LONGITUDINAL DATA ANALYSIS BASED ON RANKS

Abstract

The data measured repeatedly for same individual over time is called longitudinal data. In longitudinal data analysis, if response is continuous and normally distributed, repeated measures ANOVA is often used. But assumption required of normality is not always satisfied and validity of parametric approarch as repeated measures ANOVA is suspected. We propose rank experimental distribution method that is distribution-free to relax restriction of parametric apporoach. In case study, rank experimental distribution method and repeated measures ANOVA were different in the outcome. It is considered to be due to a difference in the underlying distribution. Then, we conducted the simulation that is supposed underlying distribution is normal or skewed to investigate whether for group effect, time effect and interaction the power of two method is different. As a result of the simulation, for group effect, time effect and interaction the power of both methods is almost the same in normally distributed data. And for group effect, time effect and interaction the power of rank experimental distribution method is higher than repeated measures ANOVA. So we have showed the rank experimental distribution method is useful for longitudinal data analysis.