In response to Frank Herrell's query about using bootstrap validation when
data are imputed:  the fundamental problem here seems to be that the
imputations based on the entire dataset incorporate information about all
of the data and therefore don't include the variability under the
bootstrap procedure.  I don't see any obvious bound on the amount of
information added by doing the imputations using the full data rather than
the bootstrap replication.  





Reply via email to