adult age differences in direct and indirect tests of …memory tests both reflect the influence of...

Journal of Gerontology: PSYCHOLOGICAL SCIENCES1991. Vol. 46, No. 1. P22-30

Copyright 1991 by The Geromological Society ofAmeric

Adult Age Differences in Direct andIndirect Tests of Memory

David F. Hultsch, Michael E. J. Masson, and Brent J. Small

Department of Psychology, University of Victoria, British Columbia.

Indirect tests of memory assess the influence of recent experience on task performance without requiring awarenessof remembering. Evidence concerning whether there are reliable age differences on such indicators of implicitmemory has been inconsistent. This inconsistency may be related either to the low power of previous studies, or thecontamination of indirect measures by conscious memory retrieval strategies. In a statistically powerful test of thisquestion, indirect and direct tests of memory were administered to 584 adults from three age groups (19-36 years, 55-69 years, 70—86 years). Significant age differences in favor of the young were found on the indirect test as well as thedirect tests, suggesting that there are small but reliable age differences in implicit memory. Correlational analysesexamining the relationship of memory performance to other cognitive variables indicated that the indirect test wassupported by different components than the direct tests.

A MONG adults the relationship between age and memory•**• performance varies widely with type of memory test. Ithas been clearly established, for example, that older adultsdo not perform as well as younger adults on tests associatedwith episodic memory, such as recall and recognition, thatrequire a subject to remember contextual information aboutan event (e.g., Burke & Light, 1981; Craik, 1977; Poon,1985). A different pattern of results has been found usingtests associated with semantic memory, such as fact re-trieval, vocabulary, and semantic priming. These tests in-volve access of general knowledge without reference tospecific contextual details. It has been suggested that olderadults perform as well as or better than younger adults on thisclass of memory tests (e.g., Burke & Harrold, 1988;Perlmutter, 1986), although there are instances in whichsmall deficits are found among older adults (e.g., Salthouse,1988). The contrast between episodic and semantic memorywith respect to age-related effects clearly shows that mem-ory abilities do not uniformly decline with age.

A third type of memory task recently has yielded yetanother pattern of age-related effects (see Howard, 1988;Light, 1988 for review). These tests have been referred to asindirect or implicit memory tests (Richardson-Klavehn &Bjork, 1988; Schacter, 1987). In contrast to episodic andsemantic memory tests, which direct a subject to retrievespecific events or facts, indirect tests require only that thesubject perform an assigned task without reference to partic-ular prior events. Indirect tests are like episodic memorytasks in that they assess the influence of some recent experi-ence on task performance. But with indirect tests, thisinfluence typically is revealed through fluent reprocessing ofa stimulus or performance of a task, rather than directrecollection of the experience. For example, in the wordstem completion task, subjects are required to complete aword stem with the first word that comes to mind (e.g.,

CON ). Prior presentation of a valid completion for astem enhances the probability of its generation, implying aform of memory for the initial presentation (Graf &

Mandler, 1984). The influence of prior experience on wordstem completion is not necessarily mediated by an explicitrecognition of that experience. Support for this assertion isseen in the fact that certain types of amnesic subjects, whoexhibit severe episodic memory deficits, produce normallevels of enhancement on indirect memory tasks (e.g.,Cohen & Squire, 1980; Graf, Squire, & Mandler, 1984).

The pattern of age-related effects in indirect memory testsis one of small differences in favor of younger subjects.Often these differences are not statistically significant (e.g.,Light & Singh, 1987; Light, Singh, & Capps, 1986; Mitch-ell, 1989; Moscovitch, 1982; Moscovitch, Winocur, &McLachlan, 1986), but in some studies they are reliable(Chiarello & Hoyer, 1988; Howard, Heisey, & Shaw, 1986;Rose, Yesavage, Hill, & Bower, 1986). There are twofactors that may contribute to lack of consistency in thestatistical significance of age differences on this class oftests. One involves the possibility that subjects performingan indirect memory task sometimes attempt explicit retrievalof information about previously presented items. This strat-egy may be induced by certain details of the experimentalprocedure such as indicating, prior to presentation of a list ofitems, that a memory test will be given afterward (Perruchet& Baveaux, 1989), or administering a direct memory testbefore an indirect one (Chiarello, Hoyer, & Frazier, 1987).Attempts to consciously remember items from the originallist, a form of episodic memory, presumably would be moresuccessful for younger subjects and could artifactually en-hance age differences on a putative indirect memory test.

A second factor that may be responsible for inconsistentage-related effects in indirect memory is the potentiallysmall magnitude of any existing age differences. Studiesbased on sample sizes of 30 or so subjects per age group maynot have sufficient power to consistently detect small ageeffects. For example, Light and Singh (1987) obtained nosignificant age effects in word stem completion, but reportedthat the power of their studies to detect a small age differencein that task was no greater than .30. In contrast, Chiarello

P22

at University of V

ictoria on May 29, 2014

http://geronj.oxfordjournals.org/D

ownloaded from

http://geronj.oxfordjournals.org/

DIRECT AND INDIRECT TESTS OF MEMORY P23

and Hoyer (1988) found a significant age difference on theirword stem completion task, and attributed their finding toenhanced power (.78) brought about by a larger sample sizethan that used by Light and Singh. There is a problem,however, in ascribing Chiarello and Hoyer's significant ageeffect to increased sample size. In Experiment 1 of Light andSingh's study there were 32 subjects per age group, whereasin Chiarello and Hoyer's study there were 36 subjects perage group, hardly a large enough change in sample size toincrease power from .30 to .78. Rather, the source of thedifference in power between these two experiments appearsto be a difference in effect size. For example, the agedifference in means on the word stem completion task for theLight and Singh experiment was .06 compared to .09 for theChiarello and Hoyer study. When translated into measuresof effect size for computation of power, the differencebetween these age effects might be magnified. A possiblereason for the larger effect size in the Chiarello and Hoyerstudy is that explicit retrieval strategies may have made acontribution to performance on the word stem completiontask. Evidence for this comes from the fact that the age effectwas of similar magnitude for the cued recall and word stemcompletion tasks. This is surprising in the context of theusual finding of larger age differences in direct memory tests(Howard, 1988; Light & Singh, 1987; Light et al., 1986;Rabinowitz, 1986).

Given the uncertainty regarding the existence of smallbut reliable age differences in indirect memory tests, one ofthe goals of this study was to provide an unusually powerfultest for an age-related effect. In order to increase power, weused a very large sample of subjects from three age groups.One indirect measure of memory (word stem completion)and three direct measures of memory reflecting both episodicand semantic memory (word, text, and fact recall) wereobtained in order to examine age and gender differences inperformance.

A second objective was to explore the possible relation-ship between indirect and direct tests of memory. Our viewis that development of a framework for understanding age-related change in memory functioning depends, in part, onspecification of the relationships among types of memoryskill and the pattern of age-related change associated witheach kind of ability. Although memory theories that advo-cate the existence of separate subsystems suggest that fewcorrelations between classes of memory tests should befound (e.g., Squire & Cohen, 1984; Tulving, 1983, 1985),there are links between indirect tests of memory and bothepisodic and semantic memory tests. Indirect and episodicmemory tests both reflect the influence of a recent experi-ence, and indirect and semantic memory tests appear to besomewhat resistant to effects of aging.

Two recent, small-sample correlational studies haveyielded conflicting results regarding the relationship be-tween indirect and direct tests of memory. Using factoranalysis, Perruchet and Baveaux (1989) obtained a factorstructure in which two indirect tests of memory, includingword stem completion, loaded on a factor with free recalland recognition, whereas two other indirect memory testsloaded on a factor of their own. The fact that subjects in thisstudy were warned of impending memory tests prior to initial

encoding of the target items may have contributed to the useof explicit memory strategies when performing some of theindirect memory tests. On the other hand, Mitchell (1989)found that measures associated with an indirect test, namingrepeatedly presented pictures, loaded on a factor that wasdistinct from two other factors characterized as episodicmemory and semantic memory. Unfortunately, Mitchell'smeasures of indirect memory were based on response time,and the factor defined by these tests also included the onlyother speeded test, a semantic retrieval efficiency measure.Rather than being an indirect memory factor, it is possiblethat Mitchell obtained a separate processing speed factor.

Our strategy in assessing the relationships among memorytests was to focus on the correlations between an indirectmemory test and multiple cognitive abilities as well as otherkinds of memory tests. As with earlier studies, we wereinterested in age-related differences in the pattern of correla-tions between the word stem completion test and othermeasures of episodic and semantic memory. We expectedthat if memory for a recent experience were accessed andused similarly when performing indirect and episodic mem-ory tests, there would be significant correlations betweenthese two types of test and, perhaps, corresponding declinesin performance with advancing age. On the other hand, if asclaimed by some (e.g., Graf & Mandler, 1984), indirecttests of verbal memory involve activation of existing repre-sentations in a similar fashion to semantic memory tests, onemight expect these two types of test to be correlated and toresist change in later life. Finally, it is possible that indirectmemory performance depends on processes or aspects of anencoding episode that are different from those used in eithertype of direct memory test and adheres to its own pattern ofage-related change. Thus, we were also interested in therelationship between performance on the indirect and directtests of memory and a number of component processes andabilities. In order to examine this question, we includedmeasures of verbal speed, working memory, and verbalability in the battery of tests administered to the subjects.

METHOD

SubjectsThe sample consisted of 584 community-dwelling adults

(343 women and 241 men) residing in a medium-sizedmetropolitan area. They were recruited through advertise-ments in the public media and through specific appeals touniversity and community groups. Subjects were paid anominal fee of $15.00 for their participation in the study.Complete data on the measures of relevance to the presentanalysis were available on 544 of the subjects (319 womenand 225 men), and it is these data that are reported here. Thesample was divided into three age groups: 19-36 years, 55-69 years, and 70-86 years. Examination of the educationallevels of the participants summarized in Table 1 showed thatthe younger adults had significantly more years of educationthan the two older groups, F(2,538) = 10.12, p < .001.Males also had significantly more years of education thanfemales, F(l,538) = 5.96, p < .015. For all age groups,however, educational attainment and other demographiccharacteristics suggested that the sample was above average

at University of V



ownloaded from


P24 HULTSCH ET AL.

Table 1. Means and Standard Deviations for Age and Education for the Sample

Variable

Age/Gender Group

19-36 55-69

Females(« = 52)

Males(n = 44)

Females(« = 149)

Males(« = 96)

70-86

Females(« = 118)

Males(n = 85)

Age

Years of education

MSD

MSD

24.314.89

14.132.06

24.954.21

15.292.15

64.493.12

13.363.02

64.912.97

13.622.98

74.413.90

12.792.66

73.823.52

13.383.72

compared to the general population. Additional details re-garding the sample, including comparison of the samplewith the general population, may be found in Hultsch,Hertzog, and Dixon (1990).

Measurement of Memory PerformanceMemory performance was measured by one indirect and

three direct memory tasks. The indirect test consisted of aword stem completion task, and the direct tests consisted ofsingle trial recall of world knowledge, narrative texts, andcategorical word lists. The first of these direct tasks may beseen as a measure of semantic memory, whereas the lattertwo tasks may be seen as measures of episodic memory.Two indicators each of the world knowledge, text recall, andword recall tasks were administered to subjects. Becauseprevious analyses indicated these measures coalesced intoseparate world knowledge, text memory, and word memoryfactors (Hultsch et al., 1990), scores on the two indicatorswithin each domain were averaged together to yield compos-ite measures that were used for the present analysis.

Stem completion. —Two lists of 30 nouns were prepared.The words contained from five to seven letters with anaverage frequency of 171.1 occurrences per million (Kucera& Francis, 1967). Each word began with a different three-letter stem which could be completed into multiple words (M= 72.8; Webster's Pocket Dictionary). Subjects were ex-posed to the words in the context of a lexical decision task(described below). Half of the subjects were exposed to listA and half to list B. Immediately following the lexicaldecision task, subjects were given the stem completion task.The first three letters from each of the 60 words were printedfollowed by a 2 cm line. The stems were presented inrandom order in a booklet. Subjects were asked to completeeach stem using the first word that came to mind. Indirectmemory was assessed by determining for each list of wordstems the proportion that was completed using the criticalword (i.e., from lists A and B). For each subject, the list ofitems that was seen in the lexical decision task was classifiedas target items. The list that was not seen was classified asbaseline items. A subject's indirect memory score wascomputed as the difference between the proportion of targetitems given as stem completions and the proportion ofbaseline items given as stem completions. The differencein these two proportions reflects the influence of memoryfor a recent experience with a word on subsequent stemcompletion.

World knowledge recall. — This indicator of semanticmemory was defined by performance on two sets of 40questions about factual knowledge in the domains of sci-ence, history, sports, geography, and entertainment drawnfrom the Nelson and Narens (1980) measure. The questionsrequired recall of the information. A total of 240 questionswere selected to construct six sets of questions equated fortopic and difficulty (based on the Nelson & Narens, 1980norms). Random thirds of the sample received blocks of twosets of questions. This was done in order to minimize retesteffects across planned future occasions of measurement (theprocedure permits complete counterbalancing of two sets ofquestions over three occasions of measurement). The pro-portion of correct responses was used as the measure in thepresent analysis.

Story recall. — Six narrative stories from a set of 25parallel texts developed by Dixon, Hultsch, and Hertzog(1989) were used. Each story described an event in the life(lives) of an older protagonist(s). The stories were designed tobe structurally equivalent. All were 24 sentences and approxi-mately 300 words in length and contained approximately 160propositions as identified by Kintsch's (1974) system forrepresenting the meaning of texts. The main theme of eachpassage appeared in the first sentence. The stories werepresented in typed booklets for study followed by writtenrecall. Subjects studied the stories for 4 minutes (reading rateof 75 wpm) and had 10 minutes to write their recall, a rate thatour previous work indicates is sufficient time for older sub-jects to perform the task. As with the world knowledge tasks,random thirds of the sample received two stories for study andrecall. The proportion of propositions recalled was used as anindicator of text memory for the present analysis. Subjects'written recall protocols were checked against the propositionsof the original text base. Propositions were scored as correctif the "gist" of the proposition's meaning was expressed (seeDixon et al., 1989 for additional information on this scoringsystem). A separate reliability analysis was conducted on thescoring system for the present data. Twelve stories werescored independently by seven different raters. Given anaverage of 160 propositions per story, each rater made ap-proximately 2,000 scoring decisions. Across all possiblepairs of scorers, the percentage agreement ranged from 90 to93%, with an average of 90.9% agreement.

Word recall. — Six categorized lists of English nounswere developed using the Howard (1980) and Battig and

at University of V



ownloaded from



Montague (1969) norms. Each list consisted of 30 nounsfrom six taxonomic categories (five words per category).Categories and exemplars were chosen to minimize potentialinterference effects within and between lists. In general,high frequency exemplars (ranks 2-9) were chosen, but tominimize guessing, the two most frequently ranked nounswere not selected. The lists were presented in unblockedorder in typed booklets for study followed by written recall.Subjects studied the words for 2 minutes and had 5 minutesto write their recall, a rate that our previous work indicates issufficient time for older subjects to perform the task. As withthe world knowledge and text tasks, random thirds of thesample received blocks of two lists for study and recall. Theproportion of words recalled was used as the measure in thepresent analysis.

CorrelatesThe correlates consisted of 14 measures tapping multiple

cognitive ability components. The components were onesthat have been found to be important correlates of perfor-mance on direct memory tasks (Hultsch, Hertzog, & Dixon,1984, 1990; Hartley, 1986, 1988). Specifically, there werefour measures of verbal speed, two measures of workingmemory, and four measures of verbal ability. In addition,two indicators of skill specific to the word and story recalltasks were available: category clustering and reading com-prehension. Additional details regarding the specific tasksdescribed below may be found in Hultsch et al. (1990).

Verbal speed. — Tasks in this domain included twospeeded decision tasks and two measures of speed of pro-cessing text materials. The lexical decision task requiredsubjects to decide as rapidly as possible whether a 5- to 7-letter stimulus that appeared on the computer screen was anEnglish word or not (e.g., island vs nabion). A total of 60decisions were made. Median latencies of correct responsesand proportion of errors were used as measures.

The semantic decision task required subjects to decidewhether a sentence appearing on the computer screen wassensible or not (in the sense of its plausible occurrence in theworld as we know it; e.g., ' 'The tree fell to the ground with aloud crash." vs "The pig gave birth to a litter of kittens thismorning."). A total of 50 decisions were made. Medianlatencies of correct responses were used, and proportion oferrors were used as measures.

In the search for answer task, subjects were asked to reada passage one sentence at a time and search for the answer toa previously presented question. The search was self-pacedand the answer to the question was stated directly in thepassage. A total of 16 passages were seen. The measure usedwas the total search time across all passages.

The reading speed measure was derived from the readingcomprehension task. Subjects were asked to read a passageone sentence at a time on the computer screen. At the end ofthe passage, the subjects were asked five questions about thecontent of the passage in order to measure reading compre-hension. There were six passages in all. The reading speedmeasure was computed as time per proposition. A trimmedmean across all six passages (48 sentences) was used (scoresequal to or greater than three standard deviations from the

subject's initial mean score were deleted, and the subject'smean score was then recomputed).

Working memory. — This domain was identified by twotasks requiring the subject to process new information whileholding old information in memory to be reported later. Inthe nonverbal number tracking task, subjects were shown aseries of digits on the computer screen one digit at a time.The numbers remained on the screen for 1 second with a 500millisecond interval between digits. The length of the serieswas indeterminate, ranging from 5 to 9 digits. Three sets of10 trials were given. In the first set, the task of the subjectwas to hold in mind the last digit while processing the nextdigit. In the second set, the subject was to hold in mind thenext-to-last digit. Finally, in the third set, the subject was tohold in mind the third-last digit. In all sets, at the end of theindeterminate series of digits, the subject was asked to reportthe target digit. In the present analysis, the number ofcorrectly reported digits from the third set was used. Theverbal sentence construction task required subjects to readaloud a series of sentences presented on flash cards. Oneword of each sentence, printed in capital letters and under-lined in red, was designated as the word subjects were tokeep in mind while processing the series of sentences. At theend of the series, the subject was to report the target words inorder. The reported target words formed a new sentence.Sets of 3, 4, 5, and 6 sentences yielding new sentences 3, 4,5, and 6 words in length were used. There were three sets ofeach length for a total of 12 items. The score consists of thenumber of sentences correctly reported.

Verbal ability. — This domain was indexed by four testsfrom the Kit of Factor Referenced Cognitive Tests (Ekstrom,French, Harman, & Dermen, 1976). Three of the tests areindicators of various aspects of verbal fluency. The Con-trolled Associations test asks subjects to write as manywords as they can that have meanings that are the same ornearly the same as a set of target words. The Opposites taskasks subjects to write as many words as they can that havemeanings that are the opposite or nearly the opposite as a setof target words. Finally, the Figures of Speech task askssubjects to write as many words or phrases as possible thatcan be used to complete figures of speech which compareone object with another (e.g., "The fur was as soft as. . . " ) .These tasks were presented in booklets to small groups ofsubjects under the time limits specified for the tests. Thefourth measure was an indicator of verbal comprehension. Atotal of 54 recognition vocabulary items from several testswere used. Included were Part 1 of Vocabulary Test II (V-2),Part 2 of Advanced Vocabulary Test I (V-4), and Part 2 ofAdvanced Vocabulary Test II (V-5). The items were pre-sented in a booklet to small groups of subjects under the timelimits specified for the three parts. The number of correctitems was used as the measure in the present analysis.

Reading comprehension. —This variable was indexed bya memory-loaded comprehension task. Subjects were askedto read short narratives presented one sentence at a time onthe computer screen for understanding. Immediately follow-ing reading, subjects were asked to answer a set of five

at University of V



ownloaded from


P26 HULTSCH ET AL.

factual questions about the passage. A total of six passageswere read and tested by five questions each. The numberof correct answers over all six passages was used as themeasure.

Clustering. — This variable was based on clusteringscores derived from the two word recall tasks. For each wordlist, ARC scores (Roenker, Thompson, & Brown, 1971)were computed from the subject's recall. An average cluster-ing score from both word lists was used.

Experimental ProceduresThe tasks were administered during three sessions sched-

uled over a period of about one month. Each session wasabout two hours in length with a rest break in the middle. Thepaper-and-pencil tasks were administered during the firsttwo sessions scheduled about a week apart. Small groups ofabout five to ten subjects were tested at a time. The computertasks were administered in the third session scheduled duringthe next two or three weeks. These tasks were administeredto individual subjects. The order of the tasks within sessionswas invariant.

RESULTS

The data analysis consisted of two parts. The first partexamined differences in performance on the four memorytasks as a function of gender and age. The second partassessed relations between the memory tasks themselves andbetween performance on the memory tasks and the variousindicators of cognitive ability.

Gender and Age DifferencesInitial univariate analyses including gender, age, and

counterbalancing group as factors revealed significant ef-fects related to counterbalancing group for fact memory andword memory, a significant Age x Counterbalancing Groupinteraction for the stem completion difference score, and sig-nificant Age x Counterbalancing Group and Gender xCounterbalancing Group interactions for text recall. Coun-terbalancing was included in the design for future plannedoccasions of measurement, and is not of theoretical interestin the present analysis. Further, inclusion of the counterbal-ancing factor in the analysis had no impact on the pattern ofgender and age effects. Therefore, the analyses we reportcollapse across counterbalancing group.

In order to examine gender and age differences, a 2(Gender) x 3 (Age Group) MANOVA with alpha set at .05was performed on the scores from the four memory tasks.This analysis showed significant effects related to gender,Wilkes \ = .79, F(4,535) = 35.86, p < .001, and age,WilkesX = .71,F(8,1070) = 22.42, p < .001. The Genderx Age Group interaction was not significant.

Examination of the univariate gender effects revealed thatfemales scored significantly higher on both story recall (Mf

= .343, Mm = .326), F(\ ,538) = 6.05, p < .05, and wordrecall (Mf = .305, Mm = .289), F(l,538) = 8.38, p < . 0 1 ,but males outperformed females on fact recall (Mf = .426,Mm = .536), F(\, 538) = 74.90,/? < .001. There were nosignificant gender differences on the stem completion task(Mf = .057, Mm = .061).

Table 2 shows the average proportion correct on the fourmemory tasks by age group. In the stem completion task,memory was defined as the amount of improvement overbaseline performance that accrued as a result of prior expo-sure to a target word. This improvement was reflected indifference scores that were calculated by subtracting propor-tion correct in the baseline condition from proportion correctin the target condition. The univariate analysis of the differ-ence scores revealed a significant age effect, F(2,538) =7.36, p < .01, w2 = .023. A Neuman-Keuls test with alphaset at .05 indicated that the youngest group obtained higherscores than either of the two older groups, who did not differfrom each other. Although the three age groups did notreliably differ in the probability of using critical words tocomplete stems in the target condition, the two older groupswere more likely to use critical words in completing stems inthe baseline condition than the youngest group, F(2,538) =13.52, p < .001, a)2 = .051, thus yielding smaller differ-ence scores for the older subjects. It is not clear why oldersubjects were more likely than younger subjects to completethe baseline stems with the critical words that we hadchosen. But the difference scores, which reflect enhance-ment due to prior exposure of the critical words, reliablyfavored the youngest group.

Univariate analyses indicated significant age effects on allthree direct memory measures (fact recall: F(2,538) = 8.23,p < .001, to2 = .026; word recall: F(2,538) = 27.34, p <.001, (o2 = .089; and story recall: F(2,538) = 51.69, p <.001, u)2 = .158). Note that although all of the age effectsare highly reliable in terms of alpha level, they vary consid-erably in terms of proportion of variance accounted for asestimated by the omega squared statistic. Further analysesusing the Neuman-Keuls procedure were conducted to ex-amine the nature of these age group differences. In the caseof the fact recall task, the middle-aged (55-69 year) groupperformed better than the oldest group who, in turn, per-formed better than the youngest group. In the case of theword recall and story recall tasks, the youngest group re-called more than the middle-aged group who, in turn, re-called more than the oldest group.

Given the fact that there were significant gender and agedifferences in educational attainment, we also computed aMANCOVA on the four memory measures using years of

Table 2. Means and Standard Deviations (in Parentheses)of Proportion Correct for Indirect and Direct Tests

of Memory by Age Group

Test

Stem CompletionTargetBaselineDifference

Fact Recall

Story Recall

Word Recall

19-36 years

.188 (.057)

.110 (.043)

.078 (.060)

.429 (.164)

.428 (.112)

.668 (.128)

Age Group

55-69 years

.189 (.048)

.134 (.043)

.055 (.052)

.494 (.145)

.329 (.100)

.614 (.138)

70-89 years

.186 (.050)

.131 (.041)

.055 (.050)

.465 (.163)

.299 (.096)

.545 (.144)

at University of V



ownloaded from



education as a covariate. The results of this analysis werevirtually identical with those reported above.

Correlational AnalysesThe correlational analyses were designed to examine,

within each age group, relationships among the four memorymeasures and between each memory measure and the set of14 additional variables. For these correlational analyses, analpha level of .01 was used as the criterion of statistical sig-nificance. This relatively stringent criterion was chosen inorder to reduce the likelihood of significant correlations dueto Type I error, given the large number being computed (62correlations in each age group). In addition, the correlationvalues we report were obtained by partialing out of the fourmemory tests variance related to gender, counterbalancingcondition, and the interaction between these two factors.This strategy was chosen because the initial analyses ofmean performance including counterbalancing group indi-cated that this factor and its interaction with gender ac-counted for significant variability in memory test scores.

The initial correlational analysis examined the relation-ships among the four memory measures. Of particular con-cern is the relationship between the indirect measure and thethree direct measures. Significant correlations of the stemcompletion task with the direct measures could indicate thatexplicit memory processes were operating in the completionof the indirect test. The results are shown in the upperportions of Tables 3, 4, and 5 for each of the three agegroups. Examination of the first column of the upper portion

Table 4. Correlations of All Measures for the55-69-Year-Old Group (N = 245)

Explicit MemoryFact recallStory recallWord recall

Verbal SpeedLexical decision task

ErrorsLatency

Semantic decision taskErrorsLatency

Search for answerReading speed

Working MemoryNumber trackingSentence construction

Verbal AbilityControlled associatesOppositesFigures of speechVerbal comprehension

Reading comprehension

Clustering

StemCompletion

.075

.101

.110

- .059- .083

- .006- .009- .066

.030

- .020.138

.089

.077

.043

.071

.166*

- .006

FactRecall

1.000.393**.395**

- . 1 7 1 *- .293**

- .037- .429**- .484**- .146

.058

.367**

.462**

.441**

.274**

.661**

.398**

.140

StoryRecall

1.000.424**

- .093-.290**

- .076- .370**-.372**- .068

.151

.302**

.296**

.319**

.290**

.394**

.439**

.119

WordRecall

1.000

- .121- .182*

- .140- .276**- .463**- .194*

.140

.369**

.402**

.385**

.334**

.403**

.421**

.484**

*p< .01;**p< .001.





ErrorsLatency






Clustering

StemCompletion

- .095- .053- .144

.081- .029

- .059.110.105

- .169

- .030- .237

- .107- .213- .038- .186

- .231

.066

FactRecall

1.000.424**.353**

- .176-.354**

- .326*-.418**- .426**- .149

.071

.327*

.391**

.319*

.199

.689**

.472**

.172

StoryRecall

1.000.539**

- .092_ 414**

- .186- .423**-.437**- .110

.077

.414**

.283*

.226- .019

.389**

.503**

.308*

WordRecall

1.000

- .114- .282*

- .044-.374**- .299*- .005

.153

.423**

.275*

.308*

.177

.245

.401**

.439**



ErrorsLatency






Clustering

StemCompletion

.071

.012

.032

.022- .068

.074- .072

.003- .051

- .150-.071

.062

.012- .007- .082

- .022

- .035

FactRecall

1.000.465**.528**

-.232**- .296**

- .220*- .425**- .574**-.414**

.138

.324**

.460**

.458**

.334**

.575**

.391**

.255**

StoryRecall

1.000.518**

- .155-.282**

- .135-.297**-.579**-.255**

.216*

.393**

.360**

.377**

.325**

.367**

.508**

.264**

WordRecall

1.000

- .195*- .313**

-.239**- .313**-.556**-.376**

.207*

.467**

.400**

.363**

.288**

.318**

.377**

.352**

*p< . .001. *p< .0\;**p< .001.

at University of V



ownloaded from


P28 HULTSCH ET AL.

of these tables shows that none of the correlations betweenthe indirect test and the three direct tests was significant. Asexpected, significant correlations were observed among thevarious direct tests.

The remainder of Tables 3,4, and 5 report the correlationsbetween the four memory measures and performance on thecognitive ability variables. Examination of the tables showsa striking paucity of significant correlations with the stemcompletion measure. Of the 14 variables examined, therewas only one significant relationship (a correlation of .17with reading comprehension in the middle-aged group). Incontrast, multiple significant relationships are seen betweenthe direct memory tests and the various cognitive abilitymeasures (25 for those aged 19-36 years; 28 for those aged55-69 years; 39 for those aged 70-86 years).

One possible explanation for the lack of significant corre-lations with the stem completion task is that this measurewas unreliable. In order to examine this possibility, wecomputed Cronbach alphas on randomly selected halves ofthe stem completion list. The proportion of target stemscorrectly completed from each of the random halves, minusthe baseline measure for the entire list of words, was used tocalculate the correlations. In addition to all 100 youngersubjects, 120 subjects from each of the two older age groupswere selected at random for this analysis. Cronbach alphaswere computed separately for each of the age/counterbalanc-ing groups. Apart from a modest Cronbach alpha of .43 (p <.01) for the oldest group tested using list A for the target set,all the other Cronbach alphas were within the range of .61 to.69 (p < .01). The square root of these measures of reliabil-ity constitutes an upper bound on the correlation between thestem completion task and any other possible measure. Forexample, given a reliability of .60, the upper bound of thecorrelation is .77. Because the magnitude of these upperbound estimates far exceeds the level of correlation neededfor significance with our sample size (.16 to .28 dependingon the age group), the stem completion measure was judgedto be acceptably reliable.

DISCUSSION

The results of the present analysis provide clear evidenceof significant age differences on one indirect test of memory,specifically a stem completion task. Younger adults' stemcompletion responses were more strongly influenced bypreviously seen words, suggesting a more robust implicitmemory for the material. The finding of a significant agedifference on the stem completion test is at odds with anumber of previous studies (e.g., Light & Singh, 1987),although it is consistent with others (e.g., Chiarello &Hoyer, 1988). Earlier, we pointed out two potential reasonsfor the lack of consistent findings related to age differenceson indirect memory tests: (a) lack of sufficient power todetect small but reliable differences, and (b) contaminationof the indirect tests by episodic memory processes in somestudies.

In the present instance, we find no evidence suggestingthat the stem completion task was contaminated by episodicmemory processes. The words were originally presented tothe subjects in the context of a lexical decision task. In thiscontext, the time available for processing the semantic

meaning of the material is very limited, and the subject is notin any way led to expect that later recall or recognition of theitems will be requested. Moreover, the correlational analy-ses provided no evidence of significant relationships be-tween the three direct tests and the stem completion mea-sure. To the extent that performance on the indirect test wascontaminated by retrieval processes intentionally directed atitems in the recently experienced list, one would expect suchrelationships.

Given that clear age differences are observed in theabsence of any evidence for contamination by episodicretrieval strategies, it is argued that designs with substantialpower are required to reliably detect significant age-relateddifferences on indirect tests such as stem completion (cf.Chiarello & Hoyer, 1988). Within this context, it is impor-tant to note that the amount of variance associated with agedifferences varied considerably across the four memorytasks. Age was associated with about 2% of the variance onboth the indirect and semantic tasks, whereas it accountedfor 8 and 15% on the two episodic tasks. Thus, it appears thatalthough there are reliable age differences on at least sometypes of indirect tests of memory (e.g., stem completion), incomparison to episodic tests, these differences account for asmall portion of the variance. A similar result was obtainedby O'Sullivan (1988) using picture fragment completion asan indirect test of memory.

The correlational results extend the inferences that can bedrawn from the analyses of mean differences in severalways. Despite the fact that age differences in mean scoresfavoring the young were observed on the indirect test and thetwo direct episodic tests, the correlational analyses clearlysuggest that these two types of tasks were supported bydifferent component processes. First, there were no signifi-cant correlations between the direct and indirect tests ofmemory in any of the age groups. In contrast, all of the directmeasures showed significant relationships with one anotherin all three age groups. This latter result is reassuring giventhe fact that the various memory tasks involved widelydifferent materials. Such a finding suggests that the lack ofrelationship between the indirect test and the direct testscannot simply be ascribed to material differences. Second,the three direct tests showed multiple significant relation-ships with various component process measures thought tobe important for memory functioning. In contrast, virtuallyno significant correlations were observed between the vari-ous measures of cognitive processing ability and the indirectstem completion measure.

The distinction between component processes involved inthe direct and indirect memory tests used here can best bemade in terms of the contrast between conceptually drivenand data driven processes. Conceptually driven processesare directed toward the meaning of a stimulus, and areresponsible for enabling accurate direct memory for events.On the other hand, data driven processes are responsible forperceptual analysis of stimuli, and are not typically adequatefor successful performance on direct memory tests. Theimportance of semantic processing of stimuli for later recalland recognition tests has been well established (e.g., Craik&Lockhart, 1972; Craik &Tulving, 1975). The influence ofprior experience with a word on the word stem completion

at University of V



ownloaded from



test, however, appears to depend on the application of datadriven processes to the word's printed form. Completion ofword stems with previously seen target words is not stronglyinfluenced by the degree of conceptual processing applied tothe words during their initial encoding (Graf & Mandler,1984; Graf, Mandler, & Haden, 1982). If words are notactually seen, but generated from a clue (e.g., tin-C ),or named in response to presentation of a picture (Roediger& Blaxton, 1987; Weldon, Roediger, & Challis, 1989), laterword stem completion shows little or no influence of theprior experience.

The correlational pattern within the three direct tasks isconsistent with earlier work (e.g., Hultsch et al., 1984;Hartley, 1986). Performance on these direct tasks is corre-lated with multiple information processing components suchas verbal speed and working memory as well as intellectualabilities such as verbal fluency and verbal comprehension.These findings are discussed in detail by Hultsch et al.(1990). We also found a pattern of somewhat higher correla-tions between the processing and ability components andperformance on the memory tasks with increasing age. Thispattern is consistent with the results of factor analytic studiesof intelligence, which have shown stronger interrelation-ships among primary mental ability factors with increasingage (Cunningham, 1987; Hertzog & Schaie, 1986).

In sum, the present results suggest that small but reliableage differences in favor of the young are observed on at leastone indirect measure of memory. Although age accounts formore variance in the case of direct tests, the present resultssuggest that the failure of previous studies to find significantage effects on tasks similar to the present stem completionmeasure may be related to the power of the design rather thanthe absence of reliable age differences on such tasks. Fi-nally, the correlational analyses indicate quite clearly thatthe component processes involved in the stem completiontask are different from those involved in fact, word, or storyrecall. Clearly, it is necessary to determine whether thispattern of results will be seen with other types of indirecttests. A related next step is to explore the possibility thatcomponents reflective of data driven processes will correlatewith indirect measures such as stem completion, or thatmore conceptually driven indirect tests of memory will ex-hibit correlational convergence with direct tests and theircomponents.

ACKNOWLEDGMENTS

This research was supported by Operating Grants from the NaturalSciences and Engineering Research Council of Canada to Drs. Hultsch andMasson.

Address correspondence and requests for reprints to David F. Hultsch,Department of Psychology, University of Victoria, P.O. Box 3050, Victo-ria, B.C., V8W 3P5, Canada.

REFERENCES

Battig, W. F., & Montague, W. E. (1969). Category norms for verbal itemsin 56 categories: A replication and extension of the Connecticut CategoryNorms. Journal of Experimental Psvcholog\ Monograph, 80, (3)(Part 2).

Burke, D. M., & Harrold, R. M. (1988). Automatic and effortful semanticprocesses in old age: Experimental and naturalistic approaches. In L. L.Light & D. M. Burke (Eds.), Language, memory and aging (pp. 100-116). New York: Cambridge University Press.

Burke, D. M., & Light, L. L. (1981). Memory and aging: The role ofretrieval processes. Psychological Bulletin, 90, 513-546.

Chiarello, C., & Hoyer, W. J. (1988). Adult age differences in implicit andexplicit memory: Time course and encoding effects. Psychology andAging, 3, 358-366.

Chiarello, C , Hoyer, W. J.,&Frazier, L. (1987). Encoding conditions andimplicit memory in young and elderly adults. Journal of Clinical andExperimental Neuropsychology, 9, 37.

Cohen, N. J., & Squire, L. R. (1980). Preserved learning and retention ofpattern analyzing skill in amnesics: Dissociation of knowing how andknowing that. Science, 210, 207-210.

Craik, F. I. M. (1977). Age differences in human memory. In J. E. Birren& K. W. Schaie (Eds.), Handbook of the psychology of aging (pp. 384—420). New York: Van Nostrand Reinhold.

Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: Aframework for memory research. Journal of Verbal Learning and VerbalBehavior, 77,671-684.

Craik, F. I. M., & Tulving, E. (1975). Depth of processing and theretention of words in episodic memory. Journal of Experimental Psy-chology: General, 104, 268-294.

Cunningham, W. R. (1987). Intellectual abilities and age. In K. W. Schaie(Ed.), Annual review of gerontology and geriatrics (Vol. 7, pp. 117-134). New York: Springer.

Dixon, R. A., Hultsch, D. F., & Hertzog, C. (April, 1989). A manual of 25three-tiered structurally equivalent texts for use in aging research.CRGCA Technical Report No. 2, Department of Psychology, Universityof Victoria, B.C., Canada.

Ekstrom, R. B., French, J. W., Harman. H. H., & Dermen, D. (1976).Manual for kit of factor referenced cognitive tests. Princeton, NJ:Educational Testing Service.

Graf, P., & Mandler, G. (1984). Activation makes words more accessible,but not necessarily more retrievable. Journal of Verbal Learning andVerbal Behavior, 23, 553-568.

Graf, P., Mandler, G., & Haden, P. E. (1982). Simulating amnesicsymptoms in normals. Science, 218, 1243-1244.

Graf, P., Squire, L. R., & Mandler, G. (1984). The information thatamnesic patients do not forget. Journal of Experimental Psychology:Learning Memory and Cognition, 10, 164—178.

Hartley, J. T. (1986). Reader and text variables as determinants of dis-course memory in adulthood. Psychology and Aging, 1, 150-158.

Hartley, J. T. (1988). Individual differences in memory for written dis-course. In L. L. Light & D. M. Burke (Eds.), Language, memory andaging (pp. 36-57). New York: Cambridge University Press.

Hertzog, C , & Schaie, K. W. (1986). Stability and change in adultintelligence: I. Analysis of longitudinal covariance structures. Psychol-ogy and Aging, 1, 159-171.

Howard, D. (1980). Category norms: A comparison of the Battig andMontague (1969) norms with the responses of adults between the ages of20 and 80. Journal of Gerontology, 35, 225-231.

Howard, D. V. (1988). Implicit and explicit assessment of cognitive aging.In M. L. Howe & C. J. Brainerd (Eds.), Cognitive development inadulthood: Progress in cognitive development research (pp. 3-37). NewYork: Springer-Verlag.

Howard, D. V., Heisey, J. G., & Shaw, R. J. (1986). Aging and thepriming of newly learned associations. Developmental Psychology, 22,78-85.

Hultsch, D. F., Hertzog, C , & Dixon, R. A. (1984). Text recall inadulthood: The role of intellectual abilities. Developmental Psychology,20, 1193-1211.

Hultsch, D. F., Hertzog, C , & Dixon, R. A. (1990). Ability correlates ofmemory performance in adulthood and aging. Psychology and Aging, 5,356-368.

Kintsch, W. (1974). The representation of meaning in memory. Hillsdale,NJ: Erlbaum.

Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American English. Providence, RI: Brown University Press.

Light, L. L. (1988). Preserved implicit memory in old age. In M. M.Gruneberg, P. E. Morris, & R. N. Sykes (Eds.), Practical aspects ofmemory: Current research and issues (Vol. 2, pp. 90-95). Chichester,England: Wiley.

Light, L. L., & Singh, A. (1987). Implicit and explicit memory in young

at University of V



ownloaded from


P30 HULTSCH ET AL.

and older adults. Journal of Experimental Psychology: Learning, Mem-ory, and Cognition, 13, 531-541.

Light, L. L., Singh, A., & Capps, J. L. (1986). Dissociation of memoryand awareness in young and older adults. Journal of Clinical andExperimental Neuropsychology, 8, 62—74.

Mitchell, D. B. (1989). How many memory systems? Evidence from aging.Journal of Experimental Psychology: Learning, Memory, and Cogni-tion, 15, 31-49.

Moscovitch, M. (1982). A neuropsychological approach to perception andmemory in normal and pathological aging. In F. I. ML Craik & S. Trehub(Eds.), Aging and cognitive processes (pp. 55-78). New York: Plenum.

Moscovitch, M., Winocur, G., & McLachlan, D. (1986). Memory asassessed by recognition and reading time in normal and memory-impaired people with Alzheimer's disease and other neurological disor-ders. Journal of Experimental Psychology: General, 115, 331-347.

Nelson, T. O., & Narens, L. (1980). Norms of 300 general-informationquestions: Accuracy of recall, latency of recall, and feeling-of-knowingratings. Journal of Verbal Learning and Verbal Behavior, 19, 338-368.

O'Sullivan, C. D. (1988). A dissociation between verbal and pictorialimplicit memory in an elderly population. Unpublished doctoral disserta-tion. University of Victoria, B.C., Canada.

Perlmutter, M. (1986). A life-span view of memory. In P. B. Bakes, D. L.Featherman, & R. M. Lerner (Eds.), Life-span development and behav-ior (Vol. 7, pp. 271-313). Hillsdale, NJ: Erlbaum.

Perruchet, P., & Baveaux, P. (1989). Correlational analyses of explicit andimplicit memory performance. Memory & Cognition, 17, 77-86.

Poon, L. W. (1985). Differences in human memory with aging: Nature,causes, and clinical implications. In J. E. Birren & K. W. Schaie (Eds.),Handbook of the psychology of aging (2nd ed., pp. 427-462). NewYork: Van Nostrand Reinhold.

Rabinowitz, J. C. (1986). Priming in episodic memory. Journal of Geron-tology, 41, 204-213.

Richardson-Klavehn, A., & Bjork, R. A. (1988). Measures of memory.Annual Review of Psychology, 39, 475-543.

Roediger, H. L., & Blaxton, T. A. (1987). Retrieval modes producedissociations in memory for surface information. In D. S. Gorfein & R.R. Hoffman (Eds.), The Ebbinghaus Centennial Conference (pp. 349-379). Hillsdale, NJ: Erlbaum.

Roenker, D. L., Thompson, C. P., & Brown, S. C. (1971). Comparison ofmeasures for the estimation of clustering in free recall. PsychologicalBulletin, 76, 45-48.

Rose, T. L., Yesavage, J. A., Hill, R. D., & Bower, G. H. (1986). Primingeffects and recognition memory in young and elderly adults. Experimen-tal Aging Research, 12, 31-37.

Salthouse, T. A. (1988). Effects of aging on verbal abilities: Examinationof the psychometric literature. In L. L. Light & D. M. Burke (Eds.),Language, memory, and aging (pp. 17-35). New York: CambridgeUniversity Press.

Schacter, D. L. (1987). Implicit memory: History and current status.Journal of Experimental Psychology: Learning, Memory, and Cogni-tion, 13, 501-518.

Squire, L. R., & Cohen, N. J. (1984). Human memory and amnesia. In G.Lynch, J. L. McGaugh, & N. M. Weinberger (Eds.), Neurobiology oflearning and memory (pp. 3-64). New York: Guilford.

Tulving, E. (1983). Elements of episodic memory. New York: OxfordUniversity Press.

Tulving, E. (1985). How many memory systems are there? AmericanPsychologist, 40, 385-398.

Weldon, M. S., Roediger, H. L., & Challis, B. H. (1989). The propertiesof retrieval cues constrain the picture superiority effect. Memory &Cognition, 17, 95-105.

Received August 22, 1989Accepted November 9, 1989

at University of V



ownloaded from


adult age differences in direct and indirect tests of …memory tests both reflect the influence of...

Documents