This study used data from a national survey of youth mentoring programs (N = 1,451) to examine training and other potential predictors of premature mentoring match closures (Garringer et al. 2017).
Mentoring programs are a popular approach to preventing problem behavior and promoting positive youth development; however, mentoring relationships that end prematurely may have negative consequences for youth. Previous research has investigated match-level indicators of premature match closure, highlighting possible individual mentor- or mentee-level characteristics that might influence the match staying together; however, less work has investigated the importance of program-level variables in match retention. Mentor training and support may be one key modifiable program-level feature that could curtail the risk of premature match closure. The current study used a Bayesian Additive Regression Trees (BART) model to predict program-reported premature match closure rates from a set of four training-related variables and 26 other covariates (e.g., program size, budget, demographic composition). Findings indicate that the set of predictors explained about one-fifth of the variation in reported rate of premature match closure (cumulative pseudo R2 = .21), and the strongest, and only statistically significant, predictor of premature match closure was the frequency of ongoing training and support contacts per month. Overall, findings indicate that there is substantial noise in predicting program-reported premature match closure, but program-reported provision of ongoing training and support seems to emerge as a relatively stable signal in the noise. (publisher abstract modified)