Memory Scanning Performance in Healthy Young Adults, Healthy Older Adults, and Individuals with Dementia of the Alzheimer Type*

A memory scanning (Sternberg, 1966, 1975) task was administered to healthy young adults, older adults, and two groups of individuals with dementia of the Alzheimer’s type (DAT) to determine ageand diseaserelated changes in the retrieval of information from short-term memory. Healthy older adults, in comparison to healthy young adults, displayed increases in both slopes and intercepts in memory scanning. Individuals at various stages of DAT (very mild, mild, moderate) displayed increases in both slopes and intercepts compared to nondemented age-matched control individuals. There was also some evidence that DAT individuals are more likely to engage in a self-terminating search instead of an exhaustive search of shortterm memory. One of the critical features of diagnosis of dementia of the Alzheimer type (DAT) is impaired memory performance (see Nebes, 1989, for a review). In addition, there is clear evidence of age-related changes in healthy adults (e.g., Craik, 1991). Of course, it is important to understand what components of memory break down in these populations. Memory researchers have identified a number of distinct memory types. These include: (a) short-term/long-term, (b) implicit/explicit, (c) primary/secondary, and (d) semantic/episodic, among others. The present study focuses primarily on issues related to short-term memory (STM) retrieval operations in healthy aged individuals and individuals with DAT. The memory scanning task (Sternberg, 1966, 1975) was developed to directly address operations involved in STM retrieval. In this task participants receive a short list of items (e.g., numbers, letters) to retain in memory. After storing these items in memory a single probe item is presented. Participants must decide as quickly and as accurately as possible whether or not the probe item was a member of the original memory set. Because the list items are presumably stored in STM, response latency presumably reflects the time taken to retrieve an item from that memory system. The typical results from this paradigm indicate that as memory set size increases, reaction time (RT) to the probe item increases in a parallel fashion for both trials in which the probe is in the memory set (present trials) and trials in which the probe is not in the memory set (absent trials). MEMORY SCANNING PERFORMANCE 261 To account for this pattern of results, Sternberg (1966, 1975) proposed a serial exhaustive scanning model in which the time taken to scan each additional memory set item requires a fixed amount of scanning time (e.g., 38 ms per item). Because this pattern holds for both yes (item present) and no (item absent) responses, Sternberg argued these results support the notion that participants compare the probe item to all items in the memory set prior to making a response. That is, regardless of ‘‘where’’ in the memory set the probe item occurs, participants continue to scan the entire memory set prior to responding. The serial-exhaustive scanning model can be contrasted with the more intuitively appealing serial self-terminating scanning model in which the participant would terminate the search process when a match is found. However, if this were the case, then one would expect that the slope values associated with yes responses would be one half the slope values associated with no responses because, on average, participants could respond based on half the number of comparisons when the item is in the set compared to when it is not in the set. Although the self-terminating model seems intuitively more plausible, the results from this paradigm consistently yield support for the serial-exhaustive model in both young and healthy older adults. According to Sternberg’s (1966) additive-factors logic, both the serial-exhaustive and serial self-terminating models can be envisioned to operate within an information processing system that includes the following discrete stages: (a) encoding (perception and encoding of the probe item), (b) comparison (serial and exhaustive scanning of the contents of STM), (c) decision (a binary decision as to whether or not the probe item was a member of the memory set), and (d) response execution (response selection and execution). There are also cascadic models of information processing and ‘‘discrete’’ is a simplifying assumption. Presumably, the slope reflects cognitive operations occurring within the comparison stage (Stage 2) whereas the intercept reflects the cognitive operations occurring within the remaining three stages (Stages 1, 3, 4). Use of the memory scanning paradigm has several advantages over other paradigms regarding the investigation and identification of specific cognitive operations. First, as mentioned above, inferences based on slopes and intercepts can be used to isolate distinct information processing stages. Second, the resulting information processing model (serialexhaustive) provides a relatively straightforward account of retrieval from STM (see Greene, 1992, for a review). Memory scanning investigations involving healthy young and older adults have revealed qualitatively similar results to those obtained from college-aged young adults. Both age groups produce increases in RT with increases in set size. However, compared to young adults, elderly adults produce increases in both slopes and intercepts (e.g., Wickens, Braune, & Stokes, 1987). More importantly, healthy older adults also appear to engage in serial-exhaustive scanning. There have also been some investigations using the memory scanning paradigm in participants who possess greater slowing in response latencies and also memory deficits. For example, individuals with multiple sclerosis reveal deficits in both slopes and intercepts compared to healthy age-matched controls (Rao, AubinFaubert, & Leo, 1989). Similar slope/intercept deficits have also been observed in individuals with Korsakoff’s amnesia (Naus, Cermak, & DeLuca, 1977). Elderly depressed individuals reveal greater intercepts compared to healthy controls, but are virtually identical to controls in scanning rate (Hart & Kwentus, 1987). Developmentally disabled individuals typically reveal greater slopes than control individuals, although intercept differences tend to vary across studies (see Silverman & Harris, 1982, for a review of this area). There is also evidence from the Parkinson’s disease (PD) literature that in comparison to healthy controls, nondemented PD individuals produce larger intercepts but display no difference in slopes (e.g., Poewe, Berger, Beuke, & Schelosky, 1991). There are only two reports (Boaz & Denney, 1993; deToledo-Morrell et al., 1991) examining memory scanning performance in individuals with DAT. DeToledo-Morrell et al. (1991) were concerned with P-300 amplitude in probable 262 F.R. FERRARO AND D.A. BALOTA DAT and healthy control individuals. In addition to P-300 latency, these authors also recorded RT in a variant of the memory scanning paradigm. The results from this study produced the predicted group by set size interaction, which occurred for both RTs and error rates. Unfortunately, the error rates of the DAT patients surpassed 20% or more. Furthermore, these authors did not perform analyses examining yes and no responses separately. Thus it is difficult to specify the characteristics of the search process (i.e., serial exhaustive vs. self-terminating) from this study. Boaz and Denny (1993) attempted to address the shortcomings observed in the deToledo-Morrell et al. (1991) study. In comparison to healthy young and nondemented elderly controls, the Alzheimer’s Disease (AD) individuals in Boaz and Denny’s study (1993) produced deficits in both scanning rate and intercepts. These authors argued that their AD participants were employing a different memory search strategy than that of the nondemented control participants. Specifically, this strategy involved AD participants first checking if the probe item was the last item of the memory set. If a match occurred, a response was made. If, however, no match occurred, participants would then proceed to scan the entire memory set. Unfortunately the results of this study are somewhat difficult to interpret because some aspects did not replicate results typically obtained with normal healthy elderly adults. That is, their healthy nondemented elderly adults did not show greater slopes and intercepts than did their healthy young adults. As noted, an increase in slopes and intercepts has been reported by a number of investigators (Anders, Fozard, & Lillyquist, 1972; Eriksen, Hamlin, & Daye, 1973; Salthouse & Somberg, 1982; Strayer, Wickens, & Braune, 1987; Wickens et al., 1987). There is one other aspect of this report that bears discussion. Specifically, a rather large memory set size was used (2, 3, 4, or 5 items). Memory span sizes greater than 4 items may be particularly difficult for mildly and moderately demented individuals to process (Storandt, Botwinick, Danziger, Berg, & Hughes, 1984). This observation is particularly relevant because Boaz and Denney (1993) displayed each set size for the same amount of time (2 s). The use of this constant exposure duration for each set size may have encouraged a differential scanning strategy similar to that described above. The present experiment was designed to further address the nature of STM retrieval in DAT patients compared to that in healthy young and older adults. The present study provides four methodological refinements that should provide further evidence regarding STM retrieval operations in DAT. First, the inclusion of two separable stages of DAT progression (very mild and mild DAT) across individuals allowed for a more precise identification of when deficits associated with memory retrieval materialize in the course of the disease. There was no such separation based on dementia severity in previous studies. Second, the present experiment employed memory set sizes (2, 3, and 4) that are clearly within the memory span capabilities of mild and moderately demented individuals (Storandt et al., 1984). Third, we allowed all groups to examine the various set sizes for a time period proportional to the exact set size. Fourth, we included a relatively large sample size to provide stable estimates across groups. We also included healthy young adult participants to insure that we replicate the standard increase in set size and slope that has been observed in the healthy aging literature. As noted above, because the Boaz and Denny (1993) study did not replicate the standard aging pattern, the inferences that can be drawn from that work are somewhat limited. On a more practical level, studying memory scanning in DAT is important from an applied perspective. For instance, Duchek, Hunt, Ball, Buckles and Morris (1998) have shown that visual search error rates and RTs were the best predictors of driving performance in DAT individuals. Furthermore, in the DAT population, measures of selective attention (which is related to visual search processes) may serve to better distinguish safe versus unsafe drivers. Older drivers are more at risk for automobile accidents and this risk could stem from deficits related to visual search processes. Thus, a task such as memory scanning could be used as part of a MEMORY SCANNING PERFORMANCE 263 larger cognitive screening battery to identify such at-risk drivers. Efficient scanning performance is critical for older drivers because they need to scan and react to relevant information (children, stop signs, other cars) while at the same time inhibiting irrelevant information.