Genetic and Environmental Influences on the Development of Intelligence

Article (PDF Available)inBehavior Genetics 32(4):237-49 · August 2002with2,778 Reads
DOI: 10.1023/A:1019772628912 · Source: PubMed
Abstract
Measures of intelligence were collected in 209 twin pairs at 5, 7, 10, and 12 years of age, as part of a longitudinal project on intelligence, brain function, and behavioral problems. Intelligence was measured at 5, 7, and 10 years of age with the RAKIT, a well-known Dutch intelligence test, consisting of 6 subscales. At 12 years of age, the complete WISC-R was administered (12 subscales). Both intelligence tests resulted in a measure of full-scale IQ (FSIQ). Participation rate is around 93% at age 12. Correlation coefficients over time are high: (r(5-7) = .65; r(5-10) = .65; r(5-12) = .64; r(7-10) = .72; r(7-12) = .69 and r(10-12) = .78). Genetic analyses show significant heritabilities at all ages, with the expected increase of genetic influences and decrease of shared environmental influences over the years. Genetic influences seem to be the main driving force behind continuity in general cognitive ability, represented by a common factor influencing FSIQ at all ages. Shared environmental influences are responsible for stability as well as change in the development of cognitive abilities, represented by a common factor influencing FSIQ at all ages and age-specific influences, respectively.
1 Figures
0001-8244/02/0700-0237/0 © 2002 Plenum Publishing Corporation
Behavior Genetics, Vol. 32, No. 4, July 2002 (© 2002)
237
Genetic and Environmental Influences on the Development
of Intelligence
M. Bartels,
1,2
M. J. H. Rietveld,
1
G. C. M. Van Baal,
1
and D. I. Boomsma
1
Received 11 Jan. 2002—Final 24 Apr. 2002
Measures of intelligence were collected in 209 twin pairs at 5, 7, 10, and 12 years of age, as part
of a longitudinal project on intelligence, brain function, and behavioral problems. Intelligence
was measured at 5, 7, and 10 years of age with the RAKIT, a well-known Dutch intelligence test,
consisting of 6 subscales. At 12 years of age, the complete WISC-R was administered (12 sub-
scales). Both intelligence tests resulted in a measure of full-scale IQ (FSIQ). Participation rate is
around 93% at age 12. Correlation coefficients over time are high: (r(5–7) .65; r(5–10) .65;
r(5–12) .64; r(7–10) .72; r(7–12) .69 and r(10–12) .78). Genetic analyses show sig-
nificant heritabilities at all ages, with the expected increase of genetic influences and decrease of
shared environmental influences over the years. Genetic influences seem to be the main driving
force behind continuity in general cognitive ability, represented by a common factor influencing
FSIQ at all ages. Shared environmental influences are responsible for stability as well as change
in the development of cognitive abilities, represented by a common factor influencing FSIQ at
all ages and age-specific influences, respectively.
KEY WORDS: General cognitive ability; longitudinal analyses; heritability; twin study; simplex.
1993; Boomsma, 1993; Plomin et al., 1997; Boomsma
and Van Baal, 1998; Alarcón, 1998, 1999). A few lon-
gitudinal studies have focused on the influences of
genes and environment on cognitive development
rather than cognition at specific ages. New genetic in-
fluences at different ages and a common factor for
shared environmental influences have been found (Col-
orado Adoption Project; e.g., Plomin and DeFries,
1985; Louisville Twin Study; e.g., Wilson, 1983; Eaves
et al., 1986).
Longitudinal twin and family data allow the study
of persistence and change of genetic, shared environ-
mental, and unique environmental influences. The ge-
netic and environmental influences may exert their
effects following several possible mechanisms. First, ge-
netic or environmental factors may exert a continuous
influences from their time of onset (common factor in-
fluences). This mechanism implies that the same genetic
or environmental factors are responsible for stability,
possibly with age-dependent factor loadings. Second, ge-
netic and environmental influences may be specific at a
certain age and exert an effect on cognition at that age
INTRODUCTION
Heritability of intelligence has been studied exten-
sively, both in adults and in children, but far less is
known about the developmental genetics of cognitive
abilities. Many behavior genetic studies yield the
largely consistent result that genetic differences account
for at least 50% of the observed variability in cogni-
tion in adults (e.g., Bouchard and McGue, 1981;
McCartney et al., 1990; Bratko, 1996; Rijsdijk et al.,
1997, 1998; Alarcón et al., 1998, 1999, Posthuma et al.,
2000). It is also well established that the genetic influ-
ences on cognitive functioning increase throughout de-
velopment, whereas influences of common environ-
ment decrease (e.g., Skodak and Skeels, 1949; Wilson,
1983; Labuda et al., 1986; Fulker et al., 1988; Loehlin
et al., 1989; McCartney et al., 1990; McGue et al.,
1
Department of Biological Psychology, Vrije Universiteit, Amster-
dam, The Netherlands.
2
To whom correspondence should be addressed at Department of
Biological Psychology, Vrije Universiteit, room 1F 57, van der Boe-
chorststraat 1, 1081 BT, Amsterdam, The Netherlands. Fax: 31
20 4448812. e-mail: m.bartels@psy.vu.nl
only. Change in cognitive development may be due to
these age specific factors. Finally, there can be a sim-
plex-like continuity in genetic and environmental effects
(Eaves et al., 1986; Boomsma and Molenaar, 1987). In
this simplex-like continuity, there are effects specific to
each age and there are “carry-over effects” or transmis-
sion effects from one age to the subsequent age (Fig. 1).
In other words, earlier influences may be transmitted
from one occasion to the next and new influences (in-
novations) may come into play at each occasion. Data
that are collected from the same subjects repeatedly in
time often display this simplex structure for the observed
correlations among the measures at different time points.
Specifically, it is observed that correlations are highest
among adjoining occasions and that they decrease sys-
tematically as the distance between time points increases
(Guttman, 1954).
Notable longitudinal studies on cognition are the
Colorado Adoption Project (CAP) (e.g., Plomin and
DeFries, 1985) and the Louisville Twin Study (LTS)
(e.g., Wilson, 1983). These studies are more or less
comparable to the current study, in which intelligence
is assessed longitudinally in twins from age 5 to age
12. We will introduce the CAP and LTS and also men-
tion other studies, which offer an insight in the genetic
and environmental patterns that account for variance in
cognitive development (for reviews, see also Thomp-
son, 1993; Patrick, 2000).
238 Bartels, Rietveld, Van Baal, and Boomsma
The CAP is a longitudinal “full” adoption study
of behavioral development. The study started in 1975
and included adopted children and their adoptive and
biological parents. Children in the sample were tested
yearly on age-appropriate cognitive measures. Until
now, longitudinal results from 1 to 16 years of age have
been published for the CAP. The CAP original sample
consisted of 245 adoptive families and 245 nonadop-
tive control families. In 1999, the CAP sample con-
sisted of 129 adopted individuals tested at 16 years of
age and their adoptive and biological parents. The non-
adoptive (control) sample included 125 sets of parents
and nonadoptive children (Alarcón et al., 1999).
The LTS was initiated in 1957 by Falkner. In the
LTS, twins were tested every 3 months in the first year
of life. Testing continued at 6-months intervals during
second and third year of life, and annually through age
9, with follow-up visits at age 15 and adulthood. In
1983, the sample of the LTS consisted of 494 pairs of
twins active in the longitudinal study, ranging in age
from 3 months to 15 years. Recruitment has been an
ongoing process, with 25–35 pairs added each year
since 1963. However, like in every longitudinal design,
the study suffers from dropouts over the years.
Sophistication in developmental behavior genet-
ics involves the formulation of models that attempt to
describe the etiology of genetic and environmental in-
fluences on variation in cognitive development. Phillips
Fig. 1. Simplex model examining developmental pattern of genetic, shared environmental and unique environmental influences on IQ.
G
(t)
4
are the innovation parameters.
G
(t) are the transmission parameters. U are time-specific influences of unique environment.
and Fulker (1989) developed a model, based on a quasi-
simplex model presented earlier by Eaves et al. (1986),
in which it was possible to distinguish between the
three possible longitudinal mechanisms (time-specific,
common factor, simplex). This model was applied to a
large data set combined from several major projects
(Cardon et al., 1992). The CAP data at ages 1, 2, 3, 4,
and 7 were combined with data from twins at ages 1,
2, and 3 from the MacArthur Longitudinal Twin Study
(MLTS) (Plomin et al., 1990) and the Twin Infant Pro-
ject (TIP) (DiLalla et al., 1990; Benson, et al., 1993).
The best model for genetic influences on IQ was a sim-
plex model, with time-specific innovations included.
That is, genetic variation initially shown at age 1 is ex-
pressed through at least age 7 with new genetic varia-
tion, independent of the initial genetic influence, at ages
2, 3, and 7, but not at age 4. For shared environment,
the best-fitting model showed only a single common
factor influence on IQ, with equal factor loadings at
each age. This longitudinal outcome suggests that
shared environmental effects contribute to continuity
only. In complete contrast is the picture that emerged
for unique environment. For unique environment, the
influences were specific to each time-point, which
implied that change in cognition is, at least partly, ac-
counted for by these influences.
In a subsequent publication involving CAP, TIP,
and MLTS subjects, including subjects from the CAP
sample at age 9, very similar results were found (Fulker
et al., 1993a). This time, a Cholesky decomposition
was used. A common genetic factor present at year 1
continued to account for observed variance in IQ, but
with diminishing impact with increasing age. Evidence
for genetic change at two important developmental
transitions was found. The first was transition from in-
fancy to early childhood (ages 2 and 3). The second
was the transition from early to middle childhood (age
7). Fulker et al. (1993a) speculated that the new ge-
netic influence at age 7 might be in response to the
‘novel environmental challenge’ of schooling. No new
genetic effect was apparent at age 9. Further, there was
one continuous source of shared environmental influ-
ence across all ages. Application of the quasi-simplex
model to the same data yielded identical results (Cherny
and Cardon, 1994). Finally, the only longitudinal
model-fitting results based on the LTS data showed
that a simplex model gave a better fit compared to a
common factor model for genetic effects from ages 1
to 9 years (Humpreys and Davey, 1988).
To summarize, the general picture that emerges
from these studies with young children is that genetic
Genetic and Environmental Influences on the Development of Intelligence 239
effects account both for stability and change in cogni-
tive performance. This is implied by the simplex-like
structure with time-specific innovation effects. Shared
environmental effects appear to account for stability in
intellectual performance, indicated by the single com-
mon factor structure, without age-specific effects. Con-
sistent across studies, the unique environment is best
modeled as exerting time-specific influences only. This
structure implies that the unique environment is im-
portant in explaining variance in cognitive performance
at each age, but not in explaining stability of cognitive
performance across ages.
In the present longitudinal study, structural mod-
eling techniques were used to examine the influences
of genetic and environmental factors on development of
full-scale IQ (FSIQ), using data of 209 Dutch twin pairs
tested at 5, 7, 10, and 12 years of age. In addition to
estimating the importance of heritability and environ-
mental influences, the focus was on the developmental
pattern of cognition. A genetic simplex model and a
common factor model were used to study continuity and
changes of genetic and environmental influences over
time. Based on previous results of longitudinal studies
on the development of cognitive functioning, a simplex
structure for genetic influences was expected. Further,
it was assumed that shared environmental factors show
continuing effects over the years and unique environ-
mental influences are age specific only.
METHODS
Participants
This study is part of an ongoing, longitudinal study
of the development of intelligence and problem behav-
ior. The study started in 1992 with recruitment of 209
twin pairs from the Netherlands Twin Register (NTR;
Boomsma, Orlebeke, & Van Baal, 1992). The initial
sample of 209 twin pairs was selected on the basis of
age and zygosity of the twins, and their city of resi-
dence. Mean age at the first measurement occasion was
5.3 years (80% ranging from 5 years and 1 month to
5 years and 6 months). At the second measurement oc-
casion, mean age was 6.8 years (80% ranging from
6 years and 6 months to 7 years and 1 month). Mean
age at the third measurement occasion was 10 years
(80% ranging from 9 years and 11 months to 10 years
and 1 month). Mean age at the fourth measurement oc-
casion was 12 years and several days (80% ranging from
11 years and 11 months to 12 years and 1 month). Zy-
gosity of the same-sex twins was established by either
blood group polymorphisms (137 pairs) or DNA analy-
ses (24 pairs) and, in a few pairs, by physical resem-
blance assessed by the test administrator (9 pairs). There
were 47 monozygotic female (MZF), 37 dizygotic fe-
male (DZF), 42 monozygotic male (MZM), 44 dizygotic
male (DZM), and 39 dizygotic pairs of opposite sex
(DOS). The intelligence test was administered to all
209 twin pairs at age 5. At the second measurement oc-
casion (age 7), 192 pairs of the original sample provided
complete data on all subtests. The number of partici-
pating twin pairs increased to 197 when the children
were tested around their 10th birthday. At the fourth
measurement occasion (age 12), 192 twin pairs partic-
ipated. A small group of four families refused consis-
tently to participate after the first measurement occa-
sion. Five families dropped out at ages 10 and 12. The
remaining nonparticipants refused participation at one
measurement occasion. At ages 5 and 12, one incom-
plete twin pair can be found in the data because of dif-
ficulties during testing (age 5) and refusal to participate
(age 12). Because of serious loss of hearing, one twin
pair was assigned missing value at all four ages for
FSIQ. This left a sample of 176 twin pairs with com-
plete data at all four ages. No significant difference in
initial FSIQ (at age 5) has been found for twins who
dropped out on one or more of the following occasions
(F
3, 415
2.25, P .082). Details on the demographic
characteristics of the sample and information on parental
occupation can be found in Rietveld et al. (2000).
Procedure and Intelligence Tests
At ages 5 and 7 years, the twins participated in a
study on the development of cognitive abilities and
brain activity (Boomsma and Van Baal, 1998). At both
measurement occasions, the twin and their family vis-
ited the laboratory at the university. While one of the
twins participated in the electrophysiological experi-
ment, the co-twin participated in an intelligence test.
At ages 10 and 12 years, a different procedure was fol-
lowed. The twins and their parents could choose
whether they preferred to come to the university or
whether they preferred to be visited at home to partic-
ipate in the intelligence test. The majority of the fam-
ilies (around 70% at both ages) preferred testing at
home. No significant difference in FSIQ was observed
between children tested at home or at the university.
The intelligence test was assessed by an experienced
test administrator. At ages 5, 7, and 10, the test took
approximately 1 hour to complete, and at age 12 the
test took 1 and
1
2
hours to complete. All children
received a present afterwards.
240 Bartels, Rietveld, Van Baal, and Boomsma
At age 5, 7, and 10, the children were tested with
the Revised Amsterdamse Kinder Intelligentie Test
(RAKIT) (Bleichrodt et al., 1984). Six subtests, with
age-appropriate items, were employed to assess cogni-
tive functioning. Raw subtest total scores are corrected
for age and transformed into standardized scores with a
mean of 15 and a standard deviation of 5. The total IQ
score is based on the combination of these transformed
subtests with a mean of 100 and standard deviation of
15. The standardization is based on a population sample
of Dutch 6- to 11-year-old children. No difference is
made for boys and girls. For further details on this well-
known Dutch intelligence test see Rietveld et al. (2000).
At age 12, the twins conducted the complete version of
the WISC-R, Dutch version (Van Haasen et al., 1986).
The WISC-R consists of 12 subtests, 6 mainly verbal
and 6 mainly nonverbal. The subtest scores are stan-
dardized, based on results of same-aged children in the
Netherlands. No differences are made for boys and girls.
Addition of the twelve standardized subtest scores results
in FSIQ. The concurrent validity of the RAKIT and the
WISC-R is .86 (Pijl et al., 1984).
Statistical Analyses
Descriptive statistics for FSIQ were calculated
using SPSS/windows 10. Twin correlations with their
95% confidence intervals at each age have been calcu-
lated. These correlations are informative on the im-
portance of genes and environment in explaining ob-
served variance at each age. To assess stability of
intelligence, phenotypic cross correlations over time
were calculated. MZ and DZ cross correlations over
time have been calculated to get a first impression of
the genetic and environmental contributions to the
covariance over time.
Genetic Modeling
Univariate model fitting procedures were used to
estimate genetic and environmental influences at each
age separately and to investigate the presence of sex-
differences and influences of sex-specific genes in these
data. Genetic model fitting of twin data allows for sep-
aration of the observed phenotypic variance into its ge-
netic and environmental components. Additive genetic
variance (A), is the variance that results from the ad-
ditive effects of alleles at each contributing genetic
locus. Shared environmental variance (C) is the vari-
ance that results from environmental events common
to both members of a twin pair. Unique environmental
variance (E) is the variance that results from environ-
mental effects that are not shared by members of a twin
pair. Estimates of the unique environmental effects also
include measurement error. To account for this source
of variance, E is always specified in the model.
The different degree of genetic relatedness between
monozygotic (MZ) and dizygotic (DZ) twin pairs was
used to estimate the contribution of these factors to the
phenotypic variation in cognitive abilities. Similarities
for MZ twins are assumed to be due to additive genetic
influences plus environmental influences that are shared
by both members of a twin pair. Experiences that make
MZ twins different from one another are unique envi-
ronmental influences. Because DZ twins share 50% of
their genetic material on average, like other siblings,
genetic factors contribute only half to their resemblance.
As for MZ twins the shared environment contributes
fully. Model fitting to twin data is based on the com-
parison of the variance-covariance matrices in MZ and
DZ twins. Exploiting the known difference in genetic
contribution to intrapair resemblance of MZ and DZ
twin pairs and the influences of additive genetic, shared
environmental and unique environmental factors are es-
timated using the computer program Mx.
Differences between boys and girls can occur in
two ways. First, a difference in the magnitude of addi-
tive genetic, shared environmental and unique envi-
ronmental influences can exist, represented in a distinct
pattern of twin correlations for boys and girls. Second,
heterogeneity, an expression of different genes in boys
and girls, can occur. This heterogeneity would be rep-
resented by a lower twin correlation in dizygotic twins
of opposite sex in comparison to dizygotic same sex
twins. Differences in magnitude of additive genetic,
shared environmental and unique environmental influ-
ences is tested by the change in fit after constraining
the parameter estimates equal for boys and girls. Test-
ing for heterogeneity is accomplished by testing the ge-
netic correlations between two members of a dizygotic
twin of opposite sex. Normally the genetic correlation
of DZ twins is fixed at .5. Heterogeneity would result
in a genetic correlation of less than .5.
Multivariate genetic model fitting techniques were
used to obtain insight in the developmental pattern of
cognitive functioning and to obtain estimates of the ge-
netic and environmental influences on cognitive devel-
opment. Parameters were estimated by maximum likeli-
hood, using the computer program Mx (Neale et al.,
1999). Rather than decomposing the variance of a mea-
surement into genetic and environmental sources of vari-
ance, multivariate genetic analysis decomposes the vari-
ance of each measurement occasion and the covariance
between the measurement occasions into genetic and en-
vironmental sources. The total variances and covariances
Genetic and Environmental Influences on the Development of Intelligence 241
were decomposed into additive genetic (A), shared en-
vironmental (C), and unique environmental (E) parts.
First, to get an initial insight in the variance and covari-
ance structure a Cholesky decomposition model was ap-
plied to the data. Next, to investigate the stability and
change in FSIQ a genetic simplex model was applied to
the data. For each source of variance (A, C, and E) a sim-
plex structure was specified. A simplex model is a first-
order autoregressive process. In the simplex model, co-
variances among the four ages of measurement are
specified by genetic and environmental factors specific
to each age and by ‘carry-over effects’ or transmission
of these factors to subsequent ages. The model specifies
the variance unique to each measurement occasion by an
innovation term that comes into play at each time point.
The variance is a product of the age-specific effects and
age-to-age transmission effect (see appendix 1 and
Fig. 1). Finally, it was investigated whether a common
factor, possibly with age-dependent factor loadings and
age-specific influences, could replace the simplex struc-
ture for genetic and shared environmental influences.
To make optimal use of all available data, analy-
ses were performed on the raw data. Submodels were
compared by hierarchic
2
tests. The
2
statistic is com-
puted by subtracting 2LL for the full model from that
for a reduced model (
2
2 (LL
1
LL
0
)). A good
model is indicated by a low nonsignificant
2
test sta-
tistic (P .05). Apart from the
2
test statistic,
Akaike’s Information Criterion (AIC
2
2 de-
grees of freedom) was computed. The lower the AIC,
the better is the fit of the model to the observed data.
Reductions of the model were based on the ex-
pectations raised by previous studies. In detail, a sim-
plex structure for genetic influences, a common factor
for shared environmental influences and time-specific
structure for unique environmental influences is ex-
pected. Estimates of genetic, shared environmental, and
unique environmental influences on the age-specific
variance and between age covariance of general cog-
nitive abilities are reported based on the Cholesky de-
composition model, the full simplex model, and the best
fitting reduced model.
RESULTS
Descriptive statistics for FSIQ at 5, 7, 10, and
12 years of age showed that the variables were approx-
imately normal distributed (Table I). Table II shows the
twin correlations for the five zygosity groups calculated
separately for each age. MZ correlations are higher than
DZ correlations, suggesting genetic influences at each
age. The low DOS correlation at age 12 suggests sex
differences and univariate model fitting procedures were
used to explore this possibility. Estimates for genetic
and shared environmental influences based on the
univariate model-fitting procedure are presented in
Table III. These results are consistent with previous
results (Boomsma and Van Baal, 1998; Bouchard and
McGue, 1981) showing increase of genetic influences
and diminishing effects of shared environment over the
years. Shared environmental influences are insignificant
at ages 10 and 12. Univariate model fitting showed no
presence of sex differences at the four ages separately
and no presence of sex-specific genes at age 12.
To get a first impression of the developmental pat-
tern of cognitive abilities, phenotypic cross correlations
over time were calculated (Table IV). All correlations
are rather large, which indicates a strong degree of sta-
bility of intellectual performance. This structure may best
be described by a common factor mechanism. Cross cor-
relations over time for monozygotic (MZ) and dizygotic
(DZ) twins were calculated separately to explore the ge-
netic and environmental influences on the observed sta-
bility. As can be seen in Table IV, the MZ cross corre-
lations over time (above the diagonal) are higher than the
DZ cross correlations over time (below the diagonal),
suggesting that stability in intelligence over time is
242 Bartels, Rietveld, Van Baal, and Boomsma
mainly due to genetic factors. Further, when the corre-
lations of the adjoining age-intervals are compared (ages
5–7; ages 7–10; ages 10–12), the increased difference
between MZ and DZ correlations suggests an increase in
the genetic contribution to stability with increasing age.
Analyses were continued with the application of
the different models to the longitudinal data. Model-
fitting procedures yielded the results presented in
Table V. The genetic simplex model without restrictions
(model 2) was taken as a reference for evaluating
changes in
2
and associated degrees of freedom of more
parsimonious models. First, reduction of the model was
based on the expectation of age-specific unique envi-
ronmental factors only (model 3). No significant change
in
2
arose. Second, model reduction was based on the
expectation of a common factor for shared environ-
mental influences (model 4). Because the order of model
reduction may influence the fit of the model, a model
with a common factor for genetic influences and a sim-
plex structure for shared environmental influences was
fitted to the data as well (model 5). No clear distinction
could be made between models 4 and 5, both being more
parsimonious than model 3 but not significantly differ-
ent. A model with a common factor for both genetic and
shared environmental influences, allowing for time-spe-
Table I. Descriptive Statistics for Full-Scale IQ at Different Ages
Skewness Kurtosis
N
a
Mean age Min Max Mean Std S.E. S.E.
FSIQ5 415 5.3 64 142 102.75 13.18 .059 .120 .209 .239
FSIQ7 382 6.8 62 145 102.90 14.67 .127 .125 .023 .249
FSIQ10 392 10.0 63 145 106.96 15.54 .066 .123 .166 .246
FSIQ12 381 12.0 61 138 100.03 13.18 .039 .125 .177 .249
a
Number of children in the study.
Table II. Twin Correlations with 95% Confidence Intervals
Age MZF
a
DZF MZM DZM DOS
5 .78 (.64–.87) .73 (.53–.85) .77 (.62–.87) .53 (.29–.72) .64 (.41–.79)
46
b
37 42 43 39
7 .77 (.61–.87) .50 (.20–.70) .56 (.29–.74) .41 (.13–.63) .56 (.30–.74)
41 34 37 41 38
10 .87 (.78–.92) .45 (.16–.67) .73 (.54–.85) .53 (.28–.72) .50 (.21–.70)
43 37 38 41 37
12 .86 (.76–.92) .67 (.46–.82) .84 (.71–.91) .57 (.32–.75) .35 (.03–.60)
43 37 36 39 35
a
MZF monozygotic female; DZF dizygotic female; MZM monozygotic male; DZM dizyotic males;
DOS dizygotic opposite sex.
b
Number of complete twin pairs.
cific influences as well, did not gave a significant worse
fit (model 6). Further, it was tested whether dropping
the age-specific influences, either genetic or shared en-
vironmental, altered the
2
significantly (model 7 and
model 8). Based on the difference in
2
and the lower
AIC, model 7 was preferred above model 8. The genetic
(co)variance is modeled as a common factor without
specifics, whereas the shared environmental (co)vari-
ance is modeled as a common factor with specifics.
These results suggest that stability in cognitive perfor-
mance is mainly due to genetic factors. Finally, a model
with a common factor, without time-specific influences
for both genetic and shared environmental influences
showed a significant increase in
2
(model 9).
Genetic and Environmental Influences on the Development of Intelligence 243
Estimates of the path coefficients for the best fit-
ting model (model 7) are presented in Fig. 2. The per-
centage of age-specific variance explained by genetic,
shared environmental and unique environmental fac-
tors based on the Cholesky decomposition (model 1),
the full simplex model (model 2), and the best fitting
model (model 7) are presented in Table VI. Table VII
contains the percentages of between-age covariances
explained by genetic, shared environmental, and unique
environmental factors based on the Cholesky decom-
position, the full simplex model, and the best fitting
model. Indicated by the observed MZ and DZ cross cor-
relations, genes become more important in explaining
stability in cognitive performance with increasing age.
Table III. Univariate Model-Fitting Results for the Four Ages
Model 2LL df ⌬␹
2
df p A C E
5 ACE SD
a
3178.20 408
ACE 3180.40 411 2.20 3 .53 .26 (.03–.52)
b
.50 (.26–.68) .24 (.18–.33)
AE 3193.71 412 13.31 1 .00
CE 3185.25 412 4.85 1 .03
7 ACE SD
a
3051.33 375
ACE 3054.37 378 3.04 3 .39 .39 (.07–.72) .30 (.00–.55) .31 (.23–.44)
AE 3058.12 379 3.75 1 .05 .70 (.60–.78) .30 (.22–.40)
CE 3059.83 379 5.46 1 .02
10 ACE SD
a
3135.48 385
ACE 3140.87 388 5.39 3 .15 .54 (.28–.83) .25 (.00–.48) .21 (.15–.29)
AE 3143.62 389 2.75 1 .10 .80 (.72–.85) .20 (.15–.28)
CE 3156.99 389 16.12 1 .00
12 ACE rg
free
c
SD
a
2903.40 373
ACE SD
a
2903.71 374 .31 1 .58
ACE 2908.72 377 5.01 3 .17 .64 (.40–.88) .21 (.00–.43) .15 (.11–22)
AE 2910.81 378 2.09 1 .15 .85 (.79–89) .15 (.11–.21)
CE 2936.28 378 27.56 1 .00
a
Model with sex differences for parameter estimates.
b
95% confidence intervals.
c
Model with sex-specific genes.
Table IV. Phenotypic Cross Correlations for FSIQ, Calculated for the Complete Dataset and
MZ (Above Diagonal) and DZ (Below Diagonal) Cross Correlations over Time for FSIQ
Total sample 5 7 10 12
5 1.00 .65 (.59–.70) .65 (.59–.70) .64 (.57–.69)
7 1.00 .72 (.67–.77) .69 (.63–.74)
10 1.00 .78 (.74–.82)
12 1.00
DZ/MZ 5b 7b 10b 12b
5a
.66 (.54–.75) .67 (.56–.76) .68 (.57–.77)
7a .42 (.30–.54) .71 (.60–.79) .68 (.57–.77)
10a .42 (.30–.54) .39 (.26–.52) .79 (.70–.85)
12a .42 (.29–.54) .42 (.29–.54) .45 (.32–.57)
As opposed to this outcome, the shared environment
accounts for a decreasing portion of the covariance be-
tween age intervals, whereas the unique environment
explains, in general, around one quarter of the total
variance at each age (Table VI). The unique environ-
ment contributes only minimally to the observed co-
variance between ages (Table VII). It should be noted
that the genetic, shared environmental and unique en-
vironmental variance components estimated from fit-
ting the multivariate models to these data are somewhat
different from what univariate analyses at each age sep-
arately might yield. These differences arise because the
multivariate models take into account the cross-sibling
cross-time covariance structure, which can affect the
within-time parameter estimates. In addition, multi-
variate model-fitting increases the power to detect
shared environmental influences as a source of famil-
ial aggregation.
244 Bartels, Rietveld, Van Baal, and Boomsma
Table V. Model Fitting Results for FSIQ
Model 2LL df
2
df Compared to model P AIC
1. 11527.470 1520
A: Cholesky
C: Cholesky
E: Cholesky
2. 11510.105 1508
A: Full simplex structure
C: Full simplex structure
E: Full simplex structure with
time-specific factors
a
3. 11517.278 1512 5.681 4 2 .22 2.319
A: Full simplex structure
C: Full simplex structure
E: Time-specific factors only
b
4. 11513.722 1511 3.617 3 2 .31 2.383
A: Full simplex structure
C: Common factor specifics
E: Time-specific factors only
b
5. 11517.303 1511 7.198 3 2 .07 1.198
A: Common factor specifics
C: Full simplex structure
E: Time-specific factors only
b
6. 11513.743 1510 3.638 2 2 .16 .362
A: Common factor specifics
C: Common factor specifics
E: Time-specific factors only
b
7. 11513.743 1514 4 6 1.00 8.000
A: Common factor
C: Common factor specifics
E: Time-specific factors only
b
8. 11521.677 1514 7.934 4 6 .09 .066
A: Common factor specifics
C: Common factor
E: Time-specific factors only
b
9. 11545.797 1518 32.05 8 6 .00 16.050
A: Common factor
C: Common factor
E: Time-specific factors only
b
a
These time-specific factors are equal at all ages.
b
These time-specific factors are estimated separately at every age.
Genetic and Environmental Influences on the Development of Intelligence 245
Fig. 2. Model 7; common factor for A, common factor with time-specific influences for C and time-specific influences for E.
Table VI. Percentage of Variance Explained by A, C, and E, Based on a Cholesky Decomposition, a Simplex Model,
and the Best Fitting Model, with 95% Confidence Intervals
Model 1 Variance A Cholesky C Cholesky E Cholesky
5 .30 (.15–.54) .46 (.24–.62) .24 (.18–30)
7 .42 (.21–.67) .28 (.05–.49) .30 (.22–.39)
10 .61 (.36–.82) .19 (.00–.43) .20 (.15–.27)
12 .62 (.39–.86) .23 (.00–.45) .15 (.11–.21)
Model 2 Variance A simplex C simplex E simplex
5 .38 (.22–.60) .39 (.18–.56) .23 (.18–.30)
7 .38 (.20–.63) .32 (.09–.50) .30 (.23–.39)
10 .72 (.33–.83) .08 (.00–.46) .20 (.15–.27)
12 .62 (.37–.85) .23 (.01–.46) .15 (.11–.21)
Model 7 Variance A common factor C total Common Specific E time specific
5 .26 (.14–.46) .51 (.31–.65) .47 .04 .23 (.18–.29)
7 .47 (.29–.62) .26 (.11–.45) .17 .09 .27 (.21–.35)
10 .69 (.49–.82) .12 (.00–.33) .10 .02 .19 (.14–.25)
12 .64 (.45–.77) .21 (.09–.40) .11 .10 .15 (.11–.20)
DISCUSSION
The influences of genes and environment on cog-
nitive development and on its developmental structure
were studied in a longitudinal sample of Dutch twins
at 5, 7, 10, and 12 years of age. It can be concluded
that the development of general cognitive abilities is a
continuous process. Continuity is represented by a com-
mon factor, with age-specific factor loadings, for both
genetic and shared environmental influences. Change
in development, represented by age-specific factors, are
presented in the shared environmental structure and, as
expected, in the unique environmental structure. Fur-
ther, decomposition of the between-age covariances in
additive genetic, shared environmental, and unique en-
vironmental influences showed that the continuity in cog-
nitive abilities is mainly due to additive genetic factors.
In this study, increasing additive genetic influences
and decreasing influences of shared environmental
factors are found in both age-specific variances and
between-age covariances. The increase of genetic in-
fluences on cognitive functioning throughout develop-
ment is already well established in U.S. samples and
is now also found in a sample of Dutch twins. In the
common factor pattern for genetic influences (model 7;
see Fig. 2), increasing influences of heritability are rep-
resented by increasing factor loadings from age 5 to 10.
Further, in the common factor pattern for shared envi-
ronmental influences, decreasing influences are repre-
sented by decreasing factor loadings from age 5 to 10
and decreasing age-specific influences from age 5 to 10.
The developmental pattern for genetic influences
found in this study is partly different from previous, com-
246 Bartels, Rietveld, Van Baal, and Boomsma
parable studies like the combined study of CAP, MLTS,
and TIP (Cardon et al., 1992; Fulker, et al., 1993). Re-
sults provided by these studies show a simplex pattern
for genetic influences with genetic innovation at 2, 3,
and 7 years of age, with the suggestion that genetic in-
novation at age 7 may be due to “the novel environ-
mental challenge of schooling”. In our study, no indi-
cation for genetic innovation is obtained. Comparison of
the different longitudinal studies is limited due to dis-
tinct ages of testing. In our study, no information is avail-
able for cognitive development prior to age 5 and the
results, mainly presented by the CAP studies, provide
no information on the development of general cognitive
ability between ages 9 and 16. Another difficulty in lon-
gitudinal studies in general and in comparing different
longitudinal studies on cognitive development in partic-
ular is the measurement of cognitive performance. There
are no cognitive assessments that are common to all ages,
so different age-appropriate instruments must be used.
One of the difficulties with this is that no distinction can
be made between true changes in development and
changes related to different measurement instruments.
A major advantage of our longitudinal study is that the
same intelligence test (RAKIT), with age-specific items,
is used at the first three measurement occasions. Further,
Table VII. Percentage of Covariance Explained by A, C, and E, Based on a Cholesky, Model,
a Simplex Model, and a Restircited Model, with their 95% Confidence Intervals
Model 1 Covariance A Cholesky C Cholesky E Cholesky
5–7 .55 (.32–.87) .40 (.10–.63) 0.5 (.00–.12)
5–10 .66 (.40–.99) .34 (.01–.58) .00 (.00–.07)
5–12 .66 (.42–.98) .34 (.02–.58) .00 (.00–.04)
7–10 .71 (.43–.93) .21 (.00–.48) .08 (.02–.15)
7–12 .73 (.46–.98) .24 (.01–.50) .03 (.00–.10)
10–12 .78 (.51–.99) .20 (.00–.45) .02 (.00–.08)
Model 2 Covariance A simplex C simplex E simplex
5–7 .59 (.33–.95) .38 (.04–.63) .03 (.00–.10)
5–10 .82 (.55–.99) .17 (.00–.45) .01 (.00–.04)
5–12 .76 (.48–.99) .24 (.01–.52) .00 (.00–.02)
7–10 .73 (.46–.97) .20 (.00–.46) .07 (.01–.15)
7–12 .69 (.42–.98) .29 (.01–.56) .02 (.00–.07)
10–12 .84 (.55–.99) .13 (.00–.42) .03 (.00–.08)
Model 7 Covariance A common factor C common factor E time specific
a
5–7 .56 (.35–.83) .44 (.17–65)
5–10 .66 (.42–.95) .34 (.05–.58)
5–12 .64 (.41–.92) .36 (.08–.59)
7–10 .81 (.55–.98) .19 (.02–.45)
7–12 .80 (.54–.98) .20 (02–.46)
10–12 .86 (.63–.99) .14 (.01–.37)
a
E is represented in time-specific influences only.
the intelligence test used at the fourth measurement
occasion (WISC-R) shows a high concurrent validity
with the RAKIT (.86 for FSIQ) (Pijl et al., 1984).
More striking is the finding for shared environ-
mental influences. Previous studies suggested a com-
mon factor for shared environmental influences. Our
study indicates that, besides a continuing influence of
shared environmental factors, age-specific influences
are present. These age-specific effects were significant,
but the proportion of variance explained is much
smaller compared to the proportion explained by the
shared environmental factor common to all ages. This
common factor could be accounted for by SES and
parental education, because these environmental as-
pects are not sensitive to large changes over a time span