Model averaging for split-plot designs

I recently published a paper (which can be found here) with Peter Dillingham, Chuen Yen Hong and others on the use of model averaging when analysing data from a split-plot design. In some split-plot studies the whole-plot variance component may be small relative to that for the sub-plots. When this is the case, we propose using model averaging to combine the results from the usual split-plot analysis with those from an analysis in which the whole-plot variance is set to zero. This leads to confidence intervals for whole-plot treatment means that are narrower than those from the usual split-plot analysis, while still having coverage close to the nominal level.