Looks worth reading...
Excerpts:
Diagnosed rates of autism spectrum disorders have grown tremendously over the last few decades. I find that assortative mating may have meaningfully contributed to the rise. I develop a general model of genes and assortative mating which shows that small changes in sorting could have large impacts on the extremes of genetic distributions. I apply my theory to autism, which I model as the extreme right tail of a genetic formal thinking ability distribution (systemizing). Using large sample data from the Centers for Disease Control and Prevention, I find strong support for theories that autism is connected to systemizing. My mating model shows that increases in the returns to systemizing, particularly for women, can contribute significantly to rising autism rates. I provide evidence that mating on systemizing has actually shifted, and conclude with a rough calculation suggesting that despite the increase in autism, increased sorting on systemizing has been socially beneficial.
Since ASDs are relatively rare (60 cases per 10,000 for ASDs overall, and between 10 and 20 cases per 10,000 for autism4 ), increases in the variance will have a large impact on the portion of the population with ASDs. Specifically, I calculate in section 3.5 that the portion of people with ASDs should grow seven times as quickly as the standard deviation, and the portion of people with autism should grown ten times as quickly as the standard deviation. This kind of rapid growth really does happen, as we can see from the example of height. In section 3.5, I show that the standard deviation of the height distribution in the US grew 10% in the last fifty years. Being shorter than 4′ 10′′ or taller than 6′ 2′′ are both about as rare as having an ASD, and my model predicts that each group should have had relative growth of about 65%. Accounting for the change in the mean, the share of those under 4′ 10′′ actually increased 92%, and the share of those over 6′ 2′′ actually increased 73%, both even higher than the already very high prediction. If the systemizing distribution had its standard deviation grow as much as the height distribution actually did, that alone would explain a doubling in autism rates.
Baron-Cohen also led a study (Roelfsema et al., 2011) which looked at the autism prevalence in three regions of the Netherlands: Eindhoven, Haarlem, and Utrecht. Eindhoven has 30% of its population employed in information technology, compared with 16% and 17% for the two other regions. As predicted, Eindhoven has a much higher prevalence of childhood autism.
Durkin, et al (2010) showed a relationship between measures of socioeconomic status and autism prevalence in a dataset that included the MADDSP data I am using.15 . I include, as controls, the same measures that they used, median household income, the poverty rate, and the percent of adults with a bachelor’s degree.16 As a placebo, I also regress all of these measures on cerebral palsy (CP) prevalence as measured by MADDSP using the same methodology. Like ASDs, impairment from CP can be mild, moderate, or severe. There is no reason to believe that systemizing is related to CP, so this can help us to see which variables are related to detection, and which are related to true prevalence.
As an example, suppose that those well-qualified for systemizing jobs like being an engineer (or economist) are those who are more than one-standard deviation above the mean in the systemizing distribution, and ησ (1) = 1.53. If full-blown autism had a prevalence of around .001 a generation ago, and those with autism represent the extreme right tail of the systemizing distribution, then those with full-blown autism were 3.09 standard deviations away from the mean, and ησ (3.09) = 10.4. So ησ (3.09)/ησ (1) = 10.4/1.53= 6.82, which means that increases in the variance of the systemizing distribution will increase the autistic population nearly seven times as much (in proportional terms) as they will increase the population of those qualified for systemizing jobs. Therefore, a shift in assortative that doubled the size of the autistic population would cause only a 100%/6.82= 14.7% increase in the population of those qualified for systemizing jobs.
However, part of this is due to a general increase in height: between 1959 and 2009, the mean height of those between age 20 and age 55 increased by .53 inches. To account for the increase in the mean, I add .53 inches to the 2009 thresholds. In 2009, 241 of 3712 people sampled were taller than 72.53 inches, which is 6.5% of the population, and 36% higher than 1959. To compute our prediction for comparison with the true value, we need to know how many standard deviations 72 inches was from the mean in 1959 (1.67 standard deviations), and how much the standard deviation grew between 1959 and 2009 (10.35%). The formula gives ησ (1.67) = 3.47, so our predicted increase would be 3.47 ∗ 10.35% = 35.9% which is very close to the observed 36%.
There are many more women in systemizing occupations than there used to be, and we see more marriages where husband and wife are both in systemizing occupations. However, it is possible that the schoolteacher wife of a male engineer from a few generations ago would have been an engineer herself if she’d had that opportunity. If that were the case, it could be that assortative mating on observables rose, but assortative mating on genes held completely steady.
There are two things to observe in the graph. First, the graph supports the idea that field of degree is a meaningful measure. The spike in degree attainment around age 60 is driven by the Vietnam war and the associated draft.26 The spike is much more muted when we look at census science degrees, and is basically absent when we look at hard science degrees. If field choice were purely random, we would see a proportional spike in each line, so the absence of a spike suggests that hard science degrees require certain abilities that are not common among those on the margin for college. Second, the portion of men with hard science degrees has remained nearly constant, at just under 10%.
Ganz (2007) estimated the lifetime social costs of a marginal case of autism, including medical and non-medical care, and lost productivity, and arrived at $3.2 million.29
While the increased risk of autism is probably not priced into mating decisions, this suggests that increased assortative mating on systemizing is probably a good thing on balance. But while direct intervention does not seem like a good idea, there are still policy implications to this view of autism. If assortative mating has significantly contributed to the rise in autism, autism is likely to keep rising for several more generations, because it takes multiple generations to reach a new genetic equilibrium. Investments in better autism care are probably more cost-effective than they appear at the current prevalence.Nov 10, 2012
