Tuesday, December 05, 2006

Who Really Cares about Arthur Brooks pt. II

I don't know why I don't read the Volokh Conspiracy more often. Heck, it's in my blogroll. Maybe it's because I don't like the font. Maybe it's because the subjects broached there only occasionally interest me. Or maybe I just put it in my blogroll to have a token conservative site in order to present the false impression that I'm balanced.

Whatever the case, this morning I chanced upon a critique of Arthur Brooks' Who Really Cares which I wrote about previously. Not to my surprise, Jim Lindgren finds some statistical jimmying:

I am skeptical of basing so much on the SCCBS, in large part because it reports that liberal families make more money than conservatives (it is not clear from Brooks’s book whether the survey is of a representative national sample). In the 2000, 2002, and 2004 General Social Surveys, which are representative samples of the US, conservative families make $2,500 to $5,600 a year more than liberal families in each one. Although I don’t have the ANES data handy, my recollection is that the economic differences between conservatives and liberals are usually in the same direction and even larger in the ANES than in the GSS.

Note that Brooks is assuming that liberals earn 6% more than conservatives when most of the data show that conservatives in fact earn significantly more than liberals. Thus when he controls for income, he's moving his numbers in the opposite direction from which they should move. Perhaps even worse, as far as I've been able to tell, Brooks doesn't control for cost of living. Liberals are more likely to live in urban areas where cost of living is high, and conservatives more likely to live in rural areas where cost of living is low. That means that disposable income can vary widely even if gross income is the same. There's more:

This problem comes to a head in Brooks’s probit and regression models analyzing SCCBS data (pp. 192-193). After controlling for a lot of things that you might not want to control for (i.e., being religious or secular), Brooks concludes that “liberals and conservatives are not distinguishable” in whether they have made any donation in the last year. This is literally true, but he fails to note that in the model liberals give significantly more than moderates, if a traditional .05 significance level is used, while conservatives do not differ significantly from moderates. Yet in Table 6, the significance level used as a threshold for identification with an asterisk is .01, not .05, as he uses in some of the other tables. In one table (p. 197), Brooks even reports significance at the .10 level, as well as at the .05 and .01 levels.

I can’t rule out the possibility that Brooks changed his reporting of the significance level so he wouldn’t have to explain why, after lots and lots of controls, liberals were more likely to have made a donation than moderates, while conservatives did not differ significantly from either liberals or moderates.

Although I don't have the book and am unlikely to take the time to go through Brooks' arguments with a fine-toothed comb, I'll repeat what I said before: Brooks is highly disingenuous and should not be taken seriously.