> I wonder if there is any reference for combining rules to the case in which > the parameters of interest are the confidence intervals themselves. > In that case I have upper and lower limits for each imputed data set, but I > want to get a combined interval estimate. Any thoughts? >
can't you use that information to deduce the point estimate and standard error from each imputation, then combine in the usual way, and turn the combined inference back into a confidence interval? This of course is assuming that everything is based on asymptotics, otherwise everything falls apart.
