Dear Statalisters,

recently there was an upate for -margeff-, so I tried to reestimate some ordered probit models for a categorical variable with four categories with the updated -margeff-. However, now I am confused:

I tried both, Stata SE 8.2. and Stata SE 9.2. with mfx and margeff, e.g.

oprobit Y X1 X2
mfx compute, predict(outcome(1))

this works fine with Stata 9.2. but does not work with Stata 8.2. - there I get an error message (198): option version() not allowed -- anyway I prefer the newer Stata version and that might not be a severe problem.

But margeff - before the update - seemed to me faster and in many respects more convenient than mfx, e.g. I was interested in average marginal effects and the options for dummy-variables. When I try to use the updated version with Stata SE 9.2. I get the following:

oprobit Y X1 X2
margeff

ERROR (504): estimates post: matrix has missing values

This does not occur with Stata SE 8.2. - however with Stata 8.2. it is so(!) slow compared with mfx and until now I was not patient enough to wait and see what the outcome looks like, because I doubt that the command works correctly - before the update it was much faster on my computer than mfx. I tried also other specifications, but with extended commands it produces even more strange erros, e.g.:

oprobit Y X1 X2
margeff, percent outcome(1 2 3 4)

ERROR (198): Only one number can be specified in the outcome() option

Again, when I first used the command this did not happen and I could produce nice outputs with -estout- and other commands.

So I am really lost, maybe I forgot something or you have an idea what I have not tried yet, any help would be appreciated!
Thank you,

Johannes Geyer