Autism is known to be extremely heritable. It is perhaps the most heritable of all psychiatric disorders. It is about as heritable as intelligence and a lot more heritable than personality. Even though autism is not 100% heritable, no behavior or skill is known to be 100% heritable or anything close to that.
Given the low concordance of autism in siblings and fraternal twins compared to the concordance in identical twins, it is likely that autism results from the combination of a number of alleles, 3 to 15 or even more according to researchers. Since the prevalence of ASD is somewhat high (and I'd argue the current prevalence is still an underestimate) it follows that the frequency of most of the alleles that result in autism must be relatively high.
Ian Parker asks why the frequency of these alleles remains high. Even if only 1% of individuals with one of these alleles became autistic, and assuming that autistics procreate much less often than the average person (which I believe is an undisputed assumption), then the frequency of the allele in question should gradually drop until it is close to zero (or whatever can be maintained through new mutations and limited inheritance), unless non-autistic individuals with the allele reproduce more often than the average individual.
The alleles that result in autism must therefore have a slight reproductive advantage by themselves, and be reproductively disadvantageous when combined.
A follow-up question is why alleles that are reproductively advantageous don't have a frequency that gradually increases until it reaches 100%. Obviously, if all autism alleles reached a frequency of 100%, then nearly every person in the planet would be autistic. So there has to be a sort of frequency equilibrium for each autism allele.
It also follows from the above that the frequency of slightly disadvantageous alleles can be high and stable.
I hypothesize that frequency equilibrium applies to alleles associated with most other spectrum disorders in the field of psychiatry. It also contributes significantly to the huge diversity of human behavior, and explains seemingly paradoxical combinations of creativity, intelligence and neurological impairment.
There are ways to test this hypothesis. The alleles associated with autism can be mapped, and reproduction rates of non-autistic individuals with these alleles can be measured.
For the time being, let us consider the reproduction rates in the families of autistics, taking into account that parents of autistics are usually non-autistic.
I will use data from Table 2 of the US Census Report for 2003. This gives 0.89 children per family, 1.82 children per family with at least one child, and 2.46 children per family with at least two children. (I assume the number of children per family with at least 4 children is 4.25, and I have verified that variations in this number don't produce large errors).
Note that these numbers do not vary a lot from state to state, according to a 2000 Census table. It also seems these proportions have not changed much since 1994.
Now consider Ghaziuddin (2005) which examined 58 subjects with AS, and found that they had a total of 64 siblings. This gives a proportion of 2.1 children per family with at least one AS child.
See also Deykin et al (1979) which is a somewhat older paper. It finds a proportion of 3.18 children per family with at least 2 children. Based on the 1980 data found in the 1994 Census report, I estimate the proportion of children in households with at least 2 children at 2.81 in 1980.
Piven et al (1990) results in a proportion of 2.81 children per family, but it is unclear if the sample is complete, and the authors only looked at adult autistics and their siblings.
By themselves, these findings are not statistically significant, but taken together they might be indicative of a trend. Unfortunately, no studies have looked at this question specifically, so there is little to go on. (I have contacted some authors of such studies but they have not been able to provide any additional useful info so far).
There are some additional caveats:
- Census data looks at all children under 18. Autism studies look at children diagnosed with autism, which rules out very young children, but at the same time could result in a youth bias. Research specifically looking into this question would find control probands matched by age.
- The autistic child in a family likely reduces the 2nd generation reproduction rates in that family. This is not relevant to the analysis, however, as the number of families with autistic alleles is probably much larger than the number of families that have an autistic child.
- There could easily be sociological reasons to explain a higher number of children in families where one child has been diagnosed with a psychiatric disorder. But this could also work in the opposite direction.
- It is not necessary for non-autistics with autistic alleles to have a preference to procreate. It might be sufficient that they form families more often.
Autism alleles are likely advantageous by themselves, but reproductively disadvantageous when combined. This results in a frequency equilibrium for each of the alleles. The frequency of most autism alleles is likely high and relatively stable. Most autism alleles should thus not be described as "disease genes". This has some implications on how researchers should look for autism loci.
There is an open question as to whether autistic families reproduce more often than average. Additionally, research on the reproduction rates of individuals with specific alleles could be carried out.
Finally, this analysis shows that eugenics is pragmatically misguided. If combinations of alleles that have a high likelihood of resulting in autism are identified, and parents abort pregnancies based on this information, it is likely that the frequency of all autism alleles will increase (as parents may simply attempt to procreate again).