Ciao!

Stats geeks only

03.08.2005

Light blogging lately. Because I'm rushing to meet a deadline for my MPSA conference paper. It's due in 10 days. Fret not, it's mostly written. But based on department-level ordinary least squares (OLS) regression data, not province-level panel estimator regressions data. It took me a while to find enough man-hours to input five years of election data (much of it in non-digital form) into a STATA format. I was able to import 1997 & 2002 (thank you, again, National Electoral Court) from Excel spreadsheets (after some manipulation).

Then, panic struck. I wasn't able to get STATA to do anything w/ my dataset. I kept getting these annoying type mismatch r(109) error messages. I finally figured out the problem: many of my variables (most of them, actually) were set to string, not numeric. Keep in mind, my dataset has 546 observations, 46 variables.

Let me just say this: Combing through an entire dataset, that size, to search for anything that might make a value not "numeric" is an absolute pain. But, yes, I found a few commas (as in 1,000), which is a STATA no-no. And, w/ that, I'm back in business. It only took all of today.

My stress level significantly reduced, I can actually start running statistical models. Yes, you may yawn now.

Oh, but since I know at least one economist (and a few political scientists) reads this. A question thrown out for advice. I'm using xtreg, clustering my observations around province, and using population-average estimators. Basically, I'm running these kind of models:

xtreg depvar expvar expvar expvar, pa i(prov) t(year)

Or do you recommend something else?

-----
UPDATE: My stats mentor (Dr. Corder) suggests that I instead use

xtreg depvar expvar expvar expvar, re i(prov)

xtreg depvar expvar expvar expvar, fe i(prov)

xtreg depvar expvar expvar expvar, be i(prov)

Posted by Miguel at 11:41 PM

Comments

Posted by: Melli at March 9, 2005 07:46 AM