A frequent claim in the IQ debates is that which family you are raised in has no lasting impact on your IQ. Jensen argues in The g Factor that the only causes of IQ similarities between adult identical twins are genetic. Many researchers go so far as to argue that by 12 years of age, the shared environment has no impact.
Based on my limited knowledge of the behavior genetic research, I used to hold this position as well. But thanks to some recent in depth reading, I have come to the conclusion that which family you are raised in matters significantly for your IQ as an adult, especially so for people of lower socioeconomic status. I’ll detail the behavior genetic evidence here, and argue that it points to significant shared environmental influences on adult IQ scores.
The most recent and comprehensive survey of twin studies on IQ comes from Haworth et al (2009). Using pooled twin data from around the world, they modeled genetic and environmental influences as a function of age. Here is what they found regarding the effects of the shared environment:
[S]hared environment shows a decrease from childhood (33%) to adolescence (18%) but remained at that modest level in young adulthood (16%).
In an email exchange with Dr. McGue (one of the co-authors of the paper) he told me that while the latest data may not fit with earlier estimates, it’s actually more reliable due to the unprecedented sample size (11,000 pairs of twins).
One failing of this study, though, is that it doesn’t go far enough into adulthood. The young adult group ranges from 14 to 34 years of age, with an average age of 17. In contrast, McGue (1993) looked seperately at data on adults over 20 years of age. He found that the shared environment diminished to zero impact at that point. Here’s his chart:
Looking at that chart, you might quickly conclude that shared environmental influence evaporates by age 20. However, this conclusion is premature. Twin studies make a great number of assumptions, some of which increase and others of which decrease estimates of the shared environment. A straightforward way of bypassing these assumptions is to compare monozygotic twins reared apart (MZAs) to monozygotic twins reared together (MZTs). The following data comes from a comparison of MZTs and MZAs, of average age 41, in Bouchard (1990):
|WAIS IQ-Full Scale
Differences between MZA’s and MZT’s on Raven’s Progressive Matrices follow the same pattern but are even more extreme. Bouchard (1981) reported a median correlation of only 0.58 for adult MZA’s on the Raven’s. Curiously, though, MZA’s are equally if not more correlated than MZT’s on the Mill-Hill vocabulary test. Apparently, the pattern is that more g-loaded tests tend to show stronger evidence of lasting shared environmental impact.
It’s worth noting that MZT vs. MZA comparisons are actually biased towards an underestimation of shared environmental impact. Bouchard’s study of twins reared apart found an environmental correlation of .22 for MZAs on various environmental measures, with some having a small but significant correlation with IQ scores. Also MZA’s share the womb. To summarize: when the assumptions of the twin method are effectively controlled for, lasting shared environmental impacts are revealed.
To date, most adoption studies of IQ have concluded that being adopted by a new and typically well-off family has no effect on adult IQ scores. Here is a chart of adoption studies from Bouchard (2009):As you can see by clicking it, the IQ correlation between unrelated individuals in the same family decreases (on average) from .26 in childhood to .04 in adulthood (which begins at age 17 for the purposes of this graph).
However, as with the previous chart, the quick conclusion that shared environmental influences don’t matter in adulthood shouldn’t be so quickly accepted. To begin with, we can see that the adoption data underestimates the shared environment relative to the twin literature. This most likely occurs because of the assumptions that go into adoption studies.
Stoolmiller (1999), for example, highlighted the issue of range restriction– the idea that the limited range of adoptee and adoptive family environments will lower estimates of the shared environment. This idea is supported by studies which make the extra effort to include individuals of lower SES. The French adoption studies that made such an effort buck the trendline seen above, in finding that nurture matters almost as much as nature for the IQ of 14 year olds. Scarr (1993) is the outlier in the adoption graph above, finding a .19 correlation between unrelated adolescent siblings. Perhaps her results differed from others because her sample was multi-racial and therefore less range restricted. Lastly, there are other lines of evidence supporting the idea of range restriction, such as Turkheimer’s work on SES and cognitive ability.
It’s worth noting, however, that McGue (2007) looked for evidence of range restriction effects within the “broad middle class” and did not find any. He used statistical methods that are over my head to estimate the effects of range restriction based on a range restricted sample and state census data. Unfortunately there are no studies which have critiqued his as of yet. Any commenters who are familiar with the statistics involved are invited to comment. Even if McGue is right about restriction of range, my point stands that assumptions inherent in the adoption studies deflate c^2 estimates.
Future work will help sort out the still unanswered question of shared environmental influences on adult IQ scores. There are large longitudinal adoption studies currently under way, and I believe that Haworth’s twin study will be followed-up on and include data on older twins. There are also interesting (albeit less methodologically agreed upon) studies coming out like this one, which find significant shared effects on IQ in adulthood.
My reading of the available evidence is that there is a significant shared environmental input to adult IQ, and that it is associated with socioeconomic status. To what extent it’s the neighborhood or the parents themselves that matters is unclear. Just as the most g-loaded tests show the most shared environmental effects in the MZA-MZT comparison, so too does the Flyn
n effect occur on the most g-loaded tests, suggesting that whatever is loading onto the “shared environment” within generations is also responsible for differences between them.
(Republished from GNXP.com
by permission of author or representative)