See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/14443719

The Processing-Speed Theory of Adult Age Differences in Cognition

Article in Psychological Review · August 1996

DOI: 10.1037//0033-295X.103.3.403 · Source: PubMed

CITATIONS READS
4,350 14,904

All content following this page was uploaded by Timothy Salthouse on 18 July 2014.

The user has requested enhancement of the downloaded file.

Psychological Review Copyright 1996 by the American Psychological Association, Inc.
1996, Vol. 103, No. 3.403-428 0033-295X/%/$3.00

The Processing-Speed Theory of Adult Age Differences in Cognition

Timothy A. Salthouse
Georgia Institute of Technology

A theory is proposed to account for some of the age-related differences reported in measures of Type
A or fluid cognition. The central hypothesis in the theory is that increased age in adulthood is
associated with a decrease in the speed with which many processing operations can be executed and
that this reduction in speed leads to impairments in cognitive functioning because of what are
termed the limited time mechanism and the simultaneity mechanism. That is, cognitive perfor-
mance is degraded when processing is slow because relevant operations cannot be successfully exe-
cuted (limited time) and because the products of early processing may no longer be available when
later processing is complete (simultaneity). Several types of evidence, such as the discovery of con-
siderable shared age-related variance across various measures of speed and large attenuation of the
age-related influences on cognitive measures after statistical control of measures of speed, are con-
sistent with this theory.

The purpose of the current article is to describe, and discuss of established reliability and span a broad range of cognitive
the evidence relevant to, the processing-speed theory of cogni- abilities, the general phenomenon of negative relations between
tive aging phenomena. The fundamental assumption in the the- age and Type A or fluid cognition can be considered quite
ory is that a major factor contributing to age-related differences robust.
in memory and other aspects of cognitive functioning is a re- Performance on tests of cognitive ability is also a meaningful
duction with increased age in the speed with which many cog- target or criterion phenomenon because cognitive batteries have
nitive operations can be executed (Salthouse, 1985b). In this proven useful for prediction and assessment outside of the lab-
article, discussion of evidence relevant to the theory is restricted oratory and in nonacademic settings (e.g., Ghiselli, 1973;
to the adult portion of the life span, but the basic mechanism Hunter & Hunter, 1984). A focus on cognitive test performance
may be relevant across the entire life span because similar ideas therefore provides a relatively parsimonious linkage to real-
have been proposed by Kail (e.g., 1986, 1991; Kail & Park, world activities. The prediction is not perfect, but significant
1992) regarding the development of cognitive functioning dur- relations to real-world functioning have been empirically estab-
ing childhood. lished; thus, if the age-related influences on these measures can
Because the success of a theory cannot be evaluated if the be explained, at least some of the age-related effects in extra-
goal one hopes to achieve is never clearly specified, I begin by laboratory activities might also be explainable (Salthouse,
briefly describing the phenomenon that the present theory is 1992c).
intended to explain. Some of the best-documented findings in Although the range of cognitive measures is extensive, it is
the literature on aging and cognition are the age-related differ- important to emphasize that the present goal is not to explain
ences in Type A (Hebb, 1942) or fluid (Cattell, 1943; Horn, all determinants of cognitive functioning but, rather, to account
1982; Horn & Cattell, 1963) cognition, which include a wide for the differences in cognitive functioning that are systemati-
variety of measures of memory, reasoning, and spatial abilities. cally related to adult age. The phenomenon to be explained is
The relations between age and cognition have been well docu- thus the age-related variation in behavior, and not the behavior
mented since the earliest mental testing of adults (e.g., Foster itself (Salthouse, 1991c, 1992d). Ultimately, of course, more
& Taylor, 1920; Jones & Conrad, 1933), and they are readily comprehensive theories should encompass all dimensions of
apparent in the results from the standardization data in psycho- cognitive phenomena, but an explanation of the relations be-
metric and neuropsychological test batteries (e.g., see Salt- tween adult age and Type A or fluid aspects of cognition is by
house, 1991c, chap. 2, for a review). Because the samples for itself an extremely formidable goal at the present time.
the standardization data in these test batteries are typically large The article is organized in two major sections. The first sec-
and representative, and because the performance measures are tion summarizes the primary assumptions of the processing-
speed theory. Evidence relevant to critical hypotheses of the the-
ory is then described in the second section.
Preparation of this article was supported by National Institute on
Aging Grant R37 AG06826.1 would like to thank John Dunlosky, Julie
Earles, Dan Fisk, Leah Light, Ulman Lindenberger, and David Madden Theoretical Assumptions
for helpful comments on earlier versions of this article.
Correspondence concerning this article should be addressed to Tim- One substantive assumption of the current perspective is that
othy A. Salthouse, School of Psychology, Georgia Institute of Technol- performance in many cognitive tasks is limited by relatively
ogy, Atlanta, Georgia 30332-0170. Electronic mail may be sent via In- general processing constraints, in addition to restrictions of
ternet to tim.salthouse@psych.gatech.edu. knowledge (declarative, procedural, and strategic), and varia-

403
404 SALTHOUSE

tions in the efficiency or effectiveness of specific processes. Some ral window, then the quality of the final product is likely to be
relevant limitations may be partially overcome by experience, impaired because later processing operations would be either
and indeed one view of expertise is that it serves to circumvent less effective or only partially completed. Some type of juggling
processing constraints or limitations (Salthouse, 1991a). Nev- activity might be a metaphor for the simultaneity mechanism
ertheless, it is assumed that general limitations frequently im- because the fundamental principle is that many complex activ-
pose constraints on many types of processing and, hence, that ities require synchronization of the constituent tasks, and syn-
they have consequences for the performance of a large variety chronization is easier when the relevant processing operations
of cognitive tasks. can be executed rapidly.
It is sometimes asserted that general mechanisms are not
plausible as explanations of adult age differences in cognitive Limited Time Mechanism
functioning because of evidence presumed to implicate selec-
tive or differential age-related effects, such as Age X Treatment The basis for the limited time mechanism is simply that the
statistical interactions. However, the assumption of one or more time to perform later operations is greatly restricted when a
fairly general age-related factors does not preclude the occur- large proportion of the available time is occupied by the execu-
rence of significant interactions because interactions can origi- tion of early operations. This mechanism is primarily relevant
nate as a result of (a) the existence of specific or local age-re- when there are external time limits or other restrictions on the
lated influences in addition to the hypothesized broad or gen- time available for processing, such as the presence of concurrent
eral influences (Salthouse, 1992d), (b) differential reliance of demands on processing.
processes or measures on the general or common factor(s) (e.g., Some cognitive tasks (or tests) have a relatively low level of
Salthouse & Coon, 1994), (c) a multiplicative or proportional difficulty, such that the primary determinant of individual
influence of the general factor(s) such that the absolute differ- differences in performance is likely to be the speed of perform-
ences between age groups increase with the magnitude of the ing relevant operations. For these types of tasks no special ex-
treatment effect (e.g., Cerella, 1990; Cerella, Poon, & Williams, planation appears to be needed to account for the relation be-
1980; Salthouse, I985a), or (d) a statistical artifact attributable tween speed and measures of very simple cognition because per-
to differential discriminating power (e.g., reliability, power, and formance on the cognitive measure could merely be another
region in the measurement range) of the variables (e.g., Salt- manifestation of slow processing.
house, 1985b, 1991c). Particularly when one does not assert The limited time mechanism may also operate in more com-
that general factors are the exclusive source of age-related plicated cognitive tasks in which the quality or accuracy of per-
differences in cognition, therefore, the existence of what appear formance is affected by the number of operations (e.g., associa-
to be selective or differential age-related effects in the form of tions, elaborations, and rehearsals) that can be carried out in
statistical interactions is not at all inconsistent with the exis- the available time (Salthouse, 1980, 1982; Salthouse & Kail,
tence of common or general factors. 1983). If complex operations are dependent on the products of
A second substantive assumption of the processing-speed the- simpler operations, and fewer of those products are available
ory is that speed of processing is a critical processing constraint because of a slower execution speed, the effects of slow process-
associated with increased age. From the current perspective, the ing can be expected to be most pronounced on the speed and
speed with which an individual performs a cognitive activity is accuracy of complex operations. A mechanism of this type may
not simply a function of the processes required in that activity therefore account for what is sometimes referred to as the com-
but also a reflection of his or her ability to rapidly carry out plexity effect, or the positive relation between task complexity
many different types of processing operations. A slower speed of and the magnitude of age differences in both speed and accuracy
executing many cognitive operations is not assumed to be the measures of task performance (Salthouse, 1982, 1985b, 1991c;
exclusive source of age-related differences, because other age- Salthouse & Kail, 1983).
related influences are also postulated to exist. Nevertheless, a Because a gradual reduction in the speed of basic processes
reduction with increased age in the speed with which many cog- with increased age is likely to be accompanied by numerous
nitive operations can be executed is hypothesized to be a major adaptations, the consequences of slower processing are not al-
contributor to the adult age differences in many measures of ways easy to predict (Salthouse, 1985b). This point can be il-
cognition (Salthouse, 1980,1985b, 1991b, 1992b, 1994d).' Be- lustrated by considering what may be an analogous situation in
cause this is a fundamental aspect of the theory, much of the reverse, in the form of the evolutions of computer programs
remainder of the article is devoted to the elaboration and justi- that have occurred as successive generations of computers have
fication of this assumption. become progressively faster and more powerful. The enormous
Two distinct mechanisms are postulated to be responsible for increases in performance have not simply been attributable to
the relation between speed and cognition. The limited time increases in the speed of executing the same programs, because
mechanism is assumed to operate because relevant cognitive major modifications in the nature of the programs have also
operations are executed too slowly to be successfully completed
in the available time, and the simultaneity mechanism is hy- 1
Note that not all cognitive operations are assumed to be necessarily
pothesized to operate because slow processing reduces the affected by slower processing and that a role for nonspeed influences is
amount of simultaneously available information needed for explicitly acknowledged. The theory described here is thus not accu-
higher level processing. A metaphor for the limited time mech- rately characterized by claims such as the following: "Slowing of infor-
anism is an assembly line because if relevant processing opera- mation processing. . . is a single master factor underlying involutional
tions are not successfully completed within a particular tempo- changes in all cognitive skills" (Nettelbeck & Rabbitt, 1992, p. 1 9 1 ) .
 PROCESSING-SPEED THEORY 405

occurred to capitalize on the faster speed (and larger memories) the rate of information loss or decay that is primarily responsi-
of newer computers. Similar types of adaptations in the form of ble for age-related consequences of the simultaneity mechanism
alterations in strategy, reliance on prestored solutions instead of (Salthouse, 1992a, 1994b; Salthouse & Babcock, 1991). In fact,
novel problem solving, and so forth could also occur in the hu- research with tasks such as continuous recognition or continu-
man processing system as it becomes progressively slower and ous paired-associates memory suggests that forgetting functions
less efficient with increased age. Despite the complications as- are very similar across the adult age range (see Salthouse,
sociated with identifying all of the consequences of a slower rate 1992a, for a review). However, it should be noted that simulta-
of processing, however, the basic principle underlying the lim- neous availability of information can also be reduced because
ited time mechanism is quite simple; namely, more processing of disruptions in the synchronization of neural signals or pat-
frequently results in higher levels of performance, and the op- terns of activation, and not only because of changes in the rates
portunity to accomplish a larger amount of processing is greater of decay or information loss. For example, alterations in the
when the speed of processing is faster. variability of timing at elementary levels might also lead to de-
creases in the quantity or quality of information based on
multiple interacting inputs. The simultaneity mechanism
Simultaneity Mechanism
should therefore not simply be viewed as attributable to forget-
The second hypothesized mechanism for the relation be- ting because any factor that affects the synchronization of rele-
tween processing speed and quality of cognitive performance is vant inputs also has the potential to alter the amount (and
based on the idea that the products of early processing may be quality) of simultaneously available information.
lost by the time that later processing is completed. To the extent The importance of simultaneous availability of relevant in-
that this is the case, relevant information may no longer be avail- formation is not a novel idea, because it has been mentioned in
able when it is needed. Processing deficits could therefore one form or another for at least 60 years. For example, the con-
emerge because of discrepancies between the time course of loss cept is similar to speculations by Eysenck (1987), Jensen (e.g.,
of information and the speed with which critical operations 1982, 1987), and Vernon( 1983, 1987). The idea of a trade-off
such as encoding, elaboration, search, rehearsal, retrieval, inte- between loss of information and speed of relevant processing is
gration, or abstraction can be executed (Salthouse, 1982, also fundamental to the notion of an articulatory loop in Bad-
1985b, 1988a, 1992b). deley's (e.g., 1986) model of working memory. Analogous ar-
A key assumption of the simultaneity mechanism is that in- guments in discussions related to aging have been made by Bir-
formation decreases in availability (i.e., quantity or quality) ren (e.g., 1965, 1974) and Jones (1956). To illustrate, Jones
over time as a function of either decay or displacement. More- (1956, p. 138) suggested that problem-solving effectiveness is
over, under rapidly changing conditions, the information could impaired when lower level operations are too slow and earlier
also become obsolete in that it may no longer be accurate or steps are lost before the relevant information can be integrated.
pertinent by the time it becomes available. In either case, when Perhaps the first discussion of the simultaneity mechanism
the rate of executing operations is slow, relevant information is was in the following passage by Lemmon (1927):
less likely to be useful because it is more impoverished or de-
It is possible that the quality of intelligence may depend upon the
graded by the time that preceding operations are finally com-
nu mber of connections, but also upon the speed with which those
pleted. Moreover, this will occur regardless of the amount of
connections are formed. Nerve centers (e.g., association centers)
time allowed for processing because the critical limitations are
cannot remain excited indefinitely at maximum intensity; conse-
based on internal dynamics rather than on the relation between quently in the case of the person who forms connections slowly it is
internal (i.e., processing speed) and external (e.g., stimulus pre- possible that the excitation of the first association centers to be
sentation time) factors. affected will have diminished and disappeared before the latter cen-
Performance on tasks assumed to assess working memory ca- ters come into play. Thus only a limited number of centers are co-
pacity might be postulated to reflect functioning of the simulta- operating at any one time. The person who forms connections
neity mechanism because working memory is sometimes con- quickly, however, is apt to have more association centers interacting
ceptualized as consisting of information that is currently avail- at once, since the later centers are aroused before the earlier ones
had a chance to lose their effectiveness. But the most intelligent
able for storage or processing, or both. However, it is important
response is, in general, the one in which the determination of which
to distinguish the amount of simultaneously available informa-
the greatest number of factors have been taken into consideration.
tion, which may be indexed by measures of working memory,
In neural terms this may well mean the response in the determina-
from possible causes of age-related reductions in that amount. tion of which the greatest number of association centers have coop-
A critical hypothesis in the processing-speed theory is that an erated, and the number of simultaneously active centers may in
age-related decrease in speed is one of the major causes of the turn depend to some extent upon the speed with which nervous
variations in working memory associated with increased age impulses are conducted from center to center and through syn-
(e.g., Salthouse, 1992a; Salthouse & Babcock, 1991). As dis- apses within the centers, (p. 35)
cussed later, there is considerable evidence in support of this
hypothesis because statistical control of measures of processing Another early description of the simultaneity mechanism was
provided by Travis and Hunter (1928):
speed has been found to greatly reduce the amount of age-re-
lated variance in measures of working memory (see also Salt- Intelligence is probably best defined as the ability to see relation-
house, 1994b). ships and meanings by having access to as many alternatives or
From the perspective of the processing-speed theory, it is the judgments as possible at approximately the same instant of time.
slower speed of activating or processing information rather than This would necessitate the reaction patterns which subserve the
406 SALTHOUSE

judgments to be active within an extremely short interval of time. count for the speed-cognition relation. The principle underly-
The "feeble-minded" individual has relatively speaking, such a ing the limited time mechanism is that necessary operations
slow conduction rate that one reaction pattern becomes inactive by may not be completed if the processing is slow. The simultaneity
the time another becomes active, thus doing away with the very
mechanism is based on the idea that if the processing is too slow,
factor, relative simultaneity of activity, which makes possible the
then not all relevant information will be available when needed,
seeing of a relationship between ideational elements, (p. 352)
leading to impairments of critical operations that could result
An important implication of the simultaneity mechanism is in either a high rate of errors or time-consuming repetitions of
that the dynamic capacity of processing "structures" or "sys- critical operations.
tems" such as working memory will be affected, with likely im-
pairments of higher order processes such as abstraction, elabo-
Evidence
ration, or integration, because not all of the relevant informa-
tion will be available in a usable form when it is needed Empirical evidence relevant to the processing speed theory is
(Salthouse, 1992b). Furthermore, degradation of these pro- discussed in the context of three major hypotheses: (a) Age-
cesses will lead either to increased errors or to time-consuming related slowing is not exclusively determined by specific and
repetitions of critical processing operations (cf. Mayr & Kliegl, independent deficits; (b) processing speed functions as a medi-
1993). Speed effects on cognitive functioning thus may be indi- ator of some of the relations between age and measures of cog-
rect because they alter the effectiveness of a process (such as nitive functioning; and (c) the limited time and simultaneity
abstraction, elaboration, or integration) that directly affects mechanisms are primarily responsible for the relations between
cognitive performance. Because the simultaneity mechanism is speed and cognitive functioning. However, before discussing
so fundamental, it could have an impact on many aspects of these hypotheses, I briefly summarize the age-related slowing
cognition, including performance in tasks without external phenomenon, and describe the method by which processing
time constraints. From the current perspective, therefore, Peak speed has been assessed.
and Boring (1926) were correct in suggesting that power tests Slowing with age is often considered one of the best-docu-
are not necessarily those that do not involve speed but may sim- mented and least controversial behavioral phenomena of aging.
ply be those that do not take speed into account. One illustration of the slowing phenomenon is the median cor-
Some evidence of the sufficiency of the simultaneity mecha- relation of .45 between age and measures of speed across a very
nism can be obtained from computational models incorporat- wide range of behavioral activities reported by Salthouse
ing variations in the speed of propagation of activation or in the (1985a). The age-related slowing phenomenon is also evident
speed of firing productions (e.g., Salthouse, 1985b, 1988a; see in analyses of the age trends from perceptual speed tests in psy-
also MacKay & Burke, 1990, for additional discussion of a very chometric test batteries such as the Digit Symbol Substitution
similar mechanism). Prediction of the specific consequences of Test from the Wechsler Adult Intelligence Scale-Revised (see
a processing-speed limitation requires a detailed understanding Salthouse, 1992e) and the Visual Matching and Cross Out Tests
of, or a willingness to make many assumptions about, the pro- from the Woodcock-Johnson Cognitive Ability Tests (see Kail
cesses involved in a particular task. It may also be difficult to & Salthouse, 1994). Performance in the Finding A's and Identi-
distinguish the contribution of other aspects of the model from cal Pictures tests (Ekstrom, French, Harman, & Dermen,
the speed parameter being manipulated for the particular con- 1976) has also been found to be negatively related to age in both
sequences that are predicted. That is, the consequences could cross-sectional and longitudinal comparison, by Schaie (1989)
vary according to the type of processing algorithm or represen- and Schaie and Willis (1993). Not only have these researchers
tation system used, even when the method of manipulating the reported pronounced age trends in each type of comparison,
critical variable, in this case processing speed, is identical (cf. but Schaie (1989) has suggested that, in contrast to the situa-
Salthouse, 1988a). tion with many cognitive variables, the age-related declines are
Despite these limitations, the absolute differences in various actually greater in longitudinal comparisons than in cross-sec-
indexes of performance between fast and slow processing sys- tional comparisons.
tems are often larger when the amount of processing increases Figure 1 illustrates the age-related slowing phenomenon with
(Salthouse, 1988a). Furthermore, if there are external time lim- data from a sample of 221 adults between 20 and 80 years of
its on the usefulness (i.e., accuracy or relevance) of the infor- age (Salthouse, 1993b, Study 1) on two paper-and-pencil per-
mation or decreases in its availability because of displacement ceptual speed tasks (i.e., Letter Comparison and Pattern Com-
or decay, then qualitative impairments in certain types of pro- parison, described subsequently). The vertical axis in this figure
cessing can also be expected. It is also noteworthy that an im- is the average of the two perceptual speed measures expressed in
portant characteristic of a reduction in processing speed exhib- standard deviation units, with higher scores representing faster
ited by computational models is graceful degradation of perfor- performance. These results are typical of many in the literature
mance (Salthouse, 1985b, 1988a). That is, reductions in the in that a strong systematic relation is usually found in which
speed of processing seldom result in the total or catastrophic increased age is associated with a largely monotonic, and ap-
loss of a particular kind of processing but, instead, tend to lead proximately linear, decrease in speed of performance.
to a broad or diffuse reduction in the efficiency of many types Because the processing speed construct is fundamental to the
of processing. At least in a relatively general manner, therefore, theory, it is important to consider how this construct has been
analyses of speed manipulations in computational models are assessed in recent studies. Several criteria have been proposed
consistent with the assumptions of the processing-speed theory. to guide the selection of measures used to assess processing
In summary, two mechanisms have been postulated to ac- speed (e.g., Salthouse, 1985b, 1991c, 1992b). For example, one
 PROCESSING-SPEED THEORY 407

same) or a D (for different) on the line between the two mem-

z-1.59-.03 (Age), r z =.371
bers of the pair and to complete as many of the items as possible
within a specified time (usually 30 s).
Two additional paper-and-pencil tests were designed to in-
volve minimal cognitive operations, but with stimulus and re-
sponse requirements similar to the perceptual speed tests. One
measure (Digit Copying) assesses how quickly individuals can
I 0 copy digits, and another (Boxes) assesses how quickly they can
$ draw lines in specified locations.
Finally, in several projects processing speed has been assessed
I -<
with two computer-administered reaction time tasks. These
tasks are based on the Digit Symbol Substitution Test in that
they consist of a code table at the top of the screen and a probe
stimulus in the middle of the screen. In the Digit Symbol ver-
20 30 40 50 60 70 80 sion of the task, the code table contains pairs of digits and sym-
Chronological Age bols, and the probe stimulus consists of a single digit-symbol
pair. In the Digit Digit version of the task, the code table con-
Figure I. Relation between age and a composite measure of process- tains pairs of identical digits, and hence is superfluous, and the
ing speed (data from Salthouse, 1993h, Study 1). probe stimulus consists of a single pair of digits. In both tasks,
the research participant is to press one key on the keyboard
if the probe stimuli match, either with respect to associational
criterion is that the tasks used to assess processing speed should equivalence (Digit Symbol) or in terms of physical identity
be relatively simple, such that most of the individual differences (Digit Digit), and to press a different key if they do not match.
in performance are attributable to how quickly one can carry In addition to exhibiting moderate to large relations with age
out the relevant operations rather than to variations in amount (see later discussion), all of these measures have been found to
of knowledge or in other cognitive abilities. When more com- have respectable test-retest reliabilities. To illustrate, in a sam-
plex measures are used, such as lexical decision speed or read- ple of 240 adults between 19 and 82 years of age (Salthouse,
ing speed (e.g., Hartley, 1986, 1993; Hultsch, Hertzog, & 1994a, Study 1), the immediate test-retest correlations were
Dixon, 1990; Hultsch, Hertzog, Small, McDonald-Miszczak, .86 for Boxes, .86 for Digit Copying, .58 for Letter Comparison,
& Dixon, 1992), it is difficult to determine how much of the .73 for Pattern Comparison, .61 for Digit Digit reaction time,
variation in the measures is due to differences in the speed with and .93 for Digit Symbol reaction time. Values from a sample
which elementary cognitive operations can be executed as op- of 131 adults between 17 and 79 years of age (Salthouse, Fristoe,
posed to differences in the quality or quantity of semantic Lineweaver, & Coon, 1995, Study 2) were .93 for Boxes, .93
knowledge or differences in the level of more general verbal abil- for Digit Copying, .60 for Letter Comparison, .78 for Pattern
ities. However, the speed measure should not merely represent Comparison, .69 for Digit Digit reaction time, and .89 for Digit
input and output processes or sensory and motor processes, or Symbol reaction time.
else it may not reflect the duration of relevant cognitive opera-
tions. Finally, as with the assessment of any theoretical con- Hypothesis 1: Age-related slowing is a broad phenomenon and
struct, it is generally desirable that the construct be evaluated is not simply attributable to specific and independent processing
with several measures to minimize the specific variance associ- deficits.
ated with single measures and to emphasize the common, con- A key hypothesis of the processing-speed theory is that age-
struct-relevant variance. Reliance on multiple measures also related speed differences of the type just discussed originate at
has the advantage of increasing the reliability of the assessment least partially because of a small number of fairly general or
because of aggregation (Rushton, Brainerd, & Pressley, 1983). common factors rather than exclusively from a large number of
Much of the research described subsequently has used vari- specific and local factors. One reason for the assumption of a
ous combinations of seven measures to assess processing speed. substantial shared or common influence is that many measures
One measure is the score on the Digit Symbol Substitution Test of processing speed have been found to be related to increased
(Wechsler, 1981). This is a paper-and-pencil test consisting of a age, and not merely those restricted to a few tasks or to a few
code table with pairs of digits and symbols and rows of double types of cognitive operations. More important, evidence dis-
boxes with a digit in the top box and nothing in the bottom box. cussed later indicates that the age-related influences on many
The task for the research participant is to refer to the code table speed measures are not independent but, instead, have consid-
to write the symbol in the bottom box that is associated with erable shared or common variance.
the digit in the top box. Performance on the test is represented There are two primary implications of the hypothesis that
by the number of correct symbols written in 90 s. common age-related influences contribute to measures of pro-
Two perceptual speed measures require comparisons of pairs cessing speed. The first is that it should be possible to predict the
of letters (Letter Comparison) or pairs of line patterns (Pattern age differences in particular speed measures from knowledge of
Comparison). In each case, the paper-and-pencil test form con- the age differences in other speed measures. The second im-
sists of pairs of items with a horizontal line between the mem- plication is that the age-related effects in different speed mea-
bers of the pair. The task for the participant is to write an S (for sures are not expected to be independent, rather, they are as-
408 SALTHOUSE

sumed to have a considerable amount of shared age-related age-related influences on the target variable (Salthouse, 1991c,
variance. Analytical procedures based on the examination of 1992b, I992d).
systematic relations and on statistical control techniques can be A similar method was applied in a recent study by Salthouse
used to examine these implications. and Kersten (1993), and it resulted in the elimination of most
of the age-related differences across a variety of speed measures.
Furthermore, very similar results were obtained even when the
Analyses of Systematic Relations
adjustment was based on different types of speed tasks (i.e.,
In recent years, there has been considerable interest, and con- those derived from paper-and-pencil procedures rather than
troversy, regarding the existence and interpretation of system- from reaction time tasks). This outcome not only is consistent
atic relations between mean levels of performance in different with the existence of the hypothesized common speed influence
age groups (e.g., Cerella, 1985, 1990,1991,1994;Fisk& Fisher, but suggests that, for some measures, nearly all of the age-re-
1994; Fisk, Fisher, & Rogers, 1992; Hale, Lima, & Myerson, lated influences may be attributable to the common or general
1991; Laver & Burke, 1993; Madden, Pierce, & Allen, 1992, factor.
1993; Mayr & Kliegl, 1993; Myerson, Hale, Wagstaff, Poon, & One possible objection to the adjustment of scores by the pa-
Smith, 1990; Myerson, Wagstaff, & Hale, 1994; Perfect, 1994). rameters of the systematic relation is that this method may not
The method of examining age-comparative data in which the be very sensitive to specific age-related effects that are small rel-
task or condition means of one group are plotted against those ative to any general age-related influences that might exist (e.g.,
of another group was originally described by Brinley (1965), Fisk et al., 1992). Although this concern is valid, exceptions to
who used it to express relations with both accuracy measures the general pattern were detected in both the Madden et al.
and speed measures. Most subsequent researchers using this (1992) and the Salthouse and Kersten (1993) studies. For ex-
method have focused on speed measures, and heated debates ample, in the Salthouse and Kersten (1993) study, the time
have arisen concerning the meaning of these relations. One of taken by older adults to perform the Digit Symbol reaction time
the major issues of contention is whether the primary contribu- task after an opportunity to learn the associations between
tion of portraying condition means of two groups as a function digit-symbol pairs was greater than that expected from the sys-
of one another is to illuminate global age-related influences or tematic relation. According to the reasoning underlying the an-
to obscure specific age-related effects. alytical method, therefore, it can be inferred that specific or
From the perspective of the processing-speed theory, system- unique age-related influences contributed to the age differences
atic relations are interesting primarily because of their potential on this measure, in addition to the general or common influ-
to generate estimates of the relative contributions of general and ences that were postulated to be responsible for the systematic
specific, or common and unique, age-related influences. That relation.
is, given certain assumptions, the relations between the mean It is important to note that the relations between the mean
levels of performance in two age groups may allow a distinction performance of young and old adults need not be described by
between common and unique age-related effects (Salthouse, a single function to produce moderate to high levels of predict-
1992b, 1992d). The rationale has been described as follows: ability. That is, although some researchers have relied on the
number of distinct quantitative functions relating the perfor-
Only if at least some of the age-related effects on each variable were mance of young and old adults as the basis for distinguishing
determined by a factor common to other relevant variables does it between general and specific age-related effects or between sin-
seem reasonable to expect age differences on one variable to be
gle and multiple speed factors (e.g., KJiegl, Mayr, & Krampe,
related to age differences on other variables. "General" or "com-
1994; Lima, Hale, & Myerson, 1991: Mayr & KJiegl, 1993; My-
mon" in this context thus implies lack of independence, in the
sense that knowing the magnitude of the age differences on one erson et al., 1990), distinctions among alternative functions are
variable provides information about the magnitude of the age only of secondary interest from the current perspective. The
differences on other variables. (Salthouse, 1992d, p. 330) question of primary importance in the processing-speed theory
is the extent to which the age-related effects in some variables
One application of systematic relation analyses to distinguish are independent of, or not predictable from, the age-related
common and unique age-related influences was described by effects in other variables. If there is considerable independence
Madden et al. (1992; see also 1993). These investigators used and lack of predictability, then specific influences would be in-
the performance of young adults and the systematic relations ferred to predominate over general or common influences. In
from a set of variables to predict the mean values of older adults contrast, if the variables were found to share a large proportion
in a given condition. The method is based on the assumption of their age-related variance, and if the age differences in some
that the systematic relation reflects the influence of the common variables were highly predictable from the age differences in
or general speed factor for that sample and those variables. If other variables, then general or common influences would be
this assumption is valid, then adjusting each score of the mem- inferred to be of greater relative importance.
bers of one age group by the parameters of the function relating This argument can be illustrated with data from two condi-
the mean scores in the two groups can be interpreted as remov- tions in a speeded verification arithmetic task (Salthouse &
ing the common or general influence. If the adjustment does in Coon, 1994, Study 2). Between zero and seven arithmetic oper-
fact eliminate the effects of the common or general factor, then ations were presented in this study in either a sequential condi-
the magnitude of the remaining differences between the original tion (e.g., 3 + 2 - 4 = 2; false) or a hierarchical condition in
scores in one group and the adjusted scores in the other group which temporary preservation of intermediate products was re-
provides an estimate of the contribution of unique or specific quired (e.g., [5 - 3) + 4] - 1 = 4; false).
 PROCESSING-SPEED THEORY 409

20 Sequential

18 Hierarchical

Old = -.34 + 1.96(Young), r*= .99

16
O
O>
14
.££•
<D
12
P
<D
in 10
O
CL
<f> 8 Old = .55 + 1.23(Young), r*= .99
<D
CC
2 6
O
4

0
4 6 8 10 12 14 16 18 20

Young Response Time (sec)

figure 2. Mean time of older adults as a function of mean time of young adults in sequential and hierar-
chical arithmetic tasks (data from Salthouse & Coon, 1994, Study 2).

The systematic relation functions for the sequential and hier- operating in these variables. However, the important point for
archical arithmetic data in this study are portrayed in Figure 2. the current argument is that the magnitude of the age differ-
Separate regression lines are illustrated for the conditions in ences in nearly every variable was greatly reduced after adjust-
each task because the interaction of condition (sequential or ing for the estimated contribution of a common speed factor.
hierarchical) with time of the young adults was statistically sig- On average, therefore, there appears to be moderate predictabil-
nificant(cf. Salthouse, 1985a, 1991c, 1992d, 1992f). ity of the age differences in certain speed measures from knowl-
Because all participants in this study also performed two pa- edge of the pattern of age differences in other speed measures.
per-and-pencil perceptual speed tasks (Letter Comparison and Because there were different systematic functions in Figure 2,
Pattern Comparison) and two reaction time tasks (Digit Digit it is informative to examine the degree to which the age-related
and Digit Symbol), ratios of perceptual speed and reaction time variance in the measures from these conditions was indepen-
speed measures were computed to serve as additional estimates dent. That is, even though there are distinct quantitative re-
of the hypothesized general speed factor for these participants. lations for the sequential and hierarchical measures, it is possi-
Table I summarizes the actual (original) group differences for ble to examine the amount of age-related variance the measures
all speed measures and the differences between actual values from the two functions have in common. Estimates of the
and predicted values for older adults after adjustments for the shared age-related variance can be derived by determining the
general influence according to each of the equations shown in proportion of age-related variance in the measures from the
the notes to the table. The values in the table are d units, which condition with the larger age differences (i.e., hierarchical) that
correspond to the mean difference between the groups divided is shared with the measures from the condition with the smaller
by the pooled standard deviation (Cohen, 1988). age differences (i.e., sequential). Moreover, these computations
Inspection of Table 1 reveals that almost all of the differences can be carried out both for the mean values across each condi-
involving an adjustment are smaller than the initial differences tion and for pairs of measures with the same number of arith-
(i.e., the mean effect size for the original scores was 1.75 units, metic operations. Results of the computations are summarized
and the grand means for all other conditions, displayed in the in Table 2.
table notes, were between —0.51 and 0.40). The degree of pre- Two points should be noted about the entries in this table.
diction is certainly not perfect, and examination of the mea- First, estimates of the shared age-related variance are moder-
sures with consistently large residual differences may be infor- ately high, with an estimate of .871 based on the computations
mative about the nature of the specific age-related influences of the mean values and a mean across estimates from different
410 SALTHOUSE

Table 1
Effect Sizes Before and After Adjustment for the Influence of a General Speed Factor

Number of arithmetic operations

Condition

Sequential
Original 3 2.15* 2.05* 1.59* 1.56* 1.43* 1.28* 1.09* 1.29* 1.56
Equation 1" 0.36 0.28 -0.47 -0.66* -0.95* -0.92* -1.06* -0.88* -0.54
Equation 2' 0.64* 0.57* -0.18 -0.36 -0.63* -0.62* -0.78* -0.60* -0.25
Equation 3d -0.25 0.17 0.04 0.04 -0.05 -0.01 -0.13 0.08 -0.01
Equation 4e 0.41 0.08 -1.01* -1.27* -1.65* -1.62* -1.71* -1.53* -1.04
Equation 5 r 1.16* 0.75* -0.36 -0.64* -1.01* -1.03* -1.18* - 1 .02* -0.42

Hierarchical
Original" 2.24* 2.30* 2.08* 1.76* 1.92* 1.76* 1.82* 1.67* 1.94
Equation l b 0.39 0.54* 0.30 0.04 0.74* 0.49 0.69* 0.43 0.45
Equation 2C 0.69* 0.81* 0.56 0.29 0.93* 0.69* 0.86* 0.62* 0.68
Equation 3d -0.19 0.48 0.79» 0.68' 1.29* 1.11* 1.27* 1.08* 0.81
Equation 4* 0.40 0.29 -0.19 -0.53 0.24 -0.07 0.18 -0.13 0.02
Equation 5f 1.17* 0.97* 0.37 0.01 0.67* 0.38 0.57 0.29 0.55

" M d = 1.75. b Ratio of perceptual speed: Old = Young X (1/.62) (M d = 0.05). c Ratio of reaction time speed: Old = Young X 1.51 (M d = 0.22).
d
Regression equation for sequential condition: Old = (\bung X 1.23) -f .55 (Md~ 0.40). e Regression equation for hierarchical condition: Old =
(YoungX 1.96)-.34(A/rf = -0.51). ' Regression equation for both conditions: Old = (YoungX 1.72) - .37 (Md = 0.07).
*p<.01.

numbers of operations of .817. Despite the quantitative differ- tribution, of a common age-related speed factor. Although those
ence in the functions, therefore, the values representing the two parameters have descriptive value and may be useful for esti-
functions shared an average of more than 80% of their age-re- mating the contribution of the hypothesized general factor(s)
lated variance.2 Second, the proportions of shared age-related on a particular variable, they do not directly indicate the extent
variance tend to decrease as the number of arithmetic opera- to which the age-related influences in different variables are dis-
tions increases. This decrease may reflect an increased involve- tinct and independent of one another.
ment of novel or distinct age-related processes (perhaps related In summary, the considerable literature documenting the ex-
to working memory and the temporary preservation of infor- istence of systematic relations between mean times of young
mation while processing other information) when the problems and old adults is viewed as consistent with the hypothesis of a
contain four or more operations. Even when it appears that general age-related slowing factor. The current perspective
other age-related influences are operating, however, it is impor- differs from that espoused by other theorists, however, in that
tant to note that more than 50% of the age-related variance is analyses of systematic relations are not necessarily assumed to
shared with speed measures that are presumably unaffected by be informative about the number of independent age-related in-
those influences.3 fluences; instead, they are postulated to be primarily useful as
The data reported in Figure 2 and in Tables 1 and 2 suggest
that the quantitative parameters of the systematic functions are 2
Mayr and Kliegl (1993) reported that slightly more than 6% of the
not necessarily informative about the existence, or relative con- variance in measures from tasks in volvingcoordinative complexity was
independent of that in measures from tasks involving sequential com-
plexity. However, because they did not report the total age-related vari-
ance in the coordinative complexity measures, estimates of the propor-
Table 2 tions of shared age-related variance could not be derived from the in-
Estimates of Proportions of Age-Related Variance formation in their article. Figure 8 in that article also illustrates results
in the Hierarchical Arithmetic Task of an adjustment analogous to those reported in Table 2 of this article
in which a similar pattern was apparent (i.e., under-prediction of the
Age-related variance effects in the complex condition from an adjustment based on results in
Number of Proportion of
the simple condition).
operations Alone After sequential variance shared 3
A very similar pattern of results was obtained in a contrast of se-
.987 quential arithmetic performed alone and performed while also remem-
0 .594 .008
1 .613 .013 .979 bering four letters (Salthouse ct al., 1995). That is, the regression
2 .547 .069 .874 equations for the two sets of variables differed significantly in slope (i.e.,
3 .478 .042 .912 Old = 0.47 + 1 . 1 1 [Young], r2 = .99, for arithmetic alone, and Old =
4 .503 .131 .740 0.33 + 1.38 [Young], r2 = .99, for arithmetic with concurrent letter
5 .459 .084 .817 memory), the magnitude of the age difference in d units was greatly
6 .470 .202 .570 reduced after adjustment for the influence of the common factor, and a
7 .417 .142 .659
large proportion of the age-related variance in the concurrent arithme-
M .518 .067 .871
tic measures was shared with the single-task arithmetic measures.
 PROCESSING-SPEED THEORY 411

the variables will be independent, and the ratio b/(b + d) will

be small.
Both correlation and hierarchical regression procedures can
be used to derive estimates of the proportions of variance illus-
trated in Figure 3. That is, the square of the correlation between
age and the criterion variable corresponds to the proportion of
the criterion variance shared with age (i.e., [b + d]/[a +/+ b
+ d ] ) . Hierarchical regression techniques, in which the vari-
ance in the controlled variable is removed before the relation of
age to the criterion variable is examined, can be used to derive
Figure 3. Illustration of regions of variance for age and two variables. the square of the semipartial correlation (i.e., d/[a +/+ b +
The circles represent the total variance in the variables, and the regions
d]). Finally, subtracting the second quantity from the first, and
of overlap correspond to proportions of shared variance.
then dividing by the first, yields an estimate of the proportion of
age-related variance in the criterion variable that is shared with
the controlled variable (i.e., b/[b + d ] ) .
a means of identifying variables with potentially specific age- These analyses can be illustrated with an example reported in
related influences. Salthouse (1994c). The primary data in that report were from a
large (N = 910) sample of adults across a wide age range. The
two speed measures were the Digit Symbol Substitution Test
Statistical Control Procedures
and a composite perceptual speed measure formed by averaging
Because many speed measures have been found to have neg- z scores from the Letter Comparison and Pattern Comparison
ative relations with age, a fundamental question within the pro- tasks. The correlation between the Digit Symbol Substitution
cessing-speed theory concerns the number of separate and dis- Test and composite perceptual speed variables was .73, indicat-
tinct age-related influences on speed. That is, are there many ing that 54% (i.e., .73 2 ) of the total variance in each variable
independent and unique age-related influences, or are there, in- was shared. The R2 value associated with age in the Digit Sym-
stead or in addition, a relatively small number of factors with bol Substitution Test was .289, but the increment in K2 associ-
fairly broad consequences? As noted earlier, a central hypothe- ated with age after control of the perceptual speed measure was
sis of the processing-speed theory is that some version of the only .008. It can therefore be inferred that 97.2%—[.289 -
latter interpretation is the most plausible and, hence, that there .008)/.289] X 100—of the age-related variance in the Digit
should be substantial commonality, or overlap, of the age-re- Symbol Substitution Test was shared with the composite per-
lated influences on many different speed measures. In other ceptual speed measure.
words, rather than being completely independent, it is hypothe- Because the proportions of total age-related variance in the
sized that much of the age-related variance in any given speed variables need not be identical (i.e., b + d is not necessarily
measure is shared with the age-related variance in other speed equal to b + c), estimates of the proportion of shared age-re-
measures. lated variance in two variables are not necessarily symmetric.
One way of investigating the degree of commonality among In fact, when the analyses in the data set just described were
speed measures is through use of statistical control methods. reversed, the age-associated R2 value in the perceptual speed
The logic of statistical control procedures in the present context measure was .412, but the increment in R2 associated with age
can be described by reference to Figure 3 (see also Salthouse, after control of the Digit Symbol Substitution Test was .086.
1992b, 1992d, 1994c, for further discussion of the rationale). This leads to an estimate that 79.1%—[.412 - .086J/.412] X
This figure represents the total variance, and the regions of 100—of the age-related variance in the composite perceptual
shared or overlapping variance, in three variables. Note that the speed measure was shared with the age-related variance in the
proportion of variance in the criterion variable that is shared Digit Symbol Substitution Test.
with the age variable corresponds to the ratio (b + d)/(a +/+ A very similar pattern of shared age-related variance was
b + d). In a similar manner, the proportion of variance in the found in the data from an independent sample of 305 adults
criterion variable that is shared with the controlled variable cor- reported in the Salthouse (1994c) article. That is, in this sample
responds to the ratio (a + b)/(a +/+ b + d). However, it is also 92.3% of the age-related variance in the Digit Symbol Substitu-
apparent in Figure 3 that the proportion of age-related variance tion Test was shared with the perceptual speed measure, and
in the criterion variable that is shared with the controlled vari- 77.3% of the age-related variance in the perceptual speed mea-
able corresponds to the ratio b/(b + d). It is this latter quantity sure was shared with the Digit Symbol Substitution Test. In con-
that is of greatest interest in the current context because the trast, because the correlation between the two measures was .68,
prediction from the processing-speed theory is that the age-re- only 46% of the total variance in each measure was shared with
lated influences on many speed variables are not independent the other measure. The results from these two data sets there-
and, hence, that the variables will share a large proportion of fore indicate that a very large percentage of the age-related vari-
their age-related variance. In other words, the ratio b/(b + d) is ance in at least these particular speed measures was shared, and
predicted to be relatively large when the two variables both re- hence relatively little was independent or unique to each
flect speed of processing. In contrast, if separate and distinct measure.
age-related influences are responsible for the age differences in Procedures similar to those just described have also been con-
every speed measure, then most of the age-related variance in ducted on a variety of different speed measures from two sepa-
412 SALTHOUSE

Table 3
Proportions of Shared Age-Related Variance in Speed Measures: Salthouse (1993d; N = 305)

Criterion variable

Controlled variable Ager

1. Horizontal Marking _ .975 .873 .763 .468 .467 .942 .745 .449 .462 .590 -.38
2. Vertical Marking .993 .879 .780 .535 .511 .981 .855 .492 .512 .617 -.40
—
3. Digit Copying .926 .894 — .919 .588 .583 .999 .909 .625 .587 .705 -.41
4. Letter Copying .980 .9(59 .994 .738 .793 .942 .999 .781 .728 .870 -.49
—
5. Digit Comparison .838 .875 .915 .877 — .989 .731 .909 .898 .858 .943 -.55
6. Letter Comparison .791 .806 .867 .881 .963 .865 .964 .852 .789 .908 -.53
—
7. Digit Transformation .439 .456 .503 .428 .399 .395 — .855 .422 .332 .433 .23
8. Letter Transformation .318 .369 .388 .398 .369 .380 .885 — .363 .301 .379 -.24
9. Letter Comparison .723 .731 .861 .826 .807 .808 .904 .982 — .827 .820 -.51
10. Pattern Comparison .912 .925 .976 .941 .930 .913 .808 .964 .973 — .885 -.59
I I . Digit Symbol .892 .881 .939 .919 .870 .877 .846 .982 .828 .720 — -.51

rate data sets; results are summarized in Tables 3 and 4. Table is obviously limited by the total age-related variance in each
3 contains results from 11 speed measures obtained from the variable.
305 adults reported in the Salthouse (1993d) article. All mea- Because all of the measures in Tables 3 and 4 are single vari-
sures from that study were derived from paper-and-pencil tests. ables, the age-related variance is not easily partitioned into that
Table 4 contains results from 4 paper-and-pencil tests and 2 re- attributable to the hypothesized construct, in this case a com-
action time tests from a total of 744 adults who had participated mon speed factor, and that specific to the particular methods,
in one of three recent studies (i.e., Salthouse, 1994d, Studies 1 materials, and measures. Nevertheless, the results summarized
and 2; Salthouse, 1994a, Study 1 ) . in the two tables reveal that there is substantial overlap of the
In all cases, the entries in Tables 3 and 4 correspond to pro- age-related variance in the individual measures of processing
portions of shared age-related variance computed by subtract- speed, with an average of nearly 75% of the age-related variance
ing the increment in R2 associated with age after eliminating in these variables shared with other variables. Very similar re-
the variance in the controlled variable from the total R2 associ- sults have also been reported in two recent studies. Salthouse
ated with age and then dividing this difference by the total R2 and Meinz (1995) found that an average of 86.3% of the age-
associated with age. As in the examples described earlier, these related variance was shared across 2 reaction time, 2 paper-and-
values indicate how much of the age-related variance in one pencil, and 10 vocal speed measures. And Salthouse (1996)
variable (represented in the columns) is shared, or in common found that an average of approximately 62% of the age-related
with, the age-related variance in another variable (represented variance in 19 different speed measures involving vocal, writ-
in the rows). ten, and reaction time responses was shared.
The median for the values in Table 3 was .842, and that for A second method that can be used to investigate the degree of
the values in Table 4 was .628. Only the values obtained after commonality of age-related variance in measures of speed is
the control for Digit Transformation and Letter Transformation based on a structural equation model with a single latent speed
measures were consistently lower than .5, and this may reflect construct related to all speed measures and with relations from
the fact that these two variables had relatively low correlations age to the latent speed construct and to each individual speed
with age (i.e., —.23 and —.24, respectively, as compared with a measure. Within a model of this type, estimates of the common
range of -.38 to -.59 for all other variables). The amount of age-related influence on each speed measure can be obtained
age-related variance that can be shared between two variables from the product of the path coefficients from age to the corn-

Table 4
Proportions of Shared Age-Related Variance in Speed Measures:
Earles and Salthouse (1995; N= 744)

Criterion variable

Controlled variable Ager

1. Boxes .808 .573 .521 .456 .430 -.42

2. Digit Copying .904 — .766 .590 .574 .554 -.46
3. Letter Comparison .551 .648 — .568 .544 .573 -.41
4. Pattern Comparison .831 .812 .901 .769 .747 -.56
—
5. Digit Digit reaction time .500 .531 .608 .511 — .756 .44
6. Digit Symbol reaction time .719 .770 .906 .749 .985 — .56
 PROCESSING-SPEED THEORY 413

mon speed factor and from the common factor to the individual ence in inspection or processing time in the two types of prob-
speed measures. The direct path coefficients from age to the lems, particularly at the third element in the sequence, where
individual speed measures serve as estimates of the specific or the problems are first distinguishable. Mean inspection times
unique age-related influences on the measures. for correct trials for young and old adults in the two types of
Salthouse (1996) recently applied this method to four inde- problems are illustrated in Figure 4.
pendent data sets and, in all cases, found very similar results. It is apparent in Figure 4 that older adults were slower than
For every variable, the estimate of the common or shared age- young adults across every item in the sequence for both types of
related influence was much larger than the estimate of the problems. Of particular interest are the much longer times in
unique or specific age-related influence. Moreover, the unique the third, fourth, and fifth elements in the second-order prob-
age-related influence was significantly greater than zero for only lems because these durations presumably reflect the time
18 of the 53 variables across the four data sets. needed to infer or abstract relations among items. Because these
It is important to point out that a general or common age- durations are longer for older adults than for young adults, some
related influence is also likely to be operating with measures of the lower accuracy of older adults on the second-order prob-
often postulated to reflect the duration of discrete and specific lems (i.e., 35.8% correct, as compared with 68.6% correct for
cognitive processes. In fact, results from several analyses suggest young adults) may be attributable to a greater probability of
that 50% or more of the age-related variance in measures of forgetting early items in the sequence during the longer period
purportedly specific information-processing components is at- needed to identify the relations among elements.
tributable to a common or general speed factor. Analyses lead- The first two rows in Table 5 contain the values leading to
ing to this conclusion can be illustrated with data from an arti- estimates of the proportion of age-related variance in the task-
cle by Salthouse and Prill (19 8 7) involving a series completion specific speed measures that was shared with a speed measure
task. In the condition of greatest interest in the present context, from a separate task (i.e., Digit Symbol Substitution Test
each element in the problem was presented successively, and the score). Note that statistical control analyses of the type de-
time taken by the participant in examining each element was scribed earlier yield estimates that 49.1 % of the age-related vari-
monitored by a computer. Two different types of problems were ance in the abstraction measure, and 62.5% of the age-related
distinguished on the basis of the abstractness of the relations variance in the mean inspection time measure, was shared with
among elements. Problems with first-order relations among el- the Digit Symbol speed measure.
ements consisted of a simple continuation sequence (e.g., 2-4- The remaining entries in Table 5 are based on studies with a
6-8-10-??). In contrast, problems with second-order relations mixture of task-specific speed measures. Two distinct patterns
had the invariance or constancy at the second level of abstrac- are evident in these data. One is similar to that described earlier
tion, in the difference among differences (e.g., 2-3-5-8-12-??). in that the age-related variance in the task-specific measure was
A measure of abstraction time can be derived from the differ- greatly reduced after control of the variance in another speed

Young Old

1st Ordar —•— —9—

2nd Order --•-- --O--

CO
T3

O
O
Q)
U)

1 2 3 4 5 6

Element Number
Figure 4. Time to inspect successive series completion elements for young and old adults in first-order and
second-order abstraction problems (data from Salthouse & Prill, 1987, Study 2).
414 SALTHOUSE

Table 5
Estimates of Shared Age-Related Variance in Different Speed Measures

Age-related variance Proportion of

variance
Study and criterion variable Alone After control of shared with

Salthouse & Prill ( 1 987), series

completion (N = 48) Digit Symbol Digit Symbol
Mean inspection time .518 .194 .625
Difference: 2nd order minus 1st order .228 .116 .491

Salthouse ( 1 987), geometric analogies

(A' =48) Digit Symbol Digit Symbol
Mean inspection-decision time .505 .054 .893
Slope of time-element: complete problem .280 .006 .979
Slope of time-element: first 2 terms .368 .053 .856

Salthouse & Coon ( 1 993), reordered letter

memory span, Study 1 (Ar= 55) RTS RTS
Mean recall time .376 .034 .910
Reorder time .187 .001 .999

Salthouse & Coon ( 1 993), reordered letter

memory span, Study 2(N = 1\) RTS RTS
Mean recall time .236 .013 .945
Reorder time .063 .001 .984
Encoding time .274 .101 .631

Salthouse & Kersten ( 1 993), symbolic

arithmetic (N = 104) PS RTS DART PS RTS DART
Symbol arithmetic RT .518 .068 .015 .047 .869 .971 .911
Digit arithmetic RT .450 .050 .001 —' .889 .998 —'

Salthouse & Coon ( 1 994), Study 1 , subtraction

RT (N = 240) PS RTS NBRT PS RTS NBRT
Borrow RT .051 .001 .004 .001 .980 .922 .980
No borrow RT .087 .001 .002 —" .989 .977 —'

Salthouse & Coon ( 1 994), Study 2, arithmetic

with 1 to 7 operations (N = 80) PS IRT Intercept PS IRT Intercept
—• a
Sequential arithmetic intercept .193 .034 .001 .824 .995
Slope .169 .036 .032 .334 .787 .811
b
Hierarchical arithmetic intercept .083 .099 .013 —• .843 "

Slope .401 .188 .026 .169 .531 .935 .579

Salthouse et al. ( 1 995), arithmetic with 1 to 4

operations (N = 131) PS IRT Intercept PS RT Intercept
Single-task arithmetic intercept .167 .074 .091 —B .557 .455 —•
Slope .018 .016 .003 .096 .111 .833
Dual-task arithmetic intercept .039 .024 .000 —a .385 .999 —a
b
Slope .069 .001 .001 .162 .986 .986

Salthouse ( 1 994d), digit symbol with 3, 6, and 9

digit-symbol pairs. Study 1 (A7 = 246) PS DDRT Intercept PS DDRT Intercept
Intercept .284 .058 .024 a
.796 .915 — fl
b
Slope .115 .008 .046 .124 .930 .600

Salthouse (1994d), digit memory search with 1 to

4 items, Study 2 (N = 258) PS RTS Intercept PS RTS Intercept
Intercept .164 .038 .004 —• .768 .976
—a
Slope .002 .000 .000 .022 1.000 1.000 —

Salthouse ( 1 994d), letter memory search with 1 to

4 items, Study 2 (N= 258) PS RTS Intercept PS RTS Intercept
Intercept .145 .026 .002 —" .821 .986
—a
Slope .001 .003 .000 .003 1.000
 PROCESSING-SPEED THEORY 415

Table 5 (continued)

Age-related variance Proportion of

Study and criterion variable Alone After control of shared with

Salthouse & Meinz (1995), Stroop interference

(incongruent — neutral; N = 242) PS RTS PS RTS
Color .217 .033 .028 .848 .871
Number .031 .002 .002 .935 .935
Position .073 .001 .001 .986 .986

Salthouse (1996), Stroop interference

(incongruent — neutral; N = 172) PS RTS PS RTS
a
Color .361 .067 —* .814

Note. RTS = reaction time speed; PS = perceptual speed; DART = digit arithmetic reaction time; RT = reaction time; NBRT = no borrow reaction
time; IRT = identification reaction time; DDRT = Digit Digit reaction time.
• Variable not available or not relevant in the analysis. " Estimates of shared variance not meaningful as a result of an apparent suppression
relationship because the age-related variance increased rather than decreased after control of the other variable.

measure (i.e., symbol arithmetic reaction time, digit arithmetic common, rather than being completely independent and dis-
reaction time, borrow reaction time, no borrow reaction time, tinct. Moreover, the age-related influences are apparently not
sequential arithmetic intercept, Digit Symbol intercept, mem- restricted to those reflecting overall performance in simple tasks
ory search intercept, and Stroop color interference). The sec- because similar patterns are evident with measures presumed
ond pattern is that the age-related variance in the criterion vari- to reflect the duration of specific cognitive operations in moder-
able increased rather than decreased after control of the vari- ately complex tasks. These results are thus consistent with the
ance in the other speed variable (i.e., sequential arithmetic hypothesis of the processing-speed theory that a small number
slope, hierarchical arithmetic intercept, single-task and dual- of common factors contribute to the age-related differences in
task arithmetic slope, Digit Symbol slope, and memory search many speed measures.
slope). I nstances with the second pattern indicate that statistical In summary, two important implications of the hypothesis
control need not always reduce the amount of age-related vari- that age-related slowing is a general phenomenon appear to have
ance because it can also "release" the age-related variance that convincing support. The implication that the direction and
had been suppressed because of a negative relation between the magnitude of the age differences in certain speed measures are
two speed measures. For example, the intercept could operate predictable from knowledge of the age differences in other speed
as a third variable that obscures the relations between age and measures is supported by the results described earlier involving
the slope unless its effects are taken into consideration. Even systematic relations. And the implication that the age-related
though the age-related variance in these cases increased rather influences on different speed measures are not independent is
than decreased after statistical control of the other measure, it supported by the finding that many speed measures share a con-
is important to note that the results are still consistent with the siderable proportion of their age-related variance. This evidence
interpretation that the age-related influences on the speed mea- is clearly consistent with the proposal that a small number of
sures were not independent. Statistical control can alter the age common factors contribute to the age-related influences in
relations, in either a negative or a positive direction, only if at many measures of speed.
least some of the age-related variance in the measures is shared, Because there has been frequent confusion on the issue of
and not all is unique or specific. general slowing, it is important to be explicit about what is not
A very similar pattern in which a large proportion of age- implied from the current perspective. Although age-related
related variance in presumably specific speed measures was slowing is assumed to be ubiquitous, it is not necessarily as-
shared with more general processing-speed measures was also sumed to be universal, uniform, or unitary. Each of these latter
reported by Salthouse (1996). In that study, factor analysis pro- characteristics is considered in turn.
cedures were used to derive an estimate of the general speed First, the hypothesized common or general speed factors are
factor, and the specific measures represented the time to search not necessarily universal. That is, some measures may not ex-
a code table and substitute items, to search and retrieve an item hibit age-related slowing because it is not assumed that every
from memory, to reorder items in memory, to articulate or re- measure scaled in units of time is affected by common or gen-
hearse items, and so forth. Statistical control of the general eral influence(s) (Kail & Salthouse, 1994; Salthouse, 1985b,
speed factor substantially reduced the age-related variance for 1992b, 1996; Salthouse & Somberg, 1982). Because not all
all of the measures with significant age relations, implying that speed measures are postulated to be influenced by the hypothe-
many of the age-related influences in the specific measures were sized common speed factor(s), a discovery that some speed
shared with the general speed factor. measures have little or no age differences would not be incon-
The results just discussed, and summarized in Tables 3-5, sistent with the hypothesis that common factors affect many
lead to the conclusion that a moderate to large proportion of the other speed measures. Ultimately, of course, explanations are
age-related variance in many speed variables is shared, or in needed to account for why some speed measures are related
416 SALTHOUSE

Study 1

Study 2

0.8

0.6

0.4
I
o
O

0.2

-0.2 0 0.2 0.4 0.6

Correlation with Word Knowledge

Figure 5. Functions illustrating the relation between the magnitude of the correlation with age and the
correlation with a measure of word knowledge for seven speed measures in studies conducted by Salthouse
(1993c).

to age, whereas others are not. For the purpose of the current creased. One interpretation of this pattern is that speed-depen-
argument, however, the important point is that universality is dent processing requirements were reduced as the amount of
not a necessary concomitant of the existence of factors with rel- relevant knowledge increased, thereby resulting in a decrease in
atively broad influences. the relation of age to the measure. For example, when more
A second assumption of the processing-speed theory is that knowledge is available, fewer transformations or novel process-
age differences on different speed measures should not neces- ing operations might be required because much of the relevant
sarily be expected to be uniform. Instead, the age-related effects information (or "solutions") may already exist in the individu-
can be expected to vary in magnitude because of the operation al's knowledge system.
of other influences, even if a common underlying mechanism is The typing results (Salthouse, 1984) and the data in Figure 5
involved (cf. Salthouse, I992b, 1992d; Salthouse & Coon, are merely two illustrations that the relation between age and
1994). Considerable evidence indicates that a variety of factors measures of speed varies as a function of other factors. Results
contribute both to the absolute level of speeded performance of this type can be viewed as confirming the well-accepted prin-
and to the relations between age and measures of speeded per- ciple that virtually every performance measure has multiple de-
formance. As an example, research on typing has revealed that terminants. That is, one-to-one relations between a particular
the relations between age and measures of speed (i.e., interkey hypothetical process and a behavioral variable are extremely
interval) systematically vary as a function of the amount of pre- rare, and thus most variables can be assumed to be affected by
view of to-be-typed text available during typing (cf. Salthouse, several different influences. Identical age relations on different
1984, Figure 4). In the Salthouse (1984) project, the correla- speed measures should therefore not be expected unless the
tion between age and median interkey interval was about .5 with measures have the same determinants, with exactly the same
a visible window of one character, and it decreased to near 0 weightings or relative importance, and the measures are equiv-
with unlimited preview. alent in reliability and sensitivity.4
Another illustration of the influence of other factors on age-
4
speed relations is available in the results of a recent project Age-speed relations will also vary according to the reliability of the
(Salthouse, 1993c), portrayed in Figure 5. The variables in the speed measure because large age relations cannot occur if there is little
two studies in this project were several measures of perceptual systematic variance in the measure that is available for association with
other variables. As an example, the finding of Madden etal. (1993) that
speed (e.g., Letter Comparison, Pattern Comparison, and Digit
priming difference scores (i.e., priming reaction time minus unrelated
Symbol Substitution Test) and measures from other timed tests
or neutral reaction time) had reliabilities of near zero suggests that lack
presumed to require word knowledge (e.g., various word flu- of systematic variance may be one reason for the low correlations be-
ency tasks and tasks such as anagrams). It is obvious in both tween age and measures of priming. Also consistent with this interpreta-
studies that the magnitude of the relations between age and the tion are the reports by Salthouse (1994a) of correlations in two studies
speed measure decreased as the involvement of word knowledge of .51 and .85 between the reliability of the measure and the absolute
(assessed by scores on two vocabulary tests) in the task in- magnitude of the correlation of the measure with age.
 PROCESSING-SPEED THEORY 417

A third assumption of the processing-speed perspective is processing speed mediates some of the age-related effects on
that there is not necessarily a single, or unitary, speed factor. measures of cognitive performance.
Instead, several common factors could exist, as long as the num- As expected from the processing-speed theory, path analyses
ber is substantially smaller than the number of relevant speed have revealed strong relations between age and speed, moderate
measures (Kail & Salthouse, 1994; Salthouse, 1992b, 1995c). relations between speed and various measures of cognitive func-
Results of studies conducted by Earles and Salthouse (1995), tioning, and either a weak relation or no direct relation between
Hertzog (1989), Salthouse (1993c, 1994d), Tomer and Cun- age and measures of cognitive functioning. Examples with var-
ningham (1993), and White and Cunningham (1987) suggest- ious cognitive measures as the criterion variable have been de-
ing that several speed factors can be distinguished in samples of scribed in Lindenberger, Mayr, and Kliegl (1993) and Salthouse
young and old adults are therefore not incompatible with the (1991b, 1992b, 1992f, 1993d, 1994a), and those with working
present perspective. memory as the criterion cognitive construct have been reported
An analogy may help to illustrate this point. Several different in Salthouse (1994c) and Salthouse and Babcock (1991).
types of speed affect the performance of a computer, including A recent project by Salthouse (1994d) can be used to illus-
the processor clock rate; the speed of specialized mathematical trate the features, and potential contributions of the path-ana-
or graphics co-processors; hard disk access time; input rate lytic approach. Results from two independent studies with
from devices such as keyboards, scanners, or modems; and out- slightly different cognitive tests were reported in this article,
put rate to devices such as display monitors, printers, plotters, with sample sizes of 246 (Study 1) and 258 (Study 2) adults
or modems. Nevertheless, knowledge of a small number of ranging from 18 to 87 years of age. The participants in each
speed "factors" allows performance on an extremely large num- study performed paper-and-pencil sensorimotor speed (Digit
ber of tasks to be predicted quite accurately. A central hypothe- Copy and Boxes) and perceptual speed (Letter Comparison and
sis of the processing-speed theory is that the human cognitive Pattern Comparison) tasks, as well as tests of memory, reason-
system can also be conceptualized as having a relatively small ing, and spatial ability. Because the cognitive tests were admin-
number of speed factors that are related to age and that contrib- istered on computers, it was also possible to obtain separate
ute to the efficiency of many cognitive processes. This is in con- measures of study or solution time, decision time, and decision
trast to the view, which is often implicit, that every process or accuracy from each of the cognitive measures to represent dis-
measure has an independent age-related influence. tinct aspects of cognitive performance.
A crucial point from the current perspective, therefore, is that Figure 6 illustrates the best-fitting path model from the proj-
the age relations in all speed measures should not be expected ect, based on composite measures formed by averaging z scores
to be of the same magnitude, in either absolute or proportional across the three cognitive tests in each study. The degrees of
terms. Instead, the proposal is that many, but not necessarily freedom in the model were low because the number of esti-
all, of those measures share substantial age-related variance be- mated parameters was close to the number of available covari-
cause of the influence of common determinants or factors. ances, but the model nevertheless provided a good fit to the data
(e.g., Adjusted Goodness of Fit Index model determination val-
Hypothesis 2: Processing speed functions as an important me- ues of .987 in Study 1 and .955 in Study 2). All path coefficients
diator of the relations between age and measures of cognitive in Figure 6 differed from zero by more than two standard errors,
functioning. except for the path between age and decision accuracy in Study
2. No paths not represented in the figure had coefficients sig-
Because it is assumed, and the available evidence seems con-
nificantly different from zero in either study.
sistent with the assumption, that there are general age-related
Several important results should be noted about this figure.
effects on speed of processing, it is hypothesized that there are
First, there were moderate to strong relations between age and
effects not only on the speed of many cognitive operations but
perceptual speed and between perceptual speed and decision
also on the quality of the products of those operations. Two cat-
accuracy, but there were weak relations between age and deci-
egories of evidence are relevant to the hypothesized mediational
sion accuracy. These features are all consistent with the key as-
role of processing speed in age-cognition relations: results from
sumption of the theory that a slower speed of processing par-
path analyses and results from statistical control of an index of
tially mediates the adult age differences in a variety of cognitive
processing speed.
tasks. In addition, however, Figure 6 illustrates that sensorimo-
tor speed was related to perceptual speed and decision time but
Path Analyses not to decision accuracy. Peripheral (i.e., input and output) as-
pects of speed therefore do not appear to be directly involved in
Because path models can illustrate the complete pattern of the mediation of the relations between age and decision accu-
interrelationships among variables, path analyses are valuable racy. Furthermore, study time was not related to perceptual
as a means of indicating the relations among several variables. speed, but it was related both to age and to decision accuracy.
Path models should not be considered definitive because they This pattern suggests that, with increased age, more time was
can vary in the degree to which they represent or fit the data, spent working on the items and that longer time devoted to the
and alternative structural models often provide equally good fits items was associated with higher accuracy. However, because
to the data (MacCallum, Wegener, Uchino, & Fabrigar, 1993). study or solution time was not related to perceptual speed, it
Nevertheless, path analyses can be informative about the pres- may reflect strategic or stylistic factors rather than effects asso-
ence or absence of particular relations, and hence they are of ciated with a slower speed of executing relevant processing op-
obvious relevance to theories postulating that a construct like erations. Finally, it is noteworthy that perceptual speed was re-
418 SALTHOUSE

.30/.45
.S1/.54

Figure 6. Path diagram illustrating relations among variables in two studies reported by Salthouse
(1994d). Coefficients are reported in the format Study 1 /Study 2.

lated to decision accuracy but not to decision time. This is ad- match individuals of different ages on measures of speed; as il-
ditional evidence that processing speed, as indexed by the lustrated later, however, this is not always successful, and it typ-
perceptual speed measures, affects the quality of cognitive pro- ically reduces the statistical power of the comparisons because
cessing and not simply the speed with which decisions about the of the decreased sample size. The most practical method of in-
products of the processing can be communicated. vestigating hypothesized mediational relations therefore ap-
Although only a small number of relevant path analyses have pears to be some type of statistical control procedure in which
been reported, the available results have been quite consistent the variance in an index of the hypothesized mediator is held
with the predictions of the processing-speed theory. In particu- constant by statistical methods.
lar, moderate relations have been found between age and mea- It is important to note that statistical control procedures will
sures of speed and between speed measures and measures of reduce the age-related effects on a criterion variable only if the
cognition, but weak to nonexistent relations have been found measure of the hypothesized mediator is related both to age and
between age and cognition. An additional contribution of path to the criterion variable. If the mediator is related only to age,
analyses apparent in the preceding example is that because the then there will be no effect of its control on the criterion vari-
path analyses indicate how several different variables are inter- able; if the mediator is related only to the criterion variable,
related, they are informative about how speed-mediated influ- then its control will have no effect on the relation between age
ences occur. and the criterion variable. This is evident in Figure 3, which
shows that the controlled variable can contribute to the media-
Statistical Control Procedures tion of age effects in the criterion variable only if the region of
One implication of the hypothesis that processing speed is a double overlap (i.e., Region b in Figure 3) is greater than zero.
mediator of age-cognition relations is that age-related effects One method of statistical control involves the use of hierar-
would be expected to be much smaller if the variation in the chical regression procedures similar to those described earlier.
hypothesized mediator could be eliminated. An ideal investiga- That is, the total amount of age-related variance in the criterion
tive procedure would probably involve experimental manipu- variable is determined, and then the amount of unique or inde-
lation of the level of the mediator, but this does not appear fea- pendent age-related variance is assessed by controlling the vari-
sible with variables such as processing speed that are presumed ance in a measure of the hypothesized mediator. If the amount
to reflect relatively stable processing characteristics of an indi- of unique or independent age-related variance is large relative
vidual (Salthouse, 1992a, I992b). Attempts can be made to to the total age-related variance, then considerable indepen-
 PROCESSING-SPEED THEORY 419

dence of the age-related influences in the controlled reported later, stronger evidence for a mediational role of pro-
(hypothesized mediator) and criterion variables can be in- cessing speed has been obtained with larger samples and
ferred. However, if the unique age-related variance is small rel- multiple indicators of both the speed and cognition constructs.
ative to the total age-related variance, then one can infer that A graphical illustration of the effects of statistical control of
there is substantial commonality of influences in the two sets of measures of speed, and of matching individuals on speed, is pre-
measures. This second type of outcome is consistent with the sented in Figure 7. The data in this figure were obtained from a
mediation of at least some of the age effects in the criterion vari- sample of 221 adults who were administered the Raven's Pro-
able through the controlled variable (e.g., Salthouse, 1992a, gressive Matrices Test with a 20-min time limit, along with the
1992b; Salthouse & Babcock, 1991). Letter Comparison and Pattern Comparison tests of perceptual
Application of statistical control methods does require sev- speed (the sample, the same as that used in Figure 1, was re-
eral assumptions ranging from statistical (Cohen & Cohen, ported as Study 1 in Salthouse, 1993b). The vertical axis repre-
1983) to substantive in nature. One of the latter is that the cur- sents performance on the Raven's test expressed in z-score
rent level of the controlled variable, but not its etiology or de- units. Panels in the figure illustrate the initial age relations in
velopmental history, is relevant (Salthouse, 1991c, 1992a).ltis the total sample (A); the age relations in the total sample after
also important to establish realistic expectations regarding the use of statistical adjustment to eliminate the variance in the
outcome of statistical control analyses because statistical sig- composite speed measure, derived by averaging z scores for the
nificance tests in these types of analyses typically refer to the Letter Comparison and Pattern Comparison measures (B); and
amount of variance that is not explained by the controlled vari- the age relations in a suhsample (n = 90) whose composite
able rather than the amount that is explained. The age-related speed scores were within 0.5 standard deviations of zero (C). It
variance would be completely eliminated, with no residual vari- is apparent that there was large attenuation of the age relations
ance, only if the controlled and criterion variables shared all with the statistical control procedure (82%) and somewhat less
of their age-related variance. This is an extremely demanding (61%) attenuation when the sample was restricted to partici-
criterion because it would require not only that all variance at- pants with a narrow range of speed. However, it is noteworthy
tributable to the theoretical construct, but also all specific vari- that despite the substantial decrease in sample size from 221 to
ance associated with methods, materials, and measures, was 90, there was still a significant correlation between age and
shared across the controlled and criterion variables (Salthouse, speed (—.35) in the subsample. This is one of the reasons why
1992a). matching is not the optimum method for these types of analyses.
A potentially more meaningful basis for evaluating statistical Statistical control results resembling those portrayed in Pan-
control outcomes is in terms of the percentage reduction of age- els A and B of Figure 7 have been obtained in an independent
related variance. For example, 1 (Salthouse, 1992a, p. 26) have sample of adults by Babcock (1994), who found a 61% reduc-
suggested that percentage reductions of less than 20% are small, tion (i.e., from an R2 value of .212 to a value of .083) in the age-
values between 20% and 40% are interesting, values between related variance in the Raven's score after control of perceptual
40% and 60% are important, and values greater than 60% speed measures. Similar results have also been reported with a
should be considered major because in that case all other deter- computer-administered, self-paced, matrix reasoning task
minants together would be responsible for less than half of the (Salthouse, 1993b, 1994d). In the latter study, the R2 in deci-
total age-related variance. These values are admittedly arbi- sion accuracy associated with age in a sample of 246 adults
trary, but even tentative guidelines may be useful in interpreting from a wide range of ages was .149, and this was reduced to
the results of statistical control analyses. .014 (91% attenuation) after control of a composite measure of
Percentage reduction in age-related variance is not the only reaction time speed and to .015 (90% attenuation) after control
possible metric of the importance of a hypothesized mediator, of a composite measure of paper-and-pencil perceptual speed.
but if the variable is found to be associated with a very small Many published studies now exist in which some index of
reduction in age-related variance, then additional justification processing speed has been controlled in examinations of age-
is probably needed to establish its importance. If the variable is related influences on a variety of cognitive measures ranging
associated with a large reduction in age-related variance, how- from reasoning and spatial abilities to working memory, asso-
ever, it can be considered important as a potential mediator for ciative memory, and free recall. To illustrate, a total of 44 com-
this reason alone. parisons with measures from individual cognitive tests were
Some of the earliest results from statistical control analyses summarized by Salthouse (1993d). These comparisons were
with speed measures were described in Salthouse (1985b). Al- extracted from studies by Hertzog (1989) and Schaie (1989,
though the results of those initial analyses were generally con- 1990), as well as several studies by Salthouse and colleagues.
sistent with the predictions from the processing-speed theory, Age was initially associated with a mean of 16.2% of the vari-
the analyses were not optimal because the samples were often ance in these analyses, but the age-related variance was reduced
small, the speed and cognition constructs were assessed with to only 3.6% after measures of perceptual speed had been con-
single measures, and partial correlation rather than semipartial trolled. The average attenuation in the comparisons summa-
(i.e., hierarchical regression) correlation analyses were used. rized in Salthouse (1993d) was therefore nearly 78%.
Numerous other studies sharing many of these characteristics Comparisons with composite (i.e., average of z scores) cog-
have also been reported with mixed results (e.g., Bieman-Cop- nition measures were summarized in Salthouse (1992b; e.g.,
land & Charness, 1994; Bors & Forrin, 1995; Bryan & Luszcz, see Fig. 3.6). Age was initially associated with a mean of 25.2%
1996; Charness, 1987; Graf &Uttl, 1995; Hartley, 1986, 1993; of the variance in these analyses, but the age-related variance
Kwong See & Ryan, 1995; Nettelbeck & Rabbitt, 1992). As was reduced to a mean of only 4.3% after control of speed. The
420 SALTHOUSE

z - 1 fa - .03 (Age). r2= .322

« 60 60 70

B Chronological Age

z = 052 - .01 (Age). f *= .058 z = 1.08- .02 C*ge), r 2 * .126

30 40 50 60 70
Chronological Age Chronological Age

Figure 7. Relations between age and Raven's Progressive Matrices Test performance in the total sample
( A ) , in the total sample after control of a measure of speed ( B ) , and in a sample of participants within a
narrow speed range (C; data from Salthouse, 199 3b, Study I ) .

average reduction of the age-related variance in these compari- measures were also composites formed by averaging z scores
sons was therefore approximately 83%. from the spans obtained in the Reading Span and Computation
Similar magnitudes of attenuation have been found when the Span working memory tasks.
criterion cognitive tasks were self-paced and when separate Across all memory measures in Table 6, the age-related vari-
measures of decision accuracy and decision time were obtained ance was reduced an average of 77.6% after control of the reac-
(i.e., Salthouse, 1993b, 1994d). In fact, the percentage attenua- tion time speed measure and was reduced an average of 85.1 %
tion of the age-related variance in the Salthouse (1994d) studies after control of the paper-and-pencil perceptual speed measure.
was actually greater for a criterion measure of decision accuracy All of the values in Table 6 are thus in the important-to-major
(81.3% in Study 1 and 70.0% in Study 2) than for a criterion range according to the guidelines mentioned earlier, because
measure of decision time (74.5% in Study 1 and 54.8% in more than 50% of the age-related variance in the measures of
Study 2). memory is shared with the variance in relatively simple mea-
Substantial reduction of the age-related variance after control sures of processing speed.
of relatively simple measures of processing speed has also been It should also be noted that, in several studies, the propor-
found when a variety of memory measures served as the crite- tional attenuation of the age-related variance was greater with
rion cognitive variable. To illustrate, results from several recent speed measures from tasks involving perceptual or cognitive op-
projects conducted in my laboratory are summarized in Table erations, such as substitution, transformation, or comparison,
6. The speed measures in these studies were composites formed than with tasks merely requiring copying or line drawing re-
by averaging z scores for the Digit Symbol and Digit Digit reac- sponses (Salthouse, 1992b, 1993d, 1994a, 1994d; Salthouse &
tion time measures to create a reaction time speed composite Kersten, 1993; Salthouse etal., 1995). As suggested by the path
measure or by averaging z scores for the Letter Comparison and analysis results illustrated in Figure 5, the speed most relevant
Pattern Comparison measures to create a perceptual speed to the mediation of adult age differences in cognition therefore
composite measure. Several of the memory measures were appears to reflect the duration of cognitive operations rather
based on a single score (e.g., percentage correct recall in 12- than simply the speed of sensory and motor processes.
word lists for free recall and percentage correct in study-test The results just described reveal that an average of 75% or
paired-associates trials). However, in the continuous associative more of the age-related variance in a wide range of memory and
memory tasks, the associative memory measures were aggre- cognitive variables is shared with measures of processing speed.
gated across all presentation times to form a composite measure Moreover, this is true for different combinations of speed and
of associative memory performance. The working memory cognitive measures, with both paper-and-pencil and computer-
 PROCESSING-SPEED THEORY 421

Table 6
Age-Related Variance in Measures of Memory Performance

Age-related variance

Age After reaction After perceptual Proportion of

Measure and study N alone time speed speed shared variance

Free recall
Salthouse (1993d) 305 .162 .021 —1X1
Salthouse(1993e) 146" .546 —
— .049 —1.91
Salthousef I995b) 172 .289 .013 — /.96
Paired associates —
Salthouse(1993d) 305 .162 .024 -/.85
Salthouse(1993e) 146" .596 — .069
— — /.8S
Long-term memory for activities
Earlesfc Coon (1994) 177 .195 — .057 —1.1 \
Associative learning
Salthouse & Kersten (1993) 104' .152 .016 .025 .S9/.84
Salthouse(1994a)
Study 1 240 .165 .032 .059 .8I/.64
Study 2 125 .117 .044 .049 .62/.5S
Continuous associative memory
Letters and Digits
Salthouse (1994a)
Study 1 240 .105 .002 .000 .98/1.00
Study 2 125 .038 .004 .000 .89/1.00
Kersten & Salthouse ( 1 99 3) 78" .265 .046 — ,83/-
Words and Digits
Salthouse(1994d)
Study 1 246 .071 .010 .006 .S6/.92
Study 2 258 .071 .002 .010 .97/.86
Salthouse (1995d) 100' .369 .087 .061 .76/.S3
Working memory
Paper-and-pencil procedures
Salthouse & Babcock (1991), Study 2 233 .211 .007 —/.97
—
Salthouse (1 99 Ib)
Study 1 221 .292 — .050 —/.83
Study 2 228 .254 .014 — /.94
—
Study 3 223 .208 — .012 — /.94
Computer-administered procedures
Salthouse (1992a)
Study 1 ISC' .279 .081 .71/—
—
Study 2 100 .146 .014 .90/—
—
Salthouse(1995b) 117 .141 .029 .031 .79/.7S
Sal thouse&Meinzf 1995) 242 .033 .001 .001 .97/.97
Combined samples from several studies 184" .155 .031 .80/—
—
Miscellaneous (Salthouse, 1995a)
Matrix memory
Verbal 173 .277 .113 .086 .S9/.69
Spatial 173 .402 .202 .148 .50/.63
Element memory
Verbal 173 .087 .040 .005 .S4/.94
Spatial 173 .070 .020 .000 .71/1.00
Keeping track
Verbal 173 .170 .050 .029 .71/.83
Spatial 173 .142 .046 .036 .68/.75

Note. Dashes indicate that measures were not available.

a
Only young and old adults.

administered tests, and across different types of tests (e.g., those for that prediction. Not only have the path analyses revealed
requiring reasoning, spatial, and memory abilities). The statis- patterns consistent with the expectations, but the statistical con-
tical control results therefore provide strong evidence for a ma- trol analyses indicated that nearly 75% of the age-related vari-
jor role of processing speed in the relations between age and ance in many cognitive measures is shared with measures of
measures of cognitive performance. processing speed. Because similar estimates of shared age-re-
In summary, research examining the relational prediction of lated variance have been obtained across timed and self-paced
the processing-speed theory has provided impressive support measures of reasoning and spatial abilities, across a wide range
422 SALTHOUSE

100 Young Old

LagO

Lag1
90

70

500 1000 1500 2000 2500

Presentation Duration (msec)

Figure S. Accuracy as a function of stimulus presentation time with zero items and one item intervening
between presentation and test of digit-letter pairs (unpublished data from Kersten & Salthouse, 1993).

of memory measures, and with different types of speed mea- years), was a continuous associative memory task involving let-
sures, it seems indisputable that processing speed is involved in ter-digit pairs. Probes, requiring a decision of whether the items
the relations between age and cognitive performance. in the test pair had been paired with one another when either
item last occurred, were presented either immediately after the
Hypothesis 3: The limited time mechanism and the simultane- letter-digit pair (Lag 0) or after one intervening pair (Lag 1).
ity mechanism are primarily responsible for the relations be- This task is interesting because it allows the influence of stimu-
tween processing speed and measures of cognitive functioning. lus presentation time to be examined not only on a relatively
simple measure (i.e., accuracy at Lag 0) but also with another
measure that presumably requires processing beyond that
Limited Time Mechanism
needed for the first measure (i.e., accuracy at Lag 1). That is,
The basic idea underlying the limited time mechanism is that in addition to the registration and encoding required for the Lag
slower speed of executing many processing operations means 0 measure, relevant information needs to be preserved during
that less processing can be completed in a given amount of time. the presentation and processing of additional items in the Lag I
One method of illustrating the hypothesized relations involves measure.
manipulating the amount of time available to process the stim- Three points should be noted about the results portrayed in
ulus. Although it is unlikely that variations in stimulus presen- Figure 8. First, all of the functions appear to have a similar neg-
tation time will affect the speed of internal processing opera- atively accelerated relation between accuracy and presentation
tions (Salthouse, 1991c, 1992b), this manipulation can still be time, but with asymptotes of less than 100%. The finding of
informative about how level of cognitive performance is related asymptotes below 100% suggests that factors other than the du-
to time available for processing. ration of the stimulus contribute to performance on this task.
Results from manipulations of stimulus presentation time Second, the functions for the Lag 0 measure (solid lines) are
can be portrayed in time-accuracy functions in which accuracy consistently above and to the left of those for the Lag 1 measure
is represented along the vertical axis and time is represented (dashed lines), indicating that, as expected, more time was
along the horizontal axis. Vertical contrasts in this type of rep- needed for the processing associated with Lag 1 decisions than
resentation reflect the level of accuracy at a given time, and hor- for that associated with Lag 0 decisions. This indicates that the
izontal contrasts reflect the amount of time needed to achieve a effects of restricted processing time propagate to more complex
given level of accuracy. If the complete function is available, forms of processing and are not simply confined to the simplest
then parameters of the mathematical function can be examined type of processing. The third point to be noted about Figure 8
and compared across experimental conditions or age groups is that the functions for older adults (open circles) are consis-
(e.g., Kliegl et al., 1994; Mayr & KJiegl, 1993; Salthouse &
Coon, 1993).
' The functions in thisfigurehave asymptotes below 100%, indicating
Figure 8, based on results of an unpublished study by Kersten that perfect performance was not achieved even at the longest available
and Salthouse (1993), illustrates a typical pattern from manip- duration. A similar finding has been reported by Salthouse and Coon
ulations of stimulus presentation time in adults of different (1993), and this may be characteristic of relatively difficult tasks in
ages.5 The task in this study, which was performed by 39 young which only limited amounts of practice are provided (cf. Kliegl et al.,
adults (Mage = 20.5 years) and 39 older adults (Mage = 67.9 1994).
 PROCESSING-SPEED THEORY 423

tently to the right and below those for younger adults (filled ing, pattern comparison, vigilance, and hand-eye coordination
circles). It can thus be inferred that older adults complete less than those variables shared with the age-related variance in the
processing in a given amount of time than young adults in this speed variables. These results are therefore more consistent
task. This, of course, is exactly what is predicted from the lim- with the interpretation that age-related variations in speed con-
ited time mechanism. tribute to the age-related variations in these other constructs
than with the view that the age effects in these other constructs
contribute to the age effects in the measures of speed. It should
Simultaneity Mechanism
also be noted that in studies in which the appropriate compari-
The key assumption in the simultaneity mechanism is that a sons could be performed, the amount of reduction of age-re-
slower speed of processing results in a smaller amount of infor- lated variance in measures of cognitive functioning was greater
mation that is in a high enough level of activation to be available after control of measures of speed than after control of measures
for other forms of processing. The concept of working memory of working memory (e.g., Salthouse, 1991 b).
is another way of referring to the amount of simultaneously ac- One factor that complicates the distinction between cause
tive information, and there are many reports of age-related de- and consequence with the construct of speed is that because the
clines in measures of working memory (e.g., see Salthouse, tasks used to assess processing speed are so simple, the principal
1994b, for a review). Furthermore, in all of the studies in which way that variations in performance can be manifested is in
measures of processing speed were available, statistical control terms of alterations in how rapidly the tasks can be performed.
of the speed index greatly reduced the age-related variance in That is, because very few normal adults would make mistakes
the working memory measures (cf. Table 6). in these simple comparison and substitution tasks if allowed un-
Although the evidence regarding age and speed influences on limited time, any factor that influenced basic processing effi-
working memory is consistent with the predictions from the ciency would probably have its effects on measures of speed of
processing-speed theory, a number of questions remain regard- performance. It is therefore possible that a slower processing
ing the simultaneity mechanism. For example, are working speed is not the critical mediating factor in many of the age-
memory tasks the best means of assessing the amount of simul- related declines in cognition; rather, rate of performance is
taneously available information, or is performance on those merely the manner in which differences in processing efficiency
tasks heavily influenced by strategies, knowledge, and other fac- are exhibited in very simple tasks. The challenge in this alter-
tors? And do the reductions in the amount of simultaneously native interpretation is to identify independent measures of
available information originate simply from slower rates of ac- other possible determinants of basic processing efficiency that
tivating information, or are there also alterations in the rate of would allow direct comparisons with the processing speed
loss of information over time? Despite these uncertainties, the interpretation.
simultaneity mechanism remains a plausible candidate for re- Another issue concerned with the age-related slowing phe-
lating processing speed to quality of cognitive functioning. nomenon is whether the slower speeds are an artifact of a rela-
tively small number of very slow responses. If this were the case,
then the slowing phenomenon might be attributable to failures
Additional Issues
to sustain concentration or to inhibit distraction. However,
A central issue concerning age-related slowing is whether it analyses reveal little support for this attentional block inter-
is best conceptualized as a cause or as a consequence of age pretation. That is, statistical control analyses similar to those
differences in more basic behavioral constructs (e.g., Mayr & described earlier, but with reaction times from different percen-
KJiegl, 1993; Salthouse, 1985b). This is not an easy question to tiles of each individual's reaction time distribution as the con-
resolve, but relevant information can be derived from compari- trolled and criterion variables, have been reported by Salthouse
sons of the relative proportions of age-related variance shared (1993a). Because almost all of the age-related variance in the
between measures of speed and measures reflecting what might slow (90th percentile) responses was shared with the age-related
be considered more basic constructs. The reasoning is that the variance in the fast (10th percentile) responses, there was no
most fundamental and basic construct, at least in terms of age- evidence for a failure to sustain high levels of attention or con-
related influences, should be one that has a large amount of centration as a factor contributing to age-related slowing. More-
overlap with the age-related variance in other variables but has over, because this same pattern was found in two reaction time
a smaller proportion of its own age-related variance overlapping tasks across four separate data sets, it can be regarded as fairly
with that of other variables. Evidence relevant to this issue can robust. These results therefore indicate that age-related slowing
be obtained by contrasting the proportions of age-related vari- is evident throughout the individual's entire distribution of re-
ance shared between a variable reflecting another construct and sponses and is not simply manifested in his or her slowest
a speed variable. Values of these comparisons from studies in- responses.
volving a variety of "other" variables are summarized in Table One final issue concerned with the age-related slowing phe-
7. Not all of the variables in Table 7 might be considered equally nomenon is whether it merely reflects sensory and motor as-
plausible as candidates for a basic construct important for cog- pects. This does not appear to be the case because independent
nition, but each reflects a factor that could potentially contrib- and distinct age-related influences have been found on mea-
ute to age-related differences in perceptual speed. sures from tasks involving comparison or substitution pro-
It is clear from the entries in Table 7 that the speed variables cesses. Relevant results are available in a contrast of sensorimo-
shared more of the age-related variance in the measures reflect- tor speed measures and perceptual speed measures. As noted
ing working memory, inhibition, pattern memory, serial learn- earlier, the sensorimotor speed measures in these projects in-
424 SALTHOUSE

Table 7
Proportions of Shared Age-Related Variance in Measures of Speed and Other Potentially Fundamental Constructs

Proportion of shared age-related

variance

Speed with Other variable

Other variable and study Speed variable other variable with speed

Backwards digit span Digit symbol

Salthouse ( 1 988b), N = 200 .95 .23
Working memory (paper and pencil procedures) Perceptual speed
Salthouse & Babcock ( 1 99 1), n = 233 .97 .60
Salthouse ( 1 99 1 a), « = 221 .83 .69
Salthouse ( 1 991 a), n = 228 .94 .55
Salthouse ( 1 991 a), « = 223 .94 .64
Working memory (computer-administered
procedures) Reaction time speed
Salthouse (1992a),n = 180 .71 .35
Salthouse (1992a),n= 100 .90 .33
Salthouse (1995b), n = 117 .79 .32
Salthouse & Meinz (1995), n = 242 .97 .10
Combined samples from several studies (A' - 184) .80 .24
Inhibition (Stroop color-word interference measure) Perceptual speed
Salthouse & Meinz ( 1995), n = 242 .85 .48
Salthouse (1995b),n= 172 .81 .65
Salthouse etal. (1994)
Backwards digit span Symbol-Digit Substitution Test
n= 165 .97 .20
n-239 1.00 .18
Pattern memory Symbol-Digit Substitution Test
n=165 .76 .1 1
n = 223 1.00 .24
n-239 .63 .16
Serial learning Symbol-Digit Substitution Test
n-223 .80 .26
« = 239 .45 .18
Pattern comparison Symbol-Digit Substitution Test
« = 165 .97 .51
« = 223 .90 .55
n = 239 .77 .43
Vigilance (continuous performance) Symbol-Digit Substitution Test
n= 165 .75 .01
n = 223 .77 .10
« = 239 .88 .19
I land-eye coordination (tracking) Symbol-Digit Substitution Test
n = 223 .68 .24
« = 239 .49 .35
Tapping speed Symbol-Digit Substitution Test
« = 223 .11 .25
« = 239 .64 .30

volve copying digits or drawing lines in specified locations, posite sensorimotor measure were 29% for a sample of 744
whereas the perceptual speed measures involve determining adults (Earles & Salthouse, 1995), 32% for a sample of 200
whether two sets of lines or two sets of letters are identical. adults (Salthouse, 1993c, Study 1 ) , and 25% for a sample of
Because of the increased reliability, and the potential of min- 154 adults (Salthouse, 1993c, Study 2).
imizing specific variance while emphasizing construct variance, Similar analyses have been conducted with the Digit Digit
comparisons of different types of speed arc most meaningful and Digit Symbol reaction time measures. Although both tasks
when expressed in terms of composite scores. Although the two require choice decisions, the choice in the Digit Digit task is
types of speed measures have a large proportion of age-related based on physical identity, whereas the choice in the Digit Sym-
variance in common, significant residual age-related variance bol task is based on the substitution of digits and symbols. The
in the composite perceptual speed measures has been found af- percentages of age-related variance in the Digit Symbol mea-
ter the variance in the composite sensorimotor speed measure sure that were distinct or independent of those in the Digit Digit
has been controlled. To illustrate, the percentages of age-related measure were 24% for a sample of 744 adults (Earles & Salt-
variance in the composite perceptual speed measure that were house, 1995), 34% for a sample of 694 adults collapsed across
unique and distinct from the age-related variance in the com- several studies, 23% for a sample of 104 (Salthouse & Kersten,
 PROCESSING-SPEED THEORY 425

1993), and 39% for a sample of 131 (Salthouse et al., 1995, relatively little knowledge of why increased age is associated
Study 1). Because these results indicate that there are unique with a slower speed of performing many activities, but little is
and distinct age-related influences on speed measures involving known about precisely how the limited time and simultaneity
more cognitive operations, it can be inferred that the age-related mechanisms, and possibly other mechanisms, relate slower pro-
slowing phenomenon is not simply restricted to sensory and cessing to lower levels of cognitive performance. Among the im-
motor aspects. portant issues to be investigated are the neurophysiological ba-
sis for age-related slowing and what the processing-speed con-
struct actually reflects. With respect to the first issue, a number
Relations to Other Theories
of neurophysiological mechanisms could be proposed to ac-
The processing-speed theory has both similarities and differ- count for age-related slowing, including
ences with respect to other theoretical perspectives within the
field of aging and cognition. The focus on processing speed as a a slower speed of transmission along single (e.g., loss of
myelination) or multiple (e.g., loss of functional cells dictating cir-
central construct is very similar to Birren's (e.g., 1965, 1974)
cuitous linkages) pathways, or. . . delayed propagation at the con-
speculation that the speed with which many cognitive opera-
nections between neural units (e.g., impairment in functioning
tions can be carried out may function as an independent vari- of neurotransmitters, reduced synchronization of activation
able for many behavioral outcomes. The current theory diifers patterns). (Salthouse, I992b, p. 116)
from Birren's perspective, however, in that specific mechanisms
are proposed to account for the speed-cognition relations, and Multidisciplinary research will almost certainly be needed to
empirical evidence derived from several different analytical distinguish among these alternatives because it is unlikely that
methods has been generated to support the predictions. The they can be differentiated solely on the basis of behavioral
processing-speed theory also shares the assumption that a gen- observations.
eral speed factor plays a major role in age-related slowing with The processing-speed construct is postulated to represent
theories proposed by Cerella (1985, 1990) and Myerson et al. how quickly many different types of processing operations can
(1990). It differs from those theories in that the primary focus be carried out. The moderate to high proportions of shared age-
here is on explaining relations between age and cognition rather related variance across a wide range of speed measures, includ-
than between age and speed and in treating analyses of system- ing those presumed to reflect the duration of task-specific cog-
atic relations (i.e., Brinley plots) as only one source of evidence nitive processes, are clearly consistent with this hypothesis.
relevant to the hypothesis of a general slowing factor. However, there is still uncertainty as to the breadth of the pro-
The processing-speed theory also has a resemblance to theo- cessing-speed construct and whether more primitive behavioral
ries attempting to account for age-cognition relations in terms constructs might contribute to the age differences in processing
of broad explanatory mechanisms such as processing resources speed. Principled bases for specifying which variables are likely
(e.g., Craik & Byrd, 1982) and aspects of attention such as in- to be exempt from age-related slowing are also currently lack-
hibition (e.g., Hasher & Zacks, 1988). Unlike those theories, ing. Further research is therefore needed before the processing-
however, the central construct in the processing-speed theory speed construct can be considered well understood.
has been reliably operationalized, and a large body of evidence Despite these limitations, there is currently strong evidence
based on statistical control and path analysis procedures has that measures hypothesized to reflect speed of processing are
accumulated indicating that the construct has a major role in involved in the adult age differences found in many measures
mediating relations between age and cognition. of cognitive functioning. The processing-speed theory thus ap-
Finally, regardless of whether one accepts the interpretation pears to have sufficient plausibility to merit serious consider-
that at least some of the age-related declines in various mea- ation as an explanation for at least some of the age-related
sures of cognitive functioning are attributable to a slower speed effects on cognition.
of carrying out relevant processing operations, the discovery Finally, two additional advantages of focusing on the process-
that measures of how quickly very simple tasks can be per- ing-speed construct in research on cognitive aging warrant
formed share large proportions of age-related variance with mention. One is that processing speed is a parsimonious target
complex measures of cognitive performance has implications construct for research concerned with distal determinants of
for the nature of virtually all theories concerned with aging and cognitive aging phenomena. That is, because speed appears to
cognition. That is, nearly every theory, including those attribut- play a central role in many age-related cognitive differences, an
ing age-related differences to impairments in specific cognitive explanation of the factors occurring earlier in one's life that are
processes or to deficits in certain types of strategies, will pre- responsible for age-related decreases in speed would probably
sumably need to take factors related to basic processing effi- account for a large proportion of the age differences in a variety
ciency into consideration or else they may run the risk of focus- of measures of cognitive functioning.
ing on merely another symptom of what could be a broader and The second advantage of focusing on the processing-speed
more fundamental phenomenon. construct is that speed may function as a bridging construct
between behavioral and neurophysiological research. Because
Summary ti me is an objective and absolute dimension rather than a norm-
reference scale, as is the case with most behavioral measures,
Additional research is needed before the mechanisms respon- it is inherently meaningful in all disciplines and thus has the
sible for the relations between age and speed, or between speed potential to function as a Rosetta stone in linking concepts from
and cognition, can be fully understood. Not only is there still different disciplines (Salthouse, 1985b).
426 SALTHOUSE

References Fisk, A. D., & Fisher, D. L. (1994). Brinley plots and theories of aging:
The explicit, muddled, and implicit debates. Journal of Gerontology:
Babcock, R. L. (1994). Analysis of adult age differences in the Raven's Psychological Sciences, 49, P8I-P89.
Advanced Progressive Matrices Test. Psychology and Aging, 9, 303- Fisk, A. D., Fisher, D. L., & Rogers, W. A. (1992). General slowing
314. alone cannot explain age-related search effects: A reply to Cerella.
Baddeley, A. D. {1986). Working memory. Oxford, England: Clarendon Journal of Experimental Psychology: General, 121, 73-78.
Press. Foster, J. C., & Taylor, G. A. (1920). The applicability of mental tests to
Bieman-Copland, S., & Charness, N. (1994). Memory knowledge and persons over 50. Journal of Applied Psychology, 4, 39-58.
memory monitoring in adulthood. Psychology and Aging, 9, 287-
Ghiselli, E. E. (1973). The validity of aptitude tests in personnel selec-
302.
tion. Personnel Psychology, 26, 461-477.
Birren, J. E. (1965). Age changes in speed of behavior: Its central nature
Graf, P., & Uttl, B. (1995). Component processes of memory: Changes
and physiological correlates. In A. T. Welford & J. E. Birren (Eds.),
across the adult lifespan. Swiss Journal of Psychology, 54, 113-138.
Behavior, aging, and the nervous system (pp. 191-216). Springfield.
Hale, S., Lima, S. D., & Myerson, J. ( 1991). General cognitive slowing
IL: Charles C. Thomas.
in the nonlexical domain: An experimental validation. Psychology
Birren, J. E. (1974). Translations in gerontology: From lab to life: Psy-
and Aging. 6, 512-521.
chophysiology and speed of response. American Psychologist, 29,
Hartley, J. T. (1986). Reader and text variables as determinants of dis-
808-815.
course memory in adulthood. Psychology and Aging, 1, 150-158.
Bors, D. A.. & Forrin, B. (1995). Age, speed of information processing,
Hartley, J. T. (1993). Aging and prose memory: Tests of the resource-
recall, and fluid intelligence. Intelligence, 20, 229-248.
deficit hypothesis. Psychology and Aging, g, 538-551.
Brinley, J. F. (1965). Cognitive sets, speed and accuracy of performance
in the elderly. In A. T. Welford & J. E. Birren (Eds.), Behavior, aging, Hasher, L., & Zacks, R. T. (1988). Working memory, comprehension,
and the nervous system (pp. 114-149). Springfield, IL: Charles C and aging: A review and a new view. In G. H. Bower (Ed.), The psy-
Thomas. chology of learning and motivation (Vol. 22, pp. 193-225). San
Bryan, J., & Luszcz, M. A. (1996). Speed of information processing as Diego, CA: Academic Press.
a mediator between age and free recall performance. Psychology and Hebh, D. O. (1942). The effect of early and late brain injury upon test
Aging, II, 3-9. scores and the nature of normal adult intelligence. Proceedings of the
Cattell, R. B. (1943). The measurement of adult intelligence. Psycho- American Philosophical Society, Si. 275-292.
logical Bulletin, 40, 153-193. Hertzog, C. (1989). Influences of cognitive slowing on age differences
Cerella, J. (1985). Information processing rates in the elderly. Psycho- in intelligence. Developmental Psychology 25, 636-651.
logical Bulletin, 98, 67-83. Horn. J. L. (1982). The theory of fluid and crystallized intelligence in
Cerella, J. (1990). Aging and information processing rates in the el- relation to concepts of cognitive psychology and aging in adulthood.
derly. In J. E. Birren & K. W. Schaie (Eds.), Handbook of the psychol- In F. I. M. Craik & S. Trehub (Eds.), Aging and cognitive processes
ogy of aging (3rd ed., pp. 201-221). New York: Academic Press. (pp. 237-278). New York: Plenum.
Cerella, J. (1991). Age effects may be global not local: Comment on Horn, J. L., & Cattell, R. B. (1963). Age differences in fluid and crys-
Fisk and Rogers (1991). Journal of Experimental Psychology: Gen- tallized intelligence. Acta Psychologica, 26, 107-129.
eral, 120. 215-223. Hultsch, D. F., Hertzog, C., & Dixon, R. A. (1990). Ability correlates
Cerella, J. (1994). Generalized slowing in Brinley plots. Journal of Ger- of memory performance in adulthood and aging. Psychology and
ontology: Psychological Sciences, 49, P65-P7I. Aging, 5. 356-368.
Cerella, J., Poon, L. W.. & Williams, D. M. (1980). Age and the com- Hultsch, D. F.. Hertzog, C., Small, B. J., McDonald-Miszczak, L., &
plexity hypothesis. In L. W. Poon (Ed.), Aging in the 1980's (pp. Dixon, R. A. (1992). Short-term longitudinal change in cognitive
332-340). Washington, DC: American Psychological Association. performance in later life. Psychology and Aging, 7, 571-584.
Charness, N. (1987). Component processes in bridge bidding and novel Hunter. J. E., & Hunter, R. F. (1984). Validity and utility of alternate
problem-solving tasks. Canadian Journal of Psychology. 41, 223- predictors of job performance. Psychological Bulletin, 96, 72-98.
243. Jensen, A. R. (1982). The chronometry of intelligence. In R. J. Slern-
Cohen, J. (1988). Statistical power analysis for the behavioral sciences. berg (Ed.), Advances in the psychology ofhuman intelligence (Vol. I ,
Hillsdale, NJ: Erlbaum. pp. 255-310). Hillsdale, NJ: Erlbaum.
Cohen. J., & Cohen, P. (1983). Applied multiple regression/correlation
Jensen, A. R. (1987). Individual differences in the Hick paradigm. In
analyses for the behavioral sciences (2nd ed.). Hillsdale, NJ: P. A. Vernon (Ed.), Speed of information processing and intelligence
Erlbaum.
(pp. 101-175). Norwood, NJ: Ablex.
Craik, F. I. M., & Byrd, M. (1982). Aging and cognitive deficits: The
Jones, H. E. (1956). Problems of aging in perceptual and intellective
role of attentional resources. In F. I. M. Craik & S. Trehub (Eds.),
functions. In J. E. Anderson (Ed.), Psychological aspects of aging (pp.
Aging and cognitive processes (pp. 191-211). New York: Plenum.
135-139). Washington, DC: American Psychological Association.
Earles, J. L., & Coon, V. E. (1994). Adult age differences in long-term
Jones, H. E., & Conrad, H. S. (1933). The growth and decline of intel-
memory for performed activities. Journal of Gerontology: Psycholog-
ligence: A study of a homogeneous group between the ages often and
ical Sciences, 49, P32-P34.
sixty. Genetic Psychology Monographs, 13, 223-298.
Earles. J. L., & Salthouse, T. A. (1995). Interrelations of age, health,
Kail, R. (1986). Sources of age differences in speed of processing. Child
and speed. Journal of Gerontology: Psychological Sciences. SOU.
Development, 57, 969-987.
P33-P41.
Kail. R. ( 1 9 9 1 ) . Developmental change in speed of processing during
Ekstrom, R. B., French, J. W., Harman. H., & Dermen, D. (1976). Kit
of factor-referenced cognitive tests (Rev. ed.). Princeton, NJ: Educa- childhood and adolescence. Psychological Bulletin, 109. 490-501.

tional Testing Service. Kail, R., & Park, Y. (1992). Global developmental change in processing
Eysenck, H. J. (1987). Speed of information processing, reaction time, times. Merrill-Palmer Quarterly, 38, 525-541.
and the theory of intelligence. In P. A. Vernon (Ed.), Speed of infor- Kail, R., & Salthouse. T. A. (1994). Processing speed as a mental ca-
mation processing and intelligence (pp. 22-67). Norwood, NJ: pacity. Acta Psychologica, 86, 199-225.
Ablex. Kcrsten, A., & Salthouse, T. A. (1993). [Relations between time and
 PROCESSING-SPEED THEORY 427

accuracy in a continuous associative memory task]. Unpublished Salthouse, T. A. (1987). The role of representations in age differences
raw data. in analogical reasoning. Psychology and Aging, 2, 357-362.
Kliegl, R., Mayr, U., & Krampe, R. T. (1994). Time-accuracy functions Salthouse, T. A. (1988a). Initializing the formalization of theories of
for determining process and person differences: An application to cognitive aging. Psychology and Aging, 3, 3-16.
cognitive aging. Cognitive Psychology, 26, 134-164. Salthouse, T. A. (1988b). The role of processing resources in cognitive
Kwong See, S. X, & Ryan, E. B. (1995). Cognitive mediation of adult aging. In M. L. Howe & C. J. Brainerd (Eds.), Cognitive development
age differences in language performance. Psychology and Aging, 10, in adulthood (pp. 185-239). New York: Springer-Verlag.
458-468. Salthouse, T. A. (1991 a). Expertise as the circumvention of human pro-
Laver, G. D., & Burke, D. M. (1993). Why do semantic priming effects cessing limitations. In A. Ericcson & J. Smith (Eds.), Toward a gen-
increase in old age? A meta-analysis Psychology and Aging, 8, 34-43. eral theory of expertise: Prospects and limits (pp. 286-300). New
Lemmon, V. W. (1927). The relation of reaction time to measures of York: Cambridge University Press.
intelligence, memory, and learning. Archives of Psychology (Whole Salthouse, T. A. (I991b). Mediation of adult age differences in cogni-
No. 94). tion by reductions in working memory and speed of processing. Psy-
Lima, S. D., Hale, S., & Myerson, J. (1991). How general is general chological Science, 2, 179-183.
slowing? Evidence from the lexical domain. Psychology and Aging, 6, Salthouse, T. A. (1991c). Theoretical perspectives on cognitive aging.
416-425. Hillsdale, NJ: Erlbaum.
Lindenberger, U., Mayr, U., & Kliegl, R. (1993). Speed and intelligence Salthouse, T. A. (1992a). Influence of processing speed on adult age
in old age. Psychology and Aging, 8. 207-220. differences in working memory. Acta Psychologica, 79, 155-170.
MacCallum, R. C, Wegener, D. T., Uchino, B. N., & Fabrigar, L. R. Salthouse, T. A. (1992b). Mechanisms of age-cognition relations in
(1993). The problem of equivalent models in applications of covari- adulthood. Hillsdale, NJ: Erlbaum.
ance structure analysis. Psychological Bulletin, 114, 185-199. Salthouse, T. A. (1992c). Reasoning and spatial abilities. In F. I. M.
MacKay, D. G., & Burke, D. M. (1990). Cognition and aging: A theory Craik & T. A. Salthouse (Eds.), Handbook of aging and cognition (pp.
of new learning and the use of old connections. In T. M. Hess (Ed.), 167-211). Hillsdale, NJ: Erlbaum.
Aging and cognition: Knowledge organization and utilization (pp. Salthouse, T. A. (1992d). Shifting levels of analysis in the investigation
213-263). Amsterdam: North-Holland. of cognitive aging. Human Development, 35, 321-342.
Madden, D. J., Pierce, T. W, & Allen, P. A. (1992). Adult age differ- Salthouse, T. A. (1992e). What do adult age differences in the Digil
ences in attentional allocation during memory search. Psychology Symbol Substitution Test reflect? Journal of Gerontology: Psycholog-
and Aging, 7,594-601. ical Sciences. 47. P121-P128.
Madden, D. J., Pierce, T. W, & Allen, P. A. (1993). Age-related slowing Salthouse, T. A. (1992f). Why do adult age differences increase with
and the time course of semantic priming in visual word identification. task complexity? Developmental Psychology, 28. 905-918.
Psychology and Aging, 8, 490-507. Salthouse, T. A.( 1993a). Attentional blocks are not responsible for age-
Mayr, U., & Kliegl, R. (1993). Sequential and coordinative complexity: related slowing. Journal of Gerontology: Psychological Sciences, 48,
Age-based processing limitations in ngural transformations. Journal P263-P270.
of Experimental Psychology: Learning, Memory, and Cognition, 19, Salthouse, T. A. (1993b). Influence of working memory on adult age
1297-1320. differences in matrix reasoning. British Journal of Psychology, 84,
Myerson, J., Hale, S., Wagstaff, D., Poon, L. W, & Smith, G. A. (1990). 171-199.
The information loss model: A mathematical theory of age-related Salthouse, T A. (1993c). Speed and knowledge as determinants ol
cognitive slowing. Psychological Review, 97, 475-487. adult age differences in verbal tasks. Journal of Gerontology: Psycho-
Myerson, J., Wagstaff, D., & Hale, S. (1994). Brinley plots, explained logical Sciences, 48. P29-P36.
variance, and the analysis of age differences in response latencies. Salthouse, T. A. (1993d). Speed mediation of adult age differences in
Journal of Gerontology: Psychological Sciences, 49, P72-P80. cognition. Developmental Psychology, 29, 722-738.
Nettelbeck, T, & Rabbitt, P. M. A. (1992). Aging, cognitive perfor- Salthouse, T. A. (1994a). Aging associations: Influence of speed oti
mance, and mental speed. Intelligence, 16, 189-205. adult age differences in associative learning. Journal ofExperimentai
Peak, H., & Boring, E. G. (1926). The factor of speed in intelligence. Psychology: Learning, Memory, and Cognition, 20, 1486-1503.
Journal of Experimental Psychology, 9, 71-94. Salthouse, T. A. (1994b). The aging of working memory. Neuropsychol-
Perfect, T. J. (1994). What can Brinley plots tell us about cognitive ogy. 8. 535-543.
aging? Journal oj Gerontology: Psychological Sciences, 49, P60-P64. Salthouse, T. A. (1994c). How many causes are there of aging-related
Rushton, J. P., Brainerd, C. J., & Pressley, M. (1983). Behavioral devel- decrements in cognitive functioning? Developmental Review, 14,413-
opment and construct validity: The principle of aggregation. Psycho- 437.
logical Bulletin, 94, 18-38. Salthouse, T. A. (1994d). The nature of the influence of speed on aduli
Salthouse, T. A. (1980). Age and memory: Strategies for localizing the age differences in cognition. Developmental Psychology, 30, 240-259
loss. In L. W. Poon, J. L. Fozard, L. Cermak, D. Arenberg, & L. W. Salthouse, T. A. (I995a). Differential age-related influences on mem-
Thompson (Eds.), New directions in memory and aging, (pp. 47- ory for verbal-symbolic information and visual-spatial information
65). Hillsdale, NJ: Erlbaum. Journal of Gerontology: Psychological Sciences, SOB, PI93-P201.
Salthouse, T. A. (1982). Adult cognition: An experimental psychology of Salthouse, T. A. (1995b). Influence of processing speed on adult agt
human aging. New York: Springer-Verlag. differences in learning. Swiss Journal of Psychology, 54, 102-112.
Salthouse, T. A. (1984). Effects of age and skill in typing. Journal of Salthouse, T. A. (1995c). Processing capacity and its role on the re-
Experimental Psychology: General, 113, 345-371. lations between age and memory. In F. E. Weinert & W. Schneidei
Salthouse, T. A. (1985a). Speed of behavior and its implications for (Eds.), Memory performance and competencies: Issues in growth one
cognition. In J. E. Birren & K. W. Schaie (Eds.), Handbook of the development (pp. 111-125). Hillsdale, NJ: Erlbaum.
psychology ofaging(2nd ed., pp. 400-426). New York: Van Nostrand Salthouse, T. A. (1995d). Selective influences of age and speed on asso-
Reinhold. ciative memory. American Journal of Psychology, 108, 381-396.
Salthouse, T. A. (1985b). A theory of cognitive aging Amsterdam: Salthouse, T. A. (1995e). [Age, speed, and memory interrelations]. Un-
North-Holland. published raw data.
 428 SALTHOUSE

Salthouse, T. A. (1996). General and specific speed mediation of adult Salthouse, T. A., & Somberg, B. L. (1982). Isolating the age deficit in
age differences in memory. Journal of Gerontology: Psychological Sci- speeded performance. Journal of Gerontology, 37, 59-63.
ences, 51B, P30-P42, Schaie, K. W. (1989). Perceptual speed in adulthood: Cross-sectional
Salthouse, T. A., & Babcock, R. L. (1991). Decomposing adult age and longitudinal analyses. Psychology and Aging. 4, 443-453.
differences in working memory. Developmental Psychology, 27, 763- Schaie, K. W. (1990). Correction to Schaie (1989). Psychology and
776. Aging, 5, nI.
Salthouse, T. A., & Coon, V. E. (1993). Influence of task-specific pro- Schaie, K. W., & Willis, S. L. (1993). Age difference patterns of psycho-
cessing speed on age differences in memory. Journal of Gerontology: metric intelligence in adulthood: Generalizability within and across
Psychological Sciences, 48, P245-P255. ability domains. Psychology and Aging, 8, 371-383.
Salthouse, T. A., & Coon, V. E. (1994). Interpretation of differential Tomer, A., & Cunningham, W. R. (1993). The structure of cognitive
deficits: The case of aging and mental arithmetic. Journal of Experi- speed measures in old and young adults. Mullivariate Behavioral Re-
mental Psychology: Learning, Memory, ana Cognition, 20, 1172- search. 28, 1-24.
1182. Travis, L. E., & Hunter, T. A. (1928). The relation between 'intelligence'
Salthouse, T. A., Fristoe, N. M., Lineweaver, T. T., & Coon, V. E. (1995). and reflex conduction rate. Journal of Experimental Psychology, 11.
Aging of attention: Does the ability to divide decline? Memory & Cog- 342-354.
nition, 23. 59-71. Vernon, P. A. (1983). Speed of information processing and general in-
Salthouse, T. A., & Kail, R. (1983). Memory development throughout telligence. Intelligence, 7. 53-70.
the life span: The role of processing rate. In P. B. Baltes & O. G. Brim Vernon, P. A. (1987). New developments in reaction time research. In
(Eds.), Life span development and behavior (Vol. 5, pp. 89-116). New P. A. Vernon (Ed.), Speed of information processing and intelligence
"Vbrk: Academic Press. (pp. 1-19). Norwood, NJ: Ablex.
Salthouse, T. A., & Kersten, A. W. (1993). Decomposing adult age Wechsler, D. (1981). Wechsler Adult Intelligence Scale-Revised. New
differences in symbol arithmetic. Memory & Cognition, 21. 699-710. York: Psychological Corporation.
Salthouse, T. A., Letz, R., & Hooisma, J. (1994). Causes and conse- White, N., & Cunningham, W. R. (1987). The age-comparative con-
quences of age-related slowing in speeded substitution performance. struct validity of speeded cognitive factors. Multivariale Behavioral
Developmental Neuropsychology, 10.203-214. Research, 22. 249-265.
Salthouse, T. A., & Meinz, E. J. (1995). Aging, inhibition, working
memory, and speed. Journal of Gerontology: Psychological Sciences,
SOB, P297-P306. Received August 22, 1994
Salthouse, T. A., & Prill, K. A. (1987). Inferences about age impair- Revision received December 28, 1995
ments in inferential reasoning. Psychology and Aging, 2. 43-51. Accepted December 28, 1995 i

Low Publication Prices for APA Members and Affiliates

Keeping you up-to-date. All APA Fellows, Members, Associates, and Student Affiliates
receive—as part of their annual dues—subscriptions to theAmerican Psychologist andAPA
Monitor. High School Teacher and International Affiliates receive subscriptions to the APA
Monitor, and they may subscribe to (he American Psychologist at a significantly reduced
rate. In addition, all Members and Student Affiliates are eligible for savings of up to 60%
(plus a journal credit) on all other APA journals, as well as significant discounts on
subscriptions from cooperating societies and publishers (e.g., the American Association for
Counseling and Development, Academic Press, and Human Sciences Press).

Essential resources. APA members and affiliates receive special rates for purchases of
APA books, including the Publication Manual of the American Psychological Association.
and on dozens of new topical books each year.

Other benefits of membership. Membership in APA also provides eligibility for

competitive insurance plans, continuing education programs, reduced APA convention fees,
and specialty divisions.

More information. Write to American Psychological Association, Membership Services,

750 First Street, NE, Washington, DC 20002-4242

View publication stats