|
Abstract:
|
When the usual assumptions fail to hold for the 2-by-2 crossover design but the dependent variable has repeated measures in each period, then those repeated measurements can aid in modeling carryover effects. Two main approaches can be used: explicitly model the carryover effect (parametric), or treat the data with longitudinal methods (non-parametric). An example of the first approach is to model the underlying carryover term with a decay factor, which in a given period contributes a multiplicative term to the treatment mean at a given week. An example of the second approach uses longitudinal methods which treat the repeated effect as a factor. The basic difference between the two approaches is the treatment of the time effect. In the second model treatment is a time varying factor rather than the usual sequence and treatment factors of a crossover analysis. We examine a study with the repeated design in which the assumptions of the 2-by-2 crossover have not held. Using this data we illustrate the features of the two different approaches and consider their impact on interpretation of results. We also use Monte-Carlo simulation studies to compare the two methods.
|
