Dear Statalisters,

I am using the Blinder-Oaxaca decomposition with the following code:

Code:
 oaxaca  faminta mfamily112_2-mfamily112_4  [pw=weightres4], by(nssec3par) categorical(mfamily112? )  noisily relax logit

where mfamily is a categorical variable with four categories. When I specify the model, I omit one category to avoid multicollinearity.

However, when I obtain the decomposition the output shows four categories (see below for an example), including the reference, due to the fact that I specified categorical(mfamily112?).
Code:
endowments   |
mfamily112_1 |  -.0250903   .0060485    -4.15   0.000    -.0369452   -.0132353
mfamily112_2 |  -.0019509   .0011123    -1.75   0.079    -.0041309    .0002292
mfamily112_3 |  -.0043752    .004209    -1.04   0.299    -.0126246    .0038742
mfamily112_4 |   .0001289   .0021492     0.06   0.952    -.0040835    .0043412


IfI had not specified this option I would have had the decomposition of marginal effects (wrt to reference category) (see below)

Code:
                        
endowments    
mfamily112_2    -.0044638    .0020395    -2.19    0.029    -.0084611    -.0004665
mfamily112_3    -.0219248    .0055668    -3.94    0.000    -.0328355    -.011014
mfamily112_4    -.0048989    .0030191    -1.62    0.105    -.0108162    .0010185
.

I was wondering whether both the decompositions are correct, depending on the research question or whether one of them is clearly wrong and I did not get it. .

Thank you and best regards,
Lydia