A while back, Steve Sailer was discussing how much of IQ research is hollow. The usage of “hollow,” here, alludes to Jensen’s use of the term with regards to IQ tests. By Jensen’s usage, score gains that don’t involve general intelligence gains are said to be hollow because they lack predictive validity. In a parallel manner, we can say that IQ research that doesn’t bother to determine if score differences are g-loaded or are predictive are likewise hollow — hollow differences, hollow research. Anyways, Steve’s point was that while there is a good amount of research showing that various interventions can increase scores, if not lastingly, virtually none examines whether this increase has any real world importance. The hollowness of this research is, no doubt, driven by the egalitarian desire of many researchers to show that interventions or education or training possibly could make an impact. Others have pointed to this hole in the research. As example, Dutch researcher Jan te Nijenhuis and Co. noted:
Ceci (1991) showed that increased schooling leads to higher IQ scores, but are these gains highly specific or predominantly generalizable? It would be interesting to apply the techniques we used in this study to the findings from previous intervention studies. It may be that biological interventions (such as diet, vitamin supplements, vaccination against infectious disease) rather than psychological or educational interventions, are the most cost-effective method of producing true changes in g and broad abilities. It may be that there is a biological barrier between the first stratum and the second stratum that restricts the effects of behavioral interventions to narrow abilities and testspecificities. (Nijenhuis et al, 2006. Score gains on g-loaded tests: No g)
The hollowness doesn’t stop with interventional research. Nor with research on IQ, per se. No one has bothered, for example, to determine the predictive validity of scores variance attributable to genes, shared environment, or unshared environment in IQ or in g. One could imagine that the predictive validity of IQ or even general intelligence is genetic-loaded. This seems to be a non-trivially important topic to look into. And it’s not intuitively obvious — to me, at any rate — that all IQ variance would have the same predictive importance.
A recent study gives us a hint of what could be the case:
A large body of research has revealed that measures of IQ are highly predictive of a wide swath of life outcomes. However, most studies examining these associations have employed correlational statistical techniques which tend to confound environmental and genetic influences. The current study addresses this gap in the literature by making use of a monozygotic twin difference scores approach to explore the association between IQ and a variety of outcome measures, including general health, substance use, relationships, sexual behaviors, educational attainment, economic well-being, and criminal justice contacts. Analysis of monozygotic twin pairs from the National Longitudinal Study of Adolescent Health (Add Health) revealed that between-twin differences in IQ have little effect on the majority of the examined outcome measures. The implications of these findings and suggestions for future research are discussed (Exploring the association between IQ and differential life outcomes: results from a longitudinal sample of monozygotic twin
As possible accounts for the seemingly anomalous findings, the authors speculate:
Given that IQ is a significant predictor of life outcomes across a wide range of heterogeneous studies, we are left to speculate why our findings diverge from the existing literature. While not an exhaustive list, we do offer two possible reasons. First, it is possible that while IQ is integral to explaining differences across unrelated persons, perhaps IQ is not as important at explaining differences between siblings/twins. Second, and relatedly, the methodology employed in the current study is highly conservative in that the MZ difference score approach tends to reduce the amount of variation that is left to be explained.
The author’s second conjecture is plausible but not there first, as it has already been shown that sibling differences are predictively important. My conjecture would be that non-genetic variance — which includes variance due to error — is less associated with outcome differences than genetic variance. But I don’t expect this issue to be thoroughly investigated any time soon.