Sunday, 13 December 2009

DeLong, then and now

DeLong and Lang's 1992 "Are all economic hypotheses false?" was a very nice contribution. Long story short: given the ease with which researchers can pick variables to ensure their chosen specification has significant results, and the near impossibility of publishing things that show statistically insignificant results, all published standard errors in economics (or any discipline where this happens) are wrong: they're all biased downwards, and a proper accounting for publication and researcher bias would have a hard time rejecting the null that all of them are insignificant.

And so all the recent shenanigans at Hadley had me thinking of DeLong and Lang.

Today DeLong approvingly cites Mark Kleiman.
Assume some climate model predicts that... temperature would rise 3° C by 2100. If the model were very accurate and precise, that might be 3°± 1°. If the mechanisms involved remain obscure and the data unclear--as is the case today – that might be 3°± 5°.... Given how bad a 3° increase would likely be, if we knew for sure that would be the outcome in the face of inaction there would be a strong agument for making big and expensive policy changes to prevent it from happening.
But following DeLong and Lang, it isn't that the confidence interval is just wider, it's that our current uncertainty (coming from Hadley) is to what extent the estimates have been picked from the top rather than the middle of the interval. Are all climate hypotheses false?

No comments:

Post a Comment