I find the problem in an up-to-date Stata 9.1 using version 6.3.5 of
tabodds. It occurs when tabodds calls cci and passes it some missing values
when cci expects 4 non-missing integer values. It looks like a bug in
tabodds, at least in that tabodds should check what it is passing to cci to
make sure it is legal. I haven't unraveled all of tabodds but there may be
a problem in properly tracking which observations should be flagged to
include in making some calculations, because the missing values come from
using the sum function to get the mean of a specific obs or group of obs.
