-oaxaca- not correctly calculating the difference between groups

[Jane Wakeford]

Hi, I am using the -oaxaca- command and appear to have errors in my results. I am calculating the decomposition of log hours between two groups. I use the heckman correction option and eform to exponentiate my coefficients. The 'overall' statistic is 29 hours for group 1 and 16 hours for group 2. Descriptively, I know that these estimates are correct. Yet -oaxaca- claims the difference between these groups is less than 2 hours! Please see my code and sample output.

oaxaca lnhours  yearsed exp if sex==2 & etime>3, by(prognorm)  swap eform model1(heckman, select(yearsed age exp tchad) twostep) model2(heckman, select(yearsed age exp tchad) twostep)

[FernandoRios]

Hi Jane
In this case, -oaxaca- is on the side of the truth, at least from what you provide
Basically, you are not using a standard OB decomposition. You are using a heckman selection model. Thus, when OB estimates the decomposition, it will use those models coefficients and characteristics.
After correcting for those, the predicted mean will be different from the actual wage mean, which is why you obtain those results
HTH

[Jane Wakeford]

Thank FernandoRios, however the problem persists even when I don't use the Heckman correction. I thought perhaps it is due to the large variance of my estimates?

Here is my code:

oaxaca lnhours  yearsed age exp tenure if sex==2 & etime>3, by(prognorm) eform swap weight(.5) relax

And here is my output:
















The problem is less pronounced if I do not use eform. Here is my code:

oaxaca lnhours  yearsed age exp tenure if sex==2 & etime>3, by(prognorm) swap weight(.5) relax

And here is my output:

[Jane Wakeford]

I think there is a problem with my interpretation. Using my previous post and logarithmic/exponential maths:

ln(wage_group1)=3.224216

ln(wage_group2)=3.065566

difference=ln(wage_group1)-ln(wage_group2)

exponentiated difference= exp(ln(wage_group1)-ln(wage_group2))=exp(3.223216/3.065566)=

2.8626629 hours

[FernandoRios]

Right, i didnt notice you were using eform. that is feasible, but as you notice, the interpretation is not quite correct