Creating Sexual Risk Levels: Application of a Cluster Analytic Technique

Many instruments have been created to measure sexual behavior, but few measure the entire spectrum of behaviors; including oral, anal and digital sex. Without consideration of all the variables involved in sexual behavior future research and interventions may be destined for failure. To maximize efficiency and efficacy, greater attention should focus on the application of contextual methods in data analysis to supplement contextual ideas (theories, models, and frameworks).

Using several different clustering methods (Ward's, K-means, EM), this study identified the typology of sexual behavior among college students at a large southern University. Respondents of an anonymous self-report questionnaire were grouped according to the following behaviors: participation in sexual behaviors (vaginal, anal, oral, and digital), number of sex partners, frequency of condom or barrier method during specific sexual activity, contraceptive use, alcohol use within the past 30 days and heavy episodic drinking within the past 2 weeks. The Bayesian Information Criterion (BIC) was applied to select the clearest clustering method, the Expectation Maximization (EM) algorithm (n= 321). A total of 3 clusters (low, medium and high risk) were confirmed and validated using the following discriminating variables: race, age, gender, relationship status, STI Diagnosis (self or partner), unplanned pregnancy (self or partner), coercion or abuse, unprotected sex due to drinking, and sex without giving or receiving consent due to drinking (n=302).

Significant predictors of cluster membership based on risk behaviors (low, medium and high) will be covered and implications for college health screenings and interventions will be discussed.