st: xtlogit, marginal effects and odds ratios

[Prashant <presearchwork@gmail.com>]

Hi,

I am calculating odds ratios and marginal effects with associated
robust SEs using GEE models. There is some missing data in my dataset
(less than 10% of the outcome data is missing), so I also used
multiple imputation and then the mim prefix when estimating the GEE
models.

To estimate marginal effects, I had (hopefully correctly) modified a
short program (introduced in
http://www.stata.com/statalist/archive/2010-03/msg01021.html)

The basics of the program are below:

/******* begin -emargins- *************/
.   program emargins3c, eclass properties(mi)
  1.         version 11
  2.         args depvar track
  3.         xtlogit `depvar' i.treatment covariates if `track' == 1,
pa i(county_number) robust		/*track is a binary variable identifying
track*/
  4.         margins, dydx(*) post
  5.   end

. /******** end -emargins- *************/
.
. mi estimate: emargins3c outcome1 science

To estimate the odds ratios, I used the following command:
. mim: xtlogit outcome1 treatment covariates if science == 1, or pa
i(county_number) robust

(the covariates are the same in both marginal effects and odds ratio
estimation scenarios)

I have two questions:

1) I am able to estimate odds ratios without problem. However,
occasionally for some outcomes, the marginal effects (dydx) cannot be
calculated. I get the error message:

"estimates diverging (missing predictions) an error occurred when mi
estimate executed emargins3c on m=1".

Is there anything I can do to obtain the marginal effects estimates in
this situation?

2) The p-values for the marginal effects and odds ratios estimates are
different--in particular the p-values for the odds ratios are slighly
lower. Is this normal?

Thank you so much for any assistance in advance,
Prashant
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/