Note: My discussion here is overly repetitious and needs some editing — but I am sure that you can make sense of it.
In Re-analysis of Jensen’s study of Capron and Duyme adoption data, Meng Hu investigated Jensen’s and Nisbett’s competing claims concerning the relationship between the adoption effect on IQ and g-loadings in the much cited Capron & Duyme adoption study. The central question involved is whether or not differences in shared environment tend to induce differences in general intelligence. Jensen investigated this issue by applying his method of correlated vectors (MCV) to Capron and Duyme’s data to determine if the adoption effect exhibited a Jensen Effect, as one would expect were the adoption induced IQ differences in fact differences in general intelligence.
As I noted in a recent post, Shared Environment and Jensen Effects, this issue, specifically whether or not shared environmental effects tend to induce Jensen Effects, is much deserving of investigation. It has frequently been claimed, when not assumed, that shared environment effects will not produce a positive correlation between g-loadings and the magnitude of subtest differences. Indeed, the eminent psychologist Earl Hunt in his chapter, “The Demography of Intelligence” recently made this argument in defense of racial genetic differences: (a) Race differences exhibit a Jensen Effect; (b) this can’t be explained by Shared environment e.g., SES.
The problem is that the second premise has not been established. It has been shown that heritability indexes g-loadings — that the more heritable subtests are the more g-loaded subtests. This logically implies that (1-heritability) or environmentality indexes the reverse. Which logically implies that environmental effects, in total, will induce anti-Jensen Effects. But this doesn’t answer the central question. It’s easy to demonstrate that unshared environment does induce anti-Jensen Effects. To given an illustration of this using some of Meng Hu’s data:
Here, I correlated the Black-White average g-loadings (because that’s what was available in Meng Hu’s data) with the change in the Black and White sibling correlations relative to the lowest correlation. Since shared environment is the same the change in correlations can only be due to genes versus unshared environment, with the former increasing the correlation and the latter decreasing it. Since there is a positive Jensen Effect on the correlations — higher correlation, more g-loaded — unshared environment must exhibit an anti-Jensen Effect.
So environmental effects on average produce an anti-Jensen Effects and unshared environmental effects do also, but what about shared environmental effects in particular? In the previous post, I looked at Luo et al. (1994) and found a modest positive association between shared environment and g-loadings. Ideally, one would repeat this for all studies which gave estimates of g-loadings and shared environment. Jensen (1998) pg. 184 reports 4 studies from which the correlations between heritability and g-loading can be ascertained. Of these only one — or rather a series of papers based on the same sample — provided both g-loadings and shared environmental estimates i.e., Plomin 1994; Pedersen et al., 1992; Pedersen et al., 1994).
Across the three papers, two different samples, containing three estimates, were provided. The first, in Plomin et al. (1994), was a longitudinal sample with MZA, MZT, and DZT. The g-loadings and variance components are shown below:
After correcting the g-loadings for unreliability, I correlated them with the reported variance components. The authors were unclear with regards to which “Digit Span” and “Names and Faces” tests the variance estimates related to i.e., Backwards versus Forwards and Delayed versus Immediate. I assumed the former in both cases, but using Forward instead of Backwards, Immediate instead of Delayed g-loadings or just removing these sub-tests didn’t change the overall results, which are shown below:
The second sample, contained in Pedersen et al. (1992) was cross sectional. The g-loadings and variance components are shown below:
I summarized these results below. The Jensen Effects for the two time periods for the first sample were averaged. And then the Jensen Effects for the two samples were averaged.
As with the prior analysis, heritability showed a Jensen Effect and environmentality showed a corresponding anti-Jensen Effect. But the picture became more complex when broken down by environmental components. Unshared environment, as we would expect, showed an anti-Jensen Effect; but the same could not be said for shared environment. This analysis, though, is ambiguous because the variance explained by shared environment was frequently zero. Nonetheless, in two out of two analyses I was unable to find a non positive Jensen Effect on shared environment. To this, when I am feeling less unwell, I will later add a hitherto undiscussed analysis of the NLSY97 unrelated sibling data, which shows, again, a Jensen Effect on shared Environment — a finding which calls into question one of the strongest arguments for the case that racial and ethnic differences are genetically conditioned.
Plomin et al., 1994. Variability and Stability in Cognitive Abilities Are Largely Genetic Later in Life