Predictive Validity of Thin Slices of Verbal and Nonverbal Behaviors: Comparison of Slice Lengths and Rating Methodologies

Thin slices, or excerpts of behavior, are commonly used by researchers to represent behaviors in their full stimulus. The present study asked how slices of different lengths and locations, as well as different measurement methodologies, influence correlations between the measured behavior and different variables (predictive validity). We collected self-rated, perceiver-rated, and objectively measured data on 60 participants who participated in a 5-min interaction with a confederate on video. These videos were split into five 1-min slices and rated for verbal and nonverbal behaviors via global impressions, using the same rater for all five slices and also using a different rater for each slice. For single slices, results indicated no clear pattern for optimal slice locations. In general, single slices had weaker predictive validity than the total. Slices of 2 or 3 min were, in general, equal to 5-min total in predictive validity. The magnitude of correlations was similar when same versus different coders were used, and the predictive validity correlations of the two methods covaried strongly across behavior-outcome variable combinations.