Powerful and Cost-Efficient Designs for Longitudinal Intervention Studies With Two Treatment Groups

Abstract

Three issues need to be decided in the design stage of a longitudinal intervention study: the number of persons, the number of repeated measurements per person, and the duration of the study. The degree to which polynomial effects vary across persons and the drop-out pattern also influence the statistical power to detect intervention effects. This article presents a framework that allows researchers to calculate the power of a proposed design and compare alternative designs on the basis of their costs and sample sizes. A multilevel regression model with polynomial effects varying across persons is used to relate response to time. The persons’ length of stay in the study is modeled using a survival function.