Fig 1.
Schematic overview of the current study and example stimuli.
Ratings for observable cues and criterion ratings (HIV riskiness, as well as other impressions) are collected from independent groups of raters for a large set of target photographs. Averaged cue and criterion ratings are then combined and correlations are assessed between each individual cue vector and the HIV risk criterion judgments. This strategy identifies cues that may be utilized to infer HIV risk and thus comprise a ‘Brunswikian Semi-Lens’.
Table 1.
Descriptives for cue ratings, correlations between cues and perceived HIV risk (‘cue utilization coefficients’), and regression model coefficients.
Table 2.
Descriptives for ratings of general impressions.
Fig 2.
Relationship between actual impressions of HIV risk and model-based predictions.
The model was trained using LASSO-Regression and cross-validated using a 10-fold strategy. We then compare the model-based predictions against actual perceptions of HIV risk obtained from different raters, finding that the model learned to successfully predict HIV risk based on cues (r = 0.75). See text for details.