Identifying Situational, Individual, and Demographic Determinants of Social Anxiety Using a Naive Bayesian Classification Algorithm

Social situations may play a role in the conceptualization and treatment of social anxiety disorder. However, situational variance is rarely examined and has not yet been directly compared to dispositional or demographic variance. We used situation-sampling methodology and naive Bayesian classification to determine the predictive power of situational features compared to individual and demographic variables. In Study 1, a sub-sample of 309 undergraduate students responded to random subsets of 680 situations to train a machine-learning algorithm that predicted whether a participant would endorse situational social anxiety. The predictions were then tested on the remaining participants in order to generate accuracy estimates. Situational variance alone improved accuracy by 21.39% (from 50%) whereas the addition of individual variance (+3.8%) and demographic information (-1.11%), did not add to the predictive accuracy. Situational features were rank-ordered by their likelihood ratios, and were grammatically edited to reflect the top, middle, and bottom features. In Study 2, a community sample of 211 people responded to these new situations and endorsement rate was compared to predictive accuracy estimations. We found consistent accuracy for high and middle social anxiety-provoking situations indicating that situational features may be particularly relevant when treating clients with higher levels of anxiety.