Developmental Psychology Copyright 1991 by the American Psychological Association, Inc.

1991, Vol. 27, No. 5,763-776 0012-1649/91/$3.00

Decomposing Adult Age Differences in Working Memory

Timothy A. Salthouse and Renee L. Babcock
School of Psychology
Georgia Institute o f Technology

Two studies, involving a total of 460 adults between 18 and 87 years of age, were conducted to
determine which of several hypothesized processing components was most responsible for age-re-
lated declines in working memory functioning. Significant negative correlations between age and
measures of working memory (i.e., from -.39 to -.52) were found in both studies, and these
relations were substantially attenuated by partialing measures hypothesized to reflect storage ca-
pacity, processing efficiency, coordination effectiveness, and simple comparison speed. Because
the greatest attenuation of the age relations occurred with measures of simple processing speed, it
was suggested that many of the age differences in working memory may be mediated by age-related
reductions in the speed of executing elementary operations.

Working memory is generally defined as the preservation of Parkinson and his colleagues (e.g., Parkinson, 1982; Inman &
information while simultaneously processing the same or other Parkinson, 1983; Parkinson, Inman, & Dannenbaum, 1985;
information. It is distinguished from other forms of memory Parkinson, Lindholm, & Inman, 1982) have argued that limita-
because the assumption that it reflects both processing and tions of storage are a major factor contributing to age differ-
storage implies that it plays an important role in many cognitive ences in various memory and, presumably, cognitive tasks. Al-
tasks (e.g., Baddeley, 1986; Carpenter & Just, 1989; Salthouse, though they have reported that age differences in certain mem-
1990). ory tasks are reduced in magnitude when young and old adults
An illustration o f the hypothesized functioning of working are matched on a digit span measure postulated to reflect stor-
memory in one cognitive task, mental arithmetic, is presented age capacity, this storage-mediation effect has only been demon-
in Figure 1. The left column in this figure indicates the opera- strated for a few tasks. Several researchers have discussed the
tions to be performed, and the right column represents the possibility that difficulties in coordinating concurrent activities
intermediate products that must be temporarily stored while are a potential source of age differences in working memory
carrying out those operations. This figure is useful because it (e.g., Kirchner, 1958; Rabbitt, 1981; Talland, 1968; Taub, 1968;
graphically illustrates both the importance and the complexity Welford, 1958), but there is apparently not yet any evidence
o f working memory. That is, it is clear from this example that directly relevant to this interpretation. Perhaps the most popu-
effective storage of information is essential in order for the suc- lar interpretation in recent years has been the view that many of
cessful performance of certain cognitive tasks. Figure 1 also the age differences in working memory are attributable to age-
suggests that it may be fruitful to think of working memory not related reductions in processing efficiency (e.g., Baddeley, 1986;
as a single discrete structure, but rather as a dynamic inter- Craik & Rabinowitz, 1984; Gick, Craik, & Morris, 1988;
change among three conceptually distinct aspects or compo- Morris, Gick, & Craik, 1988). However, empirical support for
n e n t s - p r o c e s s i n g efficiency, storage capacity, and coordina- this interpretation has been mixed. For example, it has some-
tion effectiveness. Processing is represented by the series of times been found that the magnitude of the age differences
operations in the left column, storage is represented by the en- increase as processing requirements increase (e.g., Wingfield,
tries in the right column, and coordination can be assumed to Stine, Lahar, & Aberdeen, 1988), but in other studies it has also
correspond both to the sequencing of operations and to the been reported that age differences remain constant as process-
arrows portraying the exchange of information between pro- ing demands are varied (e.g., Babcock & Salthouse, 1990; Light
cessing and storage. & Anderson, 1985; Salthouse, Babcock, & Shaw, 1991).
A primary purpose of this article was to investigate the con- In keeping with the theoretical definition of working mem-
tribution of these three hypothesized components to age-re- ory as involving the simultaneous storage and processing of
lated differences in measures of working memory. Each of the information, most of the tasks explicitly designed to assess
components has been hypothesized to be an important source working memory require the research participant to carry out
of adult age differences by one or more researchers, but few specified processing and to remember particular pieces of in-
definitive conclusions have been possible because the currently formation. Examples are the reading span and listening span
available evidence is both weak and inconsistent. To illustrate, tasks used by Daneman and Carpenter (1980), the counting
task used by Case, Kurland, and Goldberg (1982), the computa-
tion span task used by Salthouse and colleagues (e.g., Babcock &
This research was supported by National Institute on Aging Grant
Salthouse, 1990; Salthouse, Mitchell, Skovronek, & Babcock,
AG06826 to Timothy A. Salthouse.
Correspondence concerning this article should be addressed to Tim- 1989; Salthouse & Prill, 1987), and various tasks used by Turner
othy A. Salthouse, School of Psychology, Georgia Institute of Technol- and Engle (1989).
ogy, Atlanta, Georgia 30332. Although each of these tasks appears to satisfy the theoretical

763
764 TIMOTHY A. SALTHOUSE AND RENEE L. BABCOCK

Figure 1. Illustration of hypothesized influence of working memory in the performance of a cognitive

task. (The mental arithmetic example is taken from Charness, 1985, and the figure is from Theoretical
perspectives in cognitiveaging (p. 334) by T. A. Salthouse, 1991, Hillsdale, N J: Erlbaum. Copyright 1991 by
Lawrence Erlbaum Associates. Reprinted by permission.

criteria for the assessment of working memory, no single one of retical construct, variance associated with the specific m a n n e r
them is likely to provide a pure or completely accurate estimate (e.g., procedures, stimulus materials, etc.) in which the construct
of the working m e m o r y construct because o f the influence of is assessed, and unsystematic or error variance.
task-specific factors. That is, the variance o n any given measure Some indication of the proportion of c o m m o n or construct-
can be postulated to involve variance associated with the theo- relevant variance in measures of working m e m o r y from earlier
 DECOMPOSING WORKING MEMORY 765

studies is available by examining the magnitude o f the correla- Study 1

tion coefficients between different measures o f working m e m -
ory. Most o f the reported values have ranged from .38 (for a Research participants in the first study performed all the
sentence-word measure and an arithmetic-digit measure in tasks illustrated in Figure 2. The hypothesized components o f
Turner & Engle, 1989) to .88 (for oral and silent reading span working m e m o r y investigated in this study were therefore pro-
measures in D a n e m a n & Carpenter, 1980), with intermediate cessing efficiency, storage capacity, and coordination effective-
values o f .40 (for computation span and spatial line span in ness.
Babcock & Salthouse, 1990), .44 (for counting span and reading
span in Baddeley, Logic, N i m m o - S m i t h , & Brereton, 1985), and Method
.57 (for assorted tasks with complex processing requirements in
D a n e m a n & Tardif, 1987). Although these correlations are m o d - Subjects
erate in magnitude, they are likely attenuated because o f a re-
stricted range o f ability (most o f the studies involved college Newspaper advertisements requesting healthy adults to participate
students) and less than perfect reliabilities o f the measures (few in a research project concerned with aging and memory were used to
o f the cited studies reported reliabilities). Procedural variations recruit research participants. A total of 227 adults between 20 and 87
years of age contributed complete data to the project. Seven additional
associated with the methods o f ensuring that the relevant pro-
adults did not complete all the tasks or did not understand the instruc-
cessing was actually carried out, and with the specific criteria tions, and hence their data were not included in the analyses. The final
used to assess a given individual's working m e m o r y capacity, sample consisted of116 women and I 11 men, with between 24 and 67
may have also contributed to underestimates o f the true rela- individuals in each decade. The average years of education in the sam-
tions between alternative measures o f working m e m o r y in the ple was 14.8. The average self-assessed health rating, on a scale ranging
earlier studies. from excellent (1) to poor (5), was 2.10. Correlations with age were - . 17
To emphasize the variance associated with the working m e m - (p < .05) for the education variable and .22 (p < .01) for the health
ory construct, and to m i n i m i z e variance specific to the particu- variable.
lar procedures or stimulus materials used to assess working
memory, it is desirable to obtain multiple measures o f the im-
portant constructs. These measures can then be c o m b i n e d to
Procedure
form composite scores that can be presumed to be better reflec- All research participants were tested in 90-min sessions in groups of
tions o f the relevant theoretical constructs. This procedure was l0 to 40 people. Each individual received $10 compensation for his or
followed in this study by having all research participants per- her participation. The tasks, in the order in which they were adminis-
form two independent, but theoretically parallel, sets o f tasks tered to the participants, were computation span, digit span, arithme-
related to working memory. tic, listening span, word span, sentence comprehension, and the coordi-
The two tasks used to measure working memory, and the nation task, with the arithmetic problems and the sentences presented
tasks used in the assessment o f the three hypothesized compo- orally. All participants received the tasks (and the items within the
tasks) in the same order to avoid confoundings of the individuals with a
nents o f working memory, are portrayed in Figure 2. Notice that
task order that would complicate analyses of individual differences.
the two sets o f tasks are structurally similar, but one set involves
Examples of the stimulus materials used in each task are shown in
arithmetic problems and m e m o r y for numbers, whereas the Figure 2. All the visual materials were assembled in booklets distrib-
other set involves sentence comprehension questions and m e m - uted to each research participant.
ory for words. In both cases the working m e m o r y tasks required
both processing and storage, the storage capacity tasks empha-
sized storage, the processing efficiency tasks emphasized pro- Tasks
cessing, and the coordination tasks required two activities to be
Computation span. In this task a series of arithmetic problems was
performed simultaneously. That is, the computation span and
presented for the participants to solve while also remembering the last
listening span tasks required the participant to select the
digit from each problem. The arithmetic problems were presented
correct answer to the arithmetic problems or to the questions orally at a normal speaking rate. After each problem was presented, the
about sentences while simultaneously remembering digits or examinee selected the correct answer from the three alternatives listed
words. These processing requirements were eliminated in the on an answer sheet in his or her test booklet. On completion of the
digit span and word span tasks postulated to provide relatively designated number of problems, the examinee was instructed to turn
pure measures o f storage capacity. The efficiency with which to the back of the answer page in the test booklet and write the target
the individual could carry out the relevant processing was as- digits. The time allowed for recall was approximately 4 s per target
sessed in the arithmetic and sentence comprehension tasks. Fi- digit, which pilot research indicated was sufficient for people of all
nally, effectiveness o f coordinating two simultaneous activities ages. Instructions emphasized that answering the problems correctly
was to have the highest priority and that the recall would not be consid-
was evaluated by measures o f performance with two concurrent
ered correct unless the processing task was performed accurately.
processing tasks. The number of arithmetic problems presented on each trial in-
Two studies, involving independent groups o f adults between creased successively from one to seven, with three trials presented at
18 and 87 years o f age, are reported. The goal in both studies each series length. The arithmetic problems were all of the form X +
was to attempt to d e c o m p o s e age differences in working m e m - Y = or X - Y=, with the following restrictions: (a) Xand Ywere one-di-
ory into presumably m o r e fundamental components, but the git numbers between 1 and 9; (b) answers to the problems could not be
studies differed in the particular combination o f tasks per- negative; (c) the final number, Y,could not be the same for two adjacent
formed by each individual. problems in a trial; and (d) the answer to the problem could not equal Y
766 T I M O T H Y A. SALTHOUSE A N D R E N E E L. B A B C O C K

WORKING MEMORY
COMPUTATION SPAN LISTENING SPAN
HEAR SEE HEAR SEE

5 PLUS 3 EQUALS __ 3 The boy ran with the dog. Who ran?
__ 7 -- boy
__ 8 man
_ _ girl
6 MINUS 2 EQUALS __ 4
__ 2 John wrote a note Who wrote?
__ 3 wlth • c r a y o n Bob
_ _ Sam
1 PLUS 4 EQUALS __ 5 _ _ John
7
_ _ 6 Last night, Tom w e n t When?
............. to school. _ now_

- yesterday
-

TURN THE PAGE AND - - - last night

-

RECALL -- - ............
TURN THE PAGE AND
RECALL

STORAGE CAPACITY
DIGIT SPAN WORD SPAN
HEAR HEAR

3 dog
2 crayon
. . . . . . 4_. . . . . . school

TURN THE PAGE AND

TURN THE PAGE AND
RECALL
__ RECALL --

PROCESSING EFFICIENCY
ARITHMETIC SENTENCE COMPREHENSION

SEE SEE

5 + 3 = ? __ 3 The b o y ran with the dog. Who ran?

__ 7 __ boy
8 _ _ man
__girl

COORDINATION EFFECTIVENESS
COORDINATION - ARITHMETIC COORDINATION - SENTENCES

HEAR SEE HEAR SEE

The boy ran with Who ran? 9-2=? 5 PLUS 3 EQUALS ~ 3 The b o y s w e r e jumping
the dog. _ _ boy __ 9 __ 7 for joy.
_ _ man __ 3 __ 8 Who was jumping?
__gld 7 _ _ m e n

7+8=? 6 MINUS 2 EQUALS __ 4 _ _ glds

John wrote a note With what? 1 5 2 boys
with a crayon _ _ pen
1 4 3
_ _ crayon
6
_ _ pencil

Figure 2. Examples o f trials in the eight tasks performed by all research participants in Study 1.

T h e two incorrect alternatives for the problems were randomly selected ages in the pilot research. As in the c o m p u t a t i o n span task, the instruc-
n u m b e r s between 1 a n d 20. tions e m p h a s i z e d that the questions had to be answered correctly or the
Listening span. The listening span task consisted o f the oral presen- recall responses would not be evaluated.
tation o f simple sentences; the research participant was instructed to The sentences were between 6 a n d l0 words long a n d were generated
answer a question about the sentence in the booklet while also r e m e m - with the following restrictions: (a) The final word was not longer t h a n
bering the last word in each sentence. Sentences were read at a n o r m a l two syllables, (b) the final word was c o m m o n enough to be found in a
speaking rate, a n d the examinee was requested to select the correct children's dictionary (Simon & Schuster, 1984), and (c) no word ap-
answer from one o f three alternatives listed on an answer sheet in the peared more than once in the session as the final word o f a sentence.
test booklet. O n completion o f the designated n u m b e r o f problems, the A n attempt was m a d e to keep the c o m p r e h e n s i o n questions for each
examinee was instructed to t u r n to the back o f the answer page in the sentence very simple (e.g., who?, when?, a n d where?), and neither the
test booklet a n d write the target words. T h e recall time was approxi- question nor the answer alternatives contained the to-be-remembered
mately 4 s per target w o r d - - a time more t h a n adequate for people o f all target word from the sentence.
 DECOMPOSING WORKING MEMORY 767

Digitspanandwordspan. Both digits and words were presented at a procedure consisted o f scoring each recall attempt as correct
rate of approximately 1 per s, with a duration of about 4 s per presented (all items recalled in original sequence) or incorrect and then
item allowed for recall. The number of items presented on each trial designating the span as the highest n u m b e r o f target items re-
increased from 3 to 11, with three trials at each series length. Words called correctly on at least two o f the three trials with that
used in the word span task were selected with the same constraints sequence length. An additional requirement in the computa-
used in the selection of words for the listening span task.
tion span and listening span tasks was that in order for a trial to
Arithmetic and sentence comprehension. These tasks involved the
presentation of a single page of items (27 arithmetic problems or 25 be considered correct, no errors could have been c o m m i t t e d on
sentences) with three answer alternatives for each item. The examinee the relevant arithmetic (for computation span) or sentence c o m -
was instructed to answer as many of the items as possible in 20 s. There prehension (for listening span) problems. The span estimates in
were two separately timed administrations of each task. these tasks therefore represented successful recall and success-
Coordination. Two versions of the coordination task were adminis- ful performance on the required processing.
tered. One version involved participants attempting to solve visually Performance in the remaining tasks was assessed in terms o f
presented arithmetic problems while simultaneously answering ques- the number o f items answered correctly in the allotted time.
tions concerning orally presented sentences. The second version in-
Two scores were available in the coordination tasks, corre-
volved the participants attempting to answer questions about visually
sponding to the measures on the visual task and the auditory
presented sentences while simultaneously solving arithmetic problems
that were orally presented. There were two separately timed (20 s) ad- task. However, there were very few errors, and consequently
ministrations of each version. Figure 2 shows that the stimuli for the little variability, in the performance o f the auditory tasks, and as
visual portions of the coordination tasks were presented in a manner a result, this measure was not subjected to further analyses.
identical to that for the arithmetic and sentence comprehension tasks. The distribution o f the computation span and listening span
The stimuli for the auditory portions of the tasks were presented in a estimates by decade are illustrated in Figure 3. It is clear in
manner identical to that for the computation span and listening span these data that the distribution o f spans systematically shifts
tasks. Participants were instructed to give highest priority to answer- toward lower values with increased age. This shift is reflected in
ing the questions concerning the auditorily presented material but also the age correlations o f - . 4 7 for the computation span estimates
to complete as many of the visually presented items as possible in the
and - . 5 2 for the listening span estimates. These results can
time allowed.
therefore be viewed as establishing the p h e n o m e n o n o f age-re-
lated reductions in working m e m o r y functioning that we in-
R e s u l t s a n d Discussion
tend to explain.
Spans were estimated in the same m a n n e r for the computa- Table 1 contains correlations and estimated reliabilities for
tion span, listening span, digit span, and word span tasks. The the major measures in Study 1. Reliabilities o f the spans were

8 ~
o o S ~
o o 8 ~
o o 5 ~o o S ~ o ~ ~
/ I I I

,,,, ,,,,, i,,,,

=03
$
_Q

Z
E 4
z ½ B
v

5 I I I
o.
09

6
I I

7
20s 30s 40s 50s 60s 70s
Age Decade
Figure 3. Distribution ofestim~ed spans ~ r the computation span (solid b a h
and listening span ~pen b a h measures in Study 1.
768 TIMOTHY A. SALTHOUSE AND RENEE L. BABCOCK

Table 1
Correlation Matrix for Study 1
Measure 1 2 3 4 5 6 7 8 9

1. Age -- -.47 -.52 -.34 -.42 -.57 -.52 -.66 -.49

2. CS (.90) .68 .55 .59 .64 .57 .62 .49
3. LS (.86) .45 .62 .60 .62 .61 .57
4. DS (.89) .57 .42 .42 .43 .41
5. WS (.84) .53 .56 .53 .50
6. Ar (.92) .68 .79 .58
7. Se (.89) .75 .71
8. ArD (.94) .77
9. SeD (.82)
M 50.67 2.58 3.00 6.61 4.08 12.33 7.26 5.74 1.98
SD 16.82 1.57 1.35 1.50 1.16 4.67 2.66 3.82 1.85

Note. N = 227. Entries in parentheses are estimated reliabilities derived by boosting split-half correla-
tions by the Spearman-Brown formula. Valuesin boldface represent correlations between variables postu-
lated to assess the same construct. CS = computation span; LS = listening span; DS = digit span; WS =
word span; Ar = arithmetic speed; Se = sentence speed; ArD = arithmetic speed in dual condition; SeD =
sentence speed in dual condition.

computed by determining the correlation between the average the z scores from corresponding measures in the two sets of
number of correct responses on trials with even numbers of tasks, and one each on the measures from each set of tasks.
to-be-remembered items and the average number of correct re- Results of these analyses, expressed in terms of the increment
sponses on trials with odd numbers of to-be-remembered items, in R 2 for age after control of other variables, are summarized in
and then the Spearman-Brown formula was used to estimate Table 2.
the reliability of all the trials. Although the span estimates were For the sake of simplicity, we describe the major findings
not directly based on the total number of correct trials, correla- from Table 2 in terms of the results based on the composite
tions between the total correct measure and the actual span scores. However, the table can be inspected to verify that a
estimates ranged from. 90 to .94 across the four span measures. qualitatively similar pattern is evident in the measures from
(Corresponding values for these same measures in Study 2 each set of tasks. First, it is apparent that there is only slight
ranged from .87 to .92). Reliabilities for the remaining mea- attenuation of the age differences after statistical control of the
sures in Table 1 were determined by applying the Spearman- health and education variables. However, the age-associated
Brown formula to predict reliability of the average score from variance was substantially reduced, to less than 7%, by control-
the correlation between scores of two separate assessments. ling for the variable corresponding to storage capacity, to about
Three important points should be noted about Table 1. The 2% by controlling for the variable reflecting processing effi-
first is that the reliabilities for all measures were respectable, ciency, and to just over 1% by controlling for both the storage
with a median of.89 and a range from .82 to .94. The second capacity and processing efficiency variables.
point is that although all the correlations among measures are A second noteworthy aspect of the results in Table 2 is that
moderately positive, the largest correlations are generally be- the degree of attenuation of the age-related influences on the
tween measures hypothesized to reflect the same theoretical hypothesized components varied according to the other compo-
construct. In particular, the correlation of .68 between the nents treated as potential mediators. For example, statistical
computation span and listening span tasks was the highest for control of the processing efficiency component reduced the
each of these measures. This finding is consistent with the in- age-associated variance in the storage capacity component to
terpretation that the two measures reflect a common construct, 1.3%, but statistical control of the storage capacity component
in addition to any specific processes that might be involved with only reduced the age-associated variance in the processing effi-
each task. Finally, it is noteworthy that the age correlations with ciency component to 12.3%. The asymmetric nature of these
all measures were negative and moderate to large in magnitude. effects provides a clue to the causal relations among the vari-
To evaluate the potential mediating influence of different ables because variables functioning as mediators of a relation
variables on the age-related effects on other variables, a series of would be expected to result in greater attenuation of the rela-
hierarchical regression analyses were performed. The logic of tion when they are controlled than variables representing the
the analyses was that the contribution of a variable to the age outcome, or consequence, of the relation.
differences on a given measure could be estimated by contrast- The pattern of results just described suggests that efficiency
ing the amount of age-associated variance before and after re- of processing may be a major determinant of the age differences
moving the variance associated with that variable. The variable in working memory. Storage capacity and coordination effec-
can be inferred to be important as a potential mediator of the tiveness also contribute to the age differences, but since they are
age relations to the extent that there is a substantial attenuation themselves influenced by processing efficiency, processing effi-
of the age-associated variance after it is statistically controlled. ciency seems to be the primary factor responsible for many of
Three separate sets of parallel analyses were conducted; the the age differences in working memory.
primary one based on composite scores created by averaging One method of summarizing the empirical relations among
 DECOMPOSING WORKING MEMORY 769

Table 2
Increment in R 2 for Age After Statistically Controlling for Other Variables (Study 1)
Computation
Working memory span Listening span

Control R2 F R2 F R2 F

None .297 94.93* .224 64.91" .275 85.36*

Health .261 84.31" .197 57.27* .243 75.87*
Education .244 86.11" .185 56.96* .226 76.61"
H &E .222 78.43* .167 51.61" .206 69.89*
H, E, & storage .068 34.11" .070 27.00* .079 33.76*
H, E, & process .021 10.49" .017 6.54 .055 22.60*
H, E, & coord. .028 12.53" .007 2.79 .067 26.42*
H, E, storage, & process .011 6.47 .007 3.30 .032 15.04"
H, E, storage, & coord. .014 8.00* .003 1.38 .036 16.77"
H, E, process, & coord. .011 6.76* .004 1.56 .040 16.93"
H, E, storage, process, & coord. .008 4.62 .001 0.69 .025 13.67*

Storage capac~y Di~tspan Word span

None .188 52.08* .119 30.35* .180 49.43*
Health .169 46.87* .ll6 29.48* .152 42.11"
Education .146 44.41" .097 25.49* .134 41.75"
H&E .136 41.39" .098 25.61" .118 36.76*
H, E,&process .013 4.53 .018 5.02 .024 8.68*
H, E,&coord. .014 4.68 .007 2.22 .031 10.86"
H, E, process,&coord. .006 2.29 .006 1.65 .016 5.84

Coordination
effectiveness Arithmetic duff Sentence dual
None .376 135.86" .435 173.17" .244 72.70*
Health .347 125.18" .403 160.39" .224 66.45*
Education .325 125.83" .377 163.46" .210 64.88*
H&E .307 118.57" .358 154.74" .197 60.74*
H, E, & process .026 19.46" .066 48.04* .023 10.95"
H, E, & storage .161 71.14" .258 118.69" .094 32.36*
H, E, process, & storage .023 17.08" .060 43.83* .017 8.06*

Processing Sentence
ettidency Arithmetic comprehenfion
None .353 122.69" .330 110.65" .265 81.32"
Health .305 108.29" .285 97.37* .230 71.18"
Education .293 114.27" .278 101.33" .217 72.67*
H&E .261 103.08" .247 91.11" .193 65.25*
H, E, & storage .123 57.56* .164 64.83* .088 33.99*
H, E, & coord. .011 8.47* .004 2.55 .029 14.65"
H, E, storage,&coord. .008 6.06 .003 1.97 .018 9.52*

Note. H = health; E = education; coord. = coordination effectiveness; dfs = I and 225 minus the number
of controlled variables.
*p < .01.

the theoretical constructs consists o f expressing those relations storage capacity variables and between the coordination effec-
in a path diagram. To the extent that the composite measures tiveness and working memory variables had coefficients less
accurately reflect the theoretical constructs of interest, struc- than twice their standard errors. Those paths were therefore
tural diagrams o f this type are valuable in allowing all the hy- deleted, and the analysis was repeated with only the relations
pothesized causal connections to be represented, including displayed in Figure 4. This model provided a good fit to the data
those that are indirect or mediated, as well as those that are because the 90% confidence intervals ranged from .000 to. 147
direct. A path diagram illustrating the relations among the com- for the Steiger-Lind Adjusted root mean square (RMS) Index
posite variables in this study, with the path coefficients derived (Steiger, 1989) and ranged from .872 to 1.000 for the Adjusted
from the EZPath (Steiger, 1989) computer program, is por- Population G a m m a Index.
trayed in Figure 4. Because the comparisons of primary inter- These path analysis results serve to reinforce the conclusion
est are those between different variables and not those between that variations in processing efficiency are an important factor
different samples, standardized path coefficients are reported. contributing to age differences in working memory. Particu-
An initial analysis revealed that the paths between the age and larly relevant to this conclusion are the relatively small coeffi-
770 TIMOTHY A. SALTHOUSE AND RENEE L. BABCOCK

-.116

[ A~I: / -.545 ~/' PROCESSING

/ .432 ~[ WORKING/
\ .... // -\ EFFICIENCY/~ --\ ME.O.~S

-.184~ .ooa IJ

T .706
.519 U
U
Figure 4. Path analysis model of relations among age and hypothesized components of working memory
based on results of Study I.

cients (-. 116) for the direct path linking age to working mem- The results of this study suggest that processing efficiency is
ory, the relatively large coefficients for the paths linking age to the most important determinant of age-related differences in
processing efficiency (-.545) and processing efficiency to work- working memory. Statistically significant age-related influ-
ing memory (.432), and the absence o f a direct path between age ences are evident in the measures o f storage capacity and coordi-
and storage capacity. nation effectiveness, but those measures contribute little to the

o !

~ 4 ~

I t
6 D L
7
20s 30s 40s 50s 60s 70s
Age Decade
Figure 5. Distribution of estimated spans ~ r the computation span ~olid bar~
and listening span ~pen bar~ measu~s in Study 2.
 DECOMPOSING WORKING MEMORY 771

attenuation o f the age differences in working m e m o r y after par- and 94 men) ranging from 18 to 82 years of age. Each decade from the
tialing the influence o f processing efficiency. Furthermore, al- 20s to 70+ was represented by between 28 and 52 individuals. The data
though there was a small negative relation between age and from 13 additional participants were not included in the analysis be-
ability to perform two concurrent tasks after controlling the cause they failed to understand the instructions or did not complete all
variance associated with the measures o f processing efficiency, tasks. The average health rating (1 = excellent, 5 = poor) for the 233
participants was 2.16, with an age correlation of. 13. The average years
virtually all o f the age-related differences in the span measures
of education was i 5.1, with an age correlation of-.06.
were eliminated after statistical control o f the processing effi-
ciency measures.
Procedure
Study 2 Both the sequence and the identity of the tasks were varied from
Study 1, as the order of the repeated tasks was changed, several tasks
In view o f the apparent importance o f the processing effi-
were dropped, and new ones were added. The first task performed in
ciency c o m p o n e n t in the age differences in working memory, it the session was the arithmetic task, followed by the digit span and
is desirable to be m o r e explicit about precisely what that c o m p o - computation span tasks. New tasks designed to measure comparison
nent represents. The tasks used to operationalize the processing speed were then administered, with three versions involving pairs of
efficiency construct in Study 1 required the research partici- letters and three versions involving pairs of line-segment patterns. The
pants to solve arithmetic problems and to answer sentence com- sentence comprehension task was then performed, followed by the
prehension questions, which are both moderately complex word span and listening span tasks. The final task performed in the
tasks. Because o f their complexity, it was likely that the mea- session was the Digit Symbol Substitution test from the Wechsler Adult
sures from these tasks reflected a combination o f factors and Intelligence Scale--Revised (Wechsler, 1981).
Materials and procedures for the previously used tasks (i.e., those
not just the efficiency with which processing operations were
designed to measure working memory, storage capacity, and process-
executed. In an attempt to obtain a potentially purer assessment
ing efficiency) were identical to those of Study 1. The new Letter Com-
o f the efficiency o f carrying out elementary processing opera- parison and Pattern Comparison tasks consisted of pages containing
tions, participants in this study also performed two relatively pairs of letters, or pairs of line-segment patterns, that the examinee was
simple speeded c o m p a r i s o n tasks. to classify as "same" or "different" as rapidly as possible. One half of
Because no direct relation was evident between the measures the pairs in each page were same and one half were different. Pairs
o f coordination effectiveness and o f working m e m o r y in the requiring a different response were constructed by altering one of the
path analysis results o f Study 1, the tasks used to assess coordi- elements (letter or line segment) in one member of the pair. The letters
nation effectiveness were not included in Study 2. The hypothe- were randomly selected consonants, arranged in sets of three, six, or
sized components o f working m e m o r y investigated in this nine letters in each member of the pair. The line patterns were con-
nected lines in an invisible 4 × 4 matrix, with three, six, or nine line
study were therefore processing efficiency, storage capacity, and
segments in each member of the pair. There were two separately timed
simple comparison speed.
(20 s) administrations of 32 pairs for each of the versions (i.e., with
three, six, or nine items) of each task.
Method
Subjects Results and Discussion
Research participants were recruited in the same manner described The procedure for estimating the spans was identical to that
in Study 1. Complete data were obtained from 233 adults (139 women described in Study 1. Figure 5 illustrates the distribution o f the

Table 3
Correlation Matrix for Study 2
1 2 3 4 5 6 7 8 9 10

1. Age -- -.39 -.41 -.18 -.32 -.53 -.39 -.59 -.64 -.56
2. CS (.84) .49 .52 .46 .62 .50 .54 .52 .52
3. LS (.86) .45 .56 .57 .49 .47 .49 .43
4. DS (.87) .50 .44 .39 .35 .30 .29
5. WS (.79) .42 .48 .39 .37 .37
6. Ar (.91) .68 .70 .67 .72
7. Se (.90) .62 .57 .62
8. Let (.94) .80 .71
9. Pat (.94) .73
10. DigSym
M 46.92 2.89 3.35 6.88 4.30 13.72 7.68 8.61 13.13 58.79
SD 16.44 1.32 1.25 1.37 0.94 4.73 2.29 2.19 2.86 14.39

Note. N = 233. Entries in parentheses are estimated reliabilities derived by boosting split-half correla-
tions I~ythe Spearman-Brown formula. Values in bold represent correlations between variables postulated
to assess the same construct. CS = computation span; LS = listening span; DS = digit span; WS = word
span; Ar = arithmetic speed; Se = sentence speed; Let = speed of letter comparison; Pat = speed of pattern
comparison; DigSym = digit symbol substitution.
772 TIMOTHY A. SALTHOUSE AND RENEE L. BABCOCK

Table 4
Increment in R 2 for Age After Statistical Control of Other Variables (Study 2)
Computation
Working memory span Listening span

Control R2 F R2 F R2 F

None .211 61.89" .149 40.57* .166 46.07*

Health .200 58.78* .146 39.41" .154 42.87*
Education .193 64.02* .137 39.94* .151 46.16"
H& E .187 62.02* .136 39.72* .143 43.78*
H, E,&~orage .079 39.40* .086 31.98" .058 21.99"
H, E,&process .021 9.67* .006 2.37 .062 20.71"
H,E,&speed .011 4.47 .003 1.03 .017 5.46
H, E, ~orage,&process .018 10.38" .009 3.99 .035 13.67"
H, E, ~orage, & speed .008 4.36 .005 1.90 .008 3.08
H, E, process, & speed .006 2.78 .000 0.01 .017 5.70
H, E, storage, process, & speed .005 3.21 .001 0.40 .008 7.22*

Storage capacity Digit span Word span

None .086 21.66* .034 8.04* .105 26.96*
Health .077 19.67" .031 7.31" .094 24.46*
Education .076 20.62* .029 7.22* .094 25.98*
H &E .071 19.21" .028 6.80* .087 24.09*
H, E, & process .001 0.32 .003 0.90 .025 7.76*
H, E, & speed .002 0.42 .001 0.36 .011 3.05
H, E, process, & speed .000 0.03 .007 1.92 .011 3.30

Processing Sentence
etficiency Arithmetic comprehension
None .252 77.82* .282 90.81" .150 40.92*
Health .230 72.79* .265 86.02* .132 36.99*
Education .227 87.32* .258 99.83* .133 42.33*
H, E .213 82.88* .248 96.04* .120 39.04*
H, E & storage .126 58.88* .195 87.53* .060 21.67"
H, E & speed .004 2.15 .013 6.78* .000 0.00
H, E, storage & speed .003 1.75 .014 8.36* .000 0.26

Speed
None .419 166.47"
Health .404 160.49*
Education .400 171.89*
H, E .390 167.28"
H, E & storage .296 136.68"
H, E & process .102 64.43*
H, E, storage & process .099 63.79*

Note. H = health; E = education; dis = 1 and 231 minus the number of controlled variables.

computation span and listening span measures by decade. The ment letter comparisons and -.64, -.64, and -.52, respectively,
pattern o f the distributions shifting toward lower scores with for the three-, six-, and nine-element pattern comparisons. Be-
increased age and the significant negative age correlations of cause the three versions appeared to exhibit similar relations
- . 3 9 for computation span and -.41 for listening span are simi- with other variables, the scores for the three versions were aver-
lar to those observed in Study 1. aged to provide a more reliable measure of performance with
Performance in the speeded comparison tasks was repre- each type of comparison.
sented by the number o f pairs correctly classified in the allotted The correlation matrix for the major measures is shown in
time. It was anticipated that the correlations of these measures Table 3. As in Study 1, reliabilities were generally respectable,
with age, and with the measures of working memory, might and each o f the measures had negative correlations with age but
become larger as the number o f elements in each pair increased moderately positive correlations with one another.
from three to six to nine. Although the number o f correct classi- Hierarchical regression analyses similar to those conducted
fications decreased as the comparisons involved more ele- in Study 1 were performed on the primary measures, with the
ments, no systematic trend was apparent in the magnitude of results summarized in Table 4. (The digit symbol substitution
the correlations. For example, the age correlations were -.58, measure was not included in these analyses because it is
-.56, and -.43, respectively, for the three-, six-, and nine-ele- planned to be the focus of a separate decompositional analysis
 DECOMPOSING WORKING MEMORY 773

AGE

.193
-.615

SIMPLE .711 PROCESSING ~ .296 WORKING

COMPARISON EFFICIENCY MEMORY
SPEED

.518 .399
.724 .618 .538
U U

STORAGE
CAPACITY

.728 I
U

Figure 6. Path analysis model of relations among age and hypothesized components
of working memory based on results of Study 2.

to be reported in a later article.) The major results from Study 1 mance on tasks designed to assess working memory. The age
were replicated by the findings that the age-associated variance correlations for the computation span and listening span mea-
in the measures of working memory was reduced to less than sures ranged from - . 3 9 to -.52, and Figures 3 and 5 indicate
8% after controlling for the measures of storage capacity and to that the age-related reductions are characterized by a gradual
almost 2% after controlling for the measures of processing effi- shift in the entire distribution of scores. The major question
ciency. The earlier results are extended by the additional find- addressed in these studies was which hypothesized component
ing that the age-associated variance was reduced to nearly 1% of working memory is primarily responsible for these differ-
by controlling for the measures of simple speed (letter compari- ences.
son and pattern comparison). Further examination of Table 4 A primary assumption motivating this research is that there
reveals that the interrelations among age and the measures of are some aspects of working memory that transcend specific
the hypothesized components were not symmetrical, because tasks and, hence, are independent of the particular kind of
the greatest attenuation of the age relations occurred after sta- processing carried out and of the particular type of information
tistically controlling for the simple speed measure. being remembered. We are not claiming that working memory
A path diagram illustrating the relations among the compos- functioning is completely independent of the nature of the rele-
ite variables is displayed in Figure 6. As in Figure 4, paths with vant processing or of the type of information being remem-
coefficients less than twice their standard errors were deleted bered. However, the moderately large correlations between the
from the initial analysis, and the analysis was repeated with measures from the computation span and listening span tasks
only the displayed paths. The goodness-of-fit indices revealed in Tables 1 and 3 suggest that there are substantial commonali-
an excellent fit to the data in that the 90% confidence intervals ties in at least these particular measures of working memory.
for the Steiger-Lind Adjusted rms Index (Steiger, 1989) ranged Moreover, the analyses based on the composite scores appear to
from .000 to .093 and those for the Adjusted Population yield meaningful and coherent results, even when the task-spe-
G a m m a Index (Steiger, 1989) ranged from .948 to 1.000. The cific aspects are presumably minimized by averaging across
most interesting aspect of Figure 6 is that all the age-related measures from the two sets of tasks.
effects on processing efficiency, storage capacity, and working Measures hypothesized to reflect components concerned
memory are indirect, rather than direct. In other words, all the with processing efficiency, storage capacity, and coordination
significant relations between age and measures of these con- effectiveness were investigated in Study 1. The greatest attenua-
structs appear to be mediated through age-related reductions in tion of the age differences in the working memory measures
simple speed. was found after statistically controlling for the measures of pro-
cessing efficiency. Furthermore, the relations among the compo-
G e n e r a l Discussion nent measures were asymmetrical, in that the largest attenua-
A very robust finding across the two studies in this project is tion of the age-related effects occurred after statistically control-
that increased age is associated with progressively lower perfor- ling for the measures postulated to reflect processing efficiency.
774 TIMOTHY A. SALTHOUSE AND RENEE L. BABCOCK
-t-
l--
z

z
0

Activation
I-- Threshold
(.3
I~ctivatio'~ t~ Dissipation v I

TIME

as, AOivat,on ...... ............... =

~ ' . . . , . , W o r k i n g M e m o r y = 2.3

Figure 7. Schematic illustration of major assumptions of the proposed explanation for how speed of
processing might influence working memory functioning.

The path analysis results also indicated that a substantial pro- framework as the amount of information for which, at any given
portion of the age-related differences in working memory ap- time, the activation strengths are above a threshold.
pear to be mediated by reductions in processing efficiency. Within this general framework, individual differences in
Components investigated in Study 2 were storage capacity, working memory can be postulated to originate either because
processing efficiency, and simple perceptual comparison of variations in the rate of activation or because of variations in
speed. The results indicated that processing efficiency could be the rate at which information is lost or dissipated. We propose
decomposed into as-yet-unidentified complex factors and that increased age is associated with a reduction in the rate at
aspects related to simple speed and that most of the age-related which information is activated, but not in the rate at which it is
influences seemed to be mediated through simple speed. This dissipated. The basis for the claim that the dissipation rate re-
was apparent in the multiple regression analyses in which statis- mains constant across the adult years derives from tasks such as
tical control of the simple speed measures resulted in substan- continuous recognition (e.g., Erber, 1978; Ferris, Crook, Clark,
tial attenuation of the age relations in the other measures and in McCarthy, & Rae, 1980; LeBreck & Baron, 1987; Lehman &
the path analysis results that indicated that all the significant Mellinger, 1986; P o o n & Fozard, 1980; Wickelgren, 1975), con-
age-related influences on the working memory, storage capac- tinuous paired-associate recognition (e.g., Balota, Duchek, &
ity, and processing efficiency measures appeared to be me- Paullin, 1989), keeping track of changing variables (e.g., Salt-
diated through the simple speed variable. house et al., 1991), and Brown-Peterson interference paradigms
The results of these two studies clearly suggest that the effi- with variable-length retention intervals (e.g., Charness, 1981;
ciency of processing, and especially processing involving very Parkinson et al., 1985; Puckett & Lawson, 1989; Puckett &
simple operations, is an important factor contributing to age-re- Stockburger, 1988; Talland, 1967). In each of these tasks it has
lated differences in working memory However, although the been found that young and old adults have nearly parallel func-
correlational data indicate that an association exists between tions relating accuracy of performance to time, number of in-
speed-related variance and memory-related variance, these tervening items, or number of intervening operations. It there-
data do not by themselves provide an explanation for that rela- fore seems reasonable to infer from these results that the rate at
tion. A satisfactory explanation requires not only that a signifi- which information is lost or dissipated is unrelated to age across
cant relation be established, but also that mechanisms be pos- most of the adult life span.
tulated to specify why that relation exists. Our speculations in No direct evidence exists concerning the influence of age-re-
this regard are an elaboration of earlier ideas by Salthouse (e.g,, lated processes on the rate at which information is activated.
1980, 1982, 1985, 1988). However, there are well-documented findings that older adults
The basic assumptions are portrayed in Figure 7. Notice that require longer intervals to escape backward masking in tachis-
information is postulated to increase in activation strength over toscopic presentations (e.g., see Salthouse, 1982, for an early
a relatively short interval and then to dissipate gradually over a review), a phenomenon sometimes interpreted as reflecting the
longer interval. Working memory can be conceptualized in this time needed to encode a single stimulus. The major results from
 DECOMPOSING WORKING MEMORY 775

our studies supporting the interpretation that increased age is Daneman, M., & Carpenter, E A. (1980). Individual differences in
associated with a reduction in the rate of activation are that the working memory and reading. Journal of VerbalLearning and Verbal
largest attenuation of the age-related influences was with the Behavior, 19, 450-466.
measures of processing efficiency and that this tendency was Daneman, M., & Tardif, T. (1987). Working memory and reading skill
especially pronounced with very simple measures of processing re-examined. In M. Coltheart (Ed.), Attention and performance, XII
efficiency or speed. Because the attenuation of the age relations (pp. 491-508). Hillsdale, NJ: Eribaum.
Erber, J. T. (I978). Age differences in a controlled-lag recognition mem-
on working memory was greatest for the simplest measures, it is
ory task. Experimental Aging Research, 4, 195-205.
apparently not the number or complexity of the operations that Ferris, S. H., Crook, T., Clark, E., McCarthy, M., & Rae, D. (1980).
is important, but the speed with which even very elementary Facial recognition memory deficits in normal aging and senile de-
operations can be successfully executed. These simple mea- mentia. Journal of Gerontology, 35, 707-714.
sures can be postulated to provide better estimates of the time Gick, M. L., Craik, E I. M., & Morris, R. G. (1988). Task complexity
needed to activate stimuli than the more complex processing and age differences in working memory. Memory & Cognition, 16,
efficiency measures, and it may be for this reason that the great- 353-361.
est mediation of the age relations was apparent with the mea- Inman, V.W, & Parkinson, S. R. (1983). Differences in Brown-Peterson
sures of simple comparison speed. It has been suggested that recall as a function of age and retention interval. Journal of Gerontol-
linkages between measures of speed and memory functioning ogy, 38, 58-64.
might be attributable to the rate at which to-be-remembered Kirchner, W K. (1958). Age differences in short-term retention of rap-
items can be rehearsed or repetitively cycled in an articulatory idly changing information. Journal of Experimental Psychology, 55,
352-358.
loop (e.g., Baddeley, 1986; Salthouse, 1980). We suspect that in-
LeBreck, D. B., & Baron, A. (1987). Age and practice effects in continu-
formation can be activated in many ways, however, and that a ous recognition memory. Journal of Gerontology, 42, 89-91.
slower rate of articulation or rehearsal is only one of a large Lehman, E. B., & Mellinger, J. C. (1986). Forgetting rates in modality
number of consequences of a slower speed of activating internal memory for young, mid-life, and older women. Psychology and Ag-
information. ing, 1, 178-179.
The preceding speculations must, of course, still be consid- Light, L. L., & Anderson, E A. (1985). Working-memorycapacity, age
ered quite tentative, but they do suggest a mechanism that and memory for discourse. Journal of Gerontology, 40, 737-747.
might be involved in producing the relation that now seems to Morris, R. G., Gick, M. L., & Craik, E I. M. (1988). Processing re-
be reasonably well established. It may therefore be productive sources and age differences in working memory. Memory & Cogni-
for future research to investigate not only the nature of the tion, 16, 362-366.
relations among age, processing efficiency or speed, and work- Parkinson, S. R. (1982). Performance deficits in short-term memory
tasks: A comparison of amnesic Korsakoffpatients and the aged. In
ing memory, but also to examine specific hypotheses, such as
L. S. Cermak (Ed.), Human memory and amnesia (pp. 77-96). Hills-
the one just described, for why these relations exist. dale, NJ: Erlbaum.
Parkinson, S. R., Inman, V.W.,& Dannenbaum,S. E. (1985). Adult age
differences in short-term forgetting. Acta Psychologica, 60, 83-l01.
References
Parkinson, S. R., Lindhoim, J. M., & Inman, V.W (1982). An analysisof
Babcock, R. L., & Salthouse, T. A. (1990). Effects of increased process- age differences in immediate recall. Journal of Gerontology, 37, 425-
ing demands on age differences in working memory. Psychology and 431.
Aging, 5, 421--428. Pooh, L. W, & Fozard, J. L. (1980). Age and word frequency effects in
Baddeley, A. (1986). Working memory Oxford, England: Clarendon continuous recognition memory. Journal of Gerontology, 35, 77-86.
Press. Puckett, J. M., & Lawson, W M. (1989). Absence ofage differences in
Baddeley, A., Logic, R., Nimmo-Smith, I., & Brereton, N. (1985). Com- forgetting in the Brown-Peterson task. Acta Psychologica, 72, 159-
ponents of fluid reading. Journal of Memory and Language, 24, 119- 175.
131. Puckett, J. M., & Stockburger, D. W (1988). Absence of age-related
Balota, D. A., Duchek, J. M., & PauUin, R. (1989). Age-related differ- proneness to short-term retroactive interference in the absence of
ences in the impact of spacing, lag, and retention interval. Psychol- rehearsal. Psychology and Aging, 3, 342-347.
ogy and Aging, 4, 3-9. Rabbitt, E (1981). Human ageing and disturbances of memory control
Carpenter, P. A., & Just, M. A. (1989). The role of working memory in processes underlying "intelligent" performance of some cognitive
language comprehension. In D. Klahr & K. Kotovsky (Eds.), Com- tasks. In M. E Friedman, J. P. Das, & N. O'Connor(Eds.), Intelligence
plex information processing: The impact of Herbert A. Simon (pp. and learning (pp. 427--439). New York: Plenum Press.
31-68). Hillsdale, NJ: Erlbaum. Salthouse, T. A. (1980). Age and memory: Strategies for localizing the
Case, R., Kurland, D. M., & Goldberg, J. (1982). Operational efficiency loss. In L. W POoh, J. L. Fozard, L. Cermak, D. Arenberg, & L. W.
and the growth of short-term memory span. Journal of Experimental Thompson (Eds.), New directions in memory and aging (pp. 47-65).
ChiM Psychology, 33, 386--404. Hillsdale, NJ: Eribaum.
Charness, N. (1981). Visual short-term memory and aging in chess Salthouse, T. A. (1982). Adult cognition. New York: Springer-Verlag.
players. Journal of Gerontology, 36, 615-619. Salthouse, T. A. (1985). A theory of cognitive aging. Amsterdam: North-
Charness, N. (1985). Aging and problem-solving performance. In N. Holland.
Charness (Ed.), Aging and human performance (pp. 225-259). Chi- Salthouse, T. A. (1988). Initializing the formalization of theories of
chester, England: Wiley. cognitive aging. Psychology and Aging, 3, 3-16.
Craik, E I. M., & Rabinowitz, J. (1984). Age differences in the acquisi- Salthouse, T. A. (1990). Working memory as a processing resource in
tion and use ofverbal information: A tutorial review.In H. Buouma cognitive aging. Developmental Review, 10, 101-124.
& D. (3. Bouwhuis (Eds.), Attention and performance, X (pp. 471- Salthouse, T. A. (1991). Theoreticalperspectivesin cognitiveaging. Hills-
499). Hillsdale, NJ: Erlbaum. dale, NJ: Erlbaum.
776 TIMOTHY A. SALTHOUSE AND RENEE L. BABCOCK

Salthouse, T. A., Babcock, R. L., & Shaw, R. J. (1991). Effects of adult presentation, order of report, and stimulus organization. Journal of
age on structural and operational capacities in working memory. Gerontology, 23, 159-164.
Psychology and Aging, 6, 118-127. Turner, M. L., & Engle, R. W. (1989). Is working memory capacity task
Salthouse, T. A., Mitchell, D. R., Skovronek, E., & Babcock, R. L. dependent? Journal of Memory and Language, 28, 127-154.
(1989). Effects of adult age and working memory on reasoning and Wechsler, D. (198 I). Manual for the WechslerAdult Intelligence Scale
spatial abilities. Journal of Experimental Psychology: Learning, Mem- Revised. New York: Psychological Corporation.
ory, and Cognition, 15, 507-516. Welford, A. (1958). Ageing and human skill. London: Oxford Univer-
Salthouse, T. A., & Prill, K. A. (1987). Inferences about age impair- sity Press.
ments in inferential reasoning. Psychology and Aging, 2, 43-51. Wickelgren, W. A. 0975). Age and storage dynamics in continuous
Simon & Schuster. (1984). Simon & Schuster~ illustrated young reader's recognition memory. Developmental Psychology, 11, 165-169.
dictionary New York: Wanderer Books. Wingfield, A., Stine, E. A. L., Lahar, C. J., & Aberdeen, J. S. (1988).
Steiger, J. H. (1989). EZPath: Causal modeling. Evanston, IL: SYSTAT. Does the capacity of working memory change with age? Experimen-
Talland, G. A. (1967). Age and the immediate memory span. TheGeron- tal Aging Research, 14, 103-107.
tologist, 7, 4-9.
Talland, G. A. (1968). Age and the span of immediate recall. In G. A.
Talland (Ed.), Human aging and behavior (pp. 93-129). San Diego, Received August 29, 1990
CA: Academic Press. Revision received D e c e m b e r 19, 1990
Taub, H. A. (1968). Age differences in memory as a function of rate of Accepted January 18, 1991 •

al AMERICANPSYCHOLOGICALASSOCIATION
SUBSCRIPTIONCLAIMSINFORMATION Today's Date:

W e provide this form to a s s i s t members, institutions, and nonmember individuals with any subscription problems. With the
appropriate information we can begin aresolution. I f y o u u s e the services of an agent, please do N O T duplicate claims through
them and directly to us. P L E A S E P R I N T C L E A R L Y A N D I N I N K I F P O S S I B L E .

PRINTFULLNAMEORKEY NAMEOFIN~,qll UTION MEMBERORCUSTOMERNUMBER(MAYBEPOUNDONANYPASTISSUELAIEL)

DATEYOURORDERWASMAILED(ORPHONED):
ADDRI~S
P.O. NUMBER:.
PREPAID ~ CHARGE
CITY STA~UNTRY ZIP CImCK/CARDCLEAREDDATE:
(If poJble, send • copy, front and back, of your cancelledcheck to help us In our
YOURNAMEAND PHONENUMBER re~mrchof your claim.)
ISSUES: MISSING DAMAGED

TITLE VOLUME OR YEAR NUMBER OR MONTH

Thank you. Once a claim is received and resolved, delivery of replacement issues routinely takes 4.6 weeks.
(TO BE FILLED OUT BY A P A STAFF)
DATE RECEIVED: D A T E O F ACTION:
ACTION TAKEN: INV. NO. & DATE:
STAFF NAME: L A B E L NO. & DATE:

SEND THIS F O R M TO: A P A Subscription Claims, 1400 N. Uhle Street, Arlington, V A 22201-2969

PLEASE DO NOT REMOVE. A PHOTOCOPY MAY BE USED.