply different experimental signatures. Presenting identical platforms under different party labels tests information carried by identity; experimentally varying tradeoffs among issues tests compensatory choice; presenting uncertainty about implementation tests credibility judgments. A useful theory specifies which information is believed, which dimension is weighted, and how the vote is translated into action Campbell et al. (1960) Alvarez & Morrier (2025) Blais (2000).
Political Heuristics, Information Costs, and Decision Quality
Information about candidates is costly to acquire, difficult to verify, and frequently incomplete. A heuristic is an economical rule that uses a limited cue to guide inference or choice. Party identification, incumbency, endorsements, occupational history, personal appearance, and retrospective performance can all serve as heuristics. Their value depends on how reliably the cue predicts an outcome the voter cares about. Studying heuristics requires measuring both the cognitive savings they provide and the errors they can introduce under misleading or rapidly changing conditions. A shortcut's mere presence therefore cannot determine whether the resulting choice was well informed Lau & Redlawsk (1997) Todorov et al. (2005) Campbell et al. (1960).
Lau and Redlawsk introduced an operational account of voting correctly, defining it relative to a fuller representation of voters' interests and information. Their experimental work suggested that some participants using limited information could select alternatives consistent with those fuller-information benchmarks. The result is theoretically important because it recognizes adaptive information processing under realistic constraints. The benchmark nevertheless embeds assumptions about stable values, which facts would count as fully informing someone, and how preferences ought to be aggregated. Its findings should be described as performance relative to a constructed counterfactual criterion, not as an independent determination of an objectively correct political choice Lau & Redlawsk (1997).
An adequate decision-quality study compares the same individuals' judgments under altered information conditions and explains where corrections emerge. Field settings can examine whether newly encountered facts change vote intention, while laboratory or survey tasks can identify which cues dominate a controlled comparison. Cross-study generalization depends on whether candidates are familiar, issue preferences strong, and available alternatives genuine. Recent evidence that less familiar candidate information can have larger effects is compatible with a rational informational account, yet this does not establish that all shortcuts improve choice. Researchers should report both the effect of each cue and the uncertainty surrounding its informational validity Broockman & Kalla (2023) Lau & Redlawsk (1997).
Motivated Skepticism, Evidence Assessment, and Belief Updating
Political judgment involves both the appraisal of incoming claims and the regulation of conflicts between new information and established commitments. In experiments on contested political issues, Taber and Lodge found that participants evaluated congruent arguments more favorably and engaged more strongly with counterarguments against discordant positions. They also documented selective exposure when participants could choose sources. These findings established experimentally tractable mechanisms of motivated skepticism and attitude reinforcement. The constructs concern the evaluation of arguments in particular issue domains and should not be equated automatically with verified behavior at an election Taber & Lodge (2006).
An identity-based model holds that partisan affiliations can shape how people process and endorse political claims. It describes a pathway by which belonging and group-valued outcomes influence judgments of plausibility and credibility. Such a model generates predictions about differential source trust, defensive processing under partisan threat, and stronger retention of identity-congruent arguments. Nonetheless, a voter may correctly recognize that prior knowledge varies in reliability, and some source skepticism is epistemically appropriate. Experimental work must therefore distinguish a general tendency to reject contrary information from discriminating between high- and low-quality evidence Van Bavel & Pereira (2018) Taber & Lodge (2006).
Broockman and Kalla's large 2020 survey-experimental program supplies an important counterweight to uniformly rigid accounts of partisanship. Across four experiments and 291 messages, the study found variation in persuasion associated with substantive information and prior candidate familiarity, including evidence that some messages changed candidate evaluations and expressed choices. Prior attitudes still matter because a strongly held and well-informed preference often requires substantial counterevidence to move. The synthesis is that motivated evaluation and informational updating can coexist within the same electorate and must be identified by designs that isolate the informational quality, credibility, timing, and identity relevance of competing messages Broockman & Kalla (2023) Taber & Lodge (2006).
From Preference to Participation: The Behavioral Threshold
A prospective preference enters the electoral record only when a person is eligible, registered where necessary, able to act, motivated to do so, and presented with a workable voting procedure. This sequence creates several independent failure points. A citizen may become more favorable toward a candidate while remaining a nonvoter, or may vote because of civic duty without any campaign-induced preference change. A complete causal account must represent the opportunity structure alongside motivation, executive planning, habit, social identity, and practical constraints. The psychological study of voting accordingly joins decision research with behavioral implementation and the political sociology of resources Blais (2000) Verba et al. (1995) Nickerson & Rogers (2010).
Administrative turnout is a powerful measure because it identifies an enacted behavior with less dependence on self-description. Yet the binary act contains little information about the voter's choice or subjective reasons. A treatment that raises verified turnout could mobilize supporters of multiple candidates; it need not increase any one candidate's vote share. Electoral returns reveal aggregate political consequences but generally cannot connect ballots to identified participants. Researchers therefore design distinct tests for mobilization and persuasion, establish the unit of randomization, and estimate direct or spillover effects at the level actually observed. Failure to separate these outcomes has historically led to exaggerated assertions about the effectiveness of campaign interventions Gerber & Green (1999) Gerber et al. (2008) Kalla & Broockman (2018).
Participation has a temporal structure: some obstacles arise during registration, some during information gathering, and others immediately before a polling opportunity. The psychological effect of an intervention may depend on whether it arrives before a plan is formed, during an uncertain decision, or after a strong habit has already been activated. A planned ballot may still be disrupted by work, care responsibilities, illness, transportation or procedural barriers. These factors imply that intervention studies need baseline behavioral propensity, contextual information, and an explicit follow-up window. A single final turnout record is valuable, but interpreting why it changed requires intermediate measures Arceneaux & Nickerson (2009) Nickerson & Rogers (2010) Verba et al. (1995).
Canvassing, Direct Contact, and Randomized Field Evidence
Gerber and Green's 1999 study randomized nonpartisan face-to-face canvassing in New Haven around the 1998 election and used subsequent turnout as its dependent variable. The paper reported a sizeable increase associated with personal canvassing, helping establish the modern field-experimental agenda for voter mobilization. Random allocation reduces the chance that differences in motivation or political interest explain the estimated treatment effect, although actual contact may still vary by who is reachable. Interpretation must distinguish assignment to a contact attempt from successful receipt and from any causal estimate conditional on contact, since these are not equivalent estimands Gerber & Green (1999).
Personal contact can communicate practical information, recognition, social attention, and a norm of participation. A simple comparison between contacted and uncontacted individuals cannot establish which component accounts for the outcome: people who answer their doors can differ systematically from people who do not. Randomized field studies therefore compare groups assigned to different contact protocols, and can vary message content when statistical power permits. Their external validity is similarly conditional on local election stakes, institutional arrangements, field-worker fidelity, and the electorate's familiarity with contact campaigns. A positive effect under one nonpartisan canvassing design is evidence for that intervention, not proof that every political contact persuades voters Gerber & Green (1999) Arceneaux & Nickerson (2009).
Research design also raises questions about spillovers. Conversations within households may affect people never directly contacted; field organizers can change behavior as they visit neighborhoods; and social expectations may spread along interpersonal ties. Conventional individual randomization identifies an average effect more clearly when interference between treated and control units is limited. Cluster allocation, household-level records, or social-network designs can help test such processes, but introduce their own precision and balance challenges. The key achievement of turnout field experiments is the connection of a preassigned manipulation to a recorded civic behavior, with causal interpretation anchored in the specific design and assumptions Gerber & Green (1999) Gerber et al. (2008).
Implementation Intentions and the Psychology of Voting Plans
A person can sincerely intend to vote and fail to carry out that intention. The implementation-intention approach proposes that specifying the circumstances and steps of an intended action can support execution. Nickerson and Rogers tested a voting-plan intervention during the 2008 U.S. presidential election, contrasting prompts to form a concrete plan with alternative calls. The randomized experiment involved a large sample, and the reported effects were concentrated in particular household circumstances. The result links a well-defined self-regulatory mechanism to a verified civic outcome, making implementation planning an important bridge between electoral motivation and turnout Nickerson & Rogers (2010).
The intervention's design makes contextual detail central to interpretation. A prompt can activate existing knowledge about where and when to vote, expose an unresolved logistical problem, or cue an already established routine. In the published study, the benefit was greater among contacted people living in single-eligible-voter households, whereas scripts showed different effects in multi-eligible-voter households. This interaction is compatible with the idea that household routines already provide cues or coordination in some settings. The experiment does not demonstrate that planning has the same magnitude of effect in elections with different voting rules, mail-ballot systems, or levels of administrative difficulty Nickerson & Rogers (2010).
Mechanism research should assess plans directly, rather than inferring plan formation from exposure to a script. Measures can record the specificity of time, location, transport, contingency arrangements, and subsequent completion; however, asking about these details may itself change behavior. Designs comparing otherwise similar questions can help distinguish information provision, commitment, self-prediction, and concrete implementation. The relevant psychological outcome is successful execution of an intended civic action under real constraints. A model that incorporates implementation improves on analyses that assume stable vote intention reliably predicts the final administrative participation record Nickerson & Rogers (2010) Blais (2000).
Mobilization Heterogeneity and Election Salience
Electorates are not homogeneous in their likelihood of voting. Some registered citizens participate almost automatically, others have strong barriers or little connection to the election, and many occupy a condition of genuine uncertainty. Arceneaux and Nickerson reanalyzed eleven randomized face-to-face turnout experiments and proposed that mobilization is concentrated around a participation threshold. Their findings suggested that the response of baseline low-propensity voters differs between prominent elections and less salient contests. The contribution is an interaction model: treatment effects depend on the relationship among prior propensity, campaign context, and available motivation rather than on a fixed population-wide responsiveness Arceneaux & Nickerson (2009).
Such heterogeneity complicates comparisons between studies. A recruitment experiment conducted in a high-turnout presidential election may involve very different baseline behavior than one conducted in a local primary. An absolute percentage-point increase has a different statistical ceiling and theoretical meaning when initial turnout probability is low, moderate, or already near one. Selection into contact and into study participation further alters the estimand. Researchers should report intent-to-treat effects, treatment receipt, subgroup definitions established before analysis, uncertainty intervals, and whether moderation claims survive independent replication. The eleven-experiment reanalysis offers a model for interpreting conflicting results through baseline risk Arceneaux & Nickerson (2009) Gerber & Green (1999).
Mobilization research also needs to distinguish political consequence from civic participation. An intervention may increase turnout evenly among competing groups, increase it disproportionately among one group, or leave aggregate candidate shares almost unchanged because of offsetting shifts. Administrative turnout alone cannot establish electoral advantage. Candidate preference surveys can complement turnout records but introduce their own measurement error and response selection. The strongest studies make the target estimand explicit: the probability of voting, the allocation of votes, the subjective experience of participation, or their separate causal pathways. Maintaining these distinctions protects scientific conclusions from being overstated in political commentary Kalla & Broockman (2018) Gerber et al. (2008).
Participation Inequality, Recruitment, and Institutional Constraints
Participation is distributed unequally along dimensions of resources, access, and social incorporation. The civic voluntarism framework developed by Verba, Schlozman, and Brady identifies time, money, civic skills, engagement, and recruitment as ingredients of political voice. These resources influence more than the initial wish to participate: they change how easily people can learn about candidates, obtain information, navigate procedures, and coordinate voting with other duties. An analysis that treats observed voters as a neutral cross-section of eligible citizens obscures inequalities produced before the ballot is cast Verba et al. (1995).
Resource explanations and psychological explanations interact. People who repeatedly navigate a voting system acquire routines, confidence, and contacts; those whose schedules or living circumstances frequently change may need to rebuild the entire sequence of action. Civic organizations and interpersonal networks can reduce effort while making participation socially meaningful. Conversely, distrust or previous administrative difficulty can decrease expectations that another attempt will be worthwhile. Cross-sectional correlations between resources and turnout cannot identify all these mechanisms, but multiwave panels and natural or randomized institutional changes can help separate established habits from contemporaneous barriers Verba et al. (1995) Blais (2000) Nickerson & Rogers (2010).
Comparative inference requires careful attention to the electoral system itself. Registration, ballot design, compulsory-voting rules, polling accessibility, voting methods, language availability, and election competitiveness vary across jurisdictions. These conditions shape both the opportunity to vote and the meaning of a campaign intervention. The same social-pressure message, planning question, or contact protocol can have different effects where participating is easy, difficult, habitual, or institutionally mandatory. The global study of electoral psychology therefore should map each observed mechanism onto the institution that permits or constrains its expression, rather than exporting U.S. estimates as universal properties of civic behavior Blais (2000) Verba et al. (1995) Arceneaux & Nickerson (2009).
Campaign Persuasion and the Difference Between Exposure and Effect
A campaign can expose citizens to messages without changing what they believe, whom they prefer, or whether they vote. The appropriate causal question is the effect of a defined treatment against a defined counterfactual on a defined outcome at a specified time. Exposure, message recognition, knowledge gain, candidate favorability, vote intention, and actual ballot behavior form a sequence of possible changes; none is an automatic proxy for another. A message may correct a factual misunderstanding and leave the final candidate preference intact, or produce an immediate preference difference that diminishes before election day. Research must follow the outcome relevant to its claim and avoid treating all exposure as persuasion Kalla & Broockman (2018) Broockman & Kalla (2023).
Kalla and Broockman evaluated evidence from forty existing field experiments and nine additional experiments of campaign contact in U.S. general elections. Their article reported an average persuasive effect near zero on candidate choice for the examined general-election interventions, with important qualifications involving unusual candidate positions and the timing of measurement. The result concerns the tested campaigns, treatments, contexts, and vote-choice measures; it does not establish that political information never influences judgment. It provides a benchmark for the distinction between campaign activity, short-term message processing, and durable changes in candidate selection under ordinary general-election conditions Kalla & Broockman (2018).
Why might a plausible message have a small measured average effect? Many eligible voters already possess extensive partisan and candidate information; information may be redundant; treatment might reach people unlikely to change their preference; or effects could differ in direction and cancel when averaged. Experiments can investigate these explanations by measuring baseline familiarity, information content, timing, and the distribution of individual responses. However, searching for responsive subgroups after observing aggregate null effects invites multiple-comparison error. Reliable inference requires prospectively specified moderation analyses, independent replication, and measurement of persistence beyond the immediate post-message survey Kalla & Broockman (2018) Broockman & Kalla (2023).
When New Candidate Information Changes Judgment
Broockman and Kalla's study of the 2020 U.S. presidential election examined four survey experiments with a combined sample exceeding one hundred thousand participants and a large set of campaign messages. The researchers found that voters were more persuadable by information about less familiar candidates and that informational messages could change both candidate evaluations and stated vote choice. Their interpretation emphasizes a quasi-Bayesian mechanism in which people incorporate new signals in proportion to what they already believe and know. Because these were survey-experimental outcomes, the evidence for changes in expressed preference is firmer than any claim about actual secret ballots cast at a later date Broockman & Kalla (2023).
The contrast between general-election field experiments and informational survey experiments is substantively productive. Different treatments were delivered in different contexts, at different times, to different samples, with different outcome measurements. A small average field effect on ultimate candidate choice and a detectable survey effect from some novel information can both be true. The scientific explanation must specify the treatment's informational novelty, recipient familiarity, credibility, persistence, and measurement horizon. A model that merely labels the electorate persuadable or unpersuadable sacrifices the explanatory power gained by distinguishing those features Kalla & Broockman (2018) Broockman & Kalla (2023).
A rigorous synthesis consequently separates two research questions: whether a new proposition can alter an individual's evaluation, and whether a real campaign can deliver such information reliably enough to change the election-relevant behavior of an entire population. The first question is approachable through carefully controlled messages and pre/post measures; the second requires evidence about reach, self-selection, message competition, contact fidelity, and time to election. Neither can be answered conclusively by campaign spending or advertisement exposure alone. Evidence consistent with informational updating leaves open how misleading claims, counter-messages, and source reputation alter the same process Broockman & Kalla (2023) Taber & Lodge (2006).
Media Environments, Selective Exposure, and Aggregate Outcomes
Electoral communication takes place within news markets, interpersonal networks, and platform systems whose distribution mechanisms shape what citizens actually encounter. Observational studies can relate media access to vote share, but such analyses face selection effects: politically interested or ideologically aligned audiences may seek particular outlets. DellaVigna and Kaplan examined the staggered entry of Fox News into local U.S. cable markets and analyzed changes in electoral outcomes, arguing that access to a particular media outlet could be related to measurable differences in vote share. Their study illustrates both the value of quasi-experimental exposure variation and the difficulty of separating access, viewership, credibility, and message content DellaVigna & Kaplan (2006).
Selective exposure can operate before any message is interpreted. Individuals may choose identity-congruent sources, avoid disagreeable claims, or encounter algorithmically prioritized material. Experiments on motivated skepticism show why selection and evaluation should be studied together: a preferred source can increase the chance of seeing a claim and also affect how persuasive it seems. Yet media availability is not identical to use, and a platform impression is not identical to agreement with its content. Strong media research therefore tracks a sequence of exposure, reception, belief, evaluation, and electoral action, with independently measured variables at each stage Taber & Lodge (2006) Van Bavel & Pereira (2018).
Political information systems can also generate broad indirect effects through conversation, agenda setting, or changes in opportunities to participate. These are often difficult to estimate because media environments are shared and interference can occur between nominal treatment and control participants. Natural experiments and randomized platform interventions offer informative contrasts but may identify different populations and treatment bundles. A causal explanation must say whether it concerns access to an outlet, a ranking rule, a particular factual message, or interpersonal amplification. Treating those as one intervention hides the mechanisms needed for credible cross-study comparison DellaVigna & Kaplan (2006) Allcott et al. (2024).
Randomized Social-Media Deactivation and Electoral Behavior
A major randomized study of the 2020 U.S. election investigated the consequences of temporarily deactivating Facebook or Instagram. Allcott and numerous collaborators recruited large samples of platform users and assigned participants to different deactivation windows. The reported analysis found a reduction in an index of participation driven particularly by online activity, while measured effects on candidate favorability, voter turnout, issue polarization, and affective polarization were close to zero. The findings caution against inferring from extensive digital exposure that removing one platform necessarily produces large short-term changes in electoral choice Allcott et al. (2024).
The study's design increases causal credibility for the intervention actually administered, yet its scope remains specific. Participants had to be willing to deactivate their accounts under the research arrangements, deactivation lasted a limited period, and the political context was a highly salient U.S. presidential election. The intervention simultaneously removed multiple types of content, interaction, and social reinforcement, which makes attribution to a single algorithm or message impossible. Some outcomes relied on surveys while others used different measurement sources. These facts limit claims about the long-run effects of social media on political identity, information habits, or future elections Allcott et al. (2024).
The resulting comparison with prior communication research illustrates an important hierarchy of claims. Exposure correlations can suggest plausible pathways; randomized account deactivation can estimate the total effect of an altered platform-use condition; message experiments can isolate content properties; and turnout validation establishes whether a civic behavior occurred. None provides, by itself, a complete mechanism for every political outcome. A scientific synthesis should retain both the potentially meaningful role of media institutions and the evidence that some high-profile interventions produce small or null measured shifts on specific electoral endpoints DellaVigna & Kaplan (2006) Broockman & Kalla (2023) Allcott et al. (2024).
Conversational Artificial Systems and Electoral Judgment
Interactive artificial language systems introduce a distinctive communication setting because a citizen can ask questions, request explanations, and receive sequentially adapted responses. This creates new experimental questions about source attribution, accuracy, perceived responsiveness, conversational trust, and the persistence of judgments made during a dialogue. A dialogue exposes participants to information whose sequence is partly determined by the preceding interaction. Studies must therefore record prompts, model outputs, factual correctness, experimental assignment, and the specific outcome scale if they are to distinguish factual learning from candidate preference change Lin et al. (2025) Broockman & Kalla (2023).
Lin and colleagues reported preregistered experiments involving conversations with language models in the context of the 2024 U.S. presidential election and the 2025 Canadian and Polish elections. Randomized interactions were associated with changes in candidate preferences, with effects that the authors compared to conventional video-advertising benchmarks. The study also reported factual inaccuracies in some model outputs. The experimental findings are evidence about electoral attitudes under the assigned interaction conditions; they do not establish that future large-scale deployment will generate corresponding changes in verified turnout or official vote shares. The article is particularly important because it connects an emerging communication medium with previously established distinctions among persuasion, information quality, and political identity Lin et al. (2025).
Artificial-era electoral research should examine systems as instruments within identifiable human–machine interactions. The relevant scientific variables include what information is supplied, which sources are cited, how uncertainty is represented, whether a participant can challenge claims, and how the conversation changes confidence and preference. A model can function as an information retrieval interface, a conversational explanation system, or an explicitly persuasive counterpart; these are distinct treatments whose effects should not be pooled indiscriminately. The intellectual horizon of Artificial Era invites longitudinal comparison of human and artificial information processing, while the present evidence supports carefully bounded claims about measured human responses to specific technological systems Lin et al. (2025) Allcott et al. (2024).
Comparative Electoral Settings and Transportability
Much classic voting research centers on the United States, yet electoral decisions occur within diverse institutional and cultural arrangements. Candidate-centered presidential contests, party-list parliamentary elections, proportional representation, and systems with compulsory participation differ in the information provided to voters and the meaning of individual candidacies. A party label may carry especially strong information in one system, while personal qualifications or district service may dominate another. Translating psychological constructs across these systems requires measurement equivalence and clear attention to what the ballot actually asks citizens to decide Campbell et al. (1960) Blais (2000).
Comparative experiments can test the transportability of candidate evaluations more directly than uncontextualized generalizations. Kim and Kweon's U.S.–U.K. conjoint comparison illustrates how social categories and perceived qualifications can be studied across two national contexts. Lin and colleagues provide evidence from several contemporary electoral settings involving conversational AI. These studies examine different mechanisms but share a central methodological demand: the experimental stimulus must have a locally meaningful interpretation, and measured preference change must be separated from actual voting behavior. Parallel results across countries strengthen generalization only where sampling, measurement, and institutional comparability are demonstrated Kim & Kweon (2024) Lin et al. (2025).
Age, political experience, economic inequality, minority status, media ecology, and civic institutions may moderate effects within nations as much as national boundaries do. A validated treatment in one population is a starting hypothesis for another. Transportability analyses should describe treatment availability, baseline familiarity, group identity distributions, competing information, and institutional conditions. Researchers should estimate heterogeneity with sufficient precision rather than using demographic categories as explanatory mechanisms by themselves. The broader scientific objective is a conditional theory that identifies the circumstances in which electoral information, social expectation, and practical opportunity change candidate judgment or voter participation Verba et al. (1995) Arceneaux & Nickerson (2009) Alvarez & Morrier (2025).
Survey Research, Panels, and Candidate-Choice Experiments
Survey studies remain central to electoral psychology because they record political preferences, candidate perceptions, party attachment, information exposure, and participation intentions in the respondent's own terms. Their inferential quality depends on sampling, mode of recruitment, item wording, nonresponse, and whether apparently equivalent scales retain the same meaning across populations. A candidate thermometer, a vote-choice question, and a list of preferred policy positions measure different constructs. Panel surveys follow respondents across a campaign and make within-person changes observable, but attrition, panel conditioning, and changing response composition can mimic movement in public opinion. A repeated cross-section better tracks population distributions under stable sampling assumptions, while a true panel better observes particular persons; neither automatically identifies the cause of a change Campbell et al. (1960) Achen & Bartels (2017) Broockman & Kalla (2023).
Randomized survey experiments can disentangle information components by assigning respondents to otherwise comparable candidate descriptions or campaign messages. Conjoint designs independently vary several candidate attributes within structured profiles, yielding estimates conditional on the attributes, levels, and choice task provided to participants. Such studies are especially useful when social categories, issue positions, and qualifications would ordinarily be correlated. Their outcomes are nonetheless hypothetical choices made within a research interface. A conjoint estimate is informative about how participants evaluate the presented alternatives, while transport to registered candidates on a real ballot depends on prior familiarity, partisan attachments, public scrutiny, and institutional incentives Kim & Kweon (2024) Alvarez & Morrier (2025) Bauer (2017).
Interpretive precision improves when researchers report stimulus materials, preanalysis decisions, sample recruitment procedures, weighting, attrition, and the exact wording of their outcome items. Results from heterogeneous political contexts require examination of measurement equivalence, including whether a familiar U.S. distinction such as partisan identification has a comparable social meaning elsewhere. Where experiments contain many messages or attributes, investigators should account for dependence among observations and distinguish preplanned estimands from exploratory contrasts. The purpose of these disclosures is to enable researchers to reproduce the original inference and assess what a survey task can reveal about electoral judgment beyond the questionnaire itself Broockman & Kalla (2023) Alvarez & Morrier (2025) Lin et al. (2025).
Field Experiments, Treatment Assignment, and Behavioral Validation
Randomized field experiments observe how an intervention changes behavior within an actual electoral environment. Their foundational strength is the assignment of a contact attempt, reminder, or communication condition before the outcome occurs. An intent-to-treat effect estimates the consequence of being assigned to the intervention; an effect among successfully contacted participants involves additional assumptions because reachability is selective. Administrative turnout records provide a behaviorally meaningful endpoint, though matching errors, registration rules, and uneven record coverage remain relevant. Individual ballot selections are protected by ballot secrecy, so most turnout experiments cannot directly measure an individual's candidate vote Gerber & Green (1999) Gerber et al. (2008) Nickerson & Rogers (2010).
Effective design separates assignment, exposure, intermediate belief, stated preference, and verified action. The campaign field experiments reviewed by Kalla and Broockman concern candidate-choice persuasion, whereas many get-out-the-vote trials concern participation. They cannot be combined under an unspecified single effect of campaigns. Baseline candidate knowledge, election competitiveness, the contact medium, and time to election can change both receipt and impact. Treatment variation across households or neighborhoods may also create interference, requiring assumptions or allocation designs that accommodate spillovers. The general lesson is to define the outcome and unit of treatment before interpreting the magnitude of an estimated effect Kalla & Broockman (2018) Arceneaux & Nickerson (2009) Gerber & Green (1999).
Administrative validation introduces practical and ethical obligations. Matching respondent identities to records requires appropriate permission, secure handling, and clear safeguards for sensitive political data. File records may indicate participation but should not be presented as evidence of how the voter marked a secret ballot. Unmatched observations may differ systematically from matched cases, and data infrastructures vary across jurisdictions and election cycles. Sensitivity analyses should explore plausible matching errors and missing outcomes. An intervention evaluated only among easily contacted and matched respondents may not represent its effects among the wider eligible electorate Gerber et al. (2008) Verba et al. (1995) Allcott et al. (2024).
Causal Inference, Confounding, and Competing Explanations
Electoral research frequently encounters feedback between identity, information, and political activity. A partisan orientation can influence media choice; media use can reinforce the orientation; highly engaged individuals both seek campaign information and vote more often. Observing a relation between media use and candidate preference therefore leaves direction and common causes unresolved. Randomization helps identify an assigned intervention's effect, while observational strategies require clearly stated assumptions concerning omitted variables, selection, and timing. Instrumental variation in media access may be useful if its exclusion assumptions are credible; an election regression with many controls cannot guarantee those assumptions by itself DellaVigna & Kaplan (2006) Van Bavel & Pereira (2018) Allcott et al. (2024).
Statistical precision does not substitute for construct validity. A highly significant increase in a survey thermometer can be causally identified and still have limited significance for actual votes. Conversely, an imprecisely estimated change in validated turnout may matter substantively but remain inconclusive because the experiment enrolled too few participants. Researchers should report effect sizes, confidence intervals, baseline risks, prespecified primary outcomes, and the distinction between statistically supported null findings and estimates compatible with multiple important effects. Meta-analytic comparisons need an explicit decision about which studies, populations, outcomes, and follow-up intervals are commensurable Kalla & Broockman (2018) Broockman & Kalla (2023) Allcott et al. (2024).
Candidate studies introduce additional confounding through candidate selection and unequal political visibility. More attractive or familiar candidates may also possess greater experience, funding, or media exposure, and voters may infer competence from those correlated characteristics. Experiments that randomize isolated visual or descriptive cues can estimate a component of the judgment process, whereas observational vote-share correlations do not establish that the component caused the election outcome. The validity of a research synthesis depends on preserving this difference between prediction and causal explanation. The same principle applies to retrospective accountability, where external conditions and governmental spending jointly determine what voters observe Todorov et al. (2005) Masch et al. (2021) Healy & Malhotra (2009).
Replication, Statistical Heterogeneity, and Evidence Integration
Election studies differ systematically in their samples, periods, electoral institutions, and treatment intensities. An intervention effective in a low-salience local contest may have little measurable effect in a presidential election with saturated candidate information; an unfamiliar candidate may be more susceptible to novel information than a universally recognized incumbent. This heterogeneity makes a grand universal campaign effect difficult to interpret. Kalla and Broockman's field-experiment synthesis, Broockman and Kalla's information-message experiments, and Arceneaux and Nickerson's reanalysis of turnout interventions illustrate different ways of limiting or explaining apparent contradictions through treatment type, outcome selection, baseline propensity, and timing Kalla & Broockman (2018) Broockman & Kalla (2023) Arceneaux & Nickerson (2009).
Replication requires access to questionnaires, messages, assignment rules, turnout-matching procedures, model specifications, and ideally deidentified data consistent with privacy protections. Independent studies should reproduce an effect under closely matched conditions before treating broader cultural transport as established. Where a literature permits quantitative synthesis, researchers should consider selection bias, the dependence of multiple estimates within one campaign, between-study variation, and whether a pooled effect conceals directionally different responses. The present monograph is a narrative scientific synthesis and calculates no original pooled estimates. Its conclusions concern the consistency and limits of published research designs, not a newly estimated meta-analytic quantity Gerber & Green (1999) Kalla & Broockman (2018) Lin et al. (2025).
Contradictory findings can also follow from different observation windows. A campaign message may alter an immediate judgment that decays, while a voting plan can be effective only when its cues remain accessible at the moment of action. Social-media deactivation studies change an ecosystem of interactions over a defined interval, which is distinct from showing one message in a survey. Future replication programs should measure exposure, knowledge, candidate evaluation, behavioral intention, and validated turnout where feasible on a common time scale. This would allow tests of how often a shift at one stage passes through to the next and where attenuation occurs Nickerson & Rogers (2010) Broockman & Kalla (2023) Allcott et al. (2024).
Ethics, Privacy, and the Integrity of Electoral Research
Election research has a duty to protect participants from unnecessary political exposure, privacy risks, and misleading descriptions of voting confidentiality. Social-pressure turnout experiments demonstrate that anticipated observation by others may affect behavior, but they also bring into view the social costs of reputational manipulation. Research design and governance should distinguish publicly accessible participation information from secret ballot choice, avoid unauthorized release of identifiable records, and assess risks unevenly borne by communities with histories of intimidation. Electoral experiments can be scientifically informative while requiring careful institutional and ethical review Gerber et al. (2008) Verba et al. (1995).
Studies involving artificial political dialogue add questions about factual accuracy, disclosure of automated authorship, controlled access to model outputs, and auditability of adaptive conversations. A participant who engages with an artificial interlocutor may encounter inaccuracies even when the system is intended to provide verified information. Reporting should distinguish factual corrections, candidate advocacy, and general decision assistance; identify the systems used; and preserve relevant interaction conditions for replication without exposing private political beliefs. The randomized evidence from human–AI political dialogue makes these safeguards materially relevant to future experiments rather than merely speculative design considerations Lin et al. (2025).
Scientific independence also matters where findings could influence campaign allocation or public interpretation of elections. Researchers should disclose relevant funding, treatment goals, registration and reporting decisions, and whether public outcomes were independently verified. Election legitimacy requires accurate communication about what studies demonstrate: a change in reported preference is not evidence that a particular election result was manipulated, and a deactivation experiment with null effects cannot establish that every digital platform is politically irrelevant. Ethical interpretation is therefore intertwined with causal precision and transparent uncertainty Allcott et al. (2024) Kalla & Broockman (2018) Lin et al. (2025).
Integrated Psychological Organization of Electoral Judgment
The evidence supports a multistage account of electoral judgment. People enter a campaign with political identities, prior knowledge, values, lived circumstances, and social ties. Candidates and parties supply information of varying novelty and credibility, while institutions and media environments influence the information encountered. Voters generate impressions of candidates, interpret performance and policy claims, and assign weights to issues. These evaluations may become stated intentions, but enactment depends further on civic motivations, social norms, opportunity, and implementation. Each link is a separately measurable transition, and each can show substantial variation across people and elections Campbell et al. (1960) Van Bavel & Pereira (2018) Nickerson & Rogers (2010).
Several findings follow from this framework. Rapid visual judgments may predict electoral patterns without proving intrinsic political competence; partisan identity can structure evaluation while permitting updating from credible unfamiliar information; turnout interventions can influence participation without persuading voters to select another candidate; and high exposure to a platform does not guarantee that temporary deactivation changes candidate preference. The statements coexist because they concern different constructs and treatments. A unified model gains scientific strength by explaining these conditional relationships rather than resolving their differences through an oversized general category such as political influence Todorov et al. (2005) Huddy et al. (2015) Kalla & Broockman (2018) Allcott et al. (2024).
A formal program could represent individual state as prior evaluations and knowledge, incoming information as conditionally credible evidence, electoral preference as a probabilistic candidate comparison, and turnout as a thresholded implementation process under opportunity constraints. Such a model would need to distinguish empirical parameters from latent psychological constructs and identify what observations could refute its predictions. The present literature supports the explanatory relevance of those components, but no single reviewed model has been shown to estimate all transitions across elections, countries, and media environments. An integrated account is therefore a research synthesis specifying the components to measure and relations to test Riker & Ordeshook (1968) Broockman & Kalla (2023) Arceneaux & Nickerson (2009).
Open Research Questions and the Artificial Era
An immediate research priority is to connect experimentally observed candidate evaluations with later behavioral outcomes while respecting ballot secrecy. Researchers can study validated turnout, aggregate electoral patterns, and repeated private measures of preference without claiming access to individual marked ballots. Another priority is to identify the durability of persuasion under realistic counter-messaging and changing information. Longitudinal experiments with prespecified time horizons could clarify which immediate effects persist, which are absorbed into existing partisan commitments, and which result from temporary survey demand Broockman & Kalla (2023) Kalla & Broockman (2018).
Candidate appraisal also requires broader measurement across democratic institutions and social populations. The transportability of facial competence impressions, class and gender stereotypes, and party shortcuts remains conditional on candidate selection, public visibility, and national political context. Better cross-cultural designs would vary candidate attributes within comparable tasks while documenting distinct political meanings. Studies linking social identity to choice should measure whether belonging influences trust, information selection, policy perception, expressive participation, or all of these through different pathways Bauer (2017) Kim & Kweon (2024) Todorov et al. (2005).
Artificial Era research adds a growing class of adaptive interlocutors and information environments to the study of electoral judgment. It calls for longitudinal investigation of the reliability and provenance of information delivered by artificial systems, of how people decide when to delegate inquiry, and of whether conversational interaction changes factual beliefs, candidate confidence, or enacted political behavior. Randomized dialogue experiments provide initial evidence about expressed candidate preferences, while systematic measurement of long-run consequences, real-world exposure, and institutional variation remains open. The decisive methodological task is to test the human judgment and behavior associated with specified artificial interactions without assuming that observed preference shifts automatically translate into votes Lin et al. (2025) Allcott et al. (2024).
Conclusion
Electoral judgment is a system of psychological processes connecting political information, candidate appraisal, group identification, issue weighting, participation motivation, and institutional opportunity. Its modern study developed through survey-based electoral models, formal accounts of choice, experiments on social inference and partisan reasoning, and increasingly sophisticated field and digital interventions. The evidence supports neither a universal model of the perfectly informed voter nor a simple account of immutable partisan choice. It establishes that citizens use learned shortcuts, social identities, new information, retrospective assessments, and practical routines in combinations that depend on the election and the question being measured Campbell et al. (1960) Lau & Redlawsk (1997) Van Bavel & Pereira (2018).
The most important empirical boundary is between evaluations, intentions, participation, and election results. Candidate faces and profiles can shape measured impressions; some carefully controlled messages can alter expressed preferences; many general-election field contacts have small average persuasive effects; and targeted mobilization can change validated turnout through social accountability or implementation planning. These findings can be integrated when each study is interpreted by its design, target outcome, and the population to which its causal estimate applies. Methodological progress therefore depends on accurate construct measurement, randomized treatment where feasible, behavioral validation, open materials, independent replication, and explicitly conditional generalization Todorov et al. (2005) Broockman & Kalla (2023) Kalla & Broockman (2018) Gerber et al. (2008) Nickerson & Rogers (2010).
Political psychology enters the Artificial Era with new opportunities to observe and experimentally study adaptive information exchange. A scientific understanding of elections will increasingly require attention to how artificial communication systems affect what people learn, whom they trust, and what they ultimately do. The enduring object of this dossier is the psychological organization of electoral judgment and participation: the measurable sequence through which political alternatives become personally meaningful decisions within real institutions. That object remains open to more precise explanation through comparative studies, longitudinal designs, rigorous causal inference, and ethically governed research Lin et al. (2025) Allcott et al. (2024) Verba et al. (1995).
References
- Achen, C. H., & Bartels, L. M. (2017). Democracy for realists: Why elections do not produce responsive government (rev. ed.). Princeton University Press. https://doi.org/10.2307/j.ctvc7770q
- Allcott, H., Gentzkow, M., Mason, W., Wilkins, A., Barberá, P., Brown, T., Cisneros, J. C., Crespo-Tenorio, A., Dimmery, D., Freelon, D., González-Bailón, S., Guess, A. M., Kim, Y. M., Lazer, D., Malhotra, N., Moehler, D., Nair-Desai, S., El Barj, H. N., Nyhan, B., ... Tucker, J. A. (2024). The effects of Facebook and Instagram on the 2020 election: A deactivation experiment. Proceedings of the National Academy of Sciences, 121(21), e2321584121. https://doi.org/10.1073/pnas.2321584121
- Alvarez, R. M., & Morrier, J. (2025). Do polarized issues carry more weight in voter decision-making? Insights from the 2022 congressional midterm elections. American Politics Research, 53(5), 481–498. https://doi.org/10.1177/1532673X251341892
- Arceneaux, K., & Nickerson, D. W. (2009). Who is mobilized to vote? A re-analysis of 11 field experiments. American Journal of Political Science, 53(1), 1–16. https://doi.org/10.1111/j.1540-5907.2008.00354.x
- Bauer, N. M. (2017). The effects of counterstereotypic gender strategies on candidate evaluations. Political Psychology, 38(2), 279–295. https://doi.org/10.1111/pops.12351
- Blais, A. (2000). To vote or not to vote: The merits and limits of rational choice theory. University of Pittsburgh Press. https://doi.org/10.2307/j.ctt5hjrrf
- Broockman, D. E., & Kalla, J. L. (2023). When and why are campaigns’ persuasive effects small? Evidence from the 2020 U.S. presidential election. American Journal of Political Science, 67(4), 833–849. https://doi.org/10.1111/ajps.12724
- Campbell, A., Converse, P. E., Miller, W. E., & Stokes, D. E. (1960). The American voter. Wiley. https://press.uchicago.edu/ucp/books/book/chicago/A/bo24047989.html
- DellaVigna, S., & Kaplan, E. (2006). The Fox News effect: Media bias and voting (NBER Working Paper No. 12169). National Bureau of Economic Research. https://doi.org/10.3386/w12169
- Gerber, A. S., & Green, D. P. (1999). Does canvassing increase voter turnout? A field experiment. Proceedings of the National Academy of Sciences, 96(19), 10939–10942. https://doi.org/10.1073/pnas.96.19.10939
- Gerber, A. S., Green, D. P., & Larimer, C. W. (2008). Social pressure and voter turnout: Evidence from a large-scale field experiment. American Political Science Review, 102(1), 33–48. https://doi.org/10.1017/S000305540808009X
- Healy, A., & Malhotra, N. (2009). Myopic voters and natural disaster policy. American Political Science Review, 103(3), 387–406. https://doi.org/10.1017/S0003055409990104
- Huddy, L., Mason, L., & Aarøe, L. (2015). Expressive partisanship: Campaign involvement, political emotion, and partisan identity. American Political Science Review, 109(1), 1–17. https://doi.org/10.1017/S0003055414000604
- Kalla, J. L., & Broockman, D. E. (2018). The minimal persuasive effects of campaign contact in general elections: Evidence from 49 field experiments. American Political Science Review, 112(1), 148–166. https://doi.org/10.1017/S0003055417000363
- Kim, J. H., & Kweon, Y. (2024). Double penalty? How candidate class and gender influence voter evaluations. Research & Politics, 11(1). https://doi.org/10.1177/20531680241226511
- Lau, R. R., & Redlawsk, D. P. (1997). Voting correctly. American Political Science Review, 91(3), 585–598. https://doi.org/10.2307/2952076
- Lin, H., Czarnek, G., Lewis, B., White, J. P., Berinsky, A. J., Costello, T., Pennycook, G., & Rand, D. G. (2025). Persuading voters using human–artificial intelligence dialogues. Nature, 648, 394–401. https://doi.org/10.1038/s41586-025-09771-9
- Masch, L., Gassner, A., & Rosar, U. (2021). Can a beautiful smile win the vote? The role of candidates’ physical attractiveness and facial expressions in elections. Politics and the Life Sciences, 40(2), 213–223. https://doi.org/10.1017/pls.2021.17
- Nickerson, D. W., & Rogers, T. (2010). Do you have a voting plan? Implementation intentions, voter turnout, and organic plan making. Psychological Science, 21(2), 194–199. https://doi.org/10.1177/0956797609359326
- Riker, W. H., & Ordeshook, P. C. (1968). A theory of the calculus of voting. American Political Science Review, 62(1), 25–42. https://doi.org/10.2307/1953324
- 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
- Todorov, A., Mandisodza, A. N., Goren, A., & Hall, C. C. (2005). Inferences of competence from faces predict election outcomes. Science, 308(5728), 1623–1626. https://doi.org/10.1126/science.1110589
- Van Bavel, J. J., & Pereira, A. (2018). The partisan brain: An identity-based model of political belief. Trends in Cognitive Sciences, 22(3), 213–224. https://doi.org/10.1016/j.tics.2018.01.004
- Verba, S., Schlozman, K. L., & Brady, H. E. (1995). Voice and equality: Civic voluntarism in American politics. Harvard University Press. https://doi.org/10.2307/j.ctv1pnc1k7
Social Norms, Visibility, and Turnout
Voting is publicly understood in many settings as a socially valued civic act. Social norm accounts distinguish a personally endorsed obligation from an expectation of approval or disapproval by others. Gerber, Green, and Larimer tested this distinction by randomly varying messages delivered to registered voters in a large-scale field experiment. Treatments that emphasized the potential disclosure of turnout information to household members or neighbors increased validated turnout more than simpler civic-duty messages. The result identifies social accountability as a consequential mobilization mechanism under those conditions, although the presence of normative pressure creates ethical considerations for both research practice and public communication Gerber et al. (2008).
The distinction between public participation records and secret ballots is essential. Some jurisdictions record whether an individual voted, while a particular person's candidate choice remains protected by ballot secrecy. Experimental messages about participation visibility can alter expectations of reputational consequences, but researchers must represent precisely what record is public and avoid inducing false impressions about ballot disclosure. Ethical assessment considers informed consent where applicable, potential anxiety, coercive social dynamics, and the unequal effects of surveillance-like messages on vulnerable citizens. The scientific result is an explanation of one behavioral mechanism, not a normative endorsement of pressure-based election tactics Gerber et al. (2008).
Norm strength plausibly interacts with social identity and prior turnout habits. People embedded in politically active households may encounter reminders, encouragement, and reputational expectations, while citizens socially disconnected from those networks may receive fewer cues. A single mailing that makes turnout salient can thus combine several psychological pathways. Identifying mediation requires measurements of perceived obligation, anticipated visibility, and subsequent behavior under experimentally separable conditions. A general model should include the possibility of reactance or avoidance in settings where monitoring is experienced as intrusive, rather than assuming normative pressure has a constant positive effect across all populations Huddy et al. (2015) Gerber et al. (2008) Verba et al. (1995).