SAS’s PROC MIXED can be problematic when analyzing data from randomized longitudinal two-group designs when observations are missing over time. Overall (1996, 1999) and colleagues found a number of procedures that are effective in controlling the number of false positives (Type I errors) and are yet sensitive (powerful) to detect treatment effects. Two favorable methods incorporate time in study and baseline scores to model the missing data mechanism; one method was a single-stage PROC MIXED ANCOVA solution and the other was a two-stage endpoint analysis using the change scores as dependent scores. Because the twostage approach can lack sensitivity to detect effects for certain missing data mechanisms, in this article we examined variations of the single-stage approach under conditions not considered by Overall et al., in order to assess the generality of the procedure’s positive characteristics. The results indicate when and when not it is beneficial to include a baseline score as a covariate in the model. As well, we provide clarification regarding the merits of adopting an endpoint analysis as compared to the single-stage PROC MIXED procedure.
Algina, James and Keselman, H. J.
"Assessing Treatment Effects in Randomized Longitudinal Two-Group Designs with Missing Observations,"
Journal of Modern Applied Statistical Methods:
2, Article 3.
Available at: http://digitalcommons.wayne.edu/jmasm/vol3/iss2/3