Working memory and aging: A cross-sectional and longitudinal analysis using a self-ordered pointing task

NAOMI CHAYTOR; MAUREEN SCHMITTER-EDGECOMBE

doi:10.1017/S1355617704104013

Working memory and aging: A cross-sectional and longitudinal analysis using a self-ordered pointing task

Published online by Cambridge University Press: 01 July 2004

NAOMI CHAYTOR and

MAUREEN SCHMITTER-EDGECOMBE

Show author details

NAOMI CHAYTOR: Affiliation:
Department of Psychology, Washington State University, Pullman
MAUREEN SCHMITTER-EDGECOMBE: Affiliation:
Department of Psychology, Washington State University, Pullman

Article contents

Abstract
INTRODUCTION
EXPERIMENT 1: CROSS-SECTIONAL ANALYSIS
EXPERIMENT 2: CROSS-SECTIONAL AND LONGITUDINAL ANALYSES
GENERAL DISCUSSION
ACKNOWLEDGMENTS
References

Rights & Permissions

Abstract

Age-related declines in working memory performance have been associated with deficits in inhibition, strategy use, processing speed, and monitoring. In the current study, cross-sectional and longitudinal methodologies were used to investigate the relative contribution of these components to age-related changes in working memory. In Experiment 1, a sample of 140 younger and 140 older adults completed an abstract design version of the Self-Ordered Pointing Task modeled after Shimamura and Jurica (1994). Experiment 1 revealed that only processing speed and monitoring explained age differences in SOPT performance. Participants in Experiment 2 were 53 older adults who returned 4 years after the initial testing and 53 young adults. A task that assessed the ability to generate and monitor an internal series of responses as compared to an externally imposed series of responses was also administered. Experiment 2 replicated the key findings from Experiment 1 and provided some further evidence for age-related internal monitoring difficulties. Furthermore, the exploratory longitudinal analysis revealed that older age and lower intellectual abilities tended to be associated with poorer performance on the SOPT at Time 2. (JINS, 2004, 10, 489–503.)

Keywords

Aging Working memory Self-Ordered Pointing Test Inhibition Processing speed Monitoring Strategy use

Information

Type: Research Article
Information: Journal of the International Neuropsychological Society , Volume 10 , Issue 4 , July 2004 , pp. 489 - 503

DOI: https://doi.org/10.1017/S1355617704104013 [Opens in a new window]
Copyright: 2004 The International Neuropsychological Society

INTRODUCTION

Older adults often report having to work harder when completing complex everyday tasks, and working memory has been implicated in some of these difficulties (Wilson et al., 1997). For example, when cooking, one must keep the overall goal (recipe) in mind while performing various sub-goals in sequence (preheating the oven, cutting vegetables, etc.). In essence, one must guide and actively monitor one's behavior from an internal representation, or, in other words, use working memory. Baddeley and Hitch (1974) originally described working memory as consisting of three subcomponents: the phonological loop and visuospatial sketchpad, which are responsible for the transient storage of speech-based and visuospatial information respectively; and the central executive, which manipulates and controls the above slave systems. Although the central executive remains the least understood of the three subcomponents (Baddeley & Hitch, 1974), there is general agreement that the central executive is involved in the selection, planning, attentional allocation, and control of various cognitive processes (see Miyake & Shah, 1999). In addition, it is widely accepted that other factors outside of working memory, including slowed processing speed and reduced inhibitory control, can affect the amount and/or type of information available for active monitoring within working memory (see Miller & Cohen, 2001). In the present study, we use both cross-sectional and longitudinal methodologies to further explore the nature of working memory deficits in older adults.

Cross-Sectional Research

The results of cross sectional studies have consistently shown that the working memory performance of older adults is poorer than that of younger adults (Fisk & Warr, 1996; Hultsch et al., 1998; Salthouse, 1994). Within the aging literature, several theories have been postulated to explain the nature of these age-related differences. These theories include deficits in inhibition, processing speed, and strategy use. First, the inhibition theory postulates that older adults have a primary deficit in the ability to inhibit task-irrelevant thoughts (Hasher et al., 1991; Hasher & Zacks, 1988; Lustig et al., 2001; May et al., 1999). This results in the maintenance of irrelevant information in working memory. Related to this position is the hypothesis that proactive interference leads to age-related decline in working memory (Shimamura & Jurica, 1994). Proactive interference occurs when information that was relevant to task performance on previous trials interferes with current task demands. Thus, an older adult's inability to inhibit the information from previous trials (proactive interference) is thought to decrease working memory performance on the current trial. Here, working memory storage is intact, but it is overloaded by a deficient inhibitory system.

Second, deficient strategy use by older adults has been postulated to account for age changes in working memory. Numerous studies indicate that older adults do not spontaneously organize incoming information as often or as well as younger adults (Craik, 1977; Smith, 1996). Daigneault and Braun (1993) compared the performance of older and younger participants who reported spontaneously developing and not developing a task strategy during completion of a working memory task. They found that the older adults failed to benefit from the use of a strategy, suggesting that older adults may be impaired in their exploitation of effective strategies while performing a working memory task.

Third, in its strong form, the processing speed theory proposes that age differences in working memory performance are largely accounted for by differences in processing speed (Salthouse, 1994; Salthouse & Meinz, 1995). This position has been supported by the work of Salthouse and his use of statistical procedures to uncover the amount of age-related variance in working memory tasks that is accounted for by processing speed tasks (see Salthouse, 1994 for a review). In general, Salthouse has typically found that processing speed accounts for a substantial amount of the variance in working memory performance and causes age to become an insignificant predictor.

Longitudinal Research

Despite substantial cross-sectional evidence for age-related deficits in working memory, there is little research on working memory change over time within the same individuals. Although several longitudinal studies have included working memory measures (e.g., Baltimore Longitudinal Study of Aging, Austrian Stroke Prevention Study, and Atherosclerosis Risk Study), these studies have not focused on age-related working memory change in normal older adults. To our knowledge, the Victoria Longitudinal Study (VLS) is the only published longitudinal study of normal older adults that has focused on working memory change across time (Hultsch et al., 1992, 1998). The working memory measures used in the VLS included both verbal and nonverbal working memory span tasks. The second wave of testing was conducted three years after the first wave. These investigators found a significant decline in verbal working memory performance across time, with the oldest cohort (age 71–86) showing more decline than the younger cohort (age 55–70). No significant change over time was found for nonverbal working memory. The potential influence that inhibitory and strategic factors may have played in the noted verbal working memory decline was not fully addressed. These authors did, however, investigate the processing speed hypothesis. Contrary to Salthouse's work (see Salthouse, 1994, for a review), they found that controlling for processing speed did not completely eliminate the longitudinal age-related decline observed in verbal working memory performance (Hultsch et al., 1992).

Measurement of Working Memory

Researchers continue to differ on the best way to measure the construct of working memory. Many of the cross-sectional aging studies employed a few widely used working memory measures, including the computation span and reading span tasks (Salthouse, 1994). These span tasks require the participant to answer questions (or solve arithmetic problems) while also remembering information about the presented stimuli (e.g., for the reading span task, participants are asked to remember the last word in each sentence that is presented; for the computation span task, the last digit in each problem). Although span tasks are commonly used in working memory research, it is important to investigate the construct of working memory using multiple methods.

Within the neuropsychological literature, considerable attention has been paid to decomposing components of global working memory tasks (Shimamura & Jurica, 1994; West et al., 1998; Wiegersma et al., 1990), and to localizing the processes involved in working memory tasks using PET and lesion studies. One such neuropsychological task, which was developed to assess the planning and monitoring aspects of working memory, is the Self-Ordered Pointing Task (SOPT; Petrides & Milner, 1982). In this task, a set of stimuli are arranged in different spatial locations on each of multiple sheets of paper. On each sheet, the participant must point to a stimulus item that was not previously chosen. To prevent the participant from using a strategy that would make the task too easy, selecting the same location on three consecutive sheets is not allowed. Because a given item appears in a different location on each successive sheet, the SOPT requires the participant to actively monitor a sequence of self-generated responses to avoid repeating a response.

The SOPT is particularly well suited to the study of working memory decline in older adults for the following reasons. First, the SOPT is well validated as a relatively selective frontal lobe measure by both lesion and PET studies (Petrides, 1995, 1996; Petrides et al., 1993; Petrides & Milner, 1982), and the prefrontal area has been implicated in working memory ability (Petrides, 1996; Smith & Jonides, 1997). Second, performance on the SOPT is impaired in normal older adults, and in recent years there has been evidence suggesting that the prefrontal areas of the brain may show disproportionate decline with respect to other areas in normal aging (e.g., Martin et al., 1991; Raz, 2000; see West, 1996, for a review). Third, the SOPT allows for the computation of several variables that can be used to investigate the nature of the working memory deficit in older adults. These variables include measures of proactive interference (Shimamura & Jurica, 1994), perseveration and monitoring (West et al., 1998), and strategy exploitation (Daigneault & Braun, 1993).

Previous studies have attributed older adults' poorer performance on the SOPT to one or more of the following factors: proactive interference, poor monitoring, and deficient strategy exploitation. For example, in comparison to younger adults, Shimamura and Jurica (1994) found that older adults in their 60s exhibited impairment on the second but not the first SOPT trial (a trial being one completed administration of the SOPT). Because similar stimuli were used for both trials, the authors hypothesized that the older adults exhibited heightened proactive interference due to difficulties distinguishing responses made during the first trial from those made during the second trial.

Other researchers, however, have found that proactive interference does not appear to be a cause of the working memory impairment in older adults (West et al., 1998). Instead, a study by West et al. suggested that older adults have a problem with monitoring. In their study, these authors separated 1-back errors (where participants pointed to the same design on successive sheets), from 2-back errors (where one other design was pointed to between pointing to the same design again), from forgetting errors (which included the rest of the errors which were not perseverative in nature and which the authors described as a form of monitoring or strategic memory error). They found that older adults had more 1-back and forgetting (strategic monitoring) errors than the younger adults. In addition, West et al. showed that older adults tended to make more errors towards the end of a trial, regardless of the set size for the trial (e.g., more errors were made on the 9th selection of the 10-item set size than on the 9th selection of the 12-item set size). The authors interpreted this result as a failure by older adults to accurately monitor progression through the trial, and suggested that the finding was consistent with the idea that a monitoring deficit may be contributing to older adults' errors on the SOPT.

Support for deficient strategy exploitation comes from the work of Daigneault and Braun (1993). These authors compared older and younger adults who reported using a strategy to complete the SOPT with those who did not report using a strategy. They found that in contrast to younger adults, who benefited from using a strategy, older adults' performance did not improve when they reported using a strategy, suggesting that strategy exploitation was impaired in their sample of older adults.

The current study used both cross-sectional and longitudinal methodologies to investigate working memory performances in older adults, as measured by an abstract design version of the SOPT (see Shimamura & Jurica, 1994). Our primary goal was to better understand the nature of working memory deficits associated with aging on the SOPT. In Experiment 1, age differences in inhibition (proactive interference), strategy use, processing speed, and monitoring were investigated within a single study by comparing the SOPT performances of a sample of 140 older and 140 younger adults. A subset of the older adults from Experiment 1 was then re-tested approximately four years later and compared to a new sample of younger adults. In Experiment 2, we replicated and extended key findings from Experiment 1. In the cross-sectional analyses, we also further investigated whether the older adults' difficulties on the SOPT were specific to generating internal sequences of responses. The goals of the longitudinal analyses were to examine whether nonverbal working memory performance would change over the course of 4 years and, if it did, to identify variables that might be associated with the decline. However, because only 53 older adults from the original sample were retested, the longitudinal analyses should be considered exploratory.

EXPERIMENT 1: CROSS-SECTIONAL ANALYSIS

The purpose of Experiment 1 was to determine whether difficulties with inhibitory control, strategy use, processing speed, and/or monitoring contribute to the working memory deficit observed in older adults.

Methods

Research participants

A sample of 140 older adults (M = 70.65, SD = 7.53, range = 57–93 years) and 140 younger adults (M = 20.58, SD = 2.12, range = 18–28 years) were participants in this study. The older adults were recruited through a mailing to alumni, faculty and staff of Washington State University, or through local volunteer and senior citizen organizations. The data were collected as part of a larger study of cognitive aging in which extensive neuropsychological testing was conducted. The older adults were provided with information about their cognitive performance in exchange for volunteering their time. They were also compensated for parking expenses. The younger adults were recruited through the Washington State University psychology participant pool and received course credit for their participation.

All participants were native speakers of English. They were individually interviewed in order to exclude those with a significant history of substance abuse, brain surgery, cerebrovascular or cardiovascular accident, brain damage sustained earlier from a known cause, severe psychiatric disorder, or serious health problems. Persons who had difficulty seeing the SOPT stimuli (i.e., a Snellen ratio of less than .50 at 40.6 cm) were excluded from the study. The older adults had, on average, completed more years of formal education (M = 16.86, SD = 2.63) than the younger adults [M = 14.40, SD = 1.27; t(280) = 9.98], and had a higher estimated full scale IQ [older: M = 117.93, SD = 11.60; younger: M = 105.79, SD = 10.67; t (272) = 8.99], as measured by a short form of the Wechsler Adult Intelligence Scale, Revised (WAIS–R, Wechsler, 1981; Vocabulary, Block Design, Similarities, and Arithmetic subtests, Sattler, 1988).

Materials and procedure

The following tests were administered as part of a larger battery of tests that took approximately 3 hr to complete.

Self-ordered pointing task. An abstract designs form of the self-ordered pointing task was used as a measure of working memory (see Shimamura & Jurica, 1994). Participants were shown sheets of 22 × 28 cm paper with either 9 designs arranged in a 3 × 3 array or 16 designs arranged in a 4 × 4 array (see Figure 1). There were two different versions (A and B) of both the 9- and 16-item stimuli. The stimuli included in each version were similar in visual quality, that is, black-and-white computer-generated designs. A trial was defined as a series of sheets equal in number to the number of designs printed on the initial sheet (i.e., 9 sheets for 9 designs and 16 sheets for 16 designs). For each trial, the same designs appeared on each successive page but in a different spatial location. Participants were instructed to point to a design on the first sheet, turn the page and point to another design, and so on until all the designs had been pointed to without pointing to a given design more than once. Two trials were administered for each set size, for a total of four trials overall. Progression through the task was self-paced; however, the total time per administration was noted. The examiner recorded the design chosen on each sheet and provided no feedback to the participant. An error occurred each time a participant selected a pattern that had been chosen previously within a trial. Selecting the same location on successive sheets was discouraged if this appeared to be the strategy used (pointing to the same location on three or more sheets in a row), as this would increase performance without using working memory to complete the task. Participants were randomly assigned to one of eight conditions counterbalanced for set size and alternate versions. For example, for four of the eight conditions participants received two trials of set size nine followed by two trials of set size 16, and vice versa for participants in the other four conditions. Within a set size, participants either received Version A twice, Version B twice, A then B, or B then A. Thus half of the participants had the same stimuli on successive trials and half had different stimuli on successive trials within each set size.

Example sheet from the 16-item SOPT.

WAIS–R, Digit Symbol subtest. This task is a measure of visual motor processing speed (Wechsler, 1981). In this task, participants were presented with a key at the top of the page where each of a series of numbers (1–9) is paired with a different symbol. Their task was to copy the symbol that goes with each number in the space below the number, as quickly as possible. Participants were stopped after 90 s and the total number of correct responses was tabulated.

Results

The eight counterbalanced conditions were combined for all analyses, as there were no differences between task versions (A or B) or order of administration of set size (9 or 16) as a function of age group and set size.¹

1In the full factorial model, the three-way interaction of age group, set size, and version (A or B) was not significant, F = .13. The three-way interaction of Age Group × Set Size × Order (nine first or 16 first) was also not significant, F = .07.

Unless otherwise stated, all statistics were evaluated at a p-value of .05. Mixed-model repeated measures analysis of variance (ANOVA) were first used to compare the SOPT performances of the older and younger adults, and to evaluate the effects of proactive interference and strategy use. The influence of processing speed on SOPT performance was then examined with a series of hierarchical multiple regressions. Finally, to evaluate monitoring, mixed model repeated measures ANOVAs on the number of errors that fell within each of four error distance groups (i.e., 1–2, 3–4, 5–6, 7–8) for the nine-item set size and within each of five error distance groups (i.e., 1–3, 4–6, 7–9, 10–12, 13–15) for the 16-item set size were conducted.

Total errors

Errors committed on the SOPT were submitted to a group (older adult vs. younger adult) by set size (9-item and 16-item) ANOVA with repeated measures on the last factor.²

2For all analyses, the univariate and multivariate results were similar; therefore, we only present the univariate results. In addition, since the older adults in our sample had significantly higher estimated full scale IQs than the younger adults, correlational analyses were conducted to evaluate whether SOPT performance was related to IQ. The correlation between total errors on the SOPT and estimated full scale IQ was significant for both the younger (r = −.31, p < .05) and older (r = −.35, p < .05) adults, indicating that individuals with higher IQs also tended to perform better on the SOPT. When IQ was included as a covariate in the analysis, the effect of age group only increased, indicating a greater opportunity to find group differences. Therefore, we chose not to covary for IQ in the remaining analyses.

In order to control for a scaling artifact when comparing errors across set sizes, we divided the total number of errors for each set size by the number of sheets for that set size (i.e., total errors for both trials at set size 9 was divided by 18). Overall, the older adults (M = .20) had a higher proportion of errors on the SOPT than the younger adults [M = .15; F(1,278) = 45.41, MSE = .35] and the proportion of errors was greater for the 16-item set size (M = .19) compared to the nine-item set size [M = .15; F(1,278) = 93.08, MSE = 1.28]. The interaction of Set Size × Group failed to reach significance (F < 1). Therefore, for the remaining analyses, errors were combined across the nine-item and 16-item trials and raw errors were used instead of error proportions. Consistent with previous studies, the above findings indicate that older adults tend to commit more errors on the SOPT than younger adults (Shimamura & Jurica, 1994; Daigneault & Braun, 1993). Furthermore, this does not appear to be simply due to a capacity limitation in the older adults, as the Group × Set Size interaction failed to reach significance.

Proactive interference

Previous research that has investigated the role of proactive interference in SOPT performance has examined performance across trials that use the same stimuli. Poorer performance on the second trial with the same set of stimuli has been used as evidence of proactive interference (Shimamura & Jurica, 1994). In this study, half of the participants received the same stimuli on successive trials within a set size (either AA or BB) and half received different stimuli (either AB or BA). We expected that the potential for proactive interference would be greater when the same stimuli were presented across trials because of the greater opportunity for difficulty distinguishing responses made during the first trial from those made during the second trial. Therefore, if proactive interference contributed to the older adults' higher error rate, we would expect errors for older adults who had the same stimuli across trials to increase on the second trial. To investigate this, errors committed on the SOPT were submitted to a group (older adult vs. younger adult) by condition (same stimuli: AA or BB vs. different stimuli: AB or BA) by trial (first vs. second) ANOVA with repeated measures on the last factor (see Table 1). The “first trial” was composed of the first trial administered for both the nine-item and 16-item set sizes, and the “second trial” by the second trial administered for each set size. This analysis revealed a significant main effect of group [F(1,276) = 49.38, MSE = 230.15], and condition [F(1,276) = 9.21, MSE = 42.90], but not trial (F = 1.41). The interaction of Group × Trial was also not significant (F < 1). No other interaction terms were significant (F < 2.55). Contrary to expectations based on a role for proactive interference in SOPT performance, the main effect of condition indicated that those who received the same stimuli across trials actually made fewer errors than those who received different stimuli. In addition, this effect was the same for the younger and older adults. Thus, this analysis revealed no evidence to suggest that the older adults' difficulties on the SOPT were due to proactive interference.³

3Shimamura and Jurica (1994) reported that older adults in their 60s suffered from proactive interference. Since the analysis we completed included older adults in their 70s and 80s and both similar and different stimuli on successive trials, we re-ran the analysis and compared the older adults in their 60s who had the same stimuli across trials (N = 27) to the younger adults (N = 70) who had the same stimuli across trials. Neither the main effect of trial, F = 2.73, MSE = 5.24, nor the interaction of Age × Trial (F = .02, MSE = .004) were significant. The mean error rate for the older adults in their 60s on the first and second trials was 4.33 (SD = 1.75) and 4.00 (SD = 1.59), respectively.

This finding is inconsistent with some previous research which has found that age differences in verbal working memory span may be due to differences in the ability to overcome interference (e.g., Lustig et al., 2001).

Mean number of errors (and standard deviations) on the Self-Ordered Pointing Test as a function of age group and trial by condition in Experiment 1

Strategy use

Another possible contributor to the older adults' poorer performance on the SOPT is a failure to develop an effective strategy. Based on participants' responses to a question asking how they went about completing the SOPT task, participants were classified as using a strategy (83.9%) if they reported grouping the designs in a specific manner and as not using a strategy (13.6%) if they reported guessing, pointing in a random fashion, or pointing by memory.⁴

4Percentages do not add up to 100% because of missing data.

A chi-square analysis [χ²(1, N = 235) = .35, p > .05] indicated that the age groups did not differ in the frequency of strategy use, with 87% of the older adults and 81% of the younger adults reporting the use of a strategy. Errors were then submitted to a group (older adults vs. younger adults) by strategy use (yes vs. no) ANOVA. As can be seen in Figure 2, there was a significant main effect of strategy use [F(1,269) = 9.10, MSE = 83.25, p = .01] indicating that those who reported using a strategy (M = 8.60, SD = 3.28) made fewer errors than those who did not report using a strategy (M = 9.92, SD = 3.19). Consistent with previous analyses, there was a significant main effect of group [F(1,269) = 26.40, MSE = 241.49]. Most importantly, the strategy use by group interaction term was not significant (F < 1), indicating that the benefits of strategy use were similar for the older and younger adults. Due to the drastically different cell sizes in the above analysis, the assumption of equality of variance was violated. Therefore, a t test assuming unequal variance was conducted on strategy use, which confirmed the main effect of strategy use reported above [t(55.23) = 2.69, p < .01].

Mean number of errors (with standard error bars) on the SOPT as a function of age group and strategy use in Experiment 1. Errors reflect the total number of errors across all trials.

Processing speed

The hypothesis that processing speed can account for the majority of the changes in working memory with age (Salthouse, 1994) was investigated using a series of hierarchical multiple regressions (see Salthouse, 1994 for a review of this approach). Correlations for the variables in the model were as follows: age and Digit Symbol = −.74, age and SOPT = .44, SOPT and Digit Symbol = −.42. First, when total SOPT errors were regressed on age alone, the resulting model yielded a multiple R of .442, thus accounting for 19.5% of the variance in SOPT performance, F = 66.82, p < .05. The addition of Digit Symbol performance significantly increased the proportion of explained variance to 21.5% (F_change = 6.94, p < .05) and attenuated the beta weight for age (from .442–.286; Digit Symbol beta = −.210), although age still remained highly significant. Starting over, when SOPT errors were regressed on Digit Symbol performance alone, the resulting model yielded a multiple R of .422, accounting for 17.8% of the variance in SOPT performance (F = 59.71, p < .05). Adding a term for age significantly increased the proportion of explained variance by 3.7% (F_change = 12.91, p < .05). These findings indicate that processing speed is related to performance on the SOPT. However, even after removing variance associated with processing speed, age effects on SOPT performance continued to explain a significant amount of the variance.

Monitoring

To examine the influence of monitoring difficulties on the older adults' SOPT performance, similar to the strategy used by West et al. (1998), we conducted an analysis involving error distance. Error distance refers to the number of sheets that are presented after a design is selected the first time until the design is selected a second time (the error occurs at this point). Unlike West and colleagues, however, we conceptualized error distance as a continuous rather than a dichotomous variable. More specifically, we did not categorize errors as 1-back, 2-back and forgetting (monitoring), but investigated the entire range of possible error distances as an indicator of monitoring ability. We were not convinced that forgetting (monitoring) and perseverative errors could be dichotomized as West et al. proposed, because shorter error distances could also result from poor monitoring. We believe it is possible that perseverative errors (1-back errors) could reflect a severe monitoring deficit, where the participant is unable to monitor even one design within working memory. Thus, for the purposes of the current study, error distance was used as a measure of monitoring ability on the assumption that increased proximity in errors (i.e., shorter error distances) reflects greater impairment in the self-monitoring aspects of executive control (see also Glosser & Goodglass, 1990).

The error distance analyses were conducted on the 9-item and 16-item set sizes separately. For each participant, we calculated the number of errors that fell within each of four errors distance groups for the nine-item set size (1–2, 3–4, 5–6, 7–8) and each of five error distance groups (1–3, 4–6, 7–9, 10–12, 13–15) for the 16-item set size. Due to the nature of the test, the opportunity for errors at shorter distances is much greater than it is for errors at longer distances. Therefore, to confidently interpret any group differences in the error distance data, we needed to determine whether the distribution of errors within each trial was similar for the older and younger adults. That is, was the baseline probability of committing an error when choosing the second item (no error is possible for the first item), third item, fourth item and so on similar across age groups. If one of the groups committed proportionately more errors on earlier items, then this would likewise decrease the probability of that group making errors at longer error distance ranges since the probability of committing an error at the varying error distance ranges is not independent of the accuracy of prior responses. For the nine-item set size, a Group (young and old) × Item (1st item through 9th item selected) ANOVA with repeated measures on the last factor revealed no significant interaction (F = 1.87), suggesting that the baseline distribution of errors within the nine-item trial was similar for the older and younger adults. The mean number of items selected before the first error occurred also did not differ between the younger adults (M = 6.75, SD = 1.86) and older adults (M = 6.95, SD = 1.72). For the 16-item set size, the ANOVA revealed a significant Group × Item interaction [F(15,4170) = 2.0, p < .05]. Using p < .01 to establish significance, post-hoc analyses revealed that the number of errors committed by the older adults was greater than that of the younger adults for the 13th item, 14th item, and 15th item chosen. In addition, the mean number of items selected before the first error was 8.49 (SD = 2.80) for the older adults and 9.88 (M = 2.98) for the younger adults [t(278) = −4.03, p < .05]. Because group differences in the probability of committing an error occurred only for later items when the opportunity existed to make an error at nearly any error distance range, we proceeded to interpret the raw data from the error distance analysis. We were interested in determining whether the older adults' performance differed from the younger adults' performance in the errors made at some distance ranges but not others (i.e., the interaction of Age × Error Distance).

In terms of the nine-item set size error distance analysis, the group (older adults vs. younger adults) by error distance (1–2, 3–4, 5–6, 7–8) ANOVA with repeated measures on the last factor revealed a significant main effect of error distance [F(1,278) = 304.18, MSE = 150.16]. As expected, due to the nature of the task, the number of errors increased as the error distance decreased for both groups. There was also a significant main effect of age group [F(1,278) = 24.19, MSE = 11.81]. More importantly, there was a significant Error Distance × Age Group interaction [F(1,278) = 16.72, MSE = 8.25]. Post-hoc analyses (corrected for multiple comparisons, p = .01) revealed that when compared to the younger adults (M = .86), the older adults (M = 1.35) made significantly more errors in the 1–2 error distance range [t(278) = 4.27]. No other comparisons reached significance, although the difference between the older (M = .99) and younger (M = .76) adults approached significance for the 3–4 error range (p = .04).

For the 16-item set size, the group (older adults vs. younger adults) by error distance (1–3, 4–6, 7–9, 10–12, 13–15) ANOVA also revealed that the number of errors increased as error distance decreased [F(1,278) = 554.19, MSE = 869.14; see Figure 3]. There was a significant main effect of age group [F(1,278) = 39.10, MSE = 41.49], and a significant Error Distance × Age Group interaction [F(1,278) = 28.86, MSE = 45.26]. Post-hoc analyses (corrected for multiple comparisons, p = .01) revealed that the older adults (M = 3.07) made significantly more errors in the 1–3 error distance range than the younger adults [M = 1.88; t(278) = 5.79]. No other comparisons were significant, although the difference between the older (M = 1.86) and younger (M = 1.51) adults approached significance for the 4–6 error distance range (p = .02). These findings suggest that, relative to the younger adults, when older adults committed errors on the SOPT they selected the same designs within a closer proximity for both the nine-item and 16-item set sizes.

Mean number of errors (with standard error bars) as a function of error distance range for Experiment 1. Errors are based on the number of errors made at each of five error distances summed across the two trials of the 16-item set size.

Experiment 1: Discussion

Consistent with previous studies that have used tasks other than the SOPT to assess working memory (e.g., reading span task; Fisk & Warr, 1996; Hultsch et al., 1998; Salthouse, 1994), we found that the older adults exhibited poorer working memory performance than the younger adults.⁵

5In order to rule out the possibility that the older adults' difficulties on the SOPT were due to deficits in visual pattern recognition or memory dysfunction, 35 older adults (M = 68.98 years old, SD = 7.10) and 35 younger adults (M = 20.29, SD = 1.49) completed two trials with a similar set of 16 stimuli. A recognition test for the 16 designs followed (16 target designs and 16 distracter designs). As expected, the older adults (M = 5.86, SD = 1.80) committed more errors on the SOPT than the younger adults [M = 4.86, SD = 2.03; t(68) = 2.18, p < .05]. In contrast, recognition testing revealed no differences between the older and younger adults in the proportion of hits [.87 and .89, respectively; t(68) = −.94], or the proportion of false alarms [.16 and .14, respectively; t(68) = .94]. These results indicate that age differences on the SOPT cannot be fully accounted for by differential visual pattern recognition or visual memory deficits.

An examination of factors that might be contributing to these age differences on the SOPT revealed the following: First, there was no evidence to suggest that the older adults' poorer performance was related to a greater susceptibility to proactive interference. For both age groups, when the same set of stimuli was used across successive trials, error rates did not increase on the second trial. Second, there were no age differences in the amount of benefit gained from using a strategy, as both young and older adults' performances improved equally with the use of a strategy. However, among those who reported using a strategy, differences remained between the younger and older adults. This finding indicates that the older adults were able to exploit strategies for completing the SOPT, yet they still performed more poorly than the younger adults. Third, although processing speed was important for performance on the SOPT, older adults had difficulties that were not fully explained by differences in this ability. That is, age effects continued to be a significant predictor of SOPT performance even when processing speed was accounted for. Finally, age differences in our measure of monitoring ability (i.e., error distance) were observed on the SOPT. That is, when compared to the younger adults, the older adults made more errors at the shortest error distance, but not at the other error distances.

Overall, these observations suggest that proactive interference and strategy exploitation appear to contribute little towards explaining age differences in errors on the SOPT. In contrast, both slower processing speed and difficulties in monitoring a series of designs in working memory appear important. The authors of the SOPT contend that this task measures the ability to monitor a self-generated sequence of responses (Petrides & Milner, 1982). In order to further examine the nature of the older adults' difficulty with this task, in the second wave of testing, participants completed an additional task (Externally-Ordered Pointing Task) designed to separate the internally ordered component of the SOPT from the ability to keep track of a sequence of responses in an externally ordered situation.

EXPERIMENT 2: CROSS-SECTIONAL AND LONGITUDINAL ANALYSES

The cross-sectional analyses in Experiment 2 were conducted in order to replicate the proactive interference, processing speed, and monitoring analyses of Experiment 1. Strategy use was not analyzed in Experiment 2 because nearly all of the older adults (52 out of 53) in this sample reported using a strategy at Time 2.⁶

6Forty-nine of the 53 returning older adults also reported using a strategy at Time 1.

We also investigated the question of whether older adults' difficulty on the SOPT was specific to following internal sequences of responses through use of an Externally-Ordered Pointing Task (EOPT). The first trial of the EOPT examined ability to follow an internal sequence of responses, as participants were explicitly instructed to use a grouping strategy to complete the trial. The second trial examined ability to follow an external sequence, as participants were required to point to the designs in a grouping order specified by the experimenter. This grouping order was displayed to participants during the entire trial. We hypothesized that compared to the younger adults, the older adults would have difficulty following internal sequences, but not external sequences since following an external sequence does not require working memory. Given the small sample of older adults that were retested in Experiment 2, the longitudinal analyses were considered exploratory. We were primarily interested in whether nonverbal working memory performance would change over the course of 4 years and, if it did, in identifying variables that might be associated with the decline.