Analyzing Proportion Scores as Outcomes for Prevention Trials: a Statistical Primer

• DOI: 10.1007/s11121-016-0643-6
• Publication Date: 2016-03-10T06:56:25Z
• Authors: Kehui Chen; Yu Cheng; Olga Berkout; Oliver Lindhiem
• Keywords: Clinical Trials as Topic; Outcome Assessment, Health Care; Linear Models; Humans; Preventive Medicine; 0101 mathematics; 01 natural sciences

In prevention trials, outcomes of interest frequently include data that are best quantified as proportion scores. In some cases, however, proportion scores may violate the statistical assumptions underlying common analytic methods. In this paper, we provide guidelines for analyzing frequency and proportion data as primary outcomes. We describe standard methods including generalized linear regression models to compare mean proportion scores and examine tools for testing normality and other assumptions for each model. Recommendations are made for instances when the assumptions are not met, including transformations for proportion scores that are non-normal. We also discuss more sophisticated analytical tools to model change in proportion scores over time. The guidelines provide ready-to-use analytical strategies for frequency and proportion data that are commonly encountered in prevention science.