Shared variance in measures of working memory capacity

The use of control strategies in the n-back task, and how the procedure might relate to other measures of working memory capacity, is discussed in detail in Chapter 3. It is generally agreed that indices of working memory capacity denote some use of attentional control to maintain and manipulate stimulus information, resolve interference, and perform other executive tasks (Baddeley & Hitch, 1974; Engle &

Kane, 2004; Oberauer, 2009). Consequently, individual differences research will typically show a correlation between multiple measures of working memory capacity, which supports the application of this executive/attentional system for fulfilling the task requirements (Engle & Kane, 2004; Schmiedek, Hildebrandt, et al., 2009; Wilhelm et al., 2013). This modality independent process is also supported in neuroimaging, where prefrontal activations are organised according to the task performed rather than the stimulus type used (e.g. Owen, 1997).

However, whilst a verbal n-back task is used as a common measure of working memory capacity for neuroimaging research, the actual utility of this procedure as a measurement of working memory is complex and equivocal (see Chapter 3 for a full discussion). In a meta-analysis by Redick and Lindsey (2013), the n-back task was weakly correlated to complex span tasks, suggesting the two cannot be used inter-changeably in research applications. In contrast, Schmiedek et al. (2014) produced latent factors from performance in three complex span tasks (reading span, counting span, and rotation span) and from n-back task performance (using a numerical and spatial n-back), and these two factors correlated substantially. Their reasoning for commonly-found poor correlations between complex span and n-back tasks were due to

paradigm-specific variance (e.g. the use of recall or recognition procedures), content-specific variance (e.g. the requirement for rapid counting in the counting span task, compared to visuo-spatial processing the rotation span task), and measurement error (e.g. ceiling and floor effects). This finding supports a hierarchical model of working memory performance where both n-back and complex span tasks involve the same higher-order construct in working memory, though of relevance for the present study is the finding that it is unsuitable to assess cross-modal relationships using cross-task comparisons. This is because the multiple sources of variance that result in weak relationships between tasks will mask whether the different-modality tasks applied here reflect the application of similar working memory resources.

Further support from Schmiedek et al. for the role of a higher-order working memory process that drives n-back performance is seen from the task’s strong relationship with fluid intelligence (measured by Raven’s Advances Progressive Matrices), which is proposed to occur because attentional control is essential for both skills (Carpenter, Just,

& Shell, 1990; Jaeggi et al., 2008; Schmiedek, Hildebrandt, et al., 2009; cf. Wilhelm et al. 2013 for a binding explanation of working memory capacity). Indeed, transfer effects have been observed from training in the n-back task to measures of fluid intelligence, also attributed to the requirement in both tasks for control of attention (Jaeggi et al., 2008).

A general mechanism in the n-back task is supported by modality-independent brain regions implicated during the procedure. Owen, McMillan, Laird, and Bullmore (2005) assessed different-modality n-back tasks in a meta-analysis of functional neuroimaging studies. Included in the analysis were multiple verbal and non-verbal n-back tasks (e.g.

shapes, faces, numbers, words, and fractals) that required either identity or spatial judgements. Their findings saw robust activation in the dorsolateral prefrontal cortex,

associated with strategic control of working memory processing (i.e. frontal lobe damage has been associated with the application of inefficient strategies in working memory, Owen, Morris, Sahakian, & Polkey, 1996); and the ventrolateral prefrontal cortex, implicated in the mapping of stimuli to responses upon presentation of targets or non-targets (Andersen and Buneo, 2003 cited in Owen et al., 2005). Their analysis noted these prefrontal activations in the olfactory n-back task performed in Dade et al. (2001), though it has been discussed previously and in Jönsson el al. (2011) that these findings may be explained by verbal processes in the stored odour representation rather than perceptual representations per se. However, whilst there appeared to be evidence for amodal activation during the n-back, there are also findings that support modality-specific regions of activation in the n-back task. Owen et al. (2005) acknowledged hemispheric lateralisation in frontoparietal regions related to whether the stimuli was verbal or non-verbal. Furthermore, in an n-back imaging study Knops et al. (2006) showed activation in the horizontal intraparietal sulcus for numerical stimuli, that they attribute to processing of averbal semantics (i.e. assessment of magnitude). That is, it is proposed that information in a working memory task will not only be represented as a phonological code, but will benefit from additional processing of semantic information.

The relationship between performance levels on different modality n-back tasks is another method by which generalised processing can be examined, and, in general, this method reveals strong different-modality task relationships. For example, numerical and spatial n-back procedures have shown a strong relationship (r = .66 across accuracy measures, Schmiedek et al. 2014), though other comparisons, between visuospatial and auditory 2-back tasks, have revealed weaker correlations (r = .35 across accuracy measures, Jaeggi et al. 2010). This is of interest to the present task because a relationship between olfactory n-back performance and n-back performance from other

modalities can elucidate the processes engaged in these tasks when performed with olfactory stimuli. This is because verbal and visual n-back tasks have previously shown a relationship not only with each other, but also to the ability to apply controlled working memory resources in latent variable studies (Schmiedek et al., 2014; Wilhelm et al., 2013). Specifically, this means that a relationship between olfactory n-back performance and performance in verbal and visual n-back tasks can be interpreted as a general application of controlled working memory processing. It should be noted, however, that analysis of individual differences across different-modality n-back tasks is naturally limited by the reliability of the measure itself (Jaeggi, Buschkuehl, et al., 2010; Redick & Lindsey, 2013). Performance in the n-back task is most reliable when n

> 1 (Friedman et al., 2008; Jaeggi, Buschkuehl, et al., 2010; Kane et al., 2007; Shelton, Elliott, Hill, Calamia, & Gouvier, 2009), though in Jaeggi et al. (2010) the split-half reliability of 2-back tasks varied between r = .26 and r = .85. This led the authors to conclude that the n-back task is not a suitable tool for measuring participant individual difference. To be clear, if within-participant variance is high throughout the task, then cross-modality comparisons have less validity due to uncertainty as to whether the 2-back score is a true representation of ability. However, Redick and Lindsay (2013) suggest the opposite, concluding in their meta-analysis that the n-back task does produce acceptable reliabilities (r > .70) in several studies (e.g. Kane et al., 2007;

Oberauer, 2005; Schmiedek, Hildebrandt, et al., 2009; N Unsworth & Spillers, 2010).

This is also supported in Schmiedek et al. (2014), with reliability estimates of α = .92 and α = .95 for number and spatial n-backs, respectively. Taken together, the findings suggest that reliability is not problematic for an individual-differences assessment of n-back performance.