Adaptive n-back training does not improve fluid intelligence at the construct level: Gains on individual tests suggest that training may enhance visuospatial processing

Highlights

• Short-term adaptive cognitive training based on the n-back task was applied.

• The psychological constructs of interest were assessed at the construct level.

• Matched training and control groups were compared on these constructs.

• No significant group differences were observed for any construct.

• Post-hoc analyses suggested that training enhances general visuospatial processing.

Abstract

Short-term adaptive cognitive training based on the n-back task is reported to increase scores on individual ability tests, but the key question of whether such increases generalize to the intelligence construct is not clear. Here we evaluate fluid/abstract intelligence (Gf), crystallized/verbal intelligence (Gc), working memory capacity (WMC), and attention control (ATT) using diverse measures, with equivalent versions, for estimating any changes at the construct level after training. Beginning with a sample of 169 participants, two groups of twenty-eight women each were selected and matched for their general cognitive ability scores and demographic variables. Under strict supervision in the laboratory, the training group completed an intensive adaptive training program based on the n-back task (visual, auditory, and dual versions) across twenty-four sessions distributed over twelve weeks. Results showed that this group had the expected systematic improvements in n-back performance over time; this performance systematically correlated across sessions with Gf, Gc, and WMC, but not with ATT. However, the main finding showed no significant changes in the assessed psychological constructs for the training group as compared with the control group. Nevertheless, post-hoc analyses suggested that specific tests and tasks tapping visuospatial processing might be sensitive to training.

Introduction

General intelligence (g) is defined by a broad ability for reasoning, solving problems, and efficient learning (Gottfredson et al., 1997). Hunt, 1995, Hunt, 2011 underscores the distinction between fluid intelligence (Gf) and crystallized intelligence (Gc), although these two broad abilities are related with g (Carroll, 1993, Carroll, 2003, McGrew, 2009). Gc involves the intelligent use of culturally rooted knowledge and skills (such as language or math), whereas Gf requires abilities for solving novel and abstract problems (Cattell, 1987). These latter abilities are the main target of cognitive training programs.

It has been repeatedly demonstrated that tests' scores can be increased (Neisser et al., 1996, Nisbett et al., 2012) but, is it possible to improve cognitive ability? (Colom et al., 2010). For obtaining convincing evidence, the training tools must be substantially different to usual tests of cognitive ability. Test specific skills can be improved by increased familiarity (te Nijenhuis, van Vianen, & van der Flier, 2007) but this is hardly interesting. There are reports showing improvements in intelligence, as assessed by standard tests, after training information processing skills. Thus, for instance, Posner and Rothbart (2007) trained children on a visual attention task based on the management of conflict. The trained children scored higher than a control group on a standard intelligence battery. Also studying children, Irwing, Hamza, Khaleefa, and Lynn (2008) reported large improvements in the Raven Progressive Matrices Test in a group trained for several months with the abacus (requiring the reliable preservation of intermediate calculations in working memory) when compared with a non-trained group. Jaušovec and Jaušovec (2012) reported a positive effect in the RAPM test after working memory training; the change from the pretest to the posttest assessment was equivalent to thirteen IQ points (d = .88) for their training group, whereas it was null for an active control group. Digit span scores were also substantially higher for the training group (d = 0.81) than for the control group (d = 0.25). The study by von Bastian and Oberauer (2013) concluded that general reasoning ability can be improved by working memory training (self-administered at home). The positive effect was also observed six months after ending the training program. Further, training of specific working memory processes (storage and processing, relational integration, or supervision) led to transfer in specific cognitive factors.

Jaeggi, Buschkuehl, Jonides, and Perrig (2008) reported that training in a challenging adaptive dual n-back task (tapping a mixture of executive updating, working memory, and attention skills) was related to better performance on a fluid intelligence test compared to a passive control group. These results were repeated in further studies: (a) training on the single n-back task (either visual or verbal) showed similar positive effects over performance on fluid intelligence tests (Jaeggi et al., 2013, Jaeggi et al., 2010) and (b) similar findings were observed in a sample of children (Jaeggi, Buschkuehl, Jonides, & Shah, 2011). However, the conclusion that n-back training improves fluid intelligence is controversial. For instance, Moody (2009) argued that improvements on the specific fluid measure considered by Jaeggi et al. (2008) could be explained by the strict time limit imposed for solving the less difficult items. In his view, no challenge was made over the participants' Gf, and, therefore, observed changes may be fully explained by a speed factor. From a broader perspective, Shipstead, Redick, and Engle (2012) argued that these short-term training studies fail to really increase abilities required by the working memory processing system. In their view, published studies (1) generally rely on single measures for measuring predicted intelligence changes after training and (2) administer invalid measures of working memory capacity. These authors note that a wide variety of tasks measuring the constructs of interest must be systematically administered in order to avoid critiques related to task specificity issues.

A study by Chooi and Thompson (2012) was aimed at overcoming some of the reservations enumerated by Shipstead et al. (2012) and used several measures of intelligence (crystallized-verbal, spatial, and speed) for estimating changes after training on the dual n-back task modeled from Jaeggi et al. (2008). Working memory was measured by a single task (operation span). They failed to find any effect of training on either intelligence or working memory. Passive and active controls were considered along with the training group, using two time lengths (8 and 20 days) resulting in very small sample sizes for the six analyzed groups (from 9 to 23 participants). Importantly, the n-back performance level achieved by the trained participants in the 20-day training period was well below the one attained by the Jaeggi et al.'s sample.

Redick et al. (2012) reported a similar study. Fluid (six tests) intelligence and crystallized (two tests) intelligence, along with working memory (two tasks), multitasking (three tasks), and processing speed (two tasks) were the measured constructs. Training (N = 24), active (N = 29), and passive (N = 20) control groups were analyzed. Very short versions of the considered psychological tests were administered before, during, and after the training stage. This study failed to find any difference among the three groups, consistent with Chooi and Thompson (2012). Surprisingly, Redick et al. failed to find any practice effect across their three evaluations. Again, n-back performance level achieved by the trained participants was almost identical to the attained in Chooi and Thompson and well below the reported by Jaeggi et al., 2008, Jaeggi et al., 2010.

Recently, Stephenson and Halpern (2013) replicated the Jaeggi et al., 2008, Jaeggi et al., 2010 main findings. However, significant gains were observed in two out of four fluid intelligence tests (RAPM and BETA-Matrix Reasoning). Thus, for instance, after the adaptive training program based on the dual n-back (N = 28) gains were equivalent to (a) 13.3 IQ points (d = 0.89) in the BETA-Matrix Reasoning (b) 9.9 IQ points (d = 0.66) in the RAPM, (c) 8.4 IQ points (d = 0.56) in the WASI-Matrix Reasoning, and (d) 5.2 IQ points (d = 0.35) in the Culture-Fair Intelligence Test.

The theoretical framework for the present study is based on the available evidence demonstrating a very high correlation between intelligence and working memory at the latent variable level (Colom et al., 2004, Oberauer et al., 2005). The comprehensive study by Martínez et al. (2011) is a recent example considering twenty-four measures tapping eight intelligence and cognitive factors (three measures for each factor): fluid-abstract intelligence, crystallized-verbal intelligence, and spatial intelligence, along with short-term memory, working memory capacity, executive updating, attention, and processing speed. Their main findings support the view that fluid intelligence can be largely identified with basic short-term storage processes tapped by working memory tasks and executive updating. This was seen as quite consistent with neuroimaging results showing that fluid intelligence shares relevant brain structural (Colom, Jung, & Haier, 2007) and functional (Gray, Chabris, & Braver, 2003) correlates with working memory capacity. The large correlation between intelligence and working memory at the latent variable level suggests that they share substantial capacity limitations based on the amount of information that can be reliably kept active in the short-term, both within the working memory system or during the reasoning processes required on intelligence tests (Colom et al., 2008, Colom et al., 2006, Halford et al., 2007).

The proper testing of the prediction that improvements in the working memory system (short-term storage and executive updating) through adaptive cognitive training will promote increments in fluid intelligence, mainly because their common limitations for the reliable temporary storage of the relevant information will be boosted, requires straightforward analyses going well beyond the level of specific measures, as underscored by Shipstead et al. (2012). For that purpose, here we administered several diverse intelligence and cognitive measures (three measures for each psychological factor) before and after completing a challenging cognitive training program based on the adaptive n-back dual task firstly proposed by Jaeggi et al. (2008).

The main prediction is that if adaptive working memory training promotes skills relevant for the reliable temporary storage of relevant information, then fluid intelligence and working memory scores will be higher for the trained than for the control group at the posttest evaluation. Further, (a) these higher scores must be systematically observed for all Gf and WMC specific tests and tasks, and (b) crystallized intelligence and attention control will not be sensitive to training. Both fluid intelligence and working memory require the reliable preservation of the relevant information in the short-term, as demonstrated by the seminal study by Carpenter, Just, and Shell (1990). This is not the case for crystallized intelligence and attention control, because Gc requires the recovery of the relevant information from long-term memory and attention control does not requires any short-term storage.