Friday, November 02, 2007

Statistics and the WSJ Op-ed page

There is an embarrassingly bad piece in today's WSJ op-ed page:

http://online.wsj.com/article/SB119397079767680173.html?mod=opinion_main_commentaries

The two writers follow the number of executions and the number of murders across time, conclude that there is a strong correlation between executions one year and number of murders the next, and then say the fact that they haven't controlled for anything is probably OK, because of Occam's razor.

Aside from the fact that they don't deal with the special issues presented by time series data (correlations produced by two series of data with strong trends in them will be spuriously high), their Occam's razor approach pretty much ignores how serious people do social science research. Richard Freeman, the Harvard Labor economist, says something like: "it had better be there in the correlation, it has better be there in the OLS regression, it had better be there in the IV regression, and it had better be there in the high-tech Maximum Likelihood estimate.

I am open minded about the possibility that the death penalty is a deterrent. But this morning's piece does nothing to pursuade me; worse, by polluting social science, it undermines social science.

No comments: