|
“The g
factor (and highly g-loaded test scores, such as the IQ) shows a more far-reaching and
universal practical validity than any other coherent psychological construct yet discovered”
(Jensen, 1998, p. 270). The strength of g’s prediction, together with past attempts
to move
“beyond g” (i.e., the addition of specific abilities to g in the prediction
and explanation of
educational and occupational outcomes), historically have not met with consistent success. In his
APA presidential address, McNemar (1964) concluded “the worth of the multi-test batteries as
differential predictors of achievement in school has not been demonstrated” (p. 875).
Cronbach
and Snow’s (1977) review of the aptitude-treatment interaction (ATI) research similarly
demonstrated that beyond general level of intelligence (g), few, if any, meaningful specific
ability-
treatment interactions existed. Jensen also reinforced the preeminent status of g when
he stated
that “g accounts for all of the significantly predicted variance; other testable ability
factors,
independently of g, add practically nothing to the predictive validity” (Jensen, 1984, p.
101).
In
applied assessment settings, attempts to establish the importance of specific abilities above and
beyond the full scale IQ (research largely based on the Wechsler batteries) score have typically
meet with failure. As a result, assessment practitioners have been admonished to “just say
no” to
the practice of interpreting subtest scores in individual intelligence batteries (McDermott,
Fantuzzo, & Glutting, 1990; McDermott & Glutting, 1997). The inability to move beyond g has
provided little optimism for venturing beyond an individual’s full scale IQ score in the applied
practice of intelligence test interpretation. However, Daniel, (2000) believes these critics have
probably “overstated” their case given some of the techniques they have used in their research.
Despite
the “hail to the g” mantra, a number of giants in the field of intelligence continue
to
question the “conventional wisdom” of complete deference to g (Carroll, 1996).
Carroll (1993)
concluded that “there is no reason to cease efforts to search for special abilities that may be
relevant for predicting learning” (p. 676). In a subsequent publication, Carroll (1996) stated
that
“I believe that the conventional wisdom is to some extent incorrect, however, because there are
many types of learning or performance that can be shown to depend not only on the general
factor but also on lower-stratum factors…I would point out that although Spearman attached
great importance to the general factor, he regarded some lower-stratum factors as being of
educational and occupational significance” (p. 8). Snow (1998) struck a similar chord when
he
stated that
certainly it is often the case that many ability-learning
correlations can be accounted for
by an underlying general ability factor. Yet, there are clearly situations, such as spatial-
mechanical, auditory, or language learning conditions, in which special abilities play a role
aside from G (p. 99).
In
the school psychology literature, Flanagan (1999), McGrew, Flanagan, Keith and Vanderwood
(1997) and Keith (1999) have suggested that advances in theories of intelligence (viz., CHC
theory), the development of CHC theory- driven intelligence batteries (viz., WJ-R, WJ III) , and
the use of more contemporary research methods (e.g., structural equation modeling, SEM) argue
for continued efforts to investigate the effects of g and specific abilities on general and specific
achievements. A brief summary of CHC-based g+specificàachievement abilities research
follows. |
|
Using
a Gf-Gc framework, Gustafsson and Balke (1993) reported that some specific cognitive
abilities may be important in explaining school performance beyond the influence of g when: (1)
a
Gf-Gc intelligence framework is used, (2) cognitive predictor and academic criterion measures
are both operationalized in multidimensional hierarchical frameworks, and (3)
cognitiveàachievement relations
are investigated with research methods (viz., SEM) particularly
suited to understanding and explaining (versus simply predicting). The key advantage
of the
SEM method is that it allows for the simultaneous inclusion of casual paths (effects) from a latent
g factor, plus specific paths for latent factors subsumed by the g factor, to a common
dependent
variable factor (e.g., reading). This is not possible when using multiple regression methods.
Drawing
on the research approach outlined by Gustafsson and Balke (1993), a series of CHC
designed studies completed during the past decade have identified significant CHC narrow or
broad effects on academic achievement, above and beyond the effect of g. Using the Cattell-
Horn Gf-Gc based WJ-R norm data, McGrew, Flanagan, Keith, and Vanderwood (1997) and
Vanderwood, McGrew, Flanagan, and Keith (2002) found, depending on the age level (five grade-
differentiated samples from grades 1-12), that the CHC abilities of Ga, Gc, and Gs had
significant cross-validated effects on reading achievement above and beyond the large effect of
g. In the 1st-2nd grade cross- validation sample (n = 232),
McGrew et al. (1997) reported a
strong direct effect of g on reading which was accompanied by significant specific effects for Ga
(.49) on word attack skills and Gc (.47) on reading comprehension. In math, specific effects
beyond the high direct g effect were reported at moderate levels (generally .20 to .30 range)
for
Gs and Gf, while Gc demonstrated high specific effects (generally .31 to .50 range).
Using the
same WJ-R norm data, Keith (1999) employed the same g+specificàachievement SEM methods
in an investigation of general (g) and specific effects on reading and math as a function of
ethnic
group status. Keith’s (1999) findings largely replicated those of McGrew et al. (1997) and
suggested that CHC g+specificàachievement relations
are largely invariant across ethnic group
status.
In
a sample of 166 elementary aged students, Flanagan (2000) applied the same methodology
used by McGrew et al. (1997), Keith et al. (1999) and Vanderwood et al. (2002) to a WISC-
R+WJ-R “cross-battery” dataset. A strong .71 direct effect for g on reading was
found, together
with significant specific effects for Ga (.28) on word attack and Gs (.15) and Gc
(.42) on reading
comprehension. More recently, McGrew (2000) reported the results of similar modeling studies
with the CHC-based WJ III. In three age- differentiated samples (ages 6-8, 9-13, 14-19), in
addition to the ubiquitous large effect for g on reading decoding (.81 to .85), significant specific
effects were reported for Gs (.10 to .35) and Ga (.42 to .47). |
|
Collectively,
the CHC-based g+specificàachievement SEM studies
reported during the last
decade suggest that even when g (if it does exist) is included in causal modeling studies, certain
specific lower-stratum CHC abilities display significant causal effects on reading and math
achievement. Critics could argue that the trivial increases in model fit and the amount of
additional achievement variance explained (vis-à-vis the introduction of specific lower-order CHC
paths) is not statistically significant (which is the case), and thus, Occam’s Razor would argue
for
the simpler models that only include g. Alternatively, knee-jerk acceptance of Occam’s
Razor
can inhibit scientifically meaningful discoveries. As best stated by Stankov, Boyle and Cattell
(1995) in the context of research on human intelligence, “while we acknowledge the principle of
parsimony and endorse it whenever applicable, the evidence points to relative complexity rather
than simplicity. Insistence on parsimony at all costs can lead to bad science” (1995, p.
16).
In
sum, even when a Carroll g-model of the structure of human cognitive abilities is adopted,
research indicates that a number of lower-stratum CHC abilities make important contributions to
understanding academic achievement, above and beyond g. Reschly (1997) reached the same
conclusion when he stated, in response to the first McGrew et al. (1997) paper, that “the
arguments were fairly convincing regarding the need to reconsider the specific versus general
abilities conclusions. Clearly, some specific abilities appear to have potential for improving
individual diagnoses. Note, however, that it is potential that has been demonstrated...”
(p. 238).
[Note.
The g+specificà achievement studies could
be considered to represent the Carroll position on how
cognitive abilities predict/explain academic achievement. The Horn position could similarly be
operational
defined in research studies that use either SEM or multiple regression of the lower-order CHC variables
on
achievement (no g included in the models). The results of such Horn CHCàachievement models,
completed in either the WJ-R or WJ III norm data, can be found in McGrew (1993), McGrew and Hessler
(1995), McGrew and Knopik (1993), Evans, Floyd. McGrew and Leforgee (2002), Floyd, Evans and McGrew
(2003). With the exception of Gv, all broad CHC abilities (Gf, Gc, Ga, Glr, Gsm, Gs)
are reported to be
significantly associated (at different levels that often vary within each ability domain by age) with
reading,
math, and writing achievement in the Horn CHCàachievement
model. ] |
|
|
|
|
|
1. 1. Introduction
