| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Department of Social Sciences, Gainesville State College, Oconee Campus, Watkinsville, Georgia.
2 School of Psychology, Georgia Institute of Technology, Atlanta.
Address correspondence to Nina Lamson, Gainesville State College – Oconee Campus, Watkinsville, GA, 30677. E-mail: nlamson{at}gsc.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Aging • Episodic memory • Memory strategies
Typically there is more than one path to solve a problem. The path one takes (through conscious decision, habit, or the demands of the task) is one's strategy. A particular strategy ultimately may be more efficient (i.e., faster and more accurate) than another but perhaps more cognitively demanding to implement or initially less reliable. Our primary goal in the present research was to investigate age-related differences in strategy selection across trials to better understand the strategy-selection process and how it changes as participants learn a task.
Previous research has shown that older adults do in fact select different strategies relative to younger adults for completing a task. For example, in a task in which participants have to compare a word pair to a table of word pairs (referred to as a noun-pair lookup task), older adults were less likely than younger adults to use the more optimal retrieval strategy. However, when prior practice or interim tests were provided, age-related differences in the recognition of noun pairs were eliminated (Rogers & Gilbert, 1997
). These data illustrate the importance of task parameters in strategy use; when older adults were periodically tested, they were more likely to adopt a retrieval strategy—perhaps because the tests increased their learning (Roediger & Karpicke, 2006). Similarly, older adults were more likely to shift strategies in a task with high visual scanning demands and low memory demands (and thus affording a memory-retrieval strategy) than they were in a task with high scanning demands and high memory demands (Touron & Hertzog, 2004a).
Why do some older adults prefer to use a strategy that takes more time to carry out? Rogers, Hertzog, and Fisk (2000) found that cognitive abilities did mediate the adoption of a memory-retrieval strategy. Compared with older adults who continued to search noun pairs, those older adults who switched to memory retrieval had better scores on associative memory, reasoning, perceptual speed, semantic knowledge, semantic memory access, and memory span. However, Touron and Hertzog (2004a) suggested that some older adults were seemingly reluctant to shift strategies, even though their memory performance suggested they had learned the word pairs. These researchers found that memory beliefs were significantly predictive of the strategy shift, rather than actual learning (see also Touron & Hertzog, 2004b). Similarly, Lachman and Andreoletti (2006) found that memory-control beliefs influenced strategy choices, which in turn influenced performance. Thus, in addition to ability differences, individual differences in metamemory such as self-confidence and beliefs appear to influence strategy use.
In contrast, Reder and Schunn (1996) suggested that strategy selection may not be an explicit process. That is, there may not always be conscious awareness of the strategy-selection process. Intrinsic features may implicitly and differentially exert their influence on the selection process. For example, in a mathematical task, these researchers found that a feeling of knowing (a sense of familiarity) influenced participants' retrieval selection but not an actual partial retrieval of the answer (a sense of actually knowing the answer). Thus, the retrieval-activation process may be slowed for older adults who do not switch to memory-based strategies, thus minimizing their feeling of knowing (see Reder and Schunn's Source Activation Confusion model). This may in turn exert its influence on subsequent strategy selections.
We designed the present study to require participants to make an explicit strategy choice before performing the task. Previous studies either collected posttask strategy reports (Touron & Hertzog, 2004a, 2004b) or inferred strategy use from the data patterns (Rogers & Gilbert, 1997; Rogers et al., 2000). We used a task in which participants had to make an overt strategy choice at the start of every trial; they had to perform the trial and then get feedback they could use to determine if their strategy choice was efficient (i.e., both fast and accurate). Having participants explicitly choose their strategy was designed to increase their awareness of their strategy selection and its effectiveness, thereby enabling better calibration of strategy selection to performance.
Our overall goal was to assess initial strategy preferences and how these preferences affected learning (and vice versa, how learning affected strategy preferences). In addition, rather than group participants by strategy choice, we grouped them by performance outcome. We conducted a microanalysis of strategy use across trials to compare participants who were successful in reaching the criterion on the task with those who were not.
Overview of Experiment
We adapted a task developed by Reder and Ritter (1992) wherein participants were presented with a small set of multiplication problems to solve. Initially they would have to use a calculation strategy to derive the answer. However, after many presentations of the problems, the researchers expected the participants to learn the answers to the problems and switch to a retrieval strategy to solve them. A retrieval strategy is viewed as the optimal choice because there are fewer demands on working memory and the process is both quick and accurate (Touron & Hertzog, 2004a, 2004b).
One benefit of using a math-based task is that older adults have well-maintained knowledge of math facts (see Duverne & Lemaire, 2005, for review). Older adults' performance in previous studies of word-pair learning may have been influenced by well-documented differences in associative learning (reviewed in Kausler, 1994). The math task allowed us to replicate and extend our understanding of age-related differences in strategy selection with a task that may be less influenced by associative learning deficits in the sense that learning the answers would be facilitated by the availability of well-learned math facts.
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Stimuli and Materials
We administered the Metamemory in Adulthood Questionnaire (Dixon & Hultsch, 1983), the Advanced Vocabulary Test (Ekstrom, French, Harman, & Dermen, 1976), the Digit Symbol Substitution Test (Wechsler, 1981), the Alphabet Span Test (Craik, 1986, using the absolute span score from LaPointe & Engle, 1990), a modified version of the Educational Testing Service Math Test, versions N1 and N3 (Ekstrom et al.), and the computerized Wisconsin Card Sorting Task (WCST; Loong, 1981) to the participants. We included these variables to enable exploratory analyses of individual differences. The mean data are presented in Table 1<--CO?9-->.
|
|
Apparatus
An experimenter presented the stimuli on an IBM-compatible 486DX personal computer with a 14-in. (35.5 cm) Super VGA monitor. We collected the responses by using a Psychological Software Tools Serial Response and Voice Key Box. The first of the five buttons on the response box was labeled "C" for calculate, the third button "P" for proceed, and the fifth "R" for retrieve. Participants used their left index finger for the calculation key and right index finger for the retrieval key. Either finger could be used for the proceed key. A Radio Shack, Pro-302, unidirectional dynamic microphone was attached to a boom on a floor stand. A split cable leading from the computer to two monitors, one in each cubicle, allowed the experimenter to view the participants' screen and to type in the spoken answers.
Procedure
For younger adults this was a 2-day study, whereas for older adults it was a 3-day study (we wanted to make sure we prevented fatigue among the older adults). On the first day, the experimenter tested the participants in groups of no more than four people. The participants completed demographic and medication forms, took vision tests, were given general instructions, and completed the assessment tests (except for the WCST, which was administered on the last day).
Within a week from Session 1, participants returned individually for the experimental task, which occurred the second day (younger and older adults) and third day (older adults) of the study. Prior to engaging in the acquisition phase of the experimental task, participants completed a computerized choice reaction-time task wherein a "C" or an "R" was presented and they were to respond as quickly as possible, using their index fingers, by pressing the corresponding key on the response box. This task allowed participants to become familiar with the response box. There were 5 practice trials followed by 26 additional trials. We then assessed each participant's voice level to calibrate the microphone (see Lamson, 1998, for details).
For the experimental task, participants first read printed instructions describing the experimental task followed by on-screen trial instructions and illustrated sample screens for each step. The trial sequence is illustrated in Figure 1.
|
The experimenter typed the answer that was spoken, and a feedback screen informed the participant of the strategy selection made or of the fact that a selection not made on time; if the answer was spoken on time; if the answer was correct; and the amount of bonus cash earned. The feedback screen was visible for 3 s, followed by a prompt to discard paper (compliance ensured by the experimenter). Pressing the P button to proceed terminated the "discard paper" screen. On the last screen of the trial, the problem was presented with the correct answer for 3 s for study.
Following the on-screen instructions were 30 practice trials consisting of 2 two-digit by two-digit multiplication problems. For Practice Trials 1 through 5, the participant followed the verbal directions of the experimenter to select the C key, do the calculation, and speak their answer. For Practice Trials 6 through 10, the experimenter again gave verbal instructions but merely told the participant to "push" when it was time to make a selection, leaving the strategy decision to the individual. For Trials 11 and 12, the participant was asked to press the C and R buttons, respectively, upon seeing the problem screen but to remain silent for 24 s and 2.5 s to illustrate the amount of time allotted to calculate or retrieve an answer, respectively. For the remaining practice trials the participant made the strategy choices and practiced the procedure. The experimenter provided feedback and answered questions throughout practice when necessary.
At the start of the acquisition phase of the experimental task, participants were again instructed that they would have to initially calculate the problems to obtain the answers, but their goal was to learn the answer and retrieve it from memory. The participants were given no additional instructions as to how to use the retrieval strategy or why it was considered to be the optimal strategy. For each trial the participants were presented with a randomly chosen problem from the set of four to solve, with the condition that no problem would immediately be repeated. There was a potential total of 360 trials, with a minimum of a 1-minute break every 40 trials.4 The acquisition phase was terminated once the participant reached the criterion of 20 consecutive perfect trials, which we defined as selecting the R key on time and correctly answering the problem on time.
To prevent fatigue for the older adults, we had the second day of testing end after 120 acquisition trials (pilot testing revealed that older adults did not consistently retrieve answers prior to this point). When older adults returned on the third day, the experimenter gave them an abbreviated voice-level assessment and task review, followed by the second acquisition module of 240 trials. Once they either reached the criterion or the 360th trial, they completed the remainder of the study.
Younger participants completed the study on the second day. If at the end of the first 120 trials a younger participant had not reached the criterion, then the experimenter initiated the second acquisition module of 240 trials. Once the criterion or the end of the acquisition phase was reached, the experimenter initiated the post-test phase, which consisted of 20 trials. The same four problems as in the acquisition phase were randomly presented five times each; participants were informed that no strategy selection was required. Participants had 2.5 s to speak the answer, at which point the feedback screen informed them whether the response was within the time limit and whether it was accurate.
Bonus points
We designed a bonus-point system to motivate participants to use a retrieval strategy by learning the correct answers to the problems. Selecting the retrieval strategy and answering correctly within the time limit earned 50 points. Selecting the calculation strategy and responding correctly within the time limits earned only 5 points. One point was earned if the participants either made their strategy selection or gave their response in time (but not both) and were correct (50 bonus points equaled 5 cents). During the post-test phase, participants received 5 cents for correct trials spoken on time. The total bonus cash earned ranged from $1.23 to $9.64 (M = $4.55, SD = $1.99).
Design
In the acquisition phase, the dependent measures were trials to criterion, strategy selection (calculation or retrieval), selection- and reply-response times, and accuracy of the answer. For the post-test phase, the dependent variables were reply-response time and accuracy of the answer. Age was a quasi-independent grouping variable in both phases.
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
2(1, N = 55) = 25.80, p <.001. Older adults who did meet the criterion were slower to do so than younger adults, t(35) = –5.52, p <.001. The younger adult mean trials-to-criterion was 128.11 (SD = 60.93, range = 53–280), whereas that for older adults was 249.90 (SD = 55.65, range = 164–298). Thus, not only did fewer older adults reach the criterion, but those who did required nearly twice as many trials.
Three distinct groups emerged: younger adults who met the criterion (Younger Met), older adults who met the criterion (Older Met), and older adults who did not meet the criterion (Older Not Met). We designed the remaining analyses to help us understand the differences among these three groups in the context of two issues: (a) strategy differences, and (b) other differentiating factors.
Criterion measures impose constraints on analyses because participants do not complete the same number of trials. Therefore, we grouped each participant's total trial set into 20th-percentile performance blocks, providing 5 points of comparison. Thus a participant who met the criterion in 200 trials would have Trials 1 through 40 grouped for the first point, Trials 41 through 80 grouped for the second point, and so on. This method is referred to as Vincentizing (after Vincent, 1912, cited in Addis & Kahana, 2004) and is a means of making a comparison of learning patterns across participants in a trials-to-criterion paradigm. The assumption is that "early" trials are logically comparable for all participants, as are "middle" and "later" trials. We used these performance blocks to compare across groups in the subsequent analyses.
Strategy Selection
The first decision that participants were required to make on every trial was to perform that trial by calculation or by retrieving the answer from memory. Overall, 72% of the younger adult selections were retrievals. For the older adults who met the criterion, 60% of their selections were retrievals compared with only 32% for those older adults who did not meet the criterion. Figure 2 shows the proportion of retrieval selections for each criterion group across the five performance blocks (the proportion of calculation selections being the inverse).
|
Time to Select Either Retrieval or Calculation
The time that participants took to decide whether to calculate or retrieve may provide insight into their strategy (Figure 3). If selection time did not differ between the C key and the R key and it was relatively quick, then it might indicate that a decision was made in advance of the trial as to what strategy was going to be used. For younger adults, selection times were faster for the R key relative to the C key for Performance Blocks 2 through 5 (all ps <.05). For older adults who met the criterion there was a similar diversion of selection times, with faster retrieval selections for Performance Blocks 3 and 4 (ps <.05; note that there were no calculation trials for the fifth block, so a comparison was not made). Thus, for the groups who met the criterion, the pattern was similar: There were fewer calculations performed on the later blocks and the decision to calculate was slow, perhaps indicating uncertainty for some of the items.
|
One general question was whether or not the Older Not Met participants were just very slow and could not select the C or R key within 900 ms, as required to meet the criterion. Our analysis of the choice reaction-time task (given prior to the acquisition trials) revealed that all three criterion groups had mean reaction times less than 900 ms, although the Older Not Met group was significantly slower than those who met the criterion F(2, 52) = 17.57, MSE = 19362.45, p <.001 (see Table 1<--CO?4--> for the means).
Strategy Effectiveness
The participants who successfully met the criterion (younger and older) were more likely to choose to retrieval and made this selection more quickly than participants who were not able to meet the criterion. Why? Were individuals who did not reach the criterion unable to use a retrieval strategy? Perhaps those participants were selecting calculation because they knew that was the only way they could produce the correct answer. To explore this possibility, we evaluated strategy effectiveness, which we defined as choosing to calculate or retrieve based on a correct assessment of one's capabilities. We conducted a signal-detection analysis (Table 2<--CO?5-->) with d' as a choice sensitivity measure (Macmillan & Creelman, 1991). High d' values indicate the degree of sensitivity, interpreted here as effective strategy choice. To calculate d', we categorized strategy choices into hits (selected the R key on time and correctly replied in time), misses (selected the C key, but responded quickly enough and correctly that it could have been a retrieval), false alarms (selected the R key but answered incorrectly or not in time), and correct rejections (selected the C key and did not answer in less than 2,500 ms or answered incorrectly).
|
Follow-up analyses revealed the following pattern: For Blocks 1 and 5 there were no group differences; for Block 2, Younger Met < Older Not Met (p <.05); for Blocks 3 and 4, Younger Met and Older Met < Older Not Met (all ps <.05). Thus, the high sensitivity scores of the Older Not Met group were driven by their high rates of correct rejections (i.e., select calculate and require more than 2.5 s to produce the correct answer). These participants were more prone to select and succeed with the calculation strategy.
We also calculated the response-bias statistic, or c (Table 2<--CO?6-->). Negative values indicate a bias toward responding yes, which in this case is selecting retrieve; positive values indicate a bias toward selecting calculate. We calculated a correction measure for the hit and false-alarm rates to avoid the possibility of infinite values often associated with proportions of 0 and 1. We added a value of 0.5 to all the hits, misses, false alarms, and correct rejections (Snodgrass & Corwin, 1988).
Younger and older adults who met the criterion had lower values of c (and thus a bias to select retrieval as a strategy). In contrast, older adults who did not meet the criterion had higher values of c (a bias to choose to calculate). There were significant main effects of criterion group, F(2, 52) = 8.95, MSE = 0.40, p <.001, performance block, F(4, 208) = 71.25, MSE = 0.32, p <.001, and their interaction, F(8, 208) = 4.11, MSE = 0.32, p <.001. Post hoc analyses revealed the following: Block 1, no differences; Block 2, Younger Met < Older Met (p =.05) and Younger Met < Older Not Met (p <.01); Block 3, Younger Met < Older Not Met (p <.01); Blocks 4 and 5, Younger Met < Older Not Met (p <.05) and Older Met < Older Not Met (p <.01).
In all cases in which the differences were significant, younger and older participants who met the criterion had more negative estimates of c, indicating they had a bias to select the retrieval strategy. The differences increased across performance blocks. For the older adults who did not meet the criterion, c was only slightly negative even at the end of practice, indicating only a weak bias to choose retrieval, even after 360 trials on a task that involved learning the answers to four problems.
Post-Test Data
We included these trials to determine if participants had learned the answers to the problems. Participants who met the criterion correctly answered the post-test problems (see Table 3<--CO?8-->). Those who did not meet the criterion did not appear to have learned the correct answers to the four problems<--CO?7-->. An accuracy analysis revealed a significant main effect of group, F(2, 52) = 23.73, MSE = 0.05, p <.001; post hoc comparisons revealed that the Younger Met and Older Met groups were more accurate than the Older Not Met group (ps <.05). There was no significant group effect for response times on the correct trials.
|
Microanalysis of the Older Not Met Group
Because of the variability in the dependent measures for the group of older adults who did not meet criterion, especially for the later performance blocks and for the post-test measures, we conducted additional analyses on this group to explore these individual differences. There was a very clear subgrouping of participants: Subgroup 1 (n = 7) did not increase their retrieval selections across the blocks (it was roughly 17% throughout), whereas subgroup 2 (n = 11) increased their retrieval selections from 13% at Block 1 to 76% at Block 5. Subgroup 1 had a continued bias to select calculate rather than retrieve, as illustrated by a significant difference in c for these two subgroups for Blocks 4 and 5, t(16) = 2.41, p <.05 and t(16) = 4.10, p <.01. Subgroup 1 did not learn the answers to the problems. Their mean post-test accuracy was 15% (SD = 21) compared with 77% (SD = 20) for Subgroup 2, which was a significant difference, t(16) = –6.22, p <.001.
In sum, one group was consistently biased to choose to calculate and did not know the answers to the problems even after 360 trials of practice. The other group was beginning to learn the answers (evidenced by post-test accuracy) but had not learned them well enough to achieve the criterion. We compared these subgroups on the same individual difference variables found in Table 1, and the only significant differences were that Subgroup 1 had higher scores than did Subgroup 2 for meta-anxiety, t(16) = 2.54, p <.05, and the percentage correct on the WCST, t(16) = 2.38, p <.05; additional differences may have been evident with a larger sample.
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
For the most part, however, participants who met the criterion chose retrieve more often than calculate and were faster to select retrieve than calculate. Our signal-detection analysis also revealed that participants who met the criterion had a bias toward retrieval selection; this strategy choice was not always effective because they sometimes selected to make the retrieval but either did not answer on time or were wrong (both of which we coded as false alarms). However, they did meet the criterion and at the post-test phase they learned the answers. Thus, trying to retrieve the answer from memory was a good strategy for learning.
These data replicate previous findings that some older adults can and do use strategies in a manner that is qualitatively similar to that of younger adults. If the task affords learning and use of a memory-retrieval strategy, then some older adults will adopt such a strategy and ultimately learn the task components as well as do the younger adults. Quantitative differences remain in that the older adults are slower to make the strategy shift (perhaps as a result of slower learning). These findings align with those of Rogers and Gilbert (1997), Rogers and colleagues (2000), and Touron and Hertzog (2004a, 2004b).
The pattern was very different for the older adults who did not meet the criterion, even after 360 trials of practice. These participants were likely to choose calculation—they made more calculate selections throughout practice, their decision to calculate was made quickly, and the signal-detection analysis revealed an overall bias to calculate. This bias was not inappropriate, however, because their strategy effectiveness was the highest of any group; that is, they were choosing to calculate on more trials because they needed the additional time to produce the correct answer. They either could not retrieve it from memory at all or were slow to do so. Perhaps the optimal strategy of learning to retrieve the answers from memory was too cognitively demanding for these participants and it was instead easier to calculate the answer, which involved retrieving well-learned math products. The difference here is learning something new, so as to be able to retrieve it, versus relying on that which has been previously learned. Post-test data suggest the new information was not well learned; mean recall was only 53%.
Microanalyses of the individual task components suggested that the underlying cause for the different behavior of these participants was multifactorial. The older adults who did not meet the criterion had lower abilities, more concern about age-related memory changes, and a propensity to select the calculation strategy that did not change across trials. Additional ability differences may have emerged with a larger sample. This pattern is consistent with evidence that older adults are able to make accurate metacognitive judgments about memory abilities (e.g., Dunlosky & Hertzog, 2000) and metacomprehension judgments of text learning (Dunlosky, Baker, Rawson, & Hertzog, 2006). Moreover, the idea that awareness of (or at least beliefs about) one's own ability limitations can influence memory performance is consistent with the control beliefs–memory performance relationships reported by, among others, Lachman and Andreoletti (2006).
In sum, aggregate accounting of older adult performance can be misleading and hide important nuances of behavior in understanding the underlying mechanisms of cognitive performance. As in previous studies (e.g., Rogers et al., 2000; Touron & Hertzog, 2004a, 2004b), age-related differences between younger and older adults were due to the individual differences within the older adult group. If strategies are comparable, then performance and learning are comparable across age groups. For the subset of older adults who continue to use a different strategy, age-related differences remain. In short, strategic differences do account for age-related differences in memory performance, and the mechanisms for the strategic differences are multiple.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
A pilot study conducted with eight younger adults (age, M = 19.5 years) and six older adults (age, M = 76.0 years) showed that participants in both age groups were capable of making a selection in this time interval. The mean strategy selection time was 391.0 ms (SD = 160.32) for younger adults and 466.3 ms (SD = 231.67) for older adults.
The pilot study also determined that neither age group could accurately mentally calculate the answer to a two-digit math problem in 2.5 s (younger adult, M = 25.1 s, SD = 22.23; older adult, M = 29.03 s, SD = 20.31). Thus, participants could not select the retrieval strategy but then actually calculate the answer in the time allotted. Moreover, 2.5 s (2,500 ms) was sufficient for participants to retrieve an answer from memory, once it had been learned. Participants in the pilot study were required to memorize the answers to 4 two-digit problems and then perform the experimental task. The mean time to retrieve the answers from memory was 944.17 ms (SD = 483.19) for the younger adults, and 1,620.25 ms (SD = 1,075.44) for the older adults. In addition, 24 s was an adequate amount of time for the calculation using pencil and paper. Younger adults required a mean of 9,709.55 ms (SD = 5,970.62) and the older adults a required a mean of 8,981.60 ms (SD = 4447.01) to calculate their answers to the problems.
In pilot testing, three of four older adult participants were able to learn the answers to four multiplication problems in 240 trials. Thus, for the experiment, we provided 360 trials to maximize the number of trials without an additional day of testing.
Decision Editor: Thomas M. Hess
Received for publication November 28, 2006. Accepted for publication December 5, 2007.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
