Having finished reading the paper, I have to say I was quite impressed by the care taken to cover all angles and 'plug all holes' as it were. Kudos to the author.
The study basically used two different datasets - one from various Census reports (1960-1980), and another from the National Health and Nutrition Examination Survey. The author conducted varying types of regression analysis and in each case and for both datasets came to the conclusion that education (as instrumented by variables representing compulsory attendence and other compulsory law changes that starting coming into effect after the 1910s) has a postive causal impact on health (as measured by mortality rates of hypothetical cohorts).
The use of the education instruments I find fascinating. Basically, the author was able to construct a quasi-natural experiment by comparing mortality rates of cohorts in states before and after compulsory education laws went into effect, and also note the effect such law changes that effect education (at the time the cohorts were children in the 1910s-30s) have on future (adult) mortality of synthetic cohorts. The Census data was used to track more aggregated data, while the NHNES could be analyzed at the individual level.
Some of the more valid criticisms to the author's findings are posted on Mankiw's Blog. I tend to agree with at least one commenter that said that using such compulsory law changes as instruments for education may be a bit muddied by the fact that such changes took effect in a period in America's history where the labor laws (and especially the child labor laws) were quite lax. As such, what the author thinks is instrumenting for "education," may in fact be instrumenting for "time away from hazardous job," or something of that nature. In other words, the marginal year of schooling required by the compulsory law changes may be one less year working in a coal mine.
However, to be fair, the author points out this very possibility on page 26 of the paper (see right side of my blog):
"The second issue is that the effect of education is quite large, and it is important to understand
why. I categorize the potential effects of education into two categories: direct and indirect effects. First note that if the effect of compulsory schooling was to remove kids from unhealthy jobs and bring them to the classroom, then the effect of education cannot be distinguished from the effect of not working."
She still concludes that, while that may be an issue, it doesn't seem to be able to account for the quite large and significant effect of compulory education changes as a child on adult mortality. And on this point i think she may be on to something:
"An additional year of education lowers the probability of dying in the next 10 years
by approximately ... 3.6%." That's pretty big.
Of course there are other issues mentioned by the author, such as the possibility that the effect of education on mortality (health) is mediated by income or occupational choice. But again, the author notes that the effect of education on health is so large, that income and occupation can't account for all of the relationship.
All in all, this was an interesting read. I definately learned quite a bit from it (specifically on exactly what is synthetic cohort analysis). And while I don't think I 100% am ready to say that "education" is a main driving causal force of future health, I definately think this provides evidence that demands a second look.