ArticlePDF Available

Prefrontal Cognitive Ability, Intelligence, Big Five Personality, and the Prediction of Advanced Academic and Workplace Performance


Abstract and Figures

Studies 1 and 2 assessed performance on a battery of dorsolateral prefrontal cognitive ability (D-PFCA) tests, personality, psychometric intelligence, and academic performance (AP) in 2 undergraduate samples. In Studies 1 and 2, AP was correlated with D-PFCA (r=.37, p<.01, and r=.33, p<.01, respectively), IQ (r=.24, p<.05, and r=.38, p<.01, respectively), and Conscientiousness (r=.26, p<.05, and r=.37, p<.01, respectively). D-PFCA remained significant in regression analyses controlling for intelligence (or g) and personality. Studies 3 and 4 assessed D-PFCA, personality, and workplace performance among (a) managerial-administrative workers and (b) factory floor workers at a manufacturing company. Prefrontal cognitive ability correlated with supervisor ratings of manager performance at values of r ranging from .42 to .57 (ps<.001), depending on experience, and with factory floor performance at pr=.21 (p=.02), after controlling for experience, age, and education. Conscientiousness correlated with factory floor performance at r=.23.
Content may be subject to copyright.
Prefrontal Cognitive Ability, Intelligence, Big Five Personality, and the
Prediction of Advanced Academic and Workplace Performance
Daniel M. Higgins
Harvard University
Jordan B. Peterson
University of Toronto
Robert O. Pihl
McGill University
Alice G. M. Lee
University of Hawai’i at Manoa
Studies 1 and 2 assessed performance on a battery of dorsolateral prefrontal cognitive ability (D-PFCA)
tests, personality, psychometric intelligence, and academic performance (AP) in 2 undergraduate sam-
ples. In Studies 1 and 2, AP was correlated with D-PFCA (r.37, p.01, and r.33, p.01,
respectively), IQ (r.24, p.05, and r.38, p.01, respectively), and Conscientiousness (r.26,
p.05, and r.37, p.01, respectively). D-PFCA remained significant in regression analyses
controlling for intelligence (or g) and personality. Studies 3 and 4 assessed D-PFCA, personality, and
workplace performance among (a) managerial–administrative workers and (b) factory floor workers at a
manufacturing company. Prefrontal cognitive ability correlated with supervisor ratings of manager
performance at values of rranging from .42 to .57 ( ps.001), depending on experience, and with
factory floor performance at pr .21 ( p.02), after controlling for experience, age, and education.
Conscientiousness correlated with factory floor performance at r.23.
Keywords: dorsolateral prefrontal cortex, academic performance, job performance, personality, intelli-
Intelligence has been most formally described with the “ortho-
dox model” (Eysenck, 1998, p. 108), derived from factor analysis,
and conceptualized as a hierarchical structure. At the base of the
structure are individual tests of cognitive ability, all of which are
positively intercorrelated. These may be arranged at a second
stratum into group factors such as fluid intelligence, crystallized
intelligence, and visual perception (Carroll, 1993, pp. 624 625).
These group factors share a large single domain of variance, g,
uppermost in the hierarchy (Brand, 1996; Brody, 1992; Carroll,
1993; Jensen, 1998). The general factor gaccounts for 50%– 80%
of the variance of the group factors (Deary, 2001) and is associated
with a variety of real-world outcomes (Gottfredson, 1997; Herrn-
stein & Murray, 1994; Schmidt & Hunter, 1998). The orthodox
model has generated a number of key empirical findings (Neisser
et al., 1996) and provides the bulk of nonfolk understanding of the
consequences and biological– brain correlates of intelligence
(Jensen, 1998). Intelligence is importantly related to everyday
competence (Gottfredson, 1997), academic performance (Brody,
1992; Neisser et al., 1996), job performance (Schmidt & Hunter,
1998), and various other important social outcomes (Herrnstein &
Murray, 1994; Jensen, 1998). IQ has been associated with brain
size (Anderson, 2003), and nerve conduction velocity appears
associated with IQ for primary heritable reasons (Rijsdijk &
Boomsba, 1997). Reliable correlations have been reported between
IQ and various elementary cognitive tasks. IQ has been correlated
with simple reaction time, for example, at .10; with choice
reaction time at .30; and with inspection time at .40 (Deary,
2000; Jensen, 1998). Finally, the heritability of general intelligence
approximates .50 –.80 (Plomin, 2001).
Although it is difficult to overstate the success of the collabo-
rative program of psychometric intelligence research that has
played out over the past century (Jensen, 1998), there remain
several problems in the general domain of cognitive ability and
performance that are, as yet, unresolved by the psychometric
approach. First and foremost is the formal nature of the psycho-
metric conception of intelligence (the hierarchical factor analytic
model), suspended, as it were, in rarefied air because of its com-
plex statistical definition and limited grounding in other domains
of psychology and neuroscience (Deary, 2000). Ian Deary (2000)
Daniel M. Higgins, Department of Psychology, Harvard University;
Jordan B. Peterson, Department of Psychology, University of Toronto,
Toronto, Ontario, Canada; Robert O. Pihl, Department of Psychology,
McGill University, Montreal, Quebec, Canada; Alice G. M. Lee, Faculty of
Medicine, University of Hawai’i at Manoa.
We thank James Keller, computer-chip architect extraordinaire, for his
generous financial support for the research reported in this article, as well
as David Hatch (known for his flamboyant style and inspiring personality),
David Ralston, Joe Hatchell, Randall Baumgart, Ron Halverson, Steve
Christoferson, and Sondra Tingwald for their cooperation and help, par-
ticularly with regard to Studies 3 and 4. Finally, we also thank Sara
Goldman, Jana Holvay, Christy Johnson, Crystal Layne, Lisa Lee, Mariko
Lui, Irena Milosevic, Craig Nathanson, Chayim Newman, William Rupp,
and Suzanne Toole for their help with the execution of Study 2.
Correspondence concerning this article should be addressed to Jordan B.
Peterson, Department of Psychology, University of Toronto, 100 St.
George Street, Toronto, Ontario, Canada M5S 3G3. E-mail:
Journal of Personality and Social Psychology Copyright 2007 by the American Psychological Association
2007, Vol. 93, No. 2, 298–319 0022-3514/07/$12.00 DOI: 10.1037/0022-3514.93.2.298
has argued that what some researchers may take as the psycho-
metric “theory” of human intelligence, the “three-stratum theory,”
is not so much a scientific theory as it is a statistical procedure for
reliably identifying consistent “pools of inter-individual variability
in the performance of mental tasks” across data sets (p 17), and has
reminded scientists that they still lack an adequate psychological
theory to account for the remarkable empirical regularity repre-
sented by the hierarchical model.
The second central issue is the practical prediction of perfor-
mance. The use of cognitive ability to predict academic and job
performance is an important aspect of classical psychometrics,
educational psychology, and organizational psychology (Neisser et
al., 1996). However, an optimized combination of IQ and other
trait measures (e.g., personality traits) still leaves 60%–75% of the
variance in performance unaccounted for (Schmidt & Hunter,
1998), and even small increments in predictive validity, with
regard to performance prediction, can have substantial practical
(Rosenthal, 1990) and economic consequences (Schmidt &
Hunter, 1998). Furthermore, a number of researchers have sug-
gested that standard IQ tests are limited: They evolved specifically
to assess the ability to solve well-defined (academic) problems and
therefore do a rather poor job of assessing an individual’s ability to
solve ill-defined problems (Pretz, Naples, & Sternberg, 2003).
Robert Sternberg’s (1999) theory of practical intelligence repre-
sents an essentially psychometric attempt to capitalize on this
distinction. Unfortunately, so far, this work has not been trans-
formed into effective technologies of assessment or intervention
(Gottfredson, 2003, p. 343; but see Sternberg’s, 2003, reply).
If one takes a neuropsychological rather than a psychometric
approach to intelligence, one finds the problem conceptualized at
a variety of levels of sophistication. Perhaps the crudest idea in all
of psychology—that smart people have bigger brains— has
evolved from its earliest mode of measuring cranial capacity with
lead shot (Gould, 1996), through statistical correlation of intelli-
gence with head size (Jensen, 1994), to meta-analytic studies of
magnetic-resonance-imaging-based estimates of brain volume cor-
relations with intelligence (Anderson, 2003; McDaniel, 2005).
From a functional perspective, intelligence has been associated
with neural efficiency based on electroencephalography (EEG;
Neubauer & Fink, 2003; Neubauer, Grabner, Fink, & Neuper,
2005; Neubauer, Grabner, Freudenthaler, Beckmann, & Guthke,
2004) and positron emission tomography (PET) studies (Haier,
2003). This global whole-brain approach has recently given way to
a series of more anatomically specific, often theory-driven, brain
imaging studies. Anatomically, evidence has accrued suggesting
that although both grey and white matter volume is correlated with
intelligence (Gignac, Vernon, & Wickett, 2003), the association of
grey matter volume with intelligence is particularly high in the
frontal lobe (Haier, Jung, Yeo, Head, & Alkire, 2004). It appears
that this association may be genetically mediated (Toga & Thomp-
son, 2005).
Functional studies (Duncan et al., 2000; J. R. Gray, Chabris, &
Braver, 2003; Prabhakaran, Rypma, & Gabrieli, 2001; Prabhaka-
ran, Smith, Desmond, Glover, & Gabrieli, 1997) have also impli-
cated the importance of the prefrontal cortex, along with posterior
cortical regions, in solving classes of problems that might be
broadly classified as reflecting fluid intelligence (Blair, 2006).
This seems reasonable, from a performance-prediction standpoint,
as the challenges raised by practical problems are typically dele-
gated, theoretically, to the executive functions of the frontal lobes
(Fuster, 1997; Luria, 1980). Indeed, attempts to understand the
higher order cognitive processes associated with the prefrontal
cortex appear particularly relevant to augmenting an understanding
of intelligence (Duncan, 1995). If real-world success is conceptu-
alized as long-range goal framing and attainment, intelligence
might be viewed as effective goal-directed problem solving under
novel and ambiguous environmental circumstances. This notion is
consistent with Gardner’s (1993) and Sternberg’s (1999) ideas and
has been elaborated in cognitive (G. A. Miller, Eugene, & Pribram,
1960; Wiener, 1961), social-personality (Carver & Scheier, 1998)
and neuroscience-based theories (J. A. Gray & McNaughton,
2000; Peterson, 1999; Rolls, 1999).
More specifically, Duncan (1995) argued that adaptive action
involves the planning and execution of hierarchically arranged
goals and subgoals (Schank & Abelson, 1977; Wiener, 1961) and
believes that the highest level of that hierarchy is governed by the
dorsolateral prefrontal cortex (Duncan et al., 2000). Duncan pos-
ited that environmental ambiguity (context) or distracted attention
(e.g., dual-task interference) reduces the efficiency of the goal-
weighting process and leads to goal neglect and impaired behav-
ioral selection or choice. In this view, the weighting process is
believed to be particularly sensitive to prefrontal damage and to
variance in psychometric g. Duncan and his colleagues have dem-
onstrated that goal neglect (disregarding task requirements) can be
induced using dual-task interference and that such executive fail-
ures characterizes patients with frontal lobe damage and normal
participants from the lower end of the gdistribution (Duncan,
Emslie, Williams, Johnson, & Freer, 1996). Duncan et al. (2000)
empirically underscored the importance of the prefrontal cortex for
(fluid) intelligence with the finding that increased lateral frontal
activation (using PET) is associated with solving heavily g-loaded
problems but not with solving problems whose gloadings are
negligible. In short, Duncan has argued that gis the reflection of
the cognitive process of the dorsolateral prefrontal cortex in psy-
chometric space. These technologically driven advances have cre-
ated a new wave of optimism that the neurobiological basis of
Spearman’s gwill soon be elucidated (Blair, 2006; J. R. Gray &
Thompson, 2004), although this optimism is not universal (Carey,
This new wave of neuroscientific investigation into psychomet-
ric intelligence is supported on its cognitive–psychological flank
by a theoretical reconceptualization of intelligence in terms of
Cattell’s (1987) distinction between crystallized and fluid intelli-
gence (Blair, 2006; Duncan et al., 2000; J. R. Gray et al., 2003).
The key cognitive component of this view of fluid intelligence is
working memory (Blair, 2006; Duncan, 2005; J. R. Gray &
Thompson, 2004). Empirical work crystallized around an early
confirmatory factor analytic study of working memory by Kyl-
lonen and Christal (1990). This study suggested that the common
factor underlying working memory ability might be virtually in-
distinguishable from reasoning ability, which they explicitly char-
acterized as general fluid intelligence. The idea that prefrontally
mediated working memory plays a determining role in fluid intel-
ligence (Kane & Engle, 2002) has been reinforced by a series of
recent studies from several groups (Ackerman, Beier, & Boyle,
2002; Conway, Cowan, Bunting, Therriault, & Minkoff, 2002;
Conway, Kane, & Engle, 2003; Kane et al., 2004; Su¨, Oberauer,
Wittmann, Wilhelm, & Schulze, 2002) demonstrating a strong
association between working memory and (fluid) intelligence.
Indeed, the empirical (Kane, Hambrick, & Conway, 2005; Ober-
auer, Schulze, Wilhelm, & Su¨, 2005) and conceptual (Deary,
2000) overlap between working memory and psychometric intel-
ligence is so striking that researchers have had to take considerable
pains to argue that they are distinct constructs (Blair, 2006). On the
basis of a recent meta-analysis, Ackerman, Beier, and Boyle
(2005) suggested that the correlation between working memory
capacity and psychometric gapproximates .48, and in a manner
quite uncharacteristic of this new assault on psychometric intelli-
gence, they suggested a psychometric interpretation of working
memory capacity as a lower order cognitive ability in the standard
hierarchical model of intelligence (Carroll, 1993). In a critical
response to the Ackerman et al. article, both Kane et al. (2005) and
Oberauer et al. (2005) offered meta-analytically derived correla-
tions between psychometric gand working memory that are sub-
stantially higher (.85 and .72, respectively) while strongly assert-
ing their conceptual distinction. Oberauer et al. (2005) even
suggested that interpreting working memory capacity in terms of
the psychometric model of intelligence is inappropriate. They
suggested that such an interpretation reflects a conceptual misun-
derstanding of the working memory construct, and the scientific
advances researchers have made in understanding its construct
validity (Conway et al., 2005), its neural underpinnings (Blair,
2006; Duncan, 2005; Fuster, 2005; Kane & Engle, 2002), and its
consequent potential to advance understanding of the inscrutable
psychometric construct of general intelligence (J. R. Gray &
Thompson, 2004).
In general, then, cognitive neuroscientists have come to provi-
sionally understand intelligence in terms of cognitive processes
associated with the prefrontal cortex (Duncan, 2005; Fuster, 2005;
J. R. Gray & Thompson, 2004). Unfortunately, this work still
remains poorly integrated with the larger body of knowledge
derived from over 100 years of psychometric research, and much
remains to be gained from a fully bipartisan psychometric–
cognitive-neuroscientific approach (Deary, 2005). First is the issue
of convergent and divergent validation:
When neuropsychologists devise and name a test of mental function,
they can profit from the psychometric approach which construes
people’s performance on the test in terms of variance shared with
other tests (whether this is at the gor group factors level) and variance
specific to that test itself. (Deary, 2005, p. 227)
Then there is the equally important and interesting question of
criterion validity. Prefrontally mediated cognitive processes, in-
cluding working memory, may very well share substantial variance
with psychometric intelligence. This leaves open the question of
whether these prefrontal cognitive processes share substantial vari-
ance with the important life outcomes (e.g., academic perfor-
mance, job performance, social status, etc.) that make the investi-
gation of psychometric intelligence relevant in the first place
(Herrnstein & Murray, 1994).
Whereas the work described above could be characterized as an
attempt to account for individual differences in psychometric
intelligence in terms of neuropsychological constructs using the
methods of cognitive neuroscience, the work presented in this
article could be described as an attempt to define cognitive pro-
cesses associated with the prefrontal cortex in psychometric terms.
To this end, we operationally defined dorsolateral prefrontal cog-
nitive ability (D-PFCA) as performance on a computerized battery
of prefrontal cognitive tasks, established its psychometric proper-
ties and examined its relationship to psychometric intelligence and
academic and job performance. Furthermore, because of its dem-
onstrated relevance to performance prediction (Hunter & Schmidt,
1996), Big Five personality assessment was also incorporated into
these studies. The Big Five constitute a well-defined measurement
model for assessing personality across five broad dimensions:
Neuroticism, Extraversion, Openness, Agreeableness, and Consci-
entiousness (Digman, 1990; L. R. Goldberg, 1992; John & Sriv-
astava, 1999). The personality dimensions assessed by this five-
factor model appear valid cross-culturally (McCrae & Costa, 1997)
and are substantially heritable (Loehlin, McCrae, Costa, & John,
1998) and relatively stable across the life span (Costa & McCrae,
1997). Conscientiousness, in particular, has broad predictive va-
lidity for academic (Goff & Ackerman, 1992; E. K. Gray &
Watson, 2002; M. G. Rothstein, Paunonen, Rush, & King, 1994)
and job performance (Barrick & Mount, 1991; Hogan, Hogan, &
Roberts, 1996; Hunter & Schmidt, 1996; Hurtz & Donovan, 2000;
Salgado, 1997).
The first two studies presented in this article examined the
relationship between D-PFCA, psychometric intelligence, Big
Five personality, and achievement in higher academic settings.
The second two studies, which did not include measures of psy-
chometric intelligence, extended the assessment of the criterion
validity of D-PFCA and personality to the workplace in (a) com-
plex managerial–administrative positions and (b) simpler, rote-
learning, assembly-line positions.
Study 1 had two specific goals: to examine the relationship
between D-PFCA and intelligence and to predict academic perfor-
mance. Predicting academic performance was the original problem
of intelligence research, first tackled by Binet and Simon (1905).
Grade point average (GPA) has not proven itself a particularly
useful predictor of real-world performance (Schmidt & Hunter,
1998), and the distinction between academic intelligence and prac-
tical or real-world intelligence has recently been stressed (Stern-
berg, 1999). However, maintaining a good GPA at a highly com-
petitive university is a complex and multiply determined task,
requiring a high level of performance over a protracted period of
time, as well as mastery across a range of domains. In addition, for
much of this assessment period, the student inhabits a rapidly
changing environment and, in addition to handling course work,
must exercise substantial executive control over a novel and com-
plex social, athletic, or cultural life. Thus, GPA is neither artifi-
cially constrained nor a shallow indicator of performance. Finally,
because GPA reflects performance averaged across many occa-
sions, raters, and contexts, it provides a standard and well-defined
measure of performance, whose apparent psychometric properties
are very difficult to duplicate in putative real world investigations.
In consequence, the first study examined the differential validity of
D-PFCA, IQ, and Scholastic Aptitude Test (SAT) scores with
regard to the prediction of academic performance. Because of the
apparent multiple determination of academic performance, the
study also assessed Big Five personality. The trait of Conscien-
tiousness appears most relevant to the current discussion, as this
trait has consistently emerged as a predictor of academic and job
performance. We hypothesized that D-PFCA would predict IQ and
SAT scores. Furthermore, we explored the possibility that
D-PFCA would predict academic performance over and above IQ
and SAT. Finally, we hypothesized that Conscientiousness would
independently predict academic performance.
Participants in this study included 121 full-time undergraduates
in the Faculty of Arts and Science at Harvard University, Cam-
bridge, Massachusetts. All participants were enrolled in an intro-
ductory course in personality psychology and received course
credit in return for participation. Although participation was op-
tional, all students enrolled in the course completed the experi-
ment. The participants included 67 female students. The average
age of the participants was 20 years (n117, SD 1.4 years,
range: 18 –24 years). Age data were unavailable for three partici-
pants. One older student (age 28 years) was excluded from the
calculation of average age.
Dorsolateral Prefrontal Cognitive Tasks
We defined D-PFCA as performance on a computerized battery
of seven neuropsychological tasks. Originally derived from clini-
cal neuropsychological studies (Milner & Petrides, 1984; Milner,
Petrides, & Smith, 1985), these tasks have been associated as
specifically as possible with dorsolateral prefrontal cortical func-
tion, as evidenced by a minimum of one well-designed clinical
study of brain-damaged patients, lesion studies with nonhuman
primates, and neuroimaging studies of non-brain-damaged people.
Performance on these tasks has variously been associated with
alcoholic propensity (Peterson, Finn, & Pihl, 1992), alcohol-
related aggression (Lau, Pihl, & Peterson, 1995), physical aggres-
siveness in boys (Se´guin, Nagin, Assaad, & Tremblay, 2004;
Se´guin, Pihl, Harden, Tremblay, & Boulerice, 1995), and school
performance (Peterson, Pihl, Higgins, Se´guin, & Tremblay, 2003).
Administration of the battery, which includes the following tasks,
requires approximately 90 min.
Conditional associative learning: Introduction. Petrides
(1985b, 1987) designed two conditional associative tasks to dem-
onstrate that the deficits seen in delayed response tasks in monkeys
with frontal damage may reflect an inability to learn conditional
behavioral rules (if Cue A, then Response X; else, if Cue B,
Response Y) rather than a working memory deficit. In the rhesus
monkey, periarcuate lesions (Area 8, Rostral Area 6), but not
periprincipalis lesions, produce impairments on both spatial (Pet-
rides, 1987) and nonspatial (Petrides, 1982, 1985a) conditional
associative tasks. In humans, unilateral surgical excisions (for the
treatment of epilepsy) of the left or right frontal lobes produce
impairments on both the spatial (Petrides, 1985b) and nonspatial
(Petrides, 1985b, 1990) versions of this task. Using PET, Petrides,
Alivisatos, Evans, and Meyer (1993) also found selective activa-
tion in Area 8 of the left frontal cortex during performance of the
nonspatial version of this task. A symmetrically reinforced go/
no-go task may also be conceptualized as a conditional task, with
correct performance dependant on mastery of a conditional rule: If
X, go; else, if Y, do not go (Petrides, 1987). As with the condi-
tional associative tasks, monkeys with periarcuate lesions are
impaired on this go/no-go task relative to monkeys with periprin-
cipalis lesions and normal controls (Petrides, 1986, 1987).
1. Spatial conditional associative task. The participant is pre-
sented with five identical circles and five identical squares ar-
ranged in a fixed spatial array. The participant is instructed that
each square is associated with exactly one circle. On each trial, a
circle is highlighted, and the participant is required to click the
square he or she believes might be associated with that circle. If the
participant chooses the correct square, the word “Correct” appears
on the screen, the trial is scored as correct, and the next trial
begins. If the participant chooses an incorrect square, the word
“Wrong” appears on the screen, the trial is scored as incorrect, and
the participant is allowed to continue to guess until he chooses the
correct square. Only when the participant clicks the correct square
does the next trial begin. The task is terminated when the partic-
ipant completes 10 consecutive correct trials or completes 100
trials, whichever comes first. The participant completes two dif-
ferent versions of this task (differing in the spatial arrangement of
the shapes) and receives a raw score equal to the total number of
trials completed across both versions.
2. Nonspatial conditional associative task. This task is similar
to the spatial version except that the spatial aspect has been
removed. The participant is explicitly required learn arbitrary
associations between cue words and target words. There are five
cue words and five target words. The participant is instructed that
each cue word is associated with exactly one of the target words.
On each trial, the participant is presented with a cue word (in the
center of the screen) and all five target words (arranged in a circle
around the cue word). The relative position of each of the target
words varies randomly from trial to trial. The participant is in-
structed to click the target word he or she believes might be
associated with the cue word. If the participant chooses the correct
target word, the word “Correct” appears on the screen, the trial is
scored as correct, and the next trial begins. If the participant
chooses an incorrect target word, the word “Wrong” appears on the
screen, the trial is scored as incorrect, and the participant is
allowed to continue to guess until he or she selects the correct
word. Only after clicking the correct word does the next trial
begin. The task is terminated when the participant completes 10
consecutive correct trials or completes 100 trials, whichever comes
first. The participant completes two different versions of this task
(the first employs regular words, whereas the second uses non-
words [e.g., egtao]) and receives a raw score equal to the total
number of trials completed across both versions.
3. Go/no-go. The participant is presented with four stimuli
(letters) flashing sequentially, repeatedly, and in random order
on the screen. Each appearance of a letter constitutes a trial. The
letters are displayed for 2 s. The intertrial time is a Gaussian-
distributed random number with a mean of 600 ms and a
standard deviation of 300 ms. Two of the letters are go stimuli,
and two are no-go stimuli. The participant is told that he or she
should click when certain letters appear and not click when
others appear. Whenever the participant makes a correct re-
sponse (clicking on a go trial or not clicking on a no-go trial),
he or she is rewarded by the appearance of the word “Good” and
by an increase in his or her score by 1 point (the score is
displayed throughout the task). The task is terminated after 200
trials or after 20 consecutive correct trials. The raw (error) score
is the number of trials completed.
Working memory: Introduction. Patients with unilateral left
and right frontal lobe surgical excisions perform poorly on the
self-ordered pointing task relative to brain-damaged and normal
controls (Petrides & Milner, 1982; Wiegersma, van der Scheer, &
Human, 1990). In monkeys, this deficit appears to be specific to
mid-dorsolateral frontal cortex (Areas 9 and 46), rather than the
more posterior dorsolateral cortex (Areas 6 and 8; Petrides,
1995b). PET has identified activation foci in Areas 46 and 9 in
(normal) humans while performing this task (Petrides, Alivisatos,
Evans, & Meyer, 1993). Patients with frontal lobe lesions also
appear impaired on an analogous verbal working memory task, the
digit span randomization task (Wiegersma et al., 1990). Once
again, PET has revealed bilateral activation in mid-dorsolateral
frontal cortex, Areas 46 and 9 (Petrides, Alivisatos, Meyer, &
Evans, 1993). Finally, frontal lobe damage also appears associated
with poor performance on a recency discrimination task (Milner et
al., 1985). When presented with a series of concrete representa-
tional drawings or abstract figures, for example, patients with
unilateral right frontal lobe damage appear impaired relative to
normal and brain-damaged controls. When words are used instead
of pictures, both left and right frontal damage is associated with
poor performance.
Petrides (1989, 1995a, 1998, 2000) argued that all these tasks
reflect the importance of the frontal cortex for working memory.
According to his two-level hypothesis, ventrolateral frontal cortex
interacts with temporal and parietal association areas and mediates
encoding and retrieval of information in short-term and long-term
memory. The mid-dorsolateral frontal cortex, on the other hand, is
responsible for the simultaneous manipulation and monitoring of
multiple pieces of information in working memory (Petrides,
1998). Areas 46 and 9 appear critically involved in this latter
function, and performance on the self-ordered, randomization, and
recency discrimination tasks reflects the activity of this mid-
dorsolateral memory system (Petrides, 1989).
4. Self-ordered pointing task. The participant is presented with
12 stimuli and instructed to click on each stimulus exactly once.
There is no time limit. After each trial (or selection), the spatial
location of each stimulus immediately changes, thus removing any
spatial aspect of the task. The participant is allowed to make only
12 selections. The participant completes four version of the task,
each using different classes of stimuli: abstract figure pictures,
pictures of easily named objects, words, and nonwords (e.g.,
xworl). The raw score for this task is the number of distinct stimuli
selected across all four versions (i.e., 48 number of errors).
5. Randomization task. For each trial, the participant is asked
to input a random sequence of letters for a given letter span (e.g.,
participants were asked to “randomize the letters from L to O,” i.e.,
a four-letter span). The participant enters letters by using the
computer monitor and the mouse. The monitor shows one of the
letters from the letter span. If the participant moves the mouse (in
any direction), the letter will change. As the participant continues
to move the mouse, new letters are displayed. While the mouse is
moving, the display will cycle forward through all the letters in the
span, beginning again at the first letter after the last letter has been
displayed. To select a letter, the participant clicks the mouse button
while the letter is on the screen. If the participant produces an
acceptable sequence, he or she is asked to randomize a span one
letter longer than the previous span. If the participant makes an
error (omission or patterned sequence, e.g., L, M, N), he or she is
given a second chance to randomize a span of the same length.
Before beginning the task proper, the participant must successfully
randomize 2 four-letter-span practice trials. After this, the partic-
ipant begins the scored trials with a letter span of four. The task
terminates when the participant fails to randomize a given length
span two trials in a row or when he or she correctly completes all
trials (span length 4 –14). The raw score is the maximum length of
span successfully randomized.
6. Recency discrimination task. On a typical trial, the partic-
ipant is sequentially presented with a series of (either six or eight)
familiar nouns. Each word appears for 800 ms and is followed by
a delay of 100 ms before the next appears. After all the words in
the sequence have been presented, the participant is shown two
words from the sequence and asked to “click on the word that
appeared most recently.” The participant completes eight trials
with six words and 14 trials with eight words. The raw score is the
number of trials for which the participant correctly identified the
more recent word.
Word fluency: Introduction. Both Milner and Benton (re-
viewed in Damasio & Anderson, 1993, p. 435) have demonstrated
that patients with left (but not right) frontal lobe damage manifest
impairment on Thurstone’s Word Fluency Test without presenting
with typical aphasia. Although patients with left frontal damage
are impaired on a word fluency task, both left and, more particu-
larly, right frontal damage appears associated with deficits on a
design fluency task (Milner & Petrides, 1984). PET studies have
demonstrated increased activation in left dorsolateral prefrontal
cortex in normal participants while performing a verbal fluency
task (Frith, Friston, Liddle, & Frackowiak, 1991) and that the level
of bilateral frontal lobe activity measured during a resting state
scan is negatively correlated with the number of words produced
(Boivin et al., 1992).
7. Word fluency task. The participant is instructed to enter as
many words as possible beginning with ST. He or she is instructed
not to use inflected forms and is given 5 min to complete the task.
The participant enters words by clicking an on-screen keyboard
using the mouse. The letters on this keyboard are arranged in
alphabetical order, and letters appear on the screen as they are
selected. The raw score is the number of valid words entered.
D-PFCA scoring. A sample of 444 participants was used to
establish scaled scores for each task. The scaled scores were
standardized normal scores (Anastasi & Urbina, 1997), determined
by calculating a percentile rank for each participant (based on raw
scores) and computing the z-score equivalent (based on the inverse
probability function) of this percentile rank. This resulted in a raw
score to (z-score equivalent) scale score mapping. In this standard
sample of 444, the average intercorrelation among the dorsolateral
prefrontal cognitive tasks (scaled scores) was .27. The internal
consistency reliability coefficient, or coefficient , for these tests,
when taken as a composite, was .72. An aggregate dorsolateral
prefrontal cognitive performance score, D-PFCA, was calculated
for each participant as the average of the scaled scores across the
seven dorsolateral prefrontal cognitive tasks.
A short form of the Wechsler Adult Intelligence Scale—Revised
(WAIS–R; Wechsler, 1981) consisting of the block design and vo-
cabulary subtests was used to measure IQ. With validity and reliabil-
ity defined as per Tellegen and Briggs (1967), this is the most valid
(part–whole correlation) and reliable (inter- and intratest reliability)
two-subtest short form of the WAIS–R (Cyr & Brooker, 1984; Sil-
verstein, 1982). The validity coefficient for this short form is .90,
whereas the reliability coefficient is .94 (Ward & Ryan, 1996). Full
scale equivalent IQ scores (IQ) were calculated according to Brooker
and Cyr (1986). The national mean for WAIS–R IQ is 100 and the
standard deviation 15 (Wechsler, 1981).
The 240-item Revised NEO Personality Inventory (NEO-PI-R)
Form S (Costa & McCrae, 1992) was used to assess personality.
This scale measures the personality dimensions of Neuroticism,
Extraversion, Openness, Agreeableness, and Conscientiousness,
and the subscales have reliability coefficients of .92, .89, .87, .86,
and .90, respectively.
Official university transcripts were available for 72 participants.
The transcripts included a cumulative ranking, which is a number
from 1 to 6: a score of 1 reflecting higher academic performance
than 6. A variable, GRADES, was calculated as 7 cumulative
ranking, resulting in a GRADES score from 1 to 6 points. Year 1
rankings were available for 72 participants, whereas Years 2, 3,
and 4 were available for 48, 28, and 15 participants, respectively.
Coefficient was calculated on the basis of the correlation be-
tween Year 1 ranking and the average of the rankings across Years
2– 4. When calculated this way, coefficient for the GRADES
composite used in this study was .84 (n48). Only one partici-
pant scored 1 point. This was an extreme value, and the participant
was removed from any subsequent analysis involving GRADES.
The participants completed a demographics questionnaire, in
which they were asked to rate their family’s annual gross income
on a scale of 1 to 13, where 1 $1,000 –$10,000; 2 $10,000
$20,000; 3 $20,000 –$30,000; 4 $30,000 –$40,000; 5
$40,000 –$50,000; 6 $50,000 –$75,000; 7 $75,000
$100,000; 8 $100,000 –$150,000; 9 $150,000 –$500,000;
10 $500,000 –$1,000,000; 11 $1,000,000 –$5,000,000; 12
$5,000,000 –$10,000,000; 13 $10,000,000 or more. This annual
household income index (AHII) was used as a control variable in
some of the analyses reported below.
Scholastic Aptitude Test
The SAT has two subtests, a verbal subtest (SAT-V) that tests
reading comprehension, antonyms, analogies, and sentence comple-
tion and a mathematical subtest (SAT-M) that focuses on numerical
and quantitative ability. These subtests are standardized to a national
mean of 500 and standard deviation of 100 (Jensen, 1980, p. 483).
Participants were asked to enter their best SAT-V score and their best
SAT-M score on the demographics questionnaire.
During the course of the semester, participants completed the
NEO-PI-R and the demographics questionnaire at home at a time
of their choosing. A trained tester administered the two WAIS–R
subtests in the laboratory. On a separate day, participants came
into the laboratory to complete the computerized battery of dor-
solateral prefrontal cognitive tasks. Official transcripts were ob-
tained from the registrar’s office, with the student’s written per-
Descriptive statistics and intercorrelations for NEO-PI-R per-
sonality factors and GRADES are presented in Table 1. Consci-
entiousness was the only personality variable that was significantly
correlated with GRADES. Openness was significantly correlated
with IQ (r.28, n106, p.004, two-tailed) and SAT-V (r
.29, n114, p.001, two-tailed). These were the only signifi-
cant correlations between the five personality factors and the
cognitive variables (IQ, SAT-V, SAT-M, D-PFCA).
Cognitive Variables
Descriptive statistics and intercorrelations for D-PFCA, IQ,
SAT-V, SAT-M, AHII, and GRADES are presented in Table 2.
All the cognitive variables were positively intercorrelated, and
each was significantly positively correlated with GRADES. The
zero-order correlations of AHII with the cognitive variables and
GRADES were not significant (unlike Study 2, below).
D-PFCA Predicts Academic Performance After
Controlling for IQ and SAT Scores
The results of a standard multiple regression, with GRADES as
the dependent variable and D-PFCA, IQ, SAT-V, SAT-M, Con-
scientiousness, and AHII as independents, are presented in Table
3. The regression coefficients for D-PFCA, SAT-V, Conscien-
Because these self-reported SAT scores were used in the subsequent
analyses, a few comments regarding their probable accuracy are appropri-
ate. Participants were assigned a numeric ID code and were required to use
this code (rather than their true identity) for all aspects of the experiment.
Thus, they were aware that their true identity would not be associated with
the SAT scores they provided. In all, 115 participants provided their SAT
scores. Of these, 109 (95%) explicitly stated that the scores were accurate,
and 79 (69%) explicitly stated that they would be willing to provide hard
copies of their SAT reports (using their assigned numeric ID codes). Before
providing their official transcripts, participants were asked to rate their
academic performance on a scale of 1 to 12: 1 A (4.00), 2 A(3.67),
3B(3.33), 4 B (3.00), 5 B(2.67), 6 C(2.33), 7 C (2.00),
8C(1.67), 9 D(1.33), 10 D (1.00), 11 D(0.67), and 12
F(0.00). This self-report rating of academic performance correlated with
the cumulative rankings from the official transcript (which is on a scale of
1– 6, i.e., quite different from the self-report scale) at r.88 (n71, p
.001, one-tailed). Thus, participants’ self-report academic performance
matched their official transcript ranking quite closely. Taken together,
these facts support the validity of the self-report SAT scores.
tiousness, and AHII were significant (one-tailed tests), whereas
those for IQ and SAT-M were not.
Two additional regression analyses were run to compare the
traditional means of predicting academic performance (IQ,
SAT-V, and SAT-M) with D-PFCA and Conscientiousness. When
GRADES was regressed on IQ, SAT-V, and SAT-M, the Rfor
regression was .39, F(3, 62) 3.62, p.02, and the adjusted R
was .11. On the other hand, when GRADES was regressed on
PFCA and Conscientiousness, Rfor regression was .48, F(2, 65)
9.59, p.001, and the adjusted R
was .20. In this latter regres-
sion, the standardized regression coefficients (s) were .41 and .31
for D-PFCA and Conscientiousness, respectively.
Table 1
Descriptive Statistics and Intercorrelations for Revised NEO Personality Inventory Personality Factors and Academic Performance
for Studies 1 and 2
Variable MSDRange 1 2 3 4 5 6
Study 1 (N117)
1. N 99.86 24.37 42–157
2. E 115.85 21.78 66–168 .38
3. O 130.05 19.62 71–173 .15 .20
4. A 111.79 21.78 59–168 .01 .15 .30
5. C 105.50 23.26 57–157 .49
.18 .09
4.77 0.74 3–6 .03 .18 .17 .13 .26
Study 2 (N141)
1. N 100.21 25.58 50–171
2. E 117.09 21.33 56–174 .33
3. O 132.38 19.18 78–170 .12 .43
4. A 113.64 18.73 40–158 .30
.10 .22
5. C 109.81 22.08 46–164 .44
.10 .04 .23
2.93 0.60 1.29–4.12 .03 .05 .09 .07 .37
Note. NNeuroticism; E Extraversion; O Openness; A Agreeableness; C Conscientiousness; GRADES academic performance.
N70 for this variable because of missing data and removal of one extreme value.
N133 for this variable because of missing data.
p.05, two-tailed.
p.01, two-tailed.
Table 2
Descriptive Statistics and Intercorrelations for Cognitive Measures, Academic Performance, and Annual Household Income for
Studies 1 and 2
Descriptive statistics Correlations Sample size for correlations
NMSD Range 1 2 3 4 5 6 7 1234567
Study 1
1. GRADES 71 4.77 0.74 3–6
2. D-PFCA 110 0.47 0.48 0.91–1.83 .37
3. AHII 118 7.59 1.85 2–13 .08 .01 70 107
4. IQ 109 127.61 11.48 100–151 .24
.05 67 102 107
5. RPM
6. SAT-V 115 709.57 70.19 530–800 .35
.04 .57
69 105 115 106
7. SAT-M 115 727.74 55.28 560–800 .33
.07 .34
69 105 115 106 115
Study 2
1. GRADES 133 2.93 0.60 1.29–4.12
2. D-PFCA 142 0.08 0.47 1.29–1.30 .33
3. AHII 142 6.63 1.86 2–13 .20
133 142
4. IQ 142 126.85 13.92 98–155 .39
133 142 142
5. RPM 142 25.3 4.33 13–35 .18
133 142 142 142
6. SAT-V 142
7. SAT-M 142 ————
Note. AHII Annual Household Income Index; D-PFCA dorsolateral prefrontal cognitive ability; GRADES academic performance; RPM
Raven’s Advanced Progressive Matrices, Set II; SAT-M mathematical score on Scholastic Aptitude Test; SAT-V verbal score on Scholastic Aptitude
p.05, one-tailed.
p.01, one-tailed.
D-PFCA was significantly correlated with, but not identical to,
IQ and SAT scores. Although all the cognitive variables were
related to academic performance, multiple linear regression sug-
gested that D-PFCA may be related to academic performance over
and above IQ and SAT scores. Consciousness was also an inde-
pendent predictor of academic performance.
Study 2 was designed to replicate and extend the results of
Study 1 and to more thoroughly examine the relationship between
D-PFCA, intelligence, and academic performance. Because Study
1 only employed two WAIS–R subtests, it may not have included
a broad enough range of standard IQ subtests to allow for a
comprehensive comparison of IQ and D-PFCA. In Study 2, we
measured D-PFCA in tandem with (a) a five-subtest short form of
the Wechsler Adult Intelligence Scale—Third Edition (WAIS–III;
Wechsler, 1997) IQ and (b) performance on the advanced set of
Raven’s Progressive Matrices (RPM; Raven, 1998), which is
widely considered to be one of the best single measures of g
(Jensen, 1998). The RPM and WAIS–III subtests were used to
derive gfactor scores. Personality was assessed using the NEO-
PI-R. The specific hypotheses for this study were that D-PFCA
would relate to IQ and RPM scores, that D-PFCA would predict
academic performance, that D-PFCA would predict academic per-
formance over and above IQ or gfactor scores, and that Consci-
entiousness would independently predict academic performance.
The individuals who participated in this study constituted a
subset of participants described in another context (DeYoung,
Peterson, & Higgins, 2002). The comprehensive sample of 245
University of Toronto (Toronto, Ontario, Canada) students were
recruited either through poster advertisements promising native
English-speakers $30 Canadian for completing a variety of Web
questionnaires and lab tests or through the departmental Introduc-
tion to Psychology study pool, which offered course credit in
return for completing the experimental protocol. University tran-
scripts and the relevant cognitive data were available for a subset
of 177 of these students. Thirty-five of these participants were
excluded from the final analysis either because English did not
turn out to be their first language or because their academic
department or their transcripts did not meet the inclusion criteria
detailed below. There were 97 female and 45 male students in the
final sample of 142 persons. The mean age of the participants was
20.7 years (n139, SD 2.5 years, range: 18 –35 years). Only
5 participants were over 25 years, whereas 123 participants were
between 18 and 22 years. There was no difference in age between
male and female participants.
Prospective participants were instructed to visit a Web site
where, after identifying themselves as University of Toronto un-
Table 3
Regression Analyses Predicting Academic Performance With D-PFCA, IQ, g, SAT, Conscientiousness, and AHII for Studies 1 and 2
Variable BSEBtp
rprsr Regression
Study 1: Regression of GRADES on IQ, SAT-V, SAT-M, D-PFCA, C, and AHII
Constant 0.009 1.405 0.01 .995 R.59
IQ 0.001 0.009 .02 0.17 .865 .24 .02 .02 R
SAT-V 0.003 0.001 .24 1.82 .073 .31 .23 .20 adjusted R
SAT-M 0.001 0.002 .05 0.40 .692 .33 .05 .04 F(6, 57) 4.96, p.001
D-PFCA 0.546 0.178 .37 3.07 .003 .36 .38 .33
C 0.012 0.004 .33 3.02 .004 .28 .37 .32
AHII 0.082 0.049 .18 1.69 .097 .11 .22 .18
Study 2: Regression of GRADES on IQ, D-PFCA, C, and AHII
Constant 0.456 0.503 0.91 .367 R.53
IQ 0.010 0.004 .24 2.68 .008 .40 .23 .20 R
D-PFCA 0.216 0.108 .17 2.01 .047 .33 .18 .15 adjusted R
C 0.008 0.002 .31 4.10 .000 .37 .34 .31 F(4, 127) 12.68,
AHII 0.032 0.026 .10 1.25 .214 .20 .11 .09 p.001
Study 2: Regression of GRADES on g, D-PFCA, C, and AHII
Constant 1.676 0.290 5.78 .000 R.51
g0.111 0.064 .16 1.75 .083 .34 .15 .13 R
D-PFCA 0.240 0.114 .19 2.10 .038 .33 .18 .16 adjusted R
C 0.009 0.002 .33 4.26 .000 .37 .35 .32 F(4, 127) 11.32,
AHII 0.040 0.026 .12 1.54 .126 .20 .14 .12 p.001
Note. AHII Annual Household Income Index; C Revised NEO Personality Inventory Conscientiousness; D-PFCA dorsolateral prefrontal cognitive
ability; ggfactor scores; GRADES academic performance; SAT Scholastic Aptitude Test; SAT-M mathematical score on SAT; SAT-V verbal
score on SAT.
pvalues are for two-tailed test.
dergraduates and indicating that English was their first language,
they were allowed to complete a number of questionnaires and
schedule an in-lab session. The in-lab portion of the study required
participants to complete the battery of D-PFCA tasks described in
Study 1, a short form of the WAIS–III (Wechsler, 1997), RPM Set
II (Raven, 1998), the NEO-PI-R Form S personality inventory
(Costa & McCrae, 1992), and some additional questionnaires. The
order of completion of these tasks was counterbalanced across
participants to eliminate order effects. Before leaving, participants
either signed a release form for their academic transcripts or
submitted a printout of their unofficial grade report from the
University of Toronto registrar’s Web site.
A short form of the WAIS–III (Wechsler, 1997) was used to
measure IQ. The short form was composed of three verbal subtests
(Vocabulary, Similarities and Arithmetic) and two performance
subtests (Digit Symbol and Block Design). Using formulas devel-
oped by Tellegen and Briggs (1967) to correct for error contami-
nation, Ward and Ryan (1996) reported that this short form
(VASBY, in Ward & Ryan’s, 1996, terminology) offers a validity
coefficient (part–whole correlation) of .94 and a reliability coeffi-
cient of .96, while reducing testing time by 55%–59%. This short
form was somewhat overweighted (three out of five tests) toward
crystallized intelligence (Cattell, 1987). RPM Set II, a popular test
of fluid intelligence, was also administered. The advanced set is
appropriate for the differentiation of individuals of superior ability,
for example, university undergraduates (Raven, 1998). The 40-min
timed version was used. The test–retest reliability coefficient for
this test is .91, whereas its split-half reliability coefficient is
Academic Performance
Transcripts were obtained for all participants. There was a
surprising amount of variability among the transcripts in terms of
departmental affiliation, academic load, and academic perfor-
mance. Consequently, only students with transcripts from the
Faculty of Arts and Science were included in the analyses. Aca-
demic performance (GRADES) was calculated as the straight
average of the annual GPA (AGPA) across all valid years. A year
was considered valid if the student took at least three academic
courses in both spring and fall semesters and did not score an F in
either semester. On the transcripts, an F is described as “Wholly
Inadequate” and indicates performance that is qualitatively below
par. The transcripts fail to indicate whether the grade reflects a
student’s inability to learn the course material or, as is often the
case, the occurrence of personal problems that render the student
unable to perform at his or her typical level or result in him or her
unofficially withdrawing from the course. Because of the qualita-
tively distinct nature of an F and the probability that this score
reflects nonacademic problems, academic years wherein the stu-
dent scored an F were excluded from the calculation of GRADES.
Additionally, summer courses were excluded entirely, so that
AGPA was calculated as the GPA for courses taken in the fall and
spring semesters only. GRADES, therefore, reflects an average
(across academic years) of each student’s typical performance (no
Fs) in typical courses (fall and spring) while carrying a typical
workload (three to five courses per semester). In the final analysis,
only 133 of the participants were Faculty of Arts and Science
students and had completed at least one academic year that met the
valid-year criteria detailed above, and it was this final subset that
was used in all analyses pertaining to GRADES.
Determining the reliability of the GRADES composite was
complicated by the fact that Year 1 AGPAs were available for 132
participants, whereas Years 2, 3, 4, and 5 AGPAs were available
for 70, 33, 10, and 3 participants, respectively. Consequently,
coefficient was calculated on the basis of the correlation between
Year 1 AGPA and the average of the AGPAs across Years 2–5.
When calculated this way, coefficient for the GRADES com-
posite used in this study was .82 (n71).
Participants completed a Web questionnaire designed to collect
demographic information. Participants were asked to provide their
date of birth and gender. They were also asked to estimate their
family’s annual gross income (Canadian dollars; 1: $1,000
$9,999; 2: $10,000 –$19,999; 3: $20,000–$29,999; 4: $30,000
$39,999; 5: $40,000 –$49,999; 6: $50,000–$74,999; 7: $75,000
$99,999; 8: $100,000 –$149,999; 9: $150,000–$499,999; 10:
$500,000 –$999,999; 11: $1,000,000–$4,999,999; 12:
$5,000,000 –$9,999,999; 13: $10,000,000 or more), and were as-
signed a score (the AHII) between 1 and 13 that corresponded
(based on the above list) to the income level indicated by the
Table 1 shows means, standard deviations, and intercorrelations
for the five NEO-PI-R personality factors and their correlations
with GRADES. Conscientiousness was the only personality factor
that had a significant correlation with GRADES. Openness was
significantly correlated with RPM, IQ, and D-PFCA at r.22,
.18, and .17, respectively ( p.01, .03, .05, two-tailed, respec-
tively; n141). None of the correlations between the other four
personality variables and the three cognitive variables were sig-
nificant. Household income (AHII) was positively correlated with
Extraversion (r.24, n141, p.002, two-tailed) and with
Openness (r.20, n141, p.01, two-tailed). The partial
correlation between AHII and Openness, when controlling for IQ,
was r.15 (df 138, p.07, two-tailed).
Cognitive Variables
Descriptive statistics and correlations for IQ, RPM, D-PFCA,
GRADES, and AHII are presented in Table 2. IQ, RPM, and
It should be noted here that the correlation between GRADES thus
calculated and the transcript cumulative GPA (reflecting a straight average
of performance (F or otherwise) across all courses taken across all semes-
ters (fall, spring, and summer) was .98, and substituting cumulative GPA
for GRADES in the analyses described in the Results section had a
negligible effect on the results obtained.
D-PFCA were all significantly positively intercorrelated, and each
was significantly positively correlated with GRADES. In the rel-
evant literature, measures of intelligence are typically reported as
being positively correlated with socioeconomic status (Neisser et
al., 1996). In this study, AHII was significantly and positively
correlated with IQ, RPM, D-PFCA, and GRADES.
D-PFCA Predicts Academic Performance After
Controlling for IQ or g
Principle axis factoring was used to calculate gfactor scores
based on RPM scores and the five WAIS–III subtests (Vocabulary,
Similarities, Arithmetic, Digit Symbol, and Block Design). The
first factor was extracted and factor scores calculated for each
participant using the regression method. This factor accounted for
33% of the variance in the correlation matrix. The factor loadings
were .74, .70, .60, .57, .48, and .18, for Block Design, RPM,
Similarities, Vocabulary, Arithmetic, and Digit Symbol, respec-
tively. The resulting gfactor scores had a mean of 0, a standard
deviation of 0.88, a minimum of 2.33, and a maximum of 1.77.
The correlation between D-PFCA and gwas .53 (n142, p
.001, one-tailed).
Unlike IQ, D-PFCA and Conscientiousness are not traditional
predictors of academic performance. It is of some interest, then, to
note that when GRADES was regressed against D-PFCA and
Conscientiousness, the regression was significant, F(2, 129)
18.54, p.001, and yielded a multiple Rof .47 and an adjusted
of .21, with semipartial correlation coefficients of sr .29 for
D-PFCA ( p.001, one-tailed) and sr .34 for Conscientious-
ness ( p.001, one-tailed).
In Study 1, regression analysis demonstrated that D-PFCA pre-
dicts academic performance over and above IQ and SAT scores. In
this sample, when GRADES was regressed against IQ, D-PFCA,
Conscientiousness, and AHII (see Table 3), the semipartial corre-
lation of D-PFCA with GRADES was significant, replicating
D-PFCA’s validity independent of IQ. In previous research (Dun-
can et al., 2000), the dorsolateral prefrontal cortex has been
claimed as the neural substrate g. When GRADES was regressed
against gfactor scores, D-PFCA, Conscientiousness, and AHII
(see Table 3), the semipartial correlation for D-PFCA was signif-
icant, as was the semipartial correlation for g(one-tailed), indicat-
ing that both D-PFCA and gwere independently related to per-
A significant positive correlation was found between D-PFCA
and university-level academic performance, replicating the finding
in Study 1 and increasing our confidence that D-PFCA may be as
valid a predictor of performance in this context as traditional
intelligence tests. Regression analysis revealed that a combination
of D-PFCA and Conscientiousness accounted for 21% (adjusted
) of the variance in performance. Adding either IQ or gto the
predictors did not radically alter this result. These latter regressions
demonstrate that D-PFCA predicts performance independent of
either traditional IQ or factor analytically derived gfactors scores.
Conscientiousness was also significantly and independently cor-
related with performance.
D-PFCA and the personality trait of Conscientiousness thus
emerge as viable alternatives to IQ (and g) as predictors of aca-
demic performance—which, by virtue of acting as a gatekeeper to
the more prestigious occupations, is itself an important life out-
come (Brody, 1992; Neisser et al., 1996).
Having established at least provisional evidence that D-PFCA is
characterized by criterion-related validity in relationship to ad-
vanced academic performance and by reasonable divergent and
convergent validity in relationship to IQ, g, and Big Five trait
personality, we attempted to extend its validity to the industrial–
organizational domain. Different types of jobs can be rank-ordered
according to job complexity, and the correlation between job
performance and cognitive ability typically increases with com-
plexity (Hunter & Schmidt, 1996; Schmidt & Hunter, 1998). We
therefore examined the validity of D-PFCA and personality for
predicting job performance for high-complexity administrative
jobs (Study 3) and low-complexity factory floor jobs (Study 4) at
a midsized U.S. manufacturing corporation. Study 3, as well as
Study 4, also employed a novel Big Five personality trait measure
designed for distant, rapid computer administration.
We hypothesized that D-PFCA would be positively correlated
with job performance for the higher complexity jobs examined in
Study 3. We also hypothesized that Conscientiousness and Emo-
tional Stability would prove useful predictors, following the pre-
vious research literature.
Participants included 80 salaried employees from the adminis-
trative branch of a midsized manufacturing corporation. This
group included 20 employees in administration and sales, 35 in
customer service, 13 in secretarial– basic services, and 12 in man-
agement. Thirty-six of the participants were female. Age data were
available for 49 participants. The average age was 40.4 years
(SD 9.7 years, range: 22– 64 years). Participants were aware that
their supervisors would have access to the testing results.
Personality was measured with a short computerized Five-
Dimensional Temperament Inventory (FDTI). The FDTI measures
Emotional Stability (reverse Neuroticism), Extraversion, Open-
ness, Agreeableness, and Conscientiousness and is modeled after
L. R. Goldberg’s (1992) list of 100 trait-descriptive adjectives
(TDAs). There are 50 items: 10 items for each of the five factors.
Each item is a bipolar visual analogue scale. At each pole, there are
three adjectives. The adjectives at one pole are drawn from a list
of adjectives that load on the positive end of the underlying factor,
whereas the adjectives at the opposite pole load on the negative
end of the factor. For example, a typical Extraversion item might
have Unadventurous, Shy, Withdrawn on the left end of the scale
and Active, Daring, Talkative on the right end of the scale. Above
the scale sits the question “What point on the scale best describes
you?” To enter his or her response, the participant clicks on the
scale using the mouse.
Item responses from 518 participants (298 university undergrad-
uates and 220 corporate employees) were factor analyzed using
principal factor analysis with varimax rotation (the same result is
obtained if direct oblimin rotation is used). Goldberg’s original
five factors were recovered, and these were the only factors with
eigenvalues greater than one. This confirmed the five-factor struc-
ture of the inventory. Using the same sample, coefficient for the
Emotional Stability, Extraversion, Openness, Agreeableness, and
Conscientiousness factors were .97, .98, .95, .96, and .98, respec-
tively. The intercorrelations among the five variables are presented
in Table 4.
The relationships of the FDTI factors with the NEO-PI-R and
Goldberg’s TDA factors were previously investigated using a
sample of 177 university undergraduates (Lee, Higgins, Peterson,
& Pihl, 2001). The FDTI factors of Emotional Stability, Extraver-
sion, Openness, Agreeableness, and Conscientiousness correlate
with the NEO-PI-R factors at .77, .76, .49, .62, and .73 for
Neuroticism, Extraversion, Openness, Agreeableness, and Consci-
entiousness, respectively, and with Goldberg’s TDA factors at .74,
.84, .74, .82, and .82 for Emotional Stability, Extraversion, Open-
ness, Agreeableness, and Conscientiousness, respectively. The
mean test–retest reliability of the FDTI was .83 ( p.01). For the
individual factors, the test–retest reliability was found to be .84,
.88, .80, .81, and .81 for Emotional Stability, Extraversion, Open-
ness, Agreeableness, and Conscientiousness, respectively.
Supervisor- and Self-Rated Job Performance
The corporation evaluated their employees twice yearly, in April
and in November. Each evaluation included twelve 16-point rating
scales: quality, quantity, knowledge, versatility, judgment, communi-
cations, human relations, professionalism, responsiveness, punctual-
ity, attendance, and overall performance. The employee’s immediate
supervisor completed this evaluation. The April and November rat-
ings for a given year were both included on the same form, with, for
example, the quality rating scale for April immediately above the
quality rating scale for November. For this reason, April and Novem-
ber ratings were treated as part of the same scale, and we derived
annual performance ratings by averaging the 24 rating scores found
on each annual report. Complete annual evaluation forms were avail-
able for three years: 1998, 1999, and 2000. We used the Year 2000
evaluations to calculate the internal consistency of the performance
rating scale because it included evaluations for the largest subset of
participants (n76). Coefficient for Year 2000 ratings was .95. In
addition to supervisor ratings, the corporate records also included
self-rated evaluations. The self-rated forms were identical in format to
the supervisor-rated forms. Again, Year 2000 evaluations were used
to determine the internal consistency of the self-rated evaluations.
Coefficient for Year 2000 self-rated evaluations was .97 (n73)
Participants completed the FDTI and the computerized D-PFCA
battery described in Study 1. All participants were tested privately
in a quiet, isolated room. Job performance data were collected
from the corporation’s internal records.
Performance Ratings
Descriptive statistics and intercorrelations for annual supervisor
(1998, 1999, 2000) and self- (1998, 1999, 2000) ratings of job
performance are presented in Table 5. The average intercorrelation
across the three years was r
.71 for supervisor ratings, r
.72 for self-ratings, and r
.64 for both supervisor ratings and
self-ratings. Three composite variables were formed: SUPPERF is
the average of the annual supervisor ratings of performance,
SELFPERF is the average of the annual self-ratings of perfor-
mance, and PERF is the average of all annual ratings of job
performance. The internal consistencies, coefficient s, for these
composites were .88, .89, and .92 for SUPPERF, SELFPERF, and
PERF, respectively.
Personality and Performance
Descriptive statistics and the intercorrelations among personality
variables and job performance composites are presented in Table 6,
which indicates some disagreement between supervisor-rated and
self-rated performance. Supervisor-rated job performance was not
significantly correlated with any of the personality variables. Self-
rated job performance, on the other hand, was significantly correlated
with Extraversion, Openness, and Conscientiousness and had near-
significant correlations with Emotional Stability and Agreeableness
(p.09, .10, two-tailed, respectively).
Table 4
Descriptive Statistics and Intercorrelations Among Five-Dimensional Temperament Inventory
Factors for Standardization Sample (Study 3)
Variable MSD Range 12345
1. ES 0.10 1.00 2.8–2.8
2. E 0.09 1.00 2.8–2.8 .48
3. O 0.05 1.00 2.6–2.5 .32
4. A 0.08 1.00 2.8–2.2 .60
5. C 0.08 1.00 2.8–2.6 .48
Note. N 518. ES Emotional Stability; E Extraversion; O Openness; A Agreeableness; C
p.01, two-tailed.
D-PFCA Predicts Job Performance
There was a negative correlation between D-PFCA and age of
r⫽⫺.32 (n49, p.02, two-tailed). Descriptive statistics for
D-PFCA and the relationship between D-PFCA and job perfor-
mance are provided in Table 7. Three subsets of participants were
identified by experience level (Dunnette, 1966; H. R. Rothstein,
1990). The Level 1 group (n80) consisted of all participants
with 1 or more years of experience. This group contained all
participants. The Level 2 group (n53), a subset of the Level 1
group, contained all participants with 2 or more years of experi-
ence. The Level 3 group (n47), a subset of the Level 2 group,
contained all participants with 3 or more years of experience.
D-PFCA was significantly correlated with supervisor-rated job
performance in each of these overlapping groups, with r.42, r
.53, and r.57 for Levels 1, 2, and 3, respectively. Looking at this
relationship within nonoverlapping subsamples, the correlation
between D-PFCA and supervisor-rated performance was signifi-
cantly higher (Z2.11, p.035, two-tailed) for those with 3 or
more years of experience (r.57, n47, p.001, one-tailed)
than for those with less than 2 years of experience (r.12, n
Table 5
Descriptive Statistics and Intercorrelations Among Performance Ratings for Study 3
Variable NMSD Range 123456
1. SUP98 47 12.60 1.17 10.21–14.65
2. SUP99 52 12.60 1.34 8.25–14.55 .74
3. SUP00 78 12.79 1.13 10.30–14.90 .73
4. SELF98 43 12.90 1.31 9.72–15.70 .70
5. SELF99 47 12.51 1.41 9.10–15.00 .63
6. SELF00 74 12.42 1.29 7.80–15.05 .54
Note. SUP98, SUP99, and SUP00 supervisor-rated performance for the Years 1998, 1999, and 2000, respectively; SELF98, SELF99, and SELF00
self-rated performance for the Years 1998, 1999, and 2000, respectively.
p.01, two-tailed.
Table 6
Descriptive Statistics and Intercorrelations for Personality and Job Performance for Studies 3 and 4
Variable MSD Range 1 2 3 4 5 6 7 8
Study 3 (N80)
1. ES 0.43 0.93 2.0–2.2
2. E 0.16 1.00 1.9–2.8 .52
3. O 0.05 0.98 2.5–1.8 .64
4. A 0.30 1.00 2.2–2.2 .73
5. C 0.53 0.75 1.0–2.1 .63
Job performance
6. SUPPERF 12.64 1.08 9.32–14.58 .01 .08 .03 .07 .08
12.46 1.22 9.08–14.93 .20 .28
.19 .23
8. PERF 12.57 1.05 9.80–14.50 .12 .10 .14 .14 .18 .91
Study 4 (N96)
1. ES 0.56 0.97 1.9–2.8
2. E 0.46 1.01 2.8–2.8 .41
3. O 0.12 0.96 2.5–2.5 .65
4. A 0.50 0.97 2.8–2.2 .59
.13 .39
5. C 0.55 0.85 1.2–2.6 .67
Job performance
14.98 1.49 10.40–18.44 .06 .14 .13 .09 .23
Note. N 80. ES Emotional Stability; E Extraversion; O Openness; A Agreeableness; C Conscientiousness; SUPPERF supervisor-rated
performance; SELFPERF self-rated performance; PERF composite of supervisor- and self-rated performance.
N77 for this variable because of missing data.
N94 for this variable because of missing data.
p.05, two-tailed.
p.01, two-tailed.
27, p.28). Regression analysis, predicting D-PFCA using self-
ratings and supervisor performance ratings (R.45, adjusted
.18), F(2, 74) 9.6, p.001, indicated that the independent
relationship between supervisor ratings and D-PFCA (sr .28)
was significant, t(74) 2.69, p.009, whereas the relationship
between self-ratings and D-PFCA (sr .10) was not.
D-PFCA Prediction of Job Performance
On the basis of previous meta-analyses (Hunter & Hunter,
1984), Schmidt and Hunter (1998) reported that general mental
ability (GMA), or intelligence, is an excellent predictor of job
performance, with the corrected predictive validity of GMA esti-
mated at .51 for medium-complexity jobs (62% of U.S. jobs) and
.58 for professional–managerial jobs. In the present study, the
(uncorrected) correlation between D-PFCA and supervisor-rated
job performance was r.42 for the whole sample and r.57
when only those with at least 3 years of experience were included.
Similar results were found for self-rated job performance. For
supervisor performance ratings, correction for attenuation due to
the unreliability of the supervisor-rated performance composite
(␣⫽.88) and the unreliability of the D-PFCA scores (␣⫽.72)
revealed an estimated construct-level validity (correlation between
true D-PFCA scores and true performance scores) of r.52 for
the whole sample and r.72 for the most restrictive experience-
level group (Level 3). The corrected values of r.52–.72 com-
pare well with Hunter and Schmidt’s (1996) estimates, suggesting
that D-PFCA predicts job performance at least as well as GMA.
D-PFCA Prediction of Job Performance and Its
Relationship to Experience
Although it is often presumed that the relative importance of
cognitive ability decreases as experience level increases, this pre-
sumption does not appear to be correct (Hunter & Schmidt, 1996).
In fact, the empirical evidence suggests the reverse: The relation-
ship between cognitive ability and performance increases with
experience. McDaniel (cited in Hunter & Schmidt, 1996), for
example, found that the validity of cognitive ability increased from
.35 for 0 to 6 years, to .44 for 6 to 12 years, to .59 for more than
12 years. McDaniel’s results for general cognitive ability are
consistent with the results found in this study, as D-PFCA became
significantly more important in predicting job performance as level
of experience increased. This could mean that individuals become
assessed more accurately as time goes on (Dunnette, 1966; H. R.
Rothstein, 1990) or that the effect of cognitive ability on perfor-
mance actually increases over time (Hunter & Schmidt, 1996), or
D-PFCA and Aging
This study also found a significant correlation between age and
D-PFCA of r⫽⫺.32. This finding is in keeping with a large body
of research suggesting the existence of a linear decline in cognitive
function over the life span (Kramer & Willis, 2002). Age data were
available for a subset of the sample (n49). The majority of those
employees had at least 3 years of experience (n44), whereas the
remainder had at least 2. Consequently, the age analysis largely
consisted of Level 3 employees. Although there was a relationship
between D-PFCA and age, the correlation between age and
Table 7
Descriptive Statistics and Intercorrelations for Job Performance and Dorsolateral Prefrontal
Cognitive Ability by Experience for Study 3
Variable MSD Range 1 2 3 4
Level 1: Workers with 1 or more years of experience, N80
1. SUPPERF 12.64 1.08 9.33–14.58
12.46 1.22 9.08–14.93 .64
3. PERF 12.57 1.05 9.80–14.50 .90
4. D-PFCA 0.37 0.53 1.81–1.07 .42
Level 2: Workers with 2 or more years of experience, N53
1. SUPPERF 12.71 1.13 9.33–14.58
2. SELFPERF 12.56 1.30 9.08–14.93 .65
3. PERF 12.63 1.11 9.80–14.37 .90
4. D-PFCA 0.35 0.55 1.81–1.07 .53
Level 3: Workers with 3 or more years of experience, N47
1. SUPPERF 12.72 1.18 9.33–14.58
2. SELFPERF 12.74 1.26 9.08–14.93 .73
3. PERF 12.72 1.14 9.80–14.37 .93
4. D-PFCA 0.35 0.58 1.81–1.07 .57
Note. SUPPERF supervisor-rated performance; SELFPERF self-rated performance; PERF composite
of supervisor- and self-rated performance; D-PFCA dorsolateral prefrontal cognitive ability.
N77 for this variable because of missing data.
p.01, one-tailed.
supervisor-rated performance was not significant, and controlling
for age did not substantially affect the relationship between
D-PFCA and supervisor-rated performance, suggesting that the
relationship between D-PFCA and supervisor-rated performance
may hold independent of age, at least for this subgroup.
Personality and Performance
Construct-level validity for Conscientiousness (i.e., correlation
between personality trait and job performance, corrected for sam-
pling error, range restriction, and attenuation due to the unreliabili-
ties of both the personality trait and job performance measures) has
been estimated at .22 (Barrick & Mount, 1991), .25 (Salgado,
1997), and .22 (Hurtz & Donovan, 2000). On the other hand, the
mean observed correlation between Conscientiousness and perfor-
mance (without any corrections) was .13 (Barrick & Mount, 1991),
.10 (Salgado, 1997), and .14 (Hurtz & Donovan, 2000). The
uncorrected value of r.08 for Conscientiousness and supervisor-
rated performance obtained in the present study is consistent with
the other uncorrected values, although nonsignificant. Although
the uncorrected value of r.23 for Conscientiousness and self-
rated performance is high, this result appears of dubious practical
utility, given the lack of relationship with supervisor-rated perfor-
mance. It might be assumed that in performance situations, Emo-
tional Stability, Extraversion, Openness, Agreeableness, and Con-
scientiousness would each be more explicitly valued than their
polar opposites. Consistent with this assumption, self-rated perfor-
mance was positively and relatively strongly related to each per-
sonality trait, whereas supervisor-rated performance was inconsis-
tently (with regard to direction) and weakly related to the
personality traits. In terms of employment decisions, these data
suggest that self-report personality measures may tell employers a
good deal about how a prospective employee will rate, or overrate,
his or her own performance, at least under some circumstances, but
less about his or her performance as seen from the perspective of
his or her supervisor.
For Study 4, we tested a group of factory floor workers and their
immediate floor supervisors using the FDTI and the D-PFCA
battery described above. For these workers, only immediate su-
pervisor performance ratings were available. We hypothesized that
the predictive utility of D-PFCA would be attenuated in this
situation, in which performance was probably more amenable to
rote learning, but that Conscientiousness and Emotional Stability
might emerge as useful predictors.
Ninety-six factory floor workers participated in the study. Forty-six
were female. The sample included assemblers, laborers, fabricators,
machine operators, crew leaders, supervisors, and employees holding
other factory floor positions. The average age of the participants was
38.3 years (n95, SD 9.7 years, range: 20 –58 years).
As in Study 3, employees were assessed on two occasions per
year (Occasions a and b). On each occasion, employees were rated
on five categories: productivity and quality, safety and housekeep-
ing, human relations, responsibility and personal development, and
dependability and responsiveness. The maximum score for each
category was 20 points. Consequently, each performance report
contained 10 scores: five from Occasion a and five from Occasion
b. Performance evaluations were available for four years: 1998,
1999, 2000, and 2001. Year 2000 reports were available for 89
employees and were used to assess the reliability of the annual
assessment form (coefficient ␣⫽.92).
Participants completed the FDTI (Study 3) and the D-PFCA
battery (Study 1). Performance data, date of birth, date hired, and
years of education were derived from the corporation’s records.
Performance Ratings and Work Experience
Descriptive statistics and intercorrelations among the annual
performance ratings are presented in Table 8. The average inter-
correlation among the annual performance ratings was r
As in Study 3, the annual performance ratings were averaged
across the available years to produce a composite performance
variable, SUPPERF. On the basis of the intercorrelations across
the annual performance ratings, the reliability (coefficient )of
this composite variable was .91.
Table 8 also presents the mean and standard deviation for work
experience (EXPER), the number of years the employees had been
working with the corporation. There was a significant positive
correlation between work experience and job performance across
all four performance assessment years (see Table 8).
D-PFCA, Age, Experience, Education, and Job Performance
Descriptive statistics and intercorrelations for D-PFCA, age,
years of education, experience, and job performance are given in
Table 9. As in Study 3, there was a significant negative correlation
between D-PFCA and age and a near-significant positive correla-
tion between D-PFCA and years of education (r.18, n93,
p.09, two-tailed). The partial correlation between D-PFCA and
years of education, controlling for age, was pr .24 (df 90, p
.02, two-tailed).
The correlation between D-PFCA and job performance was not
significant. However, this sample was quite heterogeneous in
terms of years of experience, which was related to job perfor-
mance, and age, which was related to D-PFCA. The partial cor-
relation between D-PFCA and job performance, controlling for
experience and age, was pr .18 (df 89, p.04, one-tailed)
and was pr 0.21 (df 86, p.02, one-tailed) when years of
education were added to the list of control variables.
Personality and Job Performance
Descriptive statistics and intercorrelations among personality
variables and job performance are presented in Table 6. As in
Study 3, the intercorrelations among the personality variables were
quite high; however, only Conscientiousness had a statistically
significant correlation with job performance.
Full Regression Model
When performance was regressed on experience, D-PFCA, and
Conscientiousness, controlling for age and years of education, the
Rfor regression was .42, which was significant, F(5, 85) 3.66,
p.005. The adjusted R
was .13. The standardized regression
coefficients for experience, D-PFCA, and Conscientiousness were
␤⫽.32, .19, and .18, respectively (sr .32, .17, and .17,
respectively) and were significant ( ps.01, .04, and .04, all
one-tailed, respectively). The regression coefficients for age and
years of education (␤⫽.09 and .11, respectively) were not
A nonsignificant zero-order correlation between D-PFCA and
job performance was found in the sample of skilled and semi-
skilled workers. This result appears attributable to a combination
of factors: contamination by age and experience, attenuation for
unreliability, and, perhaps, the fact that simpler jobs are less
dependent on cognitive ability. As in Study 3, a significant nega-
tive correlation was found between D-PFCA and age, as well as a
near-significant correlation between D-PFCA and years of educa-
tion. In addition, the skilled and semiskilled workers in this study
varied substantially in years of experience, which was related to
job performance. When age, years of education, and experience
were controlled for, the partial correlation between D-PFCA and
performance was a significant pr .21.
The correlation between D-PFCA and performance found in this
study was significantly lower than the correlation found in Study 3
(Z2.11, p.03, two-tailed). The difference may be explainable
largely in terms of job complexity. Study 3 focused on predicting
performance for medium-complexity and professional–managerial
jobs, whereas Study 4 focused on low-complexity jobs. Higher com-
plexity jobs are associated with a higher range of performance output
Table 8
Descriptive Statistics and Intercorrelations for Performance Ratings and Job Experience for Study 4
Variable NMSD Range 123456
1. PERF98 64 14.93 1.49 11.80–18.40
2. PERF99 79 14.99 1.64 10.40–19.20 .82
3. PERF00 93 15.12 1.58 11.80–19.07 .67
4. PERF01 71 15.24 1.58 10.40–19.20 .64
5. SUPPERF 94 14.98 1.49 10.40–18.44 .88
6. EXPER 96 7.61 6.78 0–27 .25
Sample size for correlations
1. PERF98
2. PERF99 64
3. PERF00 64 79
4. PERF01 47 60 70
5. SUPPERF 64 79 93 71
6. EXPER 64 79 93 71 94
Note. PERF98, PERF99, PERF00, and PERF01 supervisor-rated job performance for 1998, 1999, 2000, 2001, respectively; SUPPERF average of
annual performance ratings for Years 1998 to 2001; EXPER number of years of job experience.
p.05, two-tailed.
p.01, two-tailed.
Table 9
Descriptive Statistics and Intercorrelations for D-PFCA, Age, Experience, Education, and Job Performance for Study 4
Variable NMSD Range 1 2 3 4 5
1. D-PFCA 96 0.57 0.53 1.94–1.11
2. Age (years) 95 38.25 9.73 20–58 .23
3. YEARSEDU 93 12.57 1.32 10–17 .18 .22
4. EXPER 96 7.61 6.78 0–27 .15 .11 .07
5. SUPPERF 94 14.98 1.49 10.40–18.44 .12 .06 .07 .30
Sample size for correlations
2. Age (years) 95
3. YEARSEDU 93 93
4. EXPER 96 95 93
5. SUPPERF 94 93 91 94
Note. D-PFCA dorsolateral prefrontal cognitive ability; YEARSEDU years of education; EXPER years of experience on the job; SUPPERF
supervisor-rated job performance averaged across Years 1998 –2001.
p.05, two-tailed.
p.01, two-tailed.
than lower complexity jobs, and the predictive validity of cognitive
tests for higher complexity jobs (.51–.58) tends to be higher than that
for lower complexity jobs (.38; Hunter & Schmidt, 1996).
In recent meta-analytic studies, the construct-level (corrected)
validity for Conscientiousness was .22 (Barrick & Mount, 1991),
.23 (Salgado, 1997), and .17 (Hurtz & Donovan, 2000), whereas
the mean observed correlations between Conscientiousness and
job performance were .13 (Barrick & Mount, 1991), .09 (Salgado,
1997), and .10 (Hurtz & Donovan, 2000) for the specific occupa-
tional category of skilled or semiskilled labor. In Study 4, in which
the sample consisted primarily of skilled and semiskilled workers,
the measured correlation between Conscientiousness and
supervisor-rated job performance was .23. Thus, this study under-
scores the importance of Conscientiousness for skilled and semi-
skilled job performance.
Overall, for the relatively low-complexity work completed by
these skilled and semiskilled workers, experience appears to be an
important determinant of performance level, with D-PFCA and
Conscientiousness each independently and incrementally contrib-
uting to performance.
D-PFCA and Predicting Performance
In combination, Studies 1 and 2 demonstrate the importance of
D-PFCA and Conscientiousness in predicting academic perfor-
mance. Together, D-PFCA and Conscientiousness appear to pre-
dict academic performance in very competitive environments as
well as IQ and gand independent of both. These results suggest
that a nontrivial slice of the variance in academic performance
might be accounted for using these alternative measures. In addi-
tion, Studies 1 and 2 also help clarify the relationship between
D-PFCA and intelligence as traditionally understood (in terms of
the correlation between D-PFCA and IQ and SAT, but also with
regard to the ability of D-PFCA to predict academic performance,
the traditional criterion for intelligence testing). In the process,
they demonstrate that intelligence can be thought of in part in
terms of working memory, conditional associative learning, and
word fluency—putative aspects of dorsolateral cognitive prefron-
tal functioning—as well as, or in addition to, more traditional
psychometric concepts.
The most outstanding feature of Studies 3 and 4 was the suc-
cessful generalization of the finding that D-PFCA predicts real-
world performance from the clinical and academic to the
industrial– organizational context. Among managers and adminis-
trators, D-PFCA and performance were related at uncorrected rsof
.40 –.60 and corrected rsof.50 –.70. These are large effects
(Rosenthal, 1990), with equally large potential economic implica-
tions. Schmidt and Hunter (1998) offered a formula for the calcu-
lation of productivity increase as a consequence of use of psycho-
metric testing, using their previous work on output variability as
the essential basis for the calculation (Hunter, Schmidt, & Judi-
esch, 1990). According to their formula, the addition of D-PFCA
testing to hiring in a company using unstructured interviews
(which have a corrected validity coefficient of r.38) for em-
ployees making $75,000 per annum would result in a productivity
increase of 33% of salary—$25,000 per year and $125,000 over
the 5 years that the typical individual stays in a typical high-
complexity position (Schmidt & Hunter, 1998; assuming a one out
of 10 hiring rate, a high-complexity job, and an uncorrected r
.57 between multiyear ratings of managerial–administrative per-
Conscientiousness and Performance
Conscientiousness has previously been shown to predict aca-
demic (Goff & Ackerman, 1992; E. K. Gray & Watson, 2002;
M. G. Rothstein et al., 1994) and job performance (Barrick &
Mount, 1991; Hurtz & Donovan, 2000; Salgado, 1997). Such
analyses have revealed a mean predictive validity for Conscien-
tiousness of r⫽⬃.20. Although Conscientiousness was not asso-
ciated with performance in Study 3, in Study 4, Conscientiousness
and supervisor-rated job performance were correlated at r.23,
and in Studies 1 and 2, Conscientiousness was related to academic
performance over and above IQ and D-PFCA.
It is likely that Conscientiousness exerts its effect on academic
performance indirectly. Theoretically, it is Openness that has been
associated with intelligence (L. R. Goldberg, 1990), and Openness
has been found to correlate positively with IQ in past research
(John & Srivastava, 1999). This result was replicated in Studies 1
and 2. However, in Studies 1 and 2, it was Conscientiousness, not
Openness, that was correlated with academic performance. There
is evidence that the effects of Conscientiousness on job perfor-
mance are mediated by goal setting (Barrick, Mount, & Strauss,
1993; Gellatly, 1996), autonomy (Barrick & Mount, 1993), and
motivation (Barrick, Stewart, & Piotrowski, 2002). Given the lack
of relationship between Conscientiousness and D-PFCA and in-
telligence, it is possible that Conscientiousness exerts its effect on
performance indirectly through self-management. It is possible
that Conscientiousness reflects a motivational tendency, or desire,
to be organized, thorough, planful, efficient, responsible, reliable,
dependable, and so on (trait-adjective list from John & Srivastava,
1999), whereas D-PFCA may reflect the ability to make effective
plans, to initiate them and regulate them, to inhibit distracting
activities and stimuli, and to monitor outcomes in the pursuit of
some posited goal state.
In each study, personality scores deviated from their theoretical
orthogonality. This was particularly pronounced in Studies 3 and 4, in
which the FDTI was used to predict job performance. Although the
FDTI was characterized by relatively high scale intercorrelations
during standardization and that may point to some degree of meth-
odological flaw in its construction, it also appears likely that in
employment contexts, participants may be both motivated and able to
present themselves as conscientious, emotionally stable, agreeable,
extraverted, and open-minded–intelligent—that is, as personality
types that might seem ideal to employers. This suggestion appears
consistent with the statistically significant correlation of personality
with self-rated performance found in Study 3 (r.23), as opposed to
the nonsignificant association found between personality and
supervisor-rated performance (r.08) found in the same sample. In
Studies 1 and 2, there was a strong relationship between Conscien-
tiousness and academic performance, in which presentational pres-
sures were essentially absent.
Overall, these results add credence to the notion that personality
might be usefully assessed in academic and industrial–
organizational settings. The possibility that self-presentation bias
reduces its validity should not be dismissed. Although recent
articles have suggested that correcting for presentational bias using
putative measures of response validity does not appear useful
(Hough, Eaton, Dunnette, Kame, & McCloy, 1990; Ones, Viswes-
varan, & Reiss, 1996), it still may be necessary to develop Big Five
tests that are not susceptible to self-presentation effects.
D-PFCA and the Hierarchical Model of Psychometric
Duncan (Duncan, Burgess, & Emslie, 1995) has already at-
tempted to establish the relationship between dorsolateral prefron-
tal processing and Spearman’s g, using methods different from
ours. First, he demonstrated that individuals with frontal lobe
damage are susceptible to goal neglect (Duncan et al., 1996),
which is relatively more likely to occur under conditions of nov-
elty or ambiguity, and to attendant task failure. He also demon-
strated that individuals with low scores on Cattell’s Culture-Fair
Test (an intelligence test with a gfactor loading of around .80;
Jensen, 1980, p. 650) are also susceptible to this failing. Finally,
using PET, Duncan et al. (2000) have also shown that participants
are characterized by increased activation in dorsolateral prefrontal
cortex when solving highly g-loaded problems.
The paradox of sustained IQ scores after severe frontal lobe
damage (Stuss & Benson, 1986) can, in Duncan’s opinion, be
resolved by drawing on Cattell’s (1987) distinction between fluid
and crystallized intelligence (Duncan et al., 1995). According to
this view, the WAIS primarily measures crystallized intelligence,
whereas Cattell’s Culture-Fair Test primarily measures fluid intel-
ligence. In a sample of three patients with high IQ scores after
frontal lobe damage, Duncan and his colleagues demonstrated
relatively poor performance on Cattell’s Culture-Fair Test, permit-
ting Duncan et al. (1995) to argue that fluid intelligence is dimin-
ished in the frontal patients, whereas crystallized intelligence
(WAIS IQ) is preserved. Taken together, these findings support the
argument that dorsolateral prefrontal cortex is the neural basis for
Spearman’s g, or general intelligence.
However, there are a number of conceptual problems with this
position. Psychometric gis as conceptually delicate as it is empir-
ically robust, and although Duncan has advanced the understand-
ing of intelligence, he nonetheless appears to have made a con-
ceptual error in equating Cattell’s Culture-Fair Test scores with g.
Psychometric gis a statistical regularity, reflecting the positive
intercorrelations among all tests of mental ability. Although Cat-
tell’s Culture-Fair Test and RPM are highly g-loaded tests, neither
is required to locate g. Psychometric gcan be located quite
accurately using the WAIS (Jensen, 1998)—performance on
which, as Duncan et al. (1995) argued, may not be critically
dependent on the dorsolateral prefrontal cortex. Jensen (1998, p.
91) estimated that the gloading of a standard IQ test (e.g., the
WAIS) is around .80 —a figure identical to that of Cattell’s test.
Consider for a moment the likely performance of a large sample
of prefrontal patients such as those used by Duncan et al. (1995)—
that is, patients who are impaired on Cattell’s Culture-Fair Test but
not on the WAIS. Suppose that these patients completed a number
of typical IQ tests, such as the WAIS, the Stanford-Binet, and the
Armed Services Vocational Aptitude Battery. Given (a) that per-
formance on traditional intelligence tests is preserved in the case of
prefrontal damage and (b) that the positive manifold appears
universal, it seems certain that factor analysis of the subtest scores
of all these brain-damaged patients would still allow extraction of
a first, g-like factor. It might also be argued (c) that the patients’
premorbid Cattell’s Culture-Fair Test factor scores would also be
highly correlated with their postmorbid factor scores (that is, that
the rank-ordering of the patients might be roughly preserved,
assuming rough equivalence of damage). If these three supposi-
tions are true, then the gfactor can still easily be located after
partial destruction of the frontal lobes, and the neural basis for g
has not in fact been destroyed. It is therefore no more correct to
equate Cattell’s Culture-Fair Test scores with gthan to equate
WAIS scores with g. It is also worth remembering that perfor-
mance on various intelligence tests correlates with inspection time,
choice reaction time, brain size, EEG characteristics, and possibly
nerve conduction velocity and metabolic activity (see Deary, 2000,
for review)— characteristics that are certainly not specifically de-
pendent on dorsolateral PFC function.
Duncan et al. (2000) did demonstrate that there was more
intense neural activity in the dorsolateral prefrontal cortex when
participants were solving high-grather than low-gproblems. In
that study, however, Duncan et al. used matrix problems and
letter-based problems, all of which required participants to exam-
ine four sets of patterns and choose the pattern that did not fit with
the other three. All items, high or low g, had this form. The
primary difference between the high- and low-gitems was there-
fore the complexity of the problem, or the number of rules and
variables the participant was required to process simultaneously.
This implies that the primary difference between Duncan’s puta-
tive high- and low-gitems was in fact the load placed on working
memory (Baddeley, 1986). Rather than demonstrating that the
dorsolateral prefrontal cortex is the neural basis of Spearman’s g,
Duncan et al. have likely replicated the well-established finding
that the dorsolateral prefrontal cortex is critically involved in
working memory. Although it is clear that any account of intelli-
gence that does not incorporate the concept of working memory is
likely to be incomplete (Ackerman et al., 2005; Kane & Engle,
2002; Kyllonen & Christal, 1990), the broad range of correlates of
gsuggests that it is unlikely that working memory is identical to g.
It should also be recalled that higher gscores are associated with
higher gray matter volume overall, not just with higher gray matter
volume in the frontal cortex (Thompson et al., 2001; Thompson,
Cannon, & Toga, 2002), and that the neural correlates of gprob-
ably extend beyond the prefrontal cortex (J. R. Gray et al., 2003;
MacLullich et al., 2002).
Given all this, how might the relationship between D-PFCA,
intelligence, and academic performance be conceptualized? It ap-
pears to us that the relationship is partly a consequence of the
immediate impact of D-PFCA, particularly with regard to working
memory, and partly a consequence of its delayed, developmental
impact. D-PFCA may reflect individual differences in a student’s
day-to-day and month-to-month ability to plan and organize be-
havior in the pursuit of intrinsically generated goals, allowing for
effective orientation in a novel physical and social environment,
and successful direction of behavior toward the completion of
(academic) goals (Duncan, 1995; Luria, 1980; E. K. Miller, 2000;
Stuss & Benson, 1986). The neuropsychological literature attests
to the importance of the frontal lobes in dealing with novelty (E.
Goldberg, 2001), regulating social behavior (Damasio, 1994), and
initiating and monitoring goal-directed activity (Luria, 1980). As
Jensen (1980, p. 331) pointed out, students in college— unlike
elementary and high school students—are essentially self-directed
and self-organized and do not experience constant external disci-
pline imposed by a parent or a teacher. College students are
therefore required to plan, organize, regulate, and evaluate their
own behavior while striving to attain and maintain high levels of
academic performance.
Additionally, consider something more subtle and developmen-
tal. Individuals with higher D-PFCA may be more capable of
manipulating abstractions, both perceptual and verbal, and then
organizing increasingly more automated and crystallized forms of
intelligence over the course of their developmental history. The
result would be a crystallized intelligence whose magnitude is
correlated with the power of the original programmer (D-PFCA)
but that remains psychometrically distinguishable (Cattell, 1987).
In other words, dorsolateral prefrontal cortex, highly involved in
the processing of novel information (Fletcher et al., 2001; Turner
et al., 2004), may program cognitive abilities to a level of auto-
mation that is no longer physiologically dependent on the prefron-
tal cortex (Sakai et al., 1998; Shariff & Peterson, 2005) but that
remains correlated with it. Such abilities might include vocabulary
breadth and letter recognition speed, for example— highly prac-
ticed skills that, once mastered, are relegated to specialized brain
Figure 1 presents our attempt to provide a provisional model for
the relationship between physiological–neuropsychological and
standard hierarchical models of psychometric intelligence. It is
predicated on the idea that D-PFCA broadens people’s conception
of and ability to measure something roughly equivalent, but not
identical, to fluid intelligence. Each of the levels of overlapping
circles represents a perspective on intelligence informed by differ-
ent data, from the more biologically concrete to the more statisti-
cally abstract. At the most basic level, there exists an array of
interrelated physical phenomena, partially represented at the bot-
tom level of Figure 1 as brain volume/organization, prefrontal
volume/organization, and neural transmission speed (this is in-
tended as a schematic rather than a comprehensive portrayal of
brain function and intelligence). Clearly, total brain volume/
organization and prefrontal volume/organization are both highly
correlated and somewhat separable (MacLullich et al., 2002),
allowing measurements of each phenomenon to independently
pick up some of the variance in function, in addition to the
substantive variance they necessarily share. Brain volume/
organization also appears importantly associated with processing
speed (via axonal diameter; Harrison, Hof, & Wang, 2002). The
logical net consequence of the interrelationship between these
physiological contributors to cognitive function is overlap in terms
of cognitive ability: Larger brains have larger prefrontal cortices;
larger brains with larger prefrontal cortices have thicker axons;
brains with thicker axons are faster; larger brains with thicker,
faster axons have an advantage in terms of cognitive function.
At the second level from the bottom of Figure 1 is a represen-
tation of the potential nature of the profound functional relation-
ship between prefrontal cognitive abilities and the other primarily
cognitive abilities of the brain. In light of the role of the prefrontal
cortex in the construction of novel habits, skills, and concepts, it is
reasonable to view it as the programmer of other brain areas. So,
not only is there inseparable physiological overlap between the
prefrontal cortex and other parts of the brain (in terms of size and
speed) but also there is inseparable functional overlap (except in
special, pathological cases and circumstances): Specific cognitive
abilities, less prefrontally dependent in their mature form, are still
what they are in no small part because of the novelty analysis and
general programming capacity of the prefrontal cortex. What this
means is (a) that more specific and less prefrontally dependent
cognitive abilities will inevitably be contaminated with prefrontal
ability because of the manner in which they were initially pro-
grammed and (b) that measures of nonprefrontally dependent
Figure 1. Intelligence in its putative physiological and psychometric forms and the relationship between those
forms. D-PFCA dorsolateral prefrontal cognitive ability; gSpearman’s g;gf fluid intelligence; gc
crystallized intelligence.
intelligence (crystallized IQ, perceptual function) will inevitably
reflect prefrontal function— even after prefrontal damage sus-
tained in adulthood, if they are measured in adulthood, after a long
developmental history.
The third level from the bottom in Figure 1 is the first of those
in the psychometric and statistical domain. Imagine that the set of
all possible cognitive functions is randomly sampled, with no a
priori consideration given to weighting that sampling to accurately
reflect underlying physiological architecture (Carroll, 1993). All
such randomly selected measures are nonetheless going to be
highly contaminated with prefrontal function or fluid intelligence
because of the physiological and functional overlap between that
brain area and all others, as previously described. This statistical
contamination, inevitable because of the underlying physiological
and functional overlap, finds its representation in the fourth level
from the bottom, in the factor representing fluid intelligence, and
in the fact of the high levels of correlation between all the so-called
stratum II factors. Finally, at the topmost level, gemerges, very
tightly associated, statistically, with fluid intelligence because of
all the overlap of physiology, function, and measurement between
prefrontal cognitive function and general brain function at the
lower levels. Moreover, gemerges, reliably, even if there are no
specific fluid intelligence measures at the narrow, Stratum I mea-
surement level because all the specific subdomains of intelligence
are inextricably contaminated with fluid intelligence or prefrontal
cognitive ability, and a broad sampling of such tests will inevitably
reflect that. We would therefore argue, in keeping with this model,
that the D-PFCA battery used in our studies potentially picked up
variance in academic performance independent of IQ and gbe-
cause it more accurately sampled the D-PFCA functional domain
at a psychometric level than may typically be the case with
standard IQ or gmeasures.
Ackerman, P. L., Beier, M. E., & Boyle, M. D. (2002). Individual differ-
ences in working memory within a nomological network of cognitive
and perceptual speed abilities. Journal of Experimental Psychology:
General, 131, 567–589.
Ackerman, P. L., Beier, M. E., & Boyle, M. O. (2005). Working memory
and intelligence: The same or different constructs? Psychological Bul-
letin, 131, 30 60.
Anastasi, A., & Urbina, S. (1997). Psychological testing (7th ed.). Upper
Saddle River, NJ: Prentice Hall.
Anderson, B. (2003). Brain imaging and g. In H. Nyborg (Ed.), The
scientific study of general intelligence: Tribute to Arthur R. Jensen (pp.
29 –39). Boston: Elsevier.
Baddeley, A. D. (1986). Working memory. New York: Oxford University
Barrick, M. R., & Mount, M. K. (1991). The Big Five personality dimensions
and job performance: A meta-analysis. Personnel Psychology, 44, 1–26.
Barrick, M. R., & Mount, M. K. (1993). Autonomy as a moderator of the
relationships between the Big Five personality dimensions and job
performance. Journal of Applied Psychology, 78, 111–118.
Barrick, M. R., Mount, M. K., & Strauss, J. P. (1993). Conscientiousness
and performance of sales representatives: Test of the mediating effects
of goal setting. Journal of Applied Psychology, 78, 715–722.
Barrick, M. R., Stewart, G. L., & Piotrowski, M. (2002). Personality and
job performance: Test of the mediating effects of motivation among
sales representatives. Journal of Applied Psychology, 87, 43–51.
Binet, A., & Simon, T. (1905). Me´ thodes nouvelles pour le diagnostic du
niveau intellectual des anormaux [New methods for the diagnosis of the
intellectual level of subnormals]. L’Anne´ e Psychologique, 11, 191–244.
Blair, C. (2006). How similar are fluid cognition and general intelligence?
A developmental neuroscience perspective on fluid cognition as an
aspect of human cognitive ability. Behavioral and Brain Sciences, 29,
109 –160.
Boivin, M. J., Giordani, B., Berent, S., Amato, D. A., Lehtinen, S., Koeppe,
R. A., et al. (1992). Verbal fluency and positron emission tomographic
mapping of regional cerebral glucose metabolism. Cortex, 28, 231–239.
Brand, C. R. (1996). The g factor. Chichester, England: Wiley.
Brody, N. (1992). Intelligence (2nd ed.). San Diego, CA: Academic Press.
Brooker, B. H., & Cyr, J. J. (1986). Tables for clinicians to use to convert
WAIS–R short forms. Journal of Clinical Psychology, 42, 982–986.
Carey, D. P. (2005). Executive function or g? Hmmmmm. Cortex, 41,
220 –221.
Carroll, J. B. (1993). Human cognitive abilities: A survey of factor-analytic
studies. New York: Cambridge University Press.
Carver, C. S., & Scheier, M. F. (1998). On the self-regulation of behavior.
Cambridge, England: Cambridge University Press.
Cattell, R. B. (1987). Intelligence: Its structure, growth, and action.
Amsterdam: Elsevier Science.
Conway, A. R. A., Cowan, N., Bunting, M. F., Therriault, D. J., & Minkoff,
S. R. B. (2002). A latent variable analysis of working memory capacity,
short-term memory capacity, processing speed, and general fluid intel-
ligence. Intelligence, 30, 163–184.
Conway, A. R. A., Kane, M. J., Bunting, M. F., Hambrick, D. Z., Wilhelm, O.,
& Engle, R. W. (2005). Working memory span tasks: A methodological
review and user’s guide. Psychonomic Bulletin & Review, 12, 769 –786.
Conway, A. R. A., Kane, M. J., & Engle, R. W. (2003). Working memory
capacity and its relation to general intelligence. Trends in Cognitive
Sciences, 7, 547–552.
Costa, P. T., & McCrae, R. R. (1992). Revised NEO Personality Inventory
(NEO PR-I) and NEO Five-Factor Inventory (NEO-FFI): Professional
manual. Odessa, FL: Psychological Assessment Resources.
Costa, P. T., & McCrae, R. R. (1997). Longitudinal stability of adult person-
ality. In R. Hogan, J. A. Johnson, & S. Briggs (Eds.), Handbook of
personality psychology (pp. 269 –290). San Diego, CA: Academic Press.
Cyr, J. J., & Brooker, B. H. (1984). Use of appropriate formulas for
selecting WAIS–R short forms. Journal of Consulting and Clinical
Psychology, 52, 903–905.
Damasio, A. R. (1994). Descartes’ error: Emotion, reason, and the human
brain. New York: Putnam.
Damasio, A. R., & Anderson, S. W. (1993). The frontal lobes. In K. M.
Heilman & E. Valenstein (Eds.), Clinical neuropsychology (3rd ed., pp.
409 –460). New York: Oxford University Press.
Deary, I. J. (2000). Looking down on human intelligence: From psycho-
metrics to the brain. Oxford, England: Oxford University Press.
Deary, I. J. (2001). Human intelligence differences: A recent history.
Trends in Cognitive Sciences, 5, 127–130.
Deary, I. J. (2005). The principles of cognition and the abilities of man: A
natural collaboration. Cortex, 41, 225–227.
DeYoung, C. G., Peterson, J. B., & Higgins, D. M. (2002). Higher-order
factors of the Big Five predict conformity: Are there neuroses of health?
Personality and Individual Differences, 33, 533–552.
Digman, J. M. (1990). Personality structure: Emergence of the five-factor
model. Annual Review of Psychology, 41, 417– 440.
Duncan, J. (1995). Attention, intelligence, and the frontal lobes. In M. S.
Gazzaniga (Ed.), The cognitive neurosciences (pp. 721–733). Cam-
bridge, MA: MIT Press.
Duncan, J. (2005). Frontal lobe function and general intelligence: Why it
matters. Cortex, 41, 215–217.
Duncan, J., Burgess, P., & Emslie, H. (1995). Fluid intelligence after
frontal lobe lesions. Neuropsychologia, 33, 261–268.
Duncan, J., Emslie, H., Williams, P., Johnson, R., &