Friday, March 31, 2006

Smoke Yourself Thin?

by Tom Bozzo

Yesterday was Journal of Health Economics day at the office, and I couldn't resist picking up Jonathan Gruber and Michael Frakes's "Does falling smoking lead to rising obesity?" (Vol. 25 [2006], 183-197. This has evidently been available online for a few months, but without academic-style access to online journal archives, I have to wait for the dead-tree edition.)

Gruber and Frakes, in large part, are re-assessing the analysis from a 2004 JHE paper (*) that attempted to establish a causal relationship behind the correlation between falling smoking rates and rising obesity rates. The earlier study had, in fact, purported to find an underlying causal link, which Gruber and Frakes do not confirm. Not unsurprisingly, this turns on a few econometric specification decisions that differ between the two studies.

As it turns out, one specification restriction, which Gruber and Frakes relax (at the cost of some degrees of freedom), accounts for roughly half of the previously measured effect, and makes the statistical significance of the effect go away. There is a minor kitten-fight of a reply from the authors of the 2004 paper.

As for me, I'm skeptical about this causality claim. Self-help videos hosted by Troy McClure aside, factors other than smoking must be the drivers of increases in child obesity rates, not to mention that among the nonsmoking majority. So while it's possible that reduced smoking could lead to society-wide weight gain, it's hard to imagine how the causal effect could be very large.

(*) Chou, et al., "An economic analysis of adult obesity: results from the behavioral risk factor surveillance system," JHE 23, 565-587.
So what was the restriction?
The first paper used a quadratic time trend; the newer one uses time dummy variables. That the smoking causality result is not robust to that specification choice does not give me confidence that Chou, et al., found a real relationship.
Post a Comment

<< Home

This page is powered by Blogger. Isn't yours?