Journal of Gerontology: PSYCHOLOGICAL SCIENCES Copyright 1995 by The Geromological Society of America 1995, Vol. 50B, No. 3, P162-P17O Adult Age Differences in Perceptually Based, But Not Conceptually Based Implicit Tests of

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

Department of Psychology, University of Victoria.

Implicit tests of memory assess the influence of recent experience without requiring awareness of remembering. Evidence concerning age differences on implicit tests of memory suggests small age differences in favor of younger adults. However, the majority of research examining this issue has relied upon perceptually based implicit tests. Recently, a second type of implicit test, one that relies upon conceptually based processes, has been identified. The pattern of age differences on this second type of implicit test is less clear. In the present study, we examined the pat- tern of age differences oh one conceptually based (fact completion) and one perceptually based (stem completion) implicit test of memory, as well as two explicit tests of memory (fact and word ). Tasks were administered to 403 adults from three age groups (19-34 years, 58-73 years, 74-89 years). Significant age differences in favor of the young were found on stem completion but not fact completion. Age differences were present for both word and fact recall. Correlational analyses examining the relationship of memory performance to other cognitive variables indicated that the implicit tests were supported by different components than the explicit tests, as well as being different from each other.

N RECENT years, there has been a great deal of interest in tasks rely on the semantic meaning of the study items and I implicit tests of memory (Graf & Masson, 1993; Lewan- less upon the perceptual characteristics of the items. The dosky, Dunn, & Kirsner, 1989). Unlike explicit tests of memory tasks Blaxton used were constructed so each type of memory (e.g., recall and recognition), which directly ask task, explicit or implicit, would require either conceptual or subjects to use their memory to retrieve past information, perceptual processing. The results indicated that perfor- implicit tests require only that subjects perform an assigned mance on the memory tests did not correspond to the type of task without reference to particular prior events. With im- memory task but rather to the type of processing required by plicit tests, memory is revealed as a facilitation or change in each of the tasks. Specifically, performance on the conceptu- task performance that is attributable to exposure to informa- ally based tasks was greatest when the semantic meaning of tion during a previous episode (Jacoby & Dallas, 1981). For the study words was processed. On the other hand, perfor- example, in the word stem completion task, subjects are mance on the perceptually based tasks was greatest when the required to complete a word stem with the first word that perceptual characteristics at study matched those at test. comes to mind (e.g., CON ). Prior presentation of a Blaxton concluded that distinctions cannot be made simply valid completion for a stem enhances the probability of its on the basis of whether a task employs implicit or explicit generation, implying a form of memory for the initial pre- test instructions. The nature of the processing operations is sentation. This facilitation of task performance is called critical in determining performance. (Schacter, 1987). The distinction between conceptually driven and percep- Much of the early work with implicit tests of memory tually driven implicit tests of memory has interesting impli- centered on dissociations from explicit tests of memory. For cations for age differences in memory performance. The example, whereas explicit tests typically benefit from deep pattern of age-related effects in implicit tests of memory is and elaborative semantic , the magnitude of prim- one of small differences in favor of younger subjects. Often ing effects on some implicit tests are insensitive to the levels these differences are not statistically significant (Light & of processing manipulation (Graf & Mandler, 1984; Singh, 1987; Light, Singh, & Capps, 1986; Mitchell, 1989; Hashtroudi, Ferguson, Rappold, & Crosniak, 1988). Simi- Russo & Parkin, 1993), but in some studies they are reliable larly, amnesics exhibit severe deficits on explicit tests of (Chiarello & Hoyer, 1988; Howard, Heisey, & Shaw, 1986; memory, but their performance on implicit tests is often Hultsch, Masson, & Small, 1991; Rose, Yesavage, Hill, & similar to normal subjects (Cohen & Squire, 1980; see Bower, 1986). One factor that may be responsible for Shimamura, 1986, for a review). inconsistent age-related effects in implicit tests of memory is Recently, dissociations have been observed among im- the potentially small magnitude of any existing age differ- plicit tests of memory themselves (Blaxton, 1989; Wither- ences. For example, Hultsch et al. (1991) found the age- spoon & Moscovitch, 1989). Blaxton compared perfor- related effect in stem completion accounted for roughly mance on two explicit and two implicit tests of memory. The 2.5% of variance, whereas the age effect in word recall tasks used were classified into conceptually based or percep- accounted for approximately 9% of variance. Therefore, tually based tasks. Perceptually based tasks rely on process- studies based on sample sizes of 30 or so subjects per age ing the physical features of the presented stimulus in order to group may not have sufficient power to consistently detect perform the tasks. On the other hand, conceptually based small age effects. A recent meta-analysis suggests that

P162 AGE DIFFERENCES IN P163

insufficient statistical power may be the reason inconsisten- analysis, Perruchet and Baveux (1989) obtained a factor cies in results have been observed. Light and La Voie (1993) structure in which two perceptually based implicit tests compared age differences on a number of implicit tests (stem completion, perceptual clarification) loaded on a fac- including word stem completion, word fragment comple- tor with and recognition, whereas two other tion, and perceptual identification. They found small, but perceptually based implicit tests loaded on a factor of their reliable, age differences in implicit test performance. Meta- own. However, subjects in the study were warned of im- analyses achieve great statistical power because they pool pending memory tests prior to initial encoding of the target the results of studies, thus combining subjects from different items, and this may have contributed to the use of explicit experiments to achieve a larger effective sample size. memory strategies when performing some of the implicit Although it might be tempting to conclude there are small tests of memory. On the other hand, Mitchell (1989) found age differences in implicit tests of memory, the majority of that measures associated with a perceptually based implicit tasks used to examine this question have been perceptually test (naming repeatedly presented pictures), loaded on a based. The few studies that have employed conceptually factor that was distinct from two other factors, characterized based implicit tests have found no evidence of reliable age as episodic and . Although this finding differences in performance. Light and Albertson (1989) might be suggestive of a distinct implicit memory factor, the observed no age differences in performance when a category implicit measures of memory were based on response time generation task was used as the implicit measure, despite and might reflect a processing speed factor. statistically significant age differences on the explicit corol- In the Hultsch et al. (1991) study we examined the lary of category generation, category cued recall. Similarly, patterns of relationships among implicit tests (stem comple- Rabbit (1982, 1984) failed to find age differences in implicit tion), explicit tests (word, text, and fact recall), and a memory when a category judgment task was used. Although number of cognitive ability variables in a sample of young both sets of experiments failed to find age differences in (19-36 years), middle (55-69 years), and older (70-86 performance, they also employed relatively small sample years) adults. The cognitive ability variables included mea- sizes. Therefore, it is unclear whether the results indicate age sures of verbal speed, , verbal ability, and invariance on conceptual implicit tests or reflect relatively reading comprehension. Results indicated that although the low statistical power to detect the small age differences in explicit tests of memory intercorrelated highly, correlations performance. between implicit and explicit tests were not statistically sig- Given the uncertainty regarding the existence of age nificant. Moreover, examination of the correlations of the differences on conceptually based implicit tests of memory, cognitive ability measures with stem completion showed one of the goals of this study was to provide an unusually that, of the 14 variables examined, there was only one sig- powerful test for an age-related effect. In order to increase nificant relationship (a correlation of .17 between stem power, we used a very large sample of subjects from three completion and reading comprehension in the middle-aged age groups. Two implicit tests of memory, one conceptually group). In contrast, multiple significant relationships were based (fact completion) and one perceptually based (word shown between the explicit tests of memory and the various stem completion), as well as two explicit tests of memory cognitive measures (16 significant relationships for those (word and fact recall), were obtained in order to examine age aged 19-36 years; 20 significant relationships for those aged and gender differences in performance. 55-69 years; 26 significant relationships for those aged 70- A second objective was to explore the possible relation- 86 years). Although it appears that stem completion was not ships between implicit and explicit tests of memory. Our related to either the explicit tests of memory or performance view is that understanding age-related differences in mem- on the various ability measures, it must be remembered that ory tasks can be aided by examining the interrelationships this was a perceptually based implicit test of memory. We among memory measures and other indicators of cognitive speculated that a conceptually based implicit test might be ability. Moreover, patterns of interrelationships can provide related to explicit tests and some of the cognitive ability valuable evidence when deciding among competing expla- factors. nations of implicit and explicit test performance. For exam- Our strategy in assessing the relationships among memory ple, if processing requirements affect task performance, then tests was to focus on the correlation between implicit tests of one would expect conceptually based tasks to correlate memory and multiple cognitive abilities as well as other among themselves but perhaps not with perceptually based kinds of tests. As with earlier studies, we tasks. On the other hand, memory systems theorists (Tulving were interested in age-related differences in the pattern of & Schacter, 1990) have suggested that different memory correlations between the word stem and fact completion systems are responsible for implicit and explicit test perfor- tests, and the explicit measures of memory, word, and fact mance. Specifically, conceptual priming is thought to be recall. We expected that if processing requirements dictated dependent on semantic memory, whereas perceptual prim- performance on implicit and explicit tests, then fact comple- ing is subserved by the Perceptual Representation System tion should correlate with both explicit tests of memory, (PRS). One might expect correlations between a conceptu- whereas stem completion would not correlate with any of the ally based implicit test and a measure of semantic memory, other tasks. On the other hand, if as claimed by some but not with a measure of . (Tulving & Schacter, 1990), conceptual priming is mediated by semantic memory, then correlations between fact com- Two recent, small-sample correlational studies have pletion and fact recall should be observed. Finally, it is yielded conflicting results regarding the relationship be- possible that implicit tests of memory depend on processes tween implicit and explicit tests of memory. Using factor PI 64 SMALL ETAL. or aspects of an encoding episode that are different from implicit tests may be seen as a perceptually based measure, those used in either type of explicit test. Thus, we were also whereas the second is more conceptually based (Blaxton, interested in the relationship between performance on the 1989). The explicit tests consisted of single trial recall of implicit and explicit tests of memory and a number of world knowledge items and categorical word lists. The first component processes and abilities. In order to examine this of the explicit tasks may be seen as a measure of semantic question, we included measures of working memory and memory, whereas the latter task is a measure of episodic verbal ability in the battery of tests administered to the memory (Tulving, 1983). subjects. Before describing the memory tests, a brief description must be given regarding counterbalancing of the explicit METHOD tests of memory. Three versions of the explicit memory tests were developed. Random thirds of the sample received one Subjects version of the memory tests. This was done to minimize The sample consisted of 417 community-dwelling adults practice effects for the older adults in the longitudinal study, (158 men and 259 women) residing in a medium-sized as different materials could be rotated across occasions of metropolitan area. The sample was divided into three age testing. For consistency, random thirds of the young sample groups: 17—34 years, 58—73 years, and 74—89 years. The also received different memory materials. 17-34-year-olds were undergraduate students at the Univer- sity of Victoria who participated for optional credit in their Stem completion. — Subjects were given a list of 60 word introductory psychology course. The 58-89-year-olds were stems consisting of three letters followed by a 2 cm line. volunteers participating in a longitudinal study on memory Each stem could be completed into multiple words (M = change (for a detailed description of the longitudinal sample 72.8; Webster's Pocket Dictionary). Subjects were asked to see Hultsch, Hertzog, Small, McDonald-Mizczak, & complete the stem with the first word that came to mind. The Dixon, 1992). Older subjects were paid a nominal fee of stems were presented in random order in a booklet given to $6.00 for their participation in the study. For older adults, the subjects immediately following the . this was their second occasion of testing; the first testing Half of the stems could be completed with letters forming occurred 3 years earlier (for a complete description of the English words previously seen in a lexical decision task. longitudinal changes over this 3-year period see Hultsch et Half of the subjects had been exposed to list A and the other al., 1992). Complete data on the measures of relevance to half had been exposed to list B. The lexical decision task was the present analysis were available on 403 of the subjects based on procedures used by Baddeley, Logie, and Nimmo- (151 men and 252 women), and it is these data that are Smith (1985). Subjects were shown a string of 5 to 7 letters reported here. The age, educational, and self-rated health on the computer screen and asked to decide as rapidly as characteristics of the sample are summarized in Table 1. possible whether the letters formed an English word (e.g., Examination of the educational level of the participants island) or a nonsense word (e.g., nabion). A total of 60 showed that men had significantly more years of education items were presented. than women, F(l,392) = 6.61, p < .05. Further, there were For each subject, the items that were previously presented significant age differences in self-rated health, F(2,392) = were defined as target items, whereas the items that were not 6.29, p < .01, with younger adults reporting better health previously presented were defined as baseline items. A than either of the two older groups, who did not differ from subject's implicit memory score was computed as the differ- one another. ence between the proportion of target items given as stem completions and the proportion of baseline items given as Measurement of Memory Performance stem completions. The difference in these two proportions Memory performance was measured by two implicit and reflects the influence of memory for a recent experience with two explicit tests of memory. The implicit tests consisted of a word on subsequent stem completion. a stem completion and a fact completion test. The first of the Fact completion. — In the fact completion task (Blaxton, 1989), subjects are presented with 40 relatively difficult factual knowledge questions (e.g., What small vessel sup- Table 1. Demographic Characteristics of the Sample plies oxygen and nutrients to cell tissue?). The questions 17-34 Yrs 58-73 Yrs 74-89 Yrs were presented in a booklet and subjects were asked to write Women Men Women Men Women Men their answer on the blank line printed below the question. Variable n = 59 n = 23 n = 120 n = 77 n = 73 n = 51 The task was self-paced. Half of the questions could be completed with a word presented earlier in the context of a Age M 20.64 20.83 67.73 68.38 77.93 77.37 task in which subjects were asked to judge whether a set of SD 4.19 2.25 3.24 3.50 3.78 3.32 letters was an English word. A total of 40 sets of letters were Years of M 14.05 14.44 13.66 14.86 13.31 14.32 presented in a booklet, and subjects were asked to check one education SD 1.51 1.44 2.90 3.17 2.81 4.37 of two boxes: "real word" or "not a real word." Two lists Self-rated M .63 .48 .76 .94 .93 .94 of 20 English words were combined with 20 nonsense words health" SD .58 .67 .68 .78 .75 .79 and presented to random halves of the sample. There was a "Compared to a perfect state of health on 5-point Likert scale (from 0 = short (15-min) delay between the real word and fact comple- very good to 4 = very poor). tion tasks. For younger adults, the delay was filled by the AGE DIFFERENCES IN IMPLICIT MEMORY P165 administration of the sentence construction task (described were required to keep this target word in memory while below). For older adults, the delay was occupied by a continuing to read subsequent sentences. At the end of the questionnaire (not examined in the present series, the subject was to report the target words in order. For study). each set of sentences, the target words formed a new sen- For each subject, words that had been previously seen tence. Sets of 3, 4, 5, and 6 sentences yielding new sen- were defined as target items, and words that had not been tences 3, 4, 5, and 6 words in length were used. There were previously seen were defined as baseline items. The implicit three target sentences at each length for a total of 12 items. memory score was computed as the difference between the The verbal working memory score consisted of the number proportion of target items given as fact completions and the of newly formed sentences correctly reported. proportion of baseline items given as fact completions. Verbal ability. — This domain was indexed by two tests Word recall. — Three categorized lists of common En- from the Kit of Factor Reference Cognitive Tests (Ekstrom, glish nouns were developed from the Battig and Montague French, Harman, & Dermen, 1976). The first test indexed (1969) and Howard (1980) norms. Each list contained 5 verbal fluency. The wordopposites test asks subjects to write words from each of 6 taxonomic categories for a total of 30 as many words as they can that have meanings that are the words per list. In general, high-frequency exemplars ranging opposite or nearly the opposite as a set of four target words in rank from 2 to 9 were chosen, but to reduce guessing, the (calm, wrong, fair, awkward). Subjects were given 5 min to most frequently ranked noun was not used. There was no complete the task. The number of correct items was used as overlap in either categories or exemplars among the three the measure in the present analysis. The second measure was word lists. The categories and exemplars that were chosen an indicator of verbal comprehension. A total of 54 multiple were sufficiently distinct to minimize potential interference choice recognition vocabulary items from several tests were effects within and between lists. The three lists were coun- used. Included were Part 1 of Vocabulary Tests II (V-2), terbalanced across subjects. The lists were presented in Part 2 of Advanced Vocabulary Test I (V-4), and Part 2 of unblocked order in typed booklets for study. Subjects were Advanced Vocabulary Tests II (V-5). Subjects were given instructed they had 2 min to study the words, followed 15 min to complete the task. The number of correct items immediately by a 5-min written recall test. They were was used as the measure in the present analysis. instructed to write down as many of the words as possible in any order. The proportion of words recalled was used as the Experimental Procedures measure in the present analysis. For older adults, tasks were administered during three sessions scheduled over a period of about 1 month. Each Fact recall. — This domain was indexed by recall of session was approximately 2 hours in length with a rest break factual knowledge. A total of 120 general-information ques- in the middle. In the first two sessions, small groups of about tions (e.g., What is the last name of the author of the book 5 to 10 subjects were tested at a time. In the third session, "1984"?) were selected from those developed by Nelson subjects were tested individually. In the first session, sub- and Narens (1980). Three sets of 40 questions were sorted to jects received (in order of administration) the vocabulary, roughly equate the sets in terms of content (e.g., science, word recall, real words, and fact completion tests. In the history, art, sports, geography, entertainment), and dif- second session, subjects received (in order of administra- ficulty as indexed by the Nelson and Narens (1980) norms. tion) the word opposites and fact recall tasks. Finally, in the The questions were presented in booklets, and subjects were third session, subjects completed (in order of administration) asked to write their answer on a blank line printed below the the sentence construction, lexical decision, and stem com- question. The task was self-paced. The three lists were pletion tasks. In addition to the tasks already mentioned, the counterbalanced across subjects. The proportion of correct older adult subject group completed additional questionnaire responses was used as the measure in the present analysis. and computer tasks that were not examined in the present context. A complete description of the test battery completed Cognitive Correlates by the older adults is available in Hultsch et al. (1992). The correlates consisted of 3 measures tapping multiple For the younger age group, tasks were administered in two cognitive ability components. The components were ones sessions, separated by one week. Each session was approxi- that have been found to be important correlates of perfor- mately 1.5 hours long. In both sessions, subjects were tested mance on explicit tests of memory (Hultsch, Hertzog, & individually. In the first session, subjects received (in order Dixon, 1990; Hultsch et al., 1991). Specifically, there was a of administration) the word opposites, word recall, real measure of working memory and two measures of verbal words, sentence construction, and fact completion tasks. In ability. the second session, subjects received (in order of administra- tion) the vocabulary, fact recall, sentence construction, Working memory. — This domain was identified by a lexical decision, and stem completion tasks. Two versions of single task that required the subject to process new informa- the sentence construction task were counterbalanced across tion while holding old information in memory to be reported sessions. This was done because one goal of the study was to later. The verbal sentence construction task (Hultsch et al., establish the reliability of alternate forms of the sentence 1990) required subjects to read aloud a series of sentences construction task. Only one version of the sentence construc- presented on flash cards. One of the words in each sentence tion task was used to derive performance measures for the was printed in capital letters and underlined in red. Subjects present study. P166 SMALL ETAL.

RESULTS Table 2. Means and Standard Deviations (in Parentheses) The data analysis consisted of two parts. The first part of Proportion Correct for Tests of Memory examined differences in performance on the four memory and Cognitive Ability by Age Group tasks as a function of gender and age. The second part 17-34 Yrs 58-73 Yrs 74-89 Yrs assessed relations between the memory tasks themselves and Test n = 82 n = 197 n = 124 between performance on the memory tasks and the three indicators of cognitive ability. Stem completion Target .36 (.10) .39 (.13) .40 (.12) Baseline .21 (.07) .29 (.10) .29 (.10) Cross-Sectional Gender and Age Differences Priming .15 (.10) .11 (.13) .11 (.12) Initial univariate analyses including age, gender, and Fact completion counterbalancing group as factors revealed a significant Target .27 (.15) .39 (.15) .32 (.18) effect related to counterbalancing group for fact memory. Baseline .22 (.13) .33 (.15) .28 (.15) None of the interactions with counterbalancing was signifi- Priming .05 (.12) .06 (.13) .04(.13) cant. Counterbalancing was included in the design in order Fact recall .31 (.12) .51 (.15) .45 (.17) to minimize retest effects for the longitudinal sample and is not of theoretical interest in the present analysis. The inclu- Word recall .69 (.14) .62 (.13) .54 (.16) sion of counterbalancing as a factor in the analysis had no Working memory .77 (.20) .67 (.21) .56 (.23) effect on the pattern of gender and age differences; therefore, Verbal comprehension .56 (.13) .81 (.12) .83 (.13) analyses we report are collapsed across counterbalancing Word opposites .58 (.18) .50 (.18) .44 (.19) condition. In order to examine gender and age differences, a 2 (Gender) x 3 (Age Group) multivariate analysis of vari- ance (MANOVA) with alpha set at .05 was performed on the subjects. It is not clear why older subjects were more likely scores from the four memory tasks. This analysis showed than younger subjects to complete the baseline and target significant effects related to Gender, Wilks \ = .871, items with critical words. However, the difference scores, F(4,394) = 14.53, p < .001, and Age, Wilks \ = .671, which reflect enhancement due to prior exposure to the F(7,788) = 21.72, p < .001. The Gender x Age Group words, reliably favored the youngest group. interaction was not significant. A similar pattern of age differences is seen in the target Examination of the univariate gender effects revealed that and baseline fact completion performance. In this case, subjects in the middle group (58-73 years) were more likely women scored higher on word recall than men (Mf = .624, 2 to successfully complete the target items than either of the Mm= .584),F(1,397) = 4.93,/?<.O5, co = .007,butmen two other age groups [F(2,400) = 17.78, p < .001, GO2 = outperformed women on fact recall (Mf = .410, Mm = .515),F(1,397) = 32.76,/? <.001, a)2 = .071. There were .079]. Similarly, age differences in baseline performance [F(2,400) = 16.84, /? < .001, w2 = .075] indicated the no significant effects for stem completion (Mf = .112, Mm middle age group completed more items correctly than both = .124), or fact completion (Mf = .055, Mm = .049). Because women and men did not differ on either implicit test the youngest and oldest groups. Further, the oldest group of memory, further analyses are collapsed across gender outperformed the youngest. Unlike stem completion, how- groups. ever, the difference scores, reflecting enhancement due to Table 2 shows the average proportion correct on the four prior exposure to the answers, did not reliably favor any age memory tasks by age group. In the stem completion and fact group. completion tasks, memory was defined as the amount of Univariate analyses indicated significant age effects on improvement over baseline performance that accrued as a both explicit tests of memory [word recall: F(2,400) = 2 result of prior exposure to a target word. This improvement 27.02,/? <.001, a) = . 114; fact recall: F(2,400) = 53.95, 2 was reflected in differences scores that were calculated by p < .001, co = .208]. Further analyses using the Neuman- subtracting proportion correct in the baseline condition from Keuls procedure were conducted to examine the nature of the proportion correct in the target condition. The univariate age differences. For word recall, younger adults performed analysis of the difference scores revealed a significant age better than the 58-73-year-olds, who outperformed the eld- effect for stem completion [F(2,400) = 3.48,/? <.05, a>2 = est group. In the case of fact recall, the middle group (58-73 .012], but not for fact completion [F(2,400) = 1.56, p > years) performed better than both the oldest and youngest .05]. A Neuman-Keuls test with alpha set at .05 was per- groups. Further, the oldest group outperformed the youngest formed on the stem completion scores and indicated that the group. youngest group obtained higher scores than either of the two older groups, which did not differ from each other. It Correlational Analyses appears that for both baseline and target items, older adults The correlational analyses were designed to examine, were more likely to complete the stems with critical words within each age group, relationships among the four memory 2 [F(2,400) = 22.89,/? <.001,o> = .100;F(2,400) = 3.43, measures and between each memory measure and the set of 2 p < .05, a) = .012, respectively]. However, older adults' three cognitive variables. We do not report the interrelation- baseline completion rate was far greater than the advantage ships among the cognitive ability variables themselves. they possessed in completing target items. This pattern of Readers who are interested in these relationships should see means resulted in smaller difference scores for the older Hultsch et al. (1990). For the correlational analyses, an AGE DIFFERENCES IN IMPLICIT MEMORY P167

alpha level of .01 was used as the criterion of statistical sig- Table 3. Intercorrelations of All Measures nificance. This relatively stringent criterion was chosen in for the 17-34-Year-Old Group (n = 82) order to reduce the likelihood of significant correlations due Stem Fact Word Fact to Type I error, given the large number being computed (18 Completion Completion Recall Recall in each age group). Although counterbalancing had no effect on the pattern of mean level results, we were concerned Stem completion — Fact completion — about its possible influence on the correlation coefficients. -.009 Word recall -.300* -.183 — As a result, the correlation coefficients we report were Fact recall -.074 -.109 .102 — obtained by partialing out variance related to counterbalanc- Sentence construction -.096 .009 .410** .213 ing condition for each of the four memory tasks. Word opposites -.209 -.069 .204 .066 The initial correlational analyses examined the interrela- Verbal comprehension -.008 .013 .285* .625** tionships among the four memory measures. Significant *p< .01 ;**p<. 001. correlations between the implicit tests of memory might indicate that these tests share common variance. Further, significant correlations between the implicit and explicit Table 4. Intercorrelations of All Measures tests of memory could indicate that explicit memory pro- for the 58-73-Year-Old Group (n = 197) cesses were operating on the completion of the word stems and factual knowledge questions. The results are shown in Stem Fact Word Fact Completion Completion Recall Recall the upper portions of Tables 3,4, and 5 for each of the three age groups. Examination of the correlations with stem com- Stem completion — pletion shows a significant negative correlation with word Fact completion -.040 — recall for the young group. This correlation is interesting Word recall .021 .041 — .262** — from the standpoint that if explicit memory processes were Fact recall -.036 .150 Sentence construction .020 .025 .231* .273** acting on stem completion performance, one might expect a Word opposites -.015 .106 .268** .355** significant positive correlation between the two measures. It Verbal comprehension -.014 .098 .371** .590** is unclear exactly what this negative correlation indicates. None of the other correlations with stem completion was sig- *p<. 01 ;**/><. 001. nificant. For fact completion, none of the correlations with stem completion, or the explicit tests of memory were sig- Table 5. Intercorrelations of All Measures nificant. As expected, significant correlations were observed for the 74-89-Year-Old Group (n = 124) among the two explicit tests of memory. The remainder of Tables 3,4, and 5 report the correlations Stem Fact Word Fact between the implicit and explicit tests of memory, and the Completion Completion Recall Recall measures of working memory and verbal ability. Of the nine Stem completion — possible relationships, there was only one significant rela- Fact completion .092 — tionship with fact completion (a correlation of .235 with Word recall .050 .099 — word opposites in the 74-89-year-old group), and no signifi- Fact recall .089 .156 .525** — cant relationships with stem completion. This stands in Sentence construction .143 .165 .523** .485** contrast to the multiple significant relationships for both Word opposites .063 .235* .464** .542** Verbal comprehension .080 .244* .553** word and fact recall (8 for word recall, and 7 for fact recall). .215 One possible explanation for the lack of significant corre- *p<.01 ;**/?<.001. lations with the stem and fact completion tasks is that these measures are unreliable. We have previously addressed this issue with stem completion (Hultsch et al., 1991) and found upper bound of the correlation is .84. Because the magnitude the task to be acceptably reliable (average reliability = .63). of these upper bound estimates far exceeds the level of For the fact completion task, we correlated randomly se- correlation needed for significance with our sample size (.23 lected halves of the fact completion task. The proportion of to .25 depending on the age group), the stem and fact target questions correctly answered from each of the random completion tasks were judged to be acceptably reliable. halves, minus the baseline measure for the entire list of words, was used to calculate the correlations. The subsam- DISCUSSION ple used to examine the reliability of the fact completion task The results of the analysis provide evidence for age consisted of a random sample of 50 younger subjects, and 50 differences on one implicit test of memory (stem comple- older subjects chosen from the present sample. For all tion), but not another (fact completion). Younger adults' subjects, the correlation of the two halves was .75 (p < stem completion performance was more strongly influenced .001). Computing separately by age group resulted in a by previously seen words than older adults' performance. correlation of .79 (p < .001) for the younger group and .68 (p On the other hand, no age-related advantage was present for < .001) for the older subjects. The square root of these previously seen material in the fact completion test. The correlations constitutes an upper bound on the correlation finding of a significant age difference on stem completion is between the stem and fact completion task and any other consistent with a number of previous studies (Chiarello & possible measure. For example, given a reliability of .70, the Hoyer, 1988; Hultsch et al., 1991). Similarly, the lack of P168 SMALL ETAL. age differences on the fact completion test is also consistent relatively confident that age differences in stem completion with previous research on conceptually driven implicit tests cannot simply be attributed to contamination by consciously of memory (Light & Albertson, 1989; Rabbit, 1982, 1984). controlled retrieval processes. In this case, the age invariance of fact completion cannot be Although we find no evidence of age differences in con- attributed to a lack of power to detect age differences in ceptual priming, the levels of priming are substantially performance, because the power of the significance test was smaller for the fact completion test compared to the stem estimated as .77, assuming a similar effect size as stem completion task. Although smaller in magnitude, the size of completion. The dissociation among implicit tests of mem- priming effects in fact completion is somewhat consistent ory could be a function of two things: (a) Each task has with previous findings (Blaxton, 1989). Blaxton observed different processing requirements which are differentially marginally greater levels of priming in her experiment; sensitive to age differences in performance, and (b) the stem however, the methodology was different because subjects completion task may have been contaminated by episodic were warned of an upcoming memory test. This warning memory processes. could have increased the likelihood of conscious contamina- We find no evidence suggesting that the stem completion tion, adding to the slightly greater priming she observed. task was contaminated by explicit retrieval processes. In the The relatively low priming effects may reflect the shallow correlational analyses, the only significant relationship be- processing of the study items. Blaxton (1989) found that tween stem completion and an explicit test of memory was priming on the fact completion test can be increased by for word recall, and this relationship was negative. To the making the study task more semantic in nature. In the extent that performance on the implicit test was contami- present experiment, the real word judgment test involves nated by retrieval processes intentionally directed at the relatively shallow processing and may result in the relatively items in the recently experienced list, one might expect sig- small magnitude of observed priming in fact completion. nificant positive relationships between the items. The observed dissociation among implicit tests is consist- Similarly, if the stem completion test was contaminated, ent with Tulving and Schacter's (1990) predictions from one might expect the age differences found on the task to be their PRS. They argue that different forms of priming, slightly different. Specifically, based on the age differences conceptual or perceptual, are subserved by different memory in word recall, we might expect to find age differences systems. Specifically, perceptual priming, like the stem between the two older groups in stem completion. This is completion task, is largely determined by the physical prop- based on the assumption that if superior conscious recollec- erties of test cues, and successful completion of the word tion is the source of the age differences in stem completion, stem involves relating the stimuli to information stored in then given that the 58-73-year-old group recalled signifi- PRS. On the other hand, conceptual priming, in the fact cantly more words than the 74-89-year-old group, they completion test, does not rely upon the PRS for successful should also perform better on the stem completion task. This completion. Instead, conceptual priming reflects a process was not observed. of semantic learning: "the modification of, or adding of new Finally, the words were originally presented to subjects in information to semantic memory" (Tulving & Schacter, the context of a lexical decision task. In this context, the time 1990, p. 304). However, we did observe age differences in available for processing the semantic meaning of the mate- performance on our explicit test of memory, fact recall. This rial is very limited, and the subject is not in any way led to dissociation between implicit and explicit semantic tests is expect that later recall or recognition of the items will be unusual if both tasks are assumed to measure performance requested. Recently, Toth, Reingold, and Jacoby (1994) from the same memory system. One might expect similar demonstrated that when words for implicit tests of memory patterns of results if the memory tasks are both dependent on are studied semantically, they are more susceptible to con- the performance of the semantic memory system. scious contamination. They compared estimates of auto- The results of the correlational analyses also provide matic influences from stem completion and from their pro- evidence for a dissociation of performance between the cess dissociation procedure, which allows the influence of implicit fact completion test and the explicit fact recall task. conscious and automatic influences to be estimated sepa- Specifically, these tasks do not correlate with each other, nor rately. When study words were processed semantically, do they share a similar pattern of correlations among the either generated from a sentence or judging the pleasantness cognitive ability variables. Fact recall correlated signifi- of the word, there were clear differences between the two cantly with several cognitive ability variables across the estimates of automatic influence, indicating conscious con- three age groups, whereas fact completion correlated signifi- tamination. On the other hand, when study words were cantly with only one ability measure (a correlation of .235 processed nonsemantically, either read or processed in a with word opposites for the oldest age group). Further, the vowel comparison task, the estimates from the implicit test fact completion test did not correlate significantly with stem and the process dissociation estimates did not differ from completion either. This suggests that performance on fact each other (.24 vs .19; .15 vs .16, respectively). We agree completion is dissimilar from performance on a task pur- with Jacoby and associates (Jacoby, 1991; Toth et al., 1994) ported to utilize the same memory system (fact recall), and a when they state that it is unlikely that either implicit tests or task that shares implicit test instructions (stem completion). explicit tests of memory are necessarily process pure. How- In contrast to the implicit tests of memory, the explicit tests ever, we also believe that methodological variations make it of word and fact recall show significant correlations among more or less likely that an implicit test of memory will be each other, as well as sharing similar patterns of correlations contaminated by conscious processes. Therefore, we are with the cognitive ability variables. AGE DIFFERENCES IN IMPLICIT MEMORY P169

There are two potential explanations for the pattern of Address correspondence to Dr. Brent J. Small, Department of Psychol- observed results. First, performance on implicit tests of ogy, University of Victoria, MS7525, P.O. Box 3050, Victoria, BC, memory reflects processes that are fundamentally different Canada, V8W 3P5. Electronic mail may be sent to [email protected]. from processes involved in explicit tests of memory. Evi- After September 1, 1995 address correspondence to Dr. Brent J. Small, dence for this is gained from the patterns of age differences Section of Psychology, Stockholm Gerontology Research Center, Dalaga- tan 9-11, S-l 13 82 Stockholm, Sweden. and correlations for the implicit and explicit tests of mem- ory. For example, even though significant age differences REFERENCES were observed in stem completion, the size of this effect was Baddeley, A., Logie, R., & Nimmo-Smith, I. (1985). Components of smaller than those age effects in explicit tests of memory. fluent reading. Journal of Memory and , 24, 119-131. Specifically, for stem completion the age effect accounts for Battig, W. F., & Montague, W. E. (1969). Category norms for verbal items roughly 1.5% of variance, whereas the age effects for word in 56 categories: A replication and extension of the Connecticut Cate- gory Norms. Journal of Experimental Psychology Monograph, 80, (3) recall and fact recall account for 11% and 21% of variance, (Part 2). respectively. Hence, even though age differences are found Blaxton, T. A. (1989). Investigating dissociations among memory mea- on all three measures of memory, they differ substantially in sures: Support for transfer appropriate processing framework. Journal terms of amount of variance accounted for. Additional of Experimental Psychology: Learning, Memory, & Cognition, 15, evidence for the implicit/explicit dissociation is found in the 657-668. Chiarello, C., & Hoyer, W. J. (1988). Adult age differences in implicit and pattern of correlations observed among the memory and explicit memory: Time course and encoding effects. Psychology and cognitive ability measures. In contrast to the explicit tests, Aging, 3, 358-366. which correlate with virtually every cognitive ability vari- Cohen, N. J., & Squire, L. R. (1980). Preserved learning and retention of able, the implicit tests only correlate with one. This strik- pattern analyzing skill in amnesics: Dissociation of knowing how and knowing that. Science, 210, 207-210. ingly dissimilar pattern suggests that the processes involved Ekstrom, R. B., French, J. W., Harman, H. H., & Dermen, D. (1976). in implicit and explicit tests of memory may be different Manual for kit of factor referenced cognitive tests. Princeton, NJ: from one another. Educational Testing Service. The second possibility deals with our earlier prediction of Graf, P., & Mandler, G. (1984). Activation makes words more accessible performance on the fact completion test. We had predicted but not necessarily more retrievable. Journal of Verbal Learning and Verbal Behavior, 23, 553-568. that performance on the conceptually driven tests, such as Graf, P., & Masson, M. E. J. (1993). Implicit memory: New directions fact completion and fact recall, should correlate significantly in cognition, development, and neuropsychology. Hillsdale, NJ: with each other, as well as with the cognitive ability vari- Lawrence Erlbaum Associates. ables, but failed to find evidence for this hypothesis. One Hashtroudi, S., Ferguson, S. A., Rappold, V. A., & Crosniak, L. D. (1988). Data-driven and conceptually driven processes in partial word reason for this failure could be that the processes subserving identification and recognition. Journal of Experimental Psychology: the tasks are in fact different. Alternatively, performance on Learning, Memory, & Cognition, 14, 749—757. the fact completion test may not have shared similar variance Howard, D. V. (1980). Category norms: A comparison of the Battig and with fact recall, because our study test was not sufficiently ' Montague (1969) norms with the responses of adults between the ages elaborative enough to engage conceptual processes of mem- of 20 and 80. Journal of Gerontology, 35, 225-231. Howard, D. V., Heisey, J. G., & Shaw, R. J. (1986). Aging and the ory. Had deeper, more elaborative study processing been priming of newly learned associations. Developmental Psychology, 22, implemented, perhaps a different pattern of correlations and 78-85. age differences would have been observed. Hultsch, D. F., Hertzog, C, & Dixon, R. A. (1990). Ability correlates of In sum, the present results suggest reliable age differences memory performance in adulthood and aging. Psychology and Aging, 5, for one implicit test of memory (stem completion) but not 356-368. Hultsch, D. F., Hertzog, C, Small, B. J., McDonald-Miszczak, L., & another (fact completion). The lack of age differences in fact Dixon, R. A. (1992). Short-term longitudinal change in cognitive completion does not reflect inadequate power to detect performance in later life. Psychology and Aging, 7, 571-584. differences but may reflect different processes that are re- Hultsch, D. F., Masson, M. E. J., & Small, B. J. (1991). Adult age quired by either perceptually based or conceptually based differences in direct and indirect tests of memory. Journal of Gerontol- ogy: Psychological Sciences, 46, P22-P30. implicit tests of memory. Further, the correlational analyses Jacoby, L. L. (1991). A process dissociation framework: Separating auto- indicate clearly that the component processes involved in matic from intentional uses of memory. Journal of Memory and Lan- stem completion and fact completion are not similar to each guage, 30, 513-541. other, as well as being different from those required by Jacoby, L. L., & Dallas, M. (1981). On the relationship between autobio- explicit tests of memory. A next step is to determine whether graphical memory and perceptual learning. Journal of Experimental Psychology: General, 3, 306-340. this pattern of results will be seen when more semantic Lewandosky, S., Dunn, J. C, & Kirsner, K. (1989). Implicit memory: encoding tasks are used with the fact completion test. A Theoretical issues. Hillsdale, NJ: Lawrence Erlbaum Associates. related step is to examine performance on multiple conceptu- Light, L. L., & Albertson, S. A. (1989). Direct and indirect tests of ally and perceptually based implicit tests of memory. Per- memory for category exemplars in young and older adults. Psychology haps statistically significant correlations will be observed for and Aging, 4, 487-492. Light, L. L., & La Voie, D. (1993). Direct and indirect measures of implicit tests that share similar processing requirements, memory in old age. In P. Graf & M. E. J. Masson (Eds.), Implicit suggesting distinct domains of implicit tests of memory. memory: New directions in cognition, development, and neuropsychol- ogy (pp. 207-230). Hillsdale, NJ: Lawrence Erlbaum Associates. Light, L. L., & Singh, A. (1987). Implicit and explicit memory in young and older adults. Journal of Experimental Psychology: Learning, Mem- ACKNOWLEDGMENTS ory, & Cognition, 13, 531-541. This study was supported by grants from the Natural Sciences and Light, L. L., Singh, A., & Capps, J. L. (1986). Dissociation of memory Engineering Research Council of Canada to D. Hultsch and M. Masson and and awareness in young and older adults. Journal of Clinical and a grant from the National Institute on Aging to D. Hultsch. Experimental Neuropsychology, 8, 62-1 A. P170 SMALL ETAL.

Mitchell, D. B. (1989). How many memory systems are there? Evidence Schacter, D. L. (1987). Implicit memory: History and current status. from aging. Journal of Experimental Psychology: Learning, Memory, Journal of Experimental Psychology: Learning, Memory and Cogni- & Cognition, 15, 31-49. tion, 73,501-518. Nelson, T. O., & Narens, L. (1980). Norms of 300 general-information Shimamura, A. P. (1986). Priming effects in : Evidence for a questions: Accuracy of recall, latency of recall, and feeling-of-knowing dissociable memory function. Quarterly Journal of Experimental Psy- ratings. Journal of Verbal Learning and Verbal Behavior, 19, chology, 38A, 619-644. 338-368. Toth, J. P., Reingold, E. M., & Jacoby, L. L. (1994). Toward a redefini- Perruchet, P., & Baveux, P. (1989). Correlational analyses of explicit and tion of implicit memory: Process dissociations following elaborative implicit memory performance. Memory & Cognition, 17, 77-86. processing and self-generation. Journal of Experimental Psychology: Rabbit, P. M. A. (1982). How do old people know what to do next? In Learning, Memory, and Cognition, 20, 290-303. F. I. M. Craik & S. Trehub (Eds.), Aging and cognitive processes (pp. Tulving, E. (1983). Elements of episodic memory. New York: Oxford 79-98). New York: Plenum Press. University Press. Rabbit, P. M. A. (1984). How old people prepare themselves for events Tulving, E., & Schacter, D. L. (1990). Priming and human memory which they expect. In H. Bouma & D. G. Bouwhuis (Eds.), systems. Science, 247, 301-306. and performance X: Control of language processes (pp. 515-527). Witherspoon, D., & Moscovitch, M. (1989). Stochastic independence Hillsdale, NJ: Lawrence Erlbaum Associates. between two implicit memory tasks. Journal of Experimental Psychol- Rose, T. L., Yesavage, J. A., Hill, R. D., & Bower, G. H. (1986). Priming ogy: Learning, Memory, and Cognition, 15, 22-30. effects and recognition memory in young and elderly adults. Experi- mental Aging Research, 12, 31-37. Russo, R., & Parkin, A. J. (1993). Age differences in implicit memory: Received October 6, 1994 More apparent than real. Memory & Cognition, 21, 73-80. Accepted December 19, 1994