Research Information System
ADVANCES AND APPLICATIONS IN STATISTICS, vol.49, no.3, pp.211-229, 2016 (ESCI)
No closed form expression or simulation study exists to calculate sample size/statistical power in the context of linear mixed effects models (LMMs) for longitudinal data with first-order moving average MA(1) or first-order autoregressive moving average ARMA(1,1) autocorrelated errors and random effects having non-zero covariance in the LMMs that might be occurred in observational clinical longitudinal studies. In this paper, we actually examine how the magnitude of MA(1) autocorrelation coefficient, the signs of non-zero covariance between random effects and slope in repeated responses over time interact with each other in the statistical power calculations to find the most appropriate sample size and the number of repeated measures for the two frequently used LMMs in longitudinal data analysis. Additionally, we compared the results with those for the LMMs with ARMA(1,1) errors. Especially for small sample size and the number of repeated measures, the factors in LMMs such as the sign of slope in responses over time, covariance between random effects and the autocorrelation coefficient play a major role in statistical power calculations based on the type of LMM.