|
Regardless
whether g can be proven to represent a specific essence of the human mind, those
working in the field of applied intelligence testing need to be familiar with recent research that
suggests that certain cognitive processes may lie behind the general factor. The integration
of a
centuries of psychometric research with contemporary information processing (IP) theories has
resulted in important strides in understanding human intelligence (Kyllonen, 1996). Although
slightly different IP models have been hypothesized and researched, in general, the four-source
consensus model (Kyllonen, 1996) will suffice for this chapter. According to Kyllonen (1996),
the
four primary components or sources of IP models are procedural and declarative knowledge,
processing speed (Gs), and working memory (MW).
[Note.
Another typical description of information processing models makes a distinction between: (1)
memory systems—short- term and long-term memory, (2) types of knowledge-declarative and procedural,
and (3) types of processing-controlled and automatic (Lohman, 2000).
One
of the most intriguing findings from the marriage of psychometric and IP models, first
reported by Kyllonen and Christal (1990), is that “individual differences in working memory
capacity may be what are responsible for individual differences in general ability” (Kyllonen,
1996, p. 61). This hypothesis was proposed by Kyllonen (Kyllonen, 1996; Kyllonen & Christal,
1990) based on very high latent factor correlations (.80 to mid .90’s) between measures of MW
and Gf in a variety of adult samples. Attempts to understand the relation between MW and
higher-order cognition “have occupied researchers for the past 20 years (Kane, Bleckley,
Conway & Engle, 2001, p. 169). Since 1990, the concept of MW has played a central role in
research attempting to explain individual differences in higher level cognitive abilities such as
language comprehension (Gc; Engle, Cantor, & Carullo, 1992; Just & Carpenter, 1992),
reading
and mathematics (Grw and Gq; Hitch, Towse, & Hutton, 2001; Leather & Henry, 1994),
reasoning or general intelligence (Gf and g; Ackerman. Beier, & Boyle, 2002; Conway,
Cowan,
Bunting, Themault, & Minkoff, 2002; Engle, Tuholski, Laughlin, & Conway, 1999; Fry & Hale,
1996, 2000; Kyllonen & Christal, 1990; Süß, Oberauer, Wittmann, Wilhelm, & Schulze,
2002), and
long-term memory performance (Park, Smith, Lautenshlager & Earles, 1996; Süß, Oberauer,
Wittman, Wilhelm & Schulze, 2002).
The
theoretical explanations for the consistently strong MWàcriterion relations differ primarily in
terms of the different cognitive resources proposed to underlie MW performance (Lohman,
2000). More specifically, multiple- resource and resource-sharing models have been
proposed
(Bayliss, Jarrold, Gunn, & Baddeley, 2003). A sample of resources hypothesized to influence
MW performance are storage capacity, processing efficiency, the central executive, domain-
specific processes, and controlled attention (Bayliss et al., 2003; Engle et al., 1999; Kane et al.,
2000). Researchers have hypothesized that the reason why MW is strongly associated with
complex cognition constructs (e.g., Gf) is that considerable information must be actively
maintained in MW, especially when some active transformation of information is required. Even
if the transformation “process” is effective, it must be performed within the limits of the
working
memory system. Therefore, although many different processes may be executed in the solution
of a task, individual differences in the processes may primarily reflect individual differences, not
working memory resources (Lohman, 2000, p. 325). A detailed treatment of the different
working memory theoretical explanations is beyond the scope of the current paper and is not
necessary in the current context. Figure 3 presents schematic
summaries of four of the primary
structural equation modeling (SEM) investigations (published during the past decade) that shed
additional insights on the causal relations between MW and g or Gf.
[Note.
or readability purposes, the manifest variables and certain other latent factors (age factors) were
removed from all figures. In addition, based on a reading of the description of the variables
used in each
study, the original latent factor names were changed by the current author in accordance with CHC theory
as described in this chapter. These interpretations do not necessarily reflect the interpretations
of the
authors of the original published studies.]
In
the causal models portrayed in Figure 3, MW demonstrates a significant effect on all
dependent variables (primarily Gf or g). [Note. Hambrick and Engle (2002) and Park, Smith,
Lautenshlager & Earles (1996) have reported similar causal models with memory performance as the
dependent latent variable. In these studies, the MW direct causal paths were .30 and .44.
In the Hambrick
and Engle (2002) study, MW also had an indirect effect (.31) on memory performance that was mediated
through a domain-specific knowledge (Gk) factor.] With the exception of
the Süß et al. (2002) models
(Figures 3d/e), the strength of the MWàGf/g (.38 to
.60) relations are lower than those reported
by Kyllonen and Christal (1990). The weakest MWàGf relationship
(.38) was in the only child
and adolescent sample (Figure 3a). This finding may suggest a weaker relationship between the
construct of MW and complex cognitive reasoning during childhood. In contrast, when the two
different MW components (MW1 and MW2) are considered together in the two alternative Süß
et al. (2002) models, MW collectively exerts a strong influence on g (MW1 = .65; MW2 = .40;
Figure 3d) and Gf (MW1 = .70; MW2 = .24; Figure 3e).
It
is important to note that in most studies that have explored the relation between MW and
psychometric constructs, Gs is typically included as a direct precursor to MW (see
Figures
3a/b/c).
Collectively, the MWàcriterion studies suggest
MW may be a significant causal factor
working behind the scenes when complex cognitive performance is required (e.g., Gf or g).
Missing from this literature are studies that include a broader and more complete array of CHC
indicators and factors in larger and more carefully selected samples. This limitation is addressed
below. |
|
For
the purposes of this chapter, select tests from the CHC-designed WJ III COG battery were
used to investigate the relations between measures of information processing efficiency (viz., Gs,
MS, and MW) and complex cognitive ability (operationalized in the form of g). In the causal
model, g was operationally defined as a second- order latent factor comprised of five well
identified latent CHC factors (Gf, Gc, Glr, Ga, and Gv; McGrew & Woodcock,
2001). [Note.WJ
III test indicators for the latent factors were selected based on: (1) the principle of providing at
least two
qualitatively different narrow ability indicators for each broad CHC factor, (2) using tests that were
not
factorially complex as determined from prior CFA studies (McGrew & Woodcock, 2001), and (3) using
tests
that were some of the best WJ III CHC factor indicators (McGrew & Woodcock, 2001). ] Consistent with
the extant literature, Gs was specified to be a direct precursor to MW, although all models also
tested for significant direct paths from Gs to g. In addition, given that MW subsumes
the rote
storage role of MS, a separate MS factor with a direct effect on MW was specified. The
inclusion of both MS and MW latent factors is consistent with the research models of Engle et al.
(1999). The final model is represented in Figure 4.
For
each of five age-differentiated nationally representative samples (each which ranged in size
from approximately 1,000 to 2,200 subjects; see McGrew & Woodcock, 2001), the same initial
model was specified. In addition to the direct MWàg path, a direct path from Gsàg was also
tested in each sample (see Figure 4). [Note. Given that the primary purpose of these analyses was to
explore the relations between basic information processing constructs (Gs and Gsm) and g,
no effort was
made to “tweak” the measurement models in each sample in search of slightly better fitting
models. The
same configurially invariant measurement model was used across all five samples.] The results
summarized in Figure 4 and Table 4 are important to note
as they allow for the investigation of
the MWàg relationship
in large nationally representative samples. In addition, the latent factor
constructs defined in these analyses are represented by the same indicators across all samples, a
condition rarely achieved across independent research studies (e.g., Figure 3). This later
condition provides for configural invariance of the models across samples. The parameters
presented in Figure 4 are for the 14 to 19 year old sample. Table 4 presents the key parameters
and model fit statistics for all samples.
The
results presented in Figure 4 and Table 4 are consistent with
the previously summarized
MWàg research literature.
Across all five samples, the MWà g direct effect
path ranged
from .73 to .93. Clearly, working memory (MW) potentially exerts a large causal effect on
complex cognitive performance (i.e., g) wheb defined by the combined performance on five
latent CHC factors (i.e., Gf, Gc, Glr, Ga, Gv). The trend for the MWàg path to decrease with
increasing age (.93, .90, .82, .83, .73) may be of significant substantive interest to developmental
psychologists and intelligence researchers studying the effect of aging within the CHC
framework (e.g., see Horn & Masunaga, 2000; Park et al., 1996; Salthouse , 1996). Also of
interest is the finding, consistent with prior research (Fry & Hale, 1996, 2000), that Gs
did not
demonstrate a direct effect on g in the childhood samples. However, starting at late adolescence
(ages 14-19), Gs begins to demonstrate small, yet significant, direct effects on g (.07
and .09
from ages 14-39), and a much more substantial effect at middle age and beyond (.22). These
developmental trends suggest the hypothesis that during an individuals formative years (ages 6-
13) MW exerts a singular and large (.90 to .93) direct effect on complex cognitive task
performance (i.e., g). Around adolescence, MW appears to decrease slightly in direct influence
on g, while Gs concurrently increases in importance, particularly during the later half
of most
individuals lives (40 years and above).
It
is important to note that in all models, Gs exerts indirect effects on g via two
routes (i.e.,
GsàMSàMWàg; GsàMWàg). Using standard
path model tracing rules, the total effects
(direct + indirect) of Gs on g have been calculated and are summaried in Table
4.
The range of
total Gsàg effects is
large (.60 to .81). Clearly, these analyses suggest that Gs and MW both
exert large and significant influence on complex cognitive performance (i.e., g). Collectively,
the
total effects of Gs+MW (information processing efficiency) account for 76 % to 86 % of the
CHC defined g-factor. |
|
The
WJ III CHC MWàg analyses and
research studies presented here continue to suggest an
intriguing relation between measures of cognitive efficiency (Gs and MW) and complex cognitive
performance (viz., Gf and g). As articulated by Kyllonen (1996),
“the remarkable finding is the consistency with which
the working memory capacity
factor has proven to be the central factor in cognition abilities…that working memory
capacity is more highly related to performance on other cognitive tests, and is more
highly related to learning, both short-term and long-term, than is any other cognitive
factor” (p. 72-73).
Leaping
from these findings to the conclusion that MW is the basis of Spearman’s g (Süß
et al.,
2002) or Gf (Kyllonen & Christal, 1990) is not the intent of this section of this chapter.
Alternative claims for the basis of g (e.g., processing/reaction time) exist (see Nyborg, 2003).
The important conclusion here is that appropriately designed CHC MWàoutcome studies can
make important contributions to research focused on increasing our understanding of the nature
and importance of working memory, as well as the specific cognitive resources that contribute to
a variety of cognitive and academic performances. Accoreding to Süß et al. (2002):
The strong relationship between working memory and intelligence
paves the way for a
better understanding of psychometric ability concepts through theories of cognition.
Establishing this general association, however, is only the first step. Working memory
itself is not a precisely defind construct. It is widely accepted that working- memory
capacity is an important limited resource for complex cognition; however, which
functions of working memory affect which part of the cognitive process in a given
reasoning task is not well understood…Now that the relationship between working
memory and intelligence has been established on a molar level, further research with
more fine-grained analyses need to be done (p. 285- 286).
|
|
|
|
|
|
1. 1. Introduction
