The n-Back Mania

The n-back task was first introduced by Kirchner (1958) in an experiment measuring age-related differences in “very short-term retention”. Since then its popularity significantly increased. A Web of Science search for the term “n-back” returned 1422 hits (December 2016), 250 of which were before Jaeggi’s study (Jaeggi et al., 2008) and 1172 afterwards. Jaeggi and colleagues used an adaptive dual n-back as the only training task that led them to conclude that after all, fluid intelligence can be improved. Another circumstance that added to the prominence of the n-back was the fact that its administration is amenable to the methodological constraints (stimulus and response timing, response formats) of many neuroimaging techniques (Redick and Lindsey, 2013). Thus, as the number of neuroimaging studies steadily increases, the number of n-back citations increases.

General characteristics

The n-back requires the individual to indicate whether or not the item currently presented matches the item that has been presented n items back. In the dual n-back this judgement is extended to yet another item, usually presented in another modality (e.g. auditory). Each stimulus is briefly displayed (about 500 ms) with a longer interstimulus interval (about 2.5 s). All these variables, type of presented stimulus, n, display time and interstimulus interval, can vary and in that way increase complexity and difficulty of the n-back. For instance, of special interest are so-called lure trials in which the current stimulus is the same as a recently presented one, but is not in the correct serial position to be a match (e.g., in a 2- back task: the second A in the sequence B–A–A–B is a lure because it matches the letter presented 1-back).

Figure 1. Example of the dual n-back task used in the study by Jaeggi et al. (2008). The spatial (squares) and auditory (letters) stimuli were presented simultaneously for a duration of 500 ms with an interstimulus interval of 2500 ms. Respondents were required to indicate their answer by pressing a letter on the keyboard whenever one of the presented stimuli matched the one presented n positions back in the sequence. The three tables show correct responses in a n = 1; n = 2 and n = 3 task paradigm. The original task used by Jaeggi and variants of it are available at: http://brainworkshop.sourceforge.net.

Since its introduction some 60 years ago, the n-back has been used as a working memory (WM[1]) task, despite the lack of research on its relation to WM and to similar tasks. Thus, there is little empirical evidence that would validate the n-back as a WM measure (Kane et al., 2007). Another aspect worth mentioning is the complex set of cognitive processes assumed to be involved in accomplishing the task. Because the processes are difficult to untangle, it is challenging to study their relation to WM, their contribution to n-back performance, and individual differences in this task.

Based on several research findings, Rac-Lubashevsky and Kessler (2016) provided the following list of processes assumed to be involved in its performance: (1) encoding the new item to WM, (2) binding the item to its position within the set of items held in WM, (3) comparing the new item to the one that appeared n trials before, (4) inhibiting irrelevant distraction to the comparison process which stems from other items inside or outside WM, (5) updating the item-position associations of all items in WM items, (6) and removing outdated information from positions that are no-longer relevant.

Another aspect that influences n-back performance is strategy use (Braver, 2012). At least two strategy types were identified: a proactive and a more reactive approach. The former is anticipatory where subjects are trying to retrieve the relevant representation in position n prior to the presentation of the new stimulus (e.g., in a 2-back condition A – B –  the  subject would anticipate a “YES” answer in expecting an A and a “NO” answer expecting a not A stimulus). In contrast, a reactive strategy would involve the use of the incoming stimulus as a cue for retrieval (e.g., in the same 2-back sequence, A – B – B would suggest a “NO” answer based on retrieving A from memory comparing it with the presented B, inferring A ≠ B).

Despite the number of processes involved in solving the n-back, the most often studied was working memory updating – the ability to modify the maintained information when needed. The process is supported by a gating mechanism, when the gate is open it allows updating (from internal or external perceptual sources). On the other hand, when closed it prevents interference allowing for maintenance. It is further assumed that the gate is controlled by the basal ganglia by the dopamine circuit, while the prefrontal cortex is in charge of maintenance (for a review see Ranganath and Jacob, 2015).

The n-back continuously triggers this alternation between opening and closing the gate. Because each of these processes is difficult to study in isolation, Rac-Lubashevsky and Kessler (2016) proposed another version of the  n-back which they dubbed the reference-n-back. More studies are needed to show if this new version can shed some light on the relation between the n-back and WM.

Validating the n-back: correlational and training studies

As put forward by Redick and Lindsey (2013), working memory research has mainly employed two task types. The n-back is a favorite in cognitive neuroscience, whereas the complex span task (also dubbed storage-plus-processing test) was most often used in psychological research on individual differences and cognition. Thus, if both tasks provide a measure of the same construct they should also correlate highly.

Before addressing this question let's look at the complex span task, which we briefly introduced in one of our previous posts, dealing with similarities/dissimilarities between WM and intelligence (see footnote 1).

Simple span tasks require that we recall a series of items (numbers or letters) in the same (forward span) or reverse order (backward span). In contrast, complex span tasks require subjects to engage in processing activity unrelated to the working memory task itself – tasks presented in between the to-be-remembered items. The first complex WM task was the reading span (RST), which was followed by several others (e.g., counting, operation, symmetry span, to name just some of them). The original RST was introduced by Daneman and Carpenter (1980). Subjects were asked to read series of unconnected sentences aloud and to remember the final word of each sentence. The number of sentences was incrementally increased until the maximum number of final words correctly recalled was found.

On the surface both tests, the n-back and variants of the complex span task, are to some extent similar (Redick and Lindsey, 2013):

·         Information from a set of stimuli must be maintained.

·         The items are usually presented visually or aurally.

·       Currently relevant information must be remembered preventing interference from other presented items.

·         Time constraints on the accessibility of items. 

It is worth mentioning that both task types are related to intelligence as shown in correlational (e.g., Kane et al., 2007)[2],  as well as training studies (e.g., Jaeggi, Buschkuehl et al., 2010). However as stressed by Kane et al. (2007), the main difference between the n-back and the complex span is that the former requires recognition and the latter serial recall. This may to some extent explain the low correlation reported by the meta-analysis of Redick and Lindsey (2013, p. 1102), in which they concluded: “The present meta-analysis showed that the complex span and n-back tasks are weakly correlated, although significant heterogeneity across studies was observed”. It was further suggested that the two tasks cannot be used interchangeably as WM measures. In the same direction point also several training studies using the n-back task[3].

These conclusions were recently opposed from a psychometric perspective by Schmiedek et al. (2013). They suggested that, beside the fact that low correlations can occur because tasks measure different constructs, also other reasons might influence such a relation:

·         The domination of task-specific sources of variance, which can be related to paradigms (familiarity information), or sources specific to contents, for instance, to count quickly in a counting span task.

·         Measurement error and restrictions of range (ceiling effects). 

To disentangle these effects the authors performed a confirmatory factor analysis. In the study participated 101 younger (51.5% women, age 20–31 years) and 103 older adults (49.5% women, age 65–80 years). Over several days they solved 3 complex memory span tasks, two 3-back tasks, 2 memory updating tasks, 2 sorting memory tasks and 9 reasoning tasks.

The assumed hierarchical factor model was tested with a structural equation modeling approach. The main finding of the analysis was that after excluding measurement error and content-specific sources of variance, the n-back and complex span tasks correlated substantially, with r = 0.69 in both age groups. However, this finding still does not explain why some n-back training studies show far transfer effects on intelligence test scores, but almost no transfer can be observed after complex memory span training.

Validating the n-back: functional neuroimaging studies

To date the most extensive quantitative meta-analysis of 668 sets of activation coordinates in Talairach space reported in 24 studies of n-back task variants (manipulating location vs. identity monitoring with verbal and nonverbal content) was performed by Owen et al. (2005). Their study provided robust evidence for the activation of 7 brain areas in all n-back variants (lateral premotor cortex; dorsal cingulate and medial premotor cortex; dorsolateral and ventrolateral prefrontal cortex; frontal poles; medial and lateral posterior parietal cortex and medial cerebellum). The authors concluded that the meta-analysis “provides evidence both, for broadly consistent activation of frontal and parietal cortical regions by various versions of the n-back working memory paradigm, and for process- and content-specific frontoparietal activation by working memory” (Owen et al., 2005,  p.46).

Figure 2. Medial (left) and lateral (right) views of brain areas activated in all n-back variants in the meta-analysis by Owen et al. (2005). The numbers represent Brodmann areas (BA); SMA – supplementary motor area:  (1) bilateral and medial posterior parietal cortex, including precuneus and inferior parietal lobules (BA7,40); (2) bilateral premotor cortex (BA6,8); (3) dorsal cingulate/medial premotor cortex, including supplementary motor area (SMA; BA32,6); (4) bilateral rostral prefrontal cortex or frontal pole (BA10); (5) bilateral dorsolateral prefrontal cortex (BA9,46); and (6) bilateral mid-ventrolateral prefrontal cortex or frontal operculum (BA45,47).

An independent component analysis (ICA) combined with a general-linear-model (GLM) approach employed in a recent study by Kearney-Ramos et al. (2014) confirmed all seven task-related regions described in the meta-analysis of n-back tasks by Owen et al. (2005).

All in all, we are faced with a contradiction that correlational and training studies find weak to no relationship between WM and the n-back, whereas neuroimaging studies show rather robust evidence for a relationship. A pragmatic conclusion might well be: who cares, if the n-back works as an IQ booster, why bother with its relation to WM – however, science is seldom solely focused on application in real-life settings.

References

Braver, T. S. (2012). The variable nature of cognitive control: A dual mechanisms framework. Trends in Cognitive Sciences, 16, 106–113. http://dx.doi.org/10.1016/j.tics.2011.12.010

Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and reading. Journal of Verbal Learning and Verbal Behavior, 19, 450–466. doi:10.1016/S0022-5371(80)90312-6.

Jaeggi, S. M., Buschkuehl, M., Perrig, W. J., & Meier, B. (2010). The concurrent validity of the n-back task as a working memory measure. Memory, 18, 394–412.

Jaeggi, S. M., Buschkuehl, M., Jonides, J., & Perrig, W. J. (2008). From the Cover: Improving fluid intelligence with training on working memory. Proceedings of the National Academy of Sciences, 105(19), 6829–6833. https://doi.org/10.1073/pnas.0801268105

Kane, M. J., Conway, A. R. A., Miura, T. K., & Colflesh, G. J. H. (2007). Working memory, attention control, and the n-back task: A question of construct validity. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33(3), 615–622. https://doi.org/10.1037/0278-7393.33.3.615

Kearney-Ramos, T. E., Fausett, J. S., Gess, J. L., Reno, A., Peraza, J., Kilts, C. D., & James, G. A. (2014). Merging Clinical Neuropsychology and Functional Neuroimaging to Evaluate the Construct Validity and Neural Network Engagement of the n-Back Task. Journal of the International Neuropsychological Society, 20(7), 736–750.

Kirchner, W. K. (1958). Age differences in short-term retention of rapidly changing information. Journal of Experimental Psychology, 55(4), 352–358.

Owen, A. M., McMillan, K. M., Laird, A. R., & Bullmore, E. (2005). N-back working memory paradigm: A meta-analysis of normative functional neuroimaging studies. Human Brain Mapping, 25(1), 46–59. https://doi.org/10.1002/hbm.20131

Rac-Lubashevsky, R., & Kessler, Y. (2016). Decomposing the n-back task: An individual differences study using the reference-back paradigm. Neuropsychologia, 90, 190–199. https://doi.org/10.1016/j.neuropsychologia.2016.07.013

Ranganath, A.,Jacob,S.N.,2015.Doping the mind dopaminergic modulation of prefrontal cortical cognition. Neuroscience,1–11.

Redick, T. S., & Lindsey, D. R. B. (2013). Complex span and n-back measures of working memory: A meta-analysis. Psychonomic Bulletin & Review, 20(6), 1102–1113. https://doi.org/10.3758/s13423-013-0453-9

Schmiedek, F., M., Lövdén, M. & Lindenberger, U. (2014). A task is a task is a task: putting complex span, n-back, and other working memory indicators in psychometric context. Frontiers in Psychology, 5. https://doi.org/10.3389/fpsyg.2014.01475

[1] For a detailed discussion of WM and its relation to intelligence click here

[2] The predicted variances with the intelligence test RAPM did not overlap between the two WM tests, suggesting that the performance on the two tasks relates to different sources of variance.

[3] A detailed overview will be presented in our forthcoming book: Increasing Intelligence (AP Elsevier) that will be released in February 2017.