CoChat
Sign Up
Cited by 64Cognitive Abilities and TestingAging and Gerontology ResearchCognitive Functions and Memory

Working memory training does not enhance older adults' cognitive skills: A comprehensive meta-analysis

Open access
Sala, Giovanni, Aksayli, N. Deniz, Tatlidil, K. Semir, Gondo, Yasuyuki, Gobet, Fernand·2019·Intelligence
DOI 10.1016/j.intell.2019.101386

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

In the last two decades, considerable efforts have been devoted to finding a way to enhance cognitive function by cognitive training. To date, the attempt to boost broad cognitive functions in the general population has failed. However, it is still possible that some cognitive training regimens exert a positive influence on specific populations, such as older adults. In this meta-analytic review, we investigated the effects of working memory (WM) training on older adults' cognitive skills. Three robust-variance-estimation meta-analyses (N = 2140, m = 43, and k = 698) were run to analyze the effects of the intervention on (a) the trained tasks, (b) near-transfer measures, and (c) far-transfer measures. While large effects were found for the trained tasks (g¯ = 0.877), only modest (g¯ = 0.274) and near-zero (g¯ = 0.121) effects were obtained in the near-transfer and far-transfer meta-analyses, respectively. Publication-bias analysis provided adjusted estimates that were slightly lower. Moreover, when active control groups were implemented, the far-transfer effects were null (g¯ = −0.008). Finally, the effects were highly consistent across studies (i.e., low or null true heterogeneity), especially in the near- and far-transfer models. While confirming the difficulty in obtaining transfer effects with cognitive training, these results corroborate recent empirical evidence suggesting that WM is not isomorphic with other fundamental cognitive skills such as fluid intelligence.

AI snapshot

Written from the paper’s full text.

TL;DR

This comprehensive meta-analysis of 43 studies (N = 2140) found that working memory training in older adults produced large gains on trained tasks but only modest near-transfer and near-zero far-transfer effects. With active control groups, far-transfer effects were null, leading the authors to conclude that WM training does not enhance older adults’ broader cognitive skills and that WM is not isomorphic with fluid intelligence.

Key findings

  • Trained tasks showed large effects: g = 0.877, 95% CI [0.691; 1.063], m = 28, k = 72, p < .001; publication-bias adjusted estimates ranged from 0.479 to 0.544.
  • Near-transfer effects were modest: g = 0.274, 95% CI [0.192; 0.355], m = 39, k = 214, p < .001; adjusted estimates ranged from 0.159 to 0.246. Nearer-transfer effects (g = 0.345) were significantly greater than near-transfer (g = 0.225) and less-near-transfer (g = 0.191) effects.
  • Far-transfer effects were near zero: g = 0.121, 95% CI [0.032; 0.211], m = 38, k = 248, p = .011. With non-active controls, g = 0.262; with active controls, g = -0.008, 95% CI [-0.105; 0.090], p = .854.
  • Re-analysis of younger adults found criterion g = 1.170, near-transfer g = 0.208, and far-transfer g = 0.099; active-control far-transfer was g = 0.059, p = .170.

Methods

  • Systematic PRISMA search through June 1, 2019, across multiple databases plus reference lists and author requests; included 43 studies from 2008–2019, N = 2140, with 534 post-test effect sizes and 164 follow-up effect sizes. Inclusion required mean age ≥ 60, no serious clinical condition, a control group, at least one transfer measure, and WM tasks constituting ≥ 50% of intervention.
  • Three robust-variance-estimation (RVE) meta-analyses were run for criterion, near-transfer, and far-transfer outcomes. Moderators included allocation, control-group type, training duration, and baseline differences; sensitivity analyses included influential-case removal, Cheung and Chan merging, trim-and-fill, and PET-PEESE publication-bias correction.
  • A separate re-analysis of Melby-Lervåg et al. (2016) compared younger adults using the same effect-size categorization and modeling approach.

Limitations

  • The authors state that most primary studies did not report reliability coefficients, so no correction for measurement error was applied; they argue the practical consequences are minimal, especially for near- and far-transfer models.
  • The authors state that follow-up estimates are unreliable due to selection bias: only 15 of 39 near-transfer studies and 13 of 38 far-transfer studies reported follow-up measures. They also note PET-PEESE can be unreliable with fewer than 20 observations, high true heterogeneity, or only small-sample studies.
  • Evident from the text: inclusion was restricted to healthy older adults with mean age ≥ 60 and no serious clinical condition, so findings may not generalize to clinical populations; some subgroup analyses had small numbers and limited power, and the active-control criterion led the authors to recode some primary-study “active” controls as non-active.

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

Integrity signals

Registry

Crossref · 10.1016/j.intell.2019.101386

Cites

90 of 90 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

9 of 27 cited sources checked · 3 supported · 1 flagged · accuracy not yet measured

  • Interpretationcitation

    Theories predicting far transfer are not supported.

    “Conversely, those theories predicting far transfer are not supported (e.g., Jaeggi et al., 2008;Taatgen, 2013Taatgen, , 2016)).”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites Jaeggi et al., 2008 source

      The Abstract says the authors found transfer from working-memory training to fluid-intelligence measures, despite the trained task being different from the intelligence test. The Results also report greater gains in the trained groups than in controls. This is contrary to citing the paper as evidence that far-transfer predictions are not supported. (LLM, conf=0.90)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured
    • cites Taatgen, 2013 source

      In ‘Predictive Power and Studies That Do Not Show Transfer’ (p. 30), Taatgen says the theory has not yet proved its predictive power because it has modeled existing data rather than made predictions. But the paper also describes successful far-transfer studies (pp. 3, 30) and proposes conditions under which far transfer can occur. It does not support the categorical claim that theories predicting far transfer are unsupported. (LLM, conf=0.82)

      Experimental — accuracy not yet measured
    • cites Taatgen, 2016 source
  • Interpretationcitation

    These results are in line with the prediction of theories of skill acquisition based on task domain-specificity.

    “These results are in line with the prediction of theories of skill acquisition based on task domain-specificity (Chase & Ericsson, 1982;Gathercole, Dunning, Holmes, & Norris, 2019;Gobet, 2016;Gobet & Simon, 1996).”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites Chase & Ericsson, 1982 source
    • cites Gathercole, Dunning, Holmes, & Norris, 2019 source
    • cites Gobet, 2016

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

  • Interpretationcitation

    Small near-transfer effects can be accounted for by assuming that participants learn how to carry out a particular class of memory tasks rather than enhance their memory capacity.

    “Rather, such small effects can be easily accounted for by assuming that participants learn how to carry out a particular class of memory tasks rather than enhance their memory capacity (Gathercole et al., 2019;Melby-Lervåg et al., 2016;Sala & Gobet, 2019;Shipstead, Redick, & Engle, 2012).”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites Gathercole et al., 2019 source

      The General discussion (page unavailable) attributes substantial transfer between similar tasks to learning new cognitive routines, rather than an increase in general memory capacity. But The cognitive routine framework distinguishes small transfer effects from this account: it suggests they may arise from fine-tuning established processes, while routine-mediated learning explains moderate to large effects. The citation therefore supports the task-learning explanation, but not its application to small effects generally. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites Melby-Lervåg et al., 2016 source

      The Discussion’s Theoretical Implications section says task- or stimulus-specific strategies likely explain much of the improvement on intermediate and near-transfer tasks, and that the gains may not reflect increased working-memory capacity. But the Results section describes near-transfer effects as large, not small. No page numbers are available in our copy. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites Sala & Gobet, 2019

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

  • Findingown data

    Working memory training produces near-zero far-transfer effects in older adults (g = 0.121).

    “only modest (g = 0.274) and near-zero (g = 0.121) effects were obtained in the near-transfer and far-transfer metaanalyses, respectively.”
    Quote in the paperSubstantial claim
  • Findingown data

    When active control groups were implemented, the far-transfer effects of working memory training in older adults were null (g = -0.008).

    “Moreover, when active control groups were implemented, the far-transfer effects were null (g = -0.008).”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 21
  • May combine two claims. 24
  • The quote only partly states this claim. 18

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

    Premisecitation

    Working memory training impacts the ability to perform the trained tasks and, to a lesser degree, near-transfer tasks.

    “Four comprehensive meta-analyses (Melby-Lervåg & Hulme, 2013;Melby-Lervåg, Redick, & Hulme, 2016;Schwaighofer, Fischer, & Buhner, 2015;Weicker, Villringer, & Thöne-Otto, 2016) have found that WM training impacts on the ability to perform the trained tasks and, to a lesser degree, tasks similar to the trained tasks (i.e., near transfer).”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Melby-Lervåg & Hulme, 2013 source
    • cites Melby-Lervåg, Redick, & Hulme, 2016 source

      The Results section on near transfer reports large effects on tasks similar or identical to the trained tasks. However, the paper groups those tasks together; it does not establish that effects on similar tasks were smaller than effects on the trained tasks. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
    • cites Schwaighofer, Fischer, & Buhner, 2015 source

  • May combine two claims.

    Premisecitation

    Working memory training has little or no effect on far-transfer measures, especially when treated groups are compared to active control groups.

    “However, WM training appears to have little or no effect on cognitive tests unrelated to the trained tasks (i.e., far transfer), especially when treated groups were compared to active control groups to rule out possible placebo effects (e.g., Melby-Lervåg et al., 2016).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Melby-Lervåg et al., 2016 source

  • The quote only partly states this claim.

    Findingown data

    In the re-analysis of Melby-Lervåg et al. (2016) criterion meta-analysis of younger adults, the RVE overall effect size was g = 1.170, 95% CI [0.713; 1.627], m = 12, k = 34, df = 5.99, p < .001, ω2 = 0.061, τ2 = 0.392.

    “The overall effect size was g = 1.170, 95% CI [0.713; 1.627], m = 12, k = 34, df = 5.99, p < .001,ω 2 = 0.061, τ 2 = 0.392.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Melby-Lervåg et al. (2016) source

      The Method (Moderator Variables; page unavailable) confirms that Melby-Lervåg et al. analyzed studies of adults aged 18–64, so the paper can identify the meta-analysis used for a younger-adult re-analysis. Its Results (Effects of Working Memory Training on Near Transfer Measures; page unavailable) discuss training effects, but our copy does not report the quoted RVE estimate, confidence interval, study counts, or variance statistics. Those figures cannot be attributed to the cited paper as its own results. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

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

    Premisecitation

    A set of several measures is necessary to investigate the effects of working memory training on latent factors representing cognitive skills rather than single tasks.

    “A set of several measures is necessary to investigate the effects of WM training on latent factors representing cognitive skills rather than single tasks (Noack et al., 2009;Schmiedek, Lövdén, & Lindenberger, 2010;Shipstead et al., 2012).”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites Noack et al., 2009 source
    • cites Schmiedek, Lövdén, & Lindenberger, 2010

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

    • cites Shipstead et al., 2012 source

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

    Findingown data

    Publication-bias analysis provided adjusted estimates that were slightly lower than unadjusted estimates.

    “Publication-bias analysis provided adjusted estimates that were slightly lower.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The present meta-analysis included 43 studies, 534 effect sizes, and 2140 participants.

    “We found 43 studies, conducted from 2008 to 2019, that met the inclusion criteria.These studies included 534 effect sizes and a total of 2140 participants.”
    Quote in the paperSubstantial claim

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

    Findingown data

    A subsample of the included studies reported follow-up effects (k = 164).

    “Also, a subsample of the included studies reported follow-up effects (k = 164).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In the near-transfer meta-analysis of older adults, Baseline difference was the only significant moderator (b = -0.416, p < .001) and explained nearly all the true heterogeneity (ω2 = 0.000, τ2 = 0.006).

    “Baseline difference was the only significant moderator (b = -0.416,p < .001)and explained nearly all the true heterogeneity (ω 2 = 0.000, τ 2 = 0.006).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the criterion sensitivity meta-regression of older adults, Baseline was a significant moderator and explained a significant amount of true heterogeneity (b = -0.528, p = .008; ω2 = 0.016, τ2 = 0.019).

    “The meta-regression analysis showed that, this time, Baseline was a significant moderator and explained a significant amount of true heterogeneity (b = -0.528,p = .008;ω 2 = 0.016, τ 2 = 0.019).”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the far-transfer follow-up analysis of older adults, the RVE overall effect size was g = 0.241, 95% CI [0.086; 0.396], m = 13, k = 70, df = 7.92, p = .007, ω2 = 0.000, τ2 = 0.002.

    “The RVE overall effect size was g = 0.241, 95% CI [0.086; 0.396], m = 13, k = 70, df = 7.92, p = .007,ω 2 = 0.000, τ 2 = 0.002.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the far-transfer meta-analysis of older adults, a modest positive effect (g = 0.262) was observed in studies implementing non-active control groups, while with active-control groups the effect was practically zero (g = -0.008).

    “While a modest positive effect (g = 0.262) was observed in studies implementing non-active control groups, when active-control groups were employed the effect was practically zero (g = -0.008).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the near-transfer meta-analysis of older adults, nearer-transfer effect sizes (g = 0.345) were significantly greater than near-transfer effects (p = .012) and less-near-transfer measures (p = .043).

    “but the nearer-transfer effect sizes (g = 0.345) were significantly greater than near-transfer effects (p = .012)and less-near-transfer measures (p = .043).”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the near-transfer follow-up analysis of older adults, the RVE overall effect size was g = 0.378, 95% CI [0.252; 0.504], m = 15, k = 77, df = 9.93, p < .001, ω2 = 0.000, τ2 = 0.000.

    “The RVE overall effect size was g = 0.378, 95% CI [0.252; 0.504], m = 15, k = 77, df = 9.93, p < .001,ω 2 = 0.000, τ 2 = 0.000.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    In the younger-adult near-transfer publication-bias analysis, the PEESE estimator was g = 0.169, 95% CI [0.095; 0.242], p < .001.

    “The PET and PEESE estimators were g = 0.165, 95% CI [0.060; 0.269], p = .004and g = 0.169, 95% CI [0.095; 0.242], p < .001,respectively.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the younger-adult far-transfer meta-analysis, compared to non-active controls, the overall effect size was g = 0.161, 95% CI [0.035; 0.288], m = 26, k = 113, df = 8.62, p = .018, ω2 = 0.000, τ2 = 0.016.

    “The overall effect size of the RVE model including all the effect sizes related to far-transfer measures in groups compared to non-active controls was g = 0.161, 95% CI [0.035; 0.288], m = 26, k = 113, df = 8.62, p = .018,ω 2 = 0.000, τ 2 = 0.016.”
    Quote in the paperNot a substantial claim

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    In the younger-adult far-transfer meta-analysis, the RVE overall effect size was g = 0.099, 95% CI [0.015; 0.182], m = 44, k = 247, df = 15.26, p = .024, ω2 = 0.000, τ2 < 0.001.

    “The overall effect size was g = 0.099, 95% CI [0.015; 0.182], m = 44, k = 247, df = 15.26,p = .024,ω 2 = 0.000, τ 2 < 0.001.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

    Findingown data

    In the younger-adult far-transfer publication-bias analysis, the PEESE estimator was g = 0.061, 95% CI [-0.003; 0.126], p = .069.

    “The PET and PEESE estimators were g = -0.003,95% CI [-0.119; 0.113], p = .961and g = 0.061, 95% CI [-0.003; 0.126], p = .069,respectively.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the younger-adult criterion sensitivity analysis, after excluding two influential cases and one excessively large effect, the overall effect size was g = 0.958, 95% CI [0.660; 1.256], m = 10, k = 31, df = 4.94, p < .001, ω2 = 0.020, τ2 = 0.142.

    “The results without these effect sizes were g = 0.958, 95% CI [0.660; 1.256], m = 10, k = 31, df = 4.94, p < .001,ω 2 = 0.020, τ 2 = 0.142.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the younger-adult far-transfer sensitivity analysis, after excluding three influential cases, the overall effect size was g = 0.097, 95% CI [0.024; 0.170], m = 44, k = 244, df = 15.96, p = .012, ω2 = 0.000, τ2 = 0.000.

    “The results without these effect sizes were g = 0.097, 95% CI [0.024; 0.170], m = 44, k = 244, df = 15.96,p = .012,ω 2 = 0.000, τ 2 = 0.000.”
    Quote in the paperNot a substantial claim

  • May combine two claims. The quote only partly states this claim.

    Findingown data

    In the younger-adult near-transfer meta-analysis, the RVE overall effect size was g = 0.208, 95% CI [0.129; 0.286], m = 31, k = 160, df = 11.82, p < .001, ω2 = 0.053, τ2 = 0.008.

    “The overall effect size was g = 0.208, 95% CI [0.129; 0.286], m = 31, k = 160, df = 11.82,p < .001,ω 2 = 0.053, τ 2 = 0.008.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the younger-adult far-transfer publication-bias analysis, the random-effect model overall effect size was g = 0.107, 95% CI [0.043; 0.170], p < .001, k = 44, τ2 = 0.000.

    “the overall effect size of the random-effect model was g = 0.107, 95% CI [0.043; 0.170], p < .001,k = 44, τ 2 = 0.000.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the younger-adult far-transfer meta-analysis, compared to active controls, the overall effect size was g = 0.059, 95% CI [-0.030; 0.147], m = 27, k = 134, df = 9.91, p = .170, ω2 = 0.000, τ2 = 0.000.

    “The overall effect size of the RVE model including all the effect sizes related to far-transfer measures in groups compared to active controls was g = 0.059, 95% CI [-0.030; 0.147], m = 27, k = 134, df = 9.91, p = .170,ω 2 = 0.000, τ 2 = 0.000.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    The younger-adult near-transfer effect was slightly smaller than the older-adult near-transfer effect (g = 0.208 vs. g = 0.274).

    “Overall, the results showed a slightly smaller effect than the one estimated in older adults (g = 0.208 vs. g = 0.274).”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the younger-adult criterion publication-bias analysis, the random-effect model overall effect size was g = 0.836, 95% CI [0.624; 1.049], p < .001, k = 10, τ2 = 0.039.

    “the overall effect size of the random-effect model was g = 0.836, 95% CI [0.624; 1.049], p < .001,k = 10, τ 2 = 0.039.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the younger-adult near-transfer sensitivity analysis, after excluding three influential cases, the overall effect size was g = 0.183, 95% CI [0.105; 0.260], m = 31, k = 157, df = 11.39, p < .001, ω2 = 0.023, τ2 = 0.013.

    “The results without these effect sizes were g = 0.183, 95% CI [0.105; 0.260], m = 31, k = 157, df = 11.39,p < .001,ω 2 = 0.023, τ 2 = 0.013.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The meta-analysis found some near-transfer effects and an absence of far-transfer effects, especially on measures of fluid intelligence.

    “Another interesting theoretical insight is offered by the concurrent presence of some near-transfer effects and absence of far-transfer effects, especially on measures of fluid intelligence.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

    Findingown data

    In the younger-adult criterion publication-bias analysis, the PEESE estimator was g = 0.821, 95% CI [0.598; 1.043], p < .001.

    “The PET and PEESE estimators were g = 0.841, 95% CI [0.396; 1.286], p = .006and g = 0.821, 95% CI [0.598; 1.043], p < .001,respectively.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Findingown data

    In the younger-adult near-transfer publication-bias analysis, the random-effect model overall effect size was g = 0.180, 95% CI [0.119; 0.241], p < .001, k = 31, τ2 = 0.000.

    “the overall effect size of the random-effect model was g = 0.180, 95% CI [0.119; 0.241], p < .001,k = 31, τ 2 = 0.000.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    The authors used robust variance estimation (RVE) with hierarchical weights and small-sample corrections to control for statistical dependence between effect sizes.

    “To control for statistical dependence between effect sizes, we used robust variance estimation (RVE) with hierarchical weights and smallsample corrections to calculate the overall effect size and perform metaregression analysis (Hedges, Tipton, & Johnson, 2010).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Hedges, Tipton, & Johnson, 2010 source

  • May combine two claims.

    Premisereasoning

    Previous meta-analyses on working memory training in older adults are either underpowered or run with a suboptimal modeling approach, and none has reached a definite conclusion.

    “However, these meta-analyses are either underpowered (i.e., few studies included) or run with a suboptimal modeling approach.Thus, none of these meta-analyses has reached a definite conclusion regarding the current state of the art in this literature.”
    Quote in the paperSubstantial claim

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

    Premisereasoning

    The impact of working memory training is less clear with older adults' cognitive skills than with younger populations.

    “By contrast, the impact of WM training is less clear with older adults' cognitive skills.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Working memory is correlated with reading and mathematical skills.

    “Furthermore, WM is correlated with reading (Peng et al., 2018) and mathematical skills (Peng, Namkung, Barnes, & Sun, 2016).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Peng et al., 2018 source
    • cites Peng, Namkung, Barnes, & Sun, 2016 source

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

    Study design and conductown data

    The authors used Holm's method to adjust p-values for multiple comparisons.

    “To minimize problems related to low statistical power (e.g., Type II error), two secondary categorical moderators (with df > 1) were tested separately with the Holm's method.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Viechtbauer, 2010

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

  • Classified as probably not a substantial claim.

    Premisecitation

    Working memory capacity is strongly correlated with fluid intelligence.

    “WM capacity, that is, the number of items WM can retain and manipulate, is strongly correlated with fluid intelligence (Engle, Tuholski, Laughlin, & Conway, 1999;Kane, Hambrick, & Conway, 2005)”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Engle, Tuholski, Laughlin, & Conway, 1999 source
    • cites Kane, Hambrick, & Conway, 2005 source

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

    Study design and conductnone given

    The authors chose four main moderators a priori.

    “We chose (a priori) four main moderators that were included in the meta-regression analyses:”
    Quote in the paperNot a substantial claim

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

    Study design and conductown data

    The far-transfer categorization had Cohen's kappa = 0.96.

    “The Cohen's kappa was κ = 0.96.”
    Quote in the paperNot a substantial claim

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

    Study design and conductreasoning

    The authors conducted separate analyses for studies from the University of Padova lab.

    “In addition, we conducted separate analyses for the studies carried out by the researchers of the University of Padova (Borella and Carretti).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    The authors calculated effect sizes for each relevant dependent variable reported in the studies to prevent cherry-picking.

    “To prevent any bias caused by potential cherry-picking practices, we calculated the effect sizes for each relevant dependent variable reported in the studies.”
    Quote in the paperSubstantial claim

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

    Study design and conductown data

    Two authors coded each effect size for moderator variables independently, with Cohen's kappa = 0.98.

    “Two authors coded each effect size for moderator variables independently.The Cohen's kappa was κ = 0.98.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The uncorrected overall criterion effect size in older adults was large (g = 0.877) and relatively robust to influential case analysis and publication bias analysis.

    “The uncorrected overall effect size was large g ( = 0.877) and relatively robust to influential case analysis and pub- lication bias analysis.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The far-transfer meta-analysis of older adults showed modest to null overall effect sizes that were highly homogeneous.

    “Overall, the analyses showed modest to null overall effect sizes that were highly homogeneous.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    In the far-transfer meta-analysis of older adults, the proportion of studies reporting follow-up measures was small (13 out of 38, 70 out of 248 effect sizes), which made the estimated overall effect (g = 0.241) unreliable.

    “Finally, like in the near-transfer metaanalysis, the proportion of studies reporting follow-up measures was small (13 out of 38, 70 out of 248 effect sizes), which made the estimated overall effect (g = 0.241) unreliable.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The unbiased criterion effect in older adults is statistically significant and probably between g = 0.500 and g = 0.550.

    “As suggested by trim-and-fill and the PEESE estimator, the unbiased effect is statistically significant and probably between g = 0.500 and g = 0.550.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The probable true near-transfer effect in older adults is between g = 0.159 and g = 0.246 (both significant).

    “While the uncorrected overall effect was g = 0.274, the sensitivity analysis estimated the probable true effect between g = 0.159 and g = 0.246 (both significant; as seen earlier, the PET estimator is probably overcorrected).”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Interpretationown data

    In the younger-adult far-transfer meta-analysis, the analyses showed highly homogeneous small (with non-active controls) to near-zero (with active controls) overall effect sizes.

    “Like in the meta-analysis of WM training in older adults, the analyses showed highly homogeneous small (with non-active controls) to nearzero (with active controls) overall effect sizes.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The results reported by studies assessing the impact of WM training on older adults' cognitive function are extremely consistent, and there is no reason to think that this literature has produced mixed results so far.

    “Put together, these outcomes show that the results reported by the studies assessing the impact of WM training on older adults' cognitive function are actually extremely consistent.Therefore, there is no reason to think that this literature has produced mixed results so far.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    The difference in criterion effect size between younger and older adults is probably due to younger populations being better at acquiring new skills by training.

    “This difference is probably due to younger populations being better at acquiring new skills by training.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    The present meta-analysis corroborates the hypothesis that working memory and fluid intelligence are two non-isomorphic cognitive constructs supported by distinct neural mechanisms.

    “The present meta-analysis, therefore, corroborates the hypothesis according to which WM and fluid intelligence are two non-isomorphic cognitive constructs supported by distinct neural mechanisms.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The three meta-analyses provide a picture consistent with the literature on WM training in children and young adults: strong effects in trained tasks, small effects (near transfer and episodic memory) to medium effects (nearer transfer) in memory tasks, and small (with non-active controls) to null effects (with active controls) in far-transfer tasks.

    “The three metaanalyses provide a picture consistent with the literature on WM training in children and young adults: strong effects in the trained tasks, small effects (near transfer and episodic memory) to medium effects (nearer transfer) in the memory tasks, and small (with non-active controls) to null effects (with active controls) in the far-transfer tasks.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Both episodic-memory tasks and working memory training tasks require recall skills.

    “In fact, both episodicmemory tasks and WM training tasks require recall skills (for a review, seeUnsworth & Engle, 2007).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Unsworth & Engle, 2007 source

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

    Interpretationreasoning

    Working memory training programs may contribute to developing recall strategies such as simple mnemonics, which in turn may be used in episodic memory tasks.

    “WM training programs may contribute to developing recall strategies such as simple mnemonics, which in turn may be used in episodic memory tasks.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    The classification of episodic-memory measures differs among meta-analyses.

    “The classification of episodic-memory measures differs among meta-analyses.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Weicker et al. (2016) source
    • cites Tetlow and Edwards (2017) source

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

    Study design and conductown data

    Replicating the analyses using different values of pre-post-test correlations ranging between 0.500 and 0.800 found only negligible differences.

    “We replicated the analyses using different values of pre-post-test correlations ranging between 0.500 and 0.800.Only negligible differences were found.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Memory tests may not be adequate proxies for domain-general and transferable memory skills due to similarities between trained tasks and some memory tests.

    “Due to the similarities between trained tasks and some memory tests, the latter may not be adequate proxies for domain-general and transferable memory skills.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Practicing working memory tasks may help to develop the ability to perform similar tasks without having any impact on domain-general cognitive constructs such as short-term or long-term memory.

    “Put simply, practicing WM tasks may help to develop the ability to perform similar tasks (e.g., free recall, n-back, and span tasks) without having any impact on domain-general cognitive constructs such as short-term or long-term memory.”
    Quote in the paperSubstantial claim
55 of 55

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?

  • More recent meta-analyses that distinguished between nearest and near transfer effects found significantly greater effect sizes for the former, suggesting that WM training may mainly yield task-specific transfer rather than a general improvement in WM.

    “However, more recent meta-analyses that distinguished between nearest and near transfer effects found significantly greater effect sizes for the former, suggesting that WM training may mainly yield task-specific transfer rather than a general improvement in WM [5, 15].”

    From Adaptive working memory training does not produce transfer effects in cognition and neuroimaging (2022) · cites this paper as [15]

  • Meta-analyses that explicitly distinguish between types of control group typically report a lack of far, sometimes also near, transfer effects.

    “Critically, meta-analyses that explicitly distinguish between the types of a control group typically report a lack of far, sometimes also near, transfer effects [5, 15, 17].”

    From Adaptive working memory training does not produce transfer effects in cognition and neuroimaging (2022) · cites this paper as [15]

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • ConsistentQ(38) = 38.88reported p = .43 · recomputed p = .430

    “Q (38) = 38.88,p = .430”

  • ConsistentQ(37) = 42.01reported p = .263 · recomputed p = .263

    “Q(37) = 42.01,p = .263”

  • ConsistentQ(9) = 15.16reported p = .087 · recomputed p = .087

    “Q(9) = 15.16,p = .087”

3 of 3

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.

  • Corrected criterion-effect range excludes a reported estimate

    Minor

    The stated range for older adults' corrected criterion effect begins at 0.500, but Table 1 gives a publication-bias-adjusted range beginning at 0.479.

1 of 1

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Sala, G., Aksayli, N. D., Tatlidil, K. S., Gondo, Y., & Gobet, F.
Affiliations
  • The University of Osaka (JP)
  • Brown University (US)
  • University of Liverpool (GB)
Corresponding author
Sala, Giovanni
ORCID
Venue
Intelligence
Publisher
Elsevier BV
Published
2019-11
Volume
77
Pages
101386
ISSN
0160-2896 (ISSN-L)
DOI
10.1016/j.intell.2019.101386
OpenAlex
W2899393285
Semantic Scholar
d1120cce52aa23c670d7550fc30b3fe7c2e92765
Cited by
63
References
90
Citations by year
2026: 3 · 2025: 10 · 2024: 10 · 2023: 12 · 2022: 11 · 2021: 10 · 2020: 6 · 2019: 3
FWCI
7.70
Type
journal article
Language
en
Open access
Licence
Locations
  • Intelligence (publishedVersion, cc-by)
  • RePEc: Research Papers in Economics (submittedVersion)
  • PsyArXiv (OSF Preprints) (submittedVersion)
  • PsyArXiv (OSF Preprints) (submittedVersion, cc-by)
Funders
Japan Society for the Promotion of Science
Topics
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Cognitive Abilities and Testing
  • Social Sciences › Psychology › Neuropsychology and Physiological Psychology › Aging and Gerontology Research
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Cognitive Functions and Memory
Keywords
working memory training, cognitive training, older adults, transfer effects, near transfer, far transfer, fluid intelligence, cognitive aging

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