Oaxacan model

[Katherine Oleas]

hello everyone,

I am carrying out a bias-corrected model of Oaxaca to see the gender wage gap. And I get the estimate that I show you below. However, when performing the analysis and seeing the explained and unexplained part, its p-value is not significant, I don't know if this matters when analyzing these variables.

Greetings,
Katherine

[George Ford]

the difference is small and not different from zero.

[Katherine Oleas]

Thanks George, my question is the adjusted part where the difference, what is explained and not explained is not significant. In that case, can I analyze the coefficient or not?

[George Ford]

The differences, and their decompositions, are stat insig. There's nothing there.

You are analyzing the coefficients, and they are not different from one another. It leans toward positive by the confidence interval (and it appears stats is moving in that direction).

[George Ford]

the group_1 group_2 results are just a means test = 0, which is mostly uninteresting. It's the difference you're interested in.

[Katherine Oleas]

Thank you very much George, so in my case the difference is not significant, I wouldn't be finding anything in the model, would I?

[George Ford]

Nothing much to report in those results. And, the difference is quite small.

[Katherine Oleas]

Thanks George,
One more query because if the difference is not significant, the unexplained term is significant in the following estimation.

[George Ford]

Yup. This suggests that the the treated group should have a larger mean than the untreated group.

[Katherine Oleas]

I would like to know if I am doing my estimation correctly. I correct my Oaxaca model for the selection problem, but I don't know if I am correctly estimating the inverse ratio of mills. Here I put the code.

*Step 1: selection model
probit prob_trabajar escol edad edad2 ecivil jefe i.regiones if area==1 & female==1
predict xb if e(sample), xb

*Step 1.2: Calculate the inverse Mills ratio
gen mills = normalden(-xb)/(1-normal(-xb))
label variable mills "inverse Mills ratio" //poner etiqueta

replace mills=0 if female==0

*Step 2: Estimation of the model with the inverse Mills ratio

svyset [iw=fexp]
oaxaca lingreso escol exper exper2 jefe ecivil tamaño condicion normalize(b.regi_1 regi_2 regi_3) normalize(b.raza1 raza2 raza3 raza4) normalize(b.actividad1 actividad2 actividad3 actividad4) mills if area==1, by(female) omega noisily relax swap svy adjust(mills)

[George Ford]

What is the nature of the selection problem?

And, your covariance matrix will be wrong if the mills ratio is stat sig. You'll need a Murphy-Topel Cov Matrix, unless oaxaca corrects for that.

[Katherine Oleas]

Selection problem:
The sample selection bias occurs when, in a wage function, the variables that explain the wage influence the decision to work or not and, consequently, the sample considered in the regression only includes those who participate in the labor market. In the context of the gender wage gap, this bias occurs because generally the women who participate in the labor market are not a random sample but rather have certain characteristics that make them more likely to participate.