Some interaction combinations are not showed, why?

[Elise Sobrie]

Dear all

When I want to see the results (of my probit model) of an interactions term in my model, stata doesn't show all the possible combinations. Does anybody know why?
My interaction term: i.WLB##i.Industry where WLB (=women-led-business) is a dummy (equal to 1 when the firm is led by women) and Industry is a categorical variable ranging from 1 to 6 (6 types of industry sectors.

probit Rejected i.WLB i.COUNTRY i.Industry i.Sole i.Audit i.Quality c.Size c.FirmAge c.Experience i.WLB#i.Industry

note: 2.Industry != 0 predicts failure perfectly
      2.Industry dropped and 16 obs not used

note: 5.Industry != 0 predicts failure perfectly
      5.Industry dropped and 18 obs not used

note: 1.WLB#4.Industry != 0 predicts failure perfectly
      1.WLB#4.Industry dropped and 3 obs not used

note: 1.WLB#2.Industry omitted because of collinearity
note: 1.WLB#5.Industry identifies no observations in the sample
note: 1.WLB#6.Industry identifies no observations in the sample
Iteration 0:   log likelihood =  -117.0378 
Iteration 1:   log likelihood = -108.10336 
Iteration 2:   log likelihood = -107.79465 
Iteration 3:   log likelihood = -107.79306 
Iteration 4:   log likelihood = -107.79306 

Probit regression                               Number of obs     =        272
                                                LR chi2(13)       =      18.49
                                                Prob > chi2       =     0.1398
Log likelihood = -107.79306                     Pseudo R2         =     0.0790

------------------------------------------------------------------------------
    Rejected |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     COUNTRY |
          2  |   1.076469   .3957937     2.72   0.007     .3007279    1.852211
          3  |   .8967785   .4008626     2.24   0.025     .1111022    1.682455
             |
      1.Sole |    .305684   .2187223     1.40   0.162    -.1230038    .7343717
     1.Audit |  -.3657201   .2420971    -1.51   0.131    -.8402217    .1087816
   1.Quality |   .1506484   .2530481     0.60   0.552    -.3453167    .6466135
        Size |  -.0002739   .0007695    -0.36   0.722    -.0017821    .0012343
     FirmAge |  -.0033223   .0044911    -0.74   0.459    -.0121246    .0054801
  Experience |   .0039135   .0087645     0.45   0.655    -.0132646    .0210915
       1.WLB |   -.432172   .5547126    -0.78   0.436    -1.519389    .6550448
             |
    Industry |
          2  |          0  (empty)
          3  |  -.0260483   .2275051    -0.11   0.909    -.4719501    .4198536
          4  |   .2114259   .4017578     0.53   0.599    -.5760048    .9988567
          5  |          0  (empty)
          6  |   .1902457   .5212343     0.36   0.715    -.8313547    1.211846
             |
WLB#Industry |
        0 2  |          0  (empty)
        0 5  |          0  (empty)
        1 2  |          0  (empty)
        1 3  |   .3532311   .7439421     0.47   0.635    -1.104869    1.811331
        1 4  |          0  (empty)
        1 5  |          0  (empty)
        1 6  |          0  (empty)
             |
       _cons |  -1.691725   .4586585    -3.69   0.000     -2.59068   -.7927713
------------------------------------------------------------------------------

As you can see, the combinations of WLB = 0 with Industry = 3, 4, 6 are not presented.
I thought maybe it was because there were maybe no observations in these combinations, but that's not the case, also even if this was the case, it would be presented (as is the case with WLB = 1 and industry = 5,6).

Does anybody know what I am doing wrong?

Kind regards
Elise

[Carlo Lazzaro]

Elise:

note: 1.WLB#2.Industry omitted because of collinearity
note: 1.WLB#5.Industry identifies no observations in the sample
note: 1.WLB#6.Industry identifies no observations in the sample

It may well be that the when

...WLB = 1 and industry = 5,6...

some other values was missing and Stata rules that observation out the -e(sample)- (by the way, you can type -e(sample)- after -probit- to see how many of y6our original observations were included in -probit- regression).

[Marcos Almeida]

Carlo pointed out the main aspects.

I just wish to add that: a) considering there is no reason for keeping the interaction term in the model; b) assuming that, for some reason, you wish to keep the "perfect predictions"; c) you wouldn't mind to change it into a logistic model; when, then you might try the exact logistic regression model (help exlogistic).

[Elise Sobrie]

Dear all,

Thank you for the comments!
I see that not all my observations are used in the regression (272 instead of 345), but the weird thing is that there are more observations when WLB = 0 than when WLB = 1. Thus, because o fthis I think its striking why only two options are showed instead of five when WLB = 0. Besides, no explanation is given for the zero value for WLB = 0 and industry = 2, 5

Also, I cannot leave this interaction term out, as this is part of my research. And when trying the exlogistic I get the error that I cannot use factorial variables (but I need them).

Does anybody has advice?

[Marcos Almeida]

The reasons are provided in the output. As Carlo mentioned, you can use - e(sample) - if you wish to check what remained "unscathed" after the listwise deletion.