Do human faces, bodies, and voices “tell the same story” about attractiveness, health, and dominance
– by Tobias L. Kordsmeyer
When we form impressions of others, we rarely rely on just one cue. Faces provide one stream of information, bodies provide another, and voices add yet another layer. In everyday life, these cues are usually perceived together – so it is tempting to assume they are redundant. If someone has an attractive face, we might expect an attractive body and an attractive voice as well.
But should we indeed expect that kind of cross-modal consistency, i.e. noteworthy correlations of perceptions of attractiveness and other socially relevant traits between faces, bodies, and voices? Two competing ideas about such correlated perceptions have been debated for decades.
The one ornament hypothesis (also called backup signals hypothesis or redundancy hypothesis) proposes that different body parts and modalities reflect shared underlying qualities such as immunocompetence, genetic quality, or developmental stability. If that were true, then perceptions of attractiveness (and related traits like health or dominance) should correlate across faces, bodies, and voices.
By contrast, the multiple messages hypothesis argues that what different modalities signal is more independent: faces might reveal different information (e.g. based on how average faces are, which in turn is related to perceptions of attractiveness and health), than bodies (e.g. health based on shape and fat distribution) and voices (e.g. dominance and masculinity based on voice pitch). If so, correlations among perceptions across modalities should be weak or absent – because each channel carries partly distinct information.
Our study puts these ideas to a rigorous test by asking a seemingly simple question: When people judge faces, bodies, and voices separately, do those ratings line up? Put differently, are perceptions of attractiveness, dominance, and health correlated positively across faces, bodies, and voices? Additionally, steroid hormones (testosterone, estradiol), which influence the development of facial, bodily, and vocal characteristics, were investigated as partial explanations of purportedly correlated perceptions.
To investigate these questions, we assembled a large stimulus set from 320 men and women, including neutral face photographs, 3D body scans converted into short rotating videos, and brief voice recordings. Each of these three modalities was then judged by separate groups of raters regarding perceived attractiveness, perceived health, and perceived physical dominance (the latter conceptualised as the likelihood of winning a physical fight). That separation matters – if the same raters had judged faces, bodies, and voices, halo effects could have inflated correlations (“I liked his face, so I feel like his voice must be attractive, too”). Using different groups of raters for different modalities is a tougher test of true cross-modal concordances.
A major methodological concern in older work on judgments of bodies was that body stimuli often included skin colour and texture, which themselves convey health-related information and could invoke stereotypes, which could create shared variance with facial appearance. In our study, we deliberately used grey-coloured body scans without heads, removing skin tone cues, to shift the focus to body shape. That means any face–body correlations that remain are harder to explain away as “it’s just skin quality.”
What did we find? Generally, we found consistent and medium-sized correlations between faces and bodies for all three traits: attractiveness, health, and physical dominance. On average, people who were considered attractive in terms of their facial features were also considered attractive in terms of their body shape. Correlations for perceived health and perceived physical dominance were similar, but only slightly smaller. So, at least for faces and bodies, the data look consistent, indicating a partly redundant signal and hence supporting the one ornament hypothesis.
For voices, however, the story shifts. Out of the six correlations of voices with faces and bodies for the three traits, only three were significant, and effect sizes rather small. Thus, regarding voices, the multiple messages hypothesis seems to be better supported, and voices may add information that is not strongly predictable from faces or body shape.
When we split the target sample by sex, the face–voice and body–voice correlations were somewhat stronger for men than for women (especially for perceived health and physical dominance), whereas the body-voice correlations were more comparable across the sexes. Thus, whether the one ornament hypothesis or multiple messages hypothesis is more valid may not only depend on the modality, but may also differ by target sex.
A common idea in evolutionary psychology is that sex steroid hormones contribute to the development of sexually dimorphic traits (including facial masculinity/femininity, body composition including fat distribution, and vocal characteristics), which then influence perceptions of attractiveness, health, and dominance. If that’s right, one might expect that part of the cross–modal correlations exists because the modalities are simultaneously influenced by the same endocrine factors.
We tested this using salivary hormone measures: baseline testosterone and testosterone reactivity (after having engaged in a dyadic competition in the laboratory) in men, and baseline testosterone, baseline estradiol, the baseline testosterone/estradiol ratio, as well as hair testosterone (a measure of long-term aggregated baseline levels) in women.
The insight from these additional analyses was rather simple: We found no evidence that these hormone measures explained the cross-modal correlations. The study did find a few small direct associations between men’s baseline testosterone and their vocal attractiveness and physical dominance, but not in a way that explained the broader cross-modal pattern.
One implication is that if a shared biological mechanism links facial, bodily, and vocal characteristics, it may not be well captured by current adult hormone levels. Instead, prenatal or pubertal hormonal measures with more organisational effects during development, or additional genetic factors and environmental perturbations during development may play a stronger role as explanatory mechanisms.
What is the take-away message? This study offers a nuanced update to a debate that is sometimes framed too starkly as “redundant” versus “independent” signals. For faces and bodies, we demonstrate moderate associations across perceived attractiveness, health, and physical dominance (somewhat stronger in men than women). That supports the idea that these visual channels share underlying cues – despite the body stimuli in our study lacking skin information – in support of the one ornament hypothesis. However, for voices, cross-modal correlations were rather weak, suggesting voices may communicate partly distinct information from faces and bodies – more consistent with the multiple messages hypothesis. Concerning potential explanatory mechanisms for such cross-modally correlated perceptions, at least current adult testosterone and estradiol levels do not seem to play a role here, calling for the investigation of alternative causes.
