>Granted, W's book is not a
>comprehensive literature review. Nevertheless, Jenck's
>criticisms (as I've heard them summarized) are not the
>slam dunk refutation that Doug assumes.
It's not a "slam dunk refutation" - the evidence isn't conclusive either way. Here's what Jencks wrote me last year:
>I do not think the results are at all clearcut either way. There
>MAY be something real here, but it is far from certain.
>
>With a handful of exceptions, almost everyone working on the subject
>has such strong prior views that it's hard to trust what they do.
>There are dozens of small technical decisions to make in this
>research, each of which affects the outcome. The rationale for
>preferring one rather than another is often unclear. Yet,
>amazingly, each researcher seems to keep finding the same thing over
>and over -- either income inequality affects health or it does not.
>That's like the thirteenth chime of the clock.
>
>The root of the problem is that we have a very limited amount of
>data, a weak theory about what really causes the correlation we
>(sometimes) observe, and no good way natural experiments.
>
>Wilkinson's original comparisons were cross-national. They were
>striking, but it turns out that other variables (such as country
>size) also predicted mortality about as well as inequality. That
>would be no problem if we had a good theory about how the inequality
>effect worked, but in fact it is pretty fuzzy. Also, we only have
>reliable data on inequality for a small number of rich countries in
>a small number of years, and Wilkinson's results are very sensitive
>to choices of countries, years, measures, control variables, etc. I
>can make them go away, come back, and go away by fiddling with
>seemingly innocent details.
>
>We also have data comparing US states and similar regions of a few
>other rich countries. In general, these comparisons show moderate
>to strong correlations between mortality and inequality, but it is
>usually easy to make these correlations become statistically
>insignificant by controlling other area characteristics. Opinions
>differ about which controls are appropriate.