RESEARCH DOSSIER
Attitudes, Evaluation, and the Attitude–Behavior Relation
Abstract. An attitude is an evaluative orientation toward an object or action whose expression and behavioral consequences depend on representation, context, and opportunity. This fundamental research monograph examines how favorable and unfavorable evaluations form, become accessible, persist, and contribute to observed decisions. It reconstructs the historical attitude–action problem, differentiates explicit reports from indirect performance measures, and evaluates reasoned-action and planned-behavior theories alongside automatic activation, habit, implementation intentions, and structural constraints. The review critically examines evidence from original experiments and quantitative syntheses, including meta-analyses of prospective attitude–behavior correspondence, implicit and explicit measurement, intention interventions, and the modification of indirect measures. Its central synthesis identifies measurement correspondence, accessible and stable evaluations, actual behavioral opportunity, prior action, and social context as essential conditions for interpretable prediction. It distinguishes observational association, predictive validity, experimental manipulation, and demonstrated behavioral change; examines disputes over indirect task interpretation and criterion validity; and identifies longitudinal, cross-cultural, digital, and human–artificial research priorities. The method is a critical narrative synthesis of published literature, with no original human-participant data and no claimed new meta-analysis.
Keywords: attitudes; evaluation; attitude–behavior relation; attitude strength; attitude accessibility; attitude stability; implicit attitudes; explicit attitudes; Implicit Association Test; theory of planned behavior; behavioral intention; intention–behavior gap; habit; implementation intentions; attitude measurement; causal inference; social psychology; Artificial Era
Attitudes as Scientific Objects of Evaluation
An attitude is an evaluative orientation toward an object, person, social category, practice, proposition, or prospective action. Its defining feature is the organization of favorable or unfavorable responses, whether measured as a verbal judgment, a patterned response to a task, or a disposition expressed across encounters. The same object can be evaluated in several ways because appraisal depends on the object as represented, the relevant situation, and the judgment requested. In the reasoned-action tradition, an attitude toward performing a specified behavior is analytically distinct from a general evaluation of its object Fishbein & Ajzen (1975) Ajzen (1991). This distinction is the starting point for understanding predictive success and failure.
An evaluation may incorporate affective reactions, acquired beliefs, anticipated outcomes, remembered experiences, and relational commitments. These components can converge or diverge. Someone who values environmental protection may simultaneously consider a particular transportation option expensive, inconvenient, or inappropriate for a demanding schedule. Scientific explanation requires stating exactly what is evaluated, the time horizon, the target action, and which competing demands are present. An attitude score summarizes an evaluative response under measurement conditions; its causal role requires evidence about the processes that connect representation, opportunity, intention, and action Kraus (1995) Glasman & Albarracín (2006).
This monograph investigates the construction and representation of evaluation, individual and situational variation in its expression, the measurement of explicit and implicit responses, and the prediction and modification of behavioral outcomes. Its principal questions are how evaluative orientations develop and persist, when their activation guides action, why attitude and conduct sometimes diverge, how predictive claims can be tested, and which practical interventions produce observable behavioral consequences. The method is a critical narrative synthesis of primary studies, theoretical accounts, meta-analyses, and methodological criticism. It presents no newly collected participant data or original quantitative meta-analysis.
The Historical Attitude–Action Problem
Attitude research emerged as social psychologists sought to connect expressed evaluations with practical conduct. In 1934 Richard LaPiere described discrepancies between the treatment encountered by a traveling Chinese couple and subsequent written responses from establishments, making the attitude–action discrepancy a lasting methodological problem LaPiere (1934). Its design was historically influential but involved different respondents, different times, and distinct social encounters; the observations therefore establish a provocative discrepancy rather than a precise causal estimate of attitude effects.
Allan Wicker's 1969 review of verbal and overt responses challenged the assumption that global favorable or unfavorable declarations reliably correspond to specific actions Wicker (1969). The critique became productive because it forced researchers to examine what an attitude item and a behavior criterion actually measure. General ideological endorsement can be reliably expressed while any single action depends on opportunity, perceived costs, expectations of others, and the definition of the behavior. Aggregating conduct across occasions may estimate a different psychological quantity from predicting one event.
The subsequent research program replaced a search for a universal attitude–action coefficient with a set of conditional questions. Martin Fishbein and Icek Ajzen formalized the relationships among beliefs, evaluation, intention, and behavior Fishbein & Ajzen (1975). Ajzen's planned-behavior model expanded the account to subjective norms and perceived behavioral control Ajzen (1991). Parallel work on accessibility and automatically activated evaluation proposed that some judgments influence perception and responses before extensive deliberation Fazio et al. (1986). Later quantitative syntheses established that an average predictive association exists while revealing substantial heterogeneity across methods, attitude properties, and behavioral criteria Kraus (1995) Glasman & Albarracín (2006).
Conceptual Structure: Object, Valence, Strength, and Function
An attitude object can be a concrete product, an individual, a social policy, a moral practice, a social group, or the act of carrying out a precisely defined behavior. Valence concerns the favorable–unfavorable direction of evaluation; extremity concerns distance from neutrality; importance concerns significance to the person; certainty concerns subjective confidence; accessibility concerns how readily the evaluation is retrieved. These properties are distinguishable and need not covary. The attitude-strength literature treats durability, resistance, and influence on thought or action as consequences requiring measurement rather than as interchangeable synonyms for a strong rating Kraus (1995) Glasman & Albarracín (2006).
Ambivalence arises when competing positive and negative evaluations coexist or when emotional and belief-based reactions differ. It matters because a single bipolar scale can conceal opposed reasons for liking and disliking the same object. An unfavorable assessment of risk might coexist with attraction to the immediate reward of the action. Response competition, perceived social obligations, and the particular information made salient can therefore change the observed judgment without demonstrating wholesale replacement of the underlying knowledge. The distinction between associative activation and propositional validation is developed in competing accounts of explicit and implicit evaluation Gawronski & Bodenhausen (2006) Wilson et al. (2000).
The functions attributed to attitudes include simplifying evaluation, organizing information, marking social identity, orienting approach and avoidance, and supporting decisions. These functional labels become explanatory only when linked to observable mechanisms and testable predictions. For example, if an accessible evaluation facilitates rapid categorization, response times should change under controlled activation conditions; if importance supports durable conduct, prospective action should be measured across relevant opportunities. Different functions can operate simultaneously, which explains why the same attitude can be stable at the level of declared values yet variable in local expression.
Learning, Experience, and the Formation of Evaluation
Attitudes develop through encounters with objects and consequences, social communication, observation, memory, and repeated judgments. Direct experience may supply distinctive action-relevant information and make an evaluation easier to retrieve, while observation and testimony can create evaluations before any personal contact occurs. The longitudinal meta-analysis by Glasman and Albarracín found that attitudes formed through direct experience and repeatedly reported evaluations more strongly predicted later behavior, partly in association with accessibility and stability Glasman & Albarracín (2006). Its observational moderators establish conditions associated with prediction, not a universal causal effect of requiring repeated attitude reports.
Associative learning and inferential belief revision provide different candidate mechanisms. The evaluative consequence of pairing a person or product with pleasant material can be expressed in indirect tasks, while persuasive information may alter what outcomes a respondent considers plausible or desirable. Gawronski and Bodenhausen distinguish associative processes, which reflect activated relations, from propositional processes that evaluate whether a conclusion is valid Gawronski & Bodenhausen (2006). This distinction supplies experimental hypotheses about instruction, contradiction, context, and the durability of change, although particular tasks may combine several psychological operations.
Formation studies need to distinguish novel attitudes from the activation of prior evaluations. Baseline familiarity, differential exposure, experimenter expectations, demand characteristics, and selective attendance all matter. Random assignment to evaluative experience can support causal claims about the manipulation, while later observational follow-up is needed to establish persistence and behavioral consequences. A laboratory change in a rating is an appropriate outcome for questions about judgment; it becomes evidence about future behavior only when conduct is independently observed. The attitude–behavior research program is therefore fundamentally temporal and multi-measure.
Accessibility, Certainty, and Attitude Stability
Attitude accessibility refers to how efficiently an evaluation becomes available when its object is encountered. Fazio, Sanbonmatsu, Powell, and Kardes demonstrated that object presentations can automatically facilitate evaluatively congruent responses in a priming paradigm, supporting a role for object–evaluation associations in rapid judgment Fazio et al. (1986). Reaction-time effects depend on task conditions and are not evidence that a person is permanently controlled by an attitude. The theoretically relevant claim is narrower: readily activated evaluations can become inputs to subsequent interpretation and action, especially when deliberative correction is limited.
Certainty and temporal stability contribute to prediction through distinct pathways. A respondent may express a confident judgment that persists across weeks, whereas another may be consistently ambivalent or uncertain. Glasman and Albarracín's synthesis connected confident, accessible, and stable attitudes with stronger prospective behavioral correspondence Glasman & Albarracín (2006). The important inferential caution is that measurements of confidence, strength, and stability may share sources of variance; stronger prediction does not reveal whether certainty itself caused conduct or indexed well-established experience and more consistent circumstances.
Durability should be established through repeated observations separated by meaningful intervals and analyzed with measurement error in view. Apparent stability can reflect persistent measurement wording or a stable social environment, and apparent volatility can arise from changing contexts rather than genuine reversal of evaluation. Designs that sample persons across occasions, varying cues, and actual opportunities allow investigators to distinguish within-person fluctuations from between-person differences. The best predictive evidence asks whether accessibility adds information beyond contemporaneous intentions, prior behavior, and the conditions under which the decision will be enacted.
Explicit Attitude Measurement and the Principle of Correspondence
Explicit attitude measures ask respondents to evaluate an object or action directly, often using semantic-differential items, agreement scales, feeling ratings, or behavioral belief questions. Their methodological value derives from the accessibility of interpretable reports and the possibility of studying reasons, certainty, and change. Valid measurement requires a defined referent, response scale, recall period, and population. A rating of how favorable a person feels about exercise in general estimates something different from that person's evaluation of completing a specified exercise session next week. The reasoned-action program recommends close correspondence between attitude, action, target, context, and time Fishbein & Ajzen (1975) Ajzen (1991).
Scale scores require evidence about reliability, dimensionality, sensitivity, and interpretation. Agreement scales can exhibit acquiescence and extremity differences; reports on socially sensitive subjects may reflect audience management and perceived norms. Aggregating several appropriately targeted items can reduce random error, but aggregation cannot correct a mismatch between a global attitude object and the particular action being predicted. Wicker's historical critique and Kraus's subsequent meta-analysis make this a central point: reported correlations partly reflect whether criteria and predictors describe the same behavioral domain Wicker (1969) Kraus (1995).
Well-designed prospective studies measure the attitude before the action, register objective or independently documented behavior when feasible, and separate predictors from outcomes at collection. If the behavior is reported retrospectively in the same survey, shared response style and selective recollection may inflate correspondence. Researchers should specify whether the purpose is explaining population averages, ranking individuals, forecasting event-level conduct, or estimating causal effects of interventions. Each purpose requires different reliability analyses, temporal tests, and standards of predictive success.
Indirect Measures and Evaluative Response Latency
Indirect measures infer evaluative processing from performance rather than asking directly for a favorable–unfavorable judgment. Affective-priming tasks examine the effect of briefly presented objects on subsequent evaluative responses. The Implicit Association Test, introduced by Greenwald, McGhee, and Schwartz, compares performance when target categories and evaluative attributes share response mappings Greenwald et al. (1998). These tasks estimate relative associations under structured task demands. Their observable outputs are latency patterns and accuracy differences, not direct recordings of a unitary unconscious attitude.
Fazio and Olson's methodological review emphasized that indirect tasks vary in the mechanisms they elicit and the meaning of their scores Fazio & Olson (2003). Some measures combine category familiarity, task switching, stimulus salience, response control, and learned associations. Ranganath, Smith, and Nosek further distinguished automatic and controlled components of psychological processes from the surface distinction between direct and indirect measurement methods Ranganath et al. (2008). It follows that a direct report can involve rapid or habitual responding and an indirect task can involve conscious strategy.
Interpretation requires careful target construction, counterbalancing, adequate trials, transparent exclusion rules, and reliability assessment. Differential response time across category pairings can be a useful research measure even when an individual's score is imprecise. Strong conclusions about a person's stable preference or future discriminatory behavior demand independent validity evidence that a test score alone does not provide. Data from indirect tasks are most informative when combined with explicit reports, contextual manipulations, longitudinal measures, and directly observed relevant outcomes.
Implicit and Explicit Evaluations: Theory and Measurement
The distinction between implicit and explicit attitudes refers to a theoretical concern with evaluation under different conditions of activation, awareness, reflection, and response control. Wilson, Lindsey, and Schooler's dual-attitude account proposed that older and more recently endorsed evaluations can coexist, explaining situational differences between spontaneous and deliberative responses Wilson et al. (2000). Gawronski and Bodenhausen's associative–propositional account instead differentiates activation of associative structures from the validation of propositions about the world Gawronski & Bodenhausen (2006). Both frameworks aim to explain dissociation without assuming that each task directly detects a separate psychological entity.
A meta-analysis by Hofmann and colleagues examined correlations between IAT scores and explicit self-reports, identifying potential sources of disagreement including motivated reporting, access to representations, retrieval conditions, and method variance Hofmann et al. (2005). This means that a low correlation between measures cannot alone establish two independent attitudes, while a high correlation cannot prove that both measures capture the same mechanism. Study designs should manipulate hypothesized processes and evaluate patterns across multiple tasks rather than classify a person by one contrast score.
The contemporary synthesis by Gawronski and Corneille examines evidence about awareness of an evaluation, awareness of its environmental causes, and awareness of its behavioral effects Gawronski & Corneille (2025). Their review argues that strong evidence for sweeping unawareness claims is scarce. Accordingly, scientific descriptions should distinguish a task labeled implicit, an automatically activated response, a participant's reportable knowledge, and experimentally demonstrated absence of awareness. These propositions require different operational criteria.
Measurement Invariance, Response Bias, and Social Meaning
Attitude instruments depend on language, reference groups, familiarity, and social meaning. A word used to describe approval or obligation may carry distinct normative expectations across settings; a response option can indicate certainty in one population and willingness to make a public declaration in another. Cross-cultural comparison therefore requires tests of conceptual equivalence, translation quality, and measurement invariance rather than assuming that identical questionnaire wording yields identical psychological meaning. A difference in scale means is interpretable as a difference in the intended construct only after plausible competing measurement explanations have been examined.
Social desirability is often described as distortion, but an answer provided under an anticipated audience can itself be an informative social action. The proper methodological distinction is between the private evaluation being studied, the report a respondent chooses to make, and the social consequences of the report. Dovidio and colleagues showed that different forms of measurement could relate to different race-related responses in their experimental research Dovidio et al. (1997). These findings motivate direct measurement of contextual moderators and observed outcomes rather than the assumption that one scale reports a universally privileged psychological truth.
Where attitudes concern identity, power, discrimination, or institutional trust, test construction should include multiple populations and examine item functioning. Individual-level diagnostic interpretation needs stronger evidence than group-level research. Differential prediction, calibration by subgroup, and measurement error should be evaluated separately. When an instrument is applied to digital traces or artificial systems, the inference also changes: a behavioral pattern recorded by software is evidence of that observable pattern, while claims about internal evaluation require an explicit model and independent validation.
Statistical Models and the Limits of Prediction
Predictive studies commonly report correlations, regression coefficients, explained variance, classification performance, or probabilistic forecasts. Their interpretation depends on the prevalence of the outcome, measurement reliability, sampling design, and whether predictor and criterion were obtained independently. Kraus's meta-analysis of 88 studies found an average attitude–behavior correlation of approximately .38 while emphasizing stronger correspondence under favorable measurement and attitude-strength conditions Kraus (1995). The coefficient is a pooled result across research settings, not a universal constant to be inserted into a forecast for every domain.
A coefficient can be statistically reliable while yielding modest individual-level predictive precision. If a behavior is rare, even a well-calibrated risk ranking can produce many false-positive forecasts at a chosen threshold. Multivariable models should compare the incremental information provided by attitudes against prior behavior, intentions, resources, and contextual constraints. For temporally repeated outcomes, observations within a person are dependent and require appropriate multilevel or longitudinal treatment. Random splits at the observation level may leak the same participant's stable patterns into training and testing datasets.
Prediction also needs to be distinguished from explanation. An attitude score can forecast future action because it is a marker of past experience, social environment, habits, or stable opportunities. Such predictive utility does not establish that changing the score will change conduct. External validation across periods, regions, and populations tests transportability; randomized intervention evidence tests whether changing a hypothesized determinant alters an outcome. The gap between these two questions structures the rest of the monograph Webb & Sheeran (2006) Albarracín et al. (2024).
Reasoned Action and the Theory of Planned Behavior
The reasoned-action approach makes a particular behavioral intention a proximal determinant of action. Intentions are shaped by evaluation of performing the behavior and by perceived social expectations. Ajzen's theory of planned behavior adds perceived behavioral control, recognizing that people may favor an action while believing that performing it will be difficult or impossible Fishbein & Ajzen (1975) Ajzen (1991). In this framework, the attitude toward performing a defined behavior is calculated from salient consequences and their evaluations, while normative expectations and control beliefs contribute separately. A broad liking for the domain need not supply the appropriate behavioral predictor.
The theory has generated a large cumulative literature. Armitage and Conner's meta-analysis of 185 independent studies found that its predictors explained about 39 percent of variance in intention and about 27 percent of variance in behavior within the reviewed evidence Armitage & Conner (2001). These figures are pooled estimates for the studies and measures included, not fixed parameters. Self-reported outcomes and cross-sectional intention estimates can exaggerate practical predictability, while the control construct may combine perceived feasibility with actual constraints imperfectly.
An applied design begins by specifying an action with a target, setting, and period, measuring its antecedent beliefs, and observing subsequent performance independently. Subjective norms should be separated from the researcher's own normative preferences. Perceived control must be compared with objective opportunity and resources wherever possible. Competing predictions include direct effects of habitual repetition, changes in behavioral opportunities, and reciprocal effects of action on evaluation. Successful prediction from intentions alone is insufficient to conclude that the whole causal sequence has been identified.
Motivation, Opportunity, and the MODE Framework
Attitude-guided conduct can reflect different degrees of deliberation. In deliberate episodes people compare costs, consequences, norms, and anticipated control. In rapid or familiar episodes an activated evaluation may color perception and prompt a practiced response before extended reflection. Fazio's research on automatic attitude activation established experimentally that objects can trigger evaluative information under controlled task conditions Fazio et al. (1986). The motivation-and-opportunity perspective associated with this research predicts that deliberate correction depends on resources and motivation available at the moment of response.
These processes are complementary possibilities in a specific decision, not exclusive labels for people. Even a highly accessible evaluation may fail to produce an action when an external rule, social penalty, or physical barrier prevents performance. Conversely, a behavior may occur with minimal contemporaneous evaluation because a situational cue evokes a rehearsed routine. Studies testing this account need to manipulate distraction, response opportunity, task demands, or motivational relevance and then observe behavior, distinguishing effects on intermediate evaluation from effects on the actual decision.
Research should resist treating automatic activation as a complete description of agency. An evaluation can shape which cues receive attention without determining the final action. An observer may respond favorably toward a proposed activity yet defer it because another obligation takes priority. Such cases are informative observations of an organized choice environment rather than failures of the attitude construct. The theoretical task is to identify the stage of processing at which measured evaluation contributes incremental explanatory information.
Intention Formation and the Intention–Behavior Gap
Intention is an expressed commitment to undertake a future action. It can be a strong predictor while remaining an imperfect proxy for the eventual outcome, because the interval between reporting and performance contains changing opportunities, competing goals, and lapses of memory. Sheeran and Webb synthesize this intention–behavior gap as a practical and theoretical question about translating a chosen goal into an enacted sequence Sheeran & Webb (2016). Researchers must distinguish people with weak intention, people with strong intention and insufficient opportunity, and those whose plans are displaced by later events.
The causal significance of intentions cannot be inferred solely from a correlation measured at one time. Webb and Sheeran meta-analyzed randomized tests in which interventions changed intentions and behavior, reporting that a larger average change in intentions corresponded to a smaller change in subsequent behavior Webb & Sheeran (2006). The experimental synthesis showed intention change can affect conduct, but also demonstrated that attitude- or intention-focused intervention may be insufficient for large behavioral effects. The size of a manipulated change and the difficulty of the target behavior both matter.
Longitudinal studies can measure whether intention stability, planning, environmental access, and competing demands jointly predict enactment. A report of behavior made after the respondent learns the expected answer may confound outcome with recall or reputation management. Where feasible, time-stamped records, independent observation, and repeated opportunity sampling strengthen inference. Interventions should be evaluated against an appropriate control and analyzed for both proximal changes in intention and distal effects on behavior.
Habits, Environmental Cues, and Behavioral Persistence
Habitual actions are shaped by repeated performance in recurring contexts. Stable cues can sustain a behavior even when current evaluation or intention has changed. Wood, Tam, and Guerrero Witt examined how changing circumstances can disrupt habits, illustrating why behavioral regularities often depend on the preservation of situational cues Wood et al. (2005). Habit strength, contextual continuity, and anticipated rewards can alter the relationship between a stated preference and a subsequent action. A theory that considers attitudes alone may consequently underpredict persistence and overpredict immediate change.
Behavioral interventions differ in whether they target evaluation, planning, access, or the repeating cue–action relationship. If the barrier is unavailable transportation, improved approval of a health service may have little impact; if the barrier is forgetting, a timely cue can matter more than another message about benefits. Albarracín, Fayaz-Farkhad, and Granados Samayoa's cross-domain synthesis compares interventions targeting distinct determinants and shows why changing general attitudes is not interchangeable with modifying behavioral skills and habits Albarracín et al. (2024).
Temporal data help distinguish the influence of prior behavior from that of current attitude. Repeated action can strengthen confidence and perceived identity, so an observed association may reflect reciprocal causality. Natural changes in environment and randomized cue manipulations can test whether stable context is responsible for conduct. These designs also clarify when apparent attitude–behavior inconsistency is predictable from a history of repeated action and opportunity rather than from unreliability of the evaluation itself.
Implementation Intentions and Action Planning
Implementation intentions connect a chosen goal to a specific anticipated situation and a concrete response. The common formulation specifies that, if a defined occasion occurs, the person will initiate a selected action. Gollwitzer and Sheeran synthesized 94 independent tests and reported a medium-to-large average improvement in goal attainment relative to controls Gollwitzer & Sheeran (2006). The findings establish that translation from intention into action can be altered by procedures that specify how action will be initiated, while differences in designs and outcomes still constrain generalization.
Planning is theoretically useful because it identifies a mechanism between evaluation and behavior. A person can value an action yet fail to recognize an opportunity; linking the relevant cue to a response can reduce this problem. However, planning may work poorly when the cue is rare, the action exceeds the person's control, the plan conflicts with stronger obligations, or the initial goal is unstable. Testing these qualifications requires comparing clear plans with similarly intensive control conditions and measuring actual completion, rather than relying only on expressed readiness.
Implementation intentions should be studied alongside structural feasibility. A plan to access a service has limited significance if eligibility, cost, availability, or mobility prevents use. Practical research should report who had an opportunity to act and whether an intervention changed that opportunity. Distinguishing motivational acceptance, behavioral planning, and objective resources prevents a psychological measure from absorbing explanations that properly belong to the wider social and institutional environment Ajzen (1991) Albarracín et al. (2024).
Quantitative Evidence for Attitude–Behavior Correspondence
Meta-analysis helps distinguish a systematic association from results that happen to arise in a handful of influential laboratories. Kraus's 1995 review synthesized 88 studies and reported a mean attitude–behavior correlation of approximately .38, together with substantial moderator effects of certainty, stability, accessibility, direct experience, and measurement correspondence Kraus (1995). The finding warrants the conclusion that attitudes can forecast action across studied settings. It does not establish a universal causal parameter, because designs vary in temporal order, sampling, outcome definition, and control for background predictors.
Glasman and Albarracín's 2006 synthesis examined 128 conditions including 4,598 participants, focusing specifically on factors present during attitude formation that predict subsequent correspondence Glasman & Albarracín (2006). Higher accessibility and temporal stability were associated with stronger links, and direct experience, frequent expression, confidence, and behavior-relevant information were important moderators. The formation-stage emphasis makes this review especially informative for mechanisms of future influence, but the meta-analytic associations should not be confused with interventions that randomized all these characteristics independently.
These syntheses also illustrate the difference between averaging a literature and characterizing any one decision. Publication practices, heterogeneity in settings, range restriction, selective availability of objective outcomes, and overlap among measurements may affect pooled estimates. The theoretically strongest inference combines quantitative averages with the pattern of moderators, independent temporal validation, and sensitivity to how outcomes were obtained. A research synthesis should always identify its estimand: an individual-level correlation, a causal effect of a manipulated attitude, or the improvement in forecasting beyond simpler behavioral predictors.
Intergroup Evaluation, Prejudice, and Observed Conduct
Intergroup research examines whether evaluations of social categories predict interpersonal responses, judgments, and discriminatory actions. Dovidio and colleagues' experiments contrasted measures of relatively spontaneous and more controlled racial responses, demonstrating that indirect and self-reported assessments could correspond differently to different behavioral outcomes Dovidio et al. (1997). The studies are historically significant, but the interpretation of any individual indirect score must remain limited by its reliability and predictive validation in an appropriate context.
Kurdi and colleagues synthesized 217 reports and 36,071 participants, finding nonzero associations between IAT measures and intergroup behavioral criteria and unique contributions from implicit and explicit measures in their modeling Kurdi et al. (2019). The same paper emphasized substantial heterogeneity. Oswald and colleagues' earlier meta-analysis evaluated related criteria and reached more skeptical conclusions about the practical prediction of ethnic and racial discrimination Oswald et al. (2013). Differences in eligible outcomes, how criteria were categorized, analytical assumptions, and interpretations of practical value contribute to this continuing debate.
An attitude-related measure can be informative at the group level while being insufficiently precise for screening or judging individuals. Behavioral criteria range from reaction time and ratings in laboratories to consequential decisions or repeated field conduct, and they cannot be treated as equivalent measures of discrimination. Stronger designs combine direct observation, objective criteria, prospective periods, and subgroup calibration while documenting institutional conditions. Researchers also need to ask whether social settings amplify, suppress, or redirect the expression of an evaluative response.
Changing Indirect Measures and Changing Behavior
The ability to change a test score is a separate scientific question from the ability to change harmful conduct. Forscher and colleagues used network meta-analysis to synthesize 492 studies with 87,418 participants involving procedures intended to shift indirect or implicit measures Forscher et al. (2019). The evidence supports the claim that such measures can change under experimental procedures. It offers more limited support for the proposition that the changes reliably produce corresponding, sustained changes in real-world behavior.
An intervention may change task familiarity, salience, instructions, or a particular associative response without changing durable evaluations. It may also alter explicit beliefs, which can mediate other forms of conduct. For causal analysis, the important design includes manipulation checks, independent behavioral outcomes, adequate follow-up, controls for demand effects, and analysis of intervention specificity. An effect on the same task used to train participants may represent task learning rather than a generalized psychological change.
The ethical and applied stakes are especially high when indirect scores are used for institutional programs. If a training procedure produces a reliably smaller score difference but no verified improvement in decisions affecting others, claims of demonstrated behavioral benefit overstate the available evidence. Conversely, small effects on an individual task need not imply absence of wider structural or cumulative consequences. Appropriate conclusions are indexed to studied outcomes and populations, and future trials should preregister meaningful behavioral endpoints Oswald et al. (2013) Kurdi et al. (2019) Forscher et al. (2019).
The 2025 Meta-Analytic Reassessment of Implicit and Explicit Attitudes
The synthesis by Phipps, Hagger, and Hamilton, published in Psychological Bulletin in 2025, used multilevel meta-analytic structural equation modeling across 1,108 studies to estimate the separate associations of implicit and explicit attitude measures with behavior Phipps et al. (2025). It found nonzero, small unique average effects for both, with a larger average effect associated with explicit attitudes. The analysis sought moderation by measurement, target behavior, sample characteristics, and features of behavioral control; the authors reported relatively few systematic moderator differences and substantial residual heterogeneity.
This evidence revises simplistic expectations that indirect measures should uniquely dominate prediction of spontaneous action or that explicit measures alone exhaust evaluative influences. The central scientific conclusion is conditional: both measurement families can contribute information, but the incremental magnitude is generally small and varies across contexts. A pooled structural model clarifies associations across study populations; it cannot independently resolve all causal mechanisms underlying a particular response or establish that differences in scores identify two fully separate internal attitudes.
The contrast with earlier intergroup meta-analyses is informative. The 2025 review included a much broader behavioral and methodological literature than examinations limited to prejudice criteria, while the older debates focused on the practical value of specific tests for consequential outcomes Kurdi et al. (2019) Oswald et al. (2013). A coherent synthesis therefore distinguishes universal statements about prediction from domain-specific validity, and uses new meta-analysis to refine rather than erase earlier controversies. Independent prospective tests of real behavior remain the decisive next step.
Normative Expectations, Institutions, and Social Opportunity
Attitudes are formed and enacted within interpersonal and institutional systems. A student may value speaking openly while anticipating sanctions for contradicting a teacher; an employee may support a workplace improvement while lacking authority to implement it. The planned-behavior framework recognizes normative expectations and perceived control as determinants of intention, and contemporary behavioral reviews emphasize the importance of social-structural determinants Ajzen (1991) Albarracín et al. (2024). These variables should be observed rather than introduced merely as explanations after an unsuccessful forecast.
Norms differ in what a person believes most others do, what others approve, and what sanctions might follow deviation. They can promote behavior despite weak personal endorsement or suppress conduct despite strong favorable evaluation. An institutional rule can also produce an action directly by restricting the available alternatives. Disentangling those pathways requires measuring perceived expectations and enforcement separately from personal attitude, and identifying whether the person had genuine alternatives in the observed situation.
A study that samples only those who already have access to an option may misstate the relation in the broader population. Opportunity is often patterned by financial resources, geography, organizational hierarchy, caregiving responsibilities, and social ties. Valid external inference therefore describes the eligible population and the distribution of opportunities. A rigorous attitude–behavior account is a multilevel analysis of evaluative disposition, proximate situation, and institutional conditions.
Temporal Dynamics and Reciprocal Attitude–Behavior Relations
Attitudes measured today may predict behavior tomorrow, while today's behavior may shape tomorrow's evaluation. Repeated action can produce learning about costs, competence, social reactions, or identity, thereby changing the beliefs that inform subsequent judgment. A single cross-sectional correlation cannot identify the direction of this feedback. Prospective studies should measure evaluations at several points, register actions in between, and consider how prior behavior structures both subsequent opportunity and evaluation Wood et al. (2005) Glasman & Albarracín (2006).
Within-person processes can differ markedly from between-person associations. Individuals who more consistently favor an action may perform it more often than other individuals, yet momentary fluctuations in a given person's evaluation may add little to prediction beyond schedule and opportunity. Conversely, a temporary evaluative shift can matter strongly for a particular choice even when stable between-person rank ordering is weak. Separating these effects requires repeated measures, appropriate centering, and models that recognize serial dependence.
Durable prediction must also accommodate changes in goals and environments. Students, workers, families, and communities pass through transitions that alter which behaviors are possible and socially expected. A fixed regression coefficient trained before a transition may become miscalibrated afterward. Monitoring drift, using multiple prospective waves, and reporting context-sensitive performance make it possible to distinguish psychological development from changes in measurement and opportunity.
Development, Culture, and the Generalizability of Evaluations
Across the life course, evaluation is intertwined with changing knowledge, autonomy, social roles, and opportunities. Children often acquire preferences through observation and immediate reinforcement, while later action increasingly involves explicit plans, long-term commitments, and institutional constraints. Yet the relative contribution of these processes depends on the domain. Measurement that assumes adult levels of linguistic comprehension or prospective planning cannot automatically be applied to younger respondents without validation.
Culture shapes the meanings of objects, actions, norms, and legitimate behavioral alternatives. A stated evaluation may express relational obligation in one context and individual preference in another, even when the survey wording seems equivalent. Comparisons require careful attention to translation, sampling, reporting norms, and the conditions under which a person can act on a preference. Broad claims about national groups are particularly vulnerable to confounding because institutional access, local norms, and economic resources vary within countries.
The most transportable findings are often conditional principles rather than fixed effect sizes: correspondence improves when measures refer to the same behavior and time, stability and accessibility are assessed, and opportunities are observed Kraus (1995) Glasman & Albarracín (2006). Whether these moderators hold with similar magnitudes across age groups and cultures remains an empirical question. Multi-country preregistered studies, representative sampling, and shared outcome definitions would allow stronger tests of such regularities.
Digital Choice Environments and Algorithmic Mediation
Online platforms make evaluation highly observable through ratings, likes, search behavior, subscriptions, and selections. These traces are tempting proxies for attitude, but they are actions performed under interfaces that structure what is visible, easy, socially rewarded, or required. A click can express curiosity, obligation, exploration, or accidental activation rather than approval. The scientifically relevant unit is the relation among displayed option, participant interpretation, platform affordance, and ensuing behavior, not the trace alone.
Ranking systems and recommendation algorithms alter exposure, repetition, and perceived consensus. They can change which evaluations become accessible and which intentions can be enacted. Distinguishing preference from exposure requires information about the choice set and the selection mechanism. Randomized interface changes, delayed follow-up, and analyses of behavior outside the platform are stronger tests of whether induced evaluation leads to durable action. The general behavioral-science synthesis by Albarracín and colleagues underscores why psychological targets and social-structural conditions should be analyzed jointly Albarracín et al. (2024).
In the Artificial Era, evaluations directed toward artificial interlocutors, decision systems, and collaborative agents become scientifically important examples of situated attitudes. Trust expressed in a questionnaire can predict willingness to consult a system, yet actual delegation depends on task stakes, transparency, organizational rules, availability, and past performance. Evaluation of an artificial system is observable in human responses; claims that the system itself possesses a comparable psychological attitude require separate operational definitions and evidence. The relevant interdisciplinary program concerns calibrated human–artificial relations and how technological environments reorganize human action.
Ecological Measurement and Field-Based Attitude Research
Laboratory tasks isolate evaluative processes, while field studies examine behavior under ordinary social and material constraints. Each setting has inferential advantages and costs. In a controlled laboratory, researchers can randomize an informational cue and precisely time responses, but the tested action may have low personal significance. In a field setting, the outcome may be consequential, while self-selection, context changes, and missing data complicate inference. Good scientific synthesis treats both sources as complementary rather than assigning either automatic superiority.
Ecological momentary assessment can repeatedly sample attitudes, intentions, affect, and contexts near decision occasions. Passive logs or administrative measures may independently register subsequent behavior when consent, legal authorization, and data quality allow. Such designs permit assessment of cue exposure, timing, and within-person associations. Yet the act of repeatedly asking for an attitude may itself increase its accessibility, which Glasman and Albarracín identified as relevant to predictive correspondence Glasman & Albarracín (2006). Measurement reactivity therefore requires explicit consideration.
A high-quality field protocol predefines the target action, denominators of actual opportunities, attrition handling, and plausible confounding. It should separate intention measured immediately after a persuasive event from action observed after a meaningful interval, document nonresponse, and evaluate whether participants who remain in a panel differ from those who leave. These details allow estimates to inform consequential policy or intervention rather than merely reproducing the persuasive appearance of a survey correlation.
Causal Identification and the Architecture of Evidence
The central empirical difficulty in attitude research is that a favorable evaluation, the intention to act, prior conduct, and social opportunity are commonly entangled. A correlation between favorable evaluations and later behavior may arise because evaluation helped motivate action, because prior actions shaped the evaluation, or because some prior experience caused both. Research should articulate the causal estimand before choosing a design. A randomized manipulation of evaluative information can test its local consequences, provided that assignment is preserved, compliance and attrition are documented, and the manipulation changes the proposed determinant. Longitudinal measures can clarify temporal order without automatically removing confounding Webb & Sheeran (2006) Glasman & Albarracín (2006).
A prospective cohort can improve on cross-sectional design by measuring the same attitude before a well-defined behavioral opportunity and adjusting for prior behavior and plausible background factors. It may still be vulnerable to time-varying confounding, selection into opportunities, and unmeasured common causes. Where interventions manipulate motivation or intention, the randomized comparison estimates the effect of the intervention package; attributing it exclusively to an attitude mechanism requires additional evidence about mediators. The meta-analytic experiments on intention change by Webb and Sheeran illustrate the distinction between inducing a psychological shift and establishing its effect on an action Webb & Sheeran (2006).
Rigorous interpretation also depends on outcome quality. Decisions recorded by an independent observer or operational system can reduce common-method bias, although logs may omit motives and the conditions under which a choice was made. In a setting where actions require partners or institutions, individuals cannot independently enact all favorable intentions. Sampling actual opportunities, registering refusals and obstacles, and specifying denominators may therefore be as important as refining the attitude measure. The inferential target should remain the behavior for which evidence was collected.
Replication, Heterogeneity, and Meta-Analytic Interpretation
Replication tests whether an attitude effect and its hypothesized moderators survive changes in investigator, sample, place, time, and operational definition. A close replication investigates robustness of a specified operational contrast; a conceptual replication examines whether a proposed process generalizes to a different arrangement of observations. The broad quantitative record already reveals heterogeneity: associations in the Kraus synthesis, the future-behavior studies examined by Glasman and Albarracín, and comparisons of implicit and explicit predictors by Phipps and colleagues vary with the setting and outcome Kraus (1995) Glasman & Albarracín (2006) Phipps et al. (2025).
Pooling studies can obscure that different studies measure different actions. Self-reported intentions, observed choices, social judgments, task latencies, and long-term habits have distinct error structures and practical meaning. An average meta-analytic correlation may be helpful for estimating a general research pattern, but prediction intervals and subgroup analysis are better suited to asking what future studies could reasonably observe. Small-study effects, selective outcome reporting, missing null results, and correlated effects within reports are additional reasons to favor transparent synthesis and independent registered research.
Scientific reconciliation demands explicit attention to disagreements. Oswald and colleagues' skeptical evaluation of the predictive value of the IAT for discrimination and Kurdi and colleagues' broader account of intergroup-behavior relations differ in analytic emphasis and inferential conclusions Oswald et al. (2013) Kurdi et al. (2019). Forscher and colleagues further show that shifting an indirect attitude measure need not generate durable behavioral improvement Forscher et al. (2019). These findings coexist with evidence that evaluative constructs can have predictive value. The task is to specify which outcomes are supported by which designs and whether an apparently conflicting result reflects construct, criterion, intervention, or population differences.
Research Ethics and the Interpretation of Attitude Scores
Attitude measures can concern intensely personal topics including social identity, health decisions, prejudice, and anticipated relations with institutions. Scientific use requires informed participation where applicable, data minimization, meaningful privacy safeguards, and clarity about what a score can and cannot establish. Particularly with indirect measures, a person-specific result may have uncertainty too large for consequential diagnosis or institutional sanction. An association across a sample does not license a deterministic inference about an individual's motives, values, or conduct Fazio & Olson (2003) Oswald et al. (2013).
Measurement instruments also have social consequences. Publicly requested declarations may create incentives to conceal unpopular attitudes; similarly, labeling a task score as an individual's hidden preference can induce stigma when construct validity is contested. Responsible reporting specifies the method, target category, measurement reliability, applicable population, criterion validity, and confidence or uncertainty of inference. When models are used to predict misconduct or discrimination, observed fairness and real-world calibration need to be evaluated instead of assuming the measurement method itself carries moral authority.
Experimental research that changes social evaluations should consider foreseeable harm, consent, debriefing, and the possibility of reinforcing stereotypes through repeated stimuli. Online environments create additional concerns about covert behavioral manipulation and large-scale inference from users' traces. The distinction between observing an evaluation and intervening on a person matters scientifically and ethically. Researchers should align the intervention, the informed-consent framework, and the reported causal conclusion with the actual scope of data collection and experimental control.
Evidence Quality and a Critical Assessment of Leading Models
Reasoned-action models have strong empirical utility in making attitude and behavioral intention specific and measurable. Their limits arise when investigators retrospectively explain nonperformance by attaching a poorly operationalized control variable or when self-reported intentions and self-reported behavior inflate correspondence. The meta-analytic literature supports the usefulness of the model's constructs, but variation in design means predictive variance estimates should be treated as properties of samples and outcomes rather than as universal constants Ajzen (1991) Armitage & Conner (2001).
Accessibility approaches identify conditions under which a previously acquired evaluation can enter fast judgments. Priming experiments are well matched to questions about response facilitation and retrieval, whereas claims about long-term conduct require further outcomes. Associative–propositional models clarify why indirect and direct measures can diverge but require specific causal discriminations among retrieval, validation, and response control. The evidence thus supports a layered research strategy involving proximal process tests and distal behavior, rather than the inference that a particular task directly maps an entire psychological mechanism Fazio et al. (1986) Gawronski & Bodenhausen (2006) Ranganath et al. (2008).
Intervention research offers a more exacting standard. If changing intentions produces a smaller behavioral improvement, and changing indirect scores does not consistently improve observable outcomes, then changes in an evaluated construct should be reported as such Webb & Sheeran (2006) Forscher et al. (2019). The practical implication is to pair motivational components with opportunity, action planning, skills, environmental conditions, and verified outcomes where appropriate. The cross-domain synthesis of behavioral determinants strengthens the case for evaluating such components separately and in combination Albarracín et al. (2024).
Open Methodological Problems in Attitude Science
Several unresolved questions concern measurement. It remains difficult to determine when a rapidly recorded response reflects stable evaluation, transient salience, response compatibility, or a strategy adopted for a specific task. Across cultures and social positions, apparently identical self-report items can refer to different social obligations. Researchers need richer multi-method models that test cross-time and cross-group invariance and compare direct reports with independent behavioral records. These objectives require precisely registered constructs and transparent treatment of uncertainty rather than a proliferation of vaguely differentiated attitude labels.
Causal questions concern how changing evaluations changes action after accounting for existing habits and institutional opportunity. Randomized messages, evaluative learning, planning procedures, and environmental redesign may produce different effects on the same behavior. Factorial and sequential randomized experiments can test interaction among determinants, while longitudinal observational studies reveal conditions under which these effects persist. A theoretically meaningful model must explain why an intervention changes behavior in one opportunity structure and fails in another, including possible reciprocal feedback from conduct to evaluation Webb & Sheeran (2006) Wood et al. (2005) Albarracín et al. (2024).
A further frontier is ecological and technological validity. Recommendation algorithms, conversational systems, and personalized interfaces continually transform information exposure and choice architecture, creating feedback between observed clicks, inferred preferences, and subsequent options. The field needs ethically governed data and designs that separate learned preference from platform-induced opportunity. The same principles apply across human and artificial interlocutor contexts: identify the observable unit, specify the psychological construct being inferred, and test whether a model predicts meaningful conduct beyond the measurement conditions that produced it.
An Integrated Interpretation of Evaluative and Behavioral Processes
Across experimental and longitudinal traditions, attitudes emerge as organized evaluative responses that can contribute to perception, decision making, and future action. The best-supported interpretation links specific measurements to specific psychological operations. Accessible evaluations can be activated under controlled conditions, explicitly endorsed action evaluations can predict intentions, and intentions can forecast some later conduct. Meta-analyses show that the degree of correspondence varies systematically with stability, measurement specificity, direct experience, and behavioral context Fazio et al. (1986) Ajzen (1991) Kraus (1995) Glasman & Albarracín (2006). These observations supply connected elements of a broader account of action without implying one fixed pathway for all choices.
The empirical literature supports multiple mechanisms operating at different timescales. Immediate associations can influence responses to familiar cues; reflective appraisal can integrate consequences and social expectations; stable routines can maintain conduct after evaluation has changed; opportunity determines which actions are physically and institutionally available. Comparative evidence from interventions supports distinguishing general attitudes from behavior-specific attitudes, intentions, plans, skills, habits, and structural conditions Webb & Sheeran (2006) Gollwitzer & Sheeran (2006) Albarracín et al. (2024). Scientific synthesis thus requires a model of the decision situation as well as a measure of evaluation.
An integrated research strategy uses psychometric validation to establish what a measure means, prospective studies to evaluate prediction, experiments to identify local mechanisms, and independent behavioral follow-up to assess consequences. Each stage contributes a different inference, and claims should remain attached to the design that supports them. In particular, an indirect attitude measure is a valuable record of task performance whose behavioral interpretation must be demonstrated. The large-scale 2025 meta-analysis indicates that both implicit and explicit measures can uniquely contribute to predicting behavior, while generally modest average increments limit individual prediction Phipps et al. (2025).
Practical Implications for Research and Behavioral Intervention
An intervention should begin with a precise behavioral endpoint and a diagnosis of the principal barriers to that endpoint. If people lack accurate information, educational material may change what they know; if they value an action but forget the occasion, a cue–response plan may help; if access is blocked, structural provision may be decisive. These are testable choices about which determinant to manipulate and how to measure resulting action. The literature on implementation intentions and cross-domain behavioral change documents the importance of mechanisms that enable goal enactment Gollwitzer & Sheeran (2006) Albarracín et al. (2024).
For health, consumer, civic, and organizational decisions, public declarations and favorable ratings offer useful descriptive information but do not by themselves establish achievable behavior. Ethical interventions specify informed consent, equitable access, actual opportunity, and independently measured outcomes. A campaign that increases favorable survey responses has demonstrated a shift in those responses; a stronger claim about sustained behavior needs follow-up records and an appropriate comparison. Effectiveness should be stated at the level of the actual studied behavior and the relevant population rather than attributed to attitude transformation as a whole Webb & Sheeran (2006) Forscher et al. (2019).
Digital and artificial environments intensify these challenges because interface design continually changes which options are visible and which actions are effortless. Evaluating an artificial collaborator, expressing trust in its recommendations, and delegating a consequential decision are distinct observations. Research can estimate when stated confidence predicts delegation, when prior system performance shapes preferences, and when safeguards alter the decision process. Such work extends established measurement and causal principles into the Artificial Era while preserving rigorous distinctions between psychological evidence about human participants and theoretical questions about the organization of artificial systems.
A Program of Future Empirical Research
Future studies should compare multiple operationalizations of the same evaluation in shared prospective cohorts. A preregistered design could measure direct ratings, response latencies, attitude importance, certainty, prior experience, intended action, and objective opportunity, then observe specified behavior repeatedly. Within-person and between-person components should be estimated separately, with predictive models evaluated in a held-out population and future period. Such designs could clarify which combinations of features add information beyond prior behavior and local opportunity, and whether the effects are transportable across contexts Kraus (1995) Phipps et al. (2025).
Causal studies should randomize theoretically identified determinants while preserving independently measured outcomes. Factorial manipulations of evaluative information, action planning, and opportunity could test whether these components have independent effects or interact. Repeated follow-up would reveal persistence, delayed effects, and the possibility that prior action changes later evaluation. Research on habit disruptions and intervention targets already suggests plausible mechanisms to examine, but the required estimates must come from the actual randomized contrasts Wood et al. (2005) Gollwitzer & Sheeran (2006) Albarracín et al. (2024).
Comparative research should build designs for developmental, cultural, and institutional generalization from the outset. That entails participant-appropriate scales, tests of measurement equivalence, representative or clearly defined samples, transparent handling of attrition, and outcome measures calibrated to local opportunities. Digital studies should observe interface exposure and algorithmically modified choice sets, and examine whether interventions travel beyond the platform. Across all settings, cumulative science benefits from accessible analysis plans, complete outcome reporting, independent replication, and the willingness to revise explanatory models when new behavioral evidence conflicts with favored assumptions.
Conclusion
Attitudes, Evaluation, and the Attitude–Behavior Relation concerns the scientific study of how favorable and unfavorable evaluations are represented, expressed, measured, and connected to conduct. Its history extends from early demonstrations of the discrepancy between verbal declarations and social behavior through formal intention models, research on automatic evaluative activation, indirect measures, and large-scale quantitative synthesis. The strongest results establish that attitudes can predict behavior under defined conditions and that prediction improves when researchers measure the relevant action precisely, attend to evaluative stability and accessibility, and observe the decision context LaPiere (1934) Kraus (1995) Glasman & Albarracín (2006).
Empirical disagreement is part of the structure of the field. Studies of indirect and explicit measurement support a nuanced account of evaluative processes, while debates about discrimination criteria and intervention effects show why predictive correlation and demonstrated behavioral change must be interpreted separately. The contemporary meta-analyses identify nonzero average contributions from multiple attitude measures and highlight the limited practical value of translating a small predictive effect into strong claims about a particular person's future action Kurdi et al. (2019) Oswald et al. (2013) Forscher et al. (2019) Phipps et al. (2025).
The fundamental synthesis is that evaluative orientation contributes to conduct through identifiable relationships with knowledge, accessibility, intentions, plans, learned routines, interpersonal expectations, and available opportunities. These relationships can be examined with experimental manipulation, longitudinal observation, multimethod measurement, and systematic quantitative comparison. A mature science of attitudes explains the conditions of correspondence and divergence with the same rigor, grounding practical decisions in independently observed behavior and the real social environments in which that behavior occurs. This research program remains central to social psychology and extends naturally to the changing relations of Homo and Artificial in the Artificial Era.
References
- Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179–211. https://doi.org/10.1016/0749-5978(91)90020-T
- Albarracín, D., Johnson, B. T., Fishbein, M., & Muellerleile, P. A. (2001). Theories of reasoned action and planned behavior as models of condom use: A meta-analysis. Psychological Bulletin, 127(1), 142–161. https://doi.org/10.1037/0033-2909.127.1.142
- Albarracín, D., Fayaz-Farkhad, B., & Granados Samayoa, J. A. (2024). Determinants of behaviour and their efficacy as targets of behavioural change interventions. Nature Reviews Psychology, 3(6), 377–392. https://doi.org/10.1038/s44159-024-00305-0
- Armitage, C. J., & Conner, M. (2001). Efficacy of the theory of planned behaviour: A meta-analytic review. British Journal of Social Psychology, 40(4), 471–499. https://doi.org/10.1348/014466601164939
- Dovidio, J. F., Kawakami, K., Johnson, C., Johnson, B., & Howard, A. (1997). On the nature of prejudice: Automatic and controlled processes. Journal of Experimental Social Psychology, 33(5), 510–540. https://doi.org/10.1006/jesp.1997.1331
- Fazio, R. H., Sanbonmatsu, D. M., Powell, M. C., & Kardes, F. R. (1986). On the automatic activation of attitudes. Journal of Personality and Social Psychology, 50(2), 229–238. https://doi.org/10.1037/0022-3514.50.2.229
- Fazio, R. H., & Olson, M. A. (2003). Implicit measures in social cognition research: Their meaning and use. Annual Review of Psychology, 54, 297–327. https://doi.org/10.1146/annurev.psych.54.101601.145225
- Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior: An introduction to theory and research. Addison-Wesley. https://people.umass.edu/aizen/f&a1975.html
- Forscher, P. S., Lai, C. K., Axt, J. R., Ebersole, C. R., Herman, M., Devine, P. G., & Nosek, B. A. (2019). A meta-analysis of procedures to change implicit measures. Journal of Personality and Social Psychology, 117(3), 522–559. https://doi.org/10.1037/pspa0000160
- Gawronski, B., & Bodenhausen, G. V. (2006). Associative and propositional processes in evaluation: An integrative review of implicit and explicit attitude change. Psychological Bulletin, 132(5), 692–731. https://doi.org/10.1037/0033-2909.132.5.692
- Gawronski, B., & Corneille, O. (2025). Unawareness of attitudes, their environmental causes, and their behavioral effects. Annual Review of Psychology, 76, 359–384. https://doi.org/10.1146/annurev-psych-051324-031037
- Glasman, L. R., & Albarracín, D. (2006). Forming attitudes that predict future behavior: A meta-analysis of the attitude–behavior relation. Psychological Bulletin, 132(5), 778–822. https://doi.org/10.1037/0033-2909.132.5.778
- Gollwitzer, P. M., & Sheeran, P. (2006). Implementation intentions and goal achievement: A meta-analysis of effects and processes. Advances in Experimental Social Psychology, 38, 69–119. https://doi.org/10.1016/S0065-2601(06)38002-1
- Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (1998). Measuring individual differences in implicit cognition: The Implicit Association Test. Journal of Personality and Social Psychology, 74(6), 1464–1480. https://doi.org/10.1037/0022-3514.74.6.1464
- Hofmann, W., Gawronski, B., Gschwendner, T., Le, H., & Schmitt, M. (2005). A meta-analysis on the correlation between the Implicit Association Test and explicit self-report measures. Personality and Social Psychology Bulletin, 31(10), 1369–1385. https://doi.org/10.1177/0146167205275613
- Kraus, S. J. (1995). Attitudes and the prediction of behavior: A meta-analysis of the empirical literature. Personality and Social Psychology Bulletin, 21(1), 58–75. https://doi.org/10.1177/0146167295211007
- Kurdi, B., Seitchik, A. E., Axt, J. R., Carroll, T. J., Karapetyan, A., Kaushik, N., Tomezsko, D., Greenwald, A. G., & Banaji, M. R. (2019). Relationship between the Implicit Association Test and intergroup behavior: A meta-analysis. American Psychologist, 74(5), 569–586. https://doi.org/10.1037/amp0000364
- LaPiere, R. T. (1934). Attitudes vs. actions. Social Forces, 13(2), 230–237. https://doi.org/10.2307/2570339
- Oswald, F. L., Mitchell, G., Blanton, H., Jaccard, J., & Tetlock, P. E. (2013). Predicting ethnic and racial discrimination: A meta-analysis of IAT criterion studies. Journal of Personality and Social Psychology, 105(2), 171–192. https://doi.org/10.1037/a0032734
- Phipps, D. J., Hagger, M. S., & Hamilton, K. (2025). Relative effects of implicit and explicit attitudes on behavior: A meta-analytic review and test of key moderators. Psychological Bulletin, 151(12), 1395–1447. https://doi.org/10.1037/bul0000506
- Ranganath, K. A., Smith, C. T., & Nosek, B. A. (2008). Distinguishing automatic and controlled components of attitudes from direct and indirect measurement methods. Journal of Experimental Social Psychology, 44(2), 386–396. https://doi.org/10.1016/j.jesp.2006.12.008
- Sheeran, P., & Webb, T. L. (2016). The intention–behavior gap. Social and Personality Psychology Compass, 10(9), 503–518. https://doi.org/10.1111/spc3.12265
- Webb, T. L., & Sheeran, P. (2006). Does changing behavioral intentions engender behavior change? A meta-analysis of the experimental evidence. Psychological Bulletin, 132(2), 249–268. https://doi.org/10.1037/0033-2909.132.2.249
- Wicker, A. W. (1969). Attitudes versus actions: The relationship of verbal and overt behavioral responses to attitude objects. Journal of Social Issues, 25(4), 41–78. https://doi.org/10.1111/j.1540-4560.1969.tb00619.x
- Wilson, T. D., Lindsey, S., & Schooler, T. Y. (2000). A model of dual attitudes. Psychological Review, 107(1), 101–126. https://doi.org/10.1037/0033-295X.107.1.101
- Wood, W., Tam, L., & Guerrero Witt, M. (2005). Changing circumstances, disrupting habits. Journal of Personality and Social Psychology, 88(6), 918–933. https://doi.org/10.1037/0022-3514.88.6.918