Hi Binto, you can use one of the censored regressions (tobit or double hurdle, lognormal, etc) to model non-participation. Depending on your story/theory, a Tobit model would best fit a scenario that assumes decisions and extent in labour market participation is an outcome of a single decision. If you assume whether to participate in labour market and extent of use are two separate decisions, then use double hurdle which will allow a Probit for participation in the first stage and a truncated reg in the second stage for the extent of participation. You can also choose to use Heckman two-step sample selection model if you believe zeros in dependent variables (i.e. non-participation) is as a result of incidental truncation or if zero observations does NOT represent a corner solution or an optimal choice.

You can correct for endogeneity using Instrumental Variables (IV) approach. If you choose Tobit, use IVTOBIT. If you choose the more flexible double hurdle or Heckman models, you can use control function approach. I did do something similar in this paper (https://www.nottingham.ac.uk/credit/documents/papers/2018/18-10.pdf)

Hope this helps and good luck to you!

thank you so much Hayatullah for such a great insight. so based on your advise i took some time and went through the Heckman correction model(Heckman two stage sample selection) and i believe it fits perfectly what am trying to work on that in the first stage I will run a probit labor force participation then calculate the predicted inverse mills ratio and assign to each observation and go ahead to run a wage offer equation.

Hi Binto, yes in wage estimation Hackman is a good fit, but often hard to find variables to allow for the exclusion restriction. In stata, you can run a probit and obtain the IMR and use it as an additional explanatory variable in the second stage regression, and see whether it is significant. Alternatively, Stata can do it for you automatically using the heckman command. type "help heckman" in stata to see some examples. Good luck!

Hello Binto,

I would like just to turn your attention to another interesting problem.

You want to include in your model people who are in working age but do not participate in the labor market. The econometric techniques that you are discussing are ok. But you may want to take into account also the fact that people who report zero wages may be participating in the labor market. For many reasons (for example to receive unemployment benefits) some people when surveyed will declare that they do not work while actually they work in the grey economy. So, the number of people declaring that have zero wages may not be a good variable to describe the number of people who do not participate in the labor market. The approximation will depend on the size of the grey economy in the country.

Good luck with your research!

Hello Stefan,

Thank you so much for the insight that i actually hadn't thought of from that angle. And it indeed makes a lot of sense that indeed some individuals in the sample may fail to disclose their wage incomes due to varied reasons, so its something that i have rethink and see whether what would be the implication of making a strong assumption that indeed this people who report zero wages actually. In that case what of i consider a variable that reports hours worked and then create a dummy, 1 if one had at least some hours working 0 if one reported zero hours worked . Kindly advise

I think that what you suggest is better than to relay on wages. But still the approximation will depend on who has made the survey, haw has made it and the nature of the questions asked. You have to have a good idea about the purpose of the survey and the motivation of the serveyors and the respondents.