65studies reviewed
48measured real personality
125cue–trait pairs covered
21with a clear link
Where the numbers come from
Breil, Osterholz, Nestler and Back (2021) collected 64 articles reporting 65 independent studies of nonverbal cues and personality. In 48 of them, researchers measured how a cue relates to the target person’s actual personality, known as cue validity. This site shows only those results, not what observers believed.
The studies came from three places: work the authors already knew, a database search limited to peer-reviewed journals, and the first pages of Google Scholar. Targets were real people aged 16 or older whose behaviour was not experimentally manipulated. Personality came from self-reports and/or reports by people who knew them. Cues were coded or rated from recordings, photographs and live interactions.
The tables report the 39 cues that appeared most often across studies. The authors grouped related measures into these broad categories and assigned related traits to the nearest Big Five dimension (for example, dominance and shyness to extraversion; warmth and arrogance to agreeableness). They note that other groupings are possible and share the raw data so that readers can try them.
Correlations for the same cue and trait were averaged within each sample, then pooled across samples with a random-effects model (inverse-variance weights, DerSimonian–Laird estimate of between-study variance). Where a study reported only partial correlations, such as those controlling for age and sex, the authors used them.
What counts as a link here
The explorer covers the 25 dynamic cue categories in Tables 3–7 (facial expression, body language and voice) and the Big Five, so 125 cue–trait pairs. A pair counts as a link when at least two samples were pooled and the reported 95% interval lies wholly above or below zero. Intervals that touch 0.00 do not count. That leaves 21 links. Another pairs were tested in two or more samples without a clear link, and rest on one study or none. Appearance cues and intelligence are outside the explorer’s scope.
This screen is an editorial rule, not a test the authors ran. Breil and colleagues used a different yardstick: they highlighted cues with a correlation of at least .10 in two or more studies, whether or not the interval excluded zero. The rule here is also not adjusted for the number of comparisons. With pairs tested at the 95% level, about would pass by chance even if no cue were related to any trait, so a few of the 21 links could be chance findings.
How to read the numbers
- r runs from −1 to +1. A positive value means more of the cue goes with higher trait scores; a negative value means less of it does. The links here range from .10 to .24 in size. By the usual benchmarks (.10 small, .30 medium) all of them are small; Breil and colleagues treat .10 as the threshold for at least a small effect.
- The line around each dot is the 95% confidence interval of the pooled estimate. k is the number of independent samples and n the number of people in them.
- The “100 people” panels turn r into a picture. They assume that cue and trait are normally distributed and linearly related with the pooled correlation. Then, among people above the median on the cue, the share above the median on the trait is 50% + arcsin(r)/π. The range comes from the interval bounds. It illustrates the average relationship and is not a rule for judging anyone.
Limits worth keeping in mind
- These are correlations from observational studies. They do not show that a trait causes a behaviour. A cue can also reflect the situation, or a third factor such as age or sex, which only some of the studies adjusted for.
- Evidence is thin for many pairs: of the 21 links rest on only two or three samples. Most studies measured fewer than ten cues, and many cue–trait combinations have little or no data.
- The categories are broad. “Forward lean / proximity”, for example, combines leaning in and standing close, and studies defined and measured cues in different ways.
- The authors call their review a first, preliminary overview and say it does not aim to be complete. Some studies reported only significant correlations or only adjusted ones, which can bias pooled estimates.
- Context matters but is poorly understood. Settings, observation time, recording channels and target groups varied, and the review could say little about how these change any single cue. The authors ask for replications in more diverse samples and settings. Recorded tasks may also narrow the range of behaviour seen in everyday life.
- A link can differ between groups of people. One study the authors cite found several cues that signalled extraversion in women but not in men. The pooled figures average over such differences.
- Each cue is analysed on its own, although cues co-occur. Hand gestures, for example, happen mostly while people are speaking, so the two are hard to separate.
About the illustrations
The figure and its movements were drawn for this site to show what each cue category looks like. The dial shows the direction of a link, not its size; the size is in the chart and the “100 people” panel. Voice cues are drawn as speech-bubble diagrams. Cues such as a “pleasant” voice were rated by people in the original studies, and such judgements vary between raters.
How Hall et al. (2025) used this meta-analysis
Hall, Murphy, Schlegel and Back asked 245 U.S. adults on Amazon Mechanical Turk how likely 20 everyday nonverbal behaviours are for someone high in one Big Five trait, then compared these beliefs with Breil’s tables. Matching 10 to 14 cues per trait, beliefs tracked cue validity well for four traits (profile correlations .50 to .76) but not for openness (.27). With so few cues per trait these correlations are imprecise, and the one for conscientiousness (.50, ten cues) was not statistically significant.
The authors list their own limits: about 50 raters per trait; profile correlations based on no more than 20 cues; cue wordings that did not always match Breil’s categories; questions about people in general, without age, gender or culture, compared with specific targets recorded in particular settings, where some cues may vary little; a possible role of how socially desirable a cue seems; and 20 cues that cannot cover every behaviour, with gaps in the overlap that mean the picture could change with new research. Raters also disagreed with each other quite a lot (intraclass correlations .25 to .53).
One example in their discussion is slouching: on average, extraverts tend not to slouch. In Breil’s table the matching cue, upright and open posture, has r = +.15 for extraversion, but its interval (−.01 to +.31) touches zero, so this explorer lists it under “no clear link”. The explorer uses Breil’s estimates directly and does not show the belief ratings.
Sources
- Breil, S. M., Osterholz, S., Nestler, S. & Back, M. D. (2021). Contributions of nonverbal cues to the accurate judgment of personality traits. The Oxford Handbook of Accurate Personality Judgment. Author manuscript · Data and code
- Hall, J. A., Murphy, N. A., Schlegel, K. & Back, M. D. (2025). Explicit beliefs about nonverbal behavior and the Big Five traits. Journal of Personality, 94, 678–689 (online 2025, issue 2026).
- Fonts: Young Serif and Public Sans, both under the SIL Open Font License.