Abstract
PURPOSE: To identify characteristics and behaviors among teenagers that predict sexual initiation or sexual activity and to evaluate alternative methods for predicting sexual behavior among teenagers. METHODS: We used longitudinal data from an evaluation of the Making Proud Choices! teen pregnancy prevention program. The evaluation, funded by the Office of Population Affairs, assessed academic years 2016-2017 through 2018-2019 and examined 2,138 Grade 9 and 10 students of all genders. We used ordinary least squares (OLS), least absolute shrinkage and selection operator (lasso), and stratified OLS to identify behaviors and characteristics at baseline that predict sexual initiation, recent sex, and sex without a condom. RESULTS: OLS and lasso regression show that pre-sexual behaviors and substance use are the most powerful predictors of sexual initiation among teens. Lasso additionally identified higher-order interactions between predictors of sexual activity, including variation in predictors' influence across population subgroups. Stratified OLS regression predicted behavior most accurately for sexual initiation, recent sex, and sex without a condom. However, stratified OLS also reduces the sample size, and therefore precision, for each regression. CONCLUSIONS: Current behavior, not knowledge or beliefs about sex, best predicts future behavior. It can be difficult to evaluate the impact of programs designed to delay sexual initiation among younger adolescents because sexual behaviors often occur at older ages. Our results suggest that pre-sexual behaviors are strongly predictive of subsequent sexual initiation and having sex without a condom, so these could be used as outcome measures for identifying high-risk students or evaluating interventions among younger adolescents. Additionally, our results show that lasso can be a useful technique to identify subgroup differences in the relationship between predictors and future sexual activity and for prioritizing variables to collect in a survey.