CoChat
Sign Up
Cited by 351Cognitive Abilities and TestingCognitive Functions and MemoryNeural and Behavioral Psychology Studies

Working memory training revisited: A multi-level meta-analysis of n-back training studies

Open access
Soveri, Anna, Antfolk, Jan, Karlsson, Linda, Salo, Benny, Laine, Matti·2017·Psychonomic Bulletin & Review
DOI 10.3758/s13423-016-1217-0

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

The efficacy of working memory (WM) training has been a controversial and hotly debated issue during the past years. Despite a large number of training studies and several meta-analyses, the matter has not yet been solved. We conducted a multi-level meta-analysis on the cognitive transfer effects in healthy adults who have been administered WM updating training with n-back tasks, the most common experimental WM training paradigm. Thanks to this methodological approach that has not been employed in previous meta-analyses in this field, we were able to include effect sizes from all relevant tasks used in the original studies. Altogether 203 effect sizes were derived from 33 published, randomized, controlled trials. In contrast to earlier meta-analyses, we separated task-specific transfer (here untrained n-back tasks) from other WM transfer tasks. Two additional cognitive domains of transfer that we analyzed consisted of fluid intelligence (Gf) and cognitive control tasks. A medium-sized transfer effect was observed to untrained n-back tasks. For other WM tasks, Gf, and cognitive control, the effect sizes were of similar size and very small. Moderator analyses showed no effects of age, training dose, training type (single vs. dual), or WM and Gf transfer task contents (verbal vs. visuospatial). We conclude that a substantial part of transfer following WM training with the n-back task is task-specific and discuss the implications of the results to WM training research.

AI snapshot

Written from the paper’s full text.

TL;DR

This multi-level meta-analysis of 33 randomized controlled n-back training studies in healthy adults found a medium-sized transfer effect to untrained n-back tasks but only very small effects to other WM, cognitive control, and fluid intelligence tasks. The authors conclude that much of n-back training transfer is task-specific rather than evidence of broad working memory improvement.

Key findings

  • Across 203 effect sizes (190 after data screening) from 33 studies, 41 experiments, and 2,105 participants, transfer was largest to untrained n-back tasks: g = 0.62 [0.44, 0.81], p < .001; other WM tasks: g = 0.24 [0.16, 0.32], p < .001; cognitive control: g = 0.16 [0.05, 0.27], p < .001; and Gf: g = 0.16 [0.08, 0.24], p < .001. The n-back effect differed from all other domains (all ps < .001), while the other domains did not differ from each other (all ps ≥ .134).
  • Moderator analyses showed no effects of age, training dose (sessions or hours), training type (single vs. dual n-back), or WM/Gf transfer task materials (verbal vs. nonverbal); for example, training type: QM(1) = 2.19, g = 0.09 [−0.03, 0.21], p = .139; age: QM(1) = 0.07, g = 0.03 [−0.17, 0.22], p = .794.
  • Within studies that used both control types, active and passive control groups did not differ in pre-post change, g = 0.03 [−0.07, 0.12], p = .429. When control type was included as a covariate, passive-control studies showed a slightly higher estimated training effect, g = 0.11 [0.00, 0.22], p = .048; after controlling, n-back remained g = 0.59, WM g = 0.18, cognitive control g = 0.08 (p = .202), and Gf g = 0.11 (p > .05).
  • A reanalysis of healthy-adult data from Melby-Lervåg et al. (2016) found that untrained versions of the training tasks showed a stronger effect, g = 0.62 [0.38, 0.86], than other WM tasks, g = 0.21 [0.08, 0.33], QM(1) = 8.89, p < .01.

Methods

  • Multi-level meta-analysis of randomized, controlled trials of n-back training in healthy adults. Searches of PubMed, PsycINFO, Google Scholar, Google, and ERIC (initial February 2015; updated July 2015, February 2016, October 2016) required pretest-posttest designs, n-back-only training, active control training not targeting WM, and transfer measures in n-back, WM, Gf, or cognitive control. Final sample: 33 studies, 41 experiments, 103 groups, 2,105 participants, and 203 training effects (190 after screening).
  • Effect sizes were within-group pre-post standardized mean changes (SMCR) converted to between-group training effects (g_BETWEEN). A three-level model included random intercepts for experiment and for measure within experiment (ICC = 0.499); pre-post correlations were available for 406 of 507 within-sample comparisons, deduced or conservatively estimated for 80 of the remaining 101, and set to 0 when unavailable. Moderators included control type, training type, training dose, transfer materials, and age. Publication bias was assessed with contour-enhanced funnel plots, residual funnel plots, and PET-PEESE regressions.

Limitations

  • Stated: the meta-analysis focused on only one WM training type (n-back), limiting conclusions to n-back training; the limited number of available studies prevented investigating transfer to other cognitive domains and all moderator analyses for all four domains (e.g., material-specific transfer was investigated only for WM and Gf); interactions between moderators were not investigated.
  • Evident from the text: the sample was restricted to healthy adults, so findings may not generalize to clinical samples, children, or other WM training paradigms; reliance on author-provided or estimated pre-post correlations for many effect sizes may affect precision; and publication-bias assessment was complicated by dependent effect sizes and the multilevel structure, leading the authors to use visual inspection and PET-PEESE rather than p-curve or trim-and-fill.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3758/s13423-016-1217-0

Cites

83 of 83 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

13 of 50 cited sources checked · 7 supported · 1 flagged · accuracy not yet measured

  • Premisecitation

    Previous meta-analytic near transfer measures have lumped together untrained WM tasks that differ from training tasks only by stimuli and WM tasks structurally different from training tasks.

    “Third, meta-analytic near transfer measures have lumped together untrained WM tasks that differ from the training tasks only by their stimuli and WM tasks that are structurally different from the training tasks”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Melby-Lervåg & Hulme, 2013 source
    • cites Melby-Lervåg et al., 2016 source

      The Method section (Search, Inclusion Criteria and Coding; page not available) says this meta-analysis separated near-transfer tests similar or identical to trained tasks from an intermediate category of other working-memory measures. It does not lump those categories together as the claim says. (LLM, conf=0.84)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured
    • cites Schwaighofer et al., 2015 source
  • Premisenone given

    The meta-analysis by Melby-Lervåg et al. (2016) included studies that were not randomized controlled trials.

    “For example, the meta-analysis by Melby-Lervåg et al. (2016) included studies that were not randomized, controlled trials.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Melby-Lervåg et al., 2016 source
  • Premisereasoning

    There is disagreement between previous meta-analyses, particularly regarding the size of near transfer effects.

    “As is evident from the short review above, there is some disagreement between the previous meta-analyses, particularly regarding the size of the near transfer effects.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Melby-Lervåg & Hulme, 2013 source
    • cites Melby-Lervåg et al., 2016 source

      The Prior Reviews of Working Memory Training section (page unavailable) says earlier reviews reached seemingly conflicting conclusions. The Results section on near transfer (page unavailable) reports that effect sizes vary across studies, but does not establish that previous meta-analyses disagreed particularly about the size of near-transfer effects. (LLM, conf=0.79)

      Experimental — accuracy not yet measured
    • cites Schwaighofer et al., 2015 source
  • Findingcitation

    In a reanalysis of data from Melby-Lervåg et al. (2016), the pooled effect size for all types of working memory tasks was g = 0.29, 95% CI [0.18, 0.40], p < 0.001.

    “The pooled effect size for all types of WM tasks was g = 0.29 [0.18, 0.40] p < 0.001.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Melby-Lervåg et al. (2016) source

      The Results section on immediate intermediate transfer reports significant improvements on verbal and visuo-spatial working-memory measures. But the Method section says treated and untreated controls were analyzed separately, and the passages do not report a pooled estimate across all working-memory tasks of g = 0.29, 95% CI [0.18, 0.40], p < 0.001. That reanalysis result cannot be verified from this paper's text. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • Premisecitation

    N-back tasks are highly correlated with other working memory tasks at a latent level.

    “a recent study showed that n-back tasks are in fact highly correlated with other WM tasks at a latent level (Schmiedek et al., 2014).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Schmiedek et al., 2014 source
  • Classified as probably not a substantial claim. 22
  • May combine two claims. 27

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The concurrent validity of the n-back task has been questioned because previous studies reported low correlations between n-back and other WM tasks, especially complex span tasks.

    “The concurrent validity of the n-back task, however, has been questioned as previous studies have reported low correlations between the n-back task and other WM tasks, especially complex span tasks”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Miller, Price, Okun, Montijo, & Bowers, 2009 source
    • cites Redick & Lindsey, 2013 source

  • May combine two claims.

    Premisecitation

    Previous meta-analyses found that WM training produces small to large short-term near transfer effects and null to small transfer effects to verbal and visuospatial reasoning/Gf.

    “The results from these meta-analyses (Table 1) with both clinical and healthy samples of children and adults showed that WM training can produce small to large short-term near transfer effects (verbal and visuospatial WM tasks) and null to small effects of transfer to verbal and visuospatial reasoning (or fluid intelligence; Gf) tasks.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Melby-Lervåg & Hulme, 2013 source
    • cites Melby-Lervåg et al., 2016 source
    • cites Weicker et al., 2016 source

  • May combine two claims.

    Interpretationown data

    The transfer effect to untrained working memory tasks as a whole was approximately the same size as in Melby-Lervåg et al. (2016) but considerably smaller than in other meta-analyses not focusing specifically on n-back training.

    “The transfer effect to untrained WM tasks as a whole is approximately of the same size as in the n-back training analyses by Melby-Lervåg et al. (2016; when averaged across verbal and nonverbal WM and active and passive control groups) but considerably smaller than in the other metaanalyses not focusing specifically on n-back training”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Melby-Lervåg et al. (2016) source

      The Results section on immediate intermediate transfer reports moderate effects on verbal and visuospatial working-memory measures, and the Methods section lists n-back as a training category. The passages do not give the n-back-specific estimates needed to verify the claimed average across task types and control groups. The comparison with the citing paper’s own effect cannot be checked from this source alone. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
    • cites Melby-Lervåg & Hulme, 2013 source
    • cites Weicker et al., 2016 source
    • The authors state this as an assumption: Missing pre-post correlations can be set to zero, treating pre- and posttest measures as independent groups.

  • May combine two claims.

    Findingown data

    In the reanalysis, untrained versions of the training tasks showed a significantly stronger effect, g = 0.62, 95% CI [0.38, 0.86], p < 0.001, than other working memory tasks, g = 0.21, 95% CI [0.08, 0.33], p < 0.01, with a significant difference, QM(1) = 8.89, p < 0.01.

    “the effect size was significantly stronger, Q M (1) = 8.89, p < 0.01, for the former, g = 0.62 [0.38, 0.86] p < 0.001 (number of effect sizes = 19), than the latter, g = 0.21, [0.08, 0.33] p < 0.01 (number of effect sizes = 70).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Melby-Lervåg et al. (2016) source

      Split into 2 atoms — 0 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    P-curve and variations of Egger's regression have been found to perform better than the trim-and-fill method for detecting publication bias.

    “However, more novel and promising methods, such as p-curve (Simonsohn, Nelson, & Simmons, 2014) and variations of Egger's regression (Egger, Smith, Schneider, & Minder, 1997;Moreno et al., 2009), have been found to perform better than the trim and fill method.”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites Simonsohn et al., 2014 source
    • cites Egger et al., 1997 source
    • cites Moreno et al., 2009 source

      The Abstract (Methods and Results) says several regression-based methods outperformed Trim & Fill, and the Conclusion identifies an Egger variant among the promising methods. But the comparison concerns adjustment for publication bias, not performance in detecting it. The sentence cites Simonsohn et al., rather than Moreno et al., for p-curve. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    This finding is in line with previous meta-analyses.

    “On the whole, this is in line with the previous meta-analyses (Melby-Lervåg & Hulme, 2013;Schwaighofer et al., 2015;Melby-Lervåg et al., 2016).”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites Melby-Lervåg & Hulme, 2013 source
    • cites Schwaighofer et al., 2015 source
    • cites Melby-Lervåg et al., 2016 source

  • May combine two claims.

    Findingown data

    Transfer effects to other WM tasks, Gf, and cognitive control were very small and of similar size.

    “For other WM tasks, Gf, and cognitive control, the effect sizes were of similar size and very small.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The meta-analysed pre-post correlations were r = 0.659 for the training group, r = 0.649 for the active control group, and r = 0.656 for the passive control groups.

    “A meta-analysis of the prepost correlations showed that the correlations were r = 0.659 (k = 203, p < 0.001) for the pre-post measures in the training group, r = 0.649 (k = 94, p < 0.001) for the pre-post measures in the active control group, and r = 0.656 (k = 109, p < 0.001) for the pre-post measures in the passive control groups.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missing pre-post correlations can be set to zero, treating pre- and posttest measures as independent groups.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The 507 within-group effect sizes produced 305 between-groups effect sizes.

    “These 507 within-group effect sizes yielded 305 betweengroups effect sizes (i.e., comparisons between a training group and a control group within an experiment).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The three-level model was more adequate than the single-level and two-level models, with significant likelihood ratio tests (LRT = 24.34 and LRT = 12.34, both p < 0.001).

    “The test between the single-and two-level models was significant (LRT = 24.34,p < 0.001), and so was the test between the two-and three level models (LRT = 12.34, p < 0.001).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Thirteen of the original 203 effect sizes were excluded.

    “We thus excluded 13 of the originally 203 effect sizes”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The meta-analysis included 203 training effects, reduced to 190 after data screening, drawn from 33 studies, 41 experiments, and 2,105 participants.

    “The present meta-analysis included 203 training effects (190 after data screening) from 33 studies.These studies consisted of 41 unique experiments.In total, data were obtained from 2,105 individuals.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Hours of training had no effect on training outcomes, QM(1) = 0.03, g = 0.01, 95% CI [-0.12, 0.14], p = 0.872, and with cognitive domain as a covariate, g = 0.00, 95% CI [-0.12, 0.12], p = 0.981.

    “Neither did we find any effect of hours of training, Q M (1) = 0.03, g = 0.01 [-0.12, 0.14], p = 0.872, and g = g = 0.00 [-0.12, 0.12], p = 0.981 with cognitive domain as a covariate.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Sample age did not moderate the training effect, QM(1) = 0.07, g = 0.03, 95% CI [-0.17, 0.22], p = 0.794, and with cognitive domain as a covariate, g = 0.02, 95% CI [-0.16, 0.20], p = 0.834.

    “We found no effect of age, Q M (1) = 0.07, g = 0.03 [-0.17, 0.22], p = 0.794.Adding cognitive domain as a covariate to the analysis did not alter the effect, g = 0.02 [-0.16, 0.20], p = 0.834.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The number of training sessions had no effect on training outcomes, QM(1) = 0.02, g = -0.01, 95% CI [-0.13, 0.12], p = 0.902, and with cognitive domain as a covariate, g = -0.03, 95% CI [-0.14, 0.09], p = 0.685.

    “We did not find any effect of the number of sessions, Q M (1) = 0.02, g = -0.01[-0.13, 0.12], p = 0.902, and g = -0.03[-0.14, 0.09], p = 0.685, with cognitive domain as a covariate.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    After controlling for control group type, the fluid intelligence training effect was g = 0.11, p > 0.05.

    “For Gf the effect size was g = 0.11, p > 0.05.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In this meta-analysis, the only notable transfer effect was to untrained n-back tasks; transfer to other tasks was similarly very small.

    “However, in the present study, the only noteworthy transfer effect was seen to untrained n-back tasks, while transfer to other tasks was at similar, very small levels.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the funnel plots of residuals, asymmetry was more apparent for cognitive control and fluid intelligence, with a group of high-SE effect sizes showing higher effects than the aggregated domain effect.

    “In these funnel plots, asymmetry is a little more apparent.For cognitive control and Gf, there is a group of effect sizes with high SE that show higher effect sizes than the aggregated effect size for that domain.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The sample comprised 33 studies, 41 experiments, and 103 groups; 54 groups were training and 49 were control; 23 training groups used single n-back, 30 used dual n-back, and 1 used triple n-back; 20 control groups were active and 29 were passive.

    “The study sample consisted of 33 studies, including 41 experiments, with, in total, 103 experiment groups.Of these groups, 54 (52.4%) were training groups and 49 (47.6%) were control groups.Twenty-three (42.6%) of the training groups practiced with a single n-back task, 30 (55.6%) with a dual n-back task, and 1 (1.9%) with a triple n-back task.Twenty (40.8%) of the control groups were active, and”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Although there was some asymmetry in the contour-enhanced funnel plots, there was no obvious pattern of statistically significant effect sizes being preferred over nonsignificant effect sizes.

    “While there was some asymmetry, there was no obvious pattern of statistically significant effect sizes being preferred over nonsignificant effect sizes.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    This meta-analysis found moderate transfer to untrained n-back variants and weak transfer to working memory, cognitive control, and fluid intelligence.

    “The present meta-analysis on the efficacy of n-back training shows medium transfer effects to untrained versions of the trained n-back tasks and small transfer effects to other WM tasks, cognitive control, and Gf.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The transfer tasks categorized as measuring working memory, fluid intelligence, or cognitive control validly measure those constructs.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Previous studies comparing single and dual n-back training have shown both versions are equally effective in producing generalization and have fairly similar transfer effects.

    “Previous studies comparing the effects of single and dual nback training have shown that both task versions seem to be equally effective in producing generalization and that the transfer effects are fairly similar”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Jaeggi et al., 2010 source
    • cites Jaeggi, Buschkuehl, Shah, & Jonides, 2014 source

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The present meta-analysis used a multi-level approach to analyze cognitive transfer effects in healthy adults trained with n-back tasks.

    “We conducted a multi-level meta-analysis on the cognitive transfer effects in healthy adults who have been administered WM updating training with n-back tasks”
    Quote in the paperSubstantial claim
    • Rests on an assumption: n-back tasks are the most commonly used computerized experimental WM training paradigm.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A recent study showed that at a latent level, the n-back task is highly correlated with other WM tasks (complex spans, WM updating tasks, sorting spans).

    “Nevertheless, a recent study showed that at a latent level, the n-back task is highly correlated with other WM tasks (complex spans, WM updating tasks, and sorting spans; Schmiedek, Lövdén, & Lindenberger, 2014).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Schmiedek, Lövdén, & Lindenberger, 2014

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    Limiting sources of variability (training method, age, health status) should yield results easier to interpret.

    “First, as mentioned, limiting the sources of variability (here training method, age, and health status) should yield results that are easier to interpret.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Simple span tasks were considered WM measures in Melby-Lervåg and Hulme (2013), whereas other meta-analyses coded them as STM or excluded them.

    “Also, simple span tasks, such as the digit span, were considered WM measures in the Melby-Lervåg and Hulme (2013) study, whereas the other meta-analyses coded simple spans as measures of short-term memory (STM; Schwaighofer et al., 2015;Weicker et al., 2016) or excluded them completely from the analyses (Melby-Lervåg et al., 2016).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Melby-Lervåg & Hulme, 2013 source
    • cites Schwaighofer et al., 2015 source
    • cites Weicker et al., 2016 source

  • Classified as probably not a substantial claim.

    Premisecitation

    Cognitive control was included as a transfer domain because earlier studies suggest WM is closely related to other executive functions.

    “Cognitive control was included as a transfer domain because earlier studies suggest that WM is closely related to other executive functions (Miyake et al., 2000).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Miyake et al., 2000 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The coding of transfer tasks into cognitive domains and materials was based on consensus decisions by three authors.

    “The coding of transfer tasks into cognitive domains (i.e., task-specific transfer as measured by n-back, WM, cognitive control, and Gf), and materials (i.e., verbal vs. nonverbal), was based on consensus decisions by three of the authors (A.S., M.L., and L.K.).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    The present meta-analysis compared active and passive controls only within studies that employed both.

    “Following the current discussion on the effects of the type of control group (Au et al., 2016), we compared active and passive controls only within studies (comparing active and passive control groups within those studies that have employed both).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Au et al., 2016

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Study design and conductreasoning

    The present meta-analysis used a multilevel meta-analytical approach to include effect sizes for all pertinent tasks while accounting for dependency from same sample.

    “Therefore, we made use of a multilevel meta-analytical approach to include effect sizes for all pertinent tasks from the original studies while accounting for the fact that these effect sizes came from the same sample.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    Coding of group sizes, means, and standard deviations was reliable, with interrater agreement of 97.1% for group sizes, 100.0% for means, and 99.6% for standard deviations across ten randomly selected studies.

    “The interrater agreement was 97.1% for the 34 group size values, 100.0%for the 284 mean vales, and 99.6% for the 284 standard deviation values in the ten studies.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductreasoning

    The present meta-analysis included all measures from original studies by using a multilevel approach, reducing risk of biasing the estimated training effect.

    “In the present meta-analysis, we were able to include all measures from the original studies by employing a multilevel metaanalytical approach, and thereby having less risk of biasing the estimated training effect.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    The present meta-analysis used type of control group as a covariate in analysis of training effect.

    “We also used the type of control group as a covariate in our analysis of the training effect, thereby considering the potential influence of the type of control group used in an experiment.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Melby-Lervåg and Hulme (2013), who reported the highest near-transfer effects, in some cases included the training task in their analyses, which overestimated the near-transfer effect.

    “First, Melby-Lervåg & Hulme (2013) who reported the highest effects of near transfer, in some cases included the training task in their analyses, resulting in an overestimation of the near transfer effect.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Melby-Lervåg & Hulme (2013) source

  • May combine two claims.

    Interpretationown data

    The main analyses showed a moderate effect of task-specific transfer to untrained n-back tasks and very small transfer effects to other untrained working memory measures, cognitive control, and fluid intelligence.

    “The results from the main analyses showed a moderate effect of task-specific transfer to untrained n-back tasks and very small transfer effects to other untrained WM measures, cognitive control, and Gf.”
    • Also stated in: § Main analyses · plain
      “In sum, in the present meta-analysis, the only notable transfer effect is seen to untrained n-back tasks.Despite the fact that the transfer effects to the other domains are also statistically significant, they can be considered very small.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Missing pre-post correlations can be set to zero, treating pre- and posttest measures as independent groups.

  • Classified as probably not a substantial claim.

    Premisecitation

    Simple and complex spans can be considered measures of the same cognitive processes.

    “it has been argued that simple and complex spans (the latter ones being included both in previous metaanalyses and the present one) can in fact be considered as measures of the same cognitive processes (Unsworth & Engle, 2007).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Unsworth & Engle, 2007 source

  • May combine two claims.

    Interpretationreasoning

    A failure to distinguish task-specific from other working memory transfer inflates effect sizes for near transfer and may mislead theoretical explanations of working memory training effects.

    “A failure to make a distinction between task-specific and other WM transfer not only inflates effect sizes for near transfer, but may even lead one to a wrong track when searching for theoretical explanations for WM training effects (i.e., assuming that WM training increases the effectiveness of WM in general, rather than considering also alternative hypotheses on strategy-based effects).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationown data

    This finding is in line with most previous meta-analyses.

    “again in line with most of the previous meta-analyses”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationcitation

    Working memory training is still an unfinished endeavor.

    “We concur with what Shipstead, Hicks, and Engle (2012) stated 4 years ago: BWorking memory training remains a work in progress.^”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Shipstead, Hicks, and Engle (2012) source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Weicker et al. (2016) reported a positive association between number of training sessions and working memory transfer effect size.

    “However, Weicker et al. (2016) found that the number of training sessions was positively related to the size of the transfer effect to WM.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Weicker et al. (2016) source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    There is currently no theory of the processes involved in training-induced change in working memory.

    “Furthermore, future studies should pave a way to a theory of the processes involved in a training-induced change in WM, a theory that is currently missing (Gibson, Gondoli, Johnson, Steeger, & Morrissey, 2012).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Gibson, Gondoli, Johnson, Steeger, & Morrissey, 2012 source

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    The simplifications limit each experiment to one single comparison between a training group and a control group per measure, eliminating dependency between comparisons within an experiment that would reuse training or control groups.

    “These simplifications limit each experiment to one single comparison between a training group and a control group per measure.This eliminates dependency between comparisons within an experiment that would reuse training or control groups.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    This meta-analysis focused on n-back training in healthy adults, whereas the Weicker et al. (2016) meta-analysis included studies of all kinds of working memory training paradigms in healthy and clinical samples of children and adults.

    “While the present meta-analysis focused on n-back training in healthy adults, the Weicker et al. (2016) meta-analysis included studies investigating all kinds of WM training paradigms in both healthy and clinical samples of children and adults.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Weicker et al. (2016) source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    The authors could not perform all moderator analyses for all four cognitive domains and refrained from investigating interactions between moderators.

    “For the same reason, we could not perform all the moderator analyses for all four cognitive domains (e.g., material-specific aspects of transfer were investigated only for WM and Gf), and we refrained from investigating interactions between moderators.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationown data

    This finding is in line with the Au et al. (2015) meta-analysis.

    “in line with the Au et al. (2015) meta-analysis”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Au et al. (2015)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Overly conservative estimates for standard errors cannot be included in analyses where the standard errors are of primary interest, such as publication bias analysis.

    “overly conservative estimates for the standard errors cannot be included in analyses where the standard errors are of primary interest, such as a publication bias analysis.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missing pre-post correlations can be set to zero, treating pre- and posttest measures as independent groups.
46 of 46

  • The paper relies on this without support.

    n-back tasks are the most commonly used computerized experimental WM training paradigm.

    “the most commonly used computerized experimental WM training paradigm, namely with n-back tasks”

  • The paper relies on this without support.

    The n-back task is one of the most commonly used working memory training paradigms.

    “We examined the efficacy of one of the most commonly used WM training paradigms, namely the n-back task.”

  • The authors state this as an assumption.

    Experiments within the same study can be treated as independent samples.

    “for the sake of parsimony we did not model the dependency between experiments within the same study, but treated all 41 experiments as independent samples”

  • The authors state this as an assumption.

    Missing pre-post correlations can be set to zero, treating pre- and posttest measures as independent groups.

    “Where we were unable to estimate the correlation, we entered 0 as a correlation, essentially treating the pre-and posttest measures as independent groups.”

  • The paper relies on this without support.

    Task-specific improvement can enhance performance only on tasks with a similar structure where the same strategies can be successfully employed.

    “This is because taskspecific improvement can enhance performance only on tasks with a similar structure where the same strategies can be succesfully employed.”

  • The paper relies on this without support.

    The transfer tasks categorized as measuring working memory, fluid intelligence, or cognitive control validly measure those constructs.

    “Transfer effects to other tasks (whether they measure WM, Gf, or cognitive control), albeit being observable, are small and apparently of little practical significance.”
6 of 6

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Three of the included studies had participants less than 18 years of age (Soveri et al. 2017; Vermeir et al. 2020; Gathercole et al. 2019), and all studies had participants who were older than 55 years of age.

    “Three of the included studies had participants less than 18 years of age (Soveri et al. 2017; Vermeir et al. 2020; Gathercole et al. 2019), and all studies had participants who were older than 55 years of age.”

    From Examining Working Memory Training for Healthy Adults—A Second-Order Meta-Analysis (2024) · cites this paper as Soveri et al. 2017

    Split into 2 atoms — 0 confirmed, 0 partially supported, 2 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

    Checked by a reasoning model — may be wrong.

    Experimental — accuracy not yet measured

  • Meta-analyses have reported persistent critiques of similarity-based theories' explanatory limitations.

    “persistent critiques of these theories’ explanatory limitations have been reported by meta-analyses (Soveri et al., 2017a; You et al., 2021).”

    From The asymmetry of working memory training transfer: A systematic review and meta-analysis (2026) · cites this paper as Soveri et al., 2017a

    In “Mechanisms behind transfer” (page unavailable), Soveri et al. report that transfer was substantial mainly to untrained n-back tasks and suggest task-specific strategies rather than broadly improved working memory as an explanation. “Directions for future studies” cautions that overlooking task-specific effects can mislead theoretical explanations. These passages support a critique of one explanation for transfer, but they do not report persistent critiques of similarity-based theories as such. (LLM, conf=0.74)

    Experimental — accuracy not yet measured

  • Multiple studies suggest a small, measurable, positive effect of working memory training on working memory capacity.

    “Multiple studies suggest a small, measurable, positive effect of training on WM capacity”

    From Examining Working Memory Training for Healthy Adults—A Second-Order Meta-Analysis (2024) · cites this paper as Soveri et al. 2017

    The Main analyses and Conclusions (page unavailable) report small, statistically significant positive effects on untrained working-memory tasks across 33 n-back training studies. But the Mechanisms behind transfer section says the pattern mainly suggests task-specific improvement rather than an increase in underlying working-memory capacity. The measured task effects therefore support only part of the claim. (LLM, conf=0.86)

    Experimental — accuracy not yet measured

  • In working memory training studies, near transfer effects are easier to observe and larger than far transfer effects.

    “several meta-analyses of working memory training studies have demonstrated that near transfer effects are both more readily observable and show larger effect sizes than far transfer effects13,14.”

    From Near transfer to an unrelated N-back task mediates the effect of N-back working memory training on matrix reasoning (2022) · cites this paper as [13]

    The Training effects by cognitive domain section (page unavailable) reports a much larger effect for untrained n-back tasks than for far-transfer domains, and the Main analyses call it the only notable transfer effect. But effects on other working-memory tasks were not significantly different from those on the other domains, and effects in all domains were statistically significant. The claim therefore overgeneralizes the near-versus-far contrast and its suggestion that far transfer is less readily observable. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • Transfer to untrained N-back tasks is often not reported, so information about what participants learned is often missing.

    “Transfer to untrained N-back tasks is often not reported therefore information is often missing about what participants learned, and thus, the extent of learning on measures of transfer remains unknown.”

    From Near transfer to an unrelated N-back task mediates the effect of N-back working memory training on matrix reasoning (2022) · cites this paper as [13]

    The Directions for future studies section says that closely related tasks should be analyzed systematically to clarify what drives transfer effects. But the paper does not say that transfer to untrained n-back tasks is often unreported; its Main analyses report a moderate transfer effect on those tasks. The claimed reason that information about learning is often missing is therefore not established by this source. (LLM, conf=0.78)

    Experimental — accuracy not yet measured

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • The abstract calls all 33 trials published despite including an unpublished study

    Minor

    The abstract describes the 33 included trials as published, whereas the methods explicitly say an unpublished study was included.

  • The number of studies with multiple experiments changes

    Minor

    The paper first says six studies contained multiple experiments, but later refers to four such studies. Its intervening breakdown identifies only five-one with four experiments and four with two-leaving the stated total of 41 experiments unexplained.

2 of 2

Cites

Flagged first, then by citations

Research works printed but not linked: 4

  1. *Heffernan, A. E. (2014). The generalizability of dual n-back training in younger adults (Master’s thesis). Retrieved from http://hdl.handle.net/10222/56021
  2. Lezak, M. D., Howieson, D. B., & Loring, D. W. (2004). Neuropsychological Assessment. New York: Oxford University Press.
  3. NIMH Research Domain Criteria Workshop (2010). Working memory: Workshop proceedings. Downloaded November 8, 2016 from https://www.nimh.nih.gov/research-priorities/rdoc/working-memory-workshop-proceedings.shtml
  4. *Prandl, A. (2012). Visual memory improvement in recognition (Bachelor’s thesis). Retrieved from http://ro.ecu.edu.au/theses_hons/59/

Cited by

Sorted by citations

Related

Authors
Soveri, A., Antfolk, J., Karlsson, L., Salo, B., & Laine, M.
Affiliations
Åbo Akademi University (FI)
Corresponding author
Soveri, Anna
ORCID
Venue
Psychonomic Bulletin & Review
Publisher
Springer Science and Business Media LLC
Published
2017-01-23
Volume
24
Issue
4
Pages
1077-1096
ISSN
  • 1069-9384 (ISSN-L)
  • 1531-5320
DOI
10.3758/s13423-016-1217-0
OpenAlex
W2582113845
Cited by
350
References
83
Citations by year
2026: 20 · 2025: 29 · 2024: 45 · 2023: 28 · 2022: 38 · 2021: 50 · 2020: 47 · 2019: 48 · 2018: 37 · 2017: 13
FWCI
7.53
Type
journal article
Language
en
Open access
Licence
www.springer.com/tdm (unspecified)
Locations
  • Psychonomic Bulletin & Review (publishedVersion)
  • PubMed (publishedVersion)
  • Åbo Akademi University Research Portal (submittedVersion)
  • PsyArXiv (OSF Preprints) (submittedVersion)
  • PsyArXiv (OSF Preprints) (submittedVersion, cc-by)
Topics
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Cognitive Abilities and Testing
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Cognitive Functions and Memory
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neural and Behavioral Psychology Studies
Keywords
working memory training, n-back task, transfer of learning, fluid intelligence, cognitive control, training dose, dual-task training
MeSH terms
Cognition, Humans, Learning, Memory, Short-Term, Executive Function

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-08.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-08.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review