ArticlePDF Available

Examining the Locus of Age Effects on Complex Span Tasks

September 2003
Psychology and Aging 18(3):562-72

September 2003
18(3):562-72

DOI:10.1037/0882-7974.18.3.562

Source
PubMed

Authors:

Jennifer Mccabe

Goucher College

To investigate the locus of age effects on complex span tasks, the authors evaluated the contributions of working memory functions and processing speed. Age differences were found in measures of storage capacity, language processing speed, and lower level speed. Statistically controlling for each of these in hierarchical regressions substantially reduced, but did not eliminate, the complex span age effect. Accounting for lower level speed and storage, however, removed essentially the entire age effect, suggesting that both functions play important and independent roles. Additional evidence for the role of storage capacity was the absence of complex span age differences with span size calibrated to individual word span performance. Explanations for age differences based on inhibition and concurrent task performamce were not supported.

Correlations Among Variables With Age Group Partialed Out

…

Figures - uploaded by Jennifer Mccabe

Content may be subject to copyright.

Content uploaded by Jennifer Mccabe

Content may be subject to copyright.

Examining the Locus of Age Effects on Complex Span Tasks

Jennifer McCabe and Marilyn Hartman

University of North Carolina at Chapel Hill

To investigate the locus of age effects on complex span tasks, the authors evaluated the contributions of

working memory functions and processing speed. Age differences were found in measures of storage

capacity, language processing speed, and lower level speed. Statistically controlling for each of these in

hierarchical regressions substantially reduced, but did not eliminate, the complex span age effect.

Accounting for lower level speed and storage, however, removed essentially the entire age effect,

suggesting that both functions play important and independent roles. Additional evidence for the role of

storage capacity was the absence of complex span age differences with span size calibrated to individual

word span performance. Explanations for age differences based on inhibition and concurrent task

performamce were not supported.

Working memory is a multicomponent system that combines

aspects of both storage and processing. The traditional model

(Baddeley & Hitch, 1974, 1994) posits two code-specific storage

buffers and a central executive. The storage buffers include the

phonological loop, which maintains verbal information, and the

visuospatial sketchpad, which stores visual and spatial informa-

tion. The central executive allocates attentional resources to con-

trol access to the storage buffers and to perform other processing

requirements such as the coordination of concurrent tasks and

inhibition of irrelevant information. A recent formulation of the

model (Baddeley, 2000) also includes a temporary multimodal

storage component called the episodic buffer.

It is well established that working memory is negatively affected

by normal aging (e.g., Light, 1996; Salthouse, 1990; Verhaeghen,

Marcoen, & Goossens, 1993). The type of task most widely used

to document these age differences is the complex span task (see

Verhaeghen et al., 1993, for a meta-analysis), in which participants

perform a processing task while remembering target items (e.g.,

the final words in a series of sentences) for later recall. Several

variants of complex span have all shown age differences (Chiappe,

Hasher, & Siegel, 2000; Li, 1999; Lustig, May, & Hasher, 2001;

Myerson, Hale, Rhee, & Jenkins, 1999). Studies have also shown

that complex span performance can predict age differences in

higher order tasks such as language comprehension and episodic

memory (e.g., Hess & Tate, 1992; Kwong See & Ryan, 1995).

The reading span task (Daneman & Carpenter, 1980), a common

complex span measure, requires the performance of a processing

task involving reading and comprehending sets of sentences while

remembering the final word of each sentence. Because complex

span tasks engage multiple cognitive processes related to working

memory, it has been difficult to identify the source (or sources) of

the age effect. Declines in complex span tasks may result from

impairments in verbal storage capacity, coordination of concurrent

tasks (e.g., Engle, Nations, & Cantor, 1990), inhibition of nontar-

get information (Hasher & Zacks, 1988), and in the case of the

reading span task, syntactic processing and semantic integration

(e.g., Daneman & Merikle, 1996). Additionally, lower level pro-

cessing speed (Salthouse, 1996) appears to make an important

contribution to performance on these tasks. Whereas various re-

searchers have proposed age differences in each of these constructs

as the source of the complex span age effect, there is no consensus

as to which are most important. It is the goal of the current study

to further investigate this question.

One component function of working memory (Baddeley &

Hitch, 1974, 1994) that could account for the complex span age

effect is reduced storage capacity for verbal material, both in the

phonological loop and in the episodic buffer (Baddeley, 2000).

The storage-deficit hypothesis has found support in the findings of

age-related declines in word span (e.g., Light & Anderson, 1985;

Verhaeghen et al., 1993) and digit span tasks (e.g., Dobbs & Rule,

1989; Gregoire & Van der Linden, 1997; Verhaeghen et al., 1993;

but see Belleville, Peretz, & Malenfant, 1995; Fisk & Warr, 1996).

Furthermore, although at least one study has found a greater age

effect on complex span tasks compared with simple span tasks

(e.g., Wingfield, Stine, Lahar, & Aberdeen, 1988), a meta-analysis

indicated that age effects on complex and simple word span tasks

are of a similar magnitude, suggesting that the additional process-

ing requirements of complex span tests do not increase the age

effect (Verhaeghen et al., 1993). As further evidence for the

storage-deficit hypothesis, Belleville, Rouleau, and Caza (1998)

found that controlling for word span capacity eliminated any

further effects of age on an alphabet span task requiring mental

manipulation of the words. This suggests again that younger and

older adults differ not in their ability to manipulate information in

working memory but in storage capacity (see also Babcock &

Jennifer McCabe and Marilyn Hartman, Department of Psychology,

University of North Carolina at Chapel Hill.

This study was completed in partial fulfillment of the master’s program

requirement of Jennifer McCabe. Portions of these data were presented at

the Cognitive Aging Conference, Atlanta, Georgia, April 2002. This re-

search was supported in part by National Institute on Aging Grant

AG10593 awarded to Marilyn Hartman. We gratefully acknowledge Brea

Stratton, Mandy Schleiffer, and Marcus Johnson for their assistance in

programming, data collection, and scoring.

Correspondence concerning this article should be addressed to Jennifer

McCabe, Department of Psychology, Davie Hall, CB #3270, University of

North Carolina, Chapel Hill, North Carolina 27599-3270. E-mail:

jmccabe@email.unc.edu

2003, Vol. 18, No. 3, 562–572 0882-7974/03/$12.00 DOI: 10.1037/0882-7974.18.3.562

562

Salthouse, 1990). As evidence against the storage hypothesis,

however, Gick, Craik, and Morris (1988) found that the reading

span age effect remained relatively constant as the number of

sentences presented per trial, and therefore the number of words to

remember, increased. In sum, although a large number of studies

have suggested that reduced storage capacity accounts for some

degree of age-related decline in working memory tests, it is still

unclear whether storage capacity can completely account for these

age differences.

Another component of working memory that may lead to age

differences in performance on complex span tasks is the central

executive. The central executive carries out a number of functions

(Baddeley, 1996), one of which is to coordinate the simultaneous

performance of multiple tasks (Baddeley & Hitch, 1974, 1994).

Older adults may have reduced proficiency in performing the

processing task and the memory task concurrently during complex

span tasks. Contrary to this hypothesis, however, Gick et al. (1988)

found no difference in the magnitude of the age effect in complex

span for divided-attention and single-task span conditions. For

paradigms other than complex span tasks, however, there are

conflicting data. Whereas some studies have found greater dual-

task costs for older adults compared with younger adults (e.g.,

Salthouse, Rogan, & Prill, 1984), others have found that once

single-task performance is equated for the two age groups, older

adults are not differentially affected by adding a secondary task

(Baddeley, Logie, Bressi, Della Sala, & Spinnler, 1986; Somberg

& Salthouse, 1982).

In addition to coordinating multiple tasks, the central executive

also plays a role in inhibiting irrelevant information. Conse-

quently, if older adults have reduced ability to keep irrelevant

information out of working memory and to inhibit previously

relevant information that has become irrelevant to task goals

(Hasher & Zacks, 1988), this may cause working memory to

decline. With regard to complex span performance, a decline in the

deletion function of inhibition (Hasher, Zacks, & May, 1999)

would result in older adults having more difficulty in inhibiting

nontarget words once they have been processed and in suppressing

interference from previous trials. Support for this explanation

comes from a study by Lustig et al. (2001), who in order to

examine the role of proactive interference, compared the standard

reading span procedure with a modified procedure that begins with

larger span set sizes and progressively reduces the set size. It was

hypothesized that the “descending” format would reduce proactive

interference for the more difficult larger span trials. As predicted,

the descending format significantly increased span scores for older

adults and eliminated age differences (see also May, Hasher, &

Kane, 1999). Also consistent with the inhibition-deficit hypothesis

is the finding that age differences on a complex span task are

greater when the background material is highly similar to the

material to be remembered (Li, 1999). Thus, interference from

irrelevant information may be particularly problematic for older

adults when it is difficult to discriminate from relevant, to-be-

remembered information.

Despite the evidence for the inhibition-deficit theory, other

studies offer evidence against this explanation. For instance,

Schelstraete and Hupet (2002) recently tested the association be-

tween age differences on measures of inhibition (i.e., intrusion

errors on a reading span task and Stroop interference) and reading

span performance. Although interference control contributed to

reading span, the inhibition measures could not explain the effect

of age. An additional line of evidence against the inhibition-deficit

theory is found in a meta-analysis of age differences on working

memory tasks. Whereas older adults performed more poorly than

younger adults on complex span tasks, the age effects were similar

regardless of whether the domains of the memory task and the

processing task were the same or different (Jenkins, Myerson,

Hale, & Fry, 1999; see also Jenkins, Myerson, Joerding, & Hale,

2000). In sum, there is evidence suggesting a decline in inhibition

among older adults; however, there is enough conflicting evidence

to question whether it can fully explain age differences on complex

span tasks.

Another explanation for age differences on certain complex

span tasks (i.e., reading span) is a decline in aspects of sentence

processing that depend on working memory, namely syntactic

processing and semantic integration (e.g., Just & Carpenter, 1992).

These sentence comprehension abilities are related to both the

storage and processing aspects of working memory because they

involve the maintenance of words and the integration of syntactic

structure and meaning of the sentence. Evidence for age-related

decline in sentence processing includes findings that greater de-

grees of sentence complexity differentially reduce both complex

span performance (Gick et al., 1988) and memory for prose (Nor-

man, Kemper, Kynette, Cheung, & Anagnopoulos, 1991) in older

adults. Furthermore, reduced comprehension of complex sentences

in older adults has been attributed to their inability to allocate

attentional and memory resources to parts of sentences that are

more difficult to comprehend (Stine-Morrow, Ryan, & Leonard,

2000). Recently, Waters and Caplan (2001) found evidence that

older adults are differentially affected by syntactic complexity as

measured by sentence acceptability judgments and that this mea-

sure was correlated with working memory capacity. However,

on-line sentence reading measures that required more automatic

sentence processing abilities showed equivalent effects of syntac-

tic complexity for older and younger adults and no significant

correlations with working memory. Thus, although the evidence is

mixed, the results of several studies suggest that older adults may

have problems with sentence comprehension. These in turn could

contribute to age differences in reading span.

The effect of age on complex span could also be explained by

reduced speed of processing in older adults. Salthouse’s (1991,

1996) processing speed theory of cognitive aging is a general

model used to account for age differences on many cognitive tasks,

including those tapping working memory. Although a generalized

reduction in processing speed would be expected to affect cogni-

tive operations at all levels, two mechanisms by which it could

affect working memory are to reduce the speed of rehearsal in the

phonological loop (Baddeley & Hitch, 1994) and to decrease the

availability of information from earlier processing steps (Salt-

house, 1996). Many studies have shown that processing speed,

commonly measured by perceptual comparison tasks, is associated

with a substantial amount of variance on complex span tasks and

that age differences on these tasks are largely eliminated after

statistically controlling for speed (e.g., Fisk & Warr, 1996; Salt-

house, 1994; Salthouse & Babcock, 1991; Verhaeghen & Salt-

house, 1997).

A recent study by Park et al. (2002) provides further support for

the processing speed theory by using a formal modeling approach.

To evaluate the strength of common explanatory constructs related

563

COMPLEX SPAN TASKS AND AGING

to working memory, Park et al. assessed age-related performance

on tasks of sensory functioning, processing speed, short-term

memory, working memory, long-term memory, and verbal ability.

In addition to supporting the separation of working memory into

verbal and visuospatial domains across all age groups, the resulting

model pointed to processing speed as the strongest mediator be-

tween age and cognitive tasks, including those tapping working

memory. Because rates of age-related declines were similar across

all cognitive measures, the results supported the hypothesis that

slowed speed of processing may be the fundamental construct that

drives cognitive reductions in old age. This and many other studies

provide strong evidence that the relationship between age and

working memory is mediated by an age-related decrease in pro-

cessing speed (see also Salthouse, 1992). However, there is also

some evidence that processing speed might not provide a full

account of the complex span age effect (e.g., Gick et al., 1988;

Keys & White, 2000).

This review of evidence for and against each interpretation of

age differences on complex span tasks highlights the lack of a

coherent theoretical account of this phenomenon. One reason for

the conflicting data may be the inconsistent use of methodology

capable of isolating the source(s) of age differences. Thus, al-

though complex span tasks place multiple cognitive demands on

test takers, few studies have independently measured age differ-

ences in each task component. A comprehensive approach to the

question would assess performance on each task component, de-

termine how well each component that shows age differences

predicts the complex span age effect, and explore what combina-

tion of predictors can best account for the age differences.

Overview of the Current Experiment

The goal of the current study was to contrast the predictions

made by the different hypotheses regarding the age effect on

complex span tasks. We included two verbal complex span tasks.

The reading span task (RST) required participants to read and

verify sentences aloud while remembering the sentence-final

words (Daneman & Carpenter, 1980), and the list span task (LST)

involved reading word lists aloud and identifying animal names

while remembering the list-final words (De Beni, Palladino, Paz-

zaglia, & Cornoldi, 1998). We focused on identifying the cognitive

locus of the age effect in these complex span tasks by examining

age differences on the task components that correspond to different

working memory functions, either by using independent measures

of the components or by comparing tasks that varied only in one

component. Statistical regression techniques were also applied to

determine which working memory and/or speed-related factors are

most successful in explaining the age effect on complex span tasks.

The design of the experiment addressed each of the major

hypotheses regarding age differences in working memory. The

storage-deficit hypothesis was tested in three ways. First, age

differences in word span and complex span were compared. If

storage is the critical constraint on age-related complex span

performance, there should be similar effects of age on the two

types of tasks; however, if complex span components other than

storage capacity drive the age effect, there should be larger age

differences on the complex spans. Our second test of this hypoth-

esis was to measure each participant’s complex span performance

at a span size calibrated to his or her word span ability. If storage

capacity is critical, then older and younger adults should perform

similarly on the complex span tasks when age differences in

storage are accounted for in this way. Third, we used word span

performance as a predictor of complex span to determine whether

it would significantly reduce age-related complex span variance.

We also tested the hypothesis that older adults have reduced

central executive resources, resulting in difficulties with concur-

rent task performance. We compared age differences on the word

span task and a dual-task version of word span in which partici-

pants read aloud a series of words for later recall while pressing a

key whenever they read an animal name (De Beni et al., 1998). If

the processing of concurrent tasks affects older adults, then a

differential age effect in the dual-task condition would be

expected.

To investigate an additional manifestation of a reduction in

central executive abilities, we examined age-related declines in the

ability to inhibit irrelevant information. The primary means of

testing the inhibition-deficit theory (Hasher & Zacks, 1988) was to

examine intrusion errors during the complex span tasks (i.e., RST

and LST). The intrusion measures assessed the deletion function of

inhibition (Hasher et al., 1999); if older adults are less able to

inhibit the nonfinal words from working memory, their recall

should contain a greater proportion of intrusions. In addition, they

might be expected to have an especially high rate of animal-name

intrusion errors on the LST because these words receive additional

processing (De Beni et al., 1998). As another test of the deletion

function, we compared age differences on the LST with those on

the dual-task word span. These tasks differed mainly in the re-

quirement of the LST to inhibit nonfinal words from working

memory. A decline in inhibitory function would be expected to

differentially reduce older adults’ performance on the LST com-

pared with the dual-task word span, in which all material presented

during a trial is relevant to the memory task.

We assessed the hypothesis that a syntactic deficit could explain

age differences by comparing the two versions of complex span.

The RST and LST differed solely in terms of the syntactic pro-

cessing requirement, as both required participants to read words

aloud and to recall the final word of each sentence or list. If

reduced syntactic processing is responsible for complex span age

differences, then the age effect should be greater on the RST than

on the LST.

The final hypotheses that we tested were related to the process-

ing speed theory (Salthouse, 1996). We included three independent

measures of processing speed and compared the effects of age on

complex span performance before and after statistical control of

performance on these tasks to determine whether they were able to

substantially reduce the age effect. As an extension of the process-

ing speed hypothesis, we also considered whether age differences

in the speed of processing higher level verbal material, which is

determined by lower level speed as well as by language-specific

processing speed, might be predictive of complex span perfor-

mance. Salthouse and Babcock (1991) found that the efficiency of

processing the verbal background tests of complex span tasks was

a better predictor of the complex span age effect than simple

storage and nonmemory task coordination, although not nearly as

strong a predictor as lower level processing speed. To test the role

of this higher level language processing speed construct in the

current study, we independently measured reaction time on the

verbal background tasks of the RST and the LST and then tested

564

MCCABE AND HARTMAN

the relationship of language processing speed to complex span

performance and examined the degree of overlap between the

higher level and lower level speed constructs.

The main predictions of this experiment were that the complex

span age effect would be mostly explained by age differences in

working memory storage capacity, higher level language process-

ing speed, and/or lower level processing speed. In addition, we

expected that the contribution of lower level speed would be

partially independent of the contribution of higher level speed and

storage capacity, such that either of these last two would account

for unique age-related variance above and beyond that of lower

level processing speed. We considered basic processing speed to

be the more fundamental cognitive construct because a decline in

this lower level ability would arguably affect the efficiency of all

cognitive operations (e.g., Salthouse, 1996). In comparison, de-

clines in the other constructs included in the study would affect

more specific higher level abilities. On the basis of this logic, we

decided to account for the variance associated with lower level

speed first in each of our regression analyses. Thus, our strongest

predictions were based on explanations involving storage capacity

and processing speed. The roles of concurrent processing demands

and inhibition ability were also tested, although we had less clear

predictions regarding these explanatory constructs.

Method

Participants

Participants included 48 younger adults and 48 older adults (see Table 1

for characteristics of the sample). The younger adults were undergraduate

students participating for course credit or payment. The older adults were

healthy volunteers who were paid for participation. All participants re-

ported good or excellent health and vision that was normal or corrected to

normal. None had a reported history of neurological disorders, uncon-

trolled hypertension, diabetes, heart attacks, emphysema, kidney disease,

or recent severe psychiatric illness, nor were they currently taking psycho-

active medication. In addition, no older participants reported excessive

alcohol use.

All participants were screened for depression and anxiety by using the

Beck Depression Inventory—II (BDI–II; Beck, Steer, & Brown, 1996) and

the Beck Anxiety Inventory (BAI; Beck & Steer, 1990). Participants were

excluded if they scored above the normal range of 0–13 on the BDI–II

and/or above the normal range of 0–9 on the BAI. Thirteen younger adults

and three older adults were excluded for this reason and were replaced.

There were no age differences in BDI–II or BAI scores. The older adults

were also screened for dementia with the Mini-Mental State Examination

(MMSE; Folstein, Folstein, & McHugh, 1975) and all scored above a

minimum cutoff score of 28.

Materials

All of the span tasks (RST, LST, dual-task word span, word span) and

the language tasks (sentence processing, word-list processing) were pre-

sented on a computer monitor. Practice trials were presented to participants

prior to each task. On the span tasks, there was a 10-s break between trials

to reduce proactive interference. At the end of each span trial, a blue recall

screen immediately appeared and participants responded orally. The lan-

guage tasks involved the presentation of the items sequentially without

breaks, and the computer recorded all responses.

Word span task. The word span task required the construction of three

trials for each span-level from two to eight, using a total of 105 words.

Words selected for the trials were matched to the sentence-final words

from the RST (Daneman & Carpenter, 1980) in terms of part of speech,

frequency, concreteness, and number of syllables, on the basis of published

norms (Kucˇera & Francis, 1967; Nelson, McEvoy, & Schreiber, 1998;

Toglia & Battig, 1978). The words within each trial were chosen to be

semantically and phonologically unrelated.

On each trial, words appeared one at a time on the computer monitor for

1 s each. Participants were instructed to read each word aloud and then to

recall all the words in order when the blue screen appeared. They were told

that it was better to report the material out of order than to not report it at

all; however, they were instructed not to say the last item they read as the

first word recalled unless it was the only one they could remember. The

task began with a span size of two words. After three trials at each level,

the span size increased by one until the participant failed at least two out

of three trials at a given size. The instructions and procedures for the span

tasks followed the reading span procedure of Daneman and Carpenter

(1980).

Dual-task word span task. The dual-task word span (De Beni et al.,

1998) used 105 animal and nonanimal words matched to Daneman and

Carpenter’s (1980) sentence-final words used in the RST in the same way

as for the word span task. Animal names were placed randomly in the

trials. Three trials were constructed for each span size from two to eight.

On each trial, words were presented one at a time for 1 s each on the

computer monitor. Participants were instructed to read the words aloud and

to press the Animal key whenever they read an animal name. At the

conclusion of each trial, participants recalled the words in order when the

blue screen appeared. Testing began with a span size of two and increased

until the participant failed at least two out of three sets at a given span size.

RST. The RST consisted of 60 sentences between 8 and 12 words long,

taken from Daneman and Carpenter (1980). The final, to-be-recalled word

in each sentence contained one to three syllables. Half of the sentences

were meaningful (i.e., they followed semantic and syntactic rules of En-

glish) and half were not. Three trials were constructed for each span size

from two to six.

In each trial, sentences were presented one at a time on the computer

monitor. Participants read each sentence aloud and made a response as to

whether it was meaningful. They pressed a key labeled Yes for sentences

that were meaningful and a key labeled No for sentences that were not.

After all sentences in a trial were completed, the blue screen appeared and

they recalled the sentence-final words in order. Span size began with two

sentences and increased by one sentence after three trials. This process

continued until the span size equaled the individual’s word span minus one.

At that span size, if a participant was still able to complete more than one

out of three trials correctly, testing continued until at least two out of three

Table 1

Characteristics of Participants

Characteristic

Older adults

Younger

adults

MSDMSD

Age 72.3 5.9 20.1 2.4

Years of education 16.8 2.0 13.9 1.1

MMSE 29.4 0.7 — —

BDI–II 3.0 3.2 3.1 3.0

BAI 1.2 2.0 2.5 2.2

Shipley–Hartford Vocabulary Test 36.9 2.5 31.0 3.6

Note. Mini-Mental State Examination (MMSE) scores represent the

number of points earned out of a maximum of 30. Dashes indicate that

MMSE data were not collected for younger adults. Shipley–Hartford

Vocabulary Test scores represent the number of correct responses out of a

maximum of 40. BDI–II ⫽ Beck Depression Inventory—II; BAI ⫽ Beck

Anxiety Inventory.

565

COMPLEX SPAN TASKS AND AGING

trials were failed at a given span size. This procedure was adopted so that

we could compare age differences in complex span scores with complex

span performance calibrated to each individual’s simple storage ability. We

used the word-span-minus-one level as the criterion rather than actual word

span to avoid floor effects in this calibrated complex span measure.

LST. The LST was based on the complex span task used by De Beni

et al. (1998) and required the construction of 60 five-word lists, each with

zero to two animal names. Words used in this task were selected on the

basis of the same criteria as for the dual-task word span. For lists with

animal names, the animal words were placed randomly in the lists, includ-

ing in the final position. Three trials of word lists were constructed for each

span size from two to six.

The LST followed much the same procedure as the RST, except that instead

of sentences, lists of five words were presented on the screen, arranged

horizontally. Participants read the words in each list aloud from left to right

while pressing a key labeled Animal for each animal name. They pressed

a key labeled End after reading each list, and at the end of each trial, they

saw the blue screen and recalled the final words of the lists in order. The

task began with a span size of two, and we used the same procedures as

those used in the RST to determine when to terminate testing.

Language processing speed tasks. The two language tasks corre-

sponded to the two complex span background tasks. The test corresponding

to the RST was a sentence processing task, which included 20 new

sentences from Daneman and Carpenter (1980). Again, half of the sen-

tences were meaningful and half were not. Participants read each sentence

aloud and pressed the Yes key or the No key, depending on whether the

sentence made sense or not. For the word-list processing task, correspond-

ing to the background task in the LST, 20 new five-word lists were created

by using the same criteria as described for the LST. This task required

participants to read each list aloud while pressing the Animal key whenever

they read an animal name. The End key was pressed immediately after

reading the last word in each list. For both language tasks, the importance

of both speed and accuracy was emphasized. The computer recorded

reaction times and accuracy for each trial.

Lower level processing speed tasks. The Pattern Comparison and Let-

ter Comparison tasks (Salthouse & Babcock, 1991), as well as the Digit–

Symbol Substitution task (Wechsler, 1997), are paper-and-pencil tests that

were administered to measure basic processing speed. In the Pattern

Comparison and Letter Comparison tasks, participants were instructed to

examine a series of either pattern or letter pairs and decide whether the

members of each pair were the same or different. Participants wrote an S

on the line separating a pair if the two stimuli were the same and a D if they

were different. For each task, the goal was to complete as many items as

possible within 30 s on each of two pages. The score for each task was the

total number of correct responses. The Digit–Symbol Substitution task, a

subtest of the Wechsler Adult Intelligence Scale—III (Wechsler, 1997),

consisted of a piece of paper at the top of which was a key that matched the

digits 1–9 with a set of 9 symbols. A series of digits was printed below with

an empty box beneath each one. Participants filled in these boxes with the

appropriate symbols. Standard administration instructions were used, and

the score was the number of items answered correctly in 2 min.

Procedure

Participants were tested individually in sessions lasting approximately 1

hr. Older adults were screened for health problems via telephone prior to

the testing session, and younger adults received the health screening at the

end of the testing session. Older adults completed the MMSE at the start of

the session. There were no other differences in procedure for younger and

older adults. The ordering of the tasks was as follows. The word span was

administered first to determine the participant’s simple span level, which

was needed for administering the RST and LST (see the Materials section).

The order of the remaining computerized tasks was counterbalanced across

participants. The lower level speed measures were interleaved with the

computerized tasks. The Shipley–Hartford Vocabulary Test (Shipley, 1940),

the BDI–II, and the BAI were administered at the end of the testing session.

Results

The RST, LST, word span, and dual-task word span tasks were

scored using two methods. For the span-level scoring method

(Daneman & Carpenter, 1980), participants received 1.0 point for

every span size at which they completed at least two out of three

trials accurately, and they received an additional 0.5 point if they

accurately completed one trial out of three at the next highest span

size. For the items scoring method (e.g., Chiappe et al., 2000;

Lustig et al., 2001; May et al., 1999), we added the total number

of target words accurately recalled on fully correct trials. With the

age groups combined, the two scoring methods were highly cor-

related for each of the four span tasks (range of rs ⫽ .96–.97, p ⬍

.001). Because the items scoring method allows for a greater range

of scores and because the span-level scoring method at times

resulted in floor effects for older adults, we primarily report items

scores; however, analyses using span-level scores are reported

when they differ from the analyses that used items scores. With

few exceptions, the two scoring systems produced similar results.

Older adults scored significantly higher than younger adults on

the Shipley–Hartford Vocabulary Test, t(94) ⫽ 8.78, p ⬍ .05,

partial

␩

⫽ .45 (see Table 1). With the effect of age removed,

vocabulary was significantly correlated with an average complex

span score (r ⫽ .33, p ⬍ .05). Thus, to remove the effect of verbal

ability in all analyses, vocabulary was used as a covariate in

analysis of covariance (ANCOVA) procedures and was entered as

the first predictor in regression models.

The alpha level was set at p ⫽ .05. We report partial

␩

as a

measure of effect size. Partial

␩

⫽ .01 represents a small effect;

partial

␩

⫽ .06, a medium effect; and partial

␩

⫽ .14, a large

effect (Cohen, 1988).

Age Differences on Complex Span Tasks

Before testing our hypotheses, we compared performance on the

two complex span tasks. In doing this, we also tested whether the

age effect was different for the two versions of complex span. A 2

(type of complex span: RST, LST) ⫻ 2 (age) ANCOVA with

vocabulary as a covariate was computed. As expected, there was a

significant effect of age, F(1, 93) ⫽ 23.60, MSE ⫽ 55.96, p ⬍ .05,

partial

␩

⫽ .20, indicating overall better performance by younger

adults. There was no effect of complex span type and no interac-

tion of complex span type and age. Vocabulary was a significant

covariate, F(1, 93) ⫽ 11.64, MSE ⫽ 55.96, p ⬍ .05, partial

␩

⫽

.11, but did not interact with other variables.

The correlation between the two complex span tasks with age

partialed out was significant (r ⫽ .53, p ⬍ .001). Because the

pattern of age-related performance was similar for the RST and the

LST, mean complex span scores were used in all subsequent

analyses that compared complex span tasks with other tasks or

those that used variables to predict complex span.

Analyses Relating to the Storage-Deficit Hypothesis

Performance on the word span task showed significant age

differences, t(94) ⫽ 3.78, p ⬍ .05, partial

␩

⫽ .13, with younger

adults performing better than older adults (see Table 2). To com-

566

MCCABE AND HARTMAN

pare the magnitude of age differences on simple versus complex

span tasks, a 2 (type of span: word span, average complex

span) ⫻ 2 (age) ANCOVA was computed with vocabulary as a

covariate. There was a trend toward a main effect of span type,

F(1, 93) ⫽ 3.59, MSE ⫽ 41.28, p ⫽ .06, partial

␩

⫽ .04; a

significant main effect of age, F(1, 93) ⫽ 28.92, MSE ⫽ 90.11,

p ⬍ .05, partial

␩

⫽ .24; and a significant interaction of the two

variables, F(1, 93) ⫽ 5.24, MSE ⫽ 41.28, p ⬍ .05, partial

␩

⫽

.05. The word span scores were somewhat higher than complex

span scores, younger adults performed better than older adults, and

the interaction indicated larger age differences in word span com-

pared with complex span. There was also a significant effect of the

vocabulary covariate, F(1, 93) ⫽ 10.60, MSE ⫽ 90.11, p ⬍ .05,

partial

␩

⫽ .10, but no significant interactions with other

variables.

We also compared younger and older adults’ complex span

performance at each individual’s word-span-minus-one level. The

dependent variable in this analysis was the number of sentence- or

list-final words correctly recalled at the word-span-minus-one

level divided by the maximum number of words that could have

been recalled at that level (see Table 3). A 2 (type of complex

span: RST, LST) ⫻ 2 (age) ANCOVA was computed with vocab-

ulary as the covariate. There was no effect of span type or of age

and no significant interaction between the two variables. There

were also no significant effects or interactions involving the vo-

cabulary covariate.

Analyses Relating to the Concurrent-Task-Deficit

Hypothesis

Older adults performed significantly worse on the dual-task

word span compared with younger adults, t(94) ⫽ 3.74, p ⬍ .05,

partial

␩

⫽ .13 (see Table 2). To assess whether the addition of

a concurrent task to a memory storage task increased age differ-

ences in memory span, a 2 (type of word span: single task, dual

task) ⫻ 2 (age) ANCOVA was calculated with vocabulary and

mean word-list processing reaction times as covariates. The word-

list processing covariate, as measured by the mean reaction time to

read and respond to animal names in the word-list processing task,

was included to control for baseline performance in this compo-

nent of the dual-task word span. The results showed no effects of

word span type, although the main effect of age was significant,

F(1, 92) ⫽ 9.77, MSE ⫽ 192.61, p ⬍ .05, partial

␩

⫽ .10,

indicating lower performance on both tasks by older adults. There

was no significant interaction between age and task type. There

was a trend toward a main effect of the vocabulary covariate, F(1,

92) ⫽ 3.40, MSE ⫽ 192.61, p ⫽ .07, partial

␩

⫽ .04, but it did

not interact with other variables. The effect of the word-list pro-

cessing covariate was significant, F(1, 92) ⫽ 6.18, MSE ⫽ 192.61,

p ⬍ .05, partial

␩

⫽ .06, but it did not interact with other

variables.

Analyses Relating to the Inhibition-Deficit Hypothesis

We first examined the proportion of intrusion errors on the RST

and LST. Intrusion errors were defined as nontarget words from

the current span trial that were erroneously recalled as target

words.

This proportion was computed by dividing the total num-

ber of nonfinal words recalled by the total number of words

recalled (see Table 4). In separate t tests, no age differences in

these scores were found for either complex span task. We next

examined age differences in a subset of LST intrusion errors, the

animal names that were erroneously recalled. This score was

calculated by dividing the number of intrusion errors that were

animal names by the total number of words recalled. In a t test, the

proportion of animal intrusion errors did not show age differences

(see Table 4). The overall proportion of intrusion errors and the

proportion of animal intrusion errors across both age groups were

extremely low, however, so it was difficult to assess age trends.

The final test of the inhibition-deficit hypothesis involved a

comparison between the LST and the dual-task word span task (see

Table 2). This 2 (type of span task: LST, dual-task word span) ⫻ 2

(age) ANCOVA showed an effect of span type, F(1, 93) ⫽ 3.61,

MSE ⫽ 56.38, p ⫽ .06, partial

␩

⫽ .07, with higher scores on the

An identical pattern of results was found when intrusion errors were

scored to include intrusions from all previous span trials. The number of

intrusion errors combined for the RST, LST, and LST animal-name intru-

sion measures increased only by a total of four words for younger adults

and by seven words for older adults when across-trial intrusions were

included.

Table 2

Means and Standard Deviations on Span Tasks Using Items

Scores for Each Age Group

Span measure

Older adults Younger adults

MSDMSD

Reading span task 8.88 5.82 13.13 8.31

List span task 8.38 5.20 11.60 5.90

Word span 34.77 9.28 42.83 11.51

Dual-task word span 37.88 11.19 47.42 13.67

Table 3

Means and Standard Deviations of Proportions of Recall on the

Reading Span Task and the List Span Task at the

Word-Span-Minus-One Level

Complex span task

Older adults Younger adults

MSDMSD

Reading span task .64 .16 .66 .16

List span task .59 .21 .62 .19

Table 4

Means and Standard Deviations of the Proportions of Intrusion

Errors on the Reading Span Task (RST) and the List Span Task

(LST)

Type of intrusion

error

Older adults Younger adults

MSDMSD

RST intrusions .06 .06 .05 .06

LST intrusions .09 .08 .06 .06

LST animal intrusions .02 .05 .01 .02

567

COMPLEX SPAN TASKS AND AGING

dual-task word span. There was also a main effect of age, F(1,

93) ⫽ 26.07, MSE ⫽ 119.54, p ⬍ .05, partial

␩

⫽ .22, with better

performance by younger adults, and an interaction of the span task

and age, F(1, 93) ⫽ 7.25, MSE ⫽ 56.38, p ⬍ .05, partial

␩

⫽ .07.

The interaction reflected greater age differences in the dual-task

word span task scores compared with the LST scores. There was

a significant effect of the vocabulary covariate, F(1, 93) ⫽ 9.85,

MSE ⫽ 119.54, p ⬍ .05, partial

␩

⫽ .10, but no interactions with

other variables.

Analyses Relating to the Processing Speed Theory

Significant age differences were found on all speed-related

measures (see Table 5). For the lower level processing speed tasks,

older adults completed fewer items correctly than younger adults

on the Pattern Comparison, t(94) ⫽ 6.00, p ⬍ .05, partial

␩

⫽ .28;

Letter Comparison, t(94) ⫽ 7.80, p ⬍ .05, partial

␩

⫽ .39; and

Digit–Symbol Substitution tasks, t(94) ⫽ 12.06, p ⬍ .05, partial

␩

⫽ .61.

On both higher level language tasks, the reaction times were

significantly faster for younger adults than for older adults—

sentence processing task: t(94) ⫽ 2.33, p ⬍ .05, partial

␩

⫽ .06;

word-list processing task: t(94) ⫽ 4.62, p ⬍ .05, partial

␩

⫽ .19.

Levels of accuracy on both tasks were high (M ⫽ .98, SD ⫽ .04),

and t tests showed no age differences in either task (see Table 6).

With age partialed out, the correlation of the sentence processing

and word-list processing reaction times was significant (r ⫽ .61,

p ⬍ .001).

Predictors of Complex Span Performance

A series of hierarchical regression models tested the hypotheses

that controlling for task components that showed significant age

differences would substantially reduce the age effect on complex

span tasks. To this end, we computed separate regression analyses

with performance on the simple storage measure, the higher level

language measures, and the lower level processing speed measures

as individual complex span predictors. We then explored a hier-

archical combination of lower level processing speed with each of

the task components as predictors of complex span performance.

The concurrent task and inhibition measures were not included as

predictors because they failed to show a differential effect of age.

All regression models were computed with vocabulary as the first

predictor. The mean score on the two complex span tasks was used

as the dependent variable in each analysis (see Table 7 for corre-

lations among predictors and complex span tasks).

We computed a preliminary regression model with age as the

sole predictor of average complex span score. This model, as

expected, showed age to be a significant predictor, accounting

for 20.2% of the variance (p ⬍ .05). Subsequent regression models

included the experimental tasks as primary predictors, followed by

age, allowing us to compare the amount of variance accounted for

by each task with effects of age alone (see Table 8 for regression

results).

The first of these regression models used word span as a

predictor of complex span performance, followed by age. The

results showed that word span performance was highly predictive

of complex span performance, accounting for 31.2% of complex

span task variance (p ⬍ .05). Age was still a significant predictor,

however, accounting for an additional 5.4% of the variance (p ⬍

.05). Compared with the age-only model, adding word span as a

predictor reduced the age effect by 73.3%.

The second regression model tested the ability of higher level

language processing to predict complex span performance. This

model included the mean of the reaction times for the sentence and

word-list processing speed tasks as the first predictor, followed by

age. The language task reaction time was a significant predictor,

accounting for 13.7% of the variance (p ⬍ .05). However, age was

still a significant predictor of complex span score, accounting for

an additional 9.6% of the variance (p ⬍ .05). Compared with the

age-only model, controlling for the language tasks reduced the age

effect by 52.5%.

The lower level speed tasks were used to predict complex span

in the next model, followed by age. All three speed tasks were

entered as a group in one level of the regression model, and

accounted for 26.3% of complex span variance (p ⬍ .05). The age

variable, however, still contributed 4.2% of the variance (p ⬍ .05).

Controlling for the simple speed tasks reduced the age effect

by 79.2% compared with the age-only model.

The subsequent set of hierarchical regression models tested the

prediction that lower level processing speed, plus one of the task

components, would render the age effect nonsignificant. The first

regression model included the lower level speed measures, fol-

lowed by word span and then age. As noted above, the simple

speed tasks, entered as a group, were associated with 26.3% of the

Table 5

Means and Standard Deviations of Scores for the Lower Level

Processing Speed Tasks

Lower level speed task

Older adults Younger adults

MSDMSD

Pattern Comparison 32.5 5.2 40.6 7.8

Letter Comparison 20.6 3.6 27.7 5.2

Digit–Symbol Substitution 62.8 12.2 89.5 9.3

Note. Scores on the Pattern Comparison and Letter Comparison tasks

represent the number of correct responses in 1 min. Digit–Symbol Substi-

tution scores represent the number of correct responses in 2 min.

Table 6

Means and Standard Deviations of Reaction Times (RTs) and

Accuracy Levels on the Sentence Processing and Word-List

Processing Speed Tasks

Language task

Older adults Younger adults

MSDMSD

Sentence processing

RT 4,380 879 3,968 853

Accuracy .98 .04 .97 .04

Word-list processing

RT 4,079 950 3,314 643

Accuracy .98 .04 .99 .03

Note. RTs are presented in milliseconds. Accuracy levels represent the

proportions of correct responses.

568

MCCABE AND HARTMAN

variance (p ⬍ .05). The word span task accounted for an addi-

tional 16.6% of the variance in this model (p ⬍ .05). Including

lower level processing speed as the first predictor approximately

halved the contribution of simple storage to complex span perfor-

mance, and the combination of lower level speed and storage

rendered the contribution of age-related variance nonsignificant.

Compared with the age-only model, these predictors reduced the

age effect by 95.5%. We cannot, however, strongly conclude that

these two constructs completely eliminated the age effect because

the corresponding analysis using the span-level scoring method for

complex span measures showed a small amount of significant

remaining age-related variance.

Although there is some discrep-

ancy in results, both scoring methods resulted in a substantial

reduction of the complex span age effect after statistically control-

ling for lower level speed and storage capacity.

A final model was computed by using lower level processing

speed as the first predictor, followed by average reaction time for

the language tasks and then age. The lower level processing speed

tasks were associated with 26.3% of the complex span variance

(p ⬍ .05), and although the variance associated with the language

tasks was substantially reduced compared with the model without

the simple speed tasks, it accounted for an additional 3.4% of

complex span variance (p ⬍ .05). With the addition of these two

predictors to the model, the age effect was reduced by 83.2%

compared with the age-alone model. The age effect remained

significant, however, accounting for 3.4% of the variance (p ⬍

.05).

Because our regression analyses up to this point supported the

importance of both lower level processing speed and storage

capacity in the complex span age effect, we computed a post hoc

regression model that explored the contribution of lower level

speed to word span. This model, which included vocabulary, lower

level speed tasks, and age as hierarchical predictors of word span,

showed that lower level speed accounted for 12.9% of unique

variance (p ⬍ .05). Age still contributed an additional 7.0% of

unique variance to the model (p ⬍ .05). For comparison, age alone

accounted for 18.3% (p ⬍ .05) of word span variance. Compared

with the age-only model, the addition of the speed tasks to the

regression equation decreased the age effect by 61.7%.

The hierarchical regression model with lower level processing speed,

word span, and age as predictors of complex span performance showed

slightly different results when the span-level method of scoring was used,

in that age was still a significant predictor after accounting for lower level

speed and word span, adding an additional 2.8% of variance to the model

(p ⬍ .05). Including the lower level speed measures and the word span as

predictors reduced the age effect by 89.3% compared with the regression

model using span-level scores with age as the sole predictor. Due to this

discrepancy, we cannot positively conclude that storage capacity and lower

level speed eliminate the age effect; however, the age effect was reduced

substantially, supporting our conclusion of significant and unique contri-

butions of storage in working memory and of lower level speed to the

complex span age effect.

Table 7

Correlations Among Variables With Age Group Partialed Out

Variable 1234567891011

1. Mean complex span —

2. RST .91** —

3. LST .84** .53** —

4. Dual-task word span .56** .46** .54** —

5. Word span .50** .44** .44** .64** —

6. Sentence processing ⫺.25* ⫺.22* ⫺.22* ⫺.21* ⫺.26* —

7. Word-list processing ⫺.27** ⫺.24* ⫺.23* ⫺.23* ⫺.35** .61** —

8. Pattern Comparison .05 .09 ⫺.02 .09 .05 ⫺.26* ⫺.24* —

9. Letter Comparison .34** .34** .24* .18 .15 ⫺.20 ⫺.19 .44** —

10. Digit–Symbol Substitution .19 .19 .14 .11 .07 ⫺.35** ⫺.32** .37** .60** —

11. Vocabulary .33** .23* .37** .24* .25* ⫺.29** ⫺.31** .13 .11 .08 —

Note. RST ⫽ reading span task; LST ⫽ list span task.

* p ⬍ .05. **p ⬍ .01.

Table 8

Results of Hierarchical Regression Analyses Predicting Mean

Complex Span Score Using Items Scores

Predictor R

Increase in R

Fdf