1. No category

Why Is Reaction Time Correlated with Psychometric g?
Author(s): Arthur R. Jensen
Source: Current Directions in Psychological Science, Vol. 2, No. 2 (Apr., 1993), pp. 53-56
Published by: Sage Publications, Inc. on behalf of Association for Psychological Science
Stable URL: http://www.jstor.org/stable/20182199 .
Accessed: 26/06/2014 03:35
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].
.
Sage Publications, Inc. and Association for Psychological Science are collaborating with JSTOR to digitize,
preserve and extend access to Current Directions in Psychological Science.
http://www.jstor.org
This content downloaded from 132.210.236.20 on Thu, 26 Jun 2014 03:35:46 AM
All use subject to JSTOR Terms and Conditions
SCIENCE 53
CURRENT DIRECTIONS INPSYCHOLOGICAL
in the brain's
reflect differences
more
features
specialized
design
in specific
that may be involved
are usually
types of ability, which
at
identified
the level of first-order
factors, such as verbal, spatial, and
Is Reaction Time Correlated With
Why
Psychometric g?
Arthur R. Jensen
Itseems almost incredible that in
in reaction time
dividual differences
(RT) in simple tasks that involve no
intellectual content and are so easy
as to be performed by most persons
in less than 1 s should be correlated
com
scores on nonspeeded,
with
reasoning ability, vo
and
cabulary,
general knowledge?
the kinds of content that compose
IQ
in recent years,
tests. Nevertheless,
tests of
plex
RT and IQ
between
an
has become
empirically well es
tablished fact, based on thousands
in scores of studies con
of subjects
the correlation
inmany
laboratories around
ducted
all these
the world.1 What might
studies tell us about the nature of hu
man
Before trying to
intelligence?
it is necessary
this question,
some concepts
to summarize
and
answer
empirical
generalizations
about
the
CONCEPTUAL ISSUESAND
EMPIRICAL
GENERALIZATIONS
should not use the overex
in the
word
intelligence
present context because correlations
between RT and scores on psycho
metric tests reflect only the tests' g
1.We
tended
Arthur R. Jensen is Professor of Ed
in the Uni
ucational
Psychology
at Berkeley.
versity of California
His best known books are Educa
Differences
and Group
bility
Bias in
1973),
(Harper & Row,
Mental Testing (Free Press, 1980),
and Straight Talk About Mental
1981). Address
to Arthur
R.
correspondence
Jensen, School of Education, Uni
versity of California,
Berkeley, CA
94720.
(Free Press,
The most
performance.3
tests
show
the high
highly g-loaded
est correlations with RT, and when g
is statistically removed from the cor
RT and a psycho
relation between
cognitive
metric
test, the partial correlation
is close to zero. So we
coefficient
are really talking about RT-g corre
lation.
2. The g factor per se is a product
in a wide
of individual differences
Itscon
variety of cognitive abilities.
status
and empirical
ceptual
reality
do not depend on any knowledge
of
features of brain structure
Even though we
physiology.
little about these
know exceedingly
design features, we can investigate
about the basis of indi
hypotheses
the design
and
vidual
differences
in g.
to assume
It ismost
key phenomena.2
Tests
factor
factor, that is, the general
common
to all measures of complex
parsimonious
a given
the
that, within
species,
as
main
structural and functional
are
same
of
the
brain
the
for
all
pects
normal
and
individuals,
biologically
in cogni
that individual differences
tive abilities reflect differences
in the
in long-term
stored
information
(LTM) and how that infor
memory
is represented and organized
mation
in LTM. The efficiency
of informa
tion processing
is probably a quan
or ty
not qualitative
titative,
its
whatever
variable,
pological,
or
experiential
biological
origins.
to
Because
the g factor is common
all kinds of cognitive
performance,
it cannot be attrib
however diverse,
or
utable to any specific knowledge
skill components
of performance,
but must essentially
reflect variance
in the speed and efficiency of infor
mation
The g factor is
processing.
such a distillate from a wide variety
activities as scarcely to
of cognitive
Copyright ?
1993 American
Psychological
numerical
factors.
A source of individual differences
with
such broad generality
could
some
be
conceivably
simple, even
unitary, aspect of brain function, or
it could be determined
by a number
of different properties
of the brain
that all enter into every kind of com
plex cognitive
activity. At present,
we
no knowledge
have virtually
even of the probable
number
of
brain mechanisms
alone their precise
involved in g, let
nature. Studies of
the RT-g relationship may be able to
throw some light on this question.
a
3. Simple RT (i.e., following
preparatory signal, the time to react
to the onset of a single
stimulus
nature and location are al
whose
ready known to the subject) has a
quite small negative correlation
(typ
with
-.10)
g. The
ically about
amount of information
processing
for simple RT is very small. More of
in simple RT is attribut
the variance
able to noncognitive
sensorimotor
factors than to central
information
Under conditions
of dis
processing.
in
crimination
and choice, which
volve the speed of retrieval of infor
mation
from short-term
memory
(STM)or LTM, RT (though seldom
more
than 1 s) is very significantly
is also
greater than simple RT?and
more highly correlated with g.
4. In normal young adults,
the
RT-g correlation
begins to diminish
a level of task complexity
beyond
that makes
for RTs greater
than
1 s. There seem to be two
about
main reasons for this U-shaped
rela
tion between
the RT-g correlation
and the degree of task complexity:
First, beyond some optimal point, an
in task complexity
increase
invites
the development
of special strategies
for performing
the task. The various
strategies adopted
by subjects are
less correlated
Society
This content downloaded from 132.210.236.20 on Thu, 26 Jun 2014 03:35:46 AM
All use subject to JSTOR Terms and Conditions
with
g than
is the
54 VOLUME 2, NUMBER 2, APRIL 1993
information
processing
is either using no
everyone
strategy or using the same strategy
speed
when
of
throughout all trials. Second, when
diffi
task complexity
(and hence
a certain point, the
culty) exceeds
subject makes response errors. At or
this "threshold
of break
beyond
in RT
individual differences
down,"
are transformed
into individual dif
in error rates. Hence,
there
ferences
is a reciprocal
between
relationship
the RT-g correlation and the error-g
as a function
of task
correlation,
complexity.
5. RT decreases
in a regular way
to maturity,
from
childhood
age
as
one
would
just
predict from men
rate of brain
tal growth
curves,
in
increase
the myelina
and
growth,
tion of axons from early childhood
ve
to maturity.
(Nerve conduction
with
in the factor analysis sense, that is,
to
variance which
is not common
or
tests
other
tasks
all
any
among
into the factor analy
those entered
tests. And RT is less cor
"power/'
related with scores on speeded
than
on nonspeeded
a test
tests. When
speed factor, independent of g, is ex
relatively
large amount of
non-g variance peculiar to RT tasks
is related to sensorimotor,
rather
tracted
sis. This
it im
factors. Hence,
ceiling on the corre
RT
lation between
any particular
task and a g-loaded
psychometric
than cognitive,
poses a severe
test. This correlation
ferent
in NCV. Moreover,
the information
RT is more re
of
aspect
processing
than the sen
lated to age differences
RT and
sorimotor
Choice
aspect.
forms of RT show a
other complex
has
tasks, however,
correlations with g approaching
than
of RT
of dif
shown
.70,
the average cor
approaches
relation among different IQ tests.
which
locity, NCV, is related to the degree
RT slows with age
of myelination.)
in later maturity, mirroring
the de
tests
cline in scores on psychometric
in
of fluid g, the gradual decrease
the demy
brain weight and volume,
elination of axons, and the decrease
is gen
.50 for
ceiling
erally in the range of .30 to
RT tasks of greater complexity
simple RT. A combination
measures
based on a number
"SPEED"
VERSUS "NOISE"
THEORIES
OF THE
RT-g CORRELATION
are the fewest
independent
to ex
needed
of variance
plain the RT-g correlation? Because
the RT-g relationship will have to be
in neurological
explained
ultimately
seems
most
to
it
terms,
productive
neu
that
involve
suggest hypotheses
Some of these
rological variables.
are testable
hypotheses
neurological
means.
available
by presently
tests and Chronometrie
are factor analyzed
to
measures
RT
the
show
all
up
gether,
only on the g factor and not on the
various group factors (verbal, spa
bat
tial, etc.) of the psychometric
a
RT
But
the
of
tery.
large proportion
on
variance
factors
also shows up
is
The most obvious
hypothesis
that speed of information processing
is the essential
basis of g, and one
only on Chronometrie
fac
tasks; that is, there is a common
tor among
various
Chronometrie
tests that is not shared by psycho
metric tests. Every Chronometrie
task
also has a good deal of uniqueness,
that
load
the test-speed
zero correlation
Speed
basis of speed
possible neurological
is the speed of trans
of processing
mission
through nerve pathways,
which
comprises
both
NCV
and
speed of synaptic transmission.
Itmust be emphasized
that infor
mation processing
speed should not
be confused with overt test-taking
is plenty of evidence
speed. There
tests are generally
that speeded
less
or
than nonspeeded,
g loaded
Published by Cambridge
University
battery
tests,
has virtually
with RT.
to the speed-of
According
processing
theory of the RT-g corre
lation, speed is important because of
the brain's limited capacity
for pro
factor
information. Although
cessing
be
may
multiple
independent
when
resources,
cessing
one's
there
pro
atten
tion is highly focused, as in solving a
and novel problem
that
complex
cannot be handled by automatized
in chan
skills, there is a bottleneck
nel capacity. Also, information com
unit
ing into the central processing
or
called
(often
memory,
working
or
from external
stimuli
from
WM)
LTM is lost rapidly. If all the infor
mation needed for problem solution
is not processed
it is lost, it
before
must
What
sources
much greater age gradient than sim
ple RT.
6. All RT tasks have a great deal
that is, a relatively
of specificity;
is
small proportion of their variance
common
to conventional
psycho
a number of psy
metric tests. When
chometric
(RT) tests
of
from a psychometric
and nonspeeded
speeded
in again by repetition
or repeated retrieval
a cor
from LTM. Hence,
achieving
rect or adequate
is a race
solution
be taken
of the stimulus
between
two
processing
loss, of the
variables:
speed of
or
rate of decay,
information
needed.
and
faster speed of
persons with
have
RTs on ele
faster
processing
tasks
than
persons
mentary cognitive
with slower processing
speed, and
can also acquire
and
knowledge
skills faster,
retrieve
information
reason
from LTM more efficiently,
more
and
solve
better,
complex
tests.
problems on mental
Thus,
The crucial question,
then, is this:
Does speed of information process
ing reflect NCV in the brain? If so,
we
be
should
find a correlation
and scores on g-loaded
tests. Vernon
and
psychometric
a
found
Mori4
correlation
(about
+ .40) between
peripheral NCV (in
tween NCV
nerve of the forearm)
sam
IQ, in two independent
Reed and I,5 in a
ples. However,
very similar study based on a larger
did not find a significant
sample,
the median
and
Press
This content downloaded from 132.210.236.20 on Thu, 26 Jun 2014 03:35:46 AM
All use subject to JSTOR Terms and Conditions
CURRENTDIRECTIONS IN PSYCHOLOGICALSCIENCE 55
The reason for the discorrelation.
cr?pant results has not yet been dis
covered.
Reed and I6did find a significant
= +
correlation
(r
.27, corrected for
restriction of range in IQ = +.37)
between
IQ and NCV in the visual
tract going from the retina to the vi
sual cortex
in 147 college
males.
is a part of the central
This pathway
nervous
unlike peripheral
system,
of
and the characteristics
nerves,
nerve fibers in the visual tract are
much
more
brain
centers
like those
involved
in the higher
in complex
information
processes.
are
The fact that the correlations
in nerve tracts that register
found
evoked
responses some 200 to 300
ms before the neural impulses have
is
reached the higher brain centers
that the correlation
be
evidence
tween NCV and g is a bottom-up
and not a top-down
relationship.
is, the higher mental processes
do not
reflected in test performance
in the visual tract,
influence NCV
but nerve fibers in the visual tract
That
similar to those in
properties
NCV in
the higher centers. Hence,
tract and NCV
in the
the visual
are
centers
association
posi
higher
have
studies
tively correlated. Replication
of these findings are now under way.
If the results hold up, then variance
a
in NCV will have to be considered
basic component
"Noise"
of psychometric
in Neural
g.
trials, is generally more highly cor
related (negatively) with g than is the
mean or median RT over n trials. In
traindividualvariability inRT (RTSD)
is highly correlated with median
but has much
lower split-half
test-retest
reliability.
Eysenck7
argued
than NCV,
or
noise,
in neural
errors,
transmis
sion of information in the brain. This
I have
idea seems highly plausible.
a
"neural
construct,
hypothesized
to explain RTSD.9 Ihy
oscillation,"
a longer period of os
that
pothesize
cillation
(i.e., slower oscillation)
for larger RTSD and lower g.
Noise and oscillation
could be caus
or they
related
phenomena,
ally
could even be one and the same
makes
thing.
The
is whether
critical
question
RT and RTSD are simply different in
dices of the same basic process.
to this
Much
evidence
pertinent
question has now been analyzed. RT
and RTSD are correlated only about
.75 after correction
for attenuation
error), indicating
(i.e., measurement
that they do not reflect one and the
same
process.
when
Moreover,
one
the two variables
is statistically
seen
it
is
that RT and
controlled,
RTSD are independently
correlated
of
with g, and the (partial) correlation
is
RTSD
for
than
for
There
RT.10
larger
fore, there seem to be at least two
of g: vari
independent components
ance in NCV and variance
in neural
oscillation
underlies
(or whatever
mechanism
RT,
and
has
in RTSD, rather
is the primary or most
Also, the concept of capacity of
WM
to account
is needed
for the
is the active part
RT-g relation. WM
of STM. There is now evidence
that
RT tasks become more g loaded as
they tend to strain the capacity of
the
WM,
yet not strain it beyond
threshold of breakdown
and loss of
information.
Tasks at the level of
at which
complexity
breakdown
(hence
tion or inadequate
the threshold of
incorrect solu
Copyright ?
Psychological
1993 American
are the best measures
of an
level of g.
relationship can be demon
individual's
This
strated with Chronometrie
tasks by
means of a dual-task paradigm.
For
the subject may be asked
example,
a series of five digits
to memorize
shown for 3 s on a computer moni
tor. Immediately after the digits van
ish, two letters appear. The subject
responds as quickly as possible
by
labeled
pressing one of two keys
"YES" or "NO" to indicate if the let
ters are the same.
a
Immediately,
single probe digit appears, and
subject must respond as quickly
the "YES"
possible
by pressing
"NO" key to indicate whether
the
as
or
the
probe digit was or was not included
in the set of five digits previously
RT on each of these dual
presented.
tasks
is longer than when each task
is presented
Even more
separately.
the
of g with
correlation
interesting,
RT on each task is slightly but con
sistently greater in the dual-task than
in the single-task condition.
The oc
cupation of WM by the digit series
causes a longer RT in the interposed
discrimination
task, and individual
in this increment
differences
in RT
in g. Thus, we
an
that
of
may conclude
explanation
the RT-g correlation
also requires a
inWM capac
concept of variance
reflect
differences
notion of capacity
is more
to handle
in neurological
difficult
terms than is either speed or oscilla
tion. Speed may be equated with
NCV and oscillation
with the peri
in the synchronized
action
odicity
CAPACITYOF
WORKING MEMORY
response)
reached
ity.
The
RTSD).
Transmission
in informa
The idea of "noise"
tion processing was suggested by the
finding that intraindividual variabil
as the individu
ity in RT, measured
al's standard deviation of RT over n
that variance
and that the
I basic phenomenon,
in RT ismerely a
correlated variance
of individual
differ
consequence
ences
in RTSD.8 Theoretically,
Ey
senck views RTSD as an index of
is
of large groups of neu
potentials
rons. Psychologists
of the Erlangen
in Germany,
school
have
however,
formulated the notion of capacity of
WM
in terms of neurological
mech
that can be measured
inde
that
argue
capac
They
pendently.11
as bits
of
ity (C), measured
anisms
is the product of speed
information,
transmission
(S) of information
(or
in
measured
bits
NCV),
per second,
and the duration
(D) of the neural
traces of information, measured
in
Society
This content downloaded from 132.210.236.20 on Thu, 26 Jun 2014 03:35:46 AM
All use subject to JSTOR Terms and Conditions
56 VOLUME 2, NUMBER 2, APRIL 1993
the capacity
seconds.12
Thus,
as
WM can be expressed
C bits =
of
S bits/s x D s.
adds yet another basic
D, to the theory of the
component,
of g. Because
causal underpinnings
two somewhat
ele
independent
ments, S and D, are involved in ca
This formula
pacity, we should expect C to be
more highly correlated with g than
of
speed (or RT) alone. Correlations
.67 and .88 have been reported be
measure
tween the experimental
of
C and
on a highly g-loaded
test in two large samples
scores
vocabulary
of adults.10
of the
The proposed
hypothesis
of psy
underpinnings
neurological
es
chometric
g and its empirically
tablished correlation with RT com
sources
three
of
basic
prises
(a) speed of information
related to nerve con
transmission,
or
duction velocity;
(b) oscillation,
variance:
processing,
working memory,
reflecting the rate
of decay of neural traces that origi
nated from external stimuli or from
encoded
neurally
brought
term
information
up momentarily
from
long
memory.
This working
theory is consistent
on
present empirical evidence
RT and g,
the correlation
between
as
the neurological
and although
with
the
pects are still largely speculative,
seem
mechanisms
hypothesized
in
of continued
worthy
empirical
It seems very improba
vestigation.
ble that the proposed hypothesis will
turn out to be wholly
correct. The
correct theory, however,
when
fi
established
evi
nally
by empirical
dence, will probably not be very dif
SUMMARY
intraindividual
of activation
thresholds
periodicity
of groups of neurons; and (c) dura
tion of neurally encoded
information
or
in immediate
consciousness,
in speed of
variability
related to synchronous
ferent.
Notes
1. P.A. Vernon, Ed., Speed of Information Pro
(Ablex, Norwood,
NJ,
cessing and Intelligence
1987).
2. For a more detailed explication and citations
of specific studies, see A.R. Jensen, Understanding g
in terms of information processing, Educational Psy
(1992).
chology Review, 4, 271-308
3. A.R. Jensen, Commentary: Vehicles of g, Psy
(1992).
chological Science, 3, 275-278
4. P.A. Vernon and M. Mori, Intelligence, reac
tion times, and peripheral nerve conduction veloc
(1992).
ity, Intelligence, 16, 273-288
5. T.E. Reed and A.R. Jensen, Arm nerve con
duction velocity (NCV), brain NCV, reaction time,
and intelligence, Intelligence,
(1991).
15, 33-47
ve
6. T.E. Reed and A.R. Jensen, Conduction
a
in
nerve
brain
locity
pathway of normal adults
correlates with intelligence,
16, 259
Intelligence,
272 (1992).
7. H.J. Eysenck, Intelligence and reaction time:
The contribution of Arthur Jensen, inArthur Jensen:
Consensus
and Controversy,
S. Modgil
and C.
Modgil, Eds. (Falmer, New York, 1987).
8. There ismuch less variance among persons in
their fastest RTs than their slowest RTs. The distri
bution of any person's RTs is skewed to the right
(i.e., toward longer RT). Those persons whose RTs
have the most skewed distributions consequently
show the largest RTSDs and the longest mean (or
median) RTs, and they also have relatively lower
IQs, on average. Thus, it is easier to explain RT
differences in terms of differences
in RTSD than the
other way around. In any case, the phenomenon of
in RTSD and its correlation
individual differences
with g is firmly established. See note 10.
9. A.R. Jensen, Reaction time and psychometric
g, in A Model for Intelligence, H.J. Eysenck, Ed.
(Springer-Verlag, Berlin, 1982).
10. A.R. Jensen, The importance of intraindivid
ual variability in reaction time, Personality and In
dividual Differences,
13, 869-882
(1992).
11. S. Lehrl and B. Fischer, The basic parame
ters of human information processing: Their role in
the determination
of intelligence, Personality and
Individual Differences,
9, 883-896
(1988); S. Lehrl
and B. Fischer, A basic information psychological
parameter (BIP) for the reconstruction of concepts of
4,
European Journal of Personality,
intelligence,
259-286(1990).
12. D was measured by forward digit span and
letter span, but Iwould prefer that D be measured
chronometrically,
by the increase in RT in dual as
compared with single RT tasks; that is, as the
strength of a stimulus trace diminishes over time, the
RT for scanning the trace increases, and its rate of
increase can index D.
Filling in Gaps in Perception: Part II.
Scotomas and Phantom Limbs
suai field. When
they look at a col
ored wall or a regular pattern of any
kind (e.g., a carpet or a tile floor),
V.S.
surrounding
they gaze
In this article, Ipresent some find
ings that suggest we need to radi
cally revise two of the basic con
the concept
cepts in neuroscience:
of the receptive field as a set of re
in information
ceptors
funneling
onto single sensory neurons and the
fixed
topography,
in the adult brain.
"maps,"
in this area
interest
My
idea
the scotoma
Ramachandran
of
or
began
over
15 years ago when, as a student
I encountered
in neurology
clinics,
in the
with
focal
lesions
patients
usu
visual cortex.
Such patients
as
a
is
have
what
described
ally
scotoma1?a
field within
in the visual
region
which
nothing can be
consciously
perceived.
Remarkably,
the patients themselves are often un
aware of this gaping hole in the vi
Published
by Cambridge
University
V.S.
in" by the
gets "filled
color or pattern. Or
if
seen
at a companion
Ramachandran,
M.D.,
Ph.D.,
is Professor with the Neuroscience
of
Program and the Department
of
Califor
Psychology,
University
corre
Address
nia, San Diego.
to
V.S.
Ramachandran,
spondence
Brain & Perception
Lab, Psychol
0109, University
ogy Department
San Diego,
of California,
La Jolla,
CA 92093.
Press
This content downloaded from 132.210.236.20 on Thu, 26 Jun 2014 03:35:46 AM
All use subject to JSTOR Terms and Conditions