You are on page 1of 11

ARTICLE IN PRESS

INTELL-00428; No of Pages 11

Available online at www.sciencedirect.com

Intelligence xx (2007) xxx – xxx


Facon, B. (2008). How does the strength of the relationships between
cognitive abilities evolve over the life-span for low-IQ vs high-IQ
adults ? Intelligence, 36, 339-349.

How does the strength of the relationships between cognitive abilities


evolve over the life span for low-IQ vs high-IQ adults? ☆
Bruno Facon ⁎
Université Charles De Gaulle-Lille III, Unité de Recherche sur l'Evolution des Comportements et des Apprentissages (EA 1059), France

Received 8 March 2007; received in revised form 17 November 2007; accepted 17 November 2007

Abstract

The present study was designed to examine how the correlations between cognitive abilities evolve during adulthood. Data from
1104 participants on the French version of the Wechsler Adult Intelligence Scale-Third Edition were analyzed. The entire sample
was divided into four age groups (16–24 years; 25–44 years; 45–69 years and 70–89 years), which were themselves split into two
IQ-levels using the mean standard score on Vocabulary and Block Design. For every age group, the mean correlation between
subtest scores of low-IQ participants was higher than that of high-IQ participants. There was also no interaction between age and
IQ for the strength of subtest relationships. Indeed, the effect sizes of correlation differences between low- and high-IQ participants
appeared to be relatively constant across age. A general developmental schema of how the strength of correlations between
cognitive abilities of low- and high-IQ individuals evolves during the entire life span is sketched in the discussion.
© 2007 Elsevier Inc. All rights reserved.

Keywords: Intelligence; Adulthood; Diminishing returns, Wechsler Adult Intelligence Scale; Spearman law; Differentiation; Cognitive abilities

The law of diminishing returns (Spearman, 1927), Jensen, 2003; Legree, Pifer, & Grafton, 1996; Lynn,
that is, the inverse relationship between IQ and the 1990; te Nijenhuis & Hartmann, 2006; Reynolds &
strength of correlations of cognitive abilities, has been Keith, 2007). The bulk of this research was conducted
the subject of numerous studies in recent decades (e.g., with the implicit idea that this inverse relationship does
Abad, Colom, Juan-Espinosa, & García, 2003; Carl- not develop with age. Stated otherwise, individuals with
stedt, 2001; Deary et al., 1996; Detterman, 1993; low-IQ would have less differentiated cognitive profiles
Detterman & Daniel, 1989; Evans, 1999; Facon, 2002, than individuals with high-IQ from the outset of their
2003a, 2003b; Hartmann & Reuter, 2006; Hunt, 1997; development. However, recent investigations have
challenged this “fixist” view of the law of diminishing

Author note: I am grateful to the Editions du Centre de returns. For example, in the Facon (2004) study, the
Psychologie Appliquée for providing the data analyzed in this mean correlations between cognitive abilities of chil-
research. I also thank John M. Belmont for his helpful comments on dren of low and high-IQ aged from 6 to 9 years were
the manuscript. nearly equal, suggesting that the inverse relationship
⁎ Université Charles De Gaulle-Lille III, 35 rue Sainte Barbe, B.P.
70460, 59208 Tourcoing CEDEX, France.
does not appear during childhood, and Arden and
E-mail addresses: bruno.facon@univ-lille3.fr, Plomin (2007) have now reported similar results for 7–
bruno.facon@laposte.net (B. Facon). 10 year old children. Moreover, in a study of children
0160-2896/$ - see front matter © 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.intell.2007.11.004

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
ARTICLE IN PRESS
2 B. Facon / Intelligence xx (2007) xxx–xxx

and young adolescents of high and low intellectual Indeed, one of the most prime lessons of developmental
functioning ages 7–9, 10–12 and 13–15 years, Facon psychology in general and of the above-mentioned
(2006) showed that median correlations between examples in particular, is that psychological phenomena
intellectual scores did not change with age for high-IQ have a genesis. The law of diminishing returns is
participants (.13, .15 and .14 for the three age groups), probably not an exception to this rule, and thus it is
whereas an increase was observed for low-IQ partici- important to study when the IQ-related process of
pants (.15, .16 and .25, respectively). Thus, it seems that differentiation appears and how the correlations evolve
only at the end of childhood (13–15 years) does the in low- and high-IQ groups over the course of
phenomenon described by Spearman begin to appear. development.
The idea of bringing a developmental perspective to Second, a developmental approach to studying the
the study of the law of diminishing returns is justified in law of diminishing returns could help to resolve some
at least two respects. First, there is a great deal of inconsistencies in the literature. Indeed, the inverse
evidence, in psychology, of developmental variations in relationship between IQ and the strength of correlations
the relationships between cognitive, demographic and among cognitive abilities is not always confirmed (e.g.,
background variables. For example, the correlation Fogarty & Stankov, 1995; Hartmann & Teasdale, 2004;
between parents' education or socioeconomic status Hartmann & Reuter, 2006; Nesselroade & Thompson,
and intelligence test scores of their offspring is negative 1995). These negative results have been attributed to
or close to zero during early childhood but increases to several explanatory variables such as the type of
.40–.60 in the following years (e. g., Bayley & Jones, measure and the cutoff point used to constitute IQ
1937; Bayley, 1954, 1970; Honzik, 1957, 1963, 1967). groups, the content, number, and complexity of ability
Also, the IQ variance explained by genotype increases tests, and the location of groups along the IQ continuum
with age whereas the effect of shared environmental (Abad et al., 2003 [study 1]; Deary et al., 1996; Fogarty
influences tends to diminish between childhood and & Stankov, 1995; Hartmann & Teasdale, 2004, 2005;
adolescence (Plomin, 2003; Plomin, DeFries, Nesselroade & Thompson, 1995; Pagliari, 1998). But it
McClearn, & McGuffin, 2001; Plomin & Petrill, 1997; is also possible, beyond these variables, that the age of
Plomin & Spinath, 2004). In another area of research, an participants moderates the effect of IQ on the degree of
age-related stabilization of composite intelligence test differentiation of abilities and thus makes the phenom-
scores during childhood was early demonstrated. enon more difficult to discern when, as is frequently the
Indeed, age to age correlations are very low before case, several age levels are present but not controlled
18 months, increase rapidly up to 5 or 6 years and then for. From this standpoint, it seems clearly necessary to
become both high and stable even when the psycho- study the law of diminishing returns by taking age into
metric examinations are conducted several years apart account. In some respects, this approach combines the
(Bayley, 1949; Bayley & Schaefer, 1964; Lewis & ability-related differentiation hypothesis with another
McGurk, 1972; McCall, 1976; McCall, Hogarty, & well-known hypothesis, that of a change in the structure
Hurlburt, 1972). In the same way, although this point is of cognitive abilities with age. This latter was probably
still much debated (Anstey, Hofer, & Luszcz, 2003; originated in 1919 with Burt (see Burt, 1954) and, under
Facon, 2007; Juan-Espinosa, García, Colom, & Abad, the influence of Garrett (1946), gave rise to numerous
2000; Juan-Espinosa et al., 2002; Zelinski & Lewis, works in which, surprisingly, age and ability level were
2003), some results indicate the existence of an age- never actually studied simultaneously. Nonetheless, at
related differentiation–dedifferentiation of cognitive least two studies had already been done on the
abilities over the life span (Li et al., 2004; Hertzog & differentiation of cognitive abilities as a function of
Bleckley, 2001; Tideman & Gustafsson, 2004). Finally, both age and IQ. Lienert and Faber (1963), for example,
the relationship between intelligence test scores and simultaneously tested the IQ-related differentiation
learning or job performance often changes markedly hypothesis (called, after Wewetzer, the divergence
with practice (Ackerman, 1987, 1988, 2005; Ackerman hypothesis) and the age-differentiation hypothesis.
& Cianciolo, 2000, 2002; Fleishman, 1972; Fleishman However, the age differences between their groups
& Mumford, 1989; Fleishman & Quaintance, 1984; were, on the average, only 3 years (8–10 years vs 11–
Hulin, Henry, & Noon, 1990; Kennedy, Dunlap, Turn- 13 years). Deary et al. (1996) have also crossed these
age, & Wilkes, 1993). In view of these few examples of two variables in a methodologically sound study
macro or micro developmental trends, one may ask why conducted with a sample of impressive size, but again
the relationship between IQ and correlations between with a too small age range (14–17 years). In addition,
cognitive abilities might not also evolve with age. beyond this age span problem, neither of these two

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
ARTICLE IN PRESS
B. Facon / Intelligence xx (2007) xxx–xxx 3

studies was conducted to explore the possibility of an Facon's (2006) study. Note also that Vocabulary and Block
Age × IQ interaction on the strength of relationships Design have the highest correlation with verbal IQ and
between cognitive abilities. In fact, both viewed the two performance IQ, respectively. The low-IQ groups comprised
differentiation processes as essentially independent. all participants whose mean composite score on these two
subtests was 10 or less; the high-IQ groups comprised all the
The data from Facon (2006) indicate an age-related
others. Eight groups were thus constituted: Two IQ groups at
effect of IQ on the differentiation of cognitive abilities each of four age ranges. The two subtests employed to perform
during childhood, a result which, of course, must be this IQ selection were not included in the subsequent statistical
corroborated and explained. In addition, the analysis of analyses.
this Age × IQ interaction might also be pursued beyond
adolescence to include the whole life span in order to give 1.2. Statistical analyses
a fuller picture of the evolution of the magnitude of
correlations between cognitive abilities as a function of
The probability of detecting statistically significant main or
IQ-level. This approach was taken in the present study
interaction effects depends on several parameters including the
using data from a large population-based sample of adults reliability of measures, the level of type I error (α), the sample
of low and high-IQ aged from 16 to 89 years divided into size, the variance of the independent variable(s) or of product
four age groups. The goal was to examine whether the terms, and the size of the hypothesized effect (Cohen, 1988;
difference of correlations between low- and high-IQ Cronbach, 1991; McClelland, 2000; McClelland & Judd,
participants observed at the end of childhood increases, 1993; Schmidt & Hunter, 1999). Power analyses showed that
decreases, or remains stable across the adult years. A effect size and sample size are closely related. The lower the
secondary objective was to check the reality of Spear- effect size, the larger the sample needed to detect the difference
man's law in very old adults, for whom at present the data sought. In the context of Spearman's law, there is no consensus
are few and inconclusive (e.g., Anstey et al., 2003). among researchers about the expected size of correlation
differences between low and high-IQ groups. However, some
consider that it is probably small at best (Hartmann, 2006; te
1. Methods
Nijenhuis & Hartmann, 2006). When two correlations are
compared, a small effect size corresponds to a difference of
1.1. Measures and participants about .10 between the two Fisher's z-transformed r coeffi-
cients (Cohen, 1988). Consequently, to detect a significant
The French version of the WAIS-III (Wechsler, 2000) difference of this order with type II error rate (β) set to .20,
comprises exactly the same number and types of subtests as the more than 1000 participants are needed per group. Thus, at
American version. The French version was standardized on a least 8000 participants would be needed to constitute the eight
representative sample of 1104 adults aged 16 to 89 years. This groups of the present study. Of course, this number is far
sample was stratified into 12 age groups of about 80 to 100 beyond the present sample size, as it is those of almost all of
subjects each as a function of gender, geographic region, the research conducted in behavioral sciences. In addition,
urban–rural residence, and SES. For the subjects aged from 16 knowing whether the differences among correlation coeffi-
to 19 years, parental SES was considered. cients of low- and high-IQ groups of various ages are
In the present study, the entire sample was divided into four statistically significant is far from being the most interesting
age groups (16–24 years; 25–44 years; 45–69 years and 70– focus. More important is knowing how the strength of the
89 years). Each of these age groups was then divided into two relationships among abilities evolves with age in low- and
IQ groups using the mean standard score computed across two high-IQ groups and revealing the possible existence of an age-
subtests (one drawn from the verbal scale, the other from the related variation of the size of correlation differences between
performance scale). This was done to reduce the attenuation of participants of low and high intelligence. For these reasons it
correlations which occurs when the sample is split by Full was decided, in the present study, to leave behind the logic of
Scale IQ. Indeed, as noted by Detterman and Daniel (1989, p. “null hypothesis significant testing”. Instead, as often
353) “if subjects are selected by Full Scale IQ, their subscale suggested (Cohen 1994; Kline 2004; Nickerson, 2000;
scores must balance out to equal the Full Scale score. Those Schmidt 1996; Schmidt & Hunter 1997; Wilkinson & the
with higher scores on one subtest must have a lower score on APA Task Force on statistical inference, 1999), it was decided
one or more other subtests to keep their IQ within the range”. to rely on confidence intervals of average correlation
This phenomenon is much less pronounced when only one or coefficients computed for each Age × IQ group and on the
two subtests are used to form the groups because the score is effect sizes of correlation differences between IQ groups
not a sum of scores obtained from all the subtests included in across age. In this manner, the present results could possibly be
the whole scale. The Vocabulary and Block Design were used in meta-analyses devoted to the law of diminishing
chosen because these subtests were well correlated with Full returns.
Scale IQ across age in the normative sample. Using these One correlation coefficient was computed for each pair of
subtests also ensured methodological consistency with subtests for each of the eight Age × IQ groups. Thus, 528

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
4
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ

Table 1
WAIS-III scale scores, IQs, effect sizes, Levene's tests of homogeneity of variance, and ratios of standard deviations for low- vs. high-IQ participants

16–24 years old participants (N = 263) 25–44 years old participants (N = 303) 45–69 years old participants (N = 268) 70–89 years old participants (N = 270)

Low-IQ High-IQ Effect Levene's Ratio Low-IQ High-IQ Effect Levene's Ratio Low-IQ High-IQ Effect Levene's Ratio Low-IQ High-IQ Effect Levene's Ratio

ARTICLE IN PRESS
(n = 131) (n = 132) size a test of (n = 161) (n = 142) size a test of (n = 149) (n = 119) size a test of (n = 140) (n = 130) size a test of
SDs SDs SDs SDs
Means Means F p Means Means F p Means Means F p Means Means F p

B. Facon / Intelligence xx (2007) xxx–xxx


(SD) (SD) (SD) (SD) (SD) (SD) (SD) (SD)

Vocabulary 8.1(2.2) 12.1(2.4) −1.70 1.19 .28 .93 8.3(2.5) 12.0(2.3) −1.53 1.29 .26 1.05 7.7(2.6) 12.5(2.2) −1.98 3.54 .06 1.20 8.0(2.5) 12.3(2.0) −1.92 7.06 .01 1.27
Similarities 8.6(2.5) 11.7(2.6) −1.21 .21 .64 .95 8.3(2.5) 11.7(2.5) −1.33 .61 .43 1.03 8.4(2.8) 12.0(2.5) −1.37 .65 .42 1.12 8.3(2.6) 12.0(2.4) −1.48 .42 .52 1.08
Arithmetic 8.7(2.8) 11.3(2.8) −.91 .09 .77 .99 8.8(2.7) 11.6(2.4) −1.10 2.07 .15 1.14 8.8(2.6) 11.6(2.7) −1.06 .01 .93 .98 8.4(2.7) 11.8(2.4) −1.32 3.74 .05 1.13
Digit span 9.1(2.8) 11(2.8) −.69 .05 .83 1.00 9.2(2.9) 10.8(2.8) −.56 .03 .86 1.04 9.2(3.0) 11.0(2.7) −.66 1.25 .26 1.11 9.0(2.7) 11.1(3.0) −.72 .83 .36 .92
Information 8.5(2.6) 11.5(2.8) −1.13 .52 .47 .94 8.6(2.9) 11.5(2.3) −1.12 5.59 .02 1.23 8.4(2.6) 12.1(2.2) −1.51 4.28 .04 1.15 8.3(2.4) 12.0(2.4) −1.51 .05 .82 .97
Comprehension 8.5(2.8) 11.5(2.6) −1.10 .81 .37 1.09 8.7(2.8) 11.5(2.6) −1.00 .40 .53 1.08 8.5(2.4) 12.1(2.4) −1.52 .20 .65 1.02 8.2(2.2) 12.0(2.4) −1.66 .79 .37 .91
Letter–number 8.9(2.7) 11.1(2.6) −.81 .36 .55 1.04 9.0(3.0) 11.2(2.7) −.75 .10 .75 1.08 9.1(2.9) 11.2(2.6) −.77 .19 .67 1.10 8.8(2.9) 11.3(2.6) −.91 1.16 .28 1.08
sequencing
Picture 8.9(2.8) 11.3(2.7) −.85 .15 .70 1.02 9.1(3.0) 10.7(2.7) −.56 3.57 .06 1.14 8.8(3.0) 11.4(2.5) −.91 1.68 .20 1.20 8.7(2.7) 11.4(2.5) −1.03 .69 .41 1.08
completion
Coding 9.7(2.8) 10.4(2.9) −.26 .38 .54 .97 9.3(3.5) 10.9(2.8) −.49 6.61 .01 1.27 9.1(2.8) 11.2(2.8) −.75 .02 .90 1.01 8.5(2.8) 11.6(2.5) −1.17 2.05 .15 1.13
Block design 7.9(2.4) 12.2(2.4) −1.79 .72 .40 .99 8.2(2.3) 12.3(2.5) −1.71 1.15 .29 .91 8.4(2.4) 12.1(2.4) −1.56 .61 .44 .98 8.1(2.3) 12.0(2.5) −1.65 .07 .80 .93
Matrix 8.9(2.9) 11.2(2.4) −.87 7.14 .01 1.25 8.8(2.8) 11.5(2.5) −1.03 .67 .41 1.10 8.5(2.6) 11.8(2.5) −1.27 .36 .55 1.05 8.7(2.8) 11.3(2.5) −.97 1.58 .21 1.10
reasoning
Picture 8.9(2.8) 11.1(2.7) −.81 .36 .55 1.07 8.9(2.8) 11.4(2.7) −.92 .12 .73 1.02 8.9(2.9) 11.4(2.7) −.91 .31 .58 1.07 8.7(2.8) 11.3(2.6) −.96 1.33 .25 1.08
arrangement
Symbol search 9.3(3.0) 10.6(2.8) −.42 .41 .52 1.08 9.0(2.9) 11.0(2.7) −.71 1.11 .29 1.08 9.1(2.8) 11.4(3.0) −.78 1.03 .31 .95 8.6(2.6) 11.7(2.9) −1.10 .43 .51 .92
Object 8.6(2.8) 11.6(2.7) −1.08 .02 .89 1.02 8.5(2.6) 11.7(2.6) −1.27 .08 .78 1.02 8.8(2.7) 11.5(2.8) −.98 .44 .51 .96 8.7(2.6) 11.4(2.8) −.96 1.02 .31 .93
assembly
Verbal IQ 90.8(11.5) 109.3(11.6) −1.60 .02 .89 .98 91.4(11.5) 109.1(10.3) −1.62 1.28 .26 1.11 90.5(11.7) 111.9(10.8) −1.90 1.30 .26 1.08 89.6(10.9) 111.6(11.3) −1.98 .54 .46 .97
Performance IQ 92.0(11.2) 108.3(11.0) −1.47 .10 .75 1.02 92.0(12.0) 109.3(11.9) −1.45 .02 .89 1.01 91.3(12.1) 110.9(12.1) −1.62 .05 .82 1.00 90.1(12.2) 110.5(12.1) −1.68 .05 .82 1.00
Full-scale IQ 90.6(10.5) 109.7(10.3) −1.84 .26 .61 1.02 90.9(10.9) 110.1(9.7) −1.86 1.94 .16 1.13 90.0(11.2) 112.7(10.4) −2.09 .57 .45 1.08 88.9(11.0) 112.3(11.2) −2.11 .42 .52 .97
a Mean difference/pooled SD.
ARTICLE IN PRESS
B. Facon / Intelligence xx (2007) xxx–xxx 5

coefficients were estimated ([(12 × 11 subtests) / 2] × 2 IQ- Since the variability of measures influences the magnitude
levels × 4 age groups). The 66 coefficients of each of the of correlations, Levene's test was computed for each variable
eight groups were z-transformed, averaged, and 95% con- between low- and high-IQ participants in each age group.
fidence intervals around the means were calculated using the Also, to avoid relying only on these statistical tests to check the
Hays (1994) formula. Two statistical indicators were used to homogeneity of subtest score dispersion, the ratio of standard
estimate the effect size of correlation differences between IQ deviations of low- and high-IQ participants was computed at
groups for each age level (see Cohen, 1988, pp. 109–143). the four age levels for each of the 12 subtests entered into the
First, the difference between the two mean z-transformed analyses, as well as for the two selection subtests. These ratios
correlations (low-IQ mean z minus high-IQ mean z). Second, gave an indication of the importance of the differences of
the difference between the two squared mean correlations subtest score variability between ability groups.
(low-IQ r2 minus high-IQ r2). As stated by Cohen (1988), the Finally, as the two subtests used to form the IQ groups
first one gives an estimation “equally detectable” of the size of (Vocabulary and Block Design) are not perfectly correlated
the difference between two correlations, that is, an estimation with the other twelve WAIS-III subtests, a regression toward
unaffected by the value (small, medium or large) of the two the mean of the entire sample was anticipated for them.
compared correlations. The second effect size index indicates Therefore, effect sizes were computed for all subtests and the
the amount of “change” of proportion of explained variance. three IQ scales using Cohen's d index to estimate the overall
With this index, “equal amounts of proportion of variance difference in intelligence between the IQ groups at the four age
change can be meaningfully understood as equal amounts of levels.
change in amount of relationship, anywhere along the r scale”
(p. 114). 2. Results
To facilitate comparisons with other studies on Spearman's
law (e.g., Abad et al., 2003), a principal component analysis Means and standard deviations of low- and high-IQ
was conducted on the correlation matrix of the eight groups. participants in each age group, effect sizes, Levene's tests of
The goal was to obtain, for each subtest, the percentage of homogeneity of variances, and ratios of standard deviations are
variance explained by the first unrotated principal component given in Table 1. Effect sizes indicate the sizes of the
extracted from the matrix. To get comparative evidence, the differences between performances of low and high-IQ
calculations were performed, first, by excluding the selection subjects. For the two subtests employed to select the ability
subtests (Vocabulary, Block Design) from the analysis and, groups, the mean effect sizes are comparable, standing at
second, by including them. −1.75, −1.62, − 1.77 and −1.79 for the four age groups,
The mean reliability coefficients were computed using data respectively. That also applied to the other subtests of the
supplied by the manual of the French version of the WAIS-III WAIS-III since the full-scale IQ effect sizes only increase
(Wechsler, 2000, p. 234). Because the reliabilities were highly slightly with age (−1.84 [16–24 years], − 1.86 [25–44 years],
similar across groups (16–24 years = .81; 25–44 years = .81; −2.09 [45–69 years], and − 2.11 [70–89 years]). Levene's
45–69 years = .84; 70–89 years = .85), no correction for age- tests show, on the whole, that the score variances of low and
related unreliability of subtests was applied to the correlations. high-ability groups are homogenous. On the 68 Levene's tests,
Of course, to be sure that the results of the present study were only 6 are statistically significant (Table 1). Moreover, the
not affected by measurement errors, indications about the ratios of standard deviations are generally close to 1.0, thus
reliability of the subtests within each ability group would have confirming the similarity of score dispersion in low- and high-
been useful. These were not available, however. Nevertheless, ability participants, whatever their age. For example, the
even given the reliability data, a definitive answer might not average ratio of SDs ([∑(low-IQ group's SD/high group's
have been obtained because internal reliability indices are SD)] / number of subtests) of the 12 subtest scores included in
themselves possibly influenced by Spearman's law (Jensen, the correlational analyses is equal to 1.03, 1.1, 1.06, and 1.03
2003). in the four age groups, respectively.

Table 2
Mean correlation between subtests, and effect size of difference between mean correlations for low- vs. high-IQ participants in each age group
Age- Correlations between subtests Effect size
group
Low-IQ group High-IQ group
n Mean 95% confidence n Mean 95% confidence Low-IQ z minus Low-IQ r2 minus
interval interval high-IQ z high-IQ r2
16–24 131 .279 .113–.430 132 .186 .016–.346 .099 .043
25–44 161 .295 .147–.430 142 .152 −.013–.309 .151 .064
45–69 149 .324 .172–.461 119 .194 .015–.361 .140 .067
70–89 140 .365 .212–.501 130 .247 .078–.402 .130 .072

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
ARTICLE IN PRESS
6 B. Facon / Intelligence xx (2007) xxx–xxx

Table 3
Percentage of variance of each subtest explained by the first unrotated principal component extracted from the correlation matrix of the eight age × IQ
groups (excluding the selection subtests and, in parentheses, including them)
16–24 years old 25–44 years old 45–69 years old 70–89 years old
Low-IQ High-IQ Low-IQ High-IQ Low-IQ High-IQ Low-IQ High-IQ
Similarities 44.6 (47.1) 33.5 (40.6) 33.3 (34.6) 06.4 (09.6) 47.3 (48.9) 31.2 (38.6) 42.4 (42.5) 32.0 (34.6)
Arithmetic 48.7 (47.4) 48.1 (43.1) 46.4 (45.8) 15.8 (17.1) 45.3 (44.5) 44.4 (38.9) 59.4 (59.0) 50.6 (49.2)
Digit span 33.9 (30.6) 15.4 (14.4) 48.0 (44.2) 33.8 (32.5) 28.1 (26.9) 16.0 (9.90) 35.9 (35.4) 29.3 (26.1)
Information 40.8 (41.3) 40.3 (48.1) 30.8 (31.3) 06.7 (10.4) 52.0 (54.0) 34.0 (41.0) 45.1 (44.6) 36.7 (38.9)
Comprehension 47.4 (48.6) 24.5 (31.3) 39.8 (41.7) 01.9 (03.3) 50.0 (52.1) 24.4 (28.6) 34.2 (37.5) 27.1 (29.3)
Letter–number sequencing 47.3 (44.2) 25.4 (19.3) 42.2 (38.0) 44.7 (39.4) 32.8 (30.5) 17.4 (11.6) 44.2 (43.4) 18.6 (15.4)
Picture completion 33.5 (32.2) 26.2 (23.0) 21.8 (22.2) 16.1 (17.7) 36.0 (34.6) 26.2 (24.6) 36.5 (35.9) 19.3 (17.8)
Coding 04.7 (03.9) 09.3 (06.9) 20.3 (20.8) 37.7 (30.6) 29.4 (28.8) 20.5 (16.5) 54.2 (51.9) 36.9 (33.6)
Matrix reasoning 35.4 (34.4) 34.1 (31.2) 29.6 (32.3) 28.5 (27.4) 47.7 (46.5) 25.2 (25.0) 37.5 (36.0) 38.2 (39.4)
Picture arrangement 35.9 (36.0) 11.7 (11.4) 40.8 (39.0) 27.3 (27.8) 39.5 (38.7) 25.2 (24.9) 46.0 (45.0) 38.3 (36.7)
Symbol search 16.9 (16.9) 21.8 (16.5) 40.6 (38.5) 33.9 (29.4) 32.5 (30.6) 26.9 (21.6) 45.7 (43.8) 37.6 (37.9)
Object assembly 24.7 (25.6) 20.5 (15.8) 30.4 (30.1) 21.3 (21.7) 17.8 (16.9) 21.2 (23.4) 20.5 (21.8) 11.7 (11.8)
Mean 34.5 (34.0) 25.9 (25.1) 35.3 (34.9) 22.8 (22.2) 38.2 (37.8) 26.1 (25.4) 41.8 (41.4) 31.4 (30.9)

Table 2 shows that the mean correlations between subtests correlation matrices of the eight Age × IQ groups are given
are greater for low-IQ participants at all four age levels. It also in Table 3. As can be seen, the explained variances of low-IQ
indicates that the strength of intersubtest relationships slightly participants are almost always greater than those of high-IQ
increases with age for low- and for high-IQ participants. But participants, either with or without selection subtests included
this is at least partly due to the small increase of subtest in the analyses. The difference is, on the average, on the order
reliability with age. Finally, and more importantly, effect sizes of 11% of the variance, a value very close to the 12% obtained
of correlation differences between low- and high-IQ groups do by Abad et al. (2003) with the Spanish version of the WAIS-
not change with age, particularly for participants aged from 25 III. Therefore, along with Abad et al. (2003), I conclude that
to 89 years. From this standpoint, it may be concluded that this finding strongly supports the IQ-related phenomenon of
Spearman's law of diminishing returns is corroborated by the differentiation of cognitive abilities. Moreover, it appears in
present results and that age does not seem to moderate the the present study that crystallized intelligence subtests are
effect of IQ on the differentiation of cognitive abilities during more subject to differentiation, a fact which suggests,
adulthood, even in very old persons. according to Abad et al. (2003), that Spearman's law “could
The percentages of variance of subtests explained by the be a by-product of the educational differences between the low
first unrotated principal component extracted from the and high-IQ bands” (p. 163).

Fig. 1. Panel A: Correlations between subtests of the Wechsler scales in the present study and that of Facon (2006). Panel B: Data from Jensen (2003),
Table 1, p. 99.

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
ARTICLE IN PRESS
B. Facon / Intelligence xx (2007) xxx–xxx 7

3. Discussion groups of participants who show a pre-existing (i. e.,


“naturally occurring”) difference in intellectual ability
Considered on the whole, the present results as well level. However, since this approach is a new one, further
as those of Facon (2006) perhaps constitute the first two research will be necessary to confirm its interest, even
pieces of the puzzle depicting the genesis and the its superiority to the other methods, for studying the law
development of the phenomenon of diminishing returns of diminishing returns.
across the life span. More precisely, the effect of The size of the differentiation effect noted in the
intellectual efficiency on the strength of correlations present study must not be minimized. It corresponds to
between cognitive abilities becomes apparent only at the that generally reported in the literature on Spearman's
end of childhood (Facon, 2006), reaches its maximum at law. Indeed, a recent but not yet published meta-analysis
the beginning of adult age, then, as suggested by the of nearly fifty data sets by Hartmann and Nyborg (2004)
present study, remains stable going into middle and old suggests that the difference in subtest intercorrelations
age. It may of course be premature to propose a general between low- and high-IQ subjects (differing by
schema on the development of a psychological approximately 1.5 standard deviations) is generally
phenomenon across the entire life span on the basis of greater than 0.10 but seldom reaches 0.30, at least if one
only two data sets concerning only two intelligence considers the best-designed studies. Moreover, the
scales. However, other results agree with the develop- difference observed here between high- and low-IQ
mental sketch proposed here. In a recent study by Jensen participants is nearly equal to those of Abad et al. (2003)
(2003), a similar age trend was observed and, although and Reynolds and Keith (2007), and not much lower
Jensen did not emphasize this developmental effect and than that reported by Jensen (2003). It must be
did not study it in its own right, his data appear to be emphasized, however, that the subdivision of the entire
highly congruent with Facon's (2006) data and those of sample in only two IQ groups has probably contributed
the present study (see Fig. 1). to lowering the effect size. Indeed, as hypothesized by
This study has several weaknesses. The numbers of several authors (e.g., Deary et al., 1996; Pagliari, 1998;
participants in the eight age × ability groups vary only Stankov, 2002), the relationship between IQ and
between 120 and 160. Greater numbers would have correlations between tests or subtests could be well
yielded more stable estimates of correlations between and truly stronger at lower levels of mental abilities. For
variables and would have permitted more sophisticated that matter, the strongest indications of an IQ-related
statistical analyses. A more varied set of subtests would differentiation have generally been obtained when
also have been useful as coverage of a wider range of samples near the lower limit of normal variation are
cognitive abilities. A longitudinal, rather than cross- included in the study (e.g., Detterman & Daniel, 1989).
sectional approach might also have been fruitful. From From this standpoint, it would be interesting to
this standpoint, the next results of the ongoing long- investigate the age-related development of the strength
itudinal study by Arden and Plomin (2007) will be of correlations between cognitive abilities by constitut-
welcome. To summarize, these authors were not able to ing four or five IQ-levels for each age range. In this way,
find evidence of the effect of intellectual efficiency on extreme-IQ groups would be better represented than in
the differentiation of abilities among children aged 7, 9 the present study.
and 10 years. However, they might demonstrate such an Additional studies will be necessary to clarify the
effect during adolescence and adulthood in the sub- development of the strength of correlations between
sequent follow-ups of their cohort. cognitive abilities in low- and high-IQ individuals
Finally, the method chosen to select participants of across age. If the developmental sketch hypothesized
low- and high-IQ may be questioned. The use of only here were confirmed, it would be necessary to give it a
two subtests for constituting the groups poses several theoretical interpretation. In their current formulations,
potential problems detailed and analyzed by Hartmann it seems that most explanations of the law of diminish-
and Reuter (2006). Among other things, the normality of ing returns (e.g., Abad et al., 2003; Anderson, 1998;
distributions is not always optimum, the selection is Jensen, 1998) cannot integrate the age trend depicted in
performed with only an approximate measure of g, and the present data (for more details on these theoretical
the subtests used to select the groups must be excluded analyses of Spearman's law, see Facon, 2004). Abad
from subsequent statistical analyses. Their total number et al. (2003), for example, cite schooling as the principal
is, thus, necessarily smaller than it could be. These cause of the differentiation of abilities. Following the
problems led te Nijenhuis and Hartmann (2006) to well-known literature review by Ceci (1991), they
propose another approach consisting of comparing consider the relationship between schooling and IQ as

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
ARTICLE IN PRESS
8 B. Facon / Intelligence xx (2007) xxx–xxx

reciprocal. Indeed, if IQ is a good, though non- phenomenon of diminishing returns appears gradually
exhaustive, predictor of scholastic achievement, school during the ascendant phase of development.
activities themselves increase performance on IQ tests Another good candidate could be the theoretical
through the acquisition of declarative and procedural explanation developed by Detterman (1993, 1999,
knowledge. In this way, schooling would contribute to 2002), who attributes the inverse relationship to a
reducing the complexity of intelligence scale items and, deficit of one or more central cognitive processes. This
for that reason, their g-loading. As bright individuals explanation derives from his “system” theory of general
stay on longer at school and assimilate tuition more intelligence in which, schematically, the cognitive
quickly, their abilities would be, de facto, more processes are considered to be independent but
differentiated. However, even if, in the present study, integrated into an information-processing system
the differentiation effect seems to be more pronounced where they constantly interact. Some of these processes
for crystallized intelligence subtests (see, Table 3), the are said to be “central”. They have a large spectrum of
“schooling” hypothesis doesn't appear to be able to influence and, thus, affect the functioning of the entire
explain the increase of correlations between subtests system. Therefore, and unlike the peripheral processes,
among individuals of low-IQ with increasing chron- their impairment limits the efficiency of the whole
ological age, and their relative stability among those system. It is for this reason that abilities are better
with high-IQ (Fig. 1). Indeed, if schooling were the root correlated among persons with low-IQ. They have
cause of the IQ-related differentiation, correlations inefficient central processes, which limits the global
between subtests should decrease with chronological level of functioning of their information-processing
age among children and adolescents with high-IQs system and thus gives the individual rather similar
because the educational level increases with age and IQ. scores across the different cognitive tasks. Indeed, the
Other interpretations must therefore be envisaged. In contribution of the peripheral processes to the variance
this respect, perhaps Anderson's (2001) mutation theory of intellectual performance is diminished because their
of g would be a good candidate. In this conception, the g efficiency is itself largely limited by the central deficits.
factor would have for substratum cortical functioning Conversely, since the central processes of persons with
which, itself, is viewed as partially determined by the high-IQ are intact and their efficiency probably situated
genotype of the individual. In concrete terms, the fact above the requisite minimum threshold in numerous
that cognitive abilities have a shared genetic component cognitive tasks, individual differences come principally
would explain the positive relations between test scores from peripheral processes. Thus, performance in ability
and, thus, the g factor. In that context, the average level tests is more variable and less correlated among high-IQ
of cognitive traits would be inversely proportional to the participants. In this theoretical perspective, one may
frequency of minor deleterious genetic mutations, which suppose that one or more of these central processes–for
are more common than previously estimated (Crow, example those related to executive control–develop
1999; Eyre-Walker & Keightley, 1999). These mutations with age and hold a growing role in the cognitive
would affect cortical functioning and, in turn, would functioning of the individual. The cognitive abilities of
diminish trait levels (for a discussion on this point, see low-IQ subjects should therefore become more corre-
Novartis Foundation Symposium, 2002, pp. 144–147). lated because they depend on sources of variation
Interestingly, these mutations would also increase whose importance increases in the course of child
relationships between cognitive scores and, for this development.
reason, would explain the inverse relationship between Beyond the law of diminishing returns, one needs to
IQ and strength of correlations (see Anderson, 2001 for a consider, in the present study, the pattern of change of
mathematical elaboration of this hypothesis). This correlations across age. Fig. 1 shows that the strength of
theoretical conception of the g factor and of the law of relationships between cognitive abilities clearly
diminishing returns is based on a certain number of increases with age among participants of low-IQ,
assumptions and simplifications and remains to be whereas it remains almost stable among those with
confirmed. However, it supplies a simple interpretation high-IQ. This developmental trend is in direct contra-
of Spearman's law which, in addition, could integrate the diction with the age-differentiation hypothesis which
developmental trend presented in Fig. 1. Indeed, if the states that the g saturation of intelligence markers
amount of variance of cognitive abilities explained by decreases from childhood to maturity (Garrett, 1946).
the genotype really increases during childhood (Plomin, This hypothesis gave rise to numerous studies which
2003; Plomin et al., 2001; Plomin & Petrill, 1997; have led to very contradictory results. In their recent
Plomin & Spinath, 2004), it is quite conceivable that the meta-analysis of more than 80 studies designed to test

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
ARTICLE IN PRESS
B. Facon / Intelligence xx (2007) xxx–xxx 9

this hypothesis, Hartmann and Nyborg (2004, p. 78) Ackerman, P. L. (1988). Determinants of individual differences during
noted only “a non-significant tendency towards a U- skill acquisition: Cognitive abilities and information processing.
Journal of Experimental Psychology. General, 117, 288−318.
curvilinear relationship between g saturation and age, Ackerman, P. L. (2005). Ability determinants of individual differences
with the minimum g saturation from ages 18–34.” In in skilled performance. In R. J. Sternberg & J.E. Pretz (Eds.),
fact, it appeared that “although about 50% of the studies Cognition and intelligence: Identifying the mechanisms of mind
support U-curvilinear relationship in some way or the (pp. 143−159). Cambridge, UK: Cambridge University Press.
Ackerman, P. L., & Cianciolo, A. T. (2000). Cognitive, perceptual-
other, approximately the same amount of studies
speed, and psychomotor determinants of individual differences
directly contradict this either by finding no relationship during skill acquisition. Journal of Experimental Psychology.
or the exact opposite of what the U-curvilinear Applied, 6, 259−290.
relationship–hypothesis would predict.” Given these Ackerman, P. L., & Cianciolo, A. T. (2002). Ability and task constraint
results, it is impossible to conclude that cognitive determinants of complex task performance. Journal of Experi-
abilities differentiate with age, since recent and well- mental Psychology. Applied, 8, 194−208.
Anderson, M. (1998). Mental retardation, general intelligence and
designed studies (e.g., Bickley, Keith, & Wolfle, 1995; modularity. Learning and Individual Differences, 10, 159−178.
Escorial, Juan-Espinosa, García, Rebollo, & Colom, Anderson, B. (2001). g as a consequence of shared genes. Intelligence,
2002; Juan-Espinosa et al., 2000, 2002; Keith, 1997) 29, 367−371.
show that the factorial structure of major intelligence Anstey, K. J., Hofer, S. M., & Luszcz, M. A. (2003). Cross-sectional
scales (e.g., WISC-III, Woodcock-Johnson, WAIS-III, and longitudinal patterns of dedifferentiation in late-life cognitive
and sensory function: The effects of age, ability, attrition, and
etc.) is fairly invariant across age groups. These occasion of measurement. Journal of Experimental Psychology.
contradictory results might be the consequence of the General, 132, 470−487.
lack of control of some influential variables, notably the Arden, R., & Plomin, R. (2007). Scant evidence for Spearman's law of
intellectual efficiency of participants. Indeed, the latter diminishing returns in middle childhood. Personality and
is neglected in almost all studies conducted on the age- Individual Differences, 42, 743−753.
Bayley, N. (1949). Consistency and variability in the growth of
differentiation hypothesis. Yet, if, in different studies, intelligence from birth to eighteen years. Journal of Genetic
the rate of participants of low- or high-IQ increases or Psychology, 75, 165−196.
decreases with age, there may ensue patterns of Bayley, N. (1954). Some increasing parent–child similarities during the
contradictory results inherent to the effect of intellectual growth of children. Journal of Educational Psychology, 45, 1−20.
efficiency on the degree of differentiation of abilities. Bayley, N. (1970). Development of mental abilities. In P. H. Mussen
(Ed.), (3rd ed.). Carmichael's manual of child psychology, Vol. 1
Therefore, rather than deplore the inconsistencies of the (pp. 1163−1209). New York: John Wiley & Sons.
literature, what is needed is to undertake others Bayley, N., & Jones, H. E. (1937). Environmental correlates of mental
researches including more potentially influential vari- and motor development: a cumulative study from infancy to six
ables than only the age of participants. years. Child Development, 8, 329−341.
Bayley, N., & Schaefer, E. S. (1964). Correlations of maternal and
Finally, this study indicates that the IQ-related
child behaviors with the development of mental abilities: Data
phenomenon of differentiation is still observable from the Berkeley Growth Study. Monographs of the Society for
among adults aged from 70 to 89 years, an age range Research in Child Development, 29, 2−80.
for which data on Spearman's law are very scarce and Bickley, P. G., Keith, T. Z., & Wolfle, L. M. (1995). The three stratum
not always coherent. For instance, a recent longitudinal theory of cognitive abilities: Test of the structure of intelligence
study with very old adults by Anstey et al. (2003) gives across the life span. Intelligence, 20, 309−328.
Burt, C. (1954). The differentiation of intellectual ability. British
only weak support to this law in that the inverse Journal of Educational Psychology, 24, 76−90.
relationship does not appear consistently across the Carlstedt, B. (2001). Differentiation of cognitive abilities as a function
different waves of data collection. From this standpoint, of level of general intelligence: A latent variable approach. Mul-
and even if more data in this connection are obviously tivariate Behavioral Research, 36, 589−609.
necessary, the present study provides, for the elderly, Ceci, S. J. (1991). How much does schooling influence general
intelligence and its cognitive components? A reassessment of the
results that are congruent with those of young and evidence. Developmental Psychology, 27, 703−722.
middle aged adults. Cohen, J. (1988). Statistical power analysis for the behavioral
sciences (2nd ed.). Hillsdale, NJ: Laurence Erlbaum Associates.
References Cohen, J. (1994). The earth is round (p b .05). American Psychologist,
49, 997−1003.
Abad, F. J., Colom, R., Juan-Espinosa, M., & García, L. F. (2003). Cronbach, L. J. (1991). Emerging views on methodology. In T. D.
Intelligence differentiation in adult samples. Intelligence, 31, Wachs & R. Plomin (Eds.), Conceptualization and measurement
157−166. of organism–environment interaction (pp. 87−104). Washington,
Ackerman, P. L. (1987). Individual differences in skill learning: An DC: American Psychological Association.
integration of psychometric and information processing perspec- Crow, J. F. (1999). The odds of losing at genetic roulette. Nature, 397,
tives. Psychological Bulletin, 102, 3−27. 293−294.

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
ARTICLE IN PRESS
10 B. Facon / Intelligence xx (2007) xxx–xxx

Deary, I. J., Egan, V., Gibson, G. J., Austin, E. J., Brand, C. R., & current standing of the theory. Unpublished manuscript, University
Kellaghan, T. (1996). Intelligence and the differentiation hypoth- of Aarhus, Denmark.
esis. Intelligence, 23, 105−132. Hartmann, P., & Reuter, M. (2006). Spearman's “Law of Diminishing
Detterman, D. K. (1993). Giftedness and intelligence: One and the same? Returns” tested with two methods. Intelligence, 34, 47−62.
In G. R. Bock & K. Ackrill (Eds.), The origins and development of Hartmann, P., & Teasdale, T. W. (2004). A test of Spearman's “Law of
high ability. CIBA Foundation Symposium (pp. 22−43). Chichester, Diminishing Returns” in two large samples of Danish military
UK: John Wiley & Sons. draftees. Intelligence, 32, 499−508.
Detterman, D. K. (1999). The psychology of mental retardation. In- Hartmann, P., & Teasdale, T. W. (2005). Spearman's “Law of
ternational Review of Psychiatry, 11, 26−33. Diminishing Returns” and the role of test reliability investigated
Detterman, D. K. (2002). General intelligence: Cognitive and biological in a large sample of Danish military draftees. Personality and
explanations. In R. J. Sternberg & E.L. Grigorenko (Eds.), The Individual Differences, 39, 1193−1203.
general factor of intelligence. How general is it? (pp. 223−243). Hays, W. L. (1994). Statistics. New York: Harcourt Brace.
Mahwah, NJ: Lawrence Erlbaum Associates. Hertzog, C., & Bleckley, M. K. (2001). Age differences in the structure
Detterman, D. K., & Daniel, M. H. (1989). Correlations of mental tests of intelligence: Influences of information processing speed. Intel-
with each other and with cognitive variables are highest for low-IQ ligence, 29, 191−217.
groups. Intelligence, 13, 349−359. Honzik, M. P. (1957). Developmental studies of parent–child
Escorial, S., Juan-Espinosa, M., García, L. F., Rebollo, I., & Colom, R. resemblance in intelligence. Child Development, 28, 215−228.
(2002). Does g variance change in adulthood? Testing the age de- Honzik, M. P. (1963). A sex difference in the age of onset of the
differentiation hypothesis across sex. Personality and Individual parent–child resemblance in intelligence. Journal of Educational
Differences, 30, 395−408. Psychology, 54, 231−237.
Evans, M. G. (1999). On the asymmetry of g. Psychological Reports, Honzik, M. P. (1967). Environmental correlates of mental growth:
85, 1059−1069. Prediction from the family setting at 21 months. Child Develop-
Eyre-Walker, A., & Keightley, P. D. (1999). High genomic deleterious ment, 38, 337−364.
mutation rates in hominids. Nature, 397, 344−346. Hulin, C. L., Henry, R. A., & Noon, S. L. (1990). Adding a dimension:
Facon, B. (2002). Intelligence générale, différenciation des aptitudes et Time as a factor in the generalizability of predictive relationships.
retard mental (General intelligence, ability differentiation, and Psychological Bulletin, 107, 328−340.
mental retardation). Revue Européenne de Psychologie Appliquée Hunt, E. (1997). The concept and utility of intelligence. In B.
(European Review of Applied Psychology), 52, 3−12. Devlin, S. E. Fienberg, D. P. Resnick, & K. Roeder (Eds.), In-
Facon, B. (2003). L'organisation des aptitudes cognitives est-elle telligence, genes and success. Scientists respond to the Bell
hiérarchique? [(Is the organization of cognitive abilities hierarch- Curve (pp. 157−176). New York: Springer-Verlag.
ical?)]. Revue Européenne de Psychologie Appliquée (European Jensen, A. R. (1998). The g factor: The science of mental ability.
Review of Applied Psychology), 53, 189−198. Westport, CT: Praeger Publishers.
Facon, B. (2003). Sur la loi des rendements décroissants. Efficience Jensen, A. R. (2003). Regularities in Spearman's law of diminishing
intellectuelle et facteur général [(On the law of diminishing returns: returns. Intelligence, 31, 95−105.
Intellectual efficiency and g factor)]. L'Année Psychologique, 103, Juan-Espinosa, M., García, L. F., Colom, R., & Abad, F. J. (2000). Testing
81−102. the age related differentiation hypothesis through the Wechsler's
Facon, B. (2004). Are correlations between cognitive abilities highest scales. Personality and Individual Differences, 29, 1069−1075.
in low-IQ groups during childhood? Intelligence, 32, 391−401. Juan-Espinosa, M., García, L. F., Escorial, S., Rebollo, I., Colom, R.,
Facon, B. (2006). Does age moderate the effect of IQ on the & Abad, F. J. (2002). Age dedifferentiation hypothesis. Evidence
differentiation of cognitive abilities during childhood? Intelli- from the Wais III. Intelligence, 30, 395−408.
gence, 34, 375−386. Keith, T. Z. (1997). Using confirmatory factor analysis to aid in
Facon, B. (2007). Sur la différenciation des aptitudes cognitives de 730 understanding the constructs measured by intelligence tests. In D. P.
l'enfant (On the differentiation of child's cognitive abilities). Revue Flanagan, J. L. Genshaft, & P. L. Harrison (Eds.), Contemporary
Européenne de Psychologie Appliquée (European Review of intellectual assessment. Theories, tests, and issues (pp. 373−402).
Applied Psychology), 57, 213−223. New York: The Guilford Press.
Fleishman, E. A. (1972). On the relation between abilities, learning, Kennedy, R. S., Dunlap, W. P., Turnage, J. J., & Wilkes, R. L. (1993).
and human performance. American Psychologist, 27, 1017−1032. Human intelligence in a repeated-measures setting: Communalities
Fleishman, E. A., & Mumford, M. D. (1989). Abilities as causes of of intelligence with performance. In D. K. Detterman (Ed.), Indivi-
individual differences in skill acquisition. Human Performance, 2, dual differences and cognition. Current topics in human intelligence,
201−223. Vol. 3 (pp. 259−281). Norwood, NJ: Ablex Publishing Corporation.
Fleishman, E. A., & Quaintance, M. K. (1984). Taxonomies of human Kline, R. B. (2004). Beyond significance testing: Reforming data
performance. The description of human tasks. Orlando, FL: analysis methods in behavioral research. Washington, DC:
Academic Press. American Psychological Association.
Fogarty, G. J., & Stankov, L. (1995). Challenging the “law of Legree, P. J., Pifer, M. E., & Grafton, F. C. (1996). Correlations among
diminishing returns”. Intelligence, 21, 157−174. cognitive abilities are lower for higher ability groups. Intelligence,
Garrett, H. E. (1946). A developmental theory of intelligence. Amer- 23, 45−57.
ican Psychologist, 1, 372−378. Lewis, M., & McGurk, H. (1972). Evaluation of infant intelligence.
Hartmann, P. (2006). Spearman's Law of Diminishing Returns. A look Infant intelligence scores—true or false? Science, 178, 1174−1177.
at age differentiation. Journal of Individual Differences, 27, Li, S. C., Lindenberger, U., Hommel, B., Aschersleben, G., Prinz, W.,
199−207. & Baltes, P. B. (2004). Transformations in the couplings among
Hartmann, P., & Nyborg, H. (2004). Spearman's “Law of Diminishing intellectual abilities and constituent cognitive processes across the
Returns”: A critical eye on a century of methods, results, and life span. Psychological Science, 15, 155−163.

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004
ARTICLE IN PRESS
B. Facon / Intelligence xx (2007) xxx–xxx 11

Lienert, G. A., & Faber, C. (1963). Über die faktorenstruktur des Plomin, R., & Spinath, F. M. (2004). Intelligence: Genetics, genes, and
HAWIK auf verschiedenen alters-und intelligenzniveaus (On the genomics. Journal of Personality and Social Psychology, 86,
factorial structure of the HAWIK for different age and ability 112−129.
levels). Diagnostica, 9, 3−11. Reynolds, M. R., & Keith, T. Z. (2007). Spearman's law of
Lynn, R. (1990). Does Spearman's g decline at high IQ levels? Some diminishing returns in hierarchical models of intelligence for
evidence from Scotland. The Journal of Genetic Psychology, 153, children and adolescents. Intelligence, 35, 267−281.
229−230. Schmidt, F. L. (1996). Statistical significance testing and cumulative
McCall, R. B. (1976). Toward an epigenetic conception of mental knowledge in psychology: Implications for training of researchers.
development in the first three years of life. In M. Lewis (Ed.), Psychological Methods, 1, 115−129.
Origins of intelligence (pp. 97−122). New York: Plenum Press. Schmidt, F. L., & Hunter, J. E. (1997). Eight common but false
McCall, R. B., Hogarty, P. S., & Hurlburt, N. (1972). Transitions in objections to the discontinuation of significance testing in
infant sensorimotor development and the prediction of childhood the analysis of research data. In L. L. Harlow, S. A. Mulaik, &
IQ. American Psychologist, 27, 728−748. J. H. Steiger (Eds.), What if there were no significance tests?
McClelland, G. H. (2000). Increasing statistical power without (pp. 37−64). Mahwah, NJ: Laurence Erlbaum Associates.
increasing sample size. American Psychologist, 55, 963−964. Schmidt, F. L., & Hunter, J. E. (1999). Theory testing and measure-
McClelland, G. H., & Judd, C. M. (1993). Statistical difficulties of ment error. Intelligence, 27, 183−198.
detecting interactions and moderator effects. Psychological Spearman, C. (1927). The abilities of man. Their Nature and
Bulletin, 114, 376−390. Measurement. London, UK: Macmillan and Co., Limited.
Nesselroade, J. R., & Thompson, W. W. (1995). Selection and related Stankov, L. (2002). g: A diminutive general. In R. J. Sternberg & E.L.
threats to group comparisons: An example comparing factorial Grigorenko (Eds.), The general factor of intelligence. How general
structures of higher and lower ability groups of adult twins. Psy- is it? (pp. 19−37). Mahwah, NJ: Lawrence Erlbaum Associates.
chological Bulletin, 117, 271−284. te Nijenhuis, J., & Hartmann, P. (2006). Spearman's “Law of
Nickerson, R. S. (2000). Null hypothesis significance testing: A Diminishing Returns” in samples of Dutch and immigrant children
review of an old and continuing controversy. Psychological and adults. Intelligence, 34, 437−447.
Methods, 5, 241−301. Tideman, E., & Gustafsson, J. E. (2004). Age-related differentiation of
Novartis Foundation Symposium (2002). The nature of intelligence. cognitive abilities in ages 3–7. Personality and Individual
Chichester, UK: John Wiley & Sons. Differences, 36, 1965−1974.
Pagliari, H. C. (1998). Individual and developmental differences in Wechsler, D. (2000). Echelle d'Intelligence de Wechsler pour Adultes,
positive manifold: a historical and empirical investigation of the troisième édition (WAIS-III). Paris: Les Editions du Centre de
differentiation hypothesis. Unpublished doctoral thesis, University Psychologie Appliquée.
of Edinburgh. Wilkinson, L., & Task Force on Statistical Inference (1999). Statistical
Plomin, R. (2003). General cognitive ability. In R. Plomin, J. C. methods in psychology journals: Guidelines and explanations.
DeFries, I. W. Craig, & P. McGuffin (Eds.), Behavioral genetics in American Psychologist, 54, 594−604.
the postgenomic era (pp. 183−201). Washington, DC: American Zelinski, E. M., & Lewis, K. L. (2003). Adult age differences in
Psychological Association. multiple cognitive functions: Differentiation, dedifferentiation, or
Plomin, R., DeFries, J. C., McClearn, G. E., & McGuffin, P. (2001). process-specific change? Psychology and Aging, 18, 727−745.
Behavioral genetics. New York: Worth Publishers.
Plomin, R., & Petrill, S. A. (1997). Genetics and intelligence: What's
new? Intelligence, 24, 53−77.

Please cite this article as: Facon, B., How does the strength of the relationships between cognitive abilities evolve over the life span for low-IQ
vs high-IQ adults? Intelligence (2007), doi:10.1016/j.intell.2007.11.004

You might also like