You are on page 1of 18

INTELLIGENCE 7, 53-70 (1983)

Speed of Information Processing and


General Intelligence*

PHILIP A . VERNON
University of Western Ontario

This study investigates the relationship between a number of measures of speed of


cognitive information-processingand intelligence test scores. One hundred university
students were given five tests of speed-of-processing, measuring their speed of
encoding, short-term memory scanning, long-term memory retrieval, efficiency of
short-term memory storage and processing, and simple and choice reaction time or
decision-making speed. They were also given the Wechsler Adult Intelligence Scale
and the Raven Advanced Progression Matrices. A number of multiple regression anal-
yses show that the cognitive processing measures are significantly related to IQ
scores. Other analyses indicate that this relationship cannot be attributed to the com-
mon content shared by the reaction time and the intelligence tests, nor to the fact that
parts of the WAIS are timed. It is concluded that the reaction time tests measure basic
cognitive operations which are involved in many forms of intellectual behavior, and
that individual differences in intelligence can be attributed, to a moderate extent, to
variance in the speed or efficiency with which individuals can execute these opera-
tions.

In the past few years, a number o f researchers have demonstrated that a positive
relationship exists between the speed with which individuals can perform differ-
ent cognitive processes and the scores they obtain on a variety o f measures o f
intelligence. The specific cognitive processes that have received most attention
include encoding or inspection time (Lally & Nettelbeck, 1977; Nettelbeck &
Lally, 1976); scanning information in short-term memory (Dugas & Kellas,
1974; Harris & Fleer, 1974; McCauley, Dugas, Kellas, & DeVellis, 1976);
retrieving information from long-term memory (Hunt, 1976); and simple and
choice reaction time or speed o f decision-making (Carlson & Jensen, in press;
Jensen, 1979a, 1979b, 1980, 1981, 1982; Vernon, 1981a).
Much o f the work in this area has focused on the relationship between one or
another of these measures of speed of cognitive processing and intelligence, and
the correlations obtained have ranged from quite low to quite high (i.e., about
.30 to .80). To date, few studies have investigated how much of the variance in
intelligence test scores they might collectively account for if measures were ob-

*This study was supported by the Institute for the Study of Educational Differences.

53
54 VERNON

tained on each of them from one group of subjects. Given the range of values of
the correlations between each one---when measured by itself and intelligence,
it might be anticipated that a multiple R including all of them would be quite
substantial (see Keating & Bobbitt, 1978). This would depend on how highly
intercorrelated the speed-of-processing variables are or how much unique vari-
ance each one shares with intelligence. This study has attempted to resolve these
questions by measuring each of the information-processing variables mentioned,
and others described below, and examining the intercorrelations that exist be-
tween them and the multiple R's their parameters collectively produce with
measures of intelligence.
One shortcoming of speed-of-processing or reaction time research is that no
theory has satisfactorily explained why the observed relationships between dif-
ferent speed-of-processing variables and intelligence exist. Typically, the stimuli
employed in the reaction time tests have not required subjects to draw on the
kinds of information or reasoning skills tapped by tests of intelligence. Some,
such as those used to measure speed of short-term memory (STM) scanning and
speed of retrieval of information from long-term memory (LTM), have consisted
of highly ovedearned stimuli, such as the letters of the alphabet and digits. Oth-
ers, like the speed of encoding or inspection time task, and Jensen's speed of
decision-making test, have measured, respectively, how quickly subjects can
perceive and discriminate between the lengths of two lines and how quickly they
can react to an increasing number of choices represented by an array of lights.
Clearly, the speed-of-processing tests and intelligence tests have very little
surface content in common, other than that both often involve printed
alphanumeric stimuli, yet they are correlated. One way to view their relationship
is to consider that the speed-of-processing tests are measuring the efficiency with
which persons can perform very basic cognitive operations which are themselves
involved in, or which underlie, other kinds of cognitive and intellectual behav-
ior. Further, if it is accepted that these cognitive operations are carried out in
some sort of short-term or working memory system, characterized by a limited
capacity to hold information, is a rapid decay or loss of information in the ab-
sence of rehearsal, and is a trade-off between the amount of information that can
be held and processed simultaneously, then the speed or efficiency with which
individuals can execute the cognitive operations involved in a given task or prob-
lem might be expected to have a considerable effect on the success of their per-
formance of the task.
These properties of the short-term, working memory system-limited capacity,
rapid decay or loss of information, and a storage-procesing trade-off-are well
documented in memory research. It is also not unreasonable to suggest that such
cognitive operations are encoding, STM processing, and LTM retrieval are inte-
gral components of probably all forms of intellectual behavior. This is not to sug-
gest that they are the only components, though it seems likely that they are
among the most basic and pervasive. Given this, it is proposed that individual
differences in intelligence may, to some extent, be the result of differences in the
SPEED OF INFORMATIONPROCESSINGAND GENERALINTELLIGENCE 55

efficiency or speed with which individuals can perform these basic components
of information-processing. The study and results described below go some way
toward testing the validity of this proposition.

METHOD

Subjects and Measuring Instruments


Subjects were 100 university students recruited by advertisements in the uni-
versity newspaper. Thirty-five were male, 65 female. Their ages ranged from 18
to 34 years (mean, 21.4; SD, 2.63). Subjects were paid for participating in the
study. They were given five measures of speed or efficiency of cognitive proc-
essing; the Wechsler Adult Intelligence Scale and the Raven Advanced Progres-
sive Matrices. The variables were measured as follows; the code name by which
each speed-of-processing variable is subsequently labelled appears in parenthe-
ses after its name:

Speed of Encoding or Inspection Time (IT).


Subjects were presented with images of a long and a short vertical line, con-
nected at the top by a horizontal line, for very brief exposure durations by means
of a tachistoscope. Long lines were 1.5 inches in length, short lines, 1 inch. The
distance between the lines (i.e., the length of the top horizontal bar) was .5 inch.
The lines were less than 1/16 inch thick. One image had the long line on the
right, the other on the left; the subjects' task was to state which side they thought
the long line was on after each exposure of an image. The images were presented
in a random order for durations ranging from 300 milliseconds to about 50
milliseconds. Preceding each exposure, a small dot appeared in the center of the
screen at which subjects were looking, to provide them with an area upon which
to focus their eyes. Immediately following each exposure of an image, a masking
image--consisting of two equally long (2 inches), thick (.25 inch) vertical lines,
connected at the top by an equally thick horizontal bar--appeared. This mask
completely covered the test images. Exposures of test images were continued for
shorter and shorter durations until a subject was no longer able to perform cor-
rectly on 9 out of l0 trials. Their score was the fastest exposure duration at which
they could maintain this criterion.

Speed of Processing (Scanning) Information in STM (DIGIT).


Subjects were presented with a string of digits, from 1 to 7 digits in length, on
the screen of a response console connected to a Rockwell AIM 65
microcomputer. The digits appeared for 2 seconds. After a l-second interval, a
single probe digit appeared and the subjects' task was to indicate as quickly as
they could whether or not the probe had appeared in the series they had just seen.
Subjects held the index finger of their preferred hand on a central home button on
56 VERNON

the response console and responded by moving their hand and pressing down one
of two other buttons (marked " y e s '* and " n o " ) equidistant from the home but-
ton. The computer measured and recorded the speed and correctness of their re-
sponses. After each response, the words "correct" or "incorrect" appeared on
the subjects' screen.
A subject's reaction time was measured as the time, in milliseconds, from the
appearance of the probe digit to when he lifted his finger from the home button.
The time to move his hand and press a response button was measured separately
as his movement time.
Eighty-four digit strings were presented (each string length and response
type---i.e., 1 digit-yes response, 1 digit-no response . . . t o . . . 7-yes, 7 - n o - -
was presented six times) in random order with respect to their length and whether
they required a " y e s " or a " n o " response. Within the strings of two or more
digits, no digit was repeated. For those strings requiring a " y e s " response, the
position of the digit which matched the probe was randomly determined.

Speed of Retrieving Information From LTM (SD2 and SA2).


Subjects were presented with 100 pairs of common words ~ on the screen of
the same response console described above. The words were either " s a m e " or
"different" depending on two criteria. In one test (SD2), the words were liter-
ally (or physically) same or different (e.g., D O G - D O G or DOG-LOG). Words in
this test ranged from 2 to 8 letters in length; word-pairs that were different were
always of the same length. In a second test (SA2), the 100 pairs of words were
either synonymous or antonymous, and word-pairs were not necessarily of the
same length. Subjects were told before each test the relevant criterion to apply in
making their same-different judgments. Subjects responded in the same manner
described above---the response buttons now being labelled " s a m e " or "differ-
ent". Again, and in all other tests performed with this equipment, " c o r r e c t " or
"incorrect" appeared on the screen after each response. A complete list of the
word-pairs composing SD2 and SA2 is reported in Vernon (1981b).

Efficiency of STM Storage and Processing (DT2 and DT3).


This test is essentially a combination of the two previous tests. Subjects were
presented with a short string of digits (1 to 7 digits) which they were told to re-
hearse for later recall. The digits appeared on the response console screen for 2
seconds and then flashed off. After a random interval of between 2 to 4 seconds,

JUsing Thorndike and Lorge (1944), the average frequencyof the words in test SD2 (and DT2) is
67.5 per 1,000,000 (or "A" by their classification). 67.1% of the words in this test are rated "AA"
or "A". The average frequency of the words in test SA2 (and DT3) is 63.23 per 1,000,000. 64% of
its words are rated "AA" or "A". In computingaverage frequencies, "A" and "AA" words were
counted as exactly 50 and 100 per million, respectively, so these figures are undoubtedly
underestimates.
SPEED OF INFORMATIONPROCESSINGAND GENERALINTELLIGENCE 57

a pair of words appeared on the screen to which subjects responded " s a m e " or
"different" by pressing the appropriate response button. Immediately after re-
sponding, they returned their hand to press down the home button. After a
1-second interval, a single probe digit appeared on the screen and subjects now
responded " y e s " or " n o " to indicate whether the probe had been a member of
the string of digits that appeared before the words.
In one test (DT2), 50 pairs of words (and 50 digit strings) were presented, and
all the word-pairs were physically same or different. They were selected from
among the word-pairs that had been used in SD2. In another test (DT3), the 50
word-pairs were synonymous or antonymous, and had been used in SA2.

Simple and Choice Reaction Time or Speed of Decision-Making (RT).


Subjects were seated before a black panel, at the lower center of which was a
red home button. Eight more red pushbuttons were arranged in a semicircle
around the panel, equidistant from the home button. Half an inch above each of
the eight pushbuttons was a small green light. Three different frames could be
placed over the panel so as to expose either 1, 2, 4, or 8 light-button combina-
tions and to vary the number of choice-alternatives subjects had to attend to.
Subjects were instructed to press down the home button with the index finger
of their preferred hand and to watch for one of the green lights to come on. When
a light appeared they were to press the pushbutton directly below the light as
quickly as possible. Subjects' reaction times and movement times (as defined
previously) were recorded in milliseconds by two electronic timers. Subjects
were given 15 trials at each of the four light-button combinations. On choice
trials (i.e., those involving more than one light), the particular light that come on
was random. Lights appeared at a random interval of between 0 to 2 seconds
after a warning signal was sounded.

Measures of Intelligence.
The measures used were the Wechsler Adult Intelligence Scale (WAIS) and
the Raven Advanced Progressive Matrices. Subjects were given the standard in-
structions for these tests and were allowed as long as they needed to finish the
Raven Matrices.

Procedure
With the exceptions described below, subjects were tested individually in
three sessions, each lasting approximately 1 I/2 hours, by one of two research as-
sistants. Subjects received the following tests in this order:

- - Session 1: WAIS, DIGIT, IT


- - Session 2: SD2, DT2, Raven Matrices
- - Session 3: SA2, DT3, RT, IT
58 VERNON

(Note: IT, the encoding or inspection time test, was administered twice to 50
subjects in order to obtain data with which to estimate test-retest reliability).
Due to a series of technical problems with the Rockwell computer and the
tachistoscope, some subjects were not tested in the above order. These subjects
were given the WAIS, Raven, and RT tests in two sessions and, when the com-
puter and tachistoscope were functional, they were called back for the remaining
tests, which they completed in two additional sessions.
Complete data were obtained from the 100 subjects on all tests except speed of
encoding (IT). Data on this test were collected from only 50 subjects who, as
mentioned, were given the test on two occasions.
On each of the speed-of-processing tests, subjects were given detailed instruc-
tions as to the nature of the tests and how they were to perform. They were also
given as many practice trials as they wished until they indicated that they under-
stood what do to (SD1, SA1, and DT1 were all practice tests presented before the
tests themselves were administered).
Only the author administered and scored the WAIS, to remove any intertester
differences. In accordance with the instructions in the Manual (Wechsler, 1955),
certain parts of the WAIS were timed and subjects received bonus points where
appropriate.

RESULTS

The results are presented in two sections. The first examines the kinds and de-
grees of relationships existing between the different speed-of-processing varia-
bles. The second describes the relationship between these variables and the
measures of intelligence.
Before describing the results of the various correlational analyses performed
on the data, two other issues will be dealt with. The first concerns the reliability
of the speed-of-processing tests, which it is necessary to know before computing,
or interpreting, correlation coefficients. The second concerns the possibility of a
speed-accuracy trade-off operating in the speed-of-processing tests.

Reliability.
Jensen (1979a) reported that reaction times and movement times on the simple
and choice reaction time apparatus had internal consistency (split-half)
reliabilities greater than .90, for as few as 15 trials on each light-button combina-
tion. The reliabilities of the speed-of-processing tests used in this study were also
estimated by the split-half (odd-even) method and are reported in Table 1. As can
be seen, they are all very high: with the exception of some of the standard devia-
tions (which represent intrasubject variability), they are all greater than .90.
Test-retest reliability of the speed of encoding task (IT) was found to be .80.
SPEED OF INFORMATIONPROCESSINGAND GENERALINTELLIGENCE 59

TABLE 1
Spearman-Brown Boosted Reliabilities of Speed-of-
Processing Variables
Spearman-Brown
Boosted
Internal-Consistency
Variable Parameter Reliability

DIGIT RT .99
SD of RT .95
SD2 RT .99
SD of RT .91
RT .98
SA2
SD of RT .92
RT .99
DT2 WORDS
SD of RT .79
RT .97
DT2 Digits SD of RT .89
RT .98
DT3 Words
SD of RT .83
RT .97
DT3 Digits SD of RT .87

Since the reliabilities of the tests are so high, the effect of correcting correlations
between them and other variables for attenuation due to unreliability would be
negligible, and has, consequently, not been done to any of the correlations re-
ported below.

Speed-accuracy Trade-off.
It is important to determine whether subjects who responded fastest to the
various speed-of-processing tests also made the greatest number of errors. If they
did, this would to some extent invalidate the tests as measures of speed of cogni-
tive processing. This problem was approached by computing phi coefficients be-
tween reaction times and number of correct responses, each dichtomized as
above and below the median, for each of the reaction time tests. Positive phi
coefficients would indicate that subjects who responded slower (i.e., had higher
RTs) than the median reaction time also made more correct responses (and that
faster subjects made more errors). Negative coefficients would indicate that
faster-responding subjects also made more correct responses. Phi coefficients
were computed for the seven reaction time tests in which a speed-accuracy trade-
off might have occurred (SD2, SA2, DIGIT, DT2 words, DT2 digits, DT3
words, and DT3 digits). Four of the seven coefficients were negative but none of
the coefficients was significantly different from zero (they ranged f r o m - . 1 1 3
to + . 166). It can be concluded that there is no relationship between speed and
accuracy on these tests.
60 VERNON

Correlations and Principal Factor Analyses

Speed of Processing Tests.


Correlations between subjects' mean reaction times on the speed-of-
processing tests, including the slope in test RT (in milliseconds), and SDs are
reported in Table 2. All correlations are positive and, with the exception of sim-
ple and choice reaction time (test RT), quite high.
These correlations were submitted to a principal factor analysis. Only one fac-
tor with an eigenvalue greater than 1 was extracted from the analysis; it accounts
for 65.5% of the variance between subjects' scores on these variables. The load-
ings of the variables on this factor are also reported in Table 2. All the variables
load positively, and, with the exception of RT and RT slope, all the loadings are
high. Not only is the first principal factor the only one with an eigenvalue greater
than 1, but the X2 test of the significance of the residual matrix was not signifi-
cant. This indicates that no other significant factor could be extracted from the
correlations among these variables.
Correlations between subjects' speed of encoding (IT) and the other speed-of-
processing variables, based on n = 50, are reported in Table 3. This table also
shows the results of a principal factor analysis carried out on the intercorrelations
among these variables. Note that speed of encoding correlates negatively with all
the variables except RT.
Two factors with eigenvalues greater than 1 were extracted, and, as can be
seen in Table 3, IT does not load positively or highly on the first factor, while the
other loadings follow a similar pattern to those reported in Table 2. IT does load
on the second factor, while the other variables have relatively low or negligible
loadings on this factor.
Correlations between subjects' intra individual standard deviations on the
speed-of-processing variables are reported in Table 4. As with the correlations
between means, all correlations are positive and moderately high. Also, correla-
tions between RT and the other variables are consistently the lowest.
These correlations were submitted to a principal factor analysis, which
yielded only one factor with an eigenvalue greater than 1. This factor accounts
for 57.9% of the variance; the loadings are also reported in Table 4. Though
somewhat lower than the loadings reported in Table 2, they follow a similar pat-
tern. All, with the exception of RT, are in the .60's to .80's. Again, a X2 test
showed that the residual matrix is not significant.
These analyses suggest the presence of an underlying speed or efficiency of
processing factor, which accounts for a large part of the variance among the tests.
However, subjects' mean reaction times per se may confound execution of men-
tal processes with factors such as reading speed or attention. More revealing,
perhaps, are the results of an analysis of the specific cognitive processing varia-
bles that the tests are designed to measure. These include retrieval of information
TABLE 2
Correlations, Means, Standard Deviations, and Loadings on First Principal Factor of Mean Reaction Times on Speed of Processing Tests
Loadings
on First
DT2 DT2 DT3 DT3 RT Principal
Tests SD2 DIGIT Words Digits Words Digits SA2 RT Slope Factor SD
SD2 1 .689 .841 .682 .693 .629 .665 .335 .391 .804 746.17 146.88
DIGIT 1 .683 .827 .681 .742 .555 .300 .471 .808 553.01 156.03
DT2 Words 1 .793 .782 .722 .748 .415 .330 .881 741.12 127.91
DT2 Digits 1 .756 .888 .675 .485 .459 .919 597.27 147.72
DT3 Words 1 .823 .864 .334 .425 .890 886.01 154.61
DT3 Digits 1 .657 .544 .450 .893 561.71 133.23
SA2 1 .321 .340 .792 1006. 01 185.13
RT 1 .326 .469 351.75 42.68
RT Slope 1 .487 24.54 11.23
TABLE 3
Correlations and Unrotated Factor Loadings of Speed-of-Processing
Tests Including IT
Loadings on Loadings on
Correlations First Principal Second Principal
Tests with IT Factor Factor
SD2 -. 147 .771 -.069
DIGIT -.279 .811 -.351
DT2 Words -.095 .835 .121
DT2 Digits -.196 .912 -.056
DT3 Words -.124 .906 .121
DT3 Digits -.116 .893 .076
SA2 -.018 .775 .253
RT .111 .514 .292
RT slope -.277 .519 -.281
IT 1.00 -.174 .519

TABLE 4
Correlations and Loadings on First Principal Factor of Intra-individual SDs on
Speed-of-Processing Tests
Loadings on
DT2 DT2 DT3 DT3 First Principal
Tests SD2 DIGIT Words Digits Words Digits SA2 RT Factor
SD2 1 .579 .528 .500 .541 .551 .379 .290 .668
DIGIT 1 .605 .698 .681 .653 .483 .254 .817
DT2 Words 1 .590 .586 .569 .516 .300 .742
DT2 Digits 1 .552 .684 .518 .272 .780
DT3 Words 1 .656 .643 .255 .803
DT3 Digits 1 .601 .315 .826
SA2 1 .315 .686
RT 1 .373

TABLE 5
Correlations between Speed-of-Processing Variables"
Variables 1 2 3 4 5 6 7
1 1 .765 .277 -.054 -.049 .097 -.171
2 1 .251 .189 -.117 .062 -.401
3 1 .525 .040 .147 .055
4 1 .244 .134 .196
5 1 .333 .389
6 1 .139
7 1
IVariables are defined and computed as follows:
1. STM storage-processing trade-off (DT2 digits-DIGIT)
2. STM storage-processing trade-off (DT3 digits-DIGIT)
3. LTM retrieval (SA2-SD2)
4. LTM retrieval (DT3 words-DT2 words)
5. Simple-choice reaction time (slope of reaction time in test RT)
6. Simple-choice reaction time variability (slope of intra-individual SD in
test RT)
7. STM scanning (slope of reaction time in test DIGIT)
SPEED OF INFORMATIONPROCESSINGAND GENERALINTELLIGENCE 63

from LTM (computed as the difference between reaction times on tests S A 2 - S D 2


and DT3 words--DT2 words), STM storage-processing trade-off (DT2
digits-DIGIT and DT3 digits-DIGIT), STM scanning (slope of reaction time on
set-size in DIGIT), and slope of reaction times and intra individual SDs in RT
the latter tapping differences between choice compared to simple reaction time.
Correlations between these variables are reported in Table 5. Factor analysis
revealed three factors which collectively account for 72.8% of the variance.
Loadings on the first unrotated factor and the three varimax rotated factors are
shown in Table 6. The first unrotated factor accounts for only 30.2% of the vari-
ance, and, as the loadings indicate, it cannot be interpreted as a strong general
factor. It is largely defined by the variables designed to tap storage-processing
trade-off in STM, while LTM retrieval variables have moderate positive load-
ings. The rotated factors show a clear pattern of association among variables de-
signed to tap the same processes and of differentiation between those measuring
different processes. The first two factors are related, respectively, to STM
storage-processing trade-off and LTM retrieval. The third factor is defined by
slopes of reaction time and intra individual SDs in test RT and slope of reaction
times in DIGIT, and may be interpreted as a STM processing factor.

Psychometric Tests.
Correlations between subjects' scaled scores on the WAIS subtests, their ver-
bal, performance, and full-scale IQs, and their raw scores on the Raven Matrices
are reported in Table 7. All the variables, with the exception of Digit Symbol,
correlate positively with one another, though most of the correlations are quite
low.
Principal factor analysis carded out on the correlations between the WAIS
subtests and Raven Matrices yielded three factors with eigenvalues greater than
1, which collectively account for 55.4% of the variance. The first factor alone
accounts for 35.5% of the variance and is interpreted as representing a general
intelligence factor. The variables' loading on this factor are also reported in Ta-
ble 7. Subjects' Z scores on the variables included in this analysis were weighted

TABLE 6
Factor Loadings of Speed-of-Processing Variables"
Varimax Rotated Factor Loadings
First Unrotated
Variables Factor Loadings Factor 1 Factor 2 Factor 3
1 .772 .842 .027 .055
2 .939 .909 .239 -.151
3 .401 .224 .517 .098
4 .321 -.062 .987 .177
5 -. 107 -. 074 .089 .736
6 .095 .114 .086 .418
7 -.300 -.315 .091 .487
•Variables are as defined and computed in Table 5.
TABLE 7
Intercorrelationsbetween PsychometricVariablesand FirstFactor Loadings of Subtestsand Raven*
First
Factor
INFO C O M P ARITH SIMS DIGSPAN VOCAB D I G S Y M B PICTCOMP BLKDES PICTARR OBJASSM RAVEN VIQ PIQ FSIQ Loading
INFO 1 .322 .476 .300 .055 .572 -.089 .383 .305 .100 .171 .253 .644 .250 .515 .598
COMP 1 .238 .303 .143 .362 .143 .317 .201 .369 .321 .352 .592 .404 .577 .546
ARITH 1 .251 .153 .453 .095 .303 .498 .242 .346 .517 .701 .452 .670 .639
SIMS 1 .232 .280 .057 .265 .305 ,171 .270 .309 .561 .331 .512 .449
DIGSPAN 1 .156 .036 .149 .115 .112 .0@8 .289 .546 .156 .406 ,248
VOCAB 1 -.017 .412 .332 .234 .301 .435 .704 .377 .621 .637
DIGSYMB 1 .185 .182 .205 .070 .238 .088 .537 .353 .185
PICTCOMP 1 .405 .209 .370 .453 .477 .661 .651 ,591
BLKDES 1 .332 .485 .623 .453 .727 .669 .692
PICTARR 1 .308 .453 .326 .635 .544 .473
OBJASSM 1 .524 .375 .685 .599 .575
RAVEN 1 .573 .695 .722 .797
VIQ 1 .522 .880
PIQ 1 .863
FSlq 1
"Variable n a m e s , in t h e o r d e r t h e y appear above, are: Information, Comprehension, Arithmetic, Similarities, Digit Span, Vocabulary, Digit Symbol,
Picture Completion, Block Design, Picture A r r a n g e m e n t , Object A s s e m b l y , R a v e n Matrices, Verbal IQ, P e r f o r m a n c e IQ, Full-Scale IQ.
SPEED OF INFORMATIONPROCESSINGAND GENERALINTELLIGENCE 65

by these loadings and summed to create g factor scores. This g factor is used in
some of the multiple regressions analyses reported in the following section.

Multiple Regression Analyses


In the first analysis, subjects' full-scale WAIS IQ scores were regressed on
their mean reaction times and intra individual SDs on the speed of processing
tests. Seven predictor variables produced a maximum shrunken R with IQ of
.464. Applying a standard formula (McNemar, 1949, p. 149) to this R allows it
to be corrected for attenuation due to the restricted range of IQ of the sample.
When this is done, the corrected R is .668, which reflects more accurately the
correlation that could be expected to be found in the general population. 2
Regressing WAIS IQ on reaction times and intra individual SDs separately
yielded similar results. Five reaction time variables produced a maximum
shrunken R of .434 with IQ (.636 after correction). Three of the intra individual
SDs produced a maximum shrunken R of .431 (.633 after correction). Evidently,
speed of processing--as reflected by the reaction times---and consistency or effi-
ciency of processing (intra individual SDs) have about the same degree of associ-
ation with IQ, while together they produce only marginally better prediction than
does either by itself due to high intercorrelation.
Another multiple regression analysis was performed on the data of the 50 sub-
jects who had taken the encoding test (IT) in order to assess the contribution of
this variable to the prediction of IQ. With full-scale WAIS IQ as the dependent
variable, IT was found not to add significantly to the multiple R when entered
with subjects' mean reaction times on the other speed-of-processing tests. Fur-
ther, its zero-order correlation with IQ is o n l y . 103, and it is the only speed-of-
processing variable that correlates positively with IQ.
As described previously, mean reaction times per se do not provide as good a
measure of specific cognitive processes as do the differences between reaction
times on some of the tests and the slopes of reaction times on others. Conse-
quently, WAIS IQ was regressed on a set of variables (as defined in Tables 5 and
6) designed to tap STM scanning, LTM retrieval, and STM storage-processing
trade-off. The maximum shrunken R produced in this analysis is .366, or .559
after correction for restriction of range, which is not significantly lower than the
value obtained in the analyses described above. Inspection of the variables'
standardized beta weights reveals that storage-processing trade-off (DT3
digits-DIGIT) makes the greatest contribution to the prediction of IQ, followed

qt might be argued that applying the formula for correction due to restriction of range produces
an artificially-inflatedestimate of the population correlation. It is possible to test this, to some extent,
by applying the formula to the correlation between any of the WAIS subtests and comparing the
boosted value to the value obtained from the normative sample and reported in the WAIS Manual.
This was done to the correlation betweeen the Arithmetic and Informationsubtests, as an example.
The sample correlation is .476. Boosted for restriction of range on Information,it becomes .659. The
value reported in the WAIS Manual for 25-34 year-olds is .66.
66 VERNON

by STM scanning (slope in DIGIT), and LTM retrieval (SA2-SD2), in that order.
It is possible that the relationships found here between WAIS IQ scores and
reaction times on the speed-of-processing tests are largely a result of the content
common to both. For example, all the speed-of-processing tests, with the excep-
tion of RT, contain verbal or numerical items, as do the verbal subtests of the
WAIS. More important, perhaps, many of the performance subtests of the WAIS
are timed, and subjects' scores can be increased if they perform quickly on these
tests. Perhaps, then, the main cause of the correlation between the reaction time
and IQ tests is the fact that both have verbal and numerical content and both, to
some extent, involve timed performance.
To test this, a composite reaction time variable was created by summing sub-
jects' reaction times and intra individual SDs on the speed-of-processing tests,
after converting these to Z scores, This composite variable--labelled RTg--was
then used as the dependent variable in a number of multiple regression analyses
which used different sets of psychometric variables as the independent variables.
In each of these analyses, the psychometric g factor was forced to enter the re-
gression equation at the first step, in order to see whether any of the other psy-
chomelric variables would add significantly to the equation after g. The rationale
for these analyses is that if the verbal or the timed WAIS subtests fail to add
significantly to a multiple R with RTg after g has already contributed to the R,
then it cannot be argued that these variables are responsible for the correlation
between intelligence and speed of cognitive processing as measured in this study.
The first of these analyses regressed RTg on g and the verbal subtests of the
WAIS. After g, which correlates - . 4 0 9 with RTg, none of the verbal subtests
adds significantly to the multiple R. Digit span enters at the second-to-last step,
and has a zero-order correlation of only - . 0 9 7 with RTg, despite the fact that
many of the variables composing RTg require subjects to rehearse digits for later
recall (or recognition). Vocabulary enters at the last step, despite the verbal con-
tent of many of the speed-of-processing tests.
The second analysis regressed RTg on g and the timed subtests of the WAIS
(Digit Symbol, Picture Arrangement, Block Design, and Object Assembly). In
this analysis, none of the timed tests contributed significantly to the multiple R
after g.
It might be argued that this analysis is an inappropriate way to assess the con-
tribution of the timed subtests to the relationship between intelligence and speed
of processing. By forcing g to enter the analysis at the first step, it is possible that
a certain amount of "timed variance" is simultaneously being carried into the
prediction equation. The same argument might be applied to the verbal subtests
and the analysis of the common content of the WAIS and the speed of processing
tests. There are a number of arguments against this, however.
First, g represents what is common to all the psychometric variables that went
into the initial factor analysis, not just the timed subtests nor just the verbal
subtests. Second, the amount of timed variance in g--that is, the amount of vari-
SPEED OF INFORMATIONPROCESSINGAND GENERALINTELLIGENCE 67

ance in g attributable purely to variance in speed of performance of the timed


subtests--is surely quite small relative to other sources: Digit Symbol--probably
the most time-dependent subtest in the WAIS--has the lowest g loading, while
the Raven, which was not timed at all, has the highest (see Table 7). It bears
noting that the Raven also shares no face content in common with any of the
speed-of-processing tests. Third, the amount of timed variance in the timed tests
themselves is also probably smaller than other sources of common variance. All
of the Performance subtests, including Digit Symbol, correlate more highly with
the untimed Raven than they do with each other. Picture Completion and Picture
Arrangement correlate higher with, respectively, Vocabulary and Comprehen-
sion (both untimed tests) than with any of the timed Performance subtests.
It seems unlikely, then, that the timed and verbal subtests fail to contribute to
the correlation with RTg because g itself is to a large extent time or verbally
saturated.
A final series of multiple regression analyses were performed to estimate the
relative power of prediction of full-scale WAIS IQ possessed by the different
speed-of-processing tests. In each analysis, IQ was regressed on all the parame-
ters associated with each of the speed-of-processing tests, in order to see which
test's parameters produced the largest shrunken R. For example, for the test SD2,
IQ was regressed on subjects' reaction times, movement times, and intra individ-
ual SDs of reaction times and movement times. The largest shrunken R produced
is .410. For the RT test, IQ was regressed on subjects' reaction times, movement
times, and intra-individual SDs of reaction times and movement times for each of
the four light-button combinations, and the slope and intercept of the regression
of reaction times on bits, producing the largest shrunken R of .447. The results of
the regression of IQ on the parameters of the other speed-of-processing tests are
summarized in Table 8. Insofar as the Rs produced by these analyses are compa-
rable, the parameters of the RT task produce the highest prediction of IQ. This
supports the conclusion reached from the previous series of analyses in that the
RT task, which shares no common verbal or numerical content with the WAIS
subtests, provides a better prediction of IQ than do the other speed-of-processing
tests.
TABLE 8
Summary of Multiple Regressions of Full-Scale IQ
Regressed on All Parameters Associated with Each of
the Speed-of-ProcessingVariables
Variables Largest Shrunken R
SD2 .410
DIGIT .425
DT2 (words and digits) .437
DT3 (words and digits) .443
SA2 .377
RT .447
68 VERNON

DISCUSSION
The high positive correlations between reaction times on the speed-of-processing
tests, and the initial factor analyses of these correlations, suggested the presence
of a single mental speed factor which accounts for a substantial part (65.5%) of
the variance in these tests• A similar, single factor was extracted from the corre-
lations between intra individual SDs. It, too, accounted for a large part of the
variance (57.9%) and was provisionally interpreted as representing general con-
sistency or efficiency of processing. However, a later analysis, which employed
combinations of reaction times designed to tap specific cognitive processes, did
not support the single-factor finding. The first unrotated factor accounts for only
30.2% of the variance, and the variables' loadings do not warrant its interpreta-
tion as a strong general factor. Varimax rotation produced three quite clearly de-
fined factors representing storage-processing trade-off in STM, LTM retrieval,
and STM processing.
Analysis of subjects' inspection times revealed that whatever is common to
the other reaction time tests is not shared by this variable. Despite its quite high
reliability (.80), IT correlated negatively or nonsignificantly with all reaction
times and did not load to any appreciable extent on the first factor extracted from
their intercorrelations. IT was also the only cognitive variable which did not cor-
relate with IQ. This is consistent with Nettelbeck & Lally (1981), where it was
suggested that IT might not vary to any large extent among samples of average or
above-average intelligence, although it varies considerably between samples of
average and below-average intelligence. It appears to be a threshold variable
which can successfully distinguish retarded and nonretarded samples (Lally and
Nettelbeck, 1977; Nettelbeck and Lally, 1976), while within either group, or at
least within a group of above-average intelligence, it does not appear to correlate
with measures of intelligence nor with other measures of cognitive processing.
The results of the multiple regression analyses support the hypothesis that the
speed with which persons can perform different cognitive processes is signifi-
cantly and quite highly related to their intelligence• Regressing WAIS IQ scores
on a set of reaction times and intra-individual SDs produced the largest shrunken
R in this study, equal to .464 before, and .668 after correction for restriction of
range• IQ regressed just on the variables designed to tap storage-processing
trade-off, LTM retrieval, and STM processing produced a multiple R of .366, or
•559 after correction. Notably, storage-processing trade-off made the largest con-
tribution to this R; it also accounted for more of the variance among the speed of
processing variables than did either LTM retrieval or STM processing•
Other multiple rgression analyses indicated that subjects' intra-individual var-
iability has about the same degree of association with IQ as do their reaction
times. This replicates earlier studies of reaction time and intelligence (e.g.,
Jensen, 1979a; Vernon, 1981a), which showed that it is not only the speed with
which a person can perform mental operations that is related to higher perform-
SPEED OF INFORMATIONPROCESSINGAND GENERALINTELLIGENCE 69

ance on tests of mental ability, but also the consistency with which he can per-
form at the same level over a period of time. A theoretical explanation of this
relationship remains to be developed. It may be a function of an individuals' at-
tention span (Carlson & Jensen, 1982b), or may, as Jensen (1979a) conjectured,
be related to individual differences in the frequencey of the oscillation of synaptic
excitatory potential. In the present study, speed and consistency of response were
highly correlated and appeared to be joint aspects of an individuals' information-
processing capability.
It is important that when a composite reaction time variable (RTg) was re-
gressed on g and the verbal and timed subtests of the WAIS, only g contributed
significantly to the multiple R. Whatever general intelligence and reaction times
have in common cannot be attributed to their shared verbal and numerical con-
tent, nor solely or even largely to the fact that parts of the WAIS are timed. Fur-
ther, subjects obtained very low error rates on all the reaction time tests, indicat-
ing that it is not these tests' difficulty per se that accounts for the relationship.
This is further supported by the final series of multiple regressions (reported in
Table 8) in which the parameters of the least complex test--RT--produced a
larger shrunken R with IQ (albeit not significantly larger) than did the parameters
of any of the other reaction time tests.
Rather, it is the g factor common to all the psychometric variables that ac-
counts for the bulk of the relationship between IQ and reaction time. Further,
given the degree of this relationship, it appears that a moderately large part of the
variance in g is attributable to variance in speed and efficiency of execution of a
small number of basic cognitive processes.
If this is the case, it is contrary to the notion that IQ tests measure little more
than the knowledge an individual has acquired, the problem-solving strategies he
has developed, and the opportunities he has had to learn these. Certainly, many
IQ tests contain informational items, the answers to which an individual either
knows or does not know. Similarly, the person who can use problem-solving
strategies relevant to an IQ test will obtain a higher score than the person who
cannot use them or who does not recognize the relevance of a particular strategy
to a particular problem. In terms of speed of processing, however, it is proposed
that the individual with the larger knowledge base and strategy base has acquired
these as a result of his or her basic information-processing capability. Over a
period of time the years of formal education, for example--faster cognitive
processing may allow more information to be acquired.

REFERENCES

Carlson, J. S., & Jensen, C. M. Reaction time, intelligence, and attention Paper presented at the
annual meeting of the American Educational Research Association, New York, March 1982.
Carlson, J. S., & Jensen, C. M. Reaction time, movementtime, and intelligence: A replication and
extension. Intelligence, in press.
70 VERNON

Dugas, J. L., & Kellas, G. Encoding and retrieval processes in normal children and retarded adoles-
cents. Journal of Experimental Child Psychology, 1974, 17, 177-185.
Harris, G. J., & Fleer, R. E. High speed memory scanning in mental retardates: Evidence for a cen-
tral processing deficit. Journal of Experimental Child Psychology, 1974, 17, 452-459.
Hunt, E. Varieties of cognitive power. In L. B. Resnick (Ed.), The nature of intelligence. Hillsdale,
NJ: Lawrence Erlbaum Associates, 1976.
Jensen, A. R. g: Outmoded theory or unconquered frontier? Creative Science and Technology, 1979,
2, 16-29. (a)
Jensen, A. R. Reaction time and intelligence. Invited address presented at the NATO Conference on
Intelligence and Learning, York University, York, England, 1979.(b)
Jensen, A. R. Chronometric analysis of intelligence. Journal of Social and Biological Structures,
1980, 3, 103-122.
Jensen, A. R. Reaction time and intelligence. In M. Friedman, J. P. Das, & N. O'Connor (Eds.)
Intelligence and learning. New York:Plenum, 1981.
Jensen, A. R. The chronometry of intelligence. In R. J. Sternberg (Ed.), Recent advances in research
on intelligence. Hiilsdale, NJ: Lawrence Erlbaum Associates, 1982.
Keating, D. P., & Bobbitt, B. L. Individual and developmental differences in cognitive-processing
components of mental ability. Child Development, 1978, 49, 155-167.
Lally, M., & Netteibeck, T. Intelligence, reaction time, and inspection time. American Journal of
Mental Deficiency, 1977, 82, 273-281.
McCauley, C., Dugas, J., Kellas, G., & DeVellis, R. F. Effects of serial rehearsal training on mem-
ory search. Journal of Educational Psychology, 1976, 68, 474--481.
McNemar, Q. Psychological statistics. New York: Wiley & Sons, 1949.
Nettelbeck, T., & Lally, M. IQ put to test. Nature, 1981, 290, 440.
Nettelbeck, T., & Lally, M. Inspection time and measured intelligence. British Journal of Psychol-
ogy, 1976, 67, 17-22.
Thorndike, E. L., & Lorge, I. The teachers' word book of 30,000 words. New York: Bureau of
Publications, Teacher's College, Columbia University, 1944.
Vernon, P. A. Reaction time and intelligence in the mentally retarded. Intelligence, 1981, 5,
31-36. (a)
Vernon, P. A. Speed of information processing and general intelligence. Unpublished doctoral dis-
sertation, University of California, Berkeley, 1981. (b)
Wechsler, D. WAIS manual. New York: The Psychological Corporation, 1955.

You might also like