0
$\begingroup$

I want to estimate log(wages). Most wage estimations suffer from sample selection bias. So I used the two step heckit procedure to correct for it.

The problem is that I get an insignificant inverse Mills ratio in my wage estimation which means i have no sample selection bias.

My exclusions restriction in the first stage probit model are the variables:

  • number of children between age 0 to 4
  • number of children between age 5 to 7
  • number of children between age 8 to 15
  • number of children between age 16 to 18
  • degree of disease
  • degree of disease squared
  • non labor income
  • marital status

the variables that I use in both regressions are :

  • origin
  • employment experience part time
  • employment experience full time
  • employment experience part time squared
  • employment experience full time squared
  • public sector
  • level of education

In the wage equation I also have 3 variables that are only available for working people:

  • business sector
  • number of years employment in company
  • number of working people in company

What can I do? Maybe add a variable in the probit equation or something else to get a significant inverse Mills ratio in the wage estimation?

$\endgroup$
  • $\begingroup$ You mean Mills ratio. The ratio is named for John P. Mills, who tabulated it. The forms Mills ratio, Mills' ratio and Mills's ratio all seem defensible, depending on preferences about punctuation, but his surname was Mills. Meanwhile there is a statistical question underlying your allusion to code in some unnamed software; sorry, but my knowledge here doesn't extend beyond my pedantic correction, so I can't suggest an answer or even what other information would help. $\endgroup$ – Nick Cox May 17 '19 at 15:39
  • $\begingroup$ @AdamO started editing out the previous error, so I finished the job. There is a longstanding convention that OP errors that arise from substantial misunderstanding should be left for the OP to correct; which side of the line that one falls is open to small discussion. $\endgroup$ – Nick Cox May 17 '19 at 16:09
  • $\begingroup$ @NickCox fair enough. OP should just use weighting if they have the appropriate information on the target population/non-responding sample. If they don't have that, I wouldn't trust any missing data method, no matter how convoluted (and the described approach is quite convoluted indeed). $\endgroup$ – AdamO May 17 '19 at 16:43
  • $\begingroup$ What does OP means? $\endgroup$ – MasterStudent1992 May 17 '19 at 17:14
  • $\begingroup$ @MasterStudent1992 OP means "original poster", i.e. specifically you here, generally whoever starts a thread urbandictionary.com/define.php?term=Op $\endgroup$ – Nick Cox May 17 '19 at 17:36

Your Answer

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

Browse other questions tagged or ask your own question.