3
$\begingroup$

I need help understanding this selection bias problem. I am estimating a mincer equation for a final year project and I was told I need to worry about self-selection bias IN OCCUPATIONS.

My lecturer said that, because wages vary between occupations, and individuals select occupations as a choice, the sample is selected. In that sense, I need to correct this problem. We were taught the Heckman selection model in the context of people deciding whether to work or not. I understand this case, but not the case of occupations. I mean, all is a choice, isn't? In my model I also have hours of work, public or private sector, full time or part time, region, industry, and other common variables. Why do we worry about occupations but not about everything else? I just cannot get my head around this.

Occupations are low, middle and high skill. I'm estimating the returns to education for Australian workers.

I think one thing is important, which is if the variable is included or not in the regression. For example, normally you cannot include a variable of employment or inactive, because inactive have no wage. This is why, as I understood from the lecture, we need to test for selection. But choice variables that are included, are they a concern?

$\endgroup$
  • $\begingroup$ Do you know, for each observation of your sample, what occupation is about? $\endgroup$ – Alecos Papadopoulos Aug 24 '18 at 12:30
  • $\begingroup$ To clarify a bit: What specifically is your primary hypothesis/what are you trying to use the Mincer equation to explain specifically? Returns to education and/or experience? $\endgroup$ – AndrewC Aug 24 '18 at 13:26
  • $\begingroup$ It's low, middle and high skill. I'm estimating the returns to education for Australian workers. $\endgroup$ – user928172 Aug 24 '18 at 14:31
  • $\begingroup$ You’re right. All is a choice. The lecturer just chose to have you address selectivity in occupation. Why? Because people in the field are interested in the issue. That’s what I understand. $\endgroup$ – chan1142 Aug 24 '18 at 23:55
2
$\begingroup$

There is no fundamental difference between the choice of occupation and say region. In fact, there exists a large literature in migration that builds on Roy's model, starting with Borjas (1987). I'd argue that ideally you'd like to correct for self- selection/ sorting in all choice dimension. In practice this can be hard/ cumbersome/ impossible to do.

Why do we tend to worry about self- selection more in the context of occupations than regions? I suppose it boils down to the fact that everyone had to actively choose an occupation at some point, whereas we are all born in some region. But, again, there is no fundamental difference.

$\endgroup$
0
$\begingroup$

The idea is that education is only a partial measure of ability. If high ability people select into certain occupations because the returns to ability are higher in those occupations, then a simple Mincer regression will conflate the returns to ability and the returns of the occupation.You need to account for the selection effect to get an unbiased estimate of the returns to occupation and education.

$\endgroup$
  • $\begingroup$ Ok, but take your phrase "If high ability people select into certain occupations because the returns to ability are higher in those occupations, then a simple Mincer regression will conflate the returns to ability and the returns of the occupation", and replace occupation with other variable. For instance, region, or sector. What's the diference? $\endgroup$ – user928172 Aug 24 '18 at 14:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.