KK recently wrote that his attempt to use mvsumm on a 600 000 obs panel
with 20 000+ unique ids failed due to 'too many values'. The authors
have recently been illuminated on this issue by StataCorp staff, who
warned that using -levels- or its predecessor -levels7- was going to
impose a constraint: the number of groups supported by -tabulate- (3
000 in Intercooled, 12 000 in SE). At some point the authors will
probably take this advice and rewrite -mvsumm- (and its relative
-mvcorr-). For now, as a workaround, it should be possible to massage
chunks of the dataset that contain fewer than 3 000 (12 000) panel ids.
