How to interpret this marginal effect's pairwise comparison(post-estimation command) for this logistical regression?

[karthy thik]

I ran a logistical regression to test the probability of death ( a binary variable, 1= death, 0 = death) based on independent varaibles such as aids and cancer. I gained statistical significant results for both explanatory variables Aids (coded as 1 = aids, 0= no-aids), and cancer( coded as before in the binary form), but can someone explain how to interpret the margins pairwise comparison command. I am confused with the aids#cancer comparison below the individual comparison of the independent variables, aids and cancer. How does this aids (0, 1) compare with cancer (0,0), and there are other variations in the possibilities of the two independent variables (one can see from the table below). Some tip would be helpful. I relied on the stata command documentation, but it is not so clear.


margins AIDS##cancer, pwcompare (pveffects)

Pairwise comparisons of predictive margins
Model VCE : OIM

Expression : Pr(death), predict()

------------------------------------------------------------
| Delta-method Unadjusted
| Contrast Std. Err. z P>|z|
----------------+-------------------------------------------
AIDS
1 vs 0 | .0128684 .0047574 2.70 0.007
|
cancer
1 vs 0 | .0429721 .0265377 1.62 0.105
|
AIDS#cancer |
(0 1) vs (0 0) | .0343318 .0223325 1.54 0.124
(1 0) vs (0 0) | .0127306 .0047105 2.70 0.007
(1 1) vs (0 0) | .0833103 .0456344 1.83 0.068
(1 0) vs (0 1) | -.0216012 .0232414 -0.93 0.353
(1 1) vs (0 1) | .0489785 .0273949 1.79 0.074
(1 1) vs (1 0) | .0705797 .0442074 1.60 0.110
-----------------------------------------------------------

[Clyde Schechter]

So, for example, the output row:

(0 1) vs (0 0) | .0343318 .0223325 1.54 0.124

means that the expected difference in probability of death between those who have no AIDS but do have cancer (0 1) and those who have neither AIDS nor cancer is 0.0343318. The 0.0223325 is the standard error, and the 1.54 are the z-test statistic and the p-value for the hypothesis that this difference is zero.

Similarly, for example, the (1 1) vs (0 1) row shows the expected difference in probability of death between those who have both AIDS and cancer, and those who have no AIDS but do have cancer, along with the standard error, z-statistic, and p-value.

Added: In the future, please post Stata output between code delimiters (see FAQ #12 for explanation of code delimiters) so that it aligns in an easy-to-read way.

[karthy thik]

Thanks for the reply. I know understand the results now. I'll try also to use code delimiters next time. Thanks again for your assistance.