Econometrics: Controls and Difference-in-Differences

[aridneptune]

Sorry for the double post with the SS section -- I thought this was probably a better place to ask this question.

Hello all,
I'm working on an econometrics project in which I'm using difference-in-differences approach to estimate the effects of the 1996 TANF welfare reform in the US. My main dependent variable is ln(total household income). My treated group is single women with children and my control group is single women without children.

My question is, when is it appropriate to add controls in a difference-in-differences regression framework? Is it only when one believes that the value of the control variables differs systematically (i.e. with time) between the treatment and control groups and also affects the dep. var.?

A related question: I want to control for education and want to use an instrumental variable for education. Any good ideas of IVs in the Current Population Survey (must use this dataset)?

Thanks very much!

 

[mmwave]

Dear poster,

Thank you for your question. In a difference-in-differences approach, it is appropriate to add controls when there are other factors that could potentially affect the outcome variable (in this case, ln(total household income)) and are not related to the treatment (1996 TANF welfare reform). This could include individual characteristics, such as education or employment status, or external factors, such as changes in the economy or policies unrelated to the TANF reform. Adding controls can help to isolate the effect of the treatment on the outcome variable and provide more accurate estimates.

As for your question about using an instrumental variable for education, it is important to carefully select an instrument that is strongly correlated with education but not directly related to the outcome variable. This can be challenging, but some potential IVs in the Current Population Survey could include parental education or parental occupation. It is important to thoroughly test the validity of your chosen instrument before using it in your regression analysis.

I hope this helps. Good luck with your project!