Louise H.Phillips and Colin Hamilton
In this chapter we examine the empirical evidence for adult age changes in the components of the Baddeley and Hitch working memory model, and highlight some methodological and theoretical issues raised by the study of age differences in working memory. Baddeley (1996, p. 19) argues that
‘ageing may be an interesting and productive variable to study within the context of working memory’. The idea that adult age changes in the available capacity of working memory underlie deficits in reasoning and language abilities has been extremely influential. However, the majority of work in this area has conceptualised working memory in terms of a general limited-capacity system, rather than using the notion of the Baddeley and Hitch three component model (WM model) with specialised subsystems for the maintenance of verbal and visuo-spatial information.
In the aging literature, the usual view of working memory is that age differences reflect a decrease in the amount of cognitive resources that can be shared out to deal with competing task demands. This maps onto the predominant limited-resource model of working memory in the North American literature. In the current chapter we will not discuss in detail the research that largely utilises this resource capacity approach to working memory, but will instead concentrate on research specifically relevant to the WM model. In the aging literature particularly, this notion of working memory is closely tied in to other conceptions of limitations on information-processing, such as attentional capacity and processing speed (Salthouse, 1991). Salthouse (e.g. 1992) has argued that slowed speed of processing information underlies the decline in capacity of working memory with age. Evidence from statistical partialling techniques suggests that age-related variance in working memory can largely be explained in terms of processing speed at both the beginning and end of the lifespan (Chuah & Maybery, 1999; Kail & Salthouse, 1994; Salthouse, 1992).
The WM model has been utilised less in aging research than limited general resource models. However, the WM model does have promise as a tool for understanding adult aging because it allows the possibility of a functional
account of exactly how and why cognitive changes occur with age, rather than the basic assertion that age differences are ‘resource’ differences. It seems appropriate that we should try to seek cognitive models of the effects of cognitive aging, rather than relying solely on neurobiological models, which are often the endpoint of the processing speed theories. The highly specified nature of the WM model compared to the rather nebulous concept of ‘working memory capacity’ also allows for specific predictions to be tested experimentally, rather than having to rely upon statistical partialling techniques. As will be outlined below, there are also problems with the use of the WM model to study aging—in particular, the general nature of aging deficits across a range of cognitive tasks may raise questions about the parsimony of using a multicomponent model as an explanation.
Many studies indicate age-related changes in measures of working memory. However, the magnitude and nature of age differences in working memory appear inconsistent across studies, and are influenced strongly by the type of test used in ways which are not yet fully explained (Baddeley, 1996). Baddeley (1996) argues that it is precisely because the effects of age on working memory appear inconsistent and challenging to explain that aging may be an interesting variable to study in respect of the working memory model. In this chapter we outline evidence for age-related changes in the various components of working memory, outline the methodological techniques that have generally been used in this area, and then evaluate the usefulness of the WM model in aging research.
THE EFFECTS OF ADULT AGE ON THE INDIVIDUAL COMPONENTS OF THE BADDELEY AND HITCH WORKING MEMORY MODEL
Phonological loop
The phonological loop, or verbal buffer system, is the best specified component of the working memory model. There is considerable evidence to support the idea that this system comprises an articulatory rehearsal mechanism based on inner speech, and a phonological store. There has been a lot of interest in the effects of adult age on verbal working memory, particularly in relation to language processing (see e.g. chapters 3–5 and 8–9 in Light & Burke, 1988).
Relatively few studies have used the WM model to investigate normal adult age differences in verbal processing. One aspect of the functioning of the phonological loop that has been examined in relation to age is the rate of articulation. Articulation rate is likely to influence the number of items that can be rehearsed, and hence memory influence the number of items that can be rehearsed, and hence memory span. There is evidence that as children get older there is a strong relationship between quickening speech rates and verbal memory span (Hulme, Thomson, Muir, & Lawrence, 1984). It has been argued (Kynette, Kemper, Norman, & Cheung, 1990) that slowed articulation rates may
WM model in adult aging research 103 cause impairments of verbal short term memory and language processing in older adults. Gerhand (1994) found that adult age differences in digit span could be entirely explained by statistically removing variance due to articulation rate, which was considerably slower in older adults. This raises the question as to whether articulation rate is an indicator of a more general speed of information-processing factor which underlies age differences (Chuah & Mayberry, 1999;
Kail, 1993; Smyth & Scholey, 1996). Chuah and Maybery (1999) present evidence that in children, age-related improvements in short-term memory are best explained in terms of a domain-independent increase of processing speed, rather than changes in articulatory rate having a specific impact upon verbal memory. The role of processing speed in adult age differences in the operation of a phonological rehearsal mechanism has not been fully explored.
The functioning of the phonological loop can be examined through the
‘unattended speech effect’, in which the presentation of task-irrelevant speech-related sounds interferes with the ability to retain verbal information (Salamé &
Baddeley, 1982). Rouleau and Belleville (1996) examined the functioning of the phonological loop in older adults by testing whether there were age differences in the effects of irrelevant speech on verbal memory. They hypothesised that older adults should be adversely affected by irrelevant speech because older people often report difficulty in filtering out background noise in everyday situations. Rouleau and Belleville reported a general age effect on verbal memory, with older adults recalling fewer digits. There was however no interaction between age and the effects of irrelevant speech, suggesting no particular age-related difficulty in dealing with task-irrelevant noise.
A further method of examining the phonological loop is through the use of concurrent articulatory suppression (e.g. the participant repeatedly saying ‘the’
while performing another task) which is argued to prevent subvocal rehearsal of verbal information. Gerhand (1994) gave a digit span task to old and young adults both with and without concurrent articulatory suppression. There were significant age differences in digit span in the single task condition, but no age differences in digit span during concurrent articulatory suppression, i.e. when subvocal rehearsal was prevented. It was therefore concluded that slow subvocal rehearsal may underlie age differences in simple verbal memory tasks.
Overall the evidence suggests that slowed articulation rates may impair verbal memory in older adults. Otherwise the operation of the articulatory loop remains relatively intact with age. The experiments outlined above highlight the value of the WM model in aging research: it allows investigation of well-specified cognitive processes (e.g. articulatory rehearsal) and how they change.
Visuo-spatial sketchpad
Very little research has directly examined the effects of age on the visuo-spatial sketchpad, the visuo-visuo-spatial buffer component of working memory.
However, age differences favouring younger adults have been reported on a number of imagery tasks which are thought to rely upon the visuo-spatial
slave system. A large number of studies have shown older adults to be slower and less accurate at tasks of mental rotation (e.g. Cerella, Poon, &
Fozard, 1981). Dror and Kosslyn (1994) looked at the effects of adult aging on tasks of image generation, scanning and rotation. They argue that there are substantial age differences in the ability to rotate images and activate stored images. In contrast there was little age effect on the ability to generate or scan images. In relation to the WM model it is likely that most of these tasks have some executive component, making it difficult to distinguish whether age differences on the tasks are specifically related to the operation of the sketchpad or the central executive. More data relevant to the distinction between central executive and visuo-spatial sketchpad changes with age are presented below in the section on studies comparing different aspects of the model.
There are no age differences in the benefit found for remembering concrete (imageable) versus abstract words (Dirkx & Craik, 1992), suggesting that both young and old adults spontaneously make use of visual imagery. However, it may be the case that younger adults can use imagery processes more efficiently than old. Dirkx and Craik report that when given a list of words to learn younger adults remembered considerably more than old. In contrast, when given word lists to learn during a simultaneous visual interference task the age difference was non-significant, suggesting that the younger group was making more effective use of imagery during the single task condition (Dirkx & Craik, 1992).
Taken together with the results reported above for age differences in the effects of articulatory suppression, this suggests that both verbal and visual rehearsal processes may be involved in age differences in remembering verbal material. However, there are still very few studies which address changes in visuo-spatial sketchpad functioning with age, and it would be of particular interest to investigate the role of both verbal and visuo-spatial rehearsal in age differences in different types of memory paradigm.
Central executive
Baddeley (1986) has argued that the central executive component of working memory is particularly impaired in older adults compared to the verbal and visuo-spatial slave systems, and increasing evidence has been gathered which supports this viewpoint. A number of paradigms have been designed specifically to explore central executive functioning. ‘Keeping track’ tasks are proposed to tap the ‘memory-updating’ facility of working memory (Morris & Jones, 1990). Keeping track of information has been shown to decline with age (Dobbs & Rule, 1989). Another task argued to depend heavily on the central executive component of working memory is the production of random strings of numbers (Baddeley, 1986). Producing random output demands repeated inhibition of stereotyped automatic sequences (Baddeley, 1986), and thus places considerable demands on
WM model in adult aging research 105 executive processes. There is evidence that older subjects produce less randomly distributed strings of digits than their younger counterparts (Van der Linden, Bregart, & Beerten, 1994). Also, older adults are less able to produce random sequences of tapping responses (Phillips, Gilhooly, Logie, Della Sala, & Wynn, submitted). However, there is still insufficient knowledge about how younger and older adults are performing these random generation tasks to be confident that the age differences reflect poorer inhibition for example, and not poorer understanding of the task demands.
The most widely used model of central executive function is based on the Norman and Shallice (1986) supervisory attentional system. This model suggests that the central executive is involved in a range of cognitive control processes such as planning, monitoring, and inhibition of inappropriate stimuli or responses. Further, this links the central executive to the operation of the frontal lobes of the brain. There has been a considerable amount of recent literature on aging devoted to the relationship between age, the frontal lobes and executive functioning (e.g.
Moscovitch, 1994; Parkin, 1997; West, 1996). The frontal lobes show evidence of deterioration earlier, and more rapidly in response to aging than any other brain area (Coffey et al., 1992; Coleman & Flood, 1987;
Raz, Gunning, Head, Dupuis, & Acker, 1998). There is substantial evidence of adult age differences in performance on neuropsychological tests argued to assess executive function such as the Wisconsin Card Sort test, verbal fluency and the Stroop test (see e.g. Rabbitt, 1997). However, poor performance on these tests may reflect age changes in other factors such as lower level information-processing characteristics rather than executive deficits (Phillips, 1999; Uttl & Graf, 1997).
It is difficult to distinguish empirically between ‘processing resource’
and ‘central executive’ theories of aging because both predict fairly general and widespread cognitive deficits with age. One example of this is the Stroop task, where colour words are printed in different colour inks, and in the ink-naming conditions the tendency to read the colour name must be suppressed while the ink colour is named (e.g. the answer to YELLOW would be ‘black’). Executive decline with age would be predicted to cause poorer inhibition of the inappropriate response. This should then cause age-related slowing on the colour ink naming condition of the Stroop task compared to the easier task of reading the colour word.
Equally though, the processing speed theory would predict greater age-related slowing on the colour-ink naming condition, because there are more processing components to be slowed. Verhaeghen and De Meersman (1998) reviewed twenty studies of age effects on the Stroop task, and concluded that ‘the apparent age-sensitivity of the Stroop interference effect appears to be merely an artefact of general slowing’ (p. 120).
It is difficult to isolate the effects of age on the central executive without also assessing the impact of age upon slave system operation.
Some studies have compared the effects of age on the central executive component with age effects on slave systems, and this research is considered in more detail below. One question that is likely to occupy aging researchers in the future is the extent and nature of fractionation in the central executive, and whether there are differential age trajectories of the various executive functions. As it stands, the WM model could fit in with many different patterns of executive function fractionation. A number of correlational studies have attempted to determine how executive functions might be segregated (e.g. Gnys & Willis, 1991; Miyake et al., 2000; Robbins et al., 1998). However, there seems so far to be little a g r e e m e n t a m o n g s t t h e s e s t u d i e s a s t o t h e p a t t e r n o f exe c u t ive fractionation. Further, there is little clear-cut neuropsychological evidence of localised brain areas associating with particular types of executive deficit. More detailed studies of the nature of executive changes with age, particularly if they shed light on the cognitive processes underlying older adults poor performance on particular executive tests, may increase understanding of the nature and interactions of the various functions of the central executive.
STUDIES COMPARING THE EFFECTS OF AGE ON DIFFERENT COMPONENTS OF THE MODEL
The usefulness of the WM model as a research tool lies partly in its multicomponential nature. A few studies have utilised this to investigate whether age differences in working memory might be particularly attributable to individual components within the model. Salthouse (1994) offers some data to test the model in aging research. Salthouse, Kausler, and Saults (1988) looked at age differences in verbal and spatial memory, using the same stimuli in both cases, but different forms of recall. Very similar degrees of age decline were found in both tasks, and Salthouse (1994) argues that this indicates similar rates of age deterioration of both phonological loop and spatial scratchpad systems. However, both verbal and spatial span tasks presumably make reasonable demands on the central executive, so it is difficult to ascertain from this type of data the extent to which individual components from the WM model may change with age.
A n u m b e r o f o t h e r s t u d i e s h ave ex a m i n e d a g e d i ff e r e n c e s i n components of working memory in more detail. Two main methodologies have been used—correlational and experimental—and the rest of this section is divided according to the type of methodology used. First, research studies using correlational techniques to partial out variance due to particular working memory components are examined. Two different experimental approaches are considered: looking at age effects when the load placed on particular components of working memory is reduced, and the dual task approach, where the load on particular components is increased.
WM model in adult aging research 107 The correlational approach
Resource hypotheses propose that individuals differ in the amount of fundamental cognitive resources (e.g. processing speed) available. It is therefore not possible to directly manipulate the main explanatory variable, as speed is proposed to reflect a fundamental limitation on processing. However, statistical techniques can simulate this manipulation:
for explanations of the logic of this procedure see Hertzog and Dixon (1996) and Salthouse (1991). There are problems with the use of statistical control methods that should be borne in mind when interpreting results.
The data obtained are correlational, and therefore assumptions about causality must be treated with caution. Also, these partialling techniques assume that the effects of age and working memory are additive and linear, and do not usually test for interactions between age and resource limitations (Hertzog & Dixon, 1996). There are a number of different statistical techniques to look at the reduction in shared variance between age and reasoning ability once memory measures are partialled out, but as yet no established method of testing for the probability that such a reduction occurred due to chance factors. It is therefore a matter of judgement as to whether a particular magnitude of reduction in variance is meaningful, i.e. ‘significant’.
Although very many aging studies use correlational approaches to examine the role of working memory in age changes in cognition, few have done so with the intention of examining the components of the WM model.
Fisk and Warr (1996) attempted to distinguish the effects of age on the phonological loop and central executive components of working memory u s i n g c o r r e l a t i o n a l t e c h n i q u e s . T h ey a s s e s s e d c e n t r a l exe c u t ive functioning using ability to generate random strings of letters (measured using various indices of randomness), and phonological loop function using measures of digit span and word span. They investigated whether age differences in measures of working memory (computation span and reading span tasks that demand simultaneous processing and storage) could be explained in terms of phonological loop or central executive functioning. Using hierarchical regression, it was revealed that age differences in working memory could not be explained by differences in the phonological loop measures, but could partially be explained by differences in central executive measures.
It is interesting to note that in the Fisk and Warr study, age differences in the working memory measures were poorly explained by a task which appears relatively similar in format to the criterion working memory span measures (the articulatory loop measure of word span), and relatively well explained by performance on a task which appears extremely dissimilar in format and task requirements (random generation). This supports the argument that age differences in relatively difficult working memory tasks are more related to central executive than phonological loop functioning.
Fisk and Warr also assessed processing speed using simple perceptual comparison tasks. They found that processing speed was a good predictor of age differences in working memory span, and conclude that speed of perception may underlie age differences in executive functioning which in turn influence working memory span.
It would be interesting to see this methodology applied to a wider range of tasks, in particular visuo-spatial measures, to examine the relationships between indices of slave system performance, executive functioning, and target memory tasks. However, it is important that the validity of the purported measures of working memory components is established. Future research in this area may clarify further the role of specific visuo-spatial or verbal mechanisms in age changes in memory and reasoning. For example, it would be interesting to know whether age differences in spatial span are specifically attributab le to spatial rehearsal rate or more general information-processing rate, as has been reported for developmental changes in spatial memory in children (Chuah & Maybery, 1999).
Experimental manipulations of working memory—the central executive versus slave systems
Work in this area uses experimental modifications of working memory paradigms to investigate age effects on different components of the WM model. A series of studies carried out by Morris, Gick and Craik examined the effects of age on various manipulations of verbal memory paradigms (e.g., Craik, Morris, & Gick, 1990; Morris, Craik, & Gick, 1990). The tasks used were mostly based on the ‘sentence span’ task, where sentences have to be verified while simultaneously remembering words. Various manipulations of complexity were carried out, but only some showed significant age interactions, i.e. much poorer performance by older adults under the more complex conditions. There was no age interaction with storage complexity, i.e. older adults were not differentially affected by the inclusion of more words to be remembered. It was therefore argued that age differences in verbal working memory do not reflect limitations on the phonological loop.
An increase in processing complexity, in terms of the grammatical complexity of the sentences was generally associated with much poorer performance by older adults. It was argued that these results indicate that with age ‘both the capacity and flexibility of the central executive are impaired to some degree’ (Craik et al., 1990, p. 264). They further argue that while there was little age difference in the efficiency of the articulatory rehearsal mechanism, younger participants were more effective at augmenting phonological loop functioning using central executive function.
This suggests that older adults rely more on articulatory rehearsal than young, particularly when executive processes are loaded.
Next, the effects of age on visual and spatial memory are considered in relation to the respective roles of the central executive and slave systems of
WM model in adult aging research 109 working memory. Recent research carried out at the Universities of Northumbria and Teeside is relevant to the distinction between different aspects of the visuo-spatial sketchpad and central executive components in age differences in working memory. Logie and Pearson (1998) found that children mature faster on tasks of visual span (e.g. retaining abstract visual matrix patterns) compared to the spatial span (such as the Corsi block test in which participants are required to retain a sequence of spatial locations).
This has been interpreted in relation to a more detailed model of the visuo-spatial sketchpad (Logie, 1995) in which a visual cache stores primarily visual information and a spatially orientated inner scribe retains information about movement sequences (see chapter 2, this volume).
Coates, Sanderson, Hamilton and Heffernan (1999) replicated the result that children appeared to improve faster on the visual matrix task compared to the Corsi spatial task. They also found that older adults performed significantly more poorly than young adults on both the Corsi and matrix tasks, suggesting age-related decline in both spatial and visual memory.
In terms of interpreting age differences in these tasks in relation to the WM model it is unclear the extent to which different components might be involved. Matrix and Corsi tasks were not designed to isolate components of the working memory model, and therefore age differences in performance could be due to demands upon the slave components or the more generic central executive resources in the model. There is preliminary evidence (Hamilton, Heffernan, & Coates, 1999) that concurrent verbal fluency (a task thought to tap executive processes) interferes with both Corsi and visual matrix span tasks. The increase of pattern complexity in the matrix tasks affords the possibility that participants will be able to form sophisticated representations of the p a t t e r n , a v i s u a l ‘ c h u n k i n g ’ p r o c e s s w h i c h c o u l d u n d e r l i e t h e developmental change seen in their data. Also, active spatial rehearsal could be employed to maintain the pattern representation. Such rehearsal methods might load the executive component of working memory. The requirement in the Corsi block task for sequential order representation may make significant demands upon executive processes (Farand & Jones, 1996; Smyth & Scholey, 1996).
Further research (Hamilton et al., 1999) has used a componential approach (Farah, 1984), to produce cognitive tasks suitable for all age groups which tap the known components of visuo-spatial working memory (Logie, 1995). This research involved the construction of working memory tasks that were relatively free of executive demands and made specific demands upon the slave systems. To assess visual memory, a task was developed where spot size was memorised. There was an initial brief exposure of a spot, followed by a maintenance interval, followed by the representation of a spot that was either the same size or different. The size step was reduced from 50 per cent through to 5 per cent, with twenty trials