RESEARCH DOSSIER
Self–Other Agreement and Multimethod Personality Assessment
Abstract. Self–other agreement investigates the correspondence, limits, and informational differences between individuals’ self-descriptions and the personality judgments of other people. This fundamental research monograph analyzes rank-order and mean-level agreement, interpersonal consensus, profile accuracy, differential accuracy, and the validity of multiple methods for assessing personality. Drawing on foundational multitrait–multimethod validation, social-relations approaches, Funder’s Realistic Accuracy Model, the Self–Other Knowledge Asymmetry model, original informant and behavioral-observation studies, latent-method models, and contemporary meta-analytic and cross-informant research, it evaluates the roles of observability, evaluativeness, acquaintance, relationship type, and information access. The synthesis distinguishes correlation from accuracy, source convergence from shared method bias, and laboratory performance from broader personality inference. It identifies reproducibility, cultural transport, computational judgment, and independent criterion validation as continuing scientific priorities. The method is a critical narrative synthesis of identifiable scholarly literature, not an original empirical experiment or a systematic meta-analysis.
Keywords: self–other agreement; personality assessment; informant ratings; observer reports; multitrait–multimethod; MTMM; construct validity; personality judgment; self–other knowledge asymmetry; SOKA; Realistic Accuracy Model; behavioral observation; trait observability; evaluativeness; interrater consensus; method effects; criterion validity; Social Relations Model; Artificial Era
The Scientific Problem of Knowing a Person
Self–other agreement concerns the correspondence between a person's account of their own personality and judgments supplied by people who observe that person. Multimethod personality assessment asks what is learned when self-ratings, informant ratings, trained behavioral observations, performance measures, and naturally occurring behavioral records are studied together. Agreement is a property of measured judgments under specified conditions. Accuracy additionally requires an appropriate criterion and a defensible account of how the criterion represents the target attribute. These distinctions organize a scientific field spanning personality psychology, social perception, psychometrics, and behavioral measurement. Early discussions often treated disagreement as an error to eliminate; contemporary research examines its informational structure and its possible relation to context Campbell & Fiske (1959) Funder (1995) Ashton & Lee (2025).
The unit of analysis may be a person's rank among other people, an absolute score difference, a profile of many traits, or a prediction of independent behavior. Two raters can rank people similarly while using different numerical scale levels. They can also agree about a person's general positivity while disagreeing about which trait is unusually pronounced. A correlation must therefore be interpreted with reference to the particular estimand, the rating scale, its reliability, the sampled targets, and the observers' access to relevant behavior. Biesanz et al. (2007) demonstrates how agreement statistics can conceal divergent patterns of acquaintance-related learning.
This monograph examines the historical formation of the problem, its theoretical mechanisms, the measurement architecture, the principal experimental and meta-analytic findings, and implications for research and applied interpretation. Its method is critical narrative synthesis of identifiable primary studies and research reviews, with attention to population, measurement, method effects, independent criteria, and inferential limits. It offers a testable synthesis of the available literature, without claiming an original experiment, systematic-review protocol, or universally valid personality truth criterion Connelly & Ones (2010) Mõttus & Henry (2025).
Construct Definition and Inferential Targets
Agreement between self and other can mean trait-wise rank-order correlation across persons, agreement in mean score levels, or similarity between two ratings of an individual's profile. Consensus among others refers to agreement between independent judges about the same target. Accuracy designates successful correspondence to a criterion that is informative about the characteristic being judged. Predictive validity concerns association with an independently measured outcome, preferably one observed later. Incremental validity asks whether one source contributes information beyond another. These questions share data but require distinct analyses and cannot be reduced to a single omnibus correlation Kenny (1991) Connelly & Ones (2010).
Consider two conscientiousness questionnaires completed by the target and a colleague. A moderate correlation across employees means that relatively high-scoring targets tend to be rated high by colleagues; it does not establish that every target's scores match, or that either observer correctly inferred a stable causal mechanism. An external record of work completion might provide a criterion for task-related behavior, yet it remains influenced by opportunity, work assignment, organizational resources, and supervisory expectations. The criterion captures a slice of functioning, not the entire psychological trait. This distinction follows the construct-validation tradition of Campbell & Fiske (1959) and the realism of independent behavioral criteria in Kolar et al. (1996).
Distinct traits have different observability, desirability, context dependence, and measurement reliability. Broad domains aggregate heterogeneous facets, some externally visible and some primarily accessible through private experience. An apparent self–other disagreement can arise from an inaccessible mental state, a relationship-specific response pattern, a scale-format artifact, or different interpretations of the same behavior. The scientific task is to determine which account is supported by triangulating measures, contexts, and independent outcomes John & Robins (1993) Vazire (2010).
Historical Foundations of Multimethod Validation
The systematic comparison of multiple measurements became a central problem of modern psychometrics when investigators recognized that agreement between two questionnaires might reflect a common response format. In 1959, Campbell & Fiske (1959) formulated the multitrait–multimethod matrix as an explicit argument for convergent and discriminant validation: measurements intended to represent one trait should show meaningful convergence across methods, while theoretically different traits should retain discriminability. This was a methodological intervention in the interpretation of measurement, rather than evidence that any specific trait inventory is valid in every population.
Subsequent personality research connected this validation problem to social judgment. McCrae & Costa (1987) compared self and peer ratings, adjective descriptions, and inventory scales to test five-factor representations. Across observers and instruments, the pattern suggested substantial convergence while preserving measurable informant variation. The achievement was methodological as much as taxonomic: independent judgment sources allowed researchers to question whether trait structure was restricted to self-description or also visible in social observation.
The 1980s and 1990s expanded the question from agreement to accuracy, acquaintance, and the social distribution of evidence. Funder & Colvin (1988) directly compared acquaintances with strangers observing a short video interaction. Kenny (1991) proposed a general model of interpersonal consensus, while Funder (1995) articulated the Realistic Accuracy Model. By the late 1990s, convergent personality assessment had developed into a program testing observable cues, the judgment process, relationship context, and independent behavior rather than relying solely upon correlations between self and peer questionnaires.
The Multitrait–Multimethod Logic
A multitrait–multimethod design crosses distinct psychological constructs with distinct methods of assessing them. Self-ratings, ratings by friends, and ratings by trained observers can be measured for each of several traits. The resulting covariance matrix allows assessment of convergence for one trait across methods, discrimination among traits assessed through one method, and possible shared method variance. Inference depends on reliable trait definitions, genuine variation in data-collection mechanisms, and a design in which traits and methods are not confounded. Campbell & Fiske (1959) supplies the founding validity argument; Eid (2000) develops a formal latent-variable model for the structure of these data.
Methods must be classified by their actual data-generating processes. A self-rated questionnaire and a friend-rated adaptation of the same questionnaire use different respondents but may share wording, scoring, and trait stereotypes. A friend report and a parent report may differ in relationship and observation settings while sharing the same item format. A trained observer's coded videotape differs in sampling and rating process from a continuously collected device trace. A genuinely informative multimethod design specifies both overlaps and differences instead of automatically treating every named source as statistically independent Eid et al. (2003) Funder et al. (2000).
Within an MTMM matrix, convergent evidence is strengthened when a trait is similarly represented across sufficiently independent methods. Discriminant evidence is strengthened when the same-method correlation between different traits remains distinguishable from same-trait cross-method correspondence. Neither criterion is mechanical: theoretically related traits legitimately covary, and narrow behavior samples may fail to capture internal states. Model interpretation therefore requires substantive hypotheses, sufficient indicators, explicit factor identification, and sensitivity checks. Eid et al. (2003) offers correlated-trait, correlated-method-minus-one modeling for separating trait and reference-method components.
Observer Categories and Information Access
A target can be rated by an intimate partner, sibling, parent, friend, coworker, clinician, trained laboratory observer, brief-interaction stranger, or someone interpreting digital records. Each perspective samples a different ecology of behavior. A partner may observe unguarded domestic conduct, a coworker may have repeated access to task performance, and a stranger may evaluate fluent speech, expressiveness, or visible sociability after minutes of interaction. The question of who knows the target is inseparable from the question of which characteristic and which behavior context matter Funder & Colvin (1988) John & Robins (1993).
Observation frequency, intimacy, and shared meaning do not increase in lockstep. Kenny et al. (1994) reviewed consensus across 32 studies and found that acquaintance was not a simple universal explanation of higher consensus; observer access and interpretation had to be modeled. Later, Connelly & Ones (2010) showed in a large meta-analytic integration that interpersonal intimacy increased informant accuracy beyond merely frequent contact, with important differences across traits. These findings can coexist because the studies ask different questions, use different indices and samples, and investigate different dimensions of familiarity.
Contemporary evidence makes relationship type an explicit design factor. In a Dutch university-student sample, Lee et al. (2024) examined reports from parents, siblings, friends, and partners and found high correlations between self- and observer-rated HEXACO dimensions, with particularly close correspondence for partners. Relationship differences in mean ratings also appeared. The sample design does not establish a universal ranking of relationships; it demonstrates why studies should report the social identity and context of each informant rather than treating all observers as interchangeable Lee et al. (2024).
The Realistic Accuracy Model
The Realistic Accuracy Model explains successful judgment as a sequence in which a relevant behavioral cue exists, becomes available to the judge, is detected, and is correctly utilized. Funder (1995) describes these linked stages as prerequisites for accurate inference. A highly diagnostic signal may carry little value if it never occurs in the observed situation or remains invisible to the judge. Likewise, abundant cues produce weak accuracy when the observer assigns them inappropriate meanings.
Good targets provide coherent and readable behavior; good traits have cues with observable relations to the construct; good information consists of diagnostic behavior sampled in contexts that reveal relevant differences; good judges detect and use those cues effectively. These are interactional properties rather than fixed reputations. A person who is easy to evaluate on outward sociability may be difficult to evaluate on internal distress. A brief stranger may judge a public expressive style effectively while lacking access to ordinary habitual behavior. The model predicts heterogeneous agreement according to the conditions of observation Funder (1995) Borkenau et al. (2004).
Direct behavioral observation tests the proposed pathway. Borkenau et al. (2004) combined videotaped episodes, peer reports, self-reports, and intelligence measures, finding that personality inferences from brief segments related to knowledgeable-informant judgments and improved with additional episodes. Such evidence supports cue accessibility but also highlights inference limits: when observable behavior is an imperfect criterion, convergence with another observer's rating does not by itself prove accuracy. A stronger design uses behavior sampled across occasions and evaluates the relative independence of each criterion.
The Social Relations Perspective
Personality judgments reflect the target, the judge, and the particular relationship. A rater may perceive everyone as more agreeable than another rater does, while a particular target may elicit especially high agreeableness ratings across judges. Some dyads have idiosyncratic experience that is not shared by other relationships. The social-relations approach decomposes these tendencies and encourages designs in which participants evaluate multiple others. Kenny (1991) formalized the analysis of interpersonal consensus, and Back et al. (2011) adapted social-relations and lens-model ideas to first impressions.
Round-robin designs can separate perceiver tendencies from target reputation and unique dyadic evaluation. Without such structure, a correlation between two people's reports may mingle the target's trait, the perceiver's general scale use, liking, and the history of their interactions. An individual who consistently evaluates others favorably could generate apparent agreement patterns even when target-specific differentiation is poor. Statistical designs with multiple targets per judge and multiple judges per target help identify where reliable information resides Back et al. (2011) Kenny (1991).
Agreement may emerge from shared stereotypes, common settings, or genuinely discriminating target behavior. The critical issue is whether a judge distinguishes one particular target from others after removing normative expectations about the average person. Biesanz et al. (2007) showed that conventional profile correlations can conceal acquaintance-related gains in differential accuracy. This is central to interpretation: a person can judge a typical personality profile accurately without accurately identifying which distinctive characteristics belong to the person being judged.
Trait Observability and Evaluative Charge
The content of a trait affects whether agreement is likely. Extraversion frequently has visible behavioral cues: how often a person enters conversation, speaks, approaches groups, or displays positive affect in social settings. Internal distress and anxious rumination are less directly available to an observer, while agreeableness and intellectual competence may involve value-laden judgments. John & Robins (1993) compared personality domains and found higher agreement for observable and less evaluative traits, with Extraversion showing the strongest agreement and Agreeableness the weakest in the studies examined. Their analyses demonstrate moderators of the inferential process rather than a fixed hierarchy of true personality dimensions.
The same research showed why inclusion of the self changes the comparison. Peer–peer correlations can exceed self–peer correlations in evaluatively charged domains, where self-protective interpretation or socially valued identity may influence ratings John & Robins (1993). Yet difference between sources is also produced by private information and relationship-specific observation. Self-advantage or observer-advantage cannot be inferred solely from a lower correlation with the other source; an external and genuinely pertinent criterion is required to compare predictive accuracy Vazire (2010) Kolar et al. (1996).
A careful design separates traits' actual observability from stereotype-based assumptions about observability. Independent coders may classify how often diagnostic cues occur in recorded settings; participants may judge which attributes are evaluatively sensitive; and informants can report their opportunities to observe targets. These dimensions can then be tested as predictors of agreement and criterion validity. Observability must be operationalized per behavior and setting, while evaluativeness may vary with culture, role, situation, and goals Connelly & Ones (2010) Funder (1995).
Self–Other Knowledge Asymmetry
The Self–Other Knowledge Asymmetry model, developed and tested by Vazire (2010), predicts an asymmetric distribution of useful personality information. The self has special access to less observable experiences; other people can possess a comparative advantage where evaluative motives constrain self-description. In the original round-robin investigation, 165 participants rated themselves and received ratings from friends and strangers, with independent behavioral testing supplying criterion measures. The model specifically predicted different relative accuracies for traits varying in observability and evaluativeness.
The study supported the expected pattern: the self was relatively accurate for neuroticism-related characteristics, friends for intellect-related attributes, and perspectives were comparably useful for extraversion-related traits Vazire (2010). The result gives the phrase 'blind spot' an empirical meaning: a person's judgments may omit reliably observable external information about performance or socially evaluated behavior. It does not follow that every friend has superior knowledge, nor that a person's account of private experience is automatically reliable. Accuracy is contingent on criterion quality, rating scales, and the information each judge could access.
The model changes research design by rejecting automatic primacy of either first-person testimony or reputational consensus. Studies should ask in advance which attributes require privileged inner access, which depend on observed behavior, and which invite defensiveness. Independent criteria, multiple judges, and temporal sampling can help arbitrate specific hypotheses. Vazire & Carlson (2011) integrates the implications for self-knowledge while acknowledging that different perspectives reveal different components of a complex person.
Self-Insight, Self-Presentation, and Motivated Judgment
Self-ratings can be influenced by memory reconstruction, social desirability, impression management, self-enhancement, selective attention, and uncertainty about comparative norms. An individual may observe many private intentions but remember them differently from completed actions. Conversely, an informant may overweight striking incidents, infer stable traits from role expectations, or use affective liking as a substitute for specific behavioral evidence. Thus rater-specific error has more than one psychological source and can be stable over time Beer & Watson (2008) Vazire & Carlson (2011).
Average self–other score gaps should not be treated as direct measurements of dishonesty or exaggerated self-esteem. Lee et al. (2024) explicitly assessed mean-level agreement in HEXACO ratings and reported little evidence for a generalized positive self-rating bias in the studied sample, although self-reported Openness was higher than observer reports. Such findings require attention to the traits selected, reference group, observer type, and the possibility of measurement noninvariance across rating perspectives. The empirical question concerns specific score structures, not universal claims about human vanity.
A useful analytic distinction separates deliberate self-presentation from genuine differences in information and interpretation. Participants may sincerely endorse a favorable statement because they recall different events from those recalled by colleagues. A dyad may also observe different legitimate expressions of the same person's behavior. Experimental manipulation of evaluative threat, respondent anonymity, item framing, and feedback can test motivational hypotheses. Such studies need independent behavioral criteria to determine whether a shift in ratings represents calibration or an altered social performance John & Robins (1993) Vazire (2010).
Assumed Similarity and Social Projection
Observers sometimes attribute their own traits, motives, or values to another person. Assumed similarity refers to similarity in ratings arising because a perceiver uses the self as a reference. This process can create agreement patterns that appear target-specific when they arise partly from judge-specific tendencies. Projection and normative assumptions may be especially influential when contact with a target is sparse or a trait has few visible signals. Testing assumed similarity requires both the perceiver's self-ratings and multiple target ratings, preferably within designs permitting separation of judge and target variance Funder et al. (1995) Kenny (1991).
An informative result comes from Funder et al. (1995), who studied ratings by different acquaintances, parents, and strangers. The authors found higher consensus and self–other agreement among acquaintances than among strangers, and reported that acquaintances' better performance derived more from distinctive judgments of the target than from simple assumed similarity. This counters an overly skeptical interpretation in which all apparent accuracy is attributed to projection or shared stereotypes.
Social projection remains theoretically relevant and must be evaluated as a component of judgment rather than as a universal explanation. A person who resembles the target in age or role may have a partially accurate prior about likely behaviors, yet can also miss distinctive conduct. Separating average stereotype accuracy, assumed similarity, and differential target accuracy prevents a model from mistaking broad population knowledge for precision about an individual. Biesanz et al. (2007) illustrates why these components change differently across acquaintance duration.
Acquaintance, Intimacy, and Contextual Diversity
Acquaintance operates through information exposure, not merely calendar time. Repeated contact within a highly constrained situation may show a narrow response repertoire. A close relationship may provide variation in mood, conflict, leisure, obligation, and unguarded behavior, while a long professional relationship may reveal reliability and cooperative style with little access to family or solitary life. Differences in agreement should therefore be modeled through diversity of observed behavior, opportunity to witness diagnostic events, personal relevance, and communication quality Funder & Colvin (1988) Connelly & Ones (2010).
The literature includes seemingly different findings because it asks how agreement is indexed. Kenny et al. (1994) did not find simple monotonic gains in consensus with acquaintance across the set of 32 studies it reviewed. Biesanz et al. (2007) later showed that acquaintance could improve differential accuracy even when raw profile correlations were stable. Connelly & Ones (2010) identified intimacy as a meaningful moderator of observers' agreement and predictive utility. These findings favor models that distinguish familiarity duration, type of knowledge, item content, and the statistical target rather than selecting one universal acquaintance effect.
Longitudinal designs can follow dyads from first meeting to extended familiarity, recording how behavioral exposure and perceived understanding evolve. Such designs must allow for selection: people who become close often differ from dyads who do not, and intimate partners may shape one another's opportunities and self-descriptions. Repeated observations permit separation of within-dyad learning, stable observer tendencies, changing target behavior, and shared environments. Agreement increases are interpretable only when measurement meaning remains sufficiently stable across occasions Biesanz et al. (2007) Kenny (1991).
Aggregation, Consensus, and Informant Independence
Combining several informants reduces idiosyncratic error to the extent that their errors are not perfectly shared. Kolar et al. (1996) compared self-ratings with judgments by knowledgeable others against videotaped behavioral criteria; averages of two acquaintance ratings performed better than self-reports for the particular behavioral outcomes examined. This observation motivates informant aggregation but does not make a numerical average an unconditional truth. Two judges who share a strong stereotype, are exposed to the same restricted context, or discuss ratings before completion may reproduce correlated biases.
Reliability improves with additional raters when they bring sufficiently independent information, but the effective informational gain depends on error covariance. A friend and coworker may sample more complementary behavior than two friends from the same social circle; alternatively, two observers in the same setting may produce more stable ratings of the behavior relevant to that setting. Weighting by empirically tested criterion prediction, rather than by closeness alone, can improve specific assessments, provided the weighting model is validated on held-out samples Connelly & Ones (2010) Lee et al. (2024).
Informant selection becomes part of the estimand: whose perception and which slice of behavior is the measurement meant to represent? For assessment of everyday sociability, context-diverse informants may improve coverage. For workplace collaboration, collaborators with repeated opportunities to observe relevant conduct may be more useful than a distant intimate. Published averages cannot be transported automatically across purposes, because trait salience, behavioral opportunities, and criterion definition shift by application Mount et al. (1994) Funder (1995).
Profiles, Elevation, and Distinctive Agreement
Personality inventories produce profiles containing many trait ratings. A profile correlation between judges can be high when both reproduce a culturally familiar pattern of what is generally common or desirable, even though they disagree about the target's distinctive departures. Profile comparison must therefore distinguish overall elevation, the relative ordering of traits within a profile, and differential accuracy for the specific person. Biesanz et al. (2007) used this distinction to identify learning across acquaintance that was invisible in ordinary profile-level summaries.
Mean-level agreement is another separate property. Two groups can rank targets similarly while one systematically assigns higher scores on Openness or Honesty–Humility. Lee et al. (2024) investigated both correlational and mean-level agreement across informant relationships and found differences that would be obscured by reporting only across-person correlations. Researchers should report absolute differences, score distributions, reliability, and uncertainty intervals alongside rank-order associations when the question concerns interchangeability of measures.
Analytic tools include profile centering, comparisons with normative trait means, and multilevel modeling of target, trait, and observer components. Such transformations must be declared in advance because they answer different questions, and some may remove meaningful shared information. A person-centered profile can be informative for counseling or research feedback, but a profile correlation is not a standardized clinical diagnosis or an exhaustive representation of character. Conclusions should be attached to the specific profile estimand and sample Biesanz et al. (2007) Connelly & Ones (2010).
Self-Ratings, Scales, and Response Processes
Self-report instruments efficiently collect responses about habits, preferences, emotions, and perceived dispositions. Their strengths include access to private experiences, scalable administration, standard item presentation, and comparability under appropriate norms. Their weaknesses include response styles, item misunderstanding, recollection error, social desirability, and reduced coverage of contextual behavior. A trait score usually summarizes responses to several items; reliability concerns the consistency of that summary, while validity asks what explanatory or predictive interpretation the score supports Campbell & Fiske (1959) Ashton & Lee (2025).
Scale design affects correspondence across raters. A first-person item about 'worrying a lot' and a third-person version answered by a friend may be linguistically parallel yet psychologically nonequivalent, because worry can remain unexpressed. Other items such as talking frequently have closer behavioral referents. Item-level analyses can identify which facets are especially vulnerable to asymmetric information, and measurement-invariance analyses can investigate whether the scale has similar structure for self and observer groups John & Robins (1993) Vazire (2010).
Ultra-short scales illustrate the trade-off between coverage and practicality. Gosling et al. (2003) developed very brief Big Five measures and reported evidence of convergence with longer measures and observer reports, alongside diminished psychometric precision compared with more extensive instruments. Agreement research based on sparse indicators must recognize that broad trait scores can be sensitive to item selection and to method-specific wording. Whenever practical, researchers should collect adequate facet coverage and report reliability separately for every rating perspective.
Informant Ratings and Reputation
An informant rating is a structured social perception of the target. It may reflect stable behavioral tendencies, contexts of encounter, reputational information from others, and the judge's own interpretive habits. Asking a close partner and a coworker the same adjective question does not create duplicate measurements of a single privileged psychological truth. Each report is an observation from a particular social position. Hofstee (1994) argued that aggregates of knowledgeable observers can be a strong basis for personality description, while Vazire (2010) clarified why that advantage depends on observability and evaluative demands.
Informant-based measures should specify how respondents were recruited, the depth and kind of acquaintance, the relation to the target, how ratings were elicited, and whether judges were independent. Recruitment through the target can produce selective sampling of agreeable relationships. Informants' incentives and fear of disclosure may also alter response patterns, especially in hierarchical or conflictual relationships. Comparisons should therefore report the selection mechanism and plausible shared biases rather than treat 'observer' as a homogeneous category Connelly & Ones (2010) Lee et al. (2024).
The reputation captured by agreement among peers is itself meaningful social information. A social group may consistently expect a person to be dependable, cooperative, or volatile, and those expectations may influence opportunities and treatment. Reputation can predict outcomes partly because institutions and other people respond to it, even when it is an imperfect estimate of stable dispositions. Predictive validity of reputation should thus be investigated together with social consequences and independently measured behaviors Mount et al. (1994) Oltmanns & Turkheimer (2009).
Trained Behavioral Observation and Thin Slices
Trained behavioral coding converts observed actions into defined units: speaking turns, facial expressions, interruptions, task persistence, behavioral initiative, or other prespecified categories. A laboratory interaction can be recorded and coded by multiple judges using a shared manual. Such methods can reduce dependence on a participant's retrospective summary, but their validity depends on whether the selected situation elicited diagnostic behavior and whether the coding scheme captures the intended construct. Funder et al. (2000) introduced the Riverside Behavioral Q-sort as a structured approach to describing behavior in social interaction.
Thin-slice studies test what a judge can infer from short samples. Borkenau et al. (2004) used 15 videotaped episodes for 600 participants and found meaningful links between observer impressions and informant reports, with improving performance as behavioral segments accumulated. Some information appears quickly in visible conduct, but brief exposure favors traits with strong outward cues and may increase vulnerability to shared stereotypes. The distinction between an observer seeing something and interpreting it accurately remains central Funder (1995).
Observation studies require careful attention to coder training, interrater reliability, blinding, situational representativeness, and the possibility that the camera or task changes target behavior. Observed behavior should be sampled across situations when the intended construct is broad. Using a video-judgment score as an external criterion is stronger than correlating two questionnaires, yet it is not fully independent if the same stereotypes guide video judges and personality raters. Robust triangulation combines direct behavior, temporally separated sources, and clearly separated assessment teams Kolar et al. (1996) Borkenau et al. (2004).
Behavioral Traces and Ecological Sampling
Behavioral traces can include recorded communication frequency, daily activity patterns, workplace outputs, interpersonal interactions, or momentary reports collected during ordinary life. Such sources promise repeated observations outside a single interview and may improve the temporal coverage of trait-related behavior. Their interpretation is difficult because traces are jointly produced by personality, technology, occupational obligations, social network opportunities, privacy choices, and situational demands. The measured variable needs a transparent bridge to the intended personality construct Mõttus & Henry (2025) Funder (1995).
A high message count cannot be equated directly with Extraversion: it may reflect a demanding job, crisis communication, automated systems, or a small number of unusually active relationships. Likewise, wearable activity can reflect health, transport, access to safe spaces, or working conditions. The validity of a trace requires replication of its relation to the same construct across independent settings and comparison with sources that have different biases. As Campbell & Fiske (1959) emphasized, convergence is especially informative when its contributing measurements have meaningfully distinct error structures.
Ecological sampling is valuable for characterizing within-person variability. A person may be more assertive at work than at home; a trait questionnaire may represent their broader tendency while observers judge one local role. Repeated sampling permits estimation of context-dependent states and can reveal conditions under which ratings diverge. Research should protect informed consent and participant privacy, minimize unnecessary collection, and avoid extending traces to unvalidated judgments about identity or health. These ethical conditions are part of credible design, especially in digitally mediated assessment.
External Criteria and Predictive Validity
Accuracy studies require a choice of criterion that is neither simply the target's self-rating nor a circular translation of the observer's judgment. Candidate criteria include independently coded behavior, objectively scored performance, temporally subsequent actions, or multi-informant composites under stated assumptions. Each criterion offers a limited window on the construct. A test score may reflect ability under standardized conditions, while an observed behavior reflects performance in a particular circumstance Vazire (2010) Kolar et al. (1996).
The studies of Vazire & Mehl (2008) compared self and other ratings with samples of daily behavior and examined whether each source contributed unique prediction. Mount et al. (1994) found that observer-rated personality dimensions predicted job performance ratings in a sample of 105 sales representatives, with incremental information beyond self-ratings for relevant dimensions. These studies support context-dependent value of observer information. Their samples, criterion structures, and timing limit broad claims about clinical diagnosis, moral character, or universal differences in observer superiority.
A criterion can share method variance with the predictor. For example, a supervisor's trait assessment and their performance evaluation may both reflect the same evaluator's global impression. Predictive correlations are stronger evidence when criterion measurement is independent of the original raters. When overlap cannot be eliminated, reporting it explicitly allows readers to assess possible contamination. Study design should also separate contemporaneous fit from prospective prediction and specify whether results are descriptive or intended to support decisions Mount et al. (1994) Connelly & Ones (2010).
Latent Variable Models and Method Effects
Confirmatory factor models can decompose observed scores into variance attributed to intended traits, systematic method effects, and residual measurement error under explicit assumptions. A traditional MTMM matrix provides a descriptive validity map; latent models attempt to estimate which components generate the observed covariance. Eid (2000) proposed a multitrait–multimethod model with minimal assumptions, and Eid et al. (2003) developed a multiple-indicator correlated-trait, correlated-method-minus-one approach. These frameworks are useful because method bias is frequently trait-specific, and a universal 'method factor' can oversimplify how perspectives differ.
In reference-method models, one method establishes a contrast against which other methods are interpreted. The trait component is therefore anchored to a defined reference perspective rather than automatically corresponding to an observer-independent psychological essence. A self-report reference changes the meaning of residual informant effects relative to an observer-report reference. Researchers should state why the reference was selected, whether results are sensitive to that choice, and how model fit and identification were evaluated Eid et al. (2003).
Latent methods require more than fit indices. At least some indicators must be sufficiently reliable and distinguishable; samples must support the number of estimated parameters; and local dependence, nonlinear responses, missingness, and grouping must be assessed. A mathematically identified model can still be substantively misleading if multiple methods share sources or if constructs are poorly differentiated. Investigators should report standardized estimates and uncertainty while testing plausible alternative structures Eid (2000) Campbell & Fiske (1959).
Dependability, Agreement Metrics, and Statistical Design
Pearson correlations describe rank-order association, intraclass correlations may describe agreement or reliability under particular random-effects definitions, and absolute-difference measures describe calibration at the score level. Profile correlations add another layer by comparing many variables within targets. None is an interchangeable synonym for accuracy. The appropriate statistic depends on whether raters are sampled from a population, whether systematic level differences are meaningful, and whether one or several judgments will be used in practice Kenny et al. (1994) Biesanz et al. (2007).
Studies should distinguish single-rater from averaged-rater reliability and avoid treating multiple judges rating the same target as statistically independent targets. Hierarchical models can represent judge and target clustering, while dyadic and round-robin models can represent their cross-classified structure. Sampling targets with restricted trait variability attenuates correlations even when judges use cues consistently. Reliability differences between self and other measures can also create apparent differences in validity that should be examined with measurement-error-aware models Connelly & Ones (2010) Eid et al. (2003).
Transparent analysis plans specify traits, raters, correlations, model specification, exclusion rules, and independent criteria in advance. Multiple traits and relationship groups create many possible contrasts; selective reporting of the largest agreements can inflate conclusions. Precision should be communicated with confidence intervals and alternative estimators under plausible missing-data assumptions. The goal is cumulative evidence about when methods converge, where they differ, and which differences predict consequential observations Mõttus & Henry (2025) Connelly & Ones (2010).
Landmark Empirical Evidence and Cumulative Results
Independent empirical traditions converge on a qualified conclusion: observers frequently provide information about personality that is consistent with, and sometimes incrementally predictive beyond, self-report. McCrae & Costa (1987) linked self-reports with peer ratings across instruments; Funder & Colvin (1988) demonstrated stronger judgments by acquaintances than strangers; Kolar et al. (1996) connected informant judgments to independent videotaped behavior; and Vazire (2010) identified trait-specific advantages of self and other perspectives. These designs have distinct sampling frames and outcome measures. Their value lies in the consistency of the general multimethod research program, not in treating the observed effects as interchangeable.
The most extensive early synthesis in this corpus is Connelly & Ones (2010), which combined three meta-analyses based on 44,178 targets across 263 independent samples. The authors distinguished interrater consensus, self–other correlations, and prediction of behavior and assessed informational moderators. The breadth of evidence makes a stronger case for the scientific value of informant data than any single experiment. Yet samples differed in source selection, trait content, and applied settings; pooled results should be read alongside variation across studies and the limits of available effect measures.
More recent research challenges simplistic assumptions about response bias. In a Dutch university-student sample of 5,405, Lee et al. (2024) reported strong correlational self–observer agreement across HEXACO dimensions and relationship categories, with partner reports showing close correspondence. The authors did not find a general self-enhancement pattern across the scales. Ashton & Lee (2025) reviewed contemporary self- and observer-report evidence, while Mõttus & Henry (2025) argued that combining methods can reduce biases embedded in single-method scores. These findings support method diversification but demand replication with different populations, languages, criteria, and data-generation systems.
Cross-Cultural Meaning and Measurement Transport
Personality terms, social norms, and opportunities to observe behavior vary across cultural and linguistic settings. A label such as assertive may denote desirable initiative in one context and disrespectful interruption in another. The same outward behavior may be interpreted through different status expectations or relationship obligations. Cross-cultural comparisons require carefully translated instruments, comparable construct meaning, and assessment of response styles and structural invariance rather than reliance on identical item wording alone McCrae & Terracciano (2005) Campbell & Fiske (1959).
Observer data have been collected across many national settings. McCrae & Terracciano (2005) examined observer personality descriptions across cultures and found evidence of broad structural similarities, while their related work on aggregate cultural profiles cautions against equating a nation's average scores with the personality of an individual member. Even when broad factor structure generalizes, a particular trait's observability, desirability, and contextual expression may vary. Self–other correlations in one sample should not be assumed to describe other groups without equivalent measurements and comparably selected informants.
Cross-cultural research should examine whether partner, parent, coworker, and stranger roles carry equivalent observational privileges and evaluative consequences. It should also report translation, response anchoring, demographic composition, social position, and whether norms were generated locally. Multigroup latent models can probe measurement invariance; ethnographic observation and qualitative interviews can clarify what rating words mean in a community. Agreement becomes an empirical relation between situated judgments, not a substitute for cultural equivalence.
Development, Social Roles, and Personality Change
Age changes the sources of available personality information. Children may have difficulty describing some dispositions or recalling behavior over long periods, making parent and teacher reports valuable but contextually selective. Adolescents gain private experiences and peer ecologies that parents may not observe. In adulthood, informant perspectives shift across family, work, and intimate relationships. At older ages, cognitive, physical, and social changes can alter both target behavior and observers' opportunities for contact. The meaning of self–other disagreement therefore depends partly on development and life situation Achenbach et al. (1987) Connelly & Ones (2010).
The cross-informant literature on emotional and behavioral problems in youth provides a methodological comparison. Achenbach et al. (1987) synthesized correlations among parents, teachers, and other observers and documented substantial discrepancies between contexts and types of informant. This is not direct evidence about normal adult personality traits, but it illustrates why low agreement may encode situational specificity and distinct observational access. Researchers should not label a child's or adolescent's report erroneous solely because it differs from a teacher's judgment.
Longitudinal personality assessment faces a further question: does observed agreement change because raters become better informed, because the target's traits change, or because relationships and standards change? Repeated parallel ratings and independent behavior records are needed to disentangle these explanations. Studies should distinguish stable between-person rank ordering from changing within-person manifestations and should evaluate whether instruments retain comparable meaning at successive stages Biesanz et al. (2007) Vazire & Mehl (2008).
Applied Interpretation and Clinical Boundaries
Applied assessment is strongest when its source selection is matched to a defined decision. In organizational research, colleagues may observe behaviors relevant to coordination and dependability. Mount et al. (1994) found observer ratings contributed to predicting selected job performance ratings in a small sample of sales representatives. Such results do not justify universal algorithmic selection rules or use of personality profiles as decisive evidence about an individual's worth. Prediction must be evaluated for the actual population, outcome, fairness implications, and independence of predictor and criterion.
Clinical and personality-pathology research uses informant reports to explore interpersonal patterns that may not be fully represented by self-report. Oltmanns & Turkheimer (2009) outlined the potential contribution of person perception to personality-pathology assessment and emphasized that informants often supply a different perspective. Clinical use requires professionally appropriate instruments, diagnostic boundaries, consideration of distress and functional impairment, and attention to relationships that may be coercive or conflictual. A high or low agreement score is not a diagnosis, and different information sources should be evaluated in a qualified assessment process.
Disagreement can offer a starting point for further inquiry when the assessment context permits informed and respectful feedback. A person may discover that coworkers experience their assertiveness differently from intended; an observer may learn that a target experiences unexpressed worry. Neither interpretation is established by discrepancy alone. Transparent reporting should describe the domain, informant relationship, source sampling, measurement error, and practical uncertainty. Where disclosures can affect employment, treatment, or personal safety, consent and data governance are material scientific constraints Vazire & Carlson (2011) Mõttus & Henry (2025).
Artificial Era and Computational Personality Inference
Artificial Era assessment extends the multimethod problem to machine-mediated judgments of human behavior. Language-model or statistical systems may infer trait labels from text, interaction traces, or observed behavior. Such systems are additional computational measurement procedures; their estimates require validation against clearly defined constructs and independent behavioral criteria. Training labels often originate in human self-ratings or informant reports, so agreement with those labels may reproduce their biases rather than demonstrate independently grounded personality knowledge Mõttus & Henry (2025) Campbell & Fiske (1959).
Repeated digital interactions can provide richer temporal samples but create correlated dependencies. A person's writing style in one platform is shaped by audience, genre, interface prompts, privacy expectations, and institutional setting. A model's output may change when prompt wording or evaluation rubric changes, just as a human observer's ratings change with information availability. Computational studies should report dataset provenance, leakage controls, label-source characteristics, sensitivity to setting, calibration, demographic transport, and whether predictions add information beyond simple baselines. These criteria adapt the multimethod validity problem to artificial inference Eid et al. (2003) Funder (1995).
A further question concerns artificial judges' own response processes. Their apparent consensus may arise from shared training corpora, shared labels, similar architectures, or copied prompt templates; multiple models are not automatically independent observers. Valid comparison requires variation in evidence channels, transparency about dependence, and behavioral criteria unavailable during training. Within the philosophical horizon of From Homo to Artificial, psychology can analyze how artificial and human systems distribute observation, classification, and epistemic authority while identifying which claims are testable and which remain conceptual.
Critical Limitations and Open Research Problems
The major limitation of agreement research is criterion ambiguity. If personality has no uniquely decisive observable test, then disagreement between two raters cannot always be adjudicated as one correct and one incorrect judgment. A benchmark derived from informant consensus may privilege reputation; a self-report benchmark privileges introspection; a brief recorded interaction privileges situational performance. Strong studies compare multiple criteria, state each criterion's representational scope, and quantify the consequences of uncertainty. Funder (1995) and Vazire (2010) give complementary reasons to study cue access and trait-specific accuracy.
Selection and dependency present further problems. Informants are rarely randomly selected from everyone who knows a target; close relationships are chosen and maintained through shared experiences. Individuals may recruit people they expect to agree with them, while judges may share information with one another. Limited variation in sampled targets, item wording effects, differential reliability, and weak statistical power can create or suppress observed agreement. More representative sampling, preregistered hypotheses, multilevel data analysis, independent replication, and explicit attrition modeling are needed Connelly & Ones (2010) Eid et al. (2003).
Current work should resolve how broad domains relate to narrow facets, which contexts provide diagnostic cues, how developmental changes alter who has information, how cultural semantics change perceived evaluativeness, and when digital traces improve inference. A mature program will compare prospective behavior prediction with conversational and clinical judgments, report model uncertainty, and investigate the social consequences of being categorized. Lee et al. (2024), Ashton & Lee (2025), and Mõttus & Henry (2025) offer recent foundations for such tests, without supplying a final general formula for truth across all psychological perspectives.
Scientific Synthesis and Future Research Design
The evidence supports a conditional architecture of personality assessment. A person contributes private experience, self-reflection, and an account of intentions; informants contribute observations of repeated social conduct and reputation; trained behavioral observers contribute standardized description of selected episodes; and behavioral records contribute traces across time. Each source supplies a partial measurement of a multifaceted construct. The appropriate synthesis depends on what is being inferred: ranking among people, profile distinctiveness, prediction of behavior, developmental change, or a setting-specific assessment Vazire (2010) Connelly & Ones (2010) Mõttus & Henry (2025).
A scientifically strong design begins with an operational trait definition and preregistered criterion. It then recruits targets across relevant demographic and behavioral ranges, samples several informants across relationship contexts, collects self-report and direct observation without mutual contamination, and records independent outcomes where feasible. Multitrait–multimethod analyses investigate construct convergence; judge-target models separate relational effects; longitudinal or ecological models track change; and out-of-sample predictions test applied usefulness. Researchers must report uncertainty and sensitivity to the definition of the target construct Campbell & Fiske (1959) Kenny (1991) Eid et al. (2003).
The principal theoretical implication is that knowledge of personality is distributed across perspectives and observational environments. Relative accuracy is produced by interaction among trait content, cue availability, observer experience, evaluative pressure, measurement design, and criterion choice. This explains why some self-ratings are uniquely informative, why knowledgeable informants often add predictive value, and why partial disagreement may contain actionable information. The next stage of research lies in explicit source-combination models and transparent validation against independent and diverse evidence, including machine-mediated judgments in the Artificial Era Funder (1995) Ashton & Lee (2025).
Replication Protocol and Transparent Evidence Integration
An informative replication program would define a target trait taxonomy in advance, select item sets with established self- and observer-report forms, and document the equivalence of each form. It would collect independent ratings from several relationship categories and recruit targets across diverse ages, educational backgrounds, and social settings. Behavioral tasks and naturalistic observation would be selected for their relevance to particular trait predictions, with external raters blinded to questionnaire responses. The sampling frame, exclusions, missingness handling, anticipated effect sizes, and preregistered contrasts would be publicly reported before the primary analyses Campbell & Fiske (1959) Vazire (2010) Lee et al. (2024).
The primary estimands would include rank-order self–other correlations for each trait, absolute self–other score differences, observer consensus, distinctive profile accuracy after normative centering, and incremental prediction of independently assessed outcomes. Secondary models would estimate judge, target, and relationship components, followed by latent-method analyses anchored to a declared reference method. Reliability uncertainty and measurement noninvariance would be examined directly, and the analysis would be repeated under plausible alternative item definitions. Replication would require a genuinely separate sample, with all computational preprocessing and model selection specified to avoid analytic flexibility Kenny (1991) Eid et al. (2003) Biesanz et al. (2007).
A meta-research program should distinguish research reporting transparency from formal systematic-review methodology. Database searches, eligibility criteria, protocol registration, duplicate screening, study-quality appraisal, and publication-bias evaluation would be necessary before claiming a new systematic meta-analysis. This monograph instead integrates a deliberately selected, traceable evidence corpus and makes no pooled effect-size estimate. Future reviews should stratify by trait, informant relationship, observation context, target age, criterion independence, and measurement instrument. This stratification would reveal how far robust average associations can support individual or applied inference, and where cultural and technological transformations demand renewed validation Connelly & Ones (2010) Mõttus & Henry (2025).
Conclusion
Self–other agreement is a family of measurable relations among judgments, rather than a self-evident index of personal truth. Its proper interpretation requires a clear distinction between consensus, rank-order agreement, mean-level similarity, profile accuracy, and prediction of independent behavior. Historical work in construct validation and interpersonal perception established the vocabulary and measurement designs; modern empirical studies demonstrate that the reliability and usefulness of particular perspectives vary with trait visibility, evaluativeness, acquaintance, and observational context Campbell & Fiske (1959) John & Robins (1993) Vazire (2010).
Multimethod personality assessment draws strength from the complementary information carried by self-description, knowledgeable observers, structured behavioral coding, and longitudinal records. Its validity improves when methods contribute genuinely independent information and when their systematic biases are explicitly modeled. The meta-analytic evidence demonstrates the practical value of informant judgments while remaining conditional on samples, traits, and outcomes. Contemporary research shows that simplistic universal assumptions about self-enhancement, acquaintance, or relationship superiority require replacement with source- and trait-specific estimates Connelly & Ones (2010) Lee et al. (2024).
The resulting research program unites psychometric validity, social perception, behavioral science, and computational measurement. It recommends independent criteria, multiple rater relationships, context-diverse observations, transparent latent models, replication, and careful treatment of privacy and decision consequences. These principles extend naturally to artificial assessment systems whose inferences depend on training labels and mediated behavioral evidence. The central conclusion is precise: rigorous personality knowledge emerges from adequately specified, independently validated relations among perspectives, rather than from unexamined deference to any single source Eid et al. (2003) Mõttus & Henry (2025).
References
- Achenbach, T. M., McConaughy, S. H., & Howell, C. T. (1987). Child/adolescent behavioral and emotional problems: Implications of cross-informant correlations for situational specificity. Psychological Bulletin, 101(2), 213–232. https://doi.org/10.1037/0033-2909.101.2.213
- Ashton, M. C., & Lee, K. (2025). Self- and observer reports of personality. Annual Review of Psychology, 76, 771–795. https://doi.org/10.1146/annurev-psych-020124-115044
- Back, M. D., Schmukle, S. C., & Egloff, B. (2011). A closer look at first sight: Social relations lens model analysis of personality and interpersonal attraction at zero acquaintance. European Journal of Personality, 25(3), 225–238. https://doi.org/10.1002/per.790
- Beer, A., & Watson, D. (2008). Asymmetry in judgments of personality: Others are less differentiated than the self. Journal of Personality, 76(3), 535–560. https://doi.org/10.1111/j.1467-6494.2008.00495.x
- Biesanz, J. C., West, S. G., & Millevoi, A. (2007). What do you learn about someone over time? The relationship between length of acquaintance and consensus and self-other agreement in judgments of personality. Journal of Personality and Social Psychology, 92(1), 119–135. https://doi.org/10.1037/0022-3514.92.1.119
- Borkenau, P., Mauer, N., Riemann, R., Spinath, F. M., & Angleitner, A. (2004). Thin slices of behavior as cues of personality and intelligence. Journal of Personality and Social Psychology, 86(4), 599–614. https://doi.org/10.1037/0022-3514.86.4.599
- Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 56(2), 81–105. https://doi.org/10.1037/h0046016
- Connelly, B. S., & Ones, D. S. (2010). An other perspective on personality: Meta-analytic integration of observers’ accuracy and predictive validity. Psychological Bulletin, 136(6), 1092–1122. https://doi.org/10.1037/a0021212
- Eid, M. (2000). A multitrait-multimethod model with minimal assumptions. Psychometrika, 65(2), 241–261. https://doi.org/10.1007/BF02294377
- Eid, M., Lischetzke, T., Nussbeck, F. W., & Trierweiler, L. I. (2003). Separating trait effects from trait-specific method effects in multitrait-multimethod models: A multiple-indicator CT-C(M-1) model. Psychological Methods, 8(1), 38–60. https://doi.org/10.1037/1082-989X.8.1.38
- Funder, D. C., & Colvin, C. R. (1988). Friends and strangers: Acquaintanceship, agreement, and the accuracy of personality judgment. Journal of Personality and Social Psychology, 55(1), 149–158. https://doi.org/10.1037/0022-3514.55.1.149
- Funder, D. C. (1995). On the accuracy of personality judgment: A realistic approach. Psychological Review, 102(4), 652–670. https://doi.org/10.1037/0033-295X.102.4.652
- Funder, D. C., Furr, R. M., & Colvin, C. R. (2000). The Riverside Behavioral Q-sort: A tool for the description of social behavior. Journal of Personality, 68(3), 451–489. https://doi.org/10.1111/1467-6494.00103
- Funder, D. C., Kolar, D. W., & Blackman, M. C. (1995). Agreement among judges of personality: Interpersonal relations, similarity, and acquaintanceship. Journal of Personality and Social Psychology, 69(4), 656–672. https://doi.org/10.1037/0022-3514.69.4.656
- Gosling, S. D., Rentfrow, P. J., & Swann, W. B., Jr. (2003). A very brief measure of the Big-Five personality domains. Journal of Research in Personality, 37(6), 504–528. https://doi.org/10.1016/S0092-6566(03)00046-1
- Hofstee, W. K. B. (1994). Who should own the definition of personality? European Journal of Personality, 8(3), 149–162. https://doi.org/10.1002/per.2410080302
- John, O. P., & Robins, R. W. (1993). Determinants of interjudge agreement on personality traits: The Big Five domains, observability, evaluativeness, and the unique perspective of the self. Journal of Personality, 61(4), 521–551. https://doi.org/10.1111/j.1467-6494.1993.tb00781.x
- Kenny, D. A. (1991). A general model of consensus and accuracy in interpersonal perception. Psychological Review, 98(2), 155–163. https://doi.org/10.1037/0033-295X.98.2.155
- Kenny, D. A., Albright, L., Malloy, T. E., & Kashy, D. A. (1994). Consensus in interpersonal perception: Acquaintance and the Big Five. Psychological Bulletin, 116(2), 245–258. https://doi.org/10.1037/0033-2909.116.2.245
- Kolar, D. W., Funder, D. C., & Colvin, C. R. (1996). Comparing the accuracy of personality judgments by the self and knowledgeable others. Journal of Personality, 64(2), 311–337. https://doi.org/10.1111/j.1467-6494.1996.tb00513.x
- Lee, K., de Vries, R. E., & Ashton, M. C. (2024). Self/observer agreement in personality assessment by observers’ relationship types. Journal of Research in Personality, 113, Article 104529. https://doi.org/10.1016/j.jrp.2024.104529
- McCrae, R. R., & Costa, P. T., Jr. (1987). Validation of the five-factor model of personality across instruments and observers. Journal of Personality and Social Psychology, 52(1), 81–90. https://doi.org/10.1037/0022-3514.52.1.81
- McCrae, R. R., & Terracciano, A. (2005). Universal features of personality traits from the observer’s perspective: Data from 50 cultures. Journal of Personality and Social Psychology, 88(3), 547–561. https://doi.org/10.1037/0022-3514.88.3.547
- Mõttus, R., & Henry, S. J. (2025). Using multi-method data for more accurate research findings. Current Opinion in Psychology, 65, Article 102075. https://doi.org/10.1016/j.copsyc.2025.102075
- Mount, M. K., Barrick, M. R., & Strauss, J. P. (1994). Validity of observer ratings of the Big Five personality factors. Journal of Applied Psychology, 79(2), 272–280. https://doi.org/10.1037/0021-9010.79.2.272
- Oltmanns, T. F., & Turkheimer, E. (2009). Person perception and personality pathology. Current Directions in Psychological Science, 18(1), 32–36. https://doi.org/10.1111/j.1467-8721.2009.01601.x
- Vazire, S. (2010). Who knows what about a person? The self–other knowledge asymmetry (SOKA) model. Journal of Personality and Social Psychology, 98(2), 281–300. https://doi.org/10.1037/a0017908
- Vazire, S., & Carlson, E. N. (2011). Others sometimes know us better than we know ourselves. Current Directions in Psychological Science, 20(2), 104–108. https://doi.org/10.1177/0963721411402478
- Vazire, S., & Mehl, M. R. (2008). Knowing me, knowing you: The accuracy and unique predictive validity of self-ratings and other-ratings of daily behavior. Journal of Personality and Social Psychology, 95(5), 1202–1216. https://doi.org/10.1037/a0013314