Paul R Swank wrote: > > If in fact the data points are bad, that is indefensible, and replaced by > missing values then I would suggest using a mixed models approachj with > Maximum Likelihood estimation. This will not result in listwise deletion and > is generally preferable to GLM because of its greater flexibility. SAS > MIXED, MLwin, HLM5 and SPSS 11 all have the ability to do the mixed model > approach.
That's probably what I'd do, though with small samples multiple imputation and standard analyses might be better. AFAIK the results from MI and mixed/multilevel models should be pretty similar as the key thing is the assumptions about the nature of missingness (MAR MCAR etc.). I think the mixed/multilevel approach ought to be roughly equivalent to MI assuming MAR. Thom . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
