Intellectual Interest Mediates Gene · Socioeconomic Status Interaction on Adolescent Academic Achievement Elliot M. Tucker-Drob and K. Paige Harden University of Texas at Austin Recent studies have demonstrated that genetic influences on cognitive ability and academic achievement are larger for children raised in higher socioeconomic status (SES) homes. However, little work has been done to document the psychosocial processes that underlie this Gene · Environment interaction. One process may involve the conversion of intellectual interest into academic achievement. Analyses of data from 777 pairs of 17-year-old twins indicated that Gene · SES effects on achievement scores can be accounted for by stronger influences of genes for intellectual interest on achievement at higher levels of SES. These findings are consis- tent with the hypothesis that higher SES affords greater opportunity for children to seek out and benefit from learning experiences that are congruent with their genetically influenced intellectual interests. Genetic influences are often found to account for upward of 50% of individual differences in both general cognitive ability and academic achieve- ment at the population level. Although some authors have argued that this large figure renders environmental explanations for socioeconomic disparities in cognition implausible (Hernstein & Murray, 1994; Jensen, 1969, 1973), other develop- mental theorists have posited that genetic variance in cognitive ability and academic achievement may emerge, in part, via children’s transactions with particular environmental experiences (Bron- fenbrenner & Ceci, 1994; Dickens & Flynn, 2001). These transactional models of cognitive develop- ment represent an attempt to move beyond a focus on the relative magnitude of genetic versus environmental influences, and to move toward a more integrated understanding of how genes and environments combine and interact to produce complex behavioral phenotypes (Anastasi, 1958). In this article, we review the propositions of trans- actional models of cognitive development and dis- cuss how transactional models can provide a framework for understanding recent findings that genetic variance in cognitive outcomes is moder- ated by socioeconomic status (SES). Next, we sug- gest that noncognitive factors, such as motivation, self-concept, and interests, are ‘‘driving forces’’ in children’s transactions with their environments, and that a decoupling of children’s intellectual interest and academic achievement may account for the decreased genetic variance in academic achievement for children living in lower SES homes. Finally, we present evidence from new analyses of the National Merit Twin Study sup- porting our hypothesis about the role of intellec- tual interest in Gene · SES interaction on academic achievement. Transactional Models of Cognitive Development Plomin, DeFries, and Loehlin (1977) first described the processes by which genotypes could come to be differentially associated with environ- mental exposure, i.e., gene–environment correla- tion. Passive gene–environment correlations arise when children are raised by their biological par- ents, such that their rearing environments are influ- enced by some of the same genes that they inherit. Evocative gene–environment correlations arise when children’s genetically influenced traits, features, and characteristics elicit particular envi- ronments from others. Finally, active gene–environ- Data were obtained from the Henry A. Murray Research Archive at Harvard University (http://www.murray.harvard. edu/). The Population Research Center at the University of Texas at Austin is supported by a center grant from the National Institute of Child Health and Human Development (R24 HD042849). The original collectors of the data, the Murray Research Archive, and NICHD bear no responsibility for use of the data or for interpretations or inferences based upon such uses. We thank John Loehlin for helpful comments on previous versions of this article. Correspondence concerning this article should be addressed to Elliot M. Tucker-Drob, Department of Psychology, University of Texas at Austin, 1 University Station A8000, Austin, TX 78712- 0187. Electronic mail may be sent to [email protected]. Child Development, xxxxx 2012, Volume 00, Number 0, Pages 1–15 ȑ 2012 The Authors Child Development ȑ 2012 Society for Research in Child Development, Inc. All rights reserved. 0009-3920/2012/xxxx-xxxx DOI: 10.1111/j.1467-8624.2011.01721.x
Transcript
Intellectual Interest Mediates Gene · Socioeconomic Status Interaction on
Adolescent Academic Achievement
Elliot M. Tucker-Drob and K. Paige HardenUniversity of Texas at Austin
Recent studies have demonstrated that genetic influences on cognitive ability and academic achievement arelarger for children raised in higher socioeconomic status (SES) homes. However, little work has been done todocument the psychosocial processes that underlie this Gene · Environment interaction. One process mayinvolve the conversion of intellectual interest into academic achievement. Analyses of data from 777 pairs of17-year-old twins indicated that Gene · SES effects on achievement scores can be accounted for by strongerinfluences of genes for intellectual interest on achievement at higher levels of SES. These findings are consis-tent with the hypothesis that higher SES affords greater opportunity for children to seek out and benefit fromlearning experiences that are congruent with their genetically influenced intellectual interests.
Genetic influences are often found to account forupward of 50% of individual differences in bothgeneral cognitive ability and academic achieve-ment at the population level. Although someauthors have argued that this large figure rendersenvironmental explanations for socioeconomicdisparities in cognition implausible (Hernstein &Murray, 1994; Jensen, 1969, 1973), other develop-mental theorists have posited that genetic variancein cognitive ability and academic achievementmay emerge, in part, via children’s transactionswith particular environmental experiences (Bron-fenbrenner & Ceci, 1994; Dickens & Flynn, 2001).These transactional models of cognitive develop-ment represent an attempt to move beyond afocus on the relative magnitude of genetic versusenvironmental influences, and to move toward amore integrated understanding of how genes andenvironments combine and interact to producecomplex behavioral phenotypes (Anastasi, 1958).In this article, we review the propositions of trans-actional models of cognitive development and dis-cuss how transactional models can provide a
framework for understanding recent findings thatgenetic variance in cognitive outcomes is moder-ated by socioeconomic status (SES). Next, we sug-gest that noncognitive factors, such as motivation,self-concept, and interests, are ‘‘driving forces’’ inchildren’s transactions with their environments,and that a decoupling of children’s intellectualinterest and academic achievement may accountfor the decreased genetic variance in academicachievement for children living in lower SEShomes. Finally, we present evidence from newanalyses of the National Merit Twin Study sup-porting our hypothesis about the role of intellec-tual interest in Gene · SES interaction onacademic achievement.
Transactional Models of Cognitive Development
Plomin, DeFries, and Loehlin (1977) firstdescribed the processes by which genotypes couldcome to be differentially associated with environ-mental exposure, i.e., gene–environment correla-tion. Passive gene–environment correlations arisewhen children are raised by their biological par-ents, such that their rearing environments are influ-enced by some of the same genes that they inherit.Evocative gene–environment correlations arisewhen children’s genetically influenced traits,features, and characteristics elicit particular envi-ronments from others. Finally, active gene–environ-
Data were obtained from the Henry A. Murray ResearchArchive at Harvard University (http://www.murray.harvard.edu/). The Population Research Center at the University ofTexas at Austin is supported by a center grant from the NationalInstitute of Child Health and Human Development (R24HD042849). The original collectors of the data, the MurrayResearch Archive, and NICHD bear no responsibility for use ofthe data or for interpretations or inferences based upon suchuses. We thank John Loehlin for helpful comments on previousversions of this article.
Correspondence concerning this article should be addressed toElliot M. Tucker-Drob, Department of Psychology, University ofTexas at Austin, 1 University Station A8000, Austin, TX 78712-0187. Electronic mail may be sent to [email protected].
Child Development, xxxxx 2012, Volume 00, Number 0, Pages 1–15
� 2012 The Authors
Child Development � 2012 Society for Research in Child Development, Inc.
All rights reserved. 0009-3920/2012/xxxx-xxxx
DOI: 10.1111/j.1467-8624.2011.01721.x
ment correlations arise when children seek out andchoose experiences and environments that are con-sistent with their genetically influenced traits.
The typology of passive, evocative, and activegene–environment correlation was applied todevelopmental theory by Scarr and McCartney(1983), who argued that children become increas-ingly autonomous over the course of development,such that passive gene–environment correlationsweaken and active gene–environment correlationsstrengthen. A plausible, albeit perhaps counterintu-itive, net effect of this process is that genes andenvironments become increasingly correlated withone another over the course of development. Thusgenetically similar people (like monozygotic [MZ]twins) will select increasingly similar environmen-tal niches over time, and it is this convergence ofenvironmental experience that maintains, or evenincreases, their phenotypic similarities. That is,Scarr and McCartney suggested that genetic influ-ences on complex psychological phenotypes arereinforced through on the on-going selection ofenvironmental experiences that are ‘‘correlatedwith motivational, personality, and intellectualaspects of our genotypes’’ (p. 427). Elaborating onthis perspective, Scarr (1992) later hypothesizedthat the process of individuals sorting themselvesinto learning environments ‘‘depends on peoplehaving a varied environment from which to chooseand construct experiences’’ (p. 9), a requisite thatshe argued is particularly likely to be absent for‘‘children reared in very disadvantaged circum-stances.’’
The process by which environmental experiencesproduce phenotypic differences was further elabo-rated on by Bronfenbrenner and Ceci (1994). Theirbioecological model contends that environmentalexperiences are inextricably linked to genetic differ-ences between individuals and that the dynamic bywhich children and their environments mutually actupon one another is central to the realization ofgenetic potential for healthy development. Specifi-cally, Bronfenbrenner and Ceci state that ‘‘humandevelopment takes place through processes of pro-gressively more complex reciprocal interactionsbetween an active [child] and the person, objectsand symbols in [the child’s] immediate environ-ment’’ (p. 572). These reciprocal interactions aretermed proximal processes. A critical aspect of the bio-ecological model is its prediction that proximalprocesses will differ in their availability and qualityacross macroenvironmental contexts, even in therange of ‘‘good enough’’ environments. This is a sig-nificant point of departure from Scarr (1993; also see
Scarr, 1992, 1996; Scarr & McCartney, 1983), whoargued that environments other than severe depri-vation were ‘‘functionally equivalent’’ (p. 1337).
Finally, Dickens and Flynn (2001) applied atransactional model of development specifically tothe domain of cognitive abilities. They state the pre-mise of transactional models simply: ‘‘Higher IQleads one into better environments causing stillhigher IQ, and so on.’’ A major contribution of theDickens and Flynn model is its conceptualization ofhow environmental experience aggregates overtime. They suggest that environmental influencesneed not be large, but simply need to be consistentand recurring over long periods of development inorder to have large effects on cognition andachievement. Moreover, environmental influencesthat are selected based on genetically influencedtraits and preferences, rather than serendipitouslyencountered or externally imposed, are most likelyto be consistent over time (Caspi, Roberts, & Shiner,2005; McAdams & Olson, 2010), and thus are pre-cisely those environmental experiences that aremost influential.
Considered together, transactional models ofcognitive development posit that (a) individualsselect (and are selected into) environments that areincreasingly congruent with their own geneticallyinfluenced traits, (b) environments that are congru-ent with one’s genotype are most likely to beconsistent and recurring over the course of devel-opment, (c) recurring interactions with high-qualityenvironments are necessary for the realization ofgenetic potential for healthy cognitive outcomes,and (d) macroenvironmental contexts affect theavailability and quality of environmental experi-ences important for cognition.
Gene · SES Interactions in Cognitive Ability andAcademic Achievement
One important macroenvironmental context thatmay affect an individual’s ability to select andinteract with high-quality environmental experi-ences necessary for cognitive development andlearning is SES. SES can be conceptualized as repre-senting a family’s level of ‘‘financial capital (mate-rial resources), human capital (nonmaterialresources such as education), and social capital(resources achieved through social connections)’’(Bradley & Corwyn, 2002, p. 372). The differencesin resources available to high- versus low-SES fami-lies are evident across multiple domains, includingparental responsiveness, parental teaching, andlevel of cognitive stimulation (Bradley, Corwyn,
2 Tucker-Drob and Harden
McAdoo, & Coll, 2001). Children in higher SEShomes are more likely to have developmentallyappropriate books, be read to by a family member,be taught academic concepts, receive special lessonsto improve specific skills, and be taken to themuseum or theater. SES differences in environmen-tal quality extend outside of the home, too (Kazdin,Kraemer, Kessler, Kupfer, & Offord, 1997). Com-pared to children growing up in higher SES homes,children growing up in lower SES homes are morelikely to attend schools with inadequate instruc-tional materials, fewer advanced placement classes,fewer books and computers, nonfunctional sci-ence labs, and fewer academically oriented orhigh-achieving classmates (Phillips & Chin, 2004).Moreover, lower SES children are perceived morenegatively and receive less attention from teachers(McLoyd, 1998).
Given the relation between SES and children’sopportunities to select and participate in the envi-ronmental experiences necessary for maximizingability and achievement, transactional models ofcognitive development predict that genetic variancein cognitive outcomes will be higher for childrenliving in higher SES circumstances. Consistent withthis prediction, there is an emerging body ofresearch indicating that genetic influences on cogni-tive outcomes do indeed positively vary with SES. Ina seminal article, Rowe, Jacobson, and van den Oord(1999) demonstrated that the heritability of verbalability in a national sample of adolescents rangedfrom 72% in the most educated families to 26% inthe least educated families. A subsequent reanalysisof this data by Guo and Stearns (2002), who incorpo-rated multiple indices of parental SES, found thatthe moderating effect of SES could be best accountedfor by parental ethnicity and unemployment. Next,using advances in quantitative techniques formodeling Gene · Environment interaction, Turkheim-er, Haley, Waldron, D’Onofrio, and Gottesman(2003) reported that among low-SES 7-year-olds,heritability of IQ was 10%, whereas among high-SES 7-year-olds, heritability of IQ was 72% (seealso Tucker-Drob, Harden, & Turkheimer, 2009).Harden, Turkheimer, and Loehlin (2007) reported asimilar effect of parental income and parental educa-tion on the heritability of academic achievement in asample of 17-year-olds, and subsequent analysesindicated that their result was robust to assumptionsabout assortative mating and passive gene–environ-ment correlation between adolescent genotype andfamily income (Loehlin, Harden, & Turkheimer,2009). Using a sample of middle-aged males, Kre-men et al. (2005) found that heritability for word
recognition (a component of reading ability) variedpositively with parental education, accounting for21% of the variance at the lowest parental educationand 69% at the highest. This effect, however, wasdriven by decreases in shared environmental vari-ances with increasing education, rather than byincreases in genetic variance. Very recently, Tucker-Drob, Rhemtulla, Harden, Turkheimer, and Fask(2011) demonstrated that Gene · SES effects on cog-nitive ability emerge as early as 2 years of age. Theyreported that in a nationally representative sampleof 2-year-old twins born in the United States in 2001,genetic influences on Bayley mental ability testscores approached 0 for low-SES children andapproached 50% for high-SES children. Finally, evi-dence consistent with a Gene · SES effect on cogni-tive outcomes has also been found using a moleculargenetic paradigm: Enoch, Waheed, Harris, Albaugh,and Goldman (2009) found a significant interactionbetween Val158Met, a functional polymorphism ofthe COMT gene, and educational attainment on cog-nitive abilities among adults. They found that,among Met allele carriers, educational attainmenthad a strong positive relation with test scores,whereas among Val ⁄ Val individuals, the positiverelation between educational attainment and testsscores was much less pronounced. Overall, extantresearch suggests that there are Gene · SES inter-actions for cognitive outcomes in childhood, adoles-cence, and adulthood.
Noncognitive Factors as Driving Forces in CognitiveDevelopment
Research on Gene · SES interactions is largelyconsistent with the propositions of transactionalmodels—that children actively select and respondto environmental experiences in accordance withtheir own genetically influenced traits, and that thisprocess is restricted by socioeconomic disadvan-tage, resulting in lower heritability of cognitive out-comes in lower SES homes. However, the specificfactors that govern how a child or adolescent differ-entially selects or responds to options within theenvironment remain largely unexplored. While theDickens and Flynn (2001) model emphasizes chil-dren’s preexisting levels of ability and competence,we propose that noncognitive factors—includinglevels of scholastic motivation, drive for achieve-ment, intellectual self-concept, and intellectualinterest—are also critical for the process of selectingenvironmental niches.
Figure 1 is a schematic illustration that repre-sents the proposed role of noncognitive factors in
Intellectual Interest and G·E 3
the selection of environments that influencecognition and achievement. At the corners of thetriangle are three core class of constructs: (a) inter-ests, intentions, and personality; (b) proximal envi-ronments (peer groups, coursework, activities,interpersonal interactions); and (c) cognitive abili-ties and achievement. Along the sides of the trian-gle are mechanisms that illustrate the bidirectionalnature of the relations between the three coreclasses of constructs. For example, interest in intel-lectual and academic pursuits is probabilisticallyrelated to experiencing high-quality proximal envi-ronments (e.g., enrollment in challenging course-work), through processes by which childrenactively seek out these experiences (e.g., an adoles-cent enrolling in advanced placement English inhopes of improving chances of college admission)and by which they evoke these experiences fromothers (e.g., a teacher recommending more difficultmath courses to an engaged and interested stu-dent). In turn, high-quality proximal environmentscan result in further intellectual interest as a resultof socialization processes, while also directly boost-ing achievement through the instructional process.(This schematic is not intended to be fully compre-hensive; there are, of course, many other mecha-nisms that may underlie the relations among thethree core sets of constructs.)
Furthermore, we predict that the strength of themechanisms relating noncognitive factors, achieve-ment, and proximal environments will differsystematically with SES. Only in high-SES circum-stances, where children and adolescents can take
advantage of a wide array of environmental experi-ences, interest and motivation will become tightlycoupled with achievement: Intellectually interestedadolescents will be able to invest more time andeffort into achievement-relevant behaviors (e.g.,additional time studying), and they will preferen-tially select achievement-enhancing proximal envi-ronments (e.g., high-achieving peer groups,challenging coursework). More specifically, geneticdifferences in interest and motivation will becometightly coupled with achievement. This is becauseintellectual interest will result in an advantage forachievement only when it is systematic and recur-ring over long periods of development (Dickens &Flynn, 2001), and genetically influenced aspects ofpersonality are generally more developmentallyconsistent than are aspects of personality influ-enced by the immediate environment (Caspi et al.,2005). Thus, children and adolescents in high-SEShomes will be able to ‘‘convert’’ genetic differencesin interest and motivation into achievement, result-ing in higher overall heritability for achievement. Incontrast, intellectual interest and achievement willbe decoupled for children in lower SES circum-stances, who have restricted access to achievement-enhancing proximal environments (e.g., feweropportunities for advanced math classes) and whohave fewer resources to devote to achievement-rele-vant behaviors. Without the opportunity to ‘‘actout’’ genetically influenced interest and motivationin the environment, genetic differences in noncog-nitive factors become irrelevant for academicachievement. The net effect of this process will bereduced influence of genes related to intellectualinterest on achievement, and lower overall herita-bility of achievement, for children living in lowerSES homes.
Hypotheses
This study aims to test the role of intellectualinterest in Gene · SES interactions on academicachievement, using a sample of 777 adolescent twinpairs from the National Merit Twin Study (Loehlin& Nichols, 1976, 2009). Specifically, we test whetherSES moderates the relation between geneticvariance in intellectual interest and academicachievement. Transactional models of cognitivedevelopment predict that in higher SES homes,genetic variance in intellectual interest is stronglycoupled with academic achievement, resulting inhigher overall heritability for academic achieve-ment. Alternatively, in lower SES homes, whereintellectually interested adolescents have restricted
Figure 1. A conceptual model for the mutual relations betweeninterests, proximal environments, and achievement.
4 Tucker-Drob and Harden
opportunities to select and respond to enrichingproximal environments, the association betweengenetic variance in intellectual interest and achieve-ment is predicted by transactional models to beweakly coupled, resulting in lower overall heritabil-ity of achievement.
Method
Data for the current project come from the NationalMerit Twin Study (Loehlin & Nichols, 1976).Harden et al. (2007) have previously reported thatin these data, heritability of a general achievementfactor was higher for twins growing up in higherincome homes. Here we extend this work by exam-ining the extent to which this Gene · Environmentinteraction can be accounted for by socioeconomicdifferences in the genetic basis of the interest–achievement relation.
Participants
Participants were sampled from approximately600,000 American high school students (averageage = 17 years) who took the National Merit Schol-arship Qualifying Test (NMSQT) in 1962. Of these,1,507 pairs of same-sex twins were identified, 850pairs of which agreed to participate. Zygosity wasdetermined by a questionnaire that assessed twinsimilarity in childhood and the frequency withwhich they were confused by others (Nichols &Bilbro, 1966). These determinations were cross-vali-dated using a subsample of 124 twin pairs ofknown zygosity, and found to be over 90% accu-rate. The current analyses were restricted to the 777pairs for whom family income was reported (475MZ pairs and 302 dizygotic [DZ] pairs). It is impor-tant to note that this sample size is comparable tothose of many other twin studies (Boomsma, Bus-jahn, & Peltonen, 2002), but much smaller thanmost epidemiological studies. While power todetect Gene · Environment interaction effects isalways a potential concern, previous studies usingthis data set have already found significant Gene ·Environment interaction effects on academicachievement (Harden et al., 2007), which is a testa-ment to the power of this sample size.
Measures
Academic achievement was measured with theNMSQT, which is composed of five subscales: Eng-lish Usage, Mathematics Usage, Social Science
Reading, Natural Science Reading, and WordUsage. For the purposes of this article the NMSQTselection score, which is a unit-weighted compositeof the five subscale scores, was used as an index ofgeneral academic achievement. The twin pair corre-lation for this score was r = .88 for MZ twins andr = .64 for DZ twins (both p < .01).
Intellectual interest was measured with the Intel-lectual Efficiency scale of the California Psychologi-cal Inventory. According to the Megargee (1972),‘‘the manifest content [of the Intellectual Efficiency]scale reflects an interest in and enjoyment in intel-lectual pursuits: ‘I like to read about history’–[True]; and self-confidence and assurance: ‘I seemto be as capable and smart as most others aroundme’– [True].’’ According to McAllister (1996), veryhigh scorers on the intellectual efficiency scale are‘‘conceptual and intellectually oriented, tending tothink or talk about problems more than act onthem,’’ whereas low scorers ‘‘prefer to deal withtangible and concrete issues rather than with con-cepts or abstractions.’’ Importantly, intellectual effi-ciency is a subscale that correlates moderately withobjective indices of cognition and achievement butis based entirely on subjective self-report personal-ity items. In the current data set, this scale corre-lated with academic achievement at .44 (p < .01;based on only 1 twin per pair). The twin pair corre-lation for this scale was r = .52 for MZ twins andr = .33 for DZ twins (both p < .01).
SES was indexed by parental report of familypretax income in a written questionnaire, with sevenresponse categories ranging from less than $5,000per year to over $25,000 per year. This rangeapproximately corresponds to a range of lessthan $31,250 to over $156,250 in 2004 dollars. In thecurrent data set, this index correlated with aca-demic achievement at .23 (p < .01; based on only 1twin per pair), and with intellectual interest at .11(p < .01; based on only 1 twin per pair).
Analytical Methods
Data were analyzed using a series of four struc-tural equation models of increasing complexity.First, we fit univariate main effects models sepa-rately to achievement and intellectual interest, inorder to determine the overall magnitude of geneticinfluences on these phenotypes. Second, we fitunivariate Gene · Environment interaction (G·E)models separately to achievement and intellectualinterest, in order to determine whether the herit-abilities of these phenotypes were moderated bySES. Third, we fit a bivariate main effects model, in
Intellectual Interest and G·E 5
order to test the contribution of genes to the associa-tion between intellectual interest and achievement.Finally, we fit a bivariate G·E model, in order todetermine whether the relation between geneticvariance in intellectual interest and achievementwas moderated by SES, as would be predicted bytransactional models of cognitive development.These models are described in more detail below.For all analyses, all variables were standardized rel-ative to the means and standard deviationsobserved for the first twin in each pair. Analyseswere carried out using the Mplus computer pro-gram (Muthen & Muthen, 1998-2009) with maxi-mum likelihood estimation.
Step 1: Univariate main effects model.. A conven-tional biometric model for twins reared togetherspecifies that a given phenotype is influenced bythree statistically additive and independent unob-served components: additive genes (A); the sharedenvironment (C), that is, environmental influencesthat make children raised in the same home similarto each other; and the nonshared environment (E),that is, environmental influences that make childrenraised in the same home different from each other,plus measurement error (see Neale & Cardon, 1992,for more details regarding the parameterization oftwin models). This basic model can be expanded toinclude measured covariates. Figure 2 illustratesthe univariate main effects model for intellectualinterest, including SES as a measured family-levelcovariate. In this model, the variances of the A, C,and E components are fixed to 1, and the correla-tion between A components in the first and second
members of each twin pair is fixed according togenetic theory (rs = 1.0 in MZ twins and 0.5 in DZtwins). The paths from the A, C, and E componentsare freely estimated, and the square of these pathsindicate the proportion of variance in the pheno-type attributable to genes, the shared environment,and the nonshared environment. Thus the squareof the a path gives the familiar heritability statistic(h2)—the proportion of variance in the phenotypeattributable to additive genes. It is important tonote that because SES was measured at the familylevel, it was by definition an aspect of the sharedenvironment. That is, controlling for SES reducesour estimate of the variance accounted for by theshared environment. Because SES is controlled forin all of our models, C should therefore be inter-preted as family-level influences that are incre-mental to family-level SES. Univariate maineffects models were fit separately for intellectualinterest and academic achievement, in order totest the overall magnitude of additive genetic,shared environmental, and nonshared environ-mental influences on these phenotypes at thepopulation level.
Step 2: Univariate interaction model.. As describedby Purcell (2002), the conventional univariate twinmodel can be easily expanded to test forGene · Environment interaction. The univariateG·E model for intellectual interest is shown inFigure 3. This model is identical to the model illus-trated in Figure 2, except that the paths represent-ing the influence of additive genetic, sharedenvironmental, and nonshared environmental influ-
1 1 1 1 1 1
1.0 or 0.5 1.0
A1 C1 E1 A2 C2 E2
I t ll t l ll l
ciai eiciai ei
IntellectualInterest(Twin 1)
IntellectualInterest(Twin 2)
si sSES
2
si si
s
Figure 2. A path diagram for the univariate main effects modelof intellectual interest in twins.
ci’SES
1 1 1 1 1 1
1.0 or 0.5 1.0
A1 C1 E1 A2 C2 E2
e + e ’SES’SES e + e ’SES’SES’ ci +
I t ll t l ll l
i + i SESai + ai SES i + i SESai + ai SES
IntellectualInterest(Twin 1)
IntellectualInterest(Twin 2)
si sSES
2
si
s
si
ci + ci SES
Figure 3. A path diagram for the univariate interaction model ofintellectual interests in twins.
6 Tucker-Drob and Harden
ences are allowed to vary with SES. For example,the path from the additive genetic components tointellectual interest is modeled as (a + a¢ · SES),where a significant value for a¢ would indicate thatthe amount of variance in intellectual interest attri-butable to genes varies with SES.
Step 3: Bivariate main effects model.. The univariatemain effects models fit in Step 1 test the magnitudeof genetic influence on each phenotype separately,but do not test the extent to which genes contributeto the association between intellectual interest andachievement. That is, to the extent that higher intel-lectual interest is correlated with higher achieve-ment, is this association attributable to commongenetic influences? In order to test this, we fit abivariate main effects model, shown in Figure 4.There are multiple ways to parameterize bivariatetwin data; the current project uses a Choleskymodel, which specifies an a priori ordering of thevariables. Specifically, this model estimates A, C,and E components for intellectual interest; achieve-ment is specified as a dependent variable that isregressed on the A, C, and E components of intel-lectual interest; and residual variance in achieve-ment that is independent of intellectual interest isdecomposed into a second set of A, C, and E com-ponents. Thus, in Figure 4, the ai, ci, and ei parame-ters represent the contributions of genes, the sharedenvironment, and the nonshared environment tointerest; the ab, cb, and eb parameters represent theextent to which genetic, shared environmental, andnonshared environmental influences on intellectualinterest also predict academic achievement; and theaa, ca, and ea parameters represent the contributionsof genes, the shared environment, and the non-
shared environment to the variance in achievementthat is independent of interest.
Step 4: Bivariate interaction model.. Just like theunivariate models, the bivariate main effects modelcan be expanded to allow the genetic, shared envi-ronmental, and nonshared environmental influ-ences to be moderated by SES. Such a bivariateinteraction model is shown in Figure 5. It is impor-tant to note that this model for tests for three differ-ent types of Gene · SES interaction: (a) whetherSES moderates genetic influence on intellectualinterest (ai + ai¢ · SES), (b) whether SES moderatesthe association between genetic variance inintellectual interest and academic achievement(ab + ab¢ · SES), and (c) whether SES moderatesgenetic influence on academic achievement that isindependent of intellectual interest (aa + aa¢ · SES).
The bivariate interaction model allows us to dis-tinguish between at least two different theoreticalscenarios. If the influence of genetic variance inintellectual interest on achievement is positivelymodified by SES, then Gene · SES interaction forachievement would be due to greater geneticcoupling between interest and achievement in high-SES homes. Such a scenario would be consistentwith our hypothesis that higher SES environmentsallow for greater opportunities to select and interactwith experiences congruent with one’s own geneti-cally influenced traits, leading to greater geneticcoupling of intellectual interest and academicachievement. However, if only the genetic compo-nent of achievement that is independent of interestis modified by SES, then the Gene · SES interactioncannot be accounted for by a decoupling betweeninterest and achievement in low- SES homes.
1 1 1 1 1 1
eb
Ai Ci Ei Aa Ca Ea
b
abciai ei
cb caaa ea
IntellectualInterest
Achievement
si sSES
2
sa
s
si
Figure 4. A path diagram of the bivariate Choleksy model ofintellectual interest and academic achievement.Note. Only one twin per pair is shown.
1 1 1 1 1 1
Ai Ci Ei Aa Ca Ea
ca + ca’SESaa + aa’SES ea + ea’SESci + ci’SES
ai + ai’SES ei + ei’SES
IntellectualInterest
Achievement
si sSES
2
si sa
s
Figure 5. A path diagram of the bivariate interaction model forintellectual interest and academic achievement.Note. Only one twin per pair is shown.
Intellectual Interest and G·E 7
Results
Step 1: Univariate Main Effects Models
Parameter estimates from the univariate maineffects models for academic achievement and intel-lectual interest are presented in the first and thirdcolumns of Table 1. It can be seen that there werestatistically significant genetic, shared environmen-tal, and nonshared environmental influences onboth outcomes. For academic achievement additivegenetic influences were estimated to account for45% of the variance, shared environmentalinfluences for 36%, and nonshared environmentalinfluences (plus measurement error) for 13% (SESaccounted for the remaining 6%). There were sub-stantially greater within-twin pair differences forintellectual interest than for achievement. For intel-lectual interest additive genetic influences wereestimated to account for 27% of the variance,shared environmental influences for 21%, andnonshared environmental influences for 49% (SESaccounted for the remaining 2%).
Step 2: Univariate Interaction Models
The parameter estimates for the univariate inter-action models for achievement and intellectualinterest are shown in the second and fourth col-umns of Table 1. As previously reported by Hardenet al. (2007), the univariate interaction model foracademic achievement fit the data significantly bet-ter than the univariate main effects (see Table 2 fora summary of model comparisons). Inspection ofparameter estimates from the univariate interactionmodel indicates that the only interaction parameterthat is significant is a¢—the moderating effect ofSES on additive genetic influence. This parameterwas estimated to equal 0.12 (SE = 0.05), indicatingthat genetic influences on academic achievementare higher at higher levels SES. Figure 6 plots thevariance in academic achievement attributable toadditive genetic, shared environmental, and non-shared environmental influences as functions ofSES, as implied by parameter estimates. It can beseen that the variance in achievement attributableto genes increases substantially from low to highSES levels. The same pattern can be observed in theAppendix, which reports MZ and DZ correlationsfor low-, middle-, and high-SES families.
In contrast to the positive finding of a Gene ·SES interaction on academic achievement, therewas no statistically significant evidence for such aninteraction on intellectual interest: The univariateinteraction model for interest did not fit signifi-
cantly better than the univariate main effectsmodel, and none of the interaction parameters (a¢,c¢, or e¢) were significantly different from zero.Based on these results, we cannot conclude thatgenetic influences are any more or less importantfor intellectual interest in advantaged environmentsversus disadvantaged environments.
Step 3: Bivariate Main Effects Model
The bivariate main effects model extends theresults from the univariate main effects model byestimating the genetic, shared environmental, andnonshared environmental influences on each phe-notype and on the association between interest andachievement. Parameter estimates for the bivariatemain effects model are shown in the fifth column ofTable 1. The ab, cb, and eb parameters were all sig-nificantly different from zero, indicating that bothgenetic and environmental factors contribute to theassociation between interest and achievement. Spe-cifically, genetic influences on interest accountedfor 8% of the variance in achievement, and sharedenvironmental influences on interest accounted for17% of the variance in achievement. Although sig-nificant, the effect of the nonshared environmentalcomponent of interest on achievement was verysmall, accounting for less than 1% of the variancein achievement after controlling for SES. Theseresults demonstrate that it is genetic and sharedenvironmental components of interest that accountfor the vast majority of the interest–achievementassociation. It is important to observe, however,that there was substantial genetic and environmen-tal variation in achievement that was independentof interest, as indicated by the statistically signifi-cant aa, ca, and ea parameters. Specifically, geneticinfluences independent of interest accounted for37% of the variance in achievement, shared envi-ronmental influences independent of interestaccounted for 18% of the variance in achievement,and nonshared environmental influences indepen-dent of interest accounted for 12% of the variancein achievement.
Step 4: Bivariate Interaction Models
The bivariate interaction model examines whetherSES moderates the extent to which academicachievement is influenced by the genetic, sharedenvironmental, and nonshared environmental com-ponents of intellectual interest. Moreover, it testswhether the genetic variance in academic achieve-ment that is independent of intellectual interest is
8 Tucker-Drob and Harden
Tab
le1
Par
amet
erE
stim
ates
Fro
mU
niv
aria
tean
dB
ivar
iate
Tw
inM
odel
s
Par
amet
er
Un
ivar
iate
mo
del
sB
ivar
iate
mo
del
s
(In
tere
stfi
Ach
iev
emen
t)A
chie
vem
ent
Inte
rest
1.M
ain
effe
cts
2.In
tera
ctio
n3.
Mai
nef
fect
s4.
Inte
ract
ion
5.M
ain
effe
cts
6.In
tera
ctio
n7.
Fin
alm
od
el
Est
imat
eS
EE
stim
ate
SE
Est
imat
eS
EE
stim
ate
SE
Est
imat
eS
EE
stim
ate
SE
Est
imat
eS
E
a i0
.52
30
.11
00
.503
0.1
270
.538
0.1
060
.490
0.1
240
.536
0.0
95
a i¢
0.10
10.
068
0.11
10.
068
c i0
.45
90
.11
30
.460
0.1
230
.445
0.1
170
.470
0.1
090
.447
0.1
04
c i¢
)0.
100
0.08
3)
0.10
30.
07
e i0
.70
00
.02
20
.699
0.0
230
.698
0.0
220
.700
0.0
220
.699
0.0
21
e i¢
)0.
037
0.02
0)
0.0
40
.02
a b0
.274
0.1
070
.283
0.1
050
.281
0.0
86
a b¢
0.1
350
.068
0.1
220
.043
c b0
.412
0.1
380
.381
0.1
080
.408
0.1
06
c b¢
0.01
40.
08
e b0
.061
0.0
160
.057
0.0
160
.059
0.0
16
e b¢
)0.
023
0.01
5
a a0
.66
80
.044
0.6
55
0.0
450
.609
0.0
560
.586
0.0
550
.589
0.0
55
a a¢
0.1
20
0.0
460.
055
0.04
5
c a0
.59
60
.055
0.5
92
0.0
590
.430
0.1
300
.447
0.0
980
.441
0.1
03
c a¢
)0.
079
0.06
4)
0.10
30.
079
e a0
.35
90
.012
0.3
58
0.0
120
.353
0.0
110
.352
0.0
110
.353
0.0
11
e a¢
)0.
012
0.01
1)
0.01
00.
011
s i0
.12
70
.03
00
.127
0.0
310
.130
0.0
300
.135
0.0
310
.135
0.0
30
s a0
.24
30
.032
0.2
51
0.0
340
.242
0.0
320
.253
0.0
340
.253
0.0
34
Not
e.P
aram
eter
sin
bo
ldar
esi
gn
ifica
nt
atp
<.0
5.
Intellectual Interest and G·E 9
moderated by SES. That is, are genetic differencesin intellectual interest more strongly coupled withacademic achievement in high-SES environments,and after accounting for this effect, is there any resi-dual Gene · SES interaction on academic achieve-ment?
Parameter estimates from the full bivariate inter-action model are shown in the sixth column ofTable 1. Notably, the ab¢ parameter was positiveand significantly different from zero, indicating thatgenetic variance in intellectual interest contributedto academic achievement significantly more for chil-dren in higher SES homes versus lower SES homes.That is, adolescents’ genetic predispositions to seekout and enjoy intellectually stimulating activitiescontribute to their academic achievement, but onlywhen they are raised in high SES homes. This isconsistent with our transactional model of cognitivedevelopment, which proposes that adolescents can‘‘convert’’ their genetically influenced interest and
motivation into higher achievement only when in amacroenvironmental context that allows them toselect and interact with appropriate environmentalexperiences. Moreover, the aa¢ parameter was notsignificantly different from zero, indicating thatgenetic influences on achievement that were inde-pendent of interest were not moderated by SES.This result indicates that Gene · SES interactions onacademic achievement can be accounted for bystronger influences of the genes for intellectualinterest on academic achievement in higher SEShomes. Note that in this model, there was also evi-dence for less nonshared environmental variance inintellectual interest in high-SES homes (i.e., the ei¢parameter was significantly less than 0).
The fit of the bivariate interaction model wasonly marginally significantly better than the fit ofthe main effects model. This is likely because themajority of the interactions modeled were not sig-nificantly different from zero. We therefore fit amodel in which all nonsignificant interactionparameters were fixed to zero. In this model, the ab¢remained significantly different from zero, but theei¢ parameter was not significantly different fromzero. Thus, as a final modeling step, we fixed allinteraction parameters to zero, except for ab¢ (theparameter representing the interaction between SESand the effect of genetic variance in intellectualinterest on academic achievement). As shown inTable 2, this final model did not fit significantlyworse than the full bivariate interaction model, andit fit significantly better than the bivariate maineffects model that did not allow for any Gene · SESinteraction. The final model was therefore acceptedas the best representation of the data.
Parameter estimates from final model are pre-sented in the seventh column of Table 1. Becausethere is no aa¢ interaction parameter, this modelindicates that the Gene · SES interaction on aca-demic achievement (shown above in Figure 6) canbe attributed to stronger influences of the genes forintellectual interest on academic achievement inhigher SES homes. Figure 7 illustrates how SESmoderates the etiology of academic achievement,based on the parameters from the final bivariateinteraction model. The amounts of variance in intel-lectual interest attributable to genetic, shared envi-ronmental, and nonshared environmental factorsdo not vary with SES (left panel of Figure 7), andthe same is true of variance in academic achieve-ment that is independent of intellectual interest(right panel of Figure 7). However, the geneticcomponent of intellectual interest is more stronglypredictive of achievement at higher levels of SES.
Figure 6. Genetic and environmental contributions to academicachievement as functions of socioeconomic status.
10 Tucker-Drob and Harden
The genetic component of intellectual interestaccounts for almost zero variance in achievement atlower SES ()2 SD), but it accounts for approxi-mately 30% of the variance in achievement athigher SES (+2 SD).
Discussion
Behavioral genetic analyses of data from 777 pairsof high school twins indicated a link between aca-demic achievement and genes for intellectual inter-est. At the population level, the magnitude of thislink was modest; genes for intellectual interestaccounted for only 8% of the variation in academicachievement. However, this proportion differedaccording to family SES. At very low levels of SES,the genetic component of intellectual interest pre-dicted close to 0% of the variation in academicachievement, whereas at high levels of SES, thisproportion was approximately 30%. In contrast, theshared environment component of intellectualinterested accounted for 17% of the variation in aca-demic achievement, and the nonshared environ-ment component of intellectual interest accountedfor less than 1% of the variation in academicachievement, regardless of level of SES. The neteffect was a Gene · SES interaction in the directionof greater heritability of academic achievement inadolescents living in higher SES contexts.
These results are consistent with the theoreticalproposition that socioeconomic disparities in chil-dren’s opportunities to match their intellectualinterest with congruent intellectually stimulatingproximal environments is a major mechanismunderlying the Gene · SES interaction on adoles-cent cognition and achievement that has beenobserved in recent behavioral genetic research.Given the breadth of literature documenting the
impact of SES on the day-to-day experiences of chil-dren and adolescents (Bradley & Corwyn, 2002),we contend that adolescents raised in lower SEScontexts (a) are afforded less opportunity to seekout intellectually stimulating scholastic experiences,peer groups, and interpersonal interactions thatmatch their levels of intellectual interest, and (b)receive fewer benefits to their intellectual growthand learning from work and effort put into intellec-tual and academic pursuits. Together, these mecha-nisms serve to make cognitive ability and academicachievement less related to intellectual interest inlower SES compared to higher SES groups. Specifi-cally, borrowing from Dickens and Flynn (2001),we predicted that socioeconomic differences wouldbe most evident in the genetic basis of the relationbetween intellectual interest and academic achieve-ment, because components of personality traits thatresult from genetic predispositions are far more sta-ble over development than are components thatresult from environmental experiences. That is, thenonshared environment is likely to represent short-lived, ‘‘one-time’’ effects that are inconsistent acrossdevelopment, and are therefore likely not to existwith the intensity or duration necessary to haveprofound effects on cognitive development andlearning (Dickens & Flynn, 2001; Turkheimer &Waldron, 2000). Genetic influences, on the otherhand, are often developmentally consistent andlong lasting, such that they result in cumulativeand consistent exposure to environments that haveeffects on achievement.
Limitations and Remaining Questions
Relation between academic achievement and cognitiveability. Readers are likely to wonder about theextent to which the current findings, whichare based on a composite measure of academic
Figure 7. Genetic and environmental components of intellectual interest, the regression of academic achievement on intellectual interest(Intellectual Interest fi Achievement), and the variance in academic achievement that is unique of intellectual interest(Achievement.Intellectual Interest), as functions of socioeconomic status.
Intellectual Interest and G·E 11
achievement, relate to Gene · Environment inter-action research that has been conducted with con-ventional IQ scores (e.g., Turkheimer et al., 2003).There is a substantial body of research indicatingthat a general dimension of academic achievementis highly related to, although not synonymous with,a global dimension of cognitive ability, oftentermed general intelligence, or g. A study by Deary,Strand, Smith, and Fernandes (2007), for example,places the magnitude of this correlation at .81.There is also evidence (Thompson, Detterman, &Plomin, 1991) that while academic achievementmay be more influenced by the shared environmentthan is cognitive ability, the two factors are largelyinfluenced by the same genes. Finally, Tucker-Drob(2010) has recently demonstrated that while SES-related differences in the rates of development inmultiple cognitive abilities and multiple domains ofacademic achievement outcomes can be substan-tially accounted for by way of a general develop-mental pathway, supplemental domain-specificpathways are necessary to account for SES-relateddifferences in the development of some aspects ofacademic achievement. In sum, while the currentresults are likely to strongly relate to extant G·Efindings with respect to general intelligence, it ispossible that they also represent some achievement-specific developmental processes.
The National Merit sample. One limitation of sam-ple used for these analyses is that because theNational Merit Qualifying Test was administeredas part of a competition for college scholarships,the National Merit Twin sample lacked representa-tion of adolescents who did not plan to go to col-lege. To illustrate, the average class ranking ofstudents taking the NMQST was the 21st percentile.Moreover, the NMSQT sample consisted of veryfew minorities. The current findings are thereforemost representative of the comparison betweenmiddle-class White families with upper-class Whitefamilies. This selectivity limits us from being ableto strongly generalize our findings to very poorfamilies with very low achieving students, or toracial and ethnic minorities.
A related limitation concerns the fact that theNational Merit Twin study was initiated in 1962,and may therefore be difficult to generalize to cur-rent adolescents, who have been raised in the pres-ent conditions of social stratification. Since the1970s, families with very low SES have becomeincreasingly concentrated in geographically isolatedcommunities, especially in urban areas (Massey,1996; Wilson, 1987). To the extent that opportunityfor scholastic achievement depends on the material
resources that communities can provide their chil-dren, rather than the resources of individual fami-lies (Brooks-Gunn, Duncan, Klebanowv, & Sealand,1993), then historical changes in the concentrationof poverty might be expected to exacerbate theimpact of SES on expression of genes related toachievement. This is speculative, however, andreplication in contemporary samples is certainlywarranted.
Why no G·E on intellect interest? Another out-standing issue concerns why Gene · SES effectswere not observed for intellectual interest. Wemight have expected that if dynamic interest–achievement matching processes were the basis forthe Gene · SES effect on academic achievement, asimilar Gene · SES effect would be present forintellectual interest. That is, according to our pro-posed framework, interest and achievement mutu-ally influence one another, such that interest (bothdirectly and indirectly) affects achievement, andachievement (both directly and indirectly) affectsinterest. It is of note that, although not statisticallysignificant, the a¢, c¢, and e¢ parameter estimates inTable 1 for the univariate interaction model of intel-lectual interest were in fact consistent with such aneffect. It is possible that, if the sample had been lar-ger or more diverse, we would have had the powerto detect an interaction on interest at statisticallysignificant levels. Another possibility is that whilethe academic achievement measure used was psy-chometrically very strong (it was a composite ofscores from five highly correlated subscales), theintellectual interest measure used may have beensomewhat weaker, rendering subtle G·E effectsharder to detect. Finally, it is possible that thedirectional relation from interest to achievement ismore affected by differences in socioeconomicopportunity than is the directional relation fromachievement to interest. This could help to explainwhy a robust Gene · SES interaction held for thevariance in achievement that was predicted byinterest, but did not hold for any other variancecomponents.
Power. It is also important to comment on thehow the size of the current sample may haveaffected our results. We analyzed data from 777pairs of twins, which is a sample size comparableto that of many other contemporary twin studies(Boomsma et al., 2002), but much smaller than mostepidemiological studies. It is possible that our anal-yses may have only been powered to detect largeand robust Gene · Environment interactions, andmay have missed more subtle interactions. A testa-ment to the power of this study is the fact that we
12 Tucker-Drob and Harden
were able to detect a significant and robust inter-action between SES and genes for interest thataccounted for variation in achievement. Afteraccounting for this interaction, the previouslydetected interaction between SES and genes forachievement was reduced to nonsignificant levels,suggesting that intellectual interest completelymediated the Gene · SES effect on achievement.However, it is possible that if we had obtained aneven larger sample of twins, this residual Gene· SES effect would have remained statisticallysignificant, thus indicating only partial mediation.Of course, logistical considerations inevitably forceresearchers to make difficult decisions regardingthe trade-off between obtaining large samples andobtaining high-quality, detailed, and reliable, multi-variate measurements. Typically, in order to over-come the substantial challenges to collecting largesamples of individuals, researchers reduce thebreadth, depth, and precision of measurement.Continued progress in identifying and testing thepsychosocial mechanisms underlying gene–envi-ronment effects on academic achievement willrequire large genetically informed studies withbroad arrays of high-quality measures.
Developmental specificity of mechanisms for G·E.The current results are consistent with the existenceof socioeconomic disparities in the success of agene–environment transaction in which adolescentsactively select proximal environments that are con-sistent with their levels of intellectual interest. It islikely, however, that earlier in life, SES is likely tohave its most profound effect on the success ofevocative processes, by which child temperamentselicit specific types of stimulation from caregivers(Bradley & Corwyn, 2002; McLoyd, 1998). There-fore, while the current results are likely attributableto SES differences in an active ‘‘niche-picking’’ pro-cess, it is likely that Gene · SES interactions onmental abilities in very early childhood are resultsof socioeconomic differences in evocative processes(Tucker-Drob et al., 2011).
Future Directions
Specific interests and specific cognition and achieve-ment outcomes.. The current project exemplifies thebenefit of integrating research on cognition andachievement with research on personality. The per-sonality–achievement intersection is historicallyunderstudied (Chamorro-Premuzic & Furnham,2005), yet is likely to be integral to understandingwhy children differ in their abilities to achieve theircognitive and academic potentials. Future research
will benefit from examining the specific loci ofthese effects by measuring a wide variety of per-sonality, ability, and achievement constructs, andexamining how they mediate heritable variation inone another. Although a general factor can accountfor large proportions of individual differences inmany different cognitive ability and academicachievement domains, researchers are increasinglyrecognizing that there are substantial individualdifferences in specific cognitive domains that areunaccounted for by a general factor, and that arepotentially governed by domain-specific develop-mental processes (Tucker-Drob, 2009). There is sim-ilarly a growing appreciation for the uniqueproperties of facets of larger personality traits(DeYoung, Quilty, & Peterson, 2007), and howdifferent facets may differentially relate to cognitiveabilities (Moutafi, Furnham, & Crump, 2006).
Specific aspects of SES and specific proximal environ-ments. Just as it will be necessary to measurespecific interests and specific cognition and achieve-ment outcomes, it will be necessary for futureresearch to make use of specific measures of bothmacroenvironments and proximal environments.SES is likely to serve as a proxy for a number ofmacroenvironmental contexts, such as populationdensity, residential instability, and school quality, toname a few. Future research will certainly benefitfrom examining how such indices relate to thestrength of dynamic processes by which childrenselect their proximal environments. It will be equallyimportant for future research to make use of specificmeasures of the proximal environments that chil-dren are likely to select. Such proximal environ-ments might include peer groups, course work,extracurricular activities, and time spent readingand studying.
Longitudinal changes across development. An addi-tional future direction will be to examine thehypotheses tested here using longitudinal data. Thecurrent project made use of a cross-sectionalapproach in order to make inferences about adynamic process that unfolds over time. Longitudi-nal measures of children’s interests, abilities, andboth macrocontextual and proximal environmentsfrom infancy through adolescence will be requiredto capture more fully the processes by which chil-dren evoke and select individual experiences thatare congruent with their interests, intentions, moti-vations, and self-concept. Obtaining such measuresacross a wide range of ages would be of particularvalue for examining when during childhood theintersection between intellectual interest andachievement emerges. Intellectual interest is likely
Intellectual Interest and G·E 13
to become important for academic achievementmuch earlier than adolescence, given evidencesuggesting that even young children’s learningexperiences are influenced by their own behaviors(Lugo-Gil & Tamis-LeMonda, 2008; Raine, Rey-nolds, Venables, & Mednick, 2002). Furthermore,longitudinal data would be valuable for examiningwhether interest–achievement relations strengthenover development. There are at least two reasons toexpect such a strengthening to occur, at the veryleast for children living in high SES contexts. First,as children get older they have greater opportunityto autonomously select their curricular and extra-curricular activities, as well as their peer groups(Scarr & McCartney, 1983). Second, early educa-tional choices and learning experiences are likely toboth bolster achievement and reinforce the interestsand motivations that were the basis for those initialchoices. Caspi et al. (2005) articulated this perspec-tive as their corresponsive principle: ‘‘the most likelyeffect of life experience on personality developmentis to deepen the characteristics that lead people tothose experiences in the first place.’’
Conclusion
In summary, behavioral genetic models were fitto data on intellectual interest and academicachievement from 777 pairs of MZ and DZ twinsfrom the National Merit Twin Study. There was sta-tistically significant evidence that the variance inacademic achievement explained by genes variedpositively with SES. In the context of a bivariatemodel, this effect could be accounted for by strongerinfluences of the genes for intellectual interest onacademic achievement in higher SES homes. Theseresults are consistent with the hypothesis that higherSES allows children to better convert their intellec-tual interest into academic achievement through aprocess of gene–environment correlation.
References
Anastasi, A. (1958). Heredity, environment, and the ques-tion ‘‘How?’’ Psychological Review, 65, 197–208.
Boomsma, D., Busjahn, A., & Peltonen, L. (2002). Classicaltwin studies and beyond. Nature Review Genetics, 3,872–882.
Bradley, R. H., & Corwyn, R. F. (2002). Socioeconomicstatus and child development. Annual Review of Psychol-ogy, 53, 371–399.
Bradley, R. H., Corwyn, R. F., McAdoo, H. P., & Coll, C.G. (2001). The home environments of children in the
United States: I. Variations by age, ethnicity, and pov-erty status. Child Development, 72, 1844–1867.
Bronfenbrenner, U., & Ceci, S. J. (1994). Nature-nurturereconceptualized in developmental perspective: A bio-ecological model. Psychological Review, 101, 568–586.
Brooks-Gunn, J., Duncan, G. J., Klebanowv, N., & Sea-land, N. (1993). Do neighborhoods influence child andadolescent development? American Journal of Sociology,99, 353–395.
Caspi, A., Roberts, B. W., & Shiner, R. L. (2005). Personal-ity development: Stability and change. Annual Review ofPsychology, 56, 453–484.
Deary, I. J., Strand, S., Smith, P., & Fernandes, C. (2007).Intelligence and educational achievement. Intelligence,35, 13–21.
DeYoung, C. G., Quilty, L. C., & Peterson, J. B. (2007).Between facets and domains: 10 aspects of the Big Five.Journal of Personality and Social Psychology, 93, 880–896.
Dickens, W. T., & Flynn, J. R. (2001). Heritability esti-mates versus large environmental effects: The IQ para-dox resolved. Psychological Review, 108, 346–369.
Enoch, M. A., Waheed, J. F., Harris, C. R., Albaugh, B., &Goldman, D. (2009). COMT and cognition: Main effectsand interaction with educational attainment. Genes,Brain, and Behavior, 8, 36–42.
Guo, G., & Stearns, E. (2002). The social influences on therealization of genetic potential for intellectual develop-ment. Social Forces, 80, 881–910.
Harden, K. P., Turkheimer, E., & Loehlin, J. C. (2007).Genotype-by-environment interaction in adolescents’cognitive aptitude. Behavior Genetics, 37, 273–283.
Hernstein, R. J., & Murray, C. (1994). The bell curve: Intelli-gence and class structure in American life. New York:Simon & Schuster.
Jensen, A. R. (1969). How much can we boost IQ and scho-lastic achievement? Harvard Educational Review, 39, 1–123.
Jensen, A. R. (1973). Educability and group differences. NewYork: Harper & Row.
Kazdin, A. E., Kraemer, H. C., Kessler, R. C., Kupfer, D.J., & Offord, D. R. (1997). Contributions of risk-factorresearch to developmental psychopathology. ClinicalPsychology Review, 17, 375–406.
Kremen, W. S., Jacobson, K. C., Xian, H., Eisen, S. A.,Waterman, B., Toomey, R., et al. (2005). Heritability ofword recognition in middle-aged men varies as a func-tion of parental education. Behavior Genetics, 35, 417–433.
Loehlin, J. C., Harden, K. P., & Turkheimer, E. (2009).The effects of assumptions about parental assortativemating and genotype-income correlation on estimatesof genotype-environment interaction in the NationalMerit Twin Study. Behavior Genetics, 39, 165–169.
Loehlin, J. C., & Nichols, R. C. (1976). Heredity, environ-ment, and personality: A study of 850 sets of twins. Austin:University of Texas Press.
Loehlin, J. C., & Nichols, R. C. (2009). The National MeritTwin Study. hdl:1902.1 ⁄ 13913 UNF:5:9jU6DgNdW
14 Tucker-Drob and Harden
2620G6xbOAqsA== Murray Research Archive[Distributor].
Lugo-Gil, J., & Tamis-LeMonda, C. S. (2008). Familyresources and parenting quality: Links to children’scognitive development across the first 3 years. ChildDevelopment, 79, 1065–1085.
Massey, D. S. (1996). The age of extremes: Concentratedaffluence and poverty in the twenty-first century.Demography, 33, 395–412.
McAdams, D. P., & Olson, B. D. (2010). Personality devel-opment: Continuity and change over the life course.Annual Review of Psychology, 61, 517–542.
McAllister, L. (1996). A practical guide to CPI interpretation(3rd ed.). Palo Alto, CA: Consulting Psychologists Press.
McLoyd, V. C. (1998). Socioeconomic disadvantage andchild development. American Psychologist, 53, 185–204.
Megargee, E. I. (1972). The California Psychological Inven-tory handbook. San Francisco, CA: Jossey-Bass.
Moutafi, J., Furnham, A., & Crump, J. (2006). What facets ofopenness and conscientiousness predict fluid intelli-gence score? Learning and Individual Differences, 16, 31–42.
Muthen, L. K., & Muthen, B. O. (1998-2009). Mplus user’sguide (5th ed.). Los Angeles, CA: Muthen & Muthen.
Neale, M. C., & Cardon, L. R. (1962). Methodology forgenetic studies of twins and families. Dordrecht, The Neth-erlands: Kluwer Academic.
Nichols, R. C., & Bilbro, W. C. (1966). The diagnosis of twinzygosity. Acta Genetica et Statistica Medica, 16, 265–275.
Phillips, M., & Chin, T. (2004). School inequality: Whatdo we know? In K. Neckerman (Ed.), Social inequality(pp. 467–519). NewYork: Russell Sage Foundation.
Plomin, R., DeFries, J. C., & Loehlin, J. C. (1977). Genotype-environment interaction and correlation in the analysisof human behavior. Psychological Bulletin, 88, 245–258.
Purcell, S. (2002). Variance components models for gene–environment interaction in twin analysis. Twin Research,5, 554–571.
Raine, A., Reynolds, C., Venables, P. H., & Mednick, S. A.(2002). Stimulation seeking and intelligence: A prospec-tive longitudinal study. Journal of Personality and SocialPsychology, 82, 663–674.
Rowe, D. C., Jacobson, K. E., & van den Oord, E. (1999).Genetic and environmental influences on vocabulary
IQ: Parental education level as a moderator. ChildDevelopment, 70, 1151–1162.
Scarr, S. (1992). Developmental theories for the 1990’s:Development and individual differences. Child Develop-ment, 63, 1–19.
Scarr, S. (1993). Biological and cultural diversity: The leg-acy of Darwin for development. Child Development, 64,1333–1353.
Scarr, S. (1996). How people make their own environ-ments: Implications for parents and policy makers. Psy-chology, Public Policy, and Law, 2, 204–228.
Scarr, S., & McCartney, K. (1983). How people make theirown environments: A theory of genotype fi environ-ment effects. Child Development, 54, 424–435.
Thompson, L. A., Detterman, D. K., & Plomin, R. (1991).Association between cognitive abilities and scholasticachievement: Genetic overlap, but environmental dif-ferences. Psychological Science, 2, 158–165.
Tucker-Drob, E. M. (2009). Differentiation of cognitiveabilities across the lifespan. Developmental Psychology,45, 1097–1118.
Tucker-Drob, E. M. (2010). How many pathways underliesocioeconomic differences in children’s cognition andachievement? Manuscript submitted for publication.
Tucker-Drob, E. M., Harden, K. P., & Turkheimer, E.(2009). Combining nonlinear biometric and psychomet-ric models of cognitive abilities. Behavior Genetics, 39,461–471.
Tucker-Drob, E. M., Rhemtulla, M., Harden, K. P.,Turkheimer, E., & Fask, D. (2011). Emergence of agene-by-socioeconomic status interaction on infantmental ability from 10 months to 2 years. PsychologicalScience, 22, 125–133.
Turkheimer, E., Haley, A., Waldron, M., D’Onofrio, B.M., & Gottesman, I. I. (2003). Socioeconomic statusmodifies heritability of IQ in young children. Psycholog-ical Science, 14, 623–628.
Turkheimer, E., & Waldron, M. C. (2000). Nonsharedenvironment: A theoretical, methodological and quanti-tative review. Psychological Bulletin, 126, 78–108.
Wilson, W. J. (1987). The truly disadvantaged: The inner city,the underclass, and public policy. Chicago: University ofChicago Press.
Appendix
Monozygotic (MZ) and Dizygotic (DZ) Twin Correlations for Academic Achievement in Lower, Middle, and HigherIncome Families
Group Family income (1–7 scale) Family income range (1962 dollars) NMZ NDZ rMZ rDZ h2 c2 e2
Lower SES 1, 2 Less than $5,000 ⁄ year to $7,499 ⁄ year 188 104 .840 .657 36.6% 47.4% 16.0%
Middle SES 3, 4 $7,500 ⁄ year to $14,999 ⁄ year 205 144 .882 .626 51.2% 37.0% 11.8%
Higher SES 5, 6, 7 $15,000 ⁄ year to $25,000 ⁄ year and over 82 54 .902 .573 65.8% 24.4% 9.8%
Full sample 1, 2, 3, 4, 5, 6, 7 Less than $5,000 ⁄ year to $25,000 ⁄ year and over 475 302 .880 .637 48.6% 39.4% 12.0%
Note. h2 = 2(rMZ)rDZ). c2 = rMZ)2(rMZ)rDZ). e2 = 1)rMZ. h2, c2, and e2 estimates reported here for the full sample differ slightly fromthose reported in the article, because results reported in article were produced by structural equation models that controlled for themain effects of socioeconomic status (SES).
