dealing with omitted variables in logit

[Gebretsadik Shibre]

Dear experts,

I would like to kindly ask your help in the interpretation of a regression output. A refrence categor of an intereraction is empty and one of the other category is omitted, both by the Stata automatically in my logit regression

svy: logit outcomvar  i.union i.region i.union#i.region

 union#region |
                                 0#somali  |          0  (empty)
                                   1#afar  |  -.0154212   1.076807    -0.01   0.989    -2.127109    2.096266
                                 1#amhara  |   .1662817   .7509852     0.22   0.825    -1.306449    1.639012
                                 1#oromia  |  -.6080664   .7652739    -0.79   0.427    -2.108818    .8926851
                                 1#somali  |          0  (omitted)
                               1#benisaha  |  -1.198869   .9228854    -1.30   0.194    -3.008707    .6109686
                                  1#snnpr  |  -.4225948   .7383459    -0.57   0.567    -1.870539    1.025349
                                1#gambela  |  -.7886199   .8148943    -0.97   0.333     -2.38668    .8094403
                                  1#harar  |   .9801469   .8771208     1.12   0.264    -.7399434    2.700237
                                     1#AA  |   1.153589   .7979991     1.45   0.148    -.4113382    2.718517
                                     1#DD  |   .5922947   .8978733     0.66   0.510    -1.168493    2.353082

How do I interpret the coefficients for the interaction in the presence of these two problems: the first category is empty, and the fifth category is omitted?
Do I interpret the coefficients with respect to the base of the omitted group? what about the empty one? It seems to me that this groups is empty because of lack of observations in that particular intersection, but how do ignoring this empty group affect the interpretation of the coefficients.

Thank you in advance

[Clyde Schechter]

Do I interpret the coefficients with respect to the base of the omitted group?

Yes.

what about the empty one?

Your data set has no information about the empty combinations and you draw no conclusions whatsoever about those.

The interpretation of other coefficients is not directly affected by this. However, to the extent that the lack of observations in some intersections is the result of biased sampling, the entire analysis is affected. If, however, the lack of observations in those intersections is due to structural considerations (i.e. those combinations are, in principle, impossible, at least in the universe of your study question) then there is no effect on anything.