Illustration 1: Working with Panel Data Problems
Picture this, you're working on an assignment where you're trying to figure out how union membership affects wages across a bunch of different years. You've got a panel dataset with multiple workers observed over time.
If you're working all on your own, you might try the simple wage union age education - no thought to the panel structure, though. The standard errors end up being wrong and the estimates are probably biased in some way.
When you come to us for Stata expert help, here's the kind of guidance you get:
First and foremost we sort the data out properly:
xtset id year
Then, we run a quick summary to check for any missing values, outliers or other issues. We have a discussion with you about whether fixed effects or random effects is the best way to go, and if need be, we might suggest running a Hausman test to make up our minds.
We might run:
xtreg wage union age education, fe
estimates store fixed
xtreg wage union age education, re
hausman fixed .
But we don't just stop at producing the output. We're always explaining things: "The Hausman test comes back significant. This means we are controlling for all characteristics that would not change over time among the workers, such as their innate ability or motivation, which gives us a more clean-cut estimate of the union effect."
That sort of insight is what transforms a completed assignment into a real learning experience.
SIGN UP
