Returns to education between males and females

[Rosabel Kilgour]

I am trying to determine the difference in returns to schooling between males and females. I was just wondering if anyone knew if it is better to split data in half (one half for males and one half for females) and run 2 separate regressions and compare or to run a regression with interaction terms? Any thoughts would be hugely appreciated!

[Clyde Schechter]

I think the answer to this question depends on how many observations you have for each sex, and how many (and what kind of) other variables you will be using in your model. It also depends on what specific regression commands you plan to use, as some are easily managed with -suest- and some are not supported by it. It also depends on how reasonable it is to assume that the confounding effects of the other variables in your model are the same for both males and females.