Bryan Caplan of Econ Log issues a call for papers, or perhaps for noodling around with a spreadsheet.
What happens if you regress annual global temperature 1880-2011 on CO2, linear trend, and other stuff trending positively or negative over this era?  The list of regressors should ideally include not just other climatological variables, but placebo variables like church attendance per capita, the Dow Jones, televisions per capita, etc.

Key question: Does CO2 really dominate in such a regression?
Dig further into the comments and into a follow-up post, and all the standard econometricks reveal themselves: cointegration, unit roots, autocorrelation, multiple causes.  It's representative of the state of the empirical art in economics these days, in which the technique often crowds out the story itself.

Perhaps I'm an antique, but is anyone else as troubled as I by the use of the expression "tease out" as description of a research strategy?  Here's the usual dynamics: investigator proposes a testable implication, investigator downloads from a public source some information that proxies reasonably well for the causes and the effects, investigator does the recommended laundering of the data (sorry, all this talk about filtering and screening is too much to resist), investigator runs the recommended regression package and comes up with a bunch of estimates lacking statistical significance, let alone the  correct sign.  Resulting seminar becomes a collection of suggestions for using additional scrubbing techniques or adding explanatory variables (often under the rubric of "controlling for") so as to get confirmation of the prior, rather than looking for possible reasons the prior is incorrect.

So might it be with rising global temperatures.  Until someone obtains the signature of Midgaardsormen, the carbon content hypothesis may be provisional at best.

No comments: