Personality and Individual Differences 135 (2018) 222–228

Contents lists available at ScienceDirect

Personality and Individual Differences

journal homepage: [Link]/locate/paid

g, mutualism, and development: Cross-sectional evidence from Iranian T

schoolchildren
⁎
Seyed Rouhollah Shahabia, , Francisco J. Abadb, Roberto Colomb
a
Institute for Humanities and Cultural Studies, Iran
b
Universidad Autónoma de Madrid, Spain

A R T I C LE I N FO A B S T R A C T

Keywords: Different models might explain one the most replicated findings in Psychology, namely, the positive correlation
Positive manifold among diverse tests of cognitive ability (positive manifold). The g factor and the mutualism models are examples.
Intelligence development Mutualism suggests that the manifold emerges progressively during development, as a consequence of inter-
g factor actions among originally uncorrelated cognitive process. The g factor model departs from this perspective and
Mutualism
assumes that there is an underlying latent factor remaining fairly constant across development. The present
study investigates these two approaches considering 661 children (age range = 6 to 12 years) recruited from
primary school in Tehran. A wide set of cognitive tests, selected under the CHC framework, was completed by
these individuals. First, the measurement model was tested across age for demonstrating that the factor of
interest is properly tapped by the administered measures. Afterwards, the scores were submitted to statistical
analyses for testing the models' prediction: the mutualism approach predicts stronger correlations among scores
as individuals grow up, whereas the g model does not. Results revealed metric invariance regarding the obtained
g factor across the age range. Furthermore, the amount of variance explained by this factor remained fairly
constant. In conclusion, the findings were consistent with the g model.

1. Introduction factors inherit their structure entirely from the data (Markus &
Borsboom, 2013). In contrast, reflective models posit that there is a
Beyond discrepancies in the details related with the intelligence common cause that can be identified using factor analysis and perhaps
concept and its measurement, the scientific community agrees about supported by some biological substrate (Kievit et al., 2011).
the ubiquitous presence of positive relationships among heterogeneous van der Maas et al. (2006) and van der Maas et al. (2016, 2017)
cognitive tests and tasks (positive manifold). People from the general endorse the dynamic mutualism approach when accounting for the
population can be ordered according to their performance across these positive manifold. According to this approach, cognitive processes will
tests and tasks. This well replicated fact is behind the g factor frame- be uncorrelated in early developmental stages. The positive manifold
work (Abad et al., 2015; Carroll, 1993; Jensen, 1998; Johnson et al., will emerge as a consequence of mutually beneficial interactions among
2004, 2008). these processes. This mutualism departs from the classic framework
Nevertheless, as discussed by Kovacs and Conway (2016) the posi- based on the g factor.
tive manifold behind the g model might be explained by the so-called Gignac (2014) tested the mutualism and g models analyzing the
mutualism approach (van der Maas et al., 2006): g is not an underlying correlation matrices published in the manuals of the WPPSI-III
latent factor, but an epiphenomenon that emerges during development. (Wechsler, 2002), the WISC-IV (Wechsler, 2003b), and the WAIS-IV
The interaction among independent cognitive processes may give rise to (Wechsler, 2008). Observed results departed from the mutualism model
the epiphenomenon. because the obtained g factor was (a) present, (b) fairly constant across
There is no definitive way to test the likelihood of both models from development, and (c) it explained the same amount of variance in
a statistical perspective, and, therefore, indirect evidence must be cognitive performance regardless of the age considered.
considered. In this regard, it may be important to note that models In the present study we follow Gignac's approach. We will use
based on correlations can be divided in two categories: formative and Omega (ω) for testing the main prediction of the mutualism and g
reflective (Kovacs & Conway, 2016). In formative models, constructs/ models. This statistic represents the proportion of variance explained by

⁎
Corresponding author at: Department of Psychology, Institute for Humanities and Cultural Studies, West 64 st, Kordestan Highway, Tehran, Iran.
E-mail address: [Link]@[Link] (S.R. Shahabi).

[Link]
Received 25 April 2018; Received in revised form 10 July 2018; Accepted 15 July 2018
0191-8869/ © 2018 Published by Elsevier Ltd.
S.R. Shahabi et al. Personality and Individual Differences 135 (2018) 222–228

a general factor derived from the correlations computed from a cog- 2.2.5. Symbol Search (WISC-IV)
nitive battery composed by various tests. This subtest was administered for measuring processing speed.
The cognitive measures administered here were selected following Participants must scan a search group indicating whether one of the
the Cattell-Horn-Carroll (CHC) model (McGrew, 2009). Therefore, we symbols in the target group matches. The reliability value for the nor-
measured Verbal Comprehension-Knowledge (Gc), Fluid/Novel Rea- mative sample was 0.83 (test-retest) (Abedi et al., 2015).
soning (Gf), Visual–Spatial Ability (Gv), Short-Term Memory (Gsm),
Long-Term Retrieval (Glr), Auditory Processing (Ga), Processing Speed 2.2.6. Arithmetic subtest (WISC-IV)
(Gs), Quantitative Knowledge (Gq), Reading and writing abilities This subtest measured mathematic achievement. Participants must
(Grw), Domain Specific Knowledge (Gkn), Reaction Speed (Gt), Ki- mentally solve series of arithmetic problems. The reliability value for
nestheric processing (Gk), and Tactile Processing (Gh). A general factor the normative sample was 0.80 (test-retest) (Abedi et al., 2015).
(g) was also obtained (Schneider & Newman, 2015).
Using this broad spectrum of cognitive tests, we will test the pre- 2.2.7. Delayed memory
diction made by the mutualism and g models. If the first model fits the We measured this ability by the Logical Memory and Paired Words
data, then the g factor should explain more variance as individuals grow Subtests from the Wechsler's Clinical Memory Scale (2009). In the
up. In contrast, if the second model fits the data, then the g factor Logical Memory Subtest, a short story is orally presented to the parti-
should explain the same amount of variance across ages. Relying on the cipant. The recall test is performed 20 min later. In the Paired Words
findings reported by Gignac (2014) we anticipate that the g model will Subtest, 10 paired words are orally presented. The recall test is per-
provide a better account of the data. formed 20 min later.

2. Method 2.2.8. Finger Tapping Test (Mueller & Piper, 2014)

This test was administered for measuring motor speed. Participants
2.1. Participants used their left and right index fingers. The dependent variable was the
number of correct taps delivered by both hands.
Participants were 661 schoolchildren with a mean age of 8.93 years
(SD = 1.55). They were organized in six groups: 6–7 years old 2.2.9. Tone Screening Test (Kayser, 2015)
(n = 101), 7–8 (n = 113), 8–9 (n = 119), 9–10 (n = 125), 10–11 This test was administered for measuring auditory processing.
(n = 97), and 11–12 years old (n = 106). The sample included 353 Participants indicated whether two 300-ms pure tones separated by a
boys and 254 girls (54 unknown). They were recruited from primary 100-ms interval were the same or different in pitch by pressing the S or
schools from Tehran and informed written consent was obtained from D key on the keyboard. The tone frequencies ranged from 325 Hz to
the schools. The chosen schools represented a wide range of socio- 1994 Hz. When the tones in a pair were different, the frequency ratios
economic areas of the city. Only children enrolled in regular classes were 0.67, 0.75, 0.85, 0.90, or 0.95. After 10 practice trials, 60 test
were asked to participate. trials were presented. The test trials consisted of 30 trials in which the
tones in a pair were the same pitch (tone ratio 1.0) and 30 trials in
which the tones were different, with each of the five frequency ratios
2.2. Measures
occurring once per block of 10 trials. Trial types were randomly dis-
tributed within each block.
2.2.1. Matrix Reasoning (WISC-IV)
The Matrix Reasoning subtest from the WISC IV was administered
2.2.10. Simple and Selective Reaction Time
for measuring fluid reasoning (Tamm & Juranek, 2012; De Alwis,
In the simple reaction time task, a colored square (target) is dis-
2011). The WISC-IV (Wechsler, 2003a) was standardized for Iran in
played for 2 s and the participant should press the space button on the
2009 (Abedi, Sadeghi, & Rabeei, 2015). Participants view an in-
computer keyboard as soon as possible. In the Selective Reaction Time
complete matrix (or series) and select the response option that com-
task, visual and auditory stimuli (250 Hz) were displayed or played
pletes the matrix (or series). The reliability value for the normative
randomly at two-seconds interval. Participants must press the right or
sample was 0.85 (test-retest).
left shift button on the keyboard for providing their answer. Test-retest
reliability values were 0.76 and 0.90 respectively.
2.2.2. Block Design (WISC-IV)
This subtest measured visual processing. Participants view a con-
2.2.11. Object Recognition Task
structed model and picture. Afterwards, they use red and white blocks
This task was administered for measuring tactile ability. Usual ob-
to recreate the design within a specific time limit. The reliability value
jects included paper clips, button, key, teaspoonful, pen head; geo-
for the normative sample was 0.78 (test-retest).
metric figure includes triangle, square and circle Persian alphabet letter
include‫ ﯼ‬،‫ ﻭ‬،‫ ﻡ‬،‫ ﻝ‬set in a bag.
2.2.3. Vocabulary & General information (WISC-IV)
These subtests were administered for measuring crystallized 2.3. Procedure
knowledge. In the Vocabulary subtest participants define words pre-
sented visually or orally. In the General information subtests partici- The children were tested individually in a quiet area at their
pants answer questions addressing a broad range of general knowledge schools. All computerized tasks were administered in one session,
topics. The reliability values for the normative sample were 0.94 and which lasted approximately 15–20 min. Afterwards they completed the
0.93, respectively (Abedi et al., 2015). remaining tests and tasks administered in the same order. Total testing
time was approx. 90–100 min. Each testing session included two short
2.2.4. Digit Span Forward/Backward (WISC-IV) breaks. Trained psychologists administered all tests and tasks.
These subtests were administered for measuring short-term memory
and working memory. In the Digit Span Forward subtest participants 2.4. Analyses
hear a sequence of numbers recalling them in the same order. In the
Digit Span Backward subtest participants must recall the same numbers 2.4.1. Descriptive statistics (means, standard deviations, and correlations)
in reverse order. The reliability value for the normative sample was The following subtests were included in the analyses: Matrix
0.71 (test-retest) (Abedi et al., 2015). Reasoning (MR), Block Design (BD), Vocabulary (VO), Information

223
S.R. Shahabi et al. Personality and Individual Differences 135 (2018) 222–228

Table 1
Descriptive statistics for subtests' raw scores by age group.
MR BD VO IN DS SS AR DM PW FT TS SiRT SeRT OR

6–7 years
N 101 101 101 101 101 101 101 100 100 99 100 95 99 101
Mean 10.7 19.2 9.0 8.4 11.1 15.1 8.8 6.4 2.6 80.3 11.9 467.4 671.3 14.5
SD 3.8 8.4 5.5 2.4 2.7 6.8 3.4 6.5 2.1 12.5 2.9 155.1 172.4 3.6
Skewness 0.4 0.3 1.2 1.2 −0.1 −0.7 0.1 1.2 0.5 0.0 0.4 1.7 0.8 0.3

7–8 years
N 113 112 113 113 113 113 113 112 113 112 113 101 103 111
Mean 12.7 20.8 12.2 10.1 11.8 15.6 11.3 7.5 3.2 85.6 12.6 417.7 639.2 15.5
SD 4.2 8.5 5.9 1.9 2.6 5.2 4.0 5.7 1.9 11.8 3.0 117.4 160.6 3.9
Skewness 0.4 0.3 0.5 0.8 0.9 0.4 0.1 0.7 0.2 −0.3 −0.1 0.9 0.8 −0.1

8–9 years
N 119 119 119 119 119 119 119 115 118 118 119 99 100 119
Mean 15.4 25.3 18.3 12.0 13.0 15.8 14.7 10.0 3.5 91.1 12.9 393.0 635.9 16.4
SD 5.2 8.2 8.4 2.7 2.9 4.5 4.1 5.9 2.3 12.7 2.9 120.2 187.7 4.2
Skewness 0.1 0.5 0.3 0.7 0.7 0.1 −0.2 0.5 0.5 −0.5 0.0 1.1 0.9 −0.2

9–10 years
N 125 125 125 125 125 125 125 124 125 125 125 106 107 125
Mean 17.8 29.5 24.9 13.9 14.5 18.6 16.9 12.3 3.8 100.2 14.1 371.7 578.3 18.2
SD 5.1 9.3 9.4 3.2 2.8 5.8 3.5 6.1 1.7 11.9 2.8 121.1 135.3 3.9
Skewness −0.1 0.2 0.4 0.6 0.8 −0.4 0.2 0.4 0.0 0.3 −0.8 1.7 0.5 −0.4

10–11 years
N 97 97 96 97 97 97 97 97 97 94 97 91 90 97
Mean 19.2 33.9 28.0 16.7 15.3 22.1 18.2 13.6 4.2 107.2 14.6 353.8 576.9 19.0
SD 4.6 9.0 8.7 3.4 3.2 5.6 4.0 5.4 1.9 13.2 2.4 107.1 164.0 2.9
Skewness −0.1 0.1 −0.1 −0.1 0.9 0.0 −0.6 0.1 0.3 0.0 −0.5 1.0 1.2 −1.1

11–12 years
N 106 106 105 105 106 106 106 106 106 104 106 103 103 106
Mean 19.9 34.8 32.6 17.0 15.3 22.6 19.0 13.6 4.0 110.9 14.5 336.5 511.7 19.0
SD 4.4 10.6 9.2 3.5 3.4 5.8 3.9 5.3 2.1 12.8 2.8 96.2 124.1 4.6
Skewness −0.5 −0.2 −0.1 −0.1 1.1 −0.2 −0.2 0.0 0.0 0.2 −0.5 1.5 1.4 −0.7

(IN), Digit Span (DS), Symbol Search (SS), Arithmetic (AR), Delayed modeling (Muthén & Muthén, 2017). The maximum likelihood (ML)
Memory (DM), Paired Words (PW), Finger Tapping Test (FT), Tone estimator was used (see Supplementary Information).
Screening Test (TS), Simple Reaction Time (SiRT) and Selective
Reaction Time (SeRT), and Object Recognition Task (OR) (Tables 1–4). 2.4.3. Omega coefficients
Descriptive statistics were obtained from the raw scores. However, The omega coefficient (ω) was computed for each age group using
for easiness of interpretation, scaled subtests scores were obtained and the metric invariance model. In the unidimensional model, ω indicates
analyzed (all subtests' scores were linearly transformed to have zero the proportion of total score variance due to the general factor. Note
mean and variance equal to one in the 6–7 age group). Additionally, that the total score was computed as the sum of the scaled scores.
scaled scores measures based on response time (i.e., SiRT and SeRT) Reliability was computed as (Raykov, 2004):
were reversed for easiness of interpretation. 2
J
⎛ ⎞φ
⎜ ∑ λ jg ⎟ g
2.4.2. Multi-group Confirmatory Factor Analysis (MG-CFA) ω= ⎝ j=1 ⎠
2
Measurement invariance across age was assessed by means of Multi- ⎛⎛ J J
⎞
∑ λ ⎞ φ + ⎛∑ θ ⎞ + ⎛∑ θ ⎞
Group Confirmatory Factor Analyses (MG-CFA). First, two configural ⎜ ⎜ j = 1 jg ⎟ g ⎜ j = 1 jj ′⎟ ⎜ j ≠ j ′ jj ′ ⎟ ⎟
⎝⎝ ⎠ ⎝ ⎠ ⎝ ⎠⎠
invariance models were compared: (a) a unidimensional model, and (b)
a unidimensional model with correlated residuals among tests mea- where λjg are the unstandardized loadings of the subtests on g, φg is the
suring the same specific factor (IN and VO for crystallized intelligence; variance of the g factor, and θjj′ are the elements of the error variance-
DM and PW for delayed memory; SiRT and SeRT for reaction time). The covariance matrix. Note that this is the correct formula for omega
latter model is depicted in Fig. 1. coefficient when subtest errors are correlated (i.e., θjj′ ≠ 0) and latent
Comparative Fit Index (CFI), Tucker-Lewis Index (TLI) and Root factors are not standardized (φg ≠ 1). Omega was defined as a derived
Mean Square Error of Approximation (RMSEA) were considered to as- parameter in the model and the 95% Wald confidence interval was
sess the goodness-of-fit of the models according to the conventional cut- computed with Mplus (Cheung, 2009).
off points (e.g., Liu, Rovine and Molenaar, 2012): CFI > 0.90, TLI >
0.90 and RMSEA < 0.08. 3. Results
For the best configural invariance model, the following parameters
were sequentially constrained to be equal: loadings, general factor 3.1. Descriptive statistics
variance, and specific variances and covariances. The sequentially
tested models were compared based on CFI values. The constraints Descriptive statistics are shown in Table SM1. 87% of students
imposed in the more restrictive model were considered tenable when completed all tests and tasks and 12% had missing data in one or two
CFI decrease was small (CFI decrease < 0.01; Cheung & Rensvold, tasks. Missing data were found mainly for the reaction time tasks.
2002). Akaike Information Criterion (AIC; Akaike, 1974) values were Pairwise correlations are shown in Table SM2. As expected, almost
also reported. all correlations (90%) were positive. For the successive age groups, the
Mplus Version 8 was employed for computing structural equation average correlation did not follow a meaningful pattern (0.19, 0.17,

224
S.R. Shahabi et al. Personality and Individual Differences 135 (2018) 222–228

Table 2
Subtests' correlation matrices for scaled scores by age group.
MR BD VO IN DS SS AR DM PW FT TS SiRT SeRT OR

6–7 years below diagonal\7–8 years above diagonal

MR 0.27 0.41 0.35 0.18 0.19 0.34 0.23 0.07 0.27 0.16 0.05 0.15 0.25
BD 0.50 0.36 0.40 0.25 0.19 0.38 0.30 0.17 0.07 0.21 0.03 0.07 0.06
VO 0.36 0.45 0.55 0.29 0.10 0.37 0.50 0.16 0.03 0.19 −0.01 0.09 0.11
IN 0.40 0.38 0.59 0.30 0.07 0.48 0.31 0.12 0.16 0.26 −0.02 0.05 0.29
DS 0.29 0.40 0.31 0.43 0.10 0.41 0.13 0.10 0.03 0.06 0.11 0.16 0.14
SS 0.23 0.29 0.12 0.03 −0.03 0.23 0.17 −0.03 −0.11 0.14 0.01 −0.03 0.10
AR 0.21 0.37 0.35 0.41 0.47 0.05 0.29 0.13 0.08 0.24 0.03 0.00 0.15
DM 0.37 0.30 0.46 0.40 0.22 −0.06 0.19 0.24 −0.01 0.05 0.03 0.00 0.28
PW 0.21 −0.01 0.20 0.32 0.21 0.01 0.24 0.36 0.06 0.05 0.10 −0.09 0.17
FT 0.24 0.13 0.03 0.10 0.19 −0.03 0.35 0.11 0.08 0.29 0.26 0.28 0.03
TS 0.24 0.33 0.22 0.34 0.40 −0.06 0.37 0.23 −0.06 0.23 0.26 0.09 0.14
SiRT 0.00 0.19 0.05 0.04 0.01 0.06 0.08 0.06 −0.26 0.16 0.16 0.26 0.24
SeRT 0.05 0.21 0.07 0.16 0.05 −0.03 0.09 0.00 −0.12 0.20 0.25 0.51 0.14
OR −0.03 0.22 0.27 0.40 0.24 0.02 0.30 0.14 0.15 0.19 0.04 0.11 0.13

8–9 years below diagonal\9–10 years above diagonal

MR 0.42 0.39 0.39 0.22 0.18 0.35 0.10 0.04 0.23 0.01 0.09 0.09 0.17
BD 0.57 0.41 0.43 0.25 0.26 0.47 0.16 0.07 0.16 0.16 0.00 0.09 0.06
VO 0.46 0.46 0.61 0.27 0.41 0.45 0.26 0.06 0.11 0.01 0.09 0.11 0.05
IN 0.42 0.41 0.68 0.23 0.28 0.53 0.28 0.16 0.25 0.13 0.07 0.09 0.02
DS 0.34 0.31 0.48 0.44 0.28 0.26 0.05 0.12 0.29 0.01 −0.01 0.21 −0.09
SS 0.34 0.38 0.45 0.22 0.29 0.18 0.09 0.11 0.04 −0.05 0.08 0.15 −0.04
AR 0.48 0.50 0.63 0.55 0.48 0.35 0.16 0.22 0.13 0.12 −0.09 −0.02 0.12
DM 0.30 0.33 0.44 0.39 0.40 0.26 0.36 0.31 0.05 −0.13 0.04 0.12 0.02
PW 0.15 0.11 0.32 0.36 0.25 0.13 0.34 0.46 0.15 −0.03 −0.04 0.06 0.10
FT 0.05 0.22 0.28 0.22 0.26 0.28 0.30 0.11 0.04 0.20 0.21 0.20 −0.03
TS 0.14 0.23 0.12 0.11 0.09 0.13 0.20 0.05 −0.02 0.23 0.14 0.22 −0.08
SiRT 0.09 0.13 0.11 0.19 0.09 0.05 0.23 0.05 0.09 0.22 0.26 0.54 0.01
SeRT 0.16 0.13 0.19 0.25 0.19 0.07 0.30 0.16 0.19 0.12 0.22 0.30 0.04
OR 0.24 0.19 0.21 0.19 0.37 0.20 0.29 0.30 0.15 0.19 0.12 0.20 0.26

10–11 years below diagonal\11–12 years above diagonal

MR 0.40 0.41 0.42 0.18 0.22 0.47 0.18 0.04 −0.09 0.22 0.17 0.07 0.23
BD 0.38 0.31 0.30 0.45 0.41 0.57 0.20 −0.02 0.03 0.24 0.03 0.07 0.20
VO 0.39 0.36 0.64 0.12 0.30 0.54 0.35 0.12 0.08 0.14 0.10 0.14 0.33
IN 0.28 0.41 0.56 0.20 0.24 0.44 0.23 0.06 0.14 0.23 −0.01 −0.03 0.23
DS 0.16 0.37 0.24 0.29 0.26 0.31 0.12 0.16 −0.04 0.07 −0.09 0.00 0.14
SS 0.28 0.42 0.27 0.29 0.21 0.36 0.20 −0.04 0.02 0.17 0.12 0.06 0.03
AR 0.38 0.33 0.33 0.45 0.34 0.27 0.37 0.25 0.07 0.05 0.17 0.06 0.22
DM 0.19 0.21 0.32 0.13 0.36 0.29 0.29 0.21 0.08 0.15 0.08 0.17 0.08
PW −0.04 0.13 0.22 0.10 0.20 0.21 0.16 0.25 0.02 −0.07 −0.06 0.06 0.17
FT 0.03 0.12 0.30 0.20 0.08 0.14 0.12 0.08 0.00 0.00 0.15 0.17 0.05
TS 0.18 0.16 0.39 0.30 0.21 0.09 0.10 0.33 0.11 0.17 −0.05 −0.05 0.15
SiRT 0.20 0.23 0.13 0.19 0.06 0.25 0.12 0.11 0.08 0.14 0.06 0.43 −0.12
SeRT 0.22 0.22 0.03 0.07 0.08 0.26 0.16 0.15 0.12 0.04 0.03 0.54 −0.02
OR 0.18 0.06 0.17 0.11 0.14 0.00 0.16 0.30 −0.11 0.02 0.09 0.03 0.18

Note. Scaled SiRT and SeRT scores were reversed for easiness of interpretation.

Table 3
Goodness-of-fit indices for the different models.
Model comparison

chi gl RMSEA CFI AIC Models compared chi gl p CFI decrease

1: Configural invariance: unidimensional model 796.8 462 0.081 0.803 25,096.3

2: Configural invariance: unidimensional model + correlated errors 592.5 444 0.055 0.913 24,928.0 1,2 204.3 18 < 0.001
3: Model 2 + equal loadings 673.8 509 0.054 0.903 24,879.3 2,3 81.3 65 0.083 0.010
4: Model 3 + equal general factor variance 684.5 514 0.055 0.900 24,880.0 3,4 10.7 5 0.057 0.003
5: Model 4 + equal specific variances and covariances 981.8 599 0.076 0.775 25,007.3 4,5 297.3 85 < 0.001 0.125

0.26, 0.15, 0.2 and 0.16). loadings across groups did not differ significantly; p = .083). For suc-
cessive age groups, the standard deviations for the g latent factor were
1, 0.95, 1.31, 1.11, 1.14, and 1.24. Next, imposing the g variance
3.2. Multi-group confirmatory factor analyses equality did not decrease fit significantly (p = .057) and the decrease in
CFI was very small (0.003), although it produced a model with a
Fit indexes for MG-CFA models are shown in Table SM3. The uni- slightly higher AIC. Pairwise contrasts for g variance differences were
dimensional model with correlated errors showed acceptable fit also non-significant after applying Bonferroni correction (α = 0.05).
(CFI = 0.913; RMSEA = 0.055) and better fit than the unidimensional Finally, equality of specific variances and covariances was un-
model. Thus, we chose the former as the baseline model for invariance ambiguously not tenable (imposing these constraints did decrease fit
analyses. Metric invariance was also tenable (imposing equality of g

225
S.R. Shahabi et al. Personality and Individual Differences 135 (2018) 222–228

Table 4
Subtests' g loadings estimated in each age group according to the metric invariance models.
Model 3: metric invariance Model 4: metric invariance/with equal g variance across age

Subtest 6–7 7–8 8–9 9–10 10–11 11–12 6–7 7–8 8–9 9–10 10–11 11–12
MR 0.59 0.53 0.60 0.51 0.57 0.63 0.64 0.59 0.54 0.52 0.56 0.59
BD 0.62 0.56 0.73 0.61 0.63 0.61 0.67 0.63 0.69 0.62 0.63 0.57
VO 0.74 0.70 0.72 0.55 0.60 0.61 0.77 0.75 0.65 0.55 0.58 0.57
IN 0.66 0.73 0.71 0.57 0.54 0.54 0.70 0.78 0.65 0.57 0.53 0.50
DS 0.49 0.47 0.59 0.48 0.47 0.45 0.54 0.53 0.53 0.49 0.46 0.42
SS 0.27 0.33 0.52 0.35 0.39 0.40 0.30 0.38 0.47 0.36 0.39 0.37
AR 0.65 0.58 0.76 0.72 0.65 0.77 0.69 0.65 0.71 0.73 0.65 0.74
DM 0.36 0.39 0.50 0.39 0.47 0.50 0.41 0.45 0.45 0.40 0.48 0.47
PW 0.20 0.22 0.25 0.27 0.25 0.25 0.23 0.26 0.22 0.27 0.25 0.23
FT 0.22 0.22 0.28 0.25 0.23 0.25 0.25 0.25 0.25 0.26 0.23 0.23
TS 0.25 0.22 0.31 0.27 0.34 0.30 0.29 0.27 0.27 0.28 0.34 0.28
SiRT 0.10 0.12 0.17 0.14 0.16 0.19 0.11 0.15 0.14 0.14 0.16 0.18
SeRT 0.15 0.15 0.19 0.21 0.18 0.25 0.17 0.18 0.16 0.22 0.18 0.23
OR 0.27 0.24 0.31 0.26 0.36 0.26 0.31 0.28 0.27 0.27 0.36 0.24
Average 0.40 0.39 0.47 0.40 0.42 0.43 0.43 0.44 0.43 0.41 0.41 0.40

significantly, p < .001, the decrease of CFI was large, 0.125, and the non-centrality parameter (δ) associated to a misspecified model, as-
AIC increased). Residual variances and covariances did not follow a suming equal variance for the general factor across ages, was obtained;
consistent pattern for all the subtests (see Table SM4). For instance, the and (d) the power (i.e., the probability of rejecting the false null model)
variance of SiRT and SeRT decreased with age, consistently with the was obtained from the non-central Chi-squared distribution (with
mutualism hypothesis, but the specific variance of VO, IN and DS in- noncentrality parameter equal to δ, nominal level of significance
creased. α = 0.05, and degrees of freedom equal to five, the number of variances
The standardized loadings for the metric invariance model (with constrained).
and without equality constraint in the g variance across age) are shown The power rates for each of the five conditions tested were 0.4, 0.59,
in Table SM5. The loadings of the 14 subtests were positive and ranged 0.78, 0.9 and 0.96, respectively. Thus, the power rate for a condition in
from 0.10 to 0.78. When the g variance equality was not assumed, there which omega coefficient ranged from 0.77 to 0.843 (range of 0.073)
was not a systematic trend in the average g loading across age. On the was reasonably high (i.e., 0.78).
other hand, when the equality of g variance was assumed, standardized Similar power rates were obtained when true models were gener-
g loadings slightly decreased with increased age. Therefore, according ated taking model 3 as reference.
to the last model, specific variance increased across age. However, this
average trend was unsystematic across tests/tasks and it was mainly
due to three subtests (Vocabulary, Information, and Digit Span). 3.4. Omega coefficients (ω)
Indeed, the opposite trend was found for the time reaction tasks (SiRT
and SeRT). Whatever the case, none of the two models were consistent The best fitting model according to the AIC index (model 3) was
with the mutualism prediction. selected for computing the Omega coefficient (ω). As it can be seen in
Fig. 2, the Omega values across the six age bands varied from 0.72 to
0.82, but there was not any regular trend. All the computed omega
3.3. Power to detect unequal general factor variance across groups differences among age groups were non-significant (α = 0.05), when
corrected for multiple comparisons (Bonferroni). It must be noted that
Since groups' sample sizes were relatively small, the power to reject the power to detect minor pairwise differences with this model was
the equality of general factor variance across groups was analyzed. generally small. For instance, the average standard error for omega
Power calculations were carried out by means of the Satorra and Saris's coefficient differences was 0.047 and, therefore, assuming a normal
(1985) method based on the nonnull distribution of the (normal theory) distribution of the bilateral test contrasts and the Bonferroni correction,
log-likelihood ratio test (Hessen, Dolan, and Wicherts, 2006). These minimum differences of 0.10, 0.12, 0.14, 0.16 or 0.18 were required to
were the steps: (a) taking model 4 as reference, several true models achieve power rates of 0.21, 0.35, 0.51, 0.68, or 0.81, respectively.
were generated modifying the range of the g variance across age (i.e., When Omega was computed assuming equal g latent variance across
from 1 to 1.2; from 1 to 1.4, from 1 to 1.6; from 1 to 1.8; from 1 to 2). groups, omega coefficients slightly decreased as age increased (ω for
The tested situations corresponds to different rates of increase in omega successive age groups were 0.77, 0.79, 0.77, 0.75, 0.76, and 0.74). In
coefficients across age (from 0.77 to 0.801; from 0.77 to 0.824; from this case, reliability was significantly higher for the 7–8 years old boys
0.77 to 0.843; from 0.77 to 0.858; from 0.77 to 0.87); (b) reproduced (with respect to the 9–10 and the 11–12 age groups). For this model,
means and covariance population matrices for the groups were ob- reliability estimates were more stable and the average standard error
tained for each of these true models; (c) for each population case, the for omega coefficient differences was 0.012, which implied a pretty

Fig. 1. Unidimensional model with correlated residuals.

226
S.R. Shahabi et al. Personality and Individual Differences 135 (2018) 222–228

variance accounted for by g or the main cognitive abilities, nor the number of
ability factors, will change along the life span” (Juan-Espinosa et al., 2002,
p. 406).
These researchers suggest replacing the investing metaphor (Cattell,
1987) by the anatomical metaphor: as with the human skeleton, in-
telligence shows one basic structure that is present early in life. This
structure remains fairly the same across the life span. However, like the
skeleton bones, cognitive abilities identified in the basic structure can
go up and down at different developmental stages.
Nevertheless, results reported here were based on cross-sectional
data. As properly noted by Rogier Kievit on his review of the present
report, “mutualism is ultimately a theory of within subject development over
time”. Kievit et al. (2017) and Hofman et al. (2018) found support for
the mutualism model studying longitudinal samples. Therefore, our
results, along with those reported by Gignac (2014), must be circum-
scribed to the cross-sectional nature of the data.
In closing, the key finding observed in the present cross-sectional
Fig. 2. Omega values across the six age bands. study is consistent with the g model. The critical development period
considered here (6–12 years) fails to reveal any increased trend in the
amount of variance explained by the obtained latent general factor. We
high-power rate (i.e., 0.89) for an omega coefficient difference of 0.05.
addressed the challenge raised by Gignac (2014) regarding the re-
quirement to replicate his findings increasing the scope of the con-
4. Discussion sidered abilities and the results were consistent with the g model. The
mutualism model is rather unspecific regarding the crucial age ranges
Here we have shown that the latent factor obtained from a wide set for detecting the emergence of g as a consequence of the presumed
of cognitive tests and tasks completed by 661 children can be mean- reciprocal beneficial interactions among independent cognitive pro-
ingfully interpreted across the age range considered (6–12 years). cesses. Perhaps these interactions take place before the first age band
Multi-group Confirmatory Factor Analyses results were consistent with considered here, but it will be difficult to tell because the measurement
this conclusion. The general factor (g) obtained at different ages was the of intelligence before this band might be relatively useless for obtaining
same g. reliable data. However, we acknowledge that the mutualism approach
From here, we moved one step forward computing the amount of cannot be ruled out because longitudinal data may be more appropriate
variance explained by this general factor across age using Omega for analyzing these competing models.
coefficients (ω) for isolating the variance explained by the general
factor, as recommended by Gignac (2014). We chose the Omega coef- Appendix A. Supplementary data
ficient, instead of the hierarchical omega coefficient, because we did
not have enough indicators per specific factor to compute a bi-factor Supplementary data to this article can be found online at https://
analysis. Furthermore, we separated the variance explained by specific [Link]/10.1016/[Link].2018.07.021.
factors by adding some correlated errors to the model.
The results were inconsistent with the mutualism model: the coef- References
ficients failed to show any reliably increasing trend. These coefficients
were fairly constant, and, therefore, the g model was not rejected. When Abad, F. J., Sorrel, M. A., Román, F. J., & Colom, R. (2015). The relationship between
the g variance was constrained to be equal a meager differentiation WAIS-IV factor index scores and educational level: A Bifactor model approach.
Psychological Assessment. [Link]
effect was found, but the trend was unsystematic across tests. This Abad, J., Colom, R., Juan-Espinosa, M., & García, L. F. (2003). Intelligence differentiation
differentiation effect was present for the crystallized tests, which is in adult samples. Intelligence, 31, 157–166.
congruent with results reported by Abad et al. (2003). These re- Abedi, M., Sadeghi, A., & Rabiei, M. (2015). Standardization of WISC-IV for children.
Journal of Psychological Achievement, 22(2), 99–116. [Link]
searchers suggested that the differentiation effect might be a by-product 2016.12310 (in Persian).
of educational differences. Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on
Interestingly, Gignac (2014) considered the full life span Automatic Control, 19, 716–723.
Carroll, J. B. (1993). Human cognitive abilities: A survey of factor-analytic studies.
(0–90 years) finding very small effects consistent with the mutualism
Cambridge, UK: Cambridge University Press.
model from 2.5 to 10 years of age only. Nothing emerged beyond this Cattell, R. B. (1987). Intelligence: Its Structure, Growth and Action. New York: North-
point. We considered ages ranging from 6 and 12 years, and, therefore, Holland.
Cheung, G. W., & Rensvold, R. B. (2002). Evaluating goodness of fit indexes for testing
the critical period identified by Gignac was the main focus in the pre-
measurement invariance. Structural Equation Modeling, 9(2), 233–255.
sent study. Nevertheless, we failed to find any substantial change in the Cheung, M. W. L. (2009). Constructing approximate confidence intervals for parameters
amount of variance explained by the obtained latent factor quantifying with structural equation models. Structural Equation Modeling, 16(2), 267–294.
g (Fig. 2). [Link]
De Alwis, D. (2011). Development of speed, memory and fluid reasoning in children. All
In short, our results are consistent with the g model. We underscore theses and dissertations (ETDs). Paper 567.
that the administered battery of tests and tasks sampled a broad array of Gignac, G. E. (2014). Dynamic mutualism versus g factor theory: An empirical test.
cognitive abilities. The measures were chosen under the CHC frame- Intelligence, 42, 89–97.
Hessen, D. J., Dolan, C. V., & Wicherts, J. M. (2006). The multigroup common factor
work (McGrew, 2009) and they were not circumscribed to the most model with minimal uniqueness constraints and the power to detect uniform bias.
popular batteries, such as the WISC, that tap 3 or four first order abil- Applied Psychological Measurement, 30(3), 233–246. [Link]
ities at best (Gignac, 2014). 0146621605279760.
Hofman, A., Kievit, R., Stevenson, C., Molenaar, D., Visser, I., & van der Maas, H. (2018).
Juan-Espinosa et al. (2002, 2006) investigated one closely related The dynamics of the development of mathematics skills: A comparison of theories of de-
issue for answering the next question: are cognitive abilities progres- veloping intelligence.
sively differentiated with increased age? Their findings supported the Jensen, A. R. (1998). The g factor: The science of mental ability. East (Vol. 17). Westport, CT:
Praeger.
age-indifferentiation hypothesis concluding that there is one strong Johnson, W., Bouchard, T., ... Gottesman, I. I. (2004). Just one g: Consistent results from
stability of the structure of intelligence across the life span: “neither the

227
S.R. Shahabi et al. Personality and Individual Differences 135 (2018) 222–228

three batteries. Intelligence, 32, 95–107. Muthén, L. K., & Muthén, B. O. (2017). Mplus user's guide (version 8). Los Angeles, CA:
Johnson, W., Nijenhuis, J., & Bouchard, T. (2008). Still just 1 g: Consistent results from Muthén & Muthén.
five test batteries. Intelligence, 36, 81–95. Raykov, T. (2004). Behavioral scale reliability and measurement invariance evaluation
Juan-Espinosa, M., Cuevas, L., Escorial, S., & Garcia, L. F. (2006). The differentiation using latent variable modeling. Behavior Therapy, 35(2), 299–331. [Link]
hypothesis and the Flynn effect. Psicothema, 18(2), 284–287. 10.1016/S0005-7894(04)80041-8.
Juan-Espinosa, M., Garcıa, L. F., Escorial, S., Rebollo, I., Colom, R., & Abad, F. J. (2002). Satorra, A., & Saris, W. E. (1985). Power of the likelihood ratio test in covariance
Age dedifferentiation hypothesis: Evidence from the WAIS III. Intelligence, 30(5), structure analysis. Psychometrika, 50, 83–90.
395–408. Schneider, W. J., & Newman, D. A. (2015). Intelligence is multidimensional: Theoretical
Kayser, J. (2015). Tone screening test. Retrieved June 19, 2015 from http:// review and implications of specific cognitive abilities. Human Resource Management
[Link]/software/TST/[Link]. Review, 25, 12–27.
Kievit, R. A., Lindenberger, U., Goodyer, I. M., Jones, P. B., Fonagy, P., Bullmore, E. T., ... Tamm, L., & Juranek, J. (2012). Fluid reasoning deficits in children with ADHD: Evidence
Dolan, R. J. (2017). Mutualistic coupling between vocabulary and reasoning supports from FMRI. Brain Research, 17(1465), 48–56.
cognitive development during late adolescence and early adulthood. Psychological Van Der Maas, H. L., Dolan, C. V., Grasman, R. P., Wicherts, J. M., Huizenga, H. M., &
Science, 28(10), 1419–1431. Raijmakers, M. E. (2006). A dynamical model of general intelligence: The positive
Kievit, R. A., Romeijn, J. W., Waldorp, L. J., Wicherts, J. M., Scholte, H. S., & Borsboom, manifold of intelligence by mutualism. Psychological Review, 113(4), 842–861.
D. (2011). Mind the gap: A psychometric approach to the reduction problem. van der Maas, H. L. J., & Kan, K. J. (2016). Comment on "Residual group-level factor
Psychological Inquiry, 22, 1–21. association: Possibly negative implications for the mutualism theory of general in-
Kovacs, K., & Conway, A. R. A. (2016). Process overlap theory: A unified account of the telligence" by Gilles E. Gignac. Intelligence, 57, 81–83.
general factor of intelligence. Psychological Inquiry, 27(3), 151–177. van der Maas, H. L. J., Kan, K. J., Marsman, M., & Stevenson, C. E. (2017). Network
Liu, S., Rovine, M. J., & Molenaar, P. (2012). Selecting a linear mixed model for long- Models for Cognitive Development and Intelligence. Journal of Intelligence, 5(2), 1–14.
itudinal data: Repeated measures analysis of variance, covariance pattern model, and [Link]
growth curve approaches. Psychological Methods, 17, 15–30. Wechsler, D. (2002). Wechsler Preschool and Primary Scale of Intelligence Third edition
Markus, K., & Borsboom, D. (2013). Frontiers of validity theory: Measurement, causation, technical and interpretative manual. San Antonio, TX: Psychological Corporation.
and meaning. New York, NY, USA: Routledge. Wechsler, D. (2003a). Wechsler Intelligence Scale for Children Fourth Edition. San Antonio,
McGrew, K. S. (2009). CHC theory and the human cognitive abilities project: standing on the TX: Harcourt Assessment, Inc.
shoulders of the giants of psychometric intelligence research. 37, 1–10. Wechsler, D. (2003b). Wechsler Intelligence Scale for Children Fourth edition technical and
Mueller, S. T., & Piper, B. J. (2014). The psychology experiment building language (PEBL) interpretive manual. San Antonio, TX: Psychological Corporation.
and PEBL test battery. Journal of Neuroscience Methods, 222, 250–259. [Link] Wechsler, D. (2008). Wechsler Adult Intelligence Scale Fourth edition: Technical and inter-
org/10.1016/[Link].2013.10.024. pretive manual. San Antonio, TX: Pearson Assessment.

228