Tuesday, May 29, 2012

If your observables are correlated with the variable of interest, how dare you assume your unobservables are not?

Trees reduce crime. I love trees, but I'm not sure I buy it.
http://grist.org/cities/in-baltimore-the-gods-will-not-save-you-but-the-trees-will/

OK, maybe I should actually read this paper before calling it out, but just because you control for SES, doesn't mean you fix the problem. Unless you've got a convincing identification strategy, you've still just got an interesting partial correlation.

No comments:

Post a Comment