You are on page 1of 14

Individual and Developmental

Differences in Cognilive-proeessing
Components of Mental Ability

Daniel P. Keating and Bruee L. Bobbitt


In.stitute of Child Development, University of Minnesota

KEATING, DANIEL P., and BOBBITT, BRUCE L. Individual and Developmental Differences in
Cognitive-processing Components of Mental Ability. CHILD DEVELOPMENT, 1978, 49, 155-167.
Cognitive activity has been viewt-tl from a variety of rest'are.'li perspectives, hut there have been
few attempts to integrate these different perspectives theoretically or empirically in eirder to gain
a metre general picture eif human cognition. The 3 perspectives of eievelopmental psycheifogy,
exptirimental psychology, and differential psychology are used in this research in an attempt to
understane:! better the nature e>f mental ability. Specifically, we searched for dilTerences in basic
cognitive processing which conld he systematically related to either developmcntiif or individual
differences between the subjects. In 3 experiments (simple vs. che)ice reaction time, Petsner
letter identificatie)n, and Stemljerg memory scanning), v\e looked for interactions e)f experimental
condition with age e)r abifit)' (defined hy Raven's matrices score.s). Interactions with age were
found in the reaction time and letter identificatie)n studies and witfi al)ifity in letter identification
and memory scanning. Additional anafyses revealed that a suf)stantial amount eil test-score vari-
ance wa.s aeee)unted for hy the processing variables and that the interrelations of the processing
variables provided some evidence for a sequential model of ce)gnitive activity.

Two recurrent findings permeate the lit- viduals. Second, differences may exist in the
erature of developmental and differential psy- strategies (or rules) which individuals posses.s
chology: (1) older children do better on cogni- and can bring to bear in solving problems.
tive task.s than younger children, and (2) indi- Third, there may also be difTerences in strat-
vidual performance on A variety of such tasks egy selection skills, sometimes called
is likely to be positively correlated. These two metac'Ognitive activity. These categories are
findings underlie the predictive success of presumably interdependent, and each
IQ-type tests, but they also challenge theorists contributes some variance to performance dif-
oí intellectual development to present "ex- ferences.
planations" whfch do more than recapitulate
these findings that older and brighter children This research is eoncenied with the first
typically do better than younger and less of these categories. If differences in basic
bright ehildren. cognitive-processing efficiency exist, and if
these differences are systematically as-
A major goal of research in cognitive de- sociated with differences in performance on
velopment thus should he to begin identifying more complex cognitive tasks, then at least
the sources of this pervasive variance in cog- some of the variance in mental ability is at-
nitive performance. Several categories of" tributable to these basic processing dif-
sources of variance can be hypothesized. ferences. The goal of finding such differences
First, there may be developmental or individ- is a long-standing one in psychology (Hunt
ual differences, or both, in the general pro- 1976). Using a rough application ofa distribu-
cessing efficiency of the organism. Some indi- tive menioiy model. Hunt and his colleagues
viduals may simply process basic information (Hunt 1976,' Hunt, Freist, & Lunnehorg 1973;
more rapidly and efficiently than other indi- Hunt, Lunneborg, & Lewis 1975) have

This research vv'as supported by a grant from the University of Minnesota Graduate School and
hy a Spencer Fellowship from the National Academy of Educatiein. We thank Tom Trabasso forhis
hefplnt comments on an earlier draft of this paper. Reprint recjuests should be addressed to
Daniel Keating at the Institute of Child Development, 51 East River Road, University of
Minne.sota, Minneapolis, Minnesota 55455,
\Chü(i Deiielopmenl, vns, 49, 155-167. © li>78 by the Society for Research in ChiUl Develoument, Inc.
0009-,1920/78/4901-0020$Û1.16]
156 Child Development
examined the relationship of some basic cog- ferences from a process as basic as binary de-
nitive processes to ability differences in col- cision.
lege students. We are using a similar A more directly relevant process is the re-
paradigm here, with the addition of a devel- trieval of information from long-term memory,
opmental variable. Besides yielding a more since the kinds of possible relationships
complete picture of the relationship between among features must be assessed. This was
these processes and mental ability, the addi- examined in the second experiment, adapted
tion of the developmental variable may be from the research of Posner, Boies, Eichel-
critical if the contribution of basic processing man, and Taylor (1969). The difference be-
differences to complex cognitive performanee tween the two conditions in this experiment is
varies as a function of development. that in the first a decision on similarity is to be
Two levels of ability, high and average, made on the basis of the physical appearance
were ineluded in this study. The ability selec- of two letters. In the second (name) condition,
tion variable was a composite score derived however, the name of the letters must be re-
from the Standard and Advanced Progressive trieved before a similarity decision is possi-
Matrices (Raven 1960, 1965). There are sev- ble. The hypothesis is that the name-sorting
eral advantages in the use of Raven's matrices. condition will take longer as a function of the
First, the problems are self-contained and do retrieval process, although that is not the only
not assess factors such as vocabulary knowl- interpretation of the dinerence between the
edge or general information. They are, thus, conditions (Posner & Snyder 1975). If any of
clearer measures of problem-solving ability. the groups are differentially slowed under the
Second, scores on the matrices tend to corre- name-sorting condition (i.e., if there is an
late highly with the first common factor extract- interaction of condition and group), it is pos-
ed from a battery of ability tests and thus to sible to interpret this as a difference in the
have a high "g" or gen eral-ability loading. efficiency of long-term memory retrieval.
It should be emphasized that we are using Another highly relevant process is
the scores as indicants of ability differences, examined in the third experiment in this
but the study is not a task analysis of the ma- study, and that is the efficiency of short-term
trices. It is important, however, to select basic memory scanning. In the matrices problems,
processing tasks which can be expected a one must hold in working memory a series of
priori to have some relationship to the solu- patterns while determining their inter-
tion of matrices problems. The three process- relationships. To examine this we used the
es which we examined were binary decision Stemberg (1966, 1969, 1975) paradigm which
(choice vs. simple reaction time), long-term varies memory set size and compares scan-
memory-code retrieval (Posner letter iden- ning rates as a function of set size. Other re-
tification), and short-term memory efficiency searchers have found ability differences in
(Sternberg memory scanning). The back- memory scanning efficiency on single cate-
ground of each of these tasks is discussed gory sets (Dugas & Kellas 1974; Harris &
briefly below. Fleer 1974). The ability differences in these
In the first experiment, we assessed the studies were between normals and retardates,
subject's ability to make a binary decision however, and the issue of pathology is poten-
rapidly. We presented first a simple reaction tially confounding. Among college .students,
time task, that is, pushing a button when a Chiang and Atkinson (1976) report no re-
light comes on, and then a two-choice reaction lationship between scanning efficiency and
time task, that is, pushing one button for a red SAT scores, but there is a range restriction in
light and another button for a green light. The their study which could obscure any such re-
major difference between these two tasks lationship. Whether children have slower
seems to be the additional problem in the lat- scanning rates than adults is currently a con-
ter of making one extra discrimination. If troversial issue. Hoving, Morin, and Konick
there is an interaction of the two conditions (1970) found no age differences in scanning
with either classification variable (age or abil- speed, and it has been believed that no such
ity), this indicates an efficiency difference at a differences exist (Stemberg 1975). Two more
very basic level. Such a process seems to be recent studies, however, have shown rather
employed in solving the matrices, since the striking age differences in the search rate on
relevance of a given feature must be de- single-category lists (Herrmann & Landis
termined early in the process. One might not 1977; Naus & Ornstein 1977). It should be
expect, however, especially salient dif- noted that all reported age and ability dif-
Keating and Bobbitt 157
ferences have been in efficiency (i.e., slope of averaging the within-group standard scores on
reaction time against set size) and not in strat- the SPM and APM. There were no overlap-
egy. All studies report evidence for a serial, ping scores. Interpretation of available data
exhaustive search of the short-term memory suggests the low groups were at about the
store. In the present study, tlic simultaneous 40-45 percentile range for their age, whereas
variation of age and ability (excluding the high groups were in the 90-95 percentile
pathological populations) should help to re- range. The low groups are more accurately
solve some ofthe complexities in the existing described as average in ability. For all three
literature. experiments there were five subjects iu each
cell: three ages X two ability groups x two sex
The major goal of these three experi- groups. In addition, for each experiment there
ments, then, was to discover whether reliable were within-subject experimental levels.
individual differences in cognitive processing
exist in children and, if so, whether these dif- Procedure.—The subjects who were
ferences are systematically related to age and selected to participate in the study were
ability. A second goal was to examine the re- tested individually. At this session all of the
lationships among the processing variables for children were administered the APM and
construct-validity evidence for the experi- were given the reaction time, name retrieval,
mental tasks. and memory scanning tasks. The order ot pre-
sentation of these tasks and the APM was
General Method blocked within the 12 cells. The entire ses-
sion lasted between 90 and 180 min. Subjects
In this section, the general methodology took frequent breaks between tasks to avoid
for all three experiments is outlined. The fatigue.
specific procedures and detailed discussion of
the tasks are presented separately for each ex- Experiment 1—Reaction Time: Simple
periment to allow easy c<miparison with the and Cboice*
results.
Subjects and design.—^The same 60 sub- Apparatus.—Three pieces of apparatus
jects were used for all three experiments and were used in this experiment: a subject con-
were drawn from a pool of 180 students irom sole, an electromechanical programming de-
the Minneapolis area public schools. The final vice, and a Beekman Universal Timer (Model
sample of 60 consisted of 20 subjects each 5230). The console consisted of two buttons
from grades 3 (CA - 9.2), 7 (CA = 13.2), and 22 mm in diameter mounted 14 cm apart and a
11 (CA = 17.3), Half ofthe subjects at each age piece of board at desk level. Located beyond
were of high mental ability, and half were of the buttons was a vertical board which had a
average mental ability, as described below. 38-mm-diameter opaque disc positioned at
Half of each of these groups were male, and eye level approximately 50 cm from the sub-
half were female. ject. Through the use of filters, a light behind
the disc cast either a red or green light.
The ability groups at each age were de-
termined by a combination of performance on Procedure.—Each subject was tested
Raven's (1960, 1965) Standard Progressive under simple reaction time (SRT) and
Matrices (SPM) and Advanced Progressive choice reaction time (CRT) conditions. The
Matrices (APM). Both tests were used because SRT condition always preceded the CRT con-
of the wide age range. Each subject in the dition. Each snbject was tested in a quiet
original pool of 180 was given the SPM in room with the lights out. Before testing began,
class while school was in session. On the basis the experimenter showed each subject all of
of these scores, a group ofthe highest scorers the equipment and explained the apparatus.
and the lowest scorers at each age were cho- In the SRT condition, the subject's task
sen for the study. These chiidren were con- was to push the right-hand button whenever
tacted first by postcard and later by telephone the disc was red. Subjects were seated with
and asked to participate in the group tif exper- the appropriate hand resting comfortably on
iments. At the time ofthe experimental ses- the buttou. Preferred hand was counterbal-
sion, each subject was administered the APM. anced. Subjects were instructed to push the
The final sample of 10 higher scorers and 10 button as fast as possible when the red light
lower scorers at each age was determined by came on. Reaction time was printed out to the
'Christine Richardson carried out this experiment tor her bachelor's honors thesis.
158 Child Development
nearest millisecond on tbe Beckman timer and Median reaction times for eacb subject in
was recorded by tbe experimenter. After 10 eacb condition were calculated and used as
practice trials, each subject was administered the dependent measure. Means and SDs of
one block of 42 experimental trials. Tbe trials reaction times for each age and ability group
were separated by either 4-, 6-, or 8-sec inter- on SRT and CRT are shown in table 1. In tbis
vals wbicb were randomly ordered and pro- experiment, there were three significant main
grammed on tbe electromechanical device. effects: age,F(2,48) - 92.38, p < .001; ability,
The interval was the time between the re- F(l,48) = 7.30, p < .01; and experimental
sponse and the next occurrence of tbe light. level,F(l,48)>100,p < .00Mn each case the
All 42 trials took place with no break. means were in the expected direction with
older and brighter children faster overall and
In the CRT condition following the SRT SRT faster than CRT. There was no main ef-
condition, the subject's task was to push tbe fect for sex. Only one interaction was sig-
button marked with green tape when a green nificant, age X level, F(2,48) = 49.28, p < .001,
ligbt appeared and tbe button marked with with younger children slowed differentially
red tape when the red ligbt appeared. As be- more on CRT compared with older children.
fore, preferred hand was counterbalanced. Tbe ability x level interaction was in the ex-
Following 24 practice trials, each subject was pected direction, witb higher-ability children
administered one block of 72 experimental differentially less affected by tbe CRT condi-
trials. The order of presentation was randomly tion relative to SRT, but it was not significant,
determined, and the intervals of 4, 6, or 8 see F(l,48) - 2.81, p < .10. No otber interactions
occurred equally often and in random order. were close to being significant. Since tbere
was no effect of sex or any interaction involv-
Results.—In tbis experiment, tbe first ing tbat variable, it was dropped for a second
analysis examined the total number of errors ANOVA witb 3 (age) x 2 (ability) x 2 (level)
and their distribution across the subjeet and factors. The results were similar, except that
experimental levels to determine the appro- the slightly inereased power shows the mar-
priateness of reaction time as tbe dependent ginal ability X level interaction more clearly,
measure. In tbis study, errors could occur only F(l,54) - 2.90, p < .09.
on tbe CRT condftion. Errors for each cell
ranged from 4% to 7%, which is within rea- In summary, tbere were individual dif-
sonable limits. There was a significant age ef- ferences in tbe processing parameter of bi-
fect, F(2,48) = 4.82, p < .05. Inspection of cell nary decision, and they were systematically
means sbows this to be due to a higher error related to age and perbaps to ability. ,
level (6%) in tbe middle age group, wbereas
the oldest and youngest groups are equal (4%)
in error rate. It is thus unHkely tbat errors in Experiment 2—Name Retrieval from
CRT affected the outcomes significantly. Memory
Tbe principal analysis for eacb experi- Materials.—The task used in this experi-
ment was an ANOVA-mixed design witb 3 ment was adapted from the work of Hunt et al.
(age) X 2 (ability) x 2 (sex) between-subjeets (1975) and Posner et al. (1969). Two sets of
factors and the witbin-subject factor of exper- cards for sorting were constructed, one for tbe
imental level. In tbis case, there were two ex- pbysieal-sorting condition (PS) and one for tbe
perimental levels, SRT and CRT. The major name-sorting condition (NS). Tbe cards used
interest was in the interactions of between- were ordinary 3 x 5-incb (7.7 X 12.8-cm)
subjeets factors witb experimental levels, file cards. Two "Quick Stick Rub On" letters
since such interaction constitutes evidence of were placed on eacb card. Only the letters A
processing differences between subject and B were used. In the PS condition, half of
groups (Hunt et al. 1973; Hunt et al. 1975). tbe cards bad letters wbicb were physically
Ceiling or floor effects may have artifactually identical {AA, BB, aa, or bb), and the otber
created interactions, and an estimate of tbis half differed in any pbysical feature (e.g.,Aa,
was attempted in each ease. We were less in- Ba, aA). Tbe actual combinations used were
terested in main effects of between-subjeets drawn randomly from the possible types of
levels, whicb indicate differences in overall letter combinations. In tbe NS condition, the
performance across experimental levels. A names of the two letters were the same on half
main effect of experimental level indicated of the cards (e.g., Ao, BB, bB) and difíerent on
that there was a difference in performance tbe other half (Afc, ba, etc.). As in the PS con-
under different conditions, wbich can be antic- dition, tbe actual pairs were selected at ran-
ipated in all cases. dom from the possible types of combinations.
Keating and Bobbitt 159
TABLE 1
MEANS AND STANDARD DEVIATIONS FOR EACH AGE/ABILITY GROUP
IN ALL CONDITIONS OF THE T H R E E EXPERIMENTS

AGE 9 AGE 13 AGE 17

Average High Average High Average High


Ability Ability Ability Ability Ability Ability
EXPERIMENT M (SD) M (SD) M (SD) M (SD) M (SD) M (SD)
Reaction time :"
Simple 311 (35) 286 (43) 260 (42) 242 (30) 233 (21) 235 (33)
Two-choice,,, 729 689 (110) 520 (52) 468 (27) 467 (24) 430 (30)
Letter identifica-
tion sorting
time:^
Physical, 28,9 (5,9) 25,8 (4.3) 20.8 (2.2) 19.3 (1-8) 19.3 (3.5) 17.6 (2.4)
Name 36, (4,9) 29.9 (6.7) 26.2 (4.0) 22.2 (2.7) 23.6 (3.8) 19.3 (2.3)
3. Memory scan-
ning reac-
tion time:=.
Set size 1. . 1,160 (204) 980 (315) 748 (262) 560 (240) 475 (129) 439 (113)
Set size 3 . . 1,455 (284) 1,178 (476) 930 (310) 695 (209) 617 (121) 571 (130)
Set size 5. . 1,612 (277) 1,199 (362) 1,040 (315) 793 (218) 722 (158) 675 (164)
Slope«! 113 (50) 55 (50) 73 ( 37) 58 (28) 62 (24) 59 (30)
Intercept. . 1,070 (209) 955 (366) 687 (266) 508 (247) 419 (128) 385 (125)

» In milliseconds.
b In seconds.
" In mi 11 i seconds.
^ Slope and intercept are variables derived by plotting RT against set size and calculating the best fitting line for each subject.

The total deck in each condition consisted of imenter had the subject sort the PS and NS
30 cards. The cards were all covered with decks six times each. As in Hunt et al. s (1975)
laminating film for protection. In addition to work, the PS deck always preceded the NS
these two decks, two practice decks were con- deck. Before the subject sorted each deck, the
structed. The first contained 30 laminated experimenter carefully explained how the
blank cards. The second had 10 cards, half of decks were to be sorted and showed each card
which had the letter X on them and half the to the subject and asked him or her to indicate
letter O. which pile the card belonged in. Under each
condition, all subjects easily understood how
Procedure.—All testing took place in a the cards were to be sorted aud did so with
quiet room with the subject seated next to the ease. After each of the six sorts, the experi-
experimenter. The experimenter explained menter shuffled the deck thoroughly. The
the nature of the task and had each subject sort time taken to sort each deck was recorded to
the blank deck three times, both for practice the nearest teuth of a second.
and to get an estimate of ability to sort cards.
The subject was given the deek upside down Results.—In this experiment only the la.st
and was told to turn it over and sort the cards four of the six trials of card sorting in each
into two piles as rapidly as possible. The ex- condition were used in the data analysis. The
perimenter began to time after the subject first two trials were considered as practice,
turned the deck over and began to sort the even though the subjects were unaware of
cards. This procedure was followed for each of this. The median card-sorting time for each
the subsequent decks, After the blank deck, condition for each subject was calculated and
the practice deck consisting ofX's andO's was used as the dependent measure.
sorted twice by each subject. Though these
trials were not timed, the experimenter The betweeu-subjects factors in the
stressed speed and accuracy. None of the sub- ANOVA were the same as in the previous ex-
jects had problems learning the concept of periment. The within-subject factor had two
how to make the appropriate discrimination levels, PS aud NS (physical sorting aud name
aud sort the cards into the appropriate piles. sorting). Means and SDs of sorting time for
each age and ability group are shown in table
Following both practice decks, the exper- 1. All four main effects were significant in this
160 Child Development
study: age,F(2,48) = 51.25, p < .001; ability, Procedure.—A simultaneous-presentation
F(l,48) - 15.99, p < ,001; sex,F(l,48) = 8.18, varied-.set procedure was used, based on the
p <.OO1; and level,F(l,48)> 100,p <.001.As Stemberg (1966, 1969) paradigm, with digits
expected, older and higher-ability subjects used as stimuli. In this procedure the order of
were faster overall in card sorting, and the NS the digits within each memory set varied from
condititjn took longer than the PS condition. trial to trial. The memory sets were either one,
There was an unexpeeted sex difference, with three, or five digits in length and were ran-
girls sorting faster overall than boys. As noted domly selected subsets of the digits 0 through
below, sex did not interact with experimental 9, Slides were constructed for both the mem-
level, and thus the sex difference is irrelevant ory set and the probes.
to this analysis. There was a significant ag^ x
level interaction, F(2,48) - 4.07, p < .05, and a The subjects sat facing the screen with
significant ability x fevel interaction, F( 1,48) their hands resting on the response buttons.
= 12.63, p < .001. There were nt) other sig- The room lights were on for this task because
nificant interactions in the ANOVA. The sig- the slides were highly illuminated by the pro-
nificant interactions indicated that the older jector. The trials were signaled by the high-
and brighter children were less affected by pitched tone generated by the amplifier. Im-
the added demands of the NS ceindition than mediately after the tone went off, a memory-
were younger and lower-ability children. set slide consisting of one, tliree, or five digits
appeared on the screen and remained visible
It should be noted that the interpretation for 4 sec. After another 4 sec in which the
of the difference between these two experi- stimulus was off, the tone again sounded,
mental conditions is not as clear-cut as in ex- warning the .subject that the probe was about
periment 1. Posner and Snyder (1975) suggest to appear. The subject's task once the probe
two components may be involved: one, the was prttjected on the screen was to indicate as
level of informatitjn available ft)r the re- quickly as possible by pushing the appro-
sponse, and the other, an advantage of an acti- priate button whether or not it had been a
vated pathway. The difference between con- member of the memory set.
ditions clearly involves a difference in
cognitive-processing efficiency, however, and A total of 90 trials was presented to each
alst) involves accessing verbal or name codes. subject in three blocks of 30 trials each, In half
In summary, these findings indicate that dif- of the trials the probe was a member of the
ferences among children in this processing memory set, and in half it was not. In addition,
variable are systematically related to age and each of" the five serial positions was probed
to psychometric ability differences. equally often. For the trials in which the
probe was not a member of the set, it was
selected randomly from the set of possibilities
Experiment 3—Memory Scanning after the memory set had been constructed.
The order of all 90 trials across all three trial
Apparatus.—The apparatus for this ex- blocks was the same for all subjects. Before
periment consisted of a subject console lo- the experimental trials were run, the last of
cated next to the console used in experiment the three trial blocks was used as practice.
1. The timer and programming device were During these 30 trials, the experimenter em-
the same as used in that experiment as well. phasized accuracy and speeei. Rest periods
The console contained a rear projection were taken between trial blocks for all sub-
screen, 45.5 cm wide and 30 cm high, ptjsi- jects.
tioned at eye level approximately 50 cm from
the .subject's eye. Located next to the screen Results.—In this study we first analyzed
was an amplifier which emitted a high- response errors to determine the appropriate-
pitched tone and which was used to signal the ness of reaction time (RT) as the dependent
onset of the memory slide and the probe slide. variable. The overall error rate was about 4%,
In front of the base of the screen were two which is acceptable. An ANOVA of errors of 3
buttons 12 cm in diameter with 14 cm be- (age) X 2 (sex) x 2 (ahility) revealed no sig-
tween their centers. The buttons were labeled nificant effects, indicating that alf subjects in
yes and no. For half of the subjects, the each group performed at the same level of ac-
preferred hand was used for the yes response, curacy. A second preliminary analysis in-
and for half, the preferred hand was used for vestigated the possible effects of practice by
the no response. A Kodak 700 carousel posi- an ANOVA of 3 (age) x 2 (sex) x 3 {trial
tioned behind the screen was used for the block). There was no significant effect for trial
presentation of the stimuli. block or for any interactions of trial block with
Keating and Bobhitt 161
other variables. At least within this experi- From the information presented in figure
ment, subjects did not benefit differentially 1, slope and [/-intercept data were derived and
from practice on the task during the data- analyzed in a separate ANOVA. The equation
collection period. y = Ax + B can be used to generate artificial
data points for each subject. In this study, the
The median RT for correct responses was data points from each .subject's equation com-
computed separately for yes and no responses pared to the actual data yielded a correlation
at each set size for each subject. There was no greater than ,99, accounting for over 98% of
main effect for yes and no responses, and it the variance in the actual data, and justified in
did not interact with other variables. In analysis of the slope and (/-intercept. In the
further analyses, median RTs were computed slope analysis, the effect of ability, F(l,48) =
for all correct responses combined. In this 3,63, p < .05, was the only significant finding.
ANOVA, significant main effects were found The age X ability interaction, however, ap-
for age, F(2,48) - 41.91, p < .001; ability, proached significance, F(2,48) = 2.82, p < ,07,
F(l,48) - 8.89,p < .001; and set size,F(2,96) > and this can be seen in the plot of the mean
100, p < .001. More interesting is the only slope of each age/ability group in figure 2. It is
significant interaction of set size with ability, clear that the high-ability groups are function-
F(2,96) = 3.65, p < .05. These results are ing near the adult level as early as 9 years old,
shown in figure 1, in which mean RTs are plot- about 55-60 msec, but the average ability
ted as a function of set size for each of the age groups achieve that level between ages 13 and
and ability groups. 17. Unlike the slope, the y-intercept was only

9 . AVERAGE
1600'

1400'

1200' O 9 , HIGH

1 3 . AVERAGE

1000'

< 800 O 13. HIGH

^^--"'"' _ „ > » 1 7 , AAVERAGE


VER
. HIGH

600
cr

400

1 3 5
SET SIZE
FIG. I,—Mean reaction times, in milliseconds, for each age/ability group as a function of
memory set size.
162 Child Development
marginally related to ability but strongly re- Kellas 1974; Harris & Fleer 1974; Hoving,
lated to age, F(2,48) = 35.60, p < .OOL See Morin, & Konick 1970).
table 1 for means and SDs of RTs within set The major infiuence of development ap'
size and of the slope and intercept for each age pears to be in the encoding for and the re-
and ability group. sponding to the probe stimulus in light of the
large decrease in the y-intercept with age. Our
Three interesting conclusions can be finding that the slope did not difïer with age
drawn from these data. First, the linear re- confiicts with the findings of Herrmann and
lationship between RT and set size indicates Landis (1977) and Naus and Omstein (1977).
that subjects in all groups were executing a The lack of consensus in these various studies
serial search for the probe item. The lack of is puzzling and could very well be due to task
effects for type of response and serial position differences. Herrmann and Landis {1977), for
leads to the conclusion that search was also example, used a fixed-set procedure rather
exhaustive. The finding that subjects at eacb than the varied-set procedure that was used in
age/ability group utilized a serial and exhaus- the p r e s e n t study. • '••"''• - • ' ' '
tive search is consonant with earlier findings
for single-category memory sets (Dugas & The third and most interesting finding is

120

(113)

no

100

90

80

70

60

(55)
50

13 17
AGE
FIG. 2.—Mean slopes of reaction time on memory set size for the two ability groups as a function oiage
Keating and Bohhitt 163
that tbe age X ability interaction in the slope fall into tbe former category, that is, similar
approached significance. This result suggests steps: CRT, PS; NS - PS, slope; CRT, INT;
that ability differences in basic processing INT, PS. There are eight correlations in the
might become less potent with increasing age latter category, shown in table 2. Tbe pattern
as more powerful and general processing of correlations is reasonably clear. Tbe aver-
strategies are acquired. age intercorrelation for similar steps is .66 and
for dissimilar steps is .30. Three of four
Tbere are now a number of studies whicb similar-step correlations are highly significant
demonstrate tbat the central process of mem- (p < .001), as are most (seven of eight) of the
ory scanning is clearly related to ability. Wbat dissimilar-step correlations (p < .05). The
is not so clear is why this is true. The one magnitudes of the correlations are «juite differ-
study tbat investigated rebearsal processes in ent, however. Tbe correlation for dissimilar
memory scanning (McCauley, Kellas, Dugas, steps may best be considered as a general
& DeVellis 1976) using ebildren as subjects speed factor, while the added covariance for
found tbat covert rebearsal aided the high- but similar steps across tasks can be interpreted as
not the low-ability subjects. Future research modest construct validity for tbe notion of dif-
should be devoted to utilizing designs tbat ferent steps in tbe processing seijuence which
borrow from the McCauley et al. (1976) study correlate across experimental tasks.
and the present one for tbe assessment of the
possible processes involved in both the age Tbere are two confounding variables
and ability differences in memory scanning. wbicb may be influencing these results. First,
tbe greater the number of total steps in any
Individual Differenee Analysis of Pro- pairwise comparison, the greater tbe likeli-
eessing Components hood of a general speed factor biasing the re-
sults. The average number of steps per pair in
One method of examining the construct the four similar-step correlations isfive,and
validity of experimental tasks such as those in the average for the eight dissimilar ones is
tbis study is to compare individual perfor- three. More carefully designed research to an-
mances on variables whicb are presumed to be swer tbis question should control for this po-
related tbeoretically. Cbiang and Atkinson tentially biasing factor. Second, tbe NS - PS/
(1976) reported high correlations of similar slope comparison is especially problematic.
steps in memory scanning and visual searcb Whereas slope is clearly isolated as a step (2)
(rstiipcN = .83, rmtertepts = .97) but insignificant separate from tbe otber three, NS - PS repre-
correlations for dissimilar steps. sents the difference between two kinds of op-
erations: internal comparison in PS versus re-
We constructed a similar analvsis for trieval plus internal comparison in NS. The
these data. The tasks are not as clearly related single deviant correlation of similar steps,
as in the Chiang and Atkinson study (1976), -.16 for males, may be attributable to this clis-
but tbe assignment of variables to different crepancy.
steps in the processing secjuence seems rea-
sonable. We use a four-step secj|uence: (1) en- We also examined the pattern for males
coding, (2) operation, (3) binary decision, and and females separately, given the preliminary
(4) response. Following previous work on the findings of Cbiang anci Atkinson (1976). These
Stemberg paradigm, we assigned slope to (2) are shown also in table 2. We did notfindany
and intercept (INT) to (1 + 3 -f 4). The SRT is systematic sex differences, except tbat on tbe
(1 + 4), and the CRT is (1 + 3 -i- 4); hence, dissimilar steps different correlations are sig-
C - SRT is (3). Finally, tbe NS eondition in nificant, but they all appear to be mostly simi-
the Posner task is (1 -f- 2 -t- 3 + 4); the PS lar in magnitude. It would stretch the data too
condition is (1 + 3 -I- 4), and NS - PS is (2). far to speculate about specific cotrelations.
Tbis assignment allows, then, for a com-
parison of individual differences in process- Cognitive-proeessing Differences as
ing at various steps. If these differences are Components of Psychometric Per-
systematic and are related by more tban a formance Differences
general speed factor, tben variables whicb in-
volve the same steps should correlate more It is clear from the results reported above
highly across tasks than variables which bave tbat differences in cognitive-processing ef-
no steps in common. We eliminated vvithin- ficiency are systematically related to individ-
task comparisons because of tbe biasing effect ual differences in psychometrically tested
of task-specific covariance. Four correlations ability. How mncb of the psychometric per-
Pi 0'='

aj I P,

, n •••••••. •

e .

PS D,,..^
c/1 o p

i s?

I rt ^
Keating and Bobbitt 165
formance variance is accounted for by age groups is striking and emphasizes the
cognitive-processing differences? To examine consistency of the link between the process-
this, we derived a central-processing parame- ing variables and psychometric perfonnance.
ter frt)m each of the three experiments to pre- For age groups 9, 13, and 17, the respective
dict, via multiple regression, the subjects' multiple fi's are .59, .57, and .60, with FMC
scores on the psychometric tests. The process- variance accounted for equal to 34%, 32%, and
ing parameter was intended to represent the 36%, respectively. In each case, NS - PS was
specific process assessed by each task. For ex- the most potent variable, accounting for a
periment 1, we used CRT minus SRT {C — greater portion of the PMC variance in suc-
SRT), a measure of binary-decision efficiency. cessive ages: 17%, 25%, and 32%. Since age is
For experiment 2, we used name-sorting con- already restricted, adding it as a fourth pre-
dition time minus physical-sorting condition dictor variable does nt^t alter the regression
time (NS — FS), a somewhat more ambiguous equations. It is also important to nt>te that the
measure of long-tenn memorv retrieval ef- central cognitive-processing variables con-
ficiency. For experiment 3, the slope of RT tribute different sources of covariance to the
against set size, which taps search efficiency prediction. The average intercorrelation of
in sfiort-term memory, was the measnre nsed. these three variables across all subjects is .27.
We constructed as the criterion variable a pro-
gressive matrices composite (FMC), which is There is always a danger in designs em-
the mean of a snbject's within-sample stan- plt)ying extreme groups that correlations will
dard scores on the SFM and the APM. vary in unpredictable ways. More extensive
replications of these estimates are thus
In the first stepwise analysis, using all needed, but we draw confidence from the
children (iV = 60), the multiple R = .72, which consistency of these results. Taken together,
indicates that 52% of the PMC variance was they suggest strongly that a significant portion
acct)unted for by C - SRT, NS - PS, and of the variance in psychometric test perfor-
slope, the three central processing variables. mance during these years is closely related to
Age differences correlated with these three differences in basic cognitive-processing ef-
variables acct)unt ft)r a significant pt)rtion of ficiency.
the total variance explained. This is clarified
by adding age as a fourth predictor variable, in
which case the multiple fi = .79, tmd 62% of Summary and Implieations of Find'-
PMC variance is accounted for. Age alone ac- ings
eonnts for 47% of the variance. Thus, 15% of
PMC variance is explained by the central- The major findings from these experi-
processing variables with age partialed out, a ments can be grouped into three categories:
highly significant (;? < .001) amount of added (a) the identification of meaningful individual
variance. Note alsei that with all four pre- and developmental differences in specific
dictors, including age, the multiple R begins cognitive-processing parameters; (b) the im-
to approach the reliability of the test. proved construct validity of the processing
tasks by virtue of their interrelationships; and^
Chiang and Atkinson (1976) reported a (c) the explanation of variance in the
sex difference in the relationship between psychometric test scores by the prt)cessing pa-
processing parameters and psychometric test rameters. The theoretical and practical im-
scores. We examined these regression equa- plications for the findings in each category are
titms separately for males and females. The discussed briefly below.
results are virtually indistinguishable for
males versus females, and most likely are Cognitive-processing differences.—The
sampling fluctuations. Further data will be most striking finding is the fact that individual
needed to examine possible sex differences, variation in these basic cognitive processes,
but from this sample no different relationships assessed by means of staneiard cognitive ex-
of processing parameters to test scores are perimental methodology, is systematically re-
evident. fated to other important differences between
the subjects, namely age and psychometric
In any regression equation which is not score level. Note also that these results were
cross-validated tin an independent sample, obtained within the range of normal children
chance works in favor ofa higher multiple fi. (i,e., nonpathological). Hunt et al. (1975) re-
To check for such inflation in this study, we ported differences on some tasks for high- and
examined the results for each age group (N = low-verbal college students, but Chiang and
20) separately. The similarity of results across Atkinson (1976) found no relationship be-
166 Child Development
tween the processing parameters and SAT vidual variance was entirely task specific (at
scores of Stanford undergraduates. Recall also least for the tasks selected), again contributing
that efforts early in the history of psychology no useful information about the general cogni-
to align ba.sic sensory/motor processes with tive model.
test performance were unsuccessful and even-
tually abandoned. Chiang and Atkinson's (1976) study is an
important one for the construct-validity ques-
What accounts for the clear pattern of tion. Their reported high correlations be-
results in this study? Three important compo- tween slopes from memory and visual search,
nents, which have not been brought together and of intercepts, but low or zero correlations
in previous studies, can be identified as con- between them, is the model for this analysis.
tributing to these patterns. First, the experi- Future researchers should consider their
mental tasks were more refined, drawing on study and the present one when using more
the vastly expanded research on adult human than a single experiment on the same sample,
cognition in recent years. Second, the range of because information gathered from such re-
ability on the psychometric test scores was search will be useful in generating hypothe-
broad enough and sufficiently sampled so that ses about the structure of human cognition.
such relationships would not fail to appear
because of insufficient variance. Third, the Explanation of psychometric vari-
developmental component revealed that the ance.—The consistency of the multiple-
relationships between processing parameters regression equations predicting the PMC is
and test scores may vary with age. In particu- important because it emphasizes the possibil-
lar, the nearly significant age x abilit)' interac- ity that cognitive experimental research may
tion on slope suggests that the relationship of be useful in understanding the nature of
either age or ability to cognitive processing widely used psychometric tests. The histori-
parameters may be obscured when they are cally atheoretical nature of test constmction
viewed independently. Future research, both has created a situation in which we know that
to replicate these findings and to explore re- tests work, but with little understanding of the
lated avenues, will need to bear iu mind the reasons why they do. If, in fact, they work
requirement of representative subject sam- (partly or principally) because they tap into
pliug and the possible developmental inter- systematic cognitive-processing differences,
actions. -•• • '• - ' -" then we are in a much better position to ex-
plore general issues in complex cognitive per-
It is more difficult to draw a picture of formance.
how the results from all three experiments
may be integrated. For simpler processes, The exploratory nature of these results
such as binary decision, encoding, and re- should be clearly recognized, however. The
sponding (C - SRT, (/-intercept), age effects progressive matrices are good psychometric
seem to dominate. On more complex process- instruments with high loadings on a "g" factor
ing, such as internal comparison and retrieval of ability, but psychometric theory is far be-
(slope, NS - PS), the ability interactions are yond a reliance on a single global index (Horn
striking, A comprehensive interpretation of 1976; Nunnally 1967). Future research should
age and ability differences in cognitive pro- attempt to tie specific processing parameters
cessing would, however, be premature specu- to different ability factors. The pattern of re-
lation at this point. lationships may be especially interesting,
Construct validity of the experimental since some processes may reasonably be ex-
tasks.—In this study, only one pattern of pected to be associated more closely with cer-
intercorrelations of the processing parameters tain psychometric abilities than others, Car-
could be construed as supporting a notion of roll's (1976) theoretical association of test fac-
human cognition as an integrated infor- tors with cognitive processes is an interesting
mation-processing system. That pattern, perspective for initiating empirical studies. In
higher correlations for variables tapping simi- the current phase of the research, negative re-
lar steps in the processing sequence than for sults on the global PMC would be devastat-
those tapping dissimilar steps, was the one we ing, leading sensibly to abandonment of the
found. Uniformly high correlations would in- research strategy. On the other hand, the de-
dicate a highly potent overall speed factor cidedly positive results we obtained are en-
which would contribute little to the validity of couraging, leading instead to pursuing the
a sequence model. Conversely, uniformly low strategy for purposes of replication and re-
or zero correlations would indicate that indi- finement of the relationships. . . , .. . ._,_
Keating and Bobbitt 167
References telligence. In G. Bower (Ed.), Advances in
learning and motivation. Vol. 7. New York:
Carroll, J. B. Psychometric tests as cognitive ta.sks: a Academic Press, 1973,
new "structure of intellect." In L. B. Resnick Hunt, E.; Lunneborg, C ; & Lewis, J. What does it
(Ed.), The nature of intelligence. Hillsdale, mean to be high verbal? Cognitive Psychology,
N.J.: Eribaum, 1976. 1975, 7, 194-227.
Chiang, A., & Atkinson, R. C. Individual differences McCauley, C ; Kellas, G.; Dugas, J.; & DeVellis, R.
and interrelationships among a select set of F. Effects of serial rehearsal training on mem-
cognitive skills. Memory and Cognition, 1976, ory se Arch. Journal of Educational Psychology,
4 , 661-672. 1976,68,474-481.
Dugas, J. L., & Kellas, G. Encoding and retrieval Naus, M. J., & Ornstein, P. A. Developmental dif-
processes in normal children and retarded ado- ferences in the memory search of categorized
lescents, y^ur/irt/ of Experimental Child Psy- lists. Developmental Psychology, 1977, 13,
chology, 1974, 17, 177-185. 60-68.
Harris, G. J., & Fleer, R. E. High speed memory Nunnally, J. C. Psychometric theory. New York:
scanning in mental retardates: evidence for a McGraw-Hill, 1967.
central processing deficit. Journal of Experi- Posner, M. I.; Boies, S. J.; Eichelman, W. H.; &
mental Child Psychology, 1974, 17,452-459. Taylor, R. L. Retention of visual and name
Hormiann, D. J.. & Landis, T. Y. Differences in the codes of single letters.Joumai of Experimental
search rate of children and adults in short-term Psychology, 1969, 79, 1-16. (Monograph)
memory, journal of Experimental Child Psy- Posner, M. I., & Snyder, C. R. R. Attention and cog-
chology, 1977. 2 3 , 151-161. nitive control. In R. L. Solso (Ed.), Information
Horn, J. L. Human abilities; ¡i review of research processing and cognition. Hillsdale, N.J.:
and theory in the early 1970s. In M. R. Eribaum, 1975.
Rosenzweig & L. W. Porter (Eds.), Annual Raven, J. C. Guide to the Standard Progressive Ma-
review of psychology. Vol. 27. Palo Alto, Caiif.; trices. London: Lewis, 1960.
Annual Reviews, 1976. Raven, J. C. Advanced Progressive Matrices. Lon-
Hoving, K.; Morin. R.; àc Konick, D. Recognition don: Lewis, 1965.
reac'tion time and size of the memory set: a de- Stemberg, Si High-speed scanning in human mem-
velopmental study. Psi/c/ionomic Science, 1970, ory. Science, 1966, 153, 652-654,
21,247-248. Stemberg, S. The discovery of stages: extension of
Hunt, E. Varieties of cognitive power. In L. B. Res- Donders' method. Acta Psychologica, 1969, 30,
nick (Ed.), The nature of intelligence. Hills- 276-315.
dale, N.J.: Eribaum, 1976. Stemberg, S. Memory scanning; new findings and
Hunt, E.; Frost, N.; & Lunneborg, C. Individual current controversies. Quarterly Journal of Ex-
differences ineognition: a new approach to in- perimental Psychology, 1975, 2 7 , 1-32.

You might also like