How to Compare Your View With a Close Observer’s

How should you frame the disagreement first?

Neither your account nor a close observer’s account wins by default. Start by asking what each statement is claiming. “You left early” describes an event that could be checked against what happened. “I felt overwhelmed” reports an internal experience that the other person may not have had direct access to. “You dislike people” interprets a broader pattern and its possible meaning. Those statements can all arise from one evening, but they are not three versions of the same claim.

A useful comparison therefore has a small unit: a particular behavior, in a particular setting, over a stated period. Suppose, as an invented example, that someone leaves a crowded dinner after an hour. Their friend calls this evidence that they dislike socializing; the person says the noise became tiring and they wanted a quiet evening. The departure is shared evidence. The feeling is the person’s report. The broad conclusion about liking people goes beyond that event. Calling the accounts contradictory would hide these different steps of reasoning.

A close observer may have seen the same behavior repeatedly and may notice a pattern the person overlooks. That report deserves examination when it includes relevant examples. But repeated exposure helps only if the observer encountered the settings the claim concerns, and the interpretation still needs support beyond the observed action. As the comparison becomes clearer, a conclusion can retain a broad tendency, narrow it to the situations or behaviors actually described, or remain suspended when the accounts do not yet address the same thing.

What does broad agreement research actually tell us?

A large average similarity between self-ratings and informant ratings does not tell you whether one particular pair agrees, or which person is more accurate. It answers a group-level question: when researchers compare the average rating given by targets with the average rating given by informants across many people, do those two averages tend to differ? This mean-level comparison concerns the centers of two groups of ratings. It is not the same as checking whether each target and their own informant give similar descriptions, and neither comparison alone establishes accuracy against an external criterion.

The meta-analysis “Self-Other Agreement in Personality Reports: A Meta-Analytic Comparison of Self- and Informant-Report Means” combined 152 samples covering 33,033 targets. Across Big Five report means, the average standardized difference was close to zero (δ = −.038), with no general pattern of informants rating targets lower than targets rated themselves. This result weighs against a simple explanation that personality disagreements generally happen because people, as a group, describe themselves more favorably than their informants do. It corrects a tempting story about average self-enhancement; it does not establish that self-ratings are free of bias or that informants are free of it.

A near-zero average can coexist with substantial variation around it. Some informants may rate a target higher than the target rates themself, while other informants may rate the same broad dimension lower; differences across traits, relationships, and measures can also balance in an overall mean. Even if a study found close average levels, that would not show that each matched self-informant pair agrees. Conversely, a particular pair can disagree while both ratings contribute valid information about different settings or aspects of the trait. The group result neither erases those cases nor identifies which explanation applies to them.

Agreement and accuracy are also separate questions. Two people can give similar ratings and still share a mistaken assumption; a disagreement does not by itself show that either rating is mistaken. To judge accuracy, a researcher needs a relevant criterion, such as behavior recorded under conditions that bear on the description. That criterion has limits of its own: recorded actions can test claims about observable actions more directly than claims about private feelings or broad meanings. The key shift is from counting agreement as a verdict to asking what evidence matches the exact claim being made.

The meta-analysis also cannot tell you whether a close observer has enough access to notice a recurring behavior, whether you have information about an internal experience that they lack, or whether either of you is generalizing from a narrow set of occasions. Informants in the samples varied, as did the measures. A group-level average is therefore a useful correction to blanket claims about self-favoring reports, not an answer key for one relationship. In an individual comparison, the next question is what kind of evidence could bear on the disputed description and whether both accounts refer to the same behavior, context, and time span.

Sources: Self-Other Agreement in Personality Reports: A Meta-Analytic Comparison of Self- and Informant-Report Means

What can an external accuracy criterion establish?

Criterion validity asks whether a judgment corresponds to some independently recorded outcome chosen as relevant to that judgment. The word independently matters: researchers are not simply asking whether a person or their observer feels confident. They compare a rating with a criterion outside that rating, such as behavior captured on video and later evaluated. That comparison can test a limited prediction: whether a description helps anticipate particular actions under the conditions observed. It cannot automatically test every meaning people attach to a personality word.

In “Comparing the Accuracy of Personality Judgments by the Self and Knowledgeable Others,” participants’ self-judgments and acquaintances’ judgments were compared with videotaped behavioral criteria. The acquaintances had a slight advantage over a single self-judgment, and combining judgments from two acquaintances produced a clearer advantage for criteria tied to overt behavior. This is evidence that knowledgeable observers can sometimes predict visible actions well. It does not establish that observers generally know a person better, or that their interpretation of those actions is more faithful than the person’s own account. The result is bounded by the study’s sample, judgment task, and recorded criteria.

That boundary becomes clearer when the judgment and criterion are set side by side. A claim such as “often interrupts during group discussions” can be compared with recordings in which interruptions are defined and coded. “Feels anxious before speaking” concerns an experience that a camera cannot directly reveal, even if it records outward behavior that may accompany it. “Is an anxious person” reaches further still: it suggests a general tendency across situations and time. A behavior record can support or weaken the action claim without deciding the private experience or the wider trait interpretation. Observable evidence is relevant to the steps it actually records, not a universal translation of inner life.

The naturalistic study “Evaluating the predictive validity of personality trait judgments using a naturalistic behavioral criterion: A preliminary test of the self-other knowledge asymmetry model” illustrates both the value and the scale of this approach. Eighty-seven participants were sampled in everyday life for two days using an Electronically Activated Recorder; researchers coded behavior across situations. Self and informant ratings predicted some corresponding behavior, with modest effects and partial support for the proposed pattern. The study therefore connects judgments to behavior outside a laboratory task, but a two-day sample captures a short window. It cannot establish that a pattern is enduring, that the coded occasions represent every relevant setting, or that one informant would settle a particular disagreement.

A recording is also not free of choices. Researchers decide which moments to sample, how to define a behavior, which context to include, and how coders distinguish one action from another. A person may speak little during recorded meals yet contribute extensively in a one-to-one conversation that was not sampled. Those decisions do not make behavioral criteria useless; they specify what the evidence means. An external record can test a narrow action claim more directly than either person’s confidence alone, while remaining a selected view of behavior rather than an unfiltered record of personality.

For an everyday comparison, first ask whether the disputed sentence names actions that could in principle be noticed and checked. If it does, ask what occasions and settings would count, and what observation would count against the claim as well as for it. If the sentence instead concerns a feeling, intention, or broad identity, an action record may inform the discussion but cannot decide it on its own. A mismatch between the claim and its criterion is not a verdict for either person: it is a reason to narrow what is being tested and keep conclusions within the evidence’s reach.

Sources: Comparing the Accuracy of Personality Judgments by the Self and Knowledgeable Others; Evaluating the predictive validity of personality trait judgments using a naturalistic behavioral criterion: A preliminary test of the self-other knowledge asymmetry model

When might each perspective predict different kinds of evidence?

The Self–Other Knowledge Asymmetry (SOKA) model proposes that access to useful information may differ across traits: a person can know some aspects of their own experience especially well, while people who know them can have a better view of other aspects. This is a conditional account of information, not a rule that assigns one person authority over a whole personality. Its testable implication is that self and observer judgments might predict different kinds of external evidence, depending on the trait and the criterion chosen.

“Who Knows What About a Person? The Self–Other Knowledge Asymmetry (SOKA) Model” examined 165 targets in a round-robin design that included ratings from friends and strangers and behavioral tests. The reported pattern was differentiated: self-ratings were more informative for some neuroticism-related criteria, friends’ ratings for some intellect-related criteria, and self and friend perspectives were similar for extraversion-related criteria. These findings concern prediction of the study’s behavioral criteria in that design. They do not say that a person will always know their own emotional tendencies best, that a friend can reliably judge intellect in daily life, or that a trait label resolves a disagreement between two particular people.

The proposed explanation is about access, not a hidden ranking of people. Internal experiences may supply information relevant to some judgments, while visible conduct and what others have had opportunities to observe may supply information relevant to others. But the study did not turn these possibilities into a test for identifying who is right in an individual conversation. The design’s groups, measures, and behavior tests define the scope of the observed pattern. A result about predicting one criterion is not automatically a result about private experience, every expression of a trait, or long-term consistency across contexts.

Measurement quality makes the domain distinctions more uncertain. The study included ratings with low reliability; for some trait ratings, reported alpha values were very low or negative, and some measures consisted of a single item. Reliability concerns whether a measure distinguishes people consistently enough for its intended use. When ratings contain substantial measurement noise, a difference between self and friend prediction may partly reflect unstable measurement rather than a dependable advantage in access. A single item also gives a narrow basis for representing a broad trait, so a fine-grained conclusion can rest on a fragile indicator. The headline pattern and this qualification belong together: one cannot read the domain results as precise while treating their measurement limits as incidental.

This does not erase the study’s contribution. It offers a useful hypothesis for why perspectives could differ by the kind of information available, and it identifies patterns worth testing with stronger, more dependable measures. The point is that the evidence supports a question about access, not a trait-by-trait decision chart. In a real disagreement, “which of us knows this trait better?” is too broad to answer from these group findings. A more grounded question is what information each person had about the particular claim and whether there are repeated, relevant observations to examine.

Applied cautiously, the model can help keep a conversation curious. If someone disputes a report about a private reaction, ask what that experience was like from the inside and whether the observer could have known it. If the claim concerns a visible pattern, ask whether the observer encountered enough relevant occasions and whether the examples fit the claim’s scope. These are prompts to inspect the information behind each account, not assumptions about which person wins for a particular trait. Neither a trait category nor a measure with weak reliability should substitute for the concrete evidence the individual claim would need.

The strongest use of SOKA here is therefore modest: it cautions against assuming that self and observer reports draw on identical information, while its own reliability problems caution against confident domain-level conclusions. Better measures, clearer behavioral criteria, and replication could strengthen or revise the reported pattern. Until then, it can suggest what kind of information to ask about, but it cannot settle an individual disagreement or convert a conditional study result into a verdict about someone’s personality.

Sources: Who Knows What About a Person? The Self–Other Knowledge Asymmetry (SOKA) Model

Why can reports overlap yet retain perspective-specific information?

Two reports can describe a recognizable shared pattern and still contain differences that belong to each reporter. In “What defines Traits, Reputations, and Identity? Personality item content in multi-rater judgments,” researchers analyzed two multi-rater datasets with a model that separates three sources of rating variation. Shared variance is the portion of ratings that varies together across self and informant reports. Self-unique variance is variation specific to self-ratings after the shared part is accounted for; informant-unique variance is the corresponding informant-specific part. These are statistical partitions in a model, not three kinds of fact and not measures of who knows the person better.

The paper connected those modeled components with the wording and content of personality items. Across its datasets, shared rating variance was more associated with items about observable behavior and affect, while self-unique variance was more associated with cognitive and less observable content. The informant-unique associations did not show a consistent pattern across both datasets. This offers a way to understand how agreement and difference can coexist: reporters may converge more on material that is available to notice, while some self-report variation may concern thought content that is less visible. The pattern is an association between item content and rating variance, not evidence that visibility causes agreement or that observers cannot know a person's thoughts.

The dataset comparison matters here because a component's meaning depends on what was measured and how reports were modeled. The paper found content associations across two datasets, but a statistical relation at that level does not reveal the mental process behind any one response. For instance, an item about a thought may be harder for an informant to rate because it is less observable, yet the result alone cannot tell whether the informant lacked access, interpreted the wording differently, or simply answered differently. Those remain possible explanations, not findings about a particular respondent.

That distinction also prevents a statistical result from being mistaken for a personal verdict. “Unique” means that a component was not shared in the model; it does not certify private knowledge, establish a true reputation, or prove that a unique rating is accurate. A self-specific rating could reflect useful first-person information, a different interpretation of an item, or measurement variation. An informant-specific rating could capture a behavior the person has not considered, or reflect that observer's limited view. The model identifies patterned differences among ratings; it does not adjudicate the reason for one pair's difference.

For illustration only, imagine that both a person and a close observer report that the person often pauses before answering in meetings. That repeated, visible action is an overlap worth preserving. The person also reports mentally rehearsing possible answers, a thought the observer did not report. This invented contrast is not data from the study: the thought may be a meaningful first-person account, but the observer's silence does not verify or disprove it. Nor does the shared pause alone establish what causes it.

A practical comparison can therefore keep two notes rather than force one account to absorb the other. First record the overlapping action in concrete terms, including the setting each person means. Then label the unshared statement as that person's report and ask what evidence could bear on it: a remembered episode, a description of the thought, or another comparable observation. This preserves useful convergence without converting a model's self-unique or informant-unique variance into verified knowledge. It also leaves room for later evidence to strengthen, narrow, or revise either account.

Sources: What defines Traits, Reputations, and Identity? Personality item content in multi-rater judgments

How does context change what a description means?

A context-bound description says where or when an observation applies instead of treating it as a statement about every situation. This distinction matters because people can use a global label such as “quiet” to summarize a broad impression, while an example such as “said little during Tuesday's project meeting” identifies a particular episode. Before treating two accounts as conflicting, check whether they refer to overlapping settings. If they do not, the comparison should remain limited to those settings unless the speakers can connect them with additional evidence.

“Testing a Hierarchical Model of Self-Knowledge” offers a narrow clue about why descriptions may differ in scope. In its priming procedure, participants accessed more abstract self-knowledge when asked for context-independent descriptions and more knowledge of behavioral episodes when asked for context-specific descriptions. The finding concerns retrieval under different prompts. It does not show that a global label is accurate, that an episode reveals a stable trait, or that one kind of description is superior. It does suggest that the question's scope can shape which information comes to mind: a broad prompt may elicit a summary, whereas a specific prompt may bring forward a remembered occasion.

For an invented illustration, one person is quiet in unfamiliar groups but talkative with familiar friends. The two observations can both be accurate because they concern different social settings. Together, they do not prove that the person is quiet everywhere, that familiarity caused the difference, or that either observer has a complete view. A useful record retains each observation at its known scope: “quiet in unfamiliar groups” and “talkative with familiar friends.” Leaving the apparent disagreement intact at first is more precise than stretching either example into an all-context description.

The scope rule also protects against a subtle mismatch in evidence: one speaker may be summarizing many occasions while the other recalls a single episode. Those reports may sound like opposing trait judgments even though one is broad and the other local. Ask what span of time and which kinds of setting each sentence covers, then compare only the portion both accounts actually address. This is a question about the reach of the evidence, not a demand that either person produce an exhaustive history.

The comparison rule is simple: match like settings, or name the difference in settings. If both accounts concern unfamiliar groups, ask about comparable occasions there; if one concerns friends and the other a work meeting, record that difference before interpreting it. Broaden the wording only when comparable examples recur across relevant settings, giving the broader claim a wider basis. Recurrence does not require identical behavior on every occasion; it means the tendency has support beyond a single vivid episode. Context variation does not make a recurring pattern unreal, but one memorable situation alone cannot establish how widely it applies.

This keeps the conclusion proportionate. An account tied to one known situation can still be useful: it identifies where a pattern may appear and invites a more targeted comparison. If later examples from several relevant settings converge, a wider description may fit. If the examples continue to cluster in one context, the narrower description remains the better supported one. When the accounts cover different settings and no shared comparison is available, the honest result is not a forced choice between them but two observations with their scope clearly stated.

Sources: Testing a Hierarchical Model of Self-Knowledge

Which explanations for a mismatch remain plausible?

A mismatch can arise because the two people had access to different information, attended to different parts of the same episode, or are answering different questions. In the 2011 review “Others Sometimes Know Us Better Than We Know Ourselves,” limited access and motivated processes are discussed as possible contributors to gaps in self-knowledge. Here, motivated distortion means only a proposed process by which a person’s interests or feelings might shape an interpretation. A disagreement by itself cannot show that this process occurred, much less diagnose it in either speaker. The same caution applies to an observer: a close relationship may provide repeated exposure, but it does not guarantee complete or neutral attention.

Consider an illustrative contrast. An observer says, “You regularly take over the conversation,” while the person remembers one tense dinner at which they spoke sharply. The remembered dinner is concrete, but it cannot alone confirm or refute a claim about a recurring pattern. Conversely, the observer’s broad wording may compress several unlike episodes into one judgment. To discriminate between those possibilities, ask for separate recent occasions that fit the description and an occasion that does not. Then compare like with like: what happened, in what setting, and what part of the account is an observed action versus an inferred motive? This is a way to examine the claim, not a presumption that either account is distorted.

The observer’s examples still deserve weight. If the same action recurs across several relevant situations, including situations the person had not considered, that pattern may reveal a genuine blind spot or make a broad tendency more plausible. If the examples all come from one emotionally charged event, or if the description moves from an action to an assumption about why it happened, the supported conclusion may need to be narrower. The provisional view should change with the evidence: matched recurring examples strengthen a repeated-behavior claim; examples from previously unseen contexts can reveal either wider consistency or meaningful variation; clearer separation of event and motive can show whether the disagreement concerns conduct or its interpretation.

“Accuracy of Self-Predictions Versus Judgments by Knowledgeable Others” supplies a counterexample to any blanket claim that observers predict better: its two studies found participants’ own predictions of future behavior more accurate overall than predictions by mothers or peers under specified conditions. But prediction of a future event is a different task from describing a person’s general trait. A prediction can be checked against whether a particular behavior later occurs; a trait description summarizes a tendency across occasions and requires a different scope of evidence. The prediction results therefore challenge a universal observer advantage without establishing general self-report superiority. Together, these explanations leave a disciplined conclusion: neither source of a mismatch identifies bias on its own. The examples, their match to the claim, and what each person could observe determine which explanation remains plausible.

A useful test of the tentative conclusion is whether it survives a fair chance to be wrong. Ask what comparable example would count against the description, and whether both people would recognize that example as relevant before hearing what it shows. If only confirming episodes are accepted, the exchange may be protecting an interpretation rather than testing it; that remains a possibility to examine, not a label to apply. If disconfirming examples are considered and the recurring action still appears across settings, retaining the tendency may be warranted even when its meaning remains disputed. The evidence can support a behavior pattern while leaving motive unresolved.

Sources: Others Sometimes Know Us Better Than We Know Ourselves; Accuracy of Self-Predictions Versus Judgments by Knowledgeable Others

What should you ask next, and what conclusion can you keep?

Ask for one recent occasion that fits the description and one that might not; then offer a comparable example from your own view. Keep the broad tendency when repeated, matched examples support it across relevant situations. Narrow it to an action or setting when support is local, and leave it unresolved when the accounts still concern different evidence. You can also name the remaining uncertainty plainly: perhaps the action is clear but its cause is not, or perhaps neither person has seen enough comparable occasions. That statement preserves what you know without forcing agreement about what it means. If seeing several tendencies together would help your private reflection, the Context Profile at /assessment is one prompt, not an arbiter. Explore the learning library at /topics for more profiles; neither route decides the disagreement or predicts career fit. Agreement is optional; precision is useful. The next conversation can end with that question still open.

Questions readers ask