Panel data interpretation

[David Deminion]

I need to analyze how returns to education have changed based on major selection overtime using panel data. The panel data I have surveyed the respondents in 1992 and then again in 2012. I do not understand how I could use the data to show how returns have changed overtime, as from my understanding, panel data is the same person surveyed overtime. So when the individuals were interviewed in 1992 they were all in their 30s and now they are in their 50s so their earnings are going to be different since they are older. I am sorry if this is a simple question I have never worked with STATA before. Thanks in advance!

[William Lisowski]

The question is not "did returns to education change over time", the question is "did returns to education increase more for some majors and less for others".

Another member may be able to advise you further on this, but fundamentally, this is a question of statistical modeling - once you know then appropriate model, that will dictate which of Stata's tools you use for your analysis.

I suspect the statistical modeling will be a "difference in differences" model, as Wikipedia describes it.

[David Deminion]

Hi William, thank you so much for your response, I really appreciate it.

[Carlo Lazzaro]

David:
as an aside to William's helpful reply, my guess is that your model may suffer from endogeneity, as, on average, smarter people achieve higher educational levels and, on average, negotiate better wages.

[David Deminion]

HI Carlo, thank you for pointing that out. I was going to control for ability via SAT scores. The place where I work has a room with restricted data. However, I am not sure I will have access to it in the near future. I was hoping a fixed effects model could help, but I am not familiar with it so I am not sure if that is a valid alternative yet.

[Carlo Lazzaro]

David:
thanks for clarifying.
The main drawback I see is that you have a scant number of data waves (2) too far apart in time.
I would check whether your approach was actually used in previous researches in your field.

[David Deminion]

Thank you for your insight Carlo. I agree the gap is quite odd to go from 1986 to 92 and then wait until 2014 to survey again.

[David Deminion]

Hi everyone I have another question I hope you don't mind answering. I have settled on using NLSY 97 data. I am running two separate regressions to see how returns on education have changed for some majors over time. I am using two separate years of income reported, 2010 and 2015 as my dependent variable. My model would look something like this:

log of income (2010)= b0+ b1 race, gender ect. + b2 years of education+ b3 + field of study + u and then my second regression would be log of income (2015)= b0+ b1 race, gender ect. + b2 years of education+ b3 + field of study + u

Respondents were born between 1980 and 1984. Do I need to control for age since they are five years older for the 2015 income regression? Thank you so much.

[Carlo Lazzaro]

David:
in general, splitting data and running different regression is not a good idea (but you might have sound reason to do so).
That said, controlling for age may be worth trying. Hovewer, I would more interested in investigating whether years of education have a non-linear relationship with the regressand.
As an aside, usually these models are at risk of suffering from endogeneity, since, Others things being equal, smarter people, on average, achieve higher educational level (which obviously does not exactly mean that they stay longer in the education system) and, on average, negotiate better wages, which, on average again, represent a relevant share of their income.

[David Deminion]

Hi Carlo, thank you for your feedback. I really appreciate it. I agree with the issue of endogeneity regarding ability, I was supposed to have access to the restricted data with SAT scores. I was thinking about using college selectivity as an ability control. I know it is a bit noisy, but I read in a paper that college selectivity is not a good proxy for college quality because colleges are only interested in attracting intelligent students, and not not necessarily care if they make improvements while at their college. Do you think it is unreasonable to use college selectivity as a proxy for ability given my constraints?

[Carlo Lazzaro]

David:
college proximity can be a good instruments to fix endogeneity (see https://www.stata.com/bookstore/micr...metrics-stata/, page 177-179).