Working on handling of missing values in round robin designs (21.09.2009)

We are currently working on adjusting the formulas for round robin analyses in TripleR for the presence of missing values. First simulation studies show very promising results; a TripleR version capable of handling missing values will be released next month on R-Forge.

First results indicate that estimations of variance components are rather robust for group sizes n > 20 and a ration of missing values < 10%. However, in big groups of n > 40 even up to 50 % of missing values do hardly influence the estimations. Here is a screenshot of a simulation showing the effect of different group sizes (n=5, 10, 20, 35, 54) and different ratios of missing values onto the estimation of the actor variance component:

Author: FelixS