Friday, March 26, 2010

Panel Data (for the stats nerds only)

I got way more excited by this project yesterday when we finished teaching Stata and started talking about what we could actually do with the data. There was a pretty great question about the effect of community-based health insurance on health care utilization in Rwanda which I totally thought we could answer with our panel data using individual fixed effects. Only panel data isn't any better at getting you causal inference if there are unobserved shocks that are correlated with the first-differenced x's and y's. (That is, OLS is biased if there are unobserved e's correlated with both x and y; panel data instead regresses delta-y on delta-x, and if you use individual fixed effects you eliminate the e's. Only you still don't observe delta-e, and if that's correlated with delta-x and delta-y, you've got the exact same problem you had before. Ugh.) Still a cool idea worth looking into.

ANYWAY, switched to a far swankier hotel this morning, hoping to get out of Kigali for the weekend, maybe headed to Lake Kivu, or maybe to the northwest.

No comments:

Post a Comment