Abstract:
|
Partially clustered designs, where clustering occurs in some conditions and not others, are common in psychology, particularly in area of prevention and intervention trials. Baldwin, Stice, & Rohde (2008) discussed the analysis of these designs and showed that the multilevel models adapted for partially clustered data are relatively unbiased and efficient and consistently maintain the nominal Type I error rate when using appropriate degrees of freedom the proposed design. Moerbeek and Wong (2008) produced power formulas for these partially clustered designs, which are referred to as 2-1 models, where the 2-1 model corresponds to one treatment arm has two levels (patients within group) and the other treatment is one level (patients treated alone). In this presentation, we extend the Moerbeek and Wong (2008), power formula to consist of an additional level, repeated observations per patient, corresponding to longitudinal data/repeated measures at the patient level. We refer to this structure as a 3-2 design.
|