The Genetics of Success: How Single- Nucleotide Polymorphisms

Psychological Science

2016, Vol. 27(7) 957 –972

Reprints and permissions:

sagepub.com/journalsPermissions.nav

DOI: 10.1177/0956797616643070

pss.sagepub.com

Research Article

In 2013, scientists reported the first successful genome-

wide association study (GWAS) of a social-science out-

come, educational attainment (Rietveld etal., 2013). Their

analysis of millions of genetic variants in more than

100,000 individuals hinted at the existence of a molecular

map to success in schooling written in the alphabet of

DNA. As anticipated, rather than finding a so-called gene

for education, this study revealed a genetic continuum:

Some individuals carry very few alleles associated with

educational attainment, the bulk of the population carries

643070PSS

XXX10.1177/0956797616643070Belsky et al.The Genetics of Success

research-article2016

Corresponding Author:

Daniel W. Belsky, 2020 W. Main St., Suite 201, Durham, NC 27708

E-mail: [email protected]

The Genetics of Success: How Single-

Nucleotide Polymorphisms Associated

With Educational Attainment Relate to

Life-Course Development

Daniel W. Belsky

1,2

, Terrie E. Moffitt

3,4,5,6

, David L. Corcoran

Benjamin Domingue

, HonaLee Harrington

, Sean Hogan

Renate Houts

, Sandhya Ramrakha

, Karen Sugden

Benjamin S. Williams

, Richie Poulton

, and

Avshalom Caspi

3,4,5,6

Department of Medicine, Duke University School of Medicine;

Social Science Research Institute,

Duke University;

Department of Psychology & Neuroscience, Duke University;

Department of Psychiatry

and Behavioral Sciences, Duke University School of Medicine;

Center for Genomic and Computational

Biology, Duke University;

MRC Social, Genetic & Developmental Psychiatry Research Centre,

Institute of Psychiatry, Psychology & Neuroscience, King’s College London;

Graduate School of Education,

Stanford University; and

Dunedin Multidisciplinary Health & Development Research Unit,

Department of Psychology, University of Otago

Abstract

A previous genome-wide association study (GWAS) of more than 100,000 individuals identified molecular-genetic

predictors of educational attainment. We undertook in-depth life-course investigation of the polygenic score derived

from this GWAS using the four-decade Dunedin Study (N = 918). There were five main findings. First, polygenic scores

predicted adult economic outcomes even after accounting for educational attainments. Second, genes and environments

were correlated: Children with higher polygenic scores were born into better-off homes. Third, children’s polygenic

scores predicted their adult outcomes even when analyses accounted for their social-class origins; social-mobility

analysis showed that children with higher polygenic scores were more upwardly mobile than children with lower

scores. Fourth, polygenic scores predicted behavior across the life course, from early acquisition of speech and reading

skills through geographic mobility and mate choice and on to financial planning for retirement. Fifth, polygenic-score

associations were mediated by psychological characteristics, including intelligence, self-control, and interpersonal skill.

Effect sizes were small. Factors connecting DNA sequence with life outcomes may provide targets for interventions to

promote population-wide positive development.

Keywords

genetics, behavior genetics, intelligence, personality, adult development

Received 12/28/15; Revision accepted 3/14/16

958 Belsky et al.

some such alleles, and a few people carry many. This

continuum, measured as a polygenic score (Chabris, Lee,

Cesarini, Benjamin, & Laibson, 2015), has since been

shown to predict educational attainment in cohorts on

three continents and even differences in educational

attainment between siblings in the same family (Conley

et al., 2015; de Zeeuw etal., 2014; Domingue, Belsky,

Conley, Harris, & Boardman, 2015; Rietveld, Esko, etal.,

2014; Ward et al., 2014). Although the magnitudes of

associations are small, these findings have provoked con-

troversy and concern about misuse and misinterpretation

(Henig, 2015). In an effort to provide an empirical foun-

dation for productive public discussion of the new sci-

ence of sociogenomics, we ask three questions in the

current article: (a) Do genetic discoveries for educational

attainment predict outcomes beyond schooling? (b) If so,

what are the developmental and behavioral pathways

that connect differences in DNA sequences with diver-

gent life outcomes? (c) Do psychological characteristics

act as mediators of genetic associations? Although these

questions may seem premature, it is important to ask

them now, before technologies using genetics to predict

social outcomes become possible.

These questions were addressed by examination of data

prospectively collected from a population-representative

birth cohort followed through midlife, the Dunedin Study

(Poulton, Moffitt, & Silva, 2015). Across 13 repeated in-

person assessments, Dunedin Study members were evalu-

ated for developmental milestones in childhood; for traits,

behaviors, and aspirations through adolescence; and ulti-

mately for attainments and outcomes in adulthood (Table

1). Because attrition has been minimal (5% at the latest

wave in 2012), the findings illustrate genetic associations

with life courses and life outcomes without bias from

selective attrition as a result of illness or challenging life

circumstances. We tested a series of hypotheses about

the scope, pathways, and psychological mechanisms of

genetic influence on socioeconomic attainments across

the first half of the life course. We tracked a deeply phe-

notyped cohort from early childhood through midlife,

examining preselected developmentally appropriate man-

ifestations of achievement-related behaviors. We report a

large number of outcome variables in order to provide a

complete account of these data. In the interest of repro-

ducibility the analysis plan was posted in advance.

Method

Sample

Participants were members of the Dunedin Study, a longi-

tudinal investigation of health and behavior in a complete

birth cohort. Dunedin Study members (N = 1,037; 91%

of eligible births; 52% male) were all individuals born

between April 1972 and March 1973 in Dunedin, New

Zealand, who were eligible on the basis of residence in

the province and who participated in the first assessment

at age 3. The cohort represented the full range of socio-

economic status (SES) in the general population of New

Zealand’s South Island. On adult health, the cohort

matched the New Zealand National Health and Nutrition

Survey (e.g., body mass index, smoking, visits to the doc-

tor; Poulton etal., 2015). The cohort was primarily White;

fewer than 7% self-identified as having non-European

ancestry, matching the population of the South Island

(Poulton et al., 2015). Assessments were carried out at

birth and at ages 3, 5, 7, 9, 11, 13, 15, 18, 21, 26, 32, and,

most recently, 38 years, when 95% of the 1,007 Dunedin

Study members still alive took part. At each assessment,

each Dunedin Study member was brought to the research

unit for a full day of interviews and examinations.

Genotyping and imputation

We used Illumina HumanOmni Express 12 BeadChip

arrays (Version 1.1; Illumina, Hayward, CA) to assay com-

mon single-nucleotide polymorphism (SNP) variation in

the genomes of our cohort members. We imputed addi-

tional SNPs using the IMPUTE2 software (Version 2.3.1;

https://mathgen.stats.ox.ac.uk/impute/impute_v2.html;

Howie, Donnelly, & Marchini, 2009) and the 1000

Genomes Phase 3 reference panel (1000 Genomes

Project, 2016). Imputation was conducted on autosomal

SNPs appearing in dbSNP (Version 140; http://www.ncbi

.nlm.nih.gov/SNP/; Sherry etal., 2001) that were “called”

in more than 98% of the Dunedin Study samples. Invari-

ant SNPs were excluded. Prephasing and imputation

were conducted using a 50-million-base-pair sliding win-

dow. The resulting genotype database included geno-

typed SNPs and SNPs imputed with 90% probability of a

specific genotype among the non-Maori members of the

Dunedin cohort (n = 918). We analyzed SNPs in Hardy-

Weinberg equilibrium (p > .01).

Polygenic scoring

We calculated polygenic scores according to the method

described by Dudbridge (2013) using the PRSice software

(Version 1.22; http://prsice.info/; Euesden, Lewis, &

O’Reilly, 2015). To calculate the polygenic score for edu-

cational attainment, we matched genotypes from our data

with GWAS results for educational attainment reported by

the Social Science Genetic Association Consortium

( Rietveld et al., 2013) and used the approximately 2.3

million matched genotypes to score Dunedin Study mem-

bers’ genetic predisposition to educational attainment.

For each genotype, we counted the number of educa-

tion-associated alleles (0, 1, or 2) and multiplied this

The Genetics of Success 959

count by the effect size estimated in the original GWAS.

(Most genotypes had effect sizes very near 0.) We then

summed weighted counts across all genotypes to calcu-

late each Dunedin Study member’s score. We used all

matched SNPs to compute polygenic scores, irrespective

of nominal significance for their association with educa-

tional attainment. Scores ranged from −30.51 to 73.77

(M=17.73, SD = 17.94) and were normally distributed in

the Dunedin birth cohort. We standardized scores so that

the mean was zero and the standard deviation was 1 (see

Fig. S1 in the Supplemental Material available online).

Given the original GWAS results, Dunedin Study mem-

bers with polygenic scores greater than 0 would be

expected to complete more years of schooling, and

Dunedin Study members with polygenic scores below 0

would be expected to complete fewer years of schooling.

We used the same method to calculate polygenic scores

for height (based on results from the Genetic Investiga-

tion of Anthropometric Traits Consortium’s most recent

GWAS of height; Wood et al., 2014). To account for

potential population stratification, we adjusted polygenic

score analyses for the first 10 principal components com-

puted from the genome-wide SNP data using the

EIGENSOFT smartPCA tool (Version 5.0.2; http://www

.hsph.harvard.edu/alkes-price/software/; Price etal., 2006;

Price, Zaitlen, Reich, & Patterson, 2010).

Measurement of life-course-

development phenotypes

More detailed descriptions of study measures described

later in this section and relevant citations are provided in

the Supplemental Material.

Social-class origins. We measured social class origins

as the average SES across repeated assessments through-

out Dunedin Study members’ childhoods. SES was deter-

mined from the higher of either parent’s occupational

status throughout the Dunedin Study members’

childhoods.

Attainment. We measured educational attainment as

the highest degree completed by a Dunedin Study mem-

ber through the time of the age-38 assessment. We mea-

sured attainment beyond education from Dunedin Study

members’ reports of their occupation, income, assets,

Table 1. Tracking the Development of Socioeconomic Success

Phenotype Measure or data source Age

Success in schooling

Highest degree Structured interview 15–38

Success beyond schooling

Adult-attainment factor Occupation (prestige score based on NZ Census data), income, assets, credit-

problems scale, difficulty-paying-expenses scale, days of social-welfare-benefit

use (NZ Social Welfare Administration), credit score (Veda credit bureau)

Social mobility Childhood social class based on parental occupation; adult attainment measured

using education, occupation, and the adult-attainment factor

Birth–15, 38

Pathways to success

Developmental milestones Interviews with mothers 3

Reading ability Burt Word Reading Test (Scottish Council for Research in Education, 1976) 7–18

Aspirations Questionnaire 15

Standardized testing NZ Ministry of Education test record 18

Geographic mobility Life-history calendar interview 21–38

Financial planfulness Structured interview and informant reports 32–38

Mate selection Structured interview in which Dunedin Study members reported their

relationship status and, for those in a serious relationship, their partners’

highest educational degree and income

Skills and abilities

Cognitive ability Peabody Picture Vocabulary Test (Dunn, 1965), Stanford-Binet Intelligence

Scale (Terman & Merrill, 1960), Wechsler Intelligence Scales for Children–

Revised (Wechsler, 1974)

3–13

Self-control skills Staff observations, parent and teacher reports, and interviews with Dunedin

Study members

3–11

Interpersonal skill Staff observations 3–9

Physical health Medical exams, anthropometry, lung function testing, clinical interviews with

parents

3–11

Note: NZ = New Zealand.

960 Belsky et al.

credit problems, and difficulties paying expenses when

they were 38 years old and from electronic record

searches of social-welfare and credit-score databases.

Pathways to success. We measured the age at which

Dunedin Study members achieved early developmental

milestones on the basis of data gathered from interviews

with their mothers when the members were 3 years old.

We measured reading ability from scores on the Burt

Word Reading Test (Scottish Council for Research in

Education, 1976), taken when Dunedin Study members

were 7, 9, 11, 13, 15, and 18 years old. We measured edu-

cational and socioeconomic aspirations from surveys

completed by the Dunedin Study members at the age of

15. We measured academic performance from scores on

standardized tests taken at the ages of 15 to 18. We mea-

sured geographic mobility from member life-history cal-

endar reports about place of work and residence from

the ages of 21 to 38. We measured financial planfulness

on the basis of data gathered from surveys of Dunedin

Study members’ friends and relatives and from structured

interviews with the members themselves when they were

32 and 38 years old. We measured the SES of members’

romantic partners from members’ reports of their part-

ners’ income and education in structured interviews con-

ducted when the members were 38 years old.

Life satisfaction. When they were 38 years old,

Dunedin Study members completed a five-item Satisfac-

tion With Life scale (e.g., “In most ways my life is close to

ideal,” “So far I have gotten the important things I want in

life”; Pavot & Diener, 1993).

Traits and abilities. We measured cognitive ability

and cognitive development using the Peabody Picture

Vocabulary Test (Dunn, 1965), administered when

Dunedin Study members were 3 years old; the Stanford-

Binet Intelligence Scale (Terman & Merrill, 1960), admin-

istered when members were 5 years old; and the Wechsler

Intelligence Scales for Children–Revised (WISC-R; Wechsler,

1974), administered when members were 7 to 13 years old.

We measured Dunedin Study members’ childhood self-

control skills from observational ratings of their lack of

control (when they were 3 and 5 years old) and parent,

teacher, and self-reports of impulsive aggression, hyper-

activity, lack of persistence, inattention, and impulsivity

(when they were 5–11 years old). We measured Dunedin

Study members’ childhood interpersonal skills from

reports made by trained research workers after standard-

ized testing sessions when they were 3 to 9years old. We

measured childhood health from medical exams, anthro-

pometry, lung function testing, and interviews with

parents at assessments made between Dunedin Study

members’ birth and the age of 11.

Height. Study members’ height at age 38 was measured

to the nearest millimeter using a stadiometer ( Harpenden;

Holtain, Ltd., Crosswell, Wales).

Ethical approvals

The study protocol was approved by the institutional ethi-

cal review boards of the participating universities. Dunedin

Study members gave informed consent before participat-

ing. The Otago University ethics committee provided ethi-

cal approval for the Dunedin Study. Participants gave

written consent before data were collected. When partici-

pants were children, their parents gave informed consent.

Data sharing

Dunedin Study data are available to researchers on appli-

cation. A managed-access process ensures that approval

is granted to research that comes under the terms of par-

ticipant consent and privacy (see the Supplemental Mate-

rial for data-sharing details).

Statistical analysis

We analyzed continuous dependent variables using lin-

ear regression models to estimate standardized regres-

sion coefficients (reported as Pearson’s r). We analyzed

dichotomous dependent variables using Poisson regres-

sion models to estimate relative risks (RRs). We analyzed

time-to-event data for developmental milestones using

Cox models to estimate hazard ratios. We analyzed

ordered categorical outcomes using ordered logit models

to estimate odds ratios. We analyzed repeated measures

longitudinal data on reading ability and cognitive devel-

opment using multilevel longitudinal growth models

(Singer & Willett, 2003). Finally, we conducted mediation

analyses using the system of equations described by

Baron and Kenny (1986) and the methods described by

Preacher and his colleagues (Preacher & Hayes, 2008;

Preacher & Kelley, 2011) to calculate total, direct, and

indirect effects and to estimate the proportion of effects

mediated by each of the mediators. Growth model and

mediation analyses are described further in the Supple-

mental Material. All models were adjusted for sex.

Results

Analyses included the 918 non-Maori Dunedin Study mem-

bers who provided DNA samples. Cohort members’

genomes were scored according to published GWAS results

for educational attainment (Rietveld etal., 2013; polygenic

scores were standardized so that they had a mean of 0 and

a standard deviation of 1; see Fig. S1 in the Supplemental

Material). The analyses proceeded in three parts. Part 1

The Genetics of Success 961

analyses examined divergent outcomes of high- and low-

scoring children, first in education and then in the acquisi-

tion of social and economic capital through midlife and the

social mobility those attainments reflected. Part 2 analyses

examined how higher-scoring children came to grow apart

from their lower-scoring peers. Analysis tested genetic dif-

ferences in the timing of early-life milestones; in the age at

which children learned to read; in the decision to test for

secondary-education credentials and university enrollment,

and performance on those tests; in geographic mobility in

search of training and employment; and in selection of

mates, formation of households, and forging of careers.

Part 3 analyses examined candidate psychological charac-

teristics through which genetic influences on development

and life outcomes might come about.

Part 1: What did discovered genetics of

educational attainment mean for life

outcomes beyond schooling?

Part 1 analyses tested the hypothesis that Dunedin Study

members’ polygenic scores would predict their life attain-

ments at the age of 38, roughly the midpoint in the

human life span. All analyses were adjusted for the first

10 principal components computed from genome-wide

SNP data (see Table S1 in the Supplemental Material) to

adjust for potential population stratification (i.e., genome-

wide patterning of differences in allele frequency that

might induce spurious correlations between the poly-

genic score and study outcomes). Unadjusted estimates

are reported in Table S1 in the Supplemental Material.

Did individuals with higher polygenic scores achieve

higher degrees? In replication of the original discovery

about the genetics of educational attainment, our results

showed that Dunedin cohort members with higher poly-

genic scores tended to go on to achieve higher degrees

compared with peers who had lower scores (r = .15,

p<.001; Fig. 1a). This correlation between polygenic score

and educational attainment was nearly identical to the esti-

mate from the original report (Rietveld etal., 2013). As in

previous studies, the genetic effect was small in magni-

tude; for example, having a polygenic score 1 standard

deviation above the mean was associated with a 19%

increase in likelihood of completing a university degree

(RR = 1.19, 95% confidence interval (CI) = [1.07, 1.32]).

Did individuals with higher polygenic scores go on

to achieve socioeconomic success beyond school-

ing? Adult socioeconomic attainments of Dunedin Study

members were measured using data from structured inter-

views about jobs, income, wealth, and financial difficul-

ties and by conducting administrative-record searches of

governmental and credit-bureau databases. Factor analy-

sis of these multiple measures was used to compute an

adult-attainment-factor score (see Table S2 and Fig. S2 in

the Supplemental Material). By midlife, individuals with

higher polygenic scores tended to be more socioeconom-

ically successful: They held more prestigious occupations,

earned higher incomes, had accumulated more assets,

reported fewer difficulties paying their expenses, relied

less on social-welfare benefits, and had higher credit

scores (r = .13, p < .001 for the adult-attainment factor;

Fig. 1b). It may seem unsurprising that a polygenic score

that predicted educational attainment also continued to

predict success after education was complete. However,

less than half of the genetic association with adult attain-

ment was accounted for by educational attainment; when

we repeated our genetic analysis of the adult-attainment

factor and included education as a covariate, the adjusted

effect size was .07 (p = .035). Genetic effect sizes for the

individual attainment measures and effect sizes after

adjustment for educational attainment are shown in

Fig.S3 in the Supplemental Material.

In sum, in the Dunedin cohort, individuals with higher

polygenic scores tended to grow up to become more

successful, not only in schooling, but also in their eco-

nomic and professional lives. This success depended

only partly on their educational attainment.

Were children with higher polygenic scores more

often born into socially advantaged families? In

previous research, the correlation between parent and off-

spring polygenic scores was estimated to be approximately

.6 (Conley etal., 2015). Moreover, if a generation of indi-

viduals who achieve more occupational and economic

success carry a certain genotype or set of genotypes, it

stands to reason that their own children will inherit not

only their genetics, but also their social success. This

hypothesis of social stratification of genotypes was tested

by comparing polygenic scores of children whose parents

occupied different social positions. Parents’ SES was mea-

sured from repeated assessments conducted when the

cohort members were growing up (i.e., during their first 15

years of life; see the Supplemental Material). Our findings

point to a gene-environment correlation: The polygenic

score for educational attainment was stratified by child-

hood SES such that children with higher polygenic scores

tended to have grown up in families with higher SES,

whereas children with lower polygenic scores tended to

have grown up in families with lower SES (r = .13, p<.001).

Were children with higher polygenic scores more

likely to achieve upward social mobility? Analyses of

social mobility tested whether the higher life attainments of

children with higher polygenic scores were independent of

their social origins. The analysis of adult socioeconomic

962 Belsky et al.

outcomes was repeated, but with the addition of a statisti-

cal control for the SES of a child’s family during his or her

first 15 years of life (see the Supplemental Material). Three

interrelated outcomes were considered: the Dunedin Study

member’s educational attainment; their attained adult SES,

measured as occupational prestige (in parallel to the status

of their parents); and their adult-attainment-factor score.

Children with higher polygenic scores tended to attain

more regardless of whether they began life in a family that

was well-off or one that was socially disadvantaged (more

education: r= .10, p = .002; more prestigious occupations:

r = 0.11, p<.001; higher adult-attainment-factor scores:

r = .11, p = .002). Figure 2 shows associations between the

polygenic score and adult attainment in groups of

Dunedin Study members with low, middle, and high

childhood SES.

Dunedin Study data confirmed that children with

higher polygenic scores had grown up in families with

more socioeconomic resources (Krapohl & Plomin,

2016). But the data also showed that even for children

born into socially disadvantaged circumstances, higher

polygenic scores predicted upward social mobility.

Part 2: How did children with higher

polygenic scores grow apart from their

peers?

If children with higher polygenic scores do achieve

higher levels of attainment in schooling and beyond, it is

important to know how this comes about. The intermedi-

ate phenotypes that link DNA sequence with life out-

comes can provide clues about genetic mechanisms and

can also suggest targets for interventions designed to

improve children’s outcomes (Belsky, Moffitt, & Caspi,

2013). The next analysis examined how children with

higher polygenic scores grew apart from their peers

beginning during the early school years and continuing

through midlife.

Children with higher polygenic scores were more

likely to say their first words at younger ages. When

Dunedin Study members were 3 years old, their mothers

were interviewed about the ages at which the members

achieved each of a series of developmental milestones.

The milestones, ordered by the normative age at which

–0.50

–0.25

0.00

0.25

0.50

Polygenic Score

Compulsory

Education

Only

School-

Leaving

Certiﬁcate

Sixth-Form

Certiﬁcate

or Bursary

University

Degree

Highest Education

–0.50

–0.25

0.00

0.25

0.50

Adult-Attainment Factor (z Score)

–2 –1 0 1 2

Polygenic Score

–0.75

0.75

Fig. 1. Association between polygenic score and educational and adult achievement. In (a), mean polygenic score is graphed as a function of edu-

cational attainment. Error bars represent 95% confidence intervals. For the 1972–73 birth cohort we studied, compulsory education ended at age 15

years, at which point students could elect to take a School Leaving Certificate exam. Fifteen percent of our sample obtained no educational creden-

tial; 15% obtained the School Leaving Certificate but did not progress further; 42% completed sixth-form or Bursary Certificates (roughly equivalent

to a full high school diploma in the United States); and 29% completed a university degree. In (b), the scatterplot (with best-fitting regression line)

shows the relationship between Dunedin Study members’ polygenic scores (x-axis) and their adult-attainment-factor z scores (y-axis). The adult

attainment factor was composed of occupational prestige, income, assets, credit problems, difficulties paying expenses, social-welfare-benefit use,

and credit score. Each plotted point represents mean x and y coordinates for a bin of 10 Dunedin Study members.

The Genetics of Success 963

they were reached, were smiling, walking, talking, feed-

ing oneself, daytime potty training, communicating using

sentences, and nighttime potty training (see Fig. S5 in the

Supplemental Material). Dunedin Study members with

higher polygenic scores began talking earlier, on average,

than peers with lower scores (hazard ratio = 1.12, 95%

CI = [1.05, 1.19], p < .001), and were also somewhat

quicker to begin communicating using sentences (hazard

ratio = 1.06, 95% CI = [1.00, 1.13], p = 0.052), although

this difference was not statistically significant at the α =

0.05 threshold. This accelerated development was

restricted to verbal ability; Dunedin Study members with

higher polygenic scores did not reach other developmen-

tal milestones ahead of peers.

Children with higher polygenic scores acquired

reading skills at younger ages. Study members’ read-

ing skill was assessed with the Burt Word Reading Test at

each measurement session from ages 7 to 18 years. We

used longitudinal multilevel growth models to test genetic

associations with the model intercept and linear and qua-

dratic slopes of change in reading over time (see Fig. S6

in the Supplemental Material). The model intercept

captured the cohort mean reading score at age 7

(b=30.50). The linear-slope term captured average annual

change in reading score from age 7 to age 18 (b = 12.50).

The quadratic-slope term captured deceleration of change;

that is, it captured the convexity of the trajectory across

childhood (b = −0.60). All model terms were statistically

significant (p< .001). We tested genetic influence on growth

by modeling intercept and slope terms of the growth curve

as functions of the polygenic score and covariates. Poly-

genic score coefficients measured the effect of a 1-stan-

dard-deviation difference in polygenic score on reading at

age 7 (intercept), on the linear change per year in reading

score between the ages of 7 and 18 (linear slope), and on

the deceleration of that change with increasing age (qua-

dratic slope).

Growth-curve modeling found that by age 7, children

with higher polygenic scores were already stronger read-

ers (intercept: b = 2.79, SE = 0.57, p < .001). Thereafter,

these children improved their performance at a faster rate

(linear slope: b = 0.25, SE = 0.09, p = 0.005) and reached

their peak performance at an earlier age (quadratic slope:

b = −0.03, SE = 0.01, p < .001; Fig. 3). These results show

that, on this educational fundamental, Dunedin Study

–1.0

–0.5

0.0

0.5

1.0

Adult-Attainment Factor (z Score)

–3 –2 –1 0 1 2 3

Polygenic Score

Low-SES Families

(n = 175)

0 1 2 3

Polygenic Score

Middle-SES Families

(n = 570)

Attainment Factor Z-score

0 1 2 3

Polygenic Score

High-SES Families

(n = 152)

–3 –2 –1 –3 –2 –1

Fig. 2. Scatterplots showing the association between Dunedin Study members’ polygenic scores and their adult-attainment-factor z scores,

separately for children born in low-, middle-, and high-socioeconomic status (SES) families. Each plotted point represents the mean x and y

coordinates for a bin of about 10 Dunedin Study members. The solid red line is the best-fitting regression line for the raw data. The dashed

lines show the mean level of attainment for each SES subgroup. The distribution of polygenic scores within each subgroup is shown in the

box-and-whiskers plots at the bottom of the figure. The vertical line in the center of each box marks the median, and the left and right edges

of each box correspond to the 25th and 75th percentiles, respectively. The whiskers indicate 95% of the range. The black vertical lines behind

the box plots show the cohort mean. The plots align with the scales on the x-axes of the graphs.

964 Belsky et al.

members with higher polygenic scores were often already

ahead of their peers by the second grade, and this gap in

ability tended to expand through the middle-school

years, although genetic differences were small.

Adolescents with higher polygenic scores had

higher aspirations as high school students. When

Dunedin Study members were 15 years old, they were

asked about the highest level of education they planned

to complete and also about the kind of job they hoped to

have some day. At this critical developmental juncture,

when adolescents of this New Zealand birth cohort

(1972–1973) were choosing whether to remain in school

or to begin working, adolescents in the Dunedin cohort

who had higher polygenic scores aspired to higher edu-

cational attainments (r = 0.15, p < .001; for aspiration to a

university degree, RR = 1.24, 95% CI = [1.11, 1.37]) and

more prestigious occupations (r = 0.12, p = 0.001; for

aspiration to a high-status “professional” occupation,

such as a medical doctor or engineer, RR = 1.16, 95% CI=

[1.06, 1.27]).

Adolescents with higher polygenic scores tested at

higher levels in high school. Students distinguish

themselves academically by selecting into more competi-

tive tracks and by their performance within those tracks.

At the time the Dunedin Study members were in high

school, New Zealand pupils sat for standardized exams

in the fifth, sixth, and seventh forms (ages 15–17 years).

For the 1972–1973 birth cohort, the age-15 certificate

exam was required to earn a School-Leaving Certificate

(the minimum secondary education credential at the

time); the age-16 Sixth-Form Certificate was used for

entry to various tertiary institutions; and the age-17 bur-

sary exam was the method through which the govern-

ment allocated funds (bursaries) to support living costs

during university. Dunedin Study members brought their

official exam records to the research unit, and their scores

were recorded. Adolescents with higher polygenic scores

were less likely to have left school without testing for a

credential (RR = 0.78, 95% CI = [0.66, 0.93], p = 0.006),

and were more likely to advance to the next testing level

at each age (ordered logit odds ratio = 1.32, 95%

CI=[1.12, 1.55], p = .001. They also performed better on

the tests (r = .24 for the age-15 certificate exam, p < .001;

r = .19 for the age-16 sixth-form exam, p < .001; and

r=.19 for the bursary exam, p = .032). These findings

show that adolescents with higher polygenic scores dis-

tinguished themselves from peers by more often compet-

ing at advanced academic levels and by outperforming

peers on standardized tests.

Dunedin Study members with higher polygenic

scores were more likely to pursue occupational

opportunities outside of New Zealand. Success in

competitive professional environments sometimes

depends on “going the extra mile.” The next analysis

tested whether Dunedin Study members with higher

polygenic scores did so literally, using data on where

members lived and worked from the time they were 21

years old through the end of follow-up (obtained from

life-history calendars completed by the Dunedin Study

members at each adult assessment; see the Supplemental

Material). Overseas work experience is common for New

Zealanders, including Dunedin-cohort members. By

age38, more than a third of the Dunedin cohort (42%)

had worked in a foreign country for a spell of at least 12

months. The most common destination for overseas work

experience was Australia (about 41% of those who

worked abroad did so in Australia but not elsewhere).

Work experience in a foreign country beyond Australia

has special significance in New Zealand and is known as

“the Big OE” (for overseas experience; Wikipedia, 2014).

Dunedin Study members with higher polygenic scores

were more likely to have an OE (RR = 1.17, 95% CI =

[1.05, 1.32], p = .007). Most New Zealanders who work

abroad ultimately return home to raise their families. At

the time of the age-38 interviews, 18% of Dunedin Study

members lived and worked in Australia, and an addi-

tional 7% lived and worked in another foreign country.

High Polygenic Score

Low Polygenic Score

Reading Score

100

7911 13 15 18

Age (years)

Fig. 3. Children with higher polygenic scores acquired reading skills

more rapidly. Association between age and reading skill (as mea-

sured by the Burt Word Reading Test; Scottish Council for Research

in Education, 1976), separately for children with high polygenic scores

(≥ 1 SD above the mean; n = 159) and those with low polygenic scores

(≥ 1 SD below the mean; n = 147). The shaded areas show 95% con-

fidence intervals.

The Genetics of Success 965

Dunedin Study members with higher polygenic scores

were more likely to be among these migrants (RR = 1.18,

95% CI = [1.05, 1.32], p = .005); compared with the poly-

genic scores of those living in New Zealand, scores for

migrants to Australia were higher by 0.19 SD, 95% CI =

[0.02, 0.36], p = .026, and scores for migrants to other

countries were higher by 0.27 SD, 95% CI = [0.02, 0.51],

p = .032 (Fig. 4). These findings suggest that Dunedin

Study members with higher polygenic scores distin-

guished themselves in the labor force by more often pur-

suing job opportunities beyond New Zealand.

Dunedin Study members with higher polygenic

scores were more financially planful. At ages 32 and

38, friends and relatives who knew each Dunedin Study

member well reported about the member’s ability to man-

age money (96% response rate). In addition, Dunedin

Study members were interviewed about financial building

blocks (investments and retirement savings) and saving

behaviors; scores on financial building blocks and savings

behavior scales were averaged to calculate a financial plan-

fulness score (see the Supplemental Material). Dunedin

Study members with higher polygenic scores were rated by

their informants as having fewer difficulties managing their

money (r = −.08, p = .013) and were more financially plan-

ful on average (r = .09, p = .008). These findings show that

in addition to acquiring academic credentials and profes-

sional experience to command higher earnings, Dunedin

Study members with higher polygenic scores tended to be

better managers of their financial resources.

Dunedin Study members with higher polygenic

scores selected partners with higher socioeconomic

attainments. In addition to education, wages, and

investments, so-called marriage markets contribute to a

person’s accumulation of social and financial resources

(Breen & Salazar, 2011). According to prior research, men

and women who are better-off tend to pair with one

another, and this pattern of homophilous mating also

occurs for people who are less well off (Schwartz, 2013).

By midlife, most Dunedin Study members were in a seri-

ous relationship. Dunedin Study members with higher

polygenic scores were no more likely to be in a serious

relationship than members with lower scores (RR = 1.00,

95% CI = [0.98, 1.03], p = .776). Dunedin Study members

in serious relationships were interviewed about their

partners’ education and income. This partner information

was available for 83% of the 918 Dunedin Study members

for whom we had genetic data (n = 759). Information

was used to classify partners’ SES as low (31%), middle

(49%), or high (20%; see the Supplemental Material).

Dunedin Study members with higher polygenic scores

0.20 (0.16)

0.19 (0.09)

0.60 (0.27)

Reference

0.17

(0.26)

Fig. 4. Association between polygenic score and likelihood of migrating out of New Zealand. The value at the end of each arrow is

the average standard-deviation difference in polygenic scores (with the standard error of the estimate in parentheses) between Dunedin

Study members who moved to that area (North America, n = 14; Europe, n = 41; Asia and Africa, n = 13; Australia, n = 162) and members

who remained in or returned to New Zealand. Migrants were defined as Dunedin Study members who had lived and worked abroad

for a minimum of 12 months since the age of 21 and who were still living abroad at the age-38 assessment.

966 Belsky et al.

tended to have partners with higher SES (r = .09, p = .011;

Fig. 5). These findings suggest that Dunedin Study mem-

bers with higher polygenic scores bolstered the socio-

economic advantages they accrued through their own

educational and occupational attainments by partnering

with socially advantaged mates.

Dunedin Study members with higher polygenic

scores were not more satisfied with their lives.

A higher polygenic score predicted conventional indica-

tors of success: educational achievement, occupational

prestige, financial security, even securing a socioeconomi-

cally successful partner. Yet some conceptualizations of

success extend beyond the realms of material and social

attainment. We therefore tested whether the polygenic

score predicted Dunedin Study members’ self-rated satis-

faction with life at age 38. It did not (r = .04, p = .189).

Genetic associations with pathways to socioeconomic

success were not accounted for by Dunedin Study

members’ social origins. Because of evidence that

Dunedin Study children’s polygenic scores were associated

with their families’ socioeconomic circumstances (r = .13,

p < .001), Part 2 analyses presented in this section were

repeated with statistical adjustment for the SES of Dunedin

Study members’ families when the members were children.

Genetic associations were largely independent of childhood

SES. Complete results are included in Table S3 in the Sup-

plemental Material.

Part 3: What personal characteristics

helped children with higher polygenic

scores achieve social and economic

success?

The pattern of findings described previously suggests

that the genetics uncovered in GWASs of educational

attainment contribute to certain underlying characteris-

tics that influence not only educational success, but also

success in broader social and economic domains of life.

We tested three different characteristics that might func-

tion as mediators of genetic influence on success in mul-

tiple life domains. These characteristics were higher

cognitive ability, stronger noncognitive skills, and overall

better physical health.

Children with higher polygenic scores performed

better on IQ tests and exhibited a more rapid pace

of cognitive development during childhood. Chil-

dren with higher polygenic scores did not score signifi-

cantly higher than their peers on the Peabody test at age3

(r = .05, p = .133), but thereafter they showed an increas-

ing cognitive advantage (r = 0.13 for Stanford-Binet IQ at

age 5; r = .13–.19 for WISC-R IQ at ages 7–13; all ps < .001;

Fig. 6a).

This pattern of findings indicates genetic influence over

the developmental process through which children accu-

mulate cognitive abilities, a hypothesis suggested by previ-

ous twin research on intelligence (Plomin, 2012) but, to

our knowledge, still untested in molecular data. To test

hypotheses about polygenic influence on the course of

cognitive development, data from repeated WISC-R assess-

ments were analyzed. This analysis focused on mental-age

scores, rather than IQ scores, because, whereas IQ scores

are age-corrected in order to allow comparisons between

a child and the population of children of the same chrono-

logical age (e.g., a student’s score is in the 66th percentile

for his or her age), mental-age scores express the child’s

level of performance as the chronological age for which

his or her score is normative (e.g., a 10-year-old student

might have a mental age of 12). Mental age can be used to

monitor a child’s intraindividual development over time

(e.g., a 10-year-old child with an unstandardized IQ score

equal to the average unstandardized score for 12-year-olds

would have a mental age of 12; Lezak, Howieson, Loring,

Hannay, & Fischer, 2004).

Growth-curve modeling tested whether the cognitive

development of children with higher polygenic scores dif-

fered from that of their peers (see the Supplemental Mate-

rial). The model intercept captured the cohort mean mental

age at a chronological age of 7 years (b = 7). The linear-

slope term captured average annual change in mental age

(b=1). Model terms were statistically significant (p<.001).

We tested genetic influence on growth by modeling

100

Percentage

Low Average High

Polygenic Score

High

Middle

Low

Partner SES

Fig. 5. Association between polygenic score and partner’s socioeco-

nomic status (SES). The graph shows the percentages (inside bars) of

members who had low-, middle-, and high-SES partners, separately for

Dunedin Study members with low polygenic scores (≥ 1 SD below the

mean; n = 119), average polygenic scores (within 1 SD of the mean;

n = 504), and high polygenic scores (≥ 1 SD above the mean; n = 136).

Partners’ SES was defined according to whether they had completed

a university degree and whether their income was above the national

sex-specific median: High-SES partners had a university education and

an above-median income, middle-SES partners met only one of these

criteria, and low-SES partners met neither criterion.

The Genetics of Success 967

0.77

1.80

1.88

2.38

2.71

2.39

Effect Size (IQ Points)

Chronological Age (years)

Mental Age (years)

7911 13

Chronological Age (years)

High Polygenic Score

Low Polygenic Score

357 13119

Fig. 6. Association between polygenic score and cognitive ability. The plotted points in (a) show the magnitude of the effect of a 1-SD increase in

polygenic score on standardized IQ (1 IQ point = 1/15 of 1 SD) measured at ages 3, 5, 7, 9, 11, and 13. Error bars indicate 95% confidence intervals.

Cognitive ability was measured with the Peabody Picture Vocabulary Test (Dunn, 1965) at age 3, the Stanford-Binet Intelligence Scale (Terman &

Merrill, 1960) at age 5, and Wechsler Intelligence Scales for Children–Revised (WISC-R; Wechsler, 1974) at ages 7–13. In (b), mental age is graphed

as a function of chronological age for children with high polygenic scores (≥ 1 SD above the mean; n = 159) and those with low polygenic scores

(≥ 1 SD below the mean; n = 147). The shaded areas show 95% confidence intervals. Mental age was measured with the WISC-R.

intercept and slope terms of the growth curve as functions

of the polygenic score and covariates. Polygenic-score

coefficients measure the effect of a 1-standard-deviation

difference in polygenic score on mental age at chronologi-

cal age 7 (intercept), and on the linear change per year in

mental age from chronological age 7 to 13 (linear slope).

Children with higher polygenic scores tended to have

older mental ages at the chronological age-7 baseline

(intercept: b = 0.13, SE = 0.04, p < .001), and they exhib-

ited a faster pace of cognitive development through age

13 years (slope: b = 0.05, SE = 0.01, p < .001; Fig. 6b).

Taken together, these effects mean that a child with a

genetic score 1 standard deviation above the mean would,

by the age of 13 years, accrue a 6-month advantage in

cognitive development relative to the population norm.

Children with higher polygenic scores had stronger

noncognitive skills. In addition to cognitive abilities, so-

called noncognitive skills influence individuals’ attainments

(Heckman, 2006). Genetic associations were tested for two

noncognitive skills, self-control and interpersonal skill.

As described previously (Moffitt etal., 2011), dossiers

of children’s self-control skills were compiled from obser-

vational ratings and from parent and teacher reports

when the children were between ages 3 and 11 years old,

and from self-reports when the children were 11 years

old. Children with higher polygenic scores tended to

show better self-control skills across their first decade of

life (r = .10, p = .001).

Children’s interpersonal skill was measured from

reports by trained research staff on behavioral observa-

tions of the Dunedin Study members at ages 3, 5, 7, and

9 years. At each age, children were given binary ratings if

they impressed the staff as being friendly, confident,

cooperative, or communicative. These ratings were used

to form an interpersonal skill scale (see the Supplemental

Material). Children with higher polygenic scores were

rated as having better interpersonal skill (r = .10, p = .004).

Genetic associations with children’s cognitive abil-

ities and noncognitive skills were independent of

their social origins. Analysis of childhood psychologi-

cal characteristics was repeated with statistical adjust-

ment for the SES of the children’s families. Genetic

associations were found to be independent of childhood

SES. Complete results are included in Table S3 in the

Supplemental Material.

Cognitive abilities and noncognitive skills mediated

genetic influences on educational and socioeco-

nomic attainments. Genetic associations with cognitive

and noncognitive skills suggest that these characteristics

968 Belsky et al.

could explain why children with higher polygenic scores

went on to achieve higher educational and socioeconomic

attainments. Mediation analyses tested whether cognitive

abilities and noncognitive skills accounted for genetic asso-

ciations with life attainments (see Figs. S7 and S8 and Table

S4 in the Supplemental Material). Cognitive ability, self-con-

trol, and interpersonal skill were all statistically significant

mediators of genetic associations with educational and

socioeconomic outcomes. Together, cognitive abilities and

noncognitive skills accounted for about 60% of the genetic

association with educational attainment and about 47% of

the genetic association with the adult-attainment-factor

score (p<.001 for both).

Children with higher polygenic scores were no

healthier than their peers. Genetic associations with

adult attainments might also result from general benefits

to physical integrity that make individuals healthier as

children, setting them up for success later in life (Case,

Fertig, & Paxson, 2005). Dunedin Study members’ health

was measured from repeated clinical assessments of

motor development, growth and obesity, cardiovascular

and pulmonary functioning, and infections and injuries

when the children were between the ages of 3 and 11

years (see the Supplemental Material). Dunedin Study

members with higher polygenic scores were no healthier

in childhood than their peers (r = .01, p = .806). Together

with the aforementioned lack of association between the

polygenic score and walking, feeding, and potty training,

this finding suggests that GWASs of educational attain-

ment have not identified a set of genetic influences on

overall robust functioning of the body’s physical systems.

As a second test of the physical-robustness hypothesis,

we analyzed the genetics of human height. Like educa-

tion, human height is known to be related to socioeco-

nomic attainments (Case & Paxson, 2008). For this analysis,

we substituted a polygenic score derived from GWASs of

human height for the education polygenic score in our

original analysis predicting life attainments. We used pub-

lished results from a large-scale GWAS of human height

(Wood etal., 2014) to calculate height polygenic scores

for Dunedin Study members. As expected, Dunedin Study

members’ polygenic scores for height were correlated

with their measured stature (r=.54, p < .001). However,

even though taller members did tend to do better in life

(adult-attainment factor: r=.13, p = .011), we observed

no association between the polygenic score for height

and life attainments measured by the adult-attainment

factor (r = .00, p = .952).

Discussion

This article describes how genetic discoveries made in

GWAS analysis of educational attainment were related to

the courses of human lives. We studied a population-

representative birth cohort followed over the course of

four decades. Findings showed that genome-wide DNA-

sequence differences identified from GWASs and sum-

marized in a “polygenic score” were associated with basic

processes of human social and economic success. Three

points are important in interpreting the substance of

these findings. First, genetic associations between the

polygenic score and adult socioeconomic success were

not fully accounted for by educational attainment. Sec-

ond, children’s socioeconomic origins were correlated

with their polygenic scores; however, genetic associations

with adult socioeconomic success, with the developmen-

tal and behavioral pathways to such success, and with the

psychological characteristics we studied were mostly

independent of children’s socioeconomic origins. Third,

across the board, effect sizes were small in magnitude.

The primary finding was that polygenic scores derived

from a GWAS of educational attainment predicted life

outcomes well beyond schooling. Dunedin Study mem-

bers with higher polygenic scores were geographically

mobile in search of professional opportunities, they built

more successful careers, they secured higher social status

mates, and they built stronger financial foundations for

retirement. From childhood to midlife, Dunedin Study

members’ genetic inheritance predicted their social mobil-

ity. Even among children born into socially disadvantaged

homes, those with higher polygenic scores achieved

more. Achievements of children with higher polygenic

scores were enabled in part by a suite of psychological

traits already evident from early life. Dunedin Study mem-

bers with higher polygenic scores talked earlier, did better

on cognitive tests beginning at 5 years old, and showed a

more rapid pace of cognitive development, and they

developed better self-control and interpersonal skills. Col-

lectively, these childhood psychological characteristics

accounted for about half of the genetic association with

social success in adulthood. Strikingly, the same genetic

differences that predicted children’s cognitive, emotional,

and social functioning were not related to their attainment

of nonverbal milestones or their physical health.

The substance of these findings is bolstered by evi-

dence that GWAS discoveries for educational attainment

are not genetic artifacts of a socially privileged class.

Because children born into better-off families are more

likely to earn advanced degrees (Breen & Jonsson, 2005),

a GWAS of educational attainment could have identified

the genetics of better-off families rather than the genetics

of a propensity to succeed. GWAS discoveries could be

no more than markers of socially advantaged ancestry.

Consistent with such a possibility, both previous studies

(Conley etal., 2015; Domingue etal., 2015; Krapohl &

Plomin, 2016) and the current study found that children

born into better-off homes had higher polygenic scores.

The Genetics of Success 969

But two findings suggest that the genetic associations are

nonspurious. First, studies that compare siblings within

the same family (who share identical ancestries) find that

the sibling with the higher polygenic score tends to com-

plete more years of schooling (Domingue et al., 2015;

Rietveld, Conley, etal., 2014). Second, our study shows

that polygenic scores also influence changes in social

position within a single generation, thereby suggesting a

mechanism to explain the gene-environment correlation

in which children of socially advantaged families tend to

have higher polygenic scores.

We acknowledge limitations. First, our study con-

cerned a single birth cohort of European descent in one

country, New Zealand. The extent to which findings gen-

eralize to other birth cohorts growing up under other

circumstances needs to be tested. Although New Zealand

has levels of social inequality similar to those in the

United States and Great Britain (after-tax Gini coeffi-

cients: New Zealand, .33; United Kingdom, .34; United

States, .37; Wikipedia, 2015), international comparisons

will prove informative (Tucker-Drob & Bates, 2016),

including in settings in which inequality is engineered to

be low (Firkowska et al., 1978). Second, the measure-

ment of the human genome we studied is necessarily

preliminary. We studied a polygenic score based on the

best available information about genetic correlates of

educational success. But future GWASs with larger sam-

ple sizes are expected to yield a more precise set of

genetic correlates. Replication checks with subsequent

iterations of the polygenic score for education are

needed. Third, follow-up of social and economic out-

comes in our study is right censored, extending through

the fourth decade of life, but not beyond. Extension of

findings into longitudinal cohort studies of older adults is

needed to clarify the extent of genetic associations into

the second half of the life course. Finally, the set of out-

comes, pathways, and traits that we studied is not com-

prehensive. Studies of other samples that use different

measurement batteries are needed to expand our under-

standing of how genetic correlates of educational attain-

ment relate to human life courses.

In light of these limitations, our study contributes to

public and scientific conversation about genetic discover-

ies regarding educational attainment in five ways. First,

GWAS discoveries regarding educational attainment are

not only about education. They are discoveries about

socioeconomic success more broadly (although perhaps

not about satisfaction with life). Education accounted for

under half of the relationship between genes and adult

socioeconomic attainments, suggesting that the mecha-

nisms of genetic influence are not limited to success in

schooling and do not depend on it.

Second, the psychological mediators of genetic asso-

ciations with socioeconomic success involve more than

what IQ tests measure as intelligence. Multivariate twin

research suggests that the heritability of educational

attainment reflects genetic influences on noncognitive

skills as well as intelligence (Krapohl et al., 2014). We

found molecular evidence to support this hypothesis:

Children’s polygenic scores for educational attainment

were correlated with their noncognitive self-control and

interpersonal skills as well as with their IQ scores. Our

“top-down” approach, working from an adult phenotype

backward in development toward a DNA sequence,

yielded findings that suggest behavioral mechanisms for

genetic influences on educational attainment.

Third, children with higher polygenic scores grew

apart from their peers along coherent developmental tra-

jectories that began to form even before they entered

school. Dunedin Study members with higher polygenic

scores began to talk at a younger age. Subsequently, they

learned to read before many of their peers did. This early

success was followed by loftier academic aspirations and

attainments extending into adulthood. These findings

support the logic of interventions to promote early liter-

acy, particularly those focusing on early language devel-

opment (Talbot, 2015).

In addition, and more speculatively, the life-course anal-

ysis that we report also suggests that GWAS findings for

educational attainment may provide a clue to the genetic

roots of life-history differences in free-living humans.

Unlike education, which is a relatively modern human

experience, patterns of migration, mate selection, and

resource acquisition and management are ancient human

behaviors that plausibly bear the imprint of our species’

evolutionary history. The finding that GWAS discoveries for

education predict these ancient behaviors suggests a win-

dow into genetic regulation of humans’ strategies to survive

and reproduce. Our data cannot reveal whether frequen-

cies of education-associated genotypes reflect some

Darwinian fitness strategy. Rather, the data suggest that

individuals whose genomes carry more education-associ-

ated alleles are forging life histories that achieve success in

the modern world, and the pathways to this success include

some that would be familiar to our ancestors.

Fourth, the current findings lend weight to earlier twin-

study observations that genes shape not just behavior, but

also the environment that contextualizes and constrains

behavioral choices (Plomin & Bergeman, 1991). The

molecular realization of such gene-environment correla-

tions creates opportunities for social theory and research.

Results reported in this study suggest that by incorporat-

ing DNA sequence into studies of status attainment,

migration, assortative mating, and financial behavior,

social scientists may be able to frame novel “socioge-

nomic” research questions. For example, do public pro-

grams to build human capital (such as improving teacher

salaries or providing universal access to prekindergarten

970 Belsky et al.

education) change the ways in which genes influence life

attainments? If so, are the returns greater for programs

that magnify genetic influences or for programs that

reduce them? Do the genetics of educational attainment

relate to social gradients in midlife health and aging? If so,

how is this process shaped by health-care costs, quality,

and access? As concerns about economic inequality

increase, are genes linked with socioeconomic success

becoming concentrated within social and geospatial

elites? If so, is this process influenced by exogenous

shocks such as natural disasters, policy shifts such as mul-

tinational trade and border agreements, or cultural

changes in equality of opportunity?

Finally, our findings shed light on the stakes of the pub-

lic conversation about sociogenomic discoveries that is

now emerging. For the general public, the significance of

new knowledge about how to measure and interpret DNA

sequence is uncertain and hotly debated, even in the field

of biomedicine, in which clinical applications of genetic

discoveries are already possible (Khoury & Evans, 2015;

Lander, 2015; Roberts etal., 2012). At present, genetic pre-

diction of educational outcomes and life success in general

is far from sensitive or specific enough to recommend any

translational application. Although there is movement

toward improving the predictive power of polygenic scores

through increased GWAS sample sizes and improved

genomic measurements, a precision medicine-type

approach to human capital development remains well out

of reach. And yet debate is already under way about the

possibility for genetic testing to someday be used in fore-

casting human potential. Policy action may be needed to

regulate the ethical use of genomic information in school

admissions and tracking decisions, and such actions

should be informed by realistic estimates of the magnitude

of genetic effects.

Action Editor

Ian H. Gotlib served as action editor for this article.

Author Contributions

D. W. Belsky, T. E. Moffitt, and A. Caspi conceived the research

and wrote the manuscript. T. E. Moffitt, A. Caspi, H. Harrington,

S. Hogan, S. Ramrakha, and R. Poulton collected the data. D. L.

Corcoran, K. Sugden, and B. S. Williams prepared the genotype

data. D. W. Belsky, T. E. Moffitt, D. L. Corcoran, H. Harrington,

R. Houts, and A. Caspi analyzed the data. All the authors

reviewed drafts, provided critical feedback, and approved the

final manuscript.

Acknowledgments

We thank the Dunedin Study members, along with their par-

ents, teachers, partners, and peer informants, and study founder

Phil Silva.

Declaration of Conflicting Interests

The authors declared that they had no conflicts of interest with

respect to their authorship or the publication of this article.

Funding

The Dunedin Multidisciplinary Health and Development

Research Unit is supported by the New Zealand Health Research

Council and the New Zealand Ministry of Business, Innovation

and Employment (MBIE). This research was supported by

National Institute on Aging Grants R01-AG032282, R01-AG048895,

and 1R01-AG049789, United Kingdom Medical Research Council

Grant MR/K00381X, and United Kingdom Economic and Social

Research Council Grant ES/M010309/1. Additional support was

provided by National Institute on Aging Grant P30-AG028716, by

Eunice Kennedy Shriver National Institute of Child Health and

Human Development Grant R21-HD078031, and by the Jacobs

Foundation. D. W. Belsky is supported by an Early-Career

Research Fellowship from the Jacobs Foundation.

Supplemental Material

Additional supporting information can be found at http://pss

.sagepub.com/content/by/supplemental-data

Open Practices

The analysis plan for this study can be found at https://docs

.google.com/viewer?a=v&pid=sites&srcid=ZGVmYXV

sdGRvbWFpbnxkdW5lZGluZXJpc2tjb25jZXB0cGFwZXJzfG

d4Ojc5NTA0YjNhNjMzYzY3YmE. The data are not publicly

available because (a) the size of the sample and the nature of

the data might make it possible to identify participants and (b)

consent for broad sharing of the data was not obtained from the

participants. However, the Supplemental Material provides

details about possible data sharing. The complete Open Prac-

tices Disclosure for this article can be found at http://pss.sage

pub.com/content/by/supplemental-data.

References

1000 Genomes Project. (2016). 1000 Genomes project data.

Retrieved from http://www.1000genomes.org/

Baron, R. M., & Kenny, D. A. (1986). The moderator–medi-

ator variable distinction in social psychological research:

Conceptual, strategic, and statistical considerations. Journal

of Personality and Social Psychology, 51, 1173–1182.

doi:10.1037/0022-3514.51.6.1173

Belsky, D. W., Moffitt, T. E., & Caspi, A. (2013). Genetics in

population health science: Strategies and opportunities.

American Journal of Public Health, 103(Suppl. 1), S73–S83.

doi:10.2105/AJPH.2012.301139

Breen, R., & Jonsson, J. O. (2005). Inequality of opportunity in

comparative perspective: Recent research on educational

attainment and social mobility. Annual Review of Sociology,

31, 223–243. doi:10.1146/annurev.soc.31.041304.122232

Breen, R., & Salazar, L. (2011). Educational assortative mating

and earnings inequality in the United States. American

Journal of Sociology, 117, 808–843. doi:10.1086/661778

The Genetics of Success 971

Case, A., Fertig, A., & Paxson, C. (2005). The lasting impact

of childhood health and circumstance. Journal of Health

Economics, 24, 365–389. doi:10.1016/j.jhealeco.2004.09.008

Case, A., & Paxson, C. (2008). Stature and status: Height, ability,

and labor market outcomes. Journal of Political Economy,

116, 499–532. doi:10.1086/589524

Chabris, C. F., Lee, J. J., Cesarini, D., Benjamin, D. J., &

Laibson, D. I. (2015). The fourth law of behavior genetics.

Current Directions in Psychological Science, 24, 304–312.

doi:10.1177/0963721415580430

Conley, D., Domingue, B., Cesarini, D., Dawes, C. T., Rietveld,

C. A., & Boardman, J. (2015). Is the effect of parental edu-

cation on offspring biased or moderated by genotype?

Sociological Science, 2, 82–105. doi:10.15195/v2.a6

de Zeeuw, E. L., van Beijsterveldt, C. E. M., Glasner, T. J.,

Bartels, M., Ehli, E. A., Davies, G. E., . . . Boomsma, D. I.

(2014). Polygenic scores associated with educational

attainment in adults predict educational achievement and

ADHD symptoms in children. American Journal of Medical

Genetics Part B: Neuropsychiatric Genetics, 165B, 510–520.

doi:10.1002/ajmg.b.32254

Domingue, B. W., Belsky, D. W., Conley, D., Harris, K. M.,

& Boardman, J. D. (2015). Polygenic influence on edu-

cational attainment. AERA Open, 1(3), 1–13. doi:10.1177/

2332858415599972

Dudbridge, F. (2013). Power and predictive accuracy of poly-

genic risk scores. PLoS Genetics, 9(3), Article e1003348.

doi:10.1371/journal.pgen.1003348

Dunn, L. (1965). The Peabody Picture Vocabulary Test.

Minneapolis, MN: American Guidance Service.

Euesden, J., Lewis, C. M., & O’Reilly, P. F. (2015). PRSice:

Polygenic Risk Score software. Bioinformatics, 31, 1466–

1468. doi:10.1093/bioinformatics/btu848

Firkowska, A. N., Ostrowska, A., Sokolowska, M., Stein, Z., Susser,

M., & Wald, I. (1978). Cognitive development and social pol-

icy. Science, 200, 1357–1362. doi:10.1126/science.663616

Heckman, J. J. (2006). Skill formation and the economics of

investing in disadvantaged children. Science, 312, 1900–

1902. doi:10.1126/science.1128898

Henig, R. M. (2015, December 11). Are there genes for intel-

ligence—And is it racist to ask? Retrieved from http://

news.nationalgeographic.com/2015/12/151211-genetics-

intelligence-racism-science/

Howie, B. N., Donnelly, P., & Marchini, J. (2009). A flexible and

accurate genotype imputation method for the next genera-

tion of genome-wide association studies. PLoS Genetics

5(6), Article e1000529. doi:10.1371/journal.pgen.1000529

Khoury, M. J., & Evans, J. P. (2015). A public health perspective

on a national precision medicine cohort: Balancing long-

term knowledge generation with early health benefit. The

Journal of the American Medical Association, 313, 2117–

2118. doi:10.1001/jama.2015.3382

Krapohl, E., & Plomin, R. (2016). Genetic link between fam-

ily socioeconomic status and children’s educational

achievement estimated from genome-wide SNPs. Molecular

Psychiatry, 21, 437–443. doi:10.1038/mp.2015.2

Krapohl, E., Rimfeld, K., Shakeshaft, N. G., Trzaskowski, M.,

McMillan, A., Pingault, J.-B., . . . Plomin, R. (2014). The

high heritability of educational achievement reflects

many genetically influenced traits, not just intelligence.

Proceedings of the National Academy of Sciences, USA, 111,

15273–15278. doi:10.1073/pnas.1408777111

Lander, E. S. (2015). Cutting the Gordian helix—regulating

genomic testing in the era of precision medicine. The New

England Journal of Medicine, 372, 1185–1186. doi:10.1056/

NEJMp1501964

Lezak, D. M., Howieson, D. B., Loring, D. W., Hannay, H. J.,

& Fischer, J. S. (2004). Neuropsychological assessment (4th

ed.). New York, NY: Oxford University Press.

Moffitt, T. E., Arseneault, L., Belsky, D., Dickson, N., Hancox,

R. J., Harrington, H., . . . Caspi, A. (2011). A gradient of

childhood self-control predicts health, wealth, and public

safety. Proceedings of the National Academy of Sciences,

USA, 108, 2693–2698. doi:10.1073/pnas.1010076108

Pavot, W., & Diener, E. (1993). Review of the Satisfaction With

Life Scale. Psychological Assessment, 5, 164–172.

Plomin, R. (2012). Genetics: How intelligence changes with

age. Nature, 482, 165–166. doi:10.1038/482165a

Plomin, R., & Bergeman, C. S. (1991). The nature of nur-

ture: Genetic influence on “environmental” measures.

Behavioral & Brain Sciences, 14, 414–427. doi:10.1017/

S0140525X00070588

Poulton, R., Moffitt, T. E., & Silva, P. A. (2015). The Dunedin

Multidisciplinary Health and Development Study: Overview

of the first 40 years, with an eye to the future. Social

Psychiatry & Psychiatric Epidemiology, 50, 679–693.

doi:10.1007/s00127-015-1048-8

Preacher, K. J., & Hayes, A. F. (2008). Asymptotic and resam-

pling strategies for assessing and comparing indirect effects

in multiple mediator models. Behavior Research Methods,

40, 879–891. doi:10.3758/BRM.40.3.879

Preacher, K. J., & Kelley, K. (2011). Effect size measures for

mediation models: Quantitative strategies for communi-

cating indirect effects. Psychological Methods, 16, 93–115.

doi:10.1037/A0022658

Price, A. L., Patterson, N. J., Plenge, R. M., Weinblatt, M. E.,

Shadick, N. A., & Reich, D. (2006). Principal components

analysis corrects for stratification in genome-wide associa-

tion studies. Nature Genetics, 38, 904–909. doi:10.1038/

ng1847

Price, A. L., Zaitlen, N. A., Reich, D., & Patterson, N. (2010).

New approaches to population stratification in genome-

wide association studies. Nature Reviews Genetics, 11, 459–

463. doi:10.1038/nrg2813

Rietveld, C. A., Conley, D., Eriksson, N., Esko, T., Medland,

S. E., Vinkhuyzen, A. A. E., . . . Social Science Genetics

Association Consortium. (2014). Replicability and robust-

ness of genome-wide-association studies for behavioral

traits. Psychological Science, 25, 1975–1986. doi:10.1177/

0956797614545132

Rietveld, C. A., Esko, T., Davies, G., Pers, T. H., Turley, P.,

Benyamin, B., . . . Koellinger, P. D. (2014). Common genetic

variants associated with cognitive performance identified

using the proxy-phenotype method. Proceedings of the

National Academy of Sciences, USA, 111, 13790–13794.

doi:10.1073/pnas.1404623111

Rietveld, C. A., Medland, S. E., Derringer, J., Yang, J., Esko, T.,

Martin, N. W., . . . Koellinger, P. D. (2013). GWAS of 126,559

972 Belsky et al.

individuals identifies genetic variants associated with educa-

tional attainment. Science, 340, 1467–1471. doi:10.1126/science

.1235488

Roberts, N. J., Vogelstein, J. T., Parmigiani, G., Kinzler, K. W.,

Vogelstein, B., & Velculescu, V. E. (2012). The predic-

tive capacity of personal genome sequencing. Science

Translational Medicine, 4(133), Article 133ra58. doi:10.1126/

scitranslmed.3003380

Schwartz, C. R. (2013). Trends and variation in assortative mat-

ing: Causes and consequences. Annual Review of Sociology,

39, 451–470. doi:10.1146/annurev-soc-071312-145544

Scottish Council for Research in Education. (1976). Burt Word

Reading Test (rev. ed.). London, England: Hodder &

Stoughton.

Sherry, S. S., Ward, M. M., Kholodov, M., Baker, J., Phan, L.,

Smigielski, E. E., Sirotkin, K. (2001). dbSNP: The NCBI data-

base of genetic variation. Nucleic Acids Research, 29, 308–311.

Singer, J. D., & Willett, J. B. (2003). Applied longitudinal data

analysis. New York, NY: Oxford University Press.

Talbot, M. (2015, January 12). The talking cure. The New Yorker.

Retrieved from http://www.newyorker.com/magazine/

2015/01/12/talking-cure

Terman, L. M., & Merrill, M. A. (1960). Stanford-Binet

Intelligence Scale: Manual for the third revision. Oxford,

England: Houghton Mifflin.

Tucker-Drob, E. M., & Bates, T. C. (2016). Large cross-national dif-

ferences in gene × socioeconomic status interaction on intel-

ligence. Psychological Science, 27, 138–149. doi:10.1177/

0956797615612727

Ward, M. E., McMahon, G., St Pourcain, B., Evans, D. M.,

Rietveld, C. A., Benjamin, D. J., . . . Timpson, N. J. (2014).

Genetic variation associated with differential educational

attainment in adults has anticipated associations with

school performance in children. PLoS ONE, 9(7), Article

e100248. doi:10.1371/journal.pone.0100248

Wechsler, D. (1974). Manual for the Wechsler Intelligence

Scale for Children–Revised. New York, NY: Psychological

Corporation.

Wikipedia. (2014). Overseas experience. Retrieved from

http://en.wikipedia.org/w/index.php?title=Overseas_

experience&oldid=633298303

Wikipedia. (2015). List of countries by income equality. Retrieved

from https://en.wikipedia.org/w/index.php?title=List_of_

countries_by_income_equality&oldid=686250962

Wood, A. R., Esko, T., Yang, J., Vedantam, S., Pers, T. H.,

Gustafsson, S., . . . Frayling, T. M. (2014). Defining the

role of common variation in the genomic and biological

architecture of adult human height. Nature Genetics, 46,

1173–1186. doi:10.1038/ng.3097