The Rutgers Alcohol Problem Index (RAPI) is a scale designed to identify alcohol-related problems experienced by adolescents and young adults. RAPI scores tend to be skewed due in part to a substantial proportion of zero scores.
In this study, we compared different statistical models for analyzing the RAPI as an outcome measure using data collected as part of an alcohol intervention program for college freshmen. Predictors included intervention group indicators. Model fit, predictive ability and practicality of implementation were compared among competing models and recommendations are made.
Of 558 students who completed the baseline survey (68% response rate), 76% completed a 3-month follow-up survey. Mean RAPI scores were 4.7 (SD = 6.0) at baseline and 4.6 (SD = 5.5) at 3 months. Significance of intervention group differences at 3 months adjusting for baseline RAPI were heavily dependent upon the type of statistical model employed. A zero-inflated (ZI) negative binomial model provided the best model fit.
This study demonstrated that inferences and conclusions about intervention efficacy were method-dependent in analysis of RAPI scores. There was empirical evidence of improved inference when methods accounting for zero-inflation and skewness were employed.
