Intelligence New Findings and Theoretical Developments

Richard E. Nisbett University of Michigan Joshua Aronson and Clancy Blair New York University

William Dickens Northeastern University James Flynn University of Otago

Diane F. Halpern Claremont McKenna College Eric Turkheimer University of Virginia

We review new findings and new theoretical developments in the field of intelligence. New findings include the follow- ing: (a) Heritability of IQ varies significantly by social class. (b) Almost no genetic polymorphisms have been discovered that are consistently associated with variation in IQ in the normal range. (c) Much has been learned about the biological underpinnings of intelligence. (d) “Crystallized” and “fluid” IQ are quite different aspects of intelligence at both the behavioral and biological levels. (e) The importance of the environment for IQ is established by the 12-point to 18-point increase in IQ when children are adopted from working-class to middle-class homes. (f) Even when improvements in IQ produced by the most effective early childhood interventions fail to persist, there can be very marked effects on academic achievement and life outcomes. (g) In most developed countries studied, gains on IQ tests have continued, and they are beginning in the developing world. (h) Sex differences in aspects of intelligence are due partly to identifiable biological factors and partly to socialization factors. (i) The IQ gap between Blacks and Whites has been reduced by 0.33 SD in recent years. We report theorizing concerning (a) the relationship between working memory and intelligence, (b) the appar- ent contradiction between strong heritability effects on IQ and strong secular effects on IQ, (c) whether a general intelligence factor could arise from initially largely inde- pendent cognitive skills, (d) the relation between self-reg- ulation and cognitive skills, and (e) the effects of stress on intelligence.

Keywords: intelligence, fluid and crystallized intelligence, environmental and genetic influences, heritability, race and sex differences

In 1994, a controversial book about intelligence byRichard Herrnstein and Charles Murray called The BellCurve was published. The book argued that IQ tests are an accurate measure of intelligence; that IQ is a strong predictor of school and career achievement; that IQ is highly heritable; that IQ is little influenced by environmen- tal factors; that racial differences in IQ are likely due at least in part, and perhaps in large part, to genetics; that environmental effects of all kinds have only a modest effect on IQ; and that educational and other interventions have

little impact on IQ and little effect on racial differences in IQ. The authors were skeptical about the ability of public policy initiatives to have much impact on IQ or IQ-related outcomes.

The Bell Curve sold more than 300,000 copies and was given enormous attention by the press, which was largely uncritical of the methods and conclusions of the book. The Science Directorate of the American Psycholog- ical Association felt it was important to present the con- sensus of intelligence experts on the issues raised by the book, and to that end a group of experts representing a wide range of views was commissioned to produce a summary of facts that were widely agreed upon in the field and a survey of what the experts felt were important questions requiring further research. The leader of the group was Ulrich Neis- ser, and the article that was produced was critical of The Bell Curve in some important respects (Neisser et al., 1996). The article was also an excellent summary of what the great majority of experts believed to be the facts about intelligence at the time and important future directions for research.

Fifteen years after publication of the review by Neis- ser and colleagues (1996), a great many important new facts about intelligence have been discovered. It is our intent in this review to update the Neisser et al. article (which remains in many ways a good summary of the field

This article was published Online First January 2, 2012. Richard E. Nisbett, Institute for Social Research, University of Mich-

igan; Joshua Aronson and Clancy Blair, Department of Applied Psychol- ogy, New York University; William Dickens, Department of Economics, Northeastern University; James Flynn, Department of Psychology, Uni- versity of Otago, Dunedin, New Zealand; Diane F. Halpern, Department of Psychology, Claremont McKenna College; Eric Turkheimer, Depart- ment of Psychology, University of Virginia.

The writing of this article and much of the research that went into it were supported by a generous grant from the Russell Sage Foundation, by National Institute on Aging Grant 1 R01 AG029509-01A2, and by Na- tional Science Foundation Grant 2007: BCS 0717982. The views pre- sented here are not necessarily those of the National Science Foundation.

We thank Angela Duckworth, Richard Haier, Susanne Jaeggi, John Jonides, Scott Kaufman, John Protzko, Carl Shulman, Robert Sternberg, and Oscar Ybarra for their critiques of an earlier version of this article.

Correspondence concerning this article should be addressed to Rich- ard E. Nisbett, Institute for Social Research, 3229 East Hall, University of Michigan, Ann Arbor, MI 48109. E-mail: nisbett@umich.edu

THIS ARTICLE HAS BEEN CORRECTED. SEE LAST PAGE

Vol. 67, No. 2, 130–159 DOI: 10.1037/a0026699

Th is

d oc

um en

t i s c

op yr

ig ht

ed b

y th

e A

m er

ic an

P sy

ch ol

og ic

al A

ss oc

ia tio

n or

o ne

o f i

ts a

lli ed

p ub

lis he

rs .

Th is

a rti

cl e

is in

te nd

ed so

le ly

fo r t

he p

er so

na l u

se o

f t he

in di

vi du

al u

se r a

nd is

n ot

to b

e di

ss em

in at

ed b

ro ad

ly .

of intelligence). There are three chief respects in which our review differs importantly from that of Neisser and col- leagues: (a) Due in part to imaging techniques, a great deal is now known about the biology of intelligence. (b) Much more is known about the effects of environment on intel- ligence, and a great deal of that knowledge points toward assigning a larger role to the environment than did Neisser and colleagues and toward a more optimistic attitude about intervention possibilities. (c) More is now known about the effects of genes on intelligence and on the interaction of genes and the environment. Our article also presents a wide range of new theoretical questions and reviews some at- tempted solutions to those questions. We do not claim to represent the full range of views about intelligence. We do maintain, however, that few of the findings we report have been widely contradicted. Where we are aware of contro- versy, we provide sources where readers can be exposed to alternative views. We acknowledge that the theoretical questions we raise might not be the ones that every expert would agree are the most important ones, and we recognize that not every expert will agree with our views on these questions. We have referenced alternative views where we are aware that such exist.

The article is organized under the rubrics of genes and the environment, new knowledge about the effects of the environment, new knowledge about interventions, the bi- ology of intelligence, group differences in IQ, and impor- tant unresolved issues. Our working definition of intelli- gence is essentially that offered by Linda Gottfredson (1997):

[Intelligence] . . . involves the ability to reason, plan, solve prob- lems, think abstractly, comprehend complex ideas, learn quickly and learn from experience. It is not merely book learning, a narrow academic skill, or test-taking smarts. Rather it reflects a broader and deeper capability for comprehending our surround- ings—“catching on,” “making sense” of things, or “figuring out” what to do. (p. 13)

The measurement of intelligence is one of psycholo- gy’s greatest achievements and one of its most controver- sial. Critics complain that no single test can capture the complexity of human intelligence, all measurement is im- perfect, no single measure is completely free from cultural bias, and there is the potential for misuse of scores on tests of intelligence. There is some merit to all these criticisms. But we would counter that the measurement of intelli- gence—which has been done primarily by IQ tests—has utilitarian value because it is a reasonably good predictor of grades at school, performance at work, and many other aspects of success in life (Gottfredson, 2004; Herrnstein & Murray, 1994). For example, students who score high on tests such as the SAT and the ACT, which correlate highly with IQ measures (Detterman & Daniel, 1989), tend to perform better in school than those who score lower (Coyle & Pillow, 2008). Similarly, people in professional careers, such as attorneys, accountants, and physicians, tend to have high IQs. Even within very narrowly defined jobs and on very narrowly defined tasks, those with higher IQs outper- form those with lower IQs on average, with the effects of

IQ being largest for those occupations and tasks that are most demanding of cognitive skills (F. L. Schmidt & Hunter, 1998, 2004). It is important to remain vigilant for misuse of scores on tests of intelligence or any other psychological assessment and to look for possible biases in any measure, but intelligence test scores remain useful when applied in a thoughtful and transparent manner.

IQ is also important because some group differences are large and predictive of performance in many domains. Much evidence indicates that it would be difficult to over- come racial disadvantage if IQ differences could not be ameliorated. IQ tests help us to track the changes in intel- ligence of different groups and of entire nations and to measure the impact of interventions intended to improve intelligence.

Types of intelligence other than the analytic kind examined by IQ tests certainly have a reality. Robert Stern- berg and his colleagues (Sternberg, 1999, 2006) have stud- ied practical intelligence, which they define as the ability to solve concrete problems in real life that require search- ing for information not necessarily contained in a problem statement, and for which many solutions are possible, as well as creativity, or the ability to come up with novel solutions to problems and to originate interesting questions. Sternberg and his colleagues maintain that both practical intelligence and creativity can be measured, that they cor- relate only moderately with analytic intelligence as mea- sured by IQ tests, and that they can predict significant amounts of variance in academic and occupational achieve- ment over and above what can be predicted by IQ measures alone. Early claims by Sternberg were disputed by other intelligence researchers (Brody, 2003; Gottfredson, 2003a, 2003b). Subsequent work by Sternberg (2006, 2007) im- proved on the original evidence base and showed that measuring nonanalytic aspects of intelligence could signif- icantly improve the predictive power of intelligence tests. See also Hunt (2011) and Willis, Dumont and Kaufman (2011).

The chief measure that we focus on in this article is IQ, because it is for that measure that the bulk of evidence pertinent to intelligence exists. When relevant, we distin- guish between IQ and g—often identified as the first factor extracted in a factor analysis of IQ subtests and with which all IQ subtests correlate. Many intelligence investigators place great importance on the difference between g and IQ on the grounds that g tends to predict some academic and life outcomes and group differences better than does IQ and because it correlates with some biological measures better than does IQ (Jensen, 1998). We do not in general share the view that g is importantly different from IQ, and we do not interpret correlations between g and life outcomes, group differences, and biological measures as having the same import as do many other investigators. Our differences with those scientists will be noted at several points in this article.

An important distinction commonly made in the liter- ature is between crystallized intelligence, g(C), or the in- dividual’s store of knowledge about the nature of the world and learned operations such as arithmetical ones which can be drawn on in solving problems, and fluid intelligence,

131February–March 2012 ● American Psychologist

Th is

d oc

um en

t i s c

op yr

ig ht

ed b

y th

e A

m er

ic an

P sy

ch ol

og ic

al A

ss oc

ia tio

n or

o ne

o f i

ts a

lli ed

p ub

lis he

rs .

Th is

a rti

cl e

is in

te nd

ed so

le ly

fo r t

he p

er so

na l u

se o

f t he

in di

vi du

al u

se r a

nd is

n ot

to b

e di

ss em

in at

ed b

ro ad

ly .

g(F), which is the ability to solve novel problems that depend relatively little on stored knowledge as well as the ability to learn. A test that is often considered the best available measure of g(F) is Raven’s Progressive Matrices. This test requires examination of a matrix of geometric figures that differ from one another according to a rule to be identified by the individual being tested. This rule is then used to generate an answer to a question about what new geometrical figure would satisfy the rule. Some of the recent advances in intelligence research, particularly those in the area of the neurobiology of intelligence, have tended to strongly support the distinction between g(F) and g(C) (Blair, 2006; Horn & McArdle, 2007). Although much research on intelligence continues to emphasize a single dimension, whether Full Scale IQ or factor analytically derived g, research on the neural basis for intelligence and also on the development of intelligence suggests that IQ is best represented by at least two dimensions rather than one.1 We refer the reader to a volume edited by Sternberg and Grigorenko (2002) on the topic of g in which propo- nents and opponents debated the existence and utility of the g construct.

Genes and the Environment

When the Neisser et al. (1996) article appeared, the con- troversy over whether genes influence intelligence was mainly in the past. That controversy has faded still further in the intervening years, as scientists have learned that not only intelligence but practically every aspect of behavior on which human beings differ is heritable to some extent. Several strands of evidence, however, suggest that the effects of genes on intelligence, though undeniable, are not nearly as determinative as hereditarians might have hoped or as environmentalists feared 25 years ago.

Heritability is the proportion of variability in a phe- notype that is “accounted for” (in the usual regression sense) by variation in genotype. Most studies estimate that the heritability of IQ is somewhere between .4 and .8 (and generally less for children), but it really makes no sense to talk about a single value for the heritability of intelligence. The heritability of a trait depends on the relative variances of the predictors, in this case genotype and environment. The concept of heritability has its origins in animal breed- ing, where variation in genotype and environment is under the control of the experimenter, and under these conditions the concept has some real-world applications. In free- ranging humans, however, variability is uncontrolled, there is no “true” degree of variation to estimate, and heritability can take practically any value for any trait depending on the relative variability of genetic endowment and environment in the population being studied. In any naturally occurring population, the heritability of intelligence is not zero (if genotype varies at all, it will be reflected in IQ scores) and it is not one (if environment varies at all, it will be reflected in IQ scores). That the heritability of intelligence is be- tween zero and one has one important consequence: With- out additional evidence, correlations between biologically related parents and children cannot be unambiguously in-

terpreted as either genetic or (as is more frequently at- tempted) environmental.

Social Class and Heritability of Cognitive Ability One example of the population dependence of heritability is the claim that the heritability of intelligence test scores differs as a function of age, with heritability increasing over the course of development (Plomin, DeFries, & Mc- Clearn, 1990). Another example of population dependence is that the heritability of intelligence test scores is appar- ently not constant across different races or socioeconomic classes. Sandra Scarr (Scarr-Salapatek, 1971) published a report of twins in the Philadelphia school system showing that the heritability of aptitude and achievement test scores was higher for White children than Black children and for twins raised in relatively richer homes than for twins raised in poorer ones. This report, although it received some positive attention at the time, also faced serious criticism (Eaves & Jinks, 1972).

Aside from a partial replication conducted by Scarr (1981) and a report from Sweden by Fischbein (1980), the hypothesis of group differences in heritability lay mostly fallow for a 20-year period encompassing the time when the Neisser et al. (1996) article was written. Interest in the phenomenon was rekindled in 1999 when Rowe, Jacobson, and Van den Oord (1999) analyzed data from the National Longitudinal Study of Adolescent Health, a large, repre- sentative sample of American youth, then in early adoles- cence, who were administered a version of the Peabody Picture Vocabulary Test. Rowe et al.’s analysis showed that most of the variance in families with poorly educated mothers was explained by the shared environment. (Shared environment constitutes the environment that is shared by siblings in the same family and that differs from one family to another, as opposed to aspects of the environment that can differ among siblings, such as being a firstborn versus a later-born child.) Most of the variance for children from well-educated families was explained by genes.

Turkheimer, Haley, Waldron, D’Onofrio, and Gottes- man (2003) conducted an analysis of socioeconomic status (SES) by heritability interactions in the National Collabor- ative Perinatal Project (NCPP). The NCPP is particularly well-suited for this purpose because it comprised a repre- sentative sample of twins, many of them raised in poverty. A well-validated measure of SES with good psychometric properties is available in the NCPP dataset (Myrianthopou- los & French, 1968). Structural equation modeling demon- strated large statistically significant interactions for Full Scale and Performance IQ (PIQ) but not for Verbal IQ (VIQ), although the effects for VIQ were in the same direction. For families at the lowest levels of SES, shared environment accounted for almost all of the variation in IQ, with genes accounting for practically none. As SES in-

1 Horn (1989) discussed several other dimensions of intelligence at the same level as g(F) and g(C). We do not doubt the reality of these other dimensions, but they have not been sufficiently well researched to justify considering them at length.

132 February–March 2012 ● American Psychologist

Th is

d oc

um en

t i s c

op yr

ig ht

ed b

y th

e A

m er

ic an

P sy

ch ol

og ic

al A

ss oc

ia tio

n or

o ne

o f i

ts a

lli ed

p ub

lis he

rs .

Th is

a rti

cl e

is in

te nd

ed so

le ly

fo r t

he p

er so

na l u

se o

f t he

in di

vi du

al u

se r a

nd is

n ot

to b

e di

ss em

in at

ed b

ro ad

ly .

creased, the contribution of shared environment diminished and the contribution of genes increased, crossing in lower middle-class families. Finally, in the most socioeconomi- cally advantaged families (who were not wealthy), practi- cally all of the variation in IQ was accounted for by genes, and almost none was accounted for by shared environment.

There have been several attempted replications of the SES by heritability interaction since the Turkheimer et al. (2003) study was published, with mixed results. (A study by Nagoshi and Johnson [2005] has been cited as a failure to replicate the SES interaction [Rushton & Jensen, 2010], but it is not. Nagoshi and Johnson used the Hawaii Family Study to demonstrate that there is no change in parent– child correlations for intelligence test scores as a function of family income. Parent–child correlations, however, are a combination of genetic and shared environmental factors. Every instance of the interaction that has been reported to date has shown that genetic and shared environmental components change in opposite directions as a function of SES, so one would expect them to cancel each other out in parent–child correlations.)

Harden, Turkheimer, and Loehlin (2007) reanalyzed the Loehlin and Nichols (1976) National Merit Scholar Qualifying Test (NMSQT) twin data. This sample of 839 twin pairs was drawn from the population of nearly 600,000 adolescents who completed the NMSQT in 1962. Around 40% of the variance was accounted for by both additive genetics and the shared environment in the fami- lies with the lowest incomes and levels of education, whereas in the richer and better-educated families, slightly more than 50% of the variance was accounted for by genetics and about 30% was accounted for by the shared environment. A puzzling aspect of this study is that few of the children in the sample were actually poor, as they had been selected for participation in the NMSQT.

A recent study of 750 pairs of twins in the Early Childhood Longitudinal Study, Birth Cohort (Tucker- Drob, Rhemtulla, Harden, Turkheimer, & Fask, 2011) identified a gene by SES interaction in the emergence of genetic variance in early childhood. For the Bayley Scales of Infant Development administered at 10 months, there was no heritable variation at any level of SES. But when the test was repeated at two years of age, significant genetic variance emerged, with more genetic variance at higher levels of parental SES.

There are mixed reports about SES by heritability interactions for adults. Kremen, Jacobson, and Xian (2005) identified a significant interaction between parental educa- tion and word recognition scores on the Wide Range Achievement Test among 690 adult twins in the Vietnam Era Twin Registry. Descriptive analyses showed that for the twins with the least educated parents, additive genetics and shared environment each accounted for 36% of the variability in reading scores. For the twins with the best educated parents, additive genetics accounted for 56% of the variability and shared environment for 12%. The effect appeared to depend solely on decreases in shared environ- mental variation as a function of parental education. The Kremen et al. study is the only one to date that has docu-

mented the SES by heritability interaction in adults. In contrast, a study of the cognitive ability test given at the time of military induction to a Vietnam Era Twin Registry sample that was part of a larger sample of 3,203 pairs (which included the pairs included in Kremen et al., 2005) showed no interaction with parental education (Grant et al., 2010).

There are also mixed results for European studies. As noted earlier, Fischbein (1980) found the typical SES by heritability interaction with 12-year-olds in Sweden. More recently, Asbury, Wachs, and Plomin (2005) studied a sample of 4,446 four-year-old British twins from the Twins Early Development Study in London. The only interaction they found was an interaction in the opposite direction (i.e., children in high-risk impoverished environments showed higher heritabilities). However, their measure of intelli- gence was based on a child ability scale administered to parents over the telephone. A second study (Docherty, Kovas, & Plomin, 2011), conducted using 1,800 of the children from the Asbury et al. study at age 10, examined interactions between environmental variables and a set of 10 single nucleotide polymorphisms (SNPs) identified in earlier studies in the prediction of mathematics ability measured over the Internet. The set of SNPs had a main effect on mathematics ability, accounting for 2.7% of the variance, and showed a variety of small interactions with environmental measures, mostly in the direction of larger genetic effects in negative and chaotic environments. A new analysis of the same dataset extending the analysis through age 14 (Hanscombe et al., 2012) has found signif- icant interactions in the original direction for the environ- mental term. The investigators concluded that shared envi- ronmental experiences have greater impact on intelligence in low-SES families.

Van der Sluis, Willemsen, de Geus, Boomsma, and Posthuma (2008) examined Gene � Environment interac- tions for cognitive ability in 548 adult twins and 207 of their siblings from older and younger cohorts of the Neth- erlands Twin Registry. Mostly small and nonsignificant interactions were observed between biometric components of IQ and a variety of environmental variables, including neighborhood, income, and parental and spousal education levels. There was a substantial interaction between parental education and shared environmental effects, in the opposite direction from the original findings: Older males from more highly educated families showed larger shared environ- mental effects.

A recent study suggests that the most commonly found interaction may operate on the level of the cerebral cortex. Chiang et al. (2011) assessed cortical white matter integrity with diffusion tensor imaging in 705 twins and their siblings. They demonstrated that white matter integ- rity was highly heritable and, moreover, that there were significant Gene � Environment interactions, such that heritabilities were higher among twins with higher SES and higher IQ scores.

In summary, it appears reasonable to conclude that the heritability of cognitive ability is attenuated among impov- erished children and young adults in the United States.

133February–March 2012 ● American Psychologist

Th is

d oc

um en

t i s c

op yr

ig ht

ed b

y th

e A

m er

ic an

P sy

ch ol

og ic

al A

ss oc

ia tio

n or

o ne

o f i

ts a

lli ed

p ub

lis he

rs .

Th is

a rti

cl e

is in

te nd

ed so

le ly

fo r t

he p

er so

na l u

se o

f t he

in di

vi du

al u

se r a

nd is

n ot

to b

e di

ss em

in at

ed b

ro ad

ly .

There is mixed evidence as to whether the effect occurs in other countries, and there is contradictory evidence about whether the effect persists into adulthood. One can only speculate about why these differences among studies might occur. It appears, for example, that socioeconomic differ- ences in intelligence are not as pronounced in modern Europe as they are in the United States. In the Turkheimer et al. (2003) study, the correlation of SES with IQ was �.70; in Asbury et al. (2005), it was about �.2. In van der Sluis et al. (2008), there was only a 5-point IQ difference between the groups of twins with the most and least edu- cated parents. Studies of adults must contend with the low magnitude of shared environmental components overall, and it may be difficult to detect interactions when there are low shared environment effects to work with.

One interpretation of the finding that heritability of IQ is very low for lower SES individuals is that children in poverty do not get to develop their full genetic potential. If true, there is room for interventions with that group to have large effects on IQ. That this interpretation of the finding is correct is indicated by an actual intervention study (Turkheimer, Blair, Sojourner, Protzko, & Horn, 2012). The Infant Health and Development Program (IHDP) was a broad-based intervention program designed to improve the cognitive function and school performance of approx- imately 1,000 low birth weight infants. There were 95 pairs of identical and fraternal twins who were incidentally in- cluded in the program. About a third of the pairs were randomly assigned to the experimental treatment group. Their families were assigned to receive weekly home visits from a clinically trained staff person who delivered a curriculum of child development and parenting education, a program of mental health counseling and support, and referral to social services available in the community. From ages 1 to 3, the treatment-group twin pairs were eligible to attend a free, full-day, high-quality child development cen- ter established and run by IHDP staff. The other twin pairs did not receive home visits or access to the child develop- ment centers. At 96 months, the children were administered a battery of standardized tests of cognitive ability and school achievement, including the Wechsler Intelligence Scale for Children, the Woodcock-Johnson Achievement Scales, and Raven’s Progressive Matrices. On the basis of studies of natural variation in SES, it was hypothesized that the heritability of intelligence would be higher in the group randomly selected for exposure to the enriched environ- ment. Heritabilities were significantly greater than 0 in the intervention group on seven of eight tests and were higher in the intervention group than in the control group on all seven.

The underrepresentation of low-SES individuals in behavioral genetic studies causes a problem for studies of heritability. The estimates of heritability just discussed are based on studies of twins. Another way to estimate the relative importance of genes and environment is to com- pare the correlation between adopted children’s IQ and that of their birth parents with the correlation between adopted children’s IQ and that of their adoptive parents. The former correlation is generally higher than the latter, sometimes

much higher. Many have concluded on the basis of such findings that environments are relatively unimportant in determining IQ, since variations in the environments of adoptive families are not very highly associated with vari- ation in children’s IQ. But work by Stoolmiller (1999) shows that estimates of the relative contributions of genes and environment may be very sensitive to the inclusion of disadvantaged populations in a given study. Adoption stud- ies may tend to underestimate the role of environment and overstate the role of genetics due to the restricted social class range of adoptive homes. Adoptive families are gen- erally of relatively high SES. Moreover, observation of family settings by the HOME technique (Home Observa- tion for Measurement of the Environment; Bradley et al., 1993; Phillips, Brooks-Gunn, Duncan, Klebanov, & Crane, 1998) shows that the environments of adoptive families are much more supportive of intellectual growth than are those of nonadoptive families. The restriction of range (as much as 70% in some studies; Stoolmiller, 1999) means that the possible magnitude of correlations between adoptive par- ents’ IQ and that of their children is curtailed.

Stoolmiller’s (1999) conclusions have been ques- tioned by Loehlin and Horn (2000) and McGue et al. (2007). Loehlin and Horn pointed to a number of implica- tions of Stoolmiller’s analysis that are inconsistent with facts Stoolmiller did not consider, but they provided no formal tests of the significance of these inconsistencies, so it is impossible to know how important these inconsisten- cies are. McGue et al. showed that regressions of IQ on a few characteristics that were restricted in range in their data show those variables to have no effect. But the variables they examined are certainly not the only ones with notably restricted range relative to the whole population, many of which are not represented in their data. Perhaps more important, McGue and colleagues were looking at the effects of restriction of range on a sample whose range was already restricted with respect to the population from which it was drawn. Their sample consisted of intact families of relatively high SES.

The findings on the interaction between SES and genetic influences described above suggest that the prob- lem goes beyond the concerns about restrictions of range addressed by Stoolmiller (1999). If there is heterogeneity in the importance of environment in different social groups, then exclusion of participants from poor socioeconomic backgrounds can have a much more profound effect than what might appear based on the reduction of variance in SES alone. Since environment has a larger impact on outcomes among lower SES individuals, then removing them from the sample not only reduces the variance of environment but also reduces the average impact of envi- ronment on outcomes in the sample, thereby causing a reduction in the measured role of environment for two separate reasons. Thus, in addition to the bias introduced into the estimates of environmental effects by the restric- tion of range in SES, excluding participants from the low- est SES levels would also bias the results by omitting the portion of the distribution for which environmental effects