Monday, February 21, 2005

Three things about empirical research in economics

DISCLAIMER: Today's post is solely about economics; if you are not an economist, you don't understand what I'm going to talk about at all. I apologise.

Number One: I wonder under which circumstances the use of a cross-sectional dataset, rather than a panel dataset, is justifiable in the 21st century economics. An exceptional case that I can think of is Shleifer and his co.'s legal origin studies (See 25th to 27th January here), though he seemed to be annoyed by criticism based on omitted variable bias concerns. (The best paper that tells you the importance of panel data may be Islam 1995 in QJE.)

Number Two: This is what my supervisor told me when I bumped into him at the STICERD communal area today.

Unlike theoretical research, trust and reputation matters a lot in empirical research as it is difficult to replicate what you've done. (Empirical researchers, like Edward Miguel does for his civil conflict paper published in JPE, have been beginning to post datasets and STATA do-files on their websites to allow others to replicate their results, though.) Telling the audience at your seminar talk that you drop observations in order to get higher t-values destroys your credibility as an empirical researcher.

Number Three: Reading Persson (2005) in order to find a better way of empirical strategy for my research, I finally understand what Robin told me when I gave a seminar presentation last Monday. An empirical methodology called "event study" seems to be the way to go. I haven't figured this out completely yet, but event study is a methodology in which you first identify an "event" that is likely to affect the dependent variable (like economic growth), then take the average of the LHS variable over a certain span of periods (like 5 years) before and after the event, and see if the difference is statistically significant. If it's significant, the event does affect the dependent variable. Obviously this demands a long time span in the data. That's why Robin suggested I use economic variables, like economic growth, which are available for a longer span of time.

This event study seems to become popular among macroeconomic empirical research, like Jones and Olken (2004), Wacziarg and Welch (2003), and Papaioannou and Siourounis (2004). And it seems like I need to jump on the bandwagon (in a good sense).

No comments: