CoChat
Sign Up
Cited by 107Cognitive Abilities and TestingCognitive and developmental aspects of mathematical skillsCognitive Functions and Memory

Reevaluating the effectiveness of n-back training on transfer through the Bayesian lens: Support for the null

Open access
Dougherty, Michael R., Hamovitz, Toby, Tidwell, Joe W.·2016·Psychonomic Bulletin & Review
DOI 10.3758/s13423-015-0865-9
Open this paper in CoChat — free accountOpen PDF

Abstract

A recent meta-analysis by Au et al. Psychonomic Bulletin & Review, 22, 366-377, (2015) reviewed the n-back training paradigm for working memory (WM) and evaluated whether (when aggregating across existing studies) there was evidence that gains obtained for training tasks transferred to gains in fluid intelligence (Gf). Their results revealed an overall effect size of g = 0.24 for the effect of n-back training on Gf. We reexamine the data through a Bayesian lens, to evaluate the relative strength of the evidence for the alternative versus null hypotheses, contingent on the type of control condition used. We find that studies using a noncontact (passive) control group strongly favor the alternative hypothesis that training leads to transfer but that studies using active-control groups show modest evidence in favor of the null. We discuss these findings in the context of placebo effects.

AI snapshot

Written from the paper’s full text.

TL;DR

Dougherty et al. reanalyze Au et al.’s (2015) n-back training meta-analysis using Bayesian methods and find that evidence for transfer to fluid intelligence depends heavily on control-group type. Passive-control studies strongly favor transfer, but active-control studies provide modest-to-strong evidence for the null, suggesting the apparent training effect may be a placebo or design artifact.

Key findings

  • Across all 24 comparisons from 20 papers, the meta-analytic Bayes factor ignoring control type was 152:1 in favor of the alternative hypothesis, but this masked a large difference by control condition.
  • Passive-control studies yielded a Bayes factor of 13,241:1 in favor of transfer, whereas active-control studies yielded a Bayes factor of 7.7:1 in favor of the null; with a two-sided test, these were 6,000:1 for the alternative and 11:1 for the null, respectively.
  • Hierarchical Bayesian models showed passive-control studies consistently had positive effect sizes, while active-control studies had effects near zero; the only aggregate effect whose 95% HDI excluded zero was for non-USA passive-control studies, and active-control effects within the USA shrank to essentially zero.
  • Among individual studies, 11 of 12 passive-control effect sizes were positive, but only 6 of 12 active-control effect sizes were positive; the majority of individual studies favored the null, including 6 of 12 passive-control studies.

Methods

  • Bayesian reanalysis of effect sizes from Fig. 3 of Au et al. (2015), using 24 comparisons from 20 papers with healthy adults aged 18–50 and n-back as the only training task; 12 comparisons used passive controls and 12 used active controls.
  • Effect sizes were transformed to t values using \(t = \sqrt{1/n_1 + 1/n_2} \times g\), then default Bayes factors were computed with the BayesFactor package in R (`ttestBF`, `meta.ttestBF`), using a one-sided interval and scale factor \(r = 1\), with sensitivity checks using two-sided tests and different priors.
  • Follow-up hierarchical Bayesian models estimated effect sizes as a function of control type (passive vs. active) and, in additive models, country of origin (USA vs. non-USA), using four prior distributions on population effect size (vague, small, medium, large).

Limitations

  • The paper states that conclusions do not necessarily generalize to other types of training or age groups, because Au et al. limited inclusion to n-back training and healthy adults aged 18–50.
  • The paper states that, without experiments specifically designed to test the placebo explanation, it cannot definitively conclude that the active–passive discrepancy reflects a placebo effect; it also notes causal indeterminacy in comparing placebo controls with no-contact controls across studies.
  • The paper notes that control type covaried with country of origin—most active-control studies were conducted within the USA, while most non-USA studies used passive controls—leaving open the possibility that cultural differences contribute, though the authors doubt this accounts for the 7-fold difference. Evident from the text: active-control studies were heterogeneous in the nature of their control tasks, and the reanalysis depends on the effect sizes and control-type categorization provided by Au et al.

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

Integrity signals

Registry

Crossref · 10.3758/s13423-015-0865-9

Cites

53 of 53 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

6 of 29 cited sources checked · 4 supported · 1 flagged · accuracy not yet measured

  • Premisecitation

    Because of causal indeterminacy, comparing different kinds of control conditions is not a valid way to determine whether placebo effects are present.

    “Due to causal indeterminacy, one cannot simply compare different types of control conditions to infer the presence or absence of placebo effects.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Hróbjartsson, Kaptchuk, and Miller (2011) source

      The Introduction says that assessing a placebo effect requires comparison with a suitable control group, and the Conclusion calls placebo-versus-no-treatment comparison the best available design for estimating placebo effects. The section on extrapolation uses ‘causal indeterminateness’ to describe difficulty isolating the factors responsible for an observed effect, not to rule out control-group comparisons. The paper cautions that such comparisons are crude and susceptible to bias, but does not deem them invalid for detecting placebo effects. (LLM, conf=0.83)

      Checked by a reasoning model — may be wrong.

      Experimental — accuracy not yet measured
    • The authors state this as an assumption: Studies that use active controls may also use other procedures that reduce expectancy or placebo effects.
  • Premisecitation

    Chein and Morrison (2010) found a significant effect of complex span training on executive control, t(38) = 1.81, p = .039 (one-sided).

    “Chein and Morrison (2010) reported a significant effect of complex span training on executive control with a t(38) = 1.81, which was reported as significant (p = .039,one-sided).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Chein and Morrison (2010) source

      Table 1 (page unavailable in our copy) reports a significant between-group effect on cognitive control, measured with the Stroop test: t(40) = 1.81, p = .039. The Results state that the t tests were one-tailed. The claim gives the wrong degrees of freedom, t(38); that value appears in Table 1 for reading comprehension, not cognitive control. (LLM, conf=0.94)

      Experimental — accuracy not yet measured
  • Findingreasoning

    With a one-sided prior on the effect size, the Bayes factor for Chein and Morrison's (2010) result is only 1.78 (0.98 for a two-sided test) favoring the alternative, which is essentially uninformative about both hypotheses.

    “assuming a one-sided prior on the effect size, the corresponding BF is only 1.78 (BF = 0.98 for the two-sided test) in favor of the alternative.This is basically uninformative with respect to both the alternative and the null hypothesis.”
    Quote in the paperSubstantial claim
  • Findingown data

    In hierarchical Bayesian models, the estimated effect size for active-control studies within the USA shrank to essentially zero.

    “Furthermore, the estimated effect size for the active-control studies within the USA shrink to essentially zero.”
    Quote in the paperSubstantial claim
  • Findingown data

    Even when the prior distribution assumed a large training effect, there was no evidence that n-back training improves Gf measures.

    “Strikingly, even when the prior distribution is set such that the effect of training is assumed to be large, there is still no evidence of that n-back training leads to improvements on Gf measures.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 15
  • May combine two claims. 23

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

    Study design and conductcitation

    The authors' analysis approach closely followed the methods used by Rouder and Morey (2011) and Rouder et al. (2013).

    “Our analysis approach closely followed the methods used by Rouder and Morey (2011) and Rouder et al. (2013) in their meta-analyses of psi (i.e., extrasensory perception).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Rouder and Morey (2011) source
    • cites Rouder et al. (2013) source

  • May combine two claims.

    Premisecitation

    A true placebo-control trial presumably isolates the treatment's causal mechanism through the placebo condition, but comparing a placebo control with a no-contact control does not do this.

    “In true placebo-control trials, the causal mechanism of the treatment is presumably isolated by virtue of including the placebo control condition.However, the same is not true when comparing a placebo-control with a no-contact control.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Hróbjartsson et al., 2011 source

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

    Premisecitation

    The shortcomings of null hypothesis significance testing are well documented.

    “The limitations of NHST methods are well documented and need not be rehashed here in their entirety”
    Quote in the paperNot a substantial claim2 of 3 sources checked
    • cites Raftery (1995) source
    • cites Wagenmakers (2007) source
    • cites Rouder and Morey (2011) source

  • May combine two claims.

    Findingown data

    The median effect size among active-control studies was slightly above zero but was essentially eliminated when country of origin was added as a predictor.

    “While this model estimates that the median effect size amongst the active control studies is slightly above zero, this small positive effect is essentially eliminated when country of origin is added as a predictor in the model, as shown in Figs. 5 and 6.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingcitation

    In the authors' Bayesian reanalysis, ignoring control type, the meta-analytic Bayes factor was 152:1 in favor of the alternative hypothesis, which is decisive evidence.

    “First, ignoring the type of control, the odds in favor of the alternative hypothesis is 152:1.This qualifies as Bdecisive^evidence according to Jeffreys' (1961) scheme.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Jeffreys, 1961

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

    • Rests on an assumption: NHST tends to overstate evidence for the alternative hypothesis and does not allow evaluating the relative probability that the null hypothesis is true.

  • May combine two claims.

    Findingown data

    Among individual studies, 11 of 12 passive-control effect sizes were positive, whereas only 6 of 12 active-control effect sizes were positive.

    “First, 11 of the 12 effect sizes for the passive control studies are positive, whereas only 6 of the 12 effect sizes are positive for the activecontrol studies.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    With a two-sided test, the Bayes factor was 6,000:1 favoring the alternative for passive-control studies and 11:1 favoring the null for active-control studies, the latter strong evidence for the null.

    “Note that when a two-sided test is used lieu of the one-sided test, we obtain a Bayes factor of 6,000:1 (in favor of the alternative) for the passive control studies and a Bayes factor of 11:1 (in favor of the null) for the active control studies.The later constitutes strong evidence for the null amongst studies using proper experimental controls.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Jeffreys, 1961

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

    • Rests on an assumption: NHST tends to overstate evidence for the alternative hypothesis and does not allow evaluating the relative probability that the null hypothesis is true.

  • May combine two claims.

    Findingown data

    Among the studies examined, the type of control design was associated with study location: active-control studies were mostly run in the USA, whereas studies run outside the USA mostly used passive controls.

    “choice of experimental design also covaried with whether the study was conducted within the USA or outside the USA.Most of the studies using active controls were conducted within the USA, whereas the majority of the studies conducted outside of the USA used passive controls.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In the authors' Bayesian reanalysis, passive-control studies favored the alternative hypothesis with a Bayes factor of 13,241:1, whereas active-control studies provided substantial evidence for the null with a Bayes factor of 7.7:1.

    “We find that studies using a noncontact (passive) control group strongly favor the alternative hypothesis that training leads to transfer but that studies using active-control groups show modest evidence in favor of the null.”
    • Also stated in: § Preliminaries · plain
      “While the Bayes factor for the passive control studies is a whopping 13,241:1 in favor of the alternative, the Bayes factor for the active control studies is a more modest 7.7:1 in favor of the null.As stated above, this qualifies as substantial evidence for the null.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Jeffreys, 1961

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

    • The authors state this as an assumption: A one-sided interval placing prior mass on effects greater than zero is reasonable under the hypothesis that training improves Gf.
    • Rests on an assumption: NHST tends to overstate evidence for the alternative hypothesis and does not allow evaluating the relative probability that the null hypothesis is true.

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

    Findingown data

    Among active-control studies, the training condition's effect size (d = 0.25) is numerically smaller than the effect size for control participants in passive-control studies (d = 0.28).

    “the effect size for the training condition amongst active-control studies (d = 0.25) is actually numerically smaller than the effect size amongst the control participants in the passive control studies (d = 0.28).”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    In Au et al.'s meta-analysis, studies using active controls (N = 12) had an effect size of 0.06, while studies using passive controls (N = 12) had an effect size of 0.44.

    “They also evaluated several possible mediators, including whether the studies used an active control (N = 12) or a passive control (N = 12), which yielded effect sizes of 0.06 and 0.44, respectively.”
    Quote in the paperSubstantial 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.

  • May combine two claims.

    Premisecitation

    Au et al. restricted their meta-analysis to studies that used n-back as the only training task and included healthy adults aged 18-50.

    “Two important details of the selection criteria are that Au et al. limited their analyses to studies that used a form of the n-back task as the only training task and to studies that included healthy adults aged 18-50.”
    Quote in the paperSubstantial 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.

  • May combine two claims.

    Premisecitation

    Au et al.'s (2015) meta-analysis found an overall weighted effect size of g = 0.24 for n-back training on fluid intelligence, based on 24 comparisons from 20 papers.

    “Their results revealed an overall effect size of g = 0.24 for the effect of n-back training on Gf.”
    • Also stated in: § Preliminaries · plain
      “The aggregate weighted effect size across these 24 comparisons was 0.24.”
    • Also stated in: § Preliminaries · plain
      “Au et al. (2015) presented effect sizes for 24 individual comparisons drawn from 20 papers.”
    Quote in the paperSubstantial 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.

  • May combine two claims.

    Premisereasoning

    The central disagreement in the working-memory training literature concerns whether training improves performance on non-trained tasks (far transfer), a question for which the null hypothesis is theoretically meaningful and plausible.

    “the main point of disagreement in the literature pertains to whether training leads to improvements on nontrained tasks (far transfer), where the null hypothesis is a theoretically meaningful and plausible hypothesis.”
    Quote in the paperSubstantial claim

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

    Study design and conductown data

    The hierarchical Bayesian analyses were repeated with three different prior distributions to assess sensitivity.

    “The hierarchical Bayesian analyses were repeated using three different prior distributions on the population level effect size (see Table 2) to assess the sensitivity of the posterior distribution to different prior beliefs.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    A prior Bayesian study found that transfer from other training types to untrained tasks consistently supported the null.

    “there is at least one study using Bayesian analyses that showed that transfer from other forms of training to non-trained tasks uniformly favored the null (see Sprenger et al., 2013).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Sprenger et al., 2013 source

  • May combine two claims.

    Study design and conductnone given

    The authors' conclusions were robust to large modifications of the prior distribution and to using a two-sided null interval.

    “Importantly, even large modifications to the prior distribution do not alter our conclusions in any substantive way, nor does using a two-sided null interval.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Because placebo effects are prevalent, the difference between active- and passive-control designs should not be treated merely as moderation; it must be taken seriously as a possible cause of the effect.

    “Given the prevalence of placebo effects, the discrepancy in findings between active and passive experimental designs cannot simply be described as a moderation effect; it has to be fully considered as a possible root cause of the effect.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationcitation

    Such near-transfer effects are unsurprising and of little practical value, and the cited studies found no far-transfer evidence.

    “such transfer effects are neither surprising nor of much practical interest, and neither of these studies found evidence for far transfer.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    Unlike inferring an effect's presence or absence from a p value, the Bayesian approach lets researchers quantify how strong the evidence is.

    “rather than relying on a p value to infer the presence or absence of an effect, the Bayesian approach allows one to quantify the strength of the evidence.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    Without effects relative to a proper control condition or other control for placebo effects, it is difficult to interpret passive-control studies or justify claims of effective transfer.

    “Without showing effects relative to a proper control condition, or otherwise controlling for possible placebo effects, it is difficult to move forward with interpreting results from the passive control studies, let alone justify claims of effective transfer.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: NHST tends to overstate evidence for the alternative hypothesis and does not allow evaluating the relative probability that the null hypothesis is true.

  • May combine two claims.

    Premisecitation

    Deviation from double-blind procedure is the norm in the cognitive training literature, with only a small number of studies attempting a double-blind placebo-control procedure.

    “Unfortunately, deviation from the double-blind procedure is the norm within the cognitive training literature: We know of only a small number of studies that attempted to use a double blind placebo-control procedure (Sprenger et al., 2013, Study 2;von Bastian & Eschen, 2015).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Sprenger et al., 2013 source
    • cites von Bastian & Eschen, 2015

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

  • May combine two claims.

    Interpretationreasoning

    The authors' findings are limited in scope and may not extend to different training tasks or age ranges.

    “Thus, the conclusions we draw below do not necessarily generalize to other types of training or age groups.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    A genuine effect of n-back training on Gf would also appear in studies that use active controls.

    “if nback training has a true effect on Gf, then these effects should hold even for studies that use active controls.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    Evidence strength can be assessed against any theoretically justified hypothesis, the null included.

    “the strength of the evidence can be evaluated in relation to any theoretically justified hypothesis, including the null hypothesis.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The pattern indicates two distinct data-generating processes, and neither is due to n-back training.

    “it is clear that the data reflect two separate data-generating processes, neither of which can be attributed to n-back training.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The effects observed in passive-control studies reflect something other than a training effect.

  • May combine two claims.

    Interpretationreasoning

    The authors doubt that cultural differences explain the seven-fold larger training effect, particularly because the studies outside the USA were mostly from Westernized cultures.

    “we doubt cultural differences would account for the 7-fold increase in the training effect, especially since the non-USA studies were primarily conducted in Westernized cultures (e.g., Europe).”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    If n-back training produces gains in fluid abilities, this should hold for both passive-control and active-control studies.

    “If n-back training does indeed produce gains in fluid abilities, as claimed by Au et al., then this should hold both for studies that include passive-control groups and for studies that use active-control groups.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: NHST tends to overstate evidence for the alternative hypothesis and does not allow evaluating the relative probability that the null hypothesis is true.

  • May combine two claims.

    Interpretationnone given

    The authors propose that the effect Au et al. observed is an experimental design artifact and is consistent with a placebo effect interpretation.

    “Although their analysis does indeed illustrate an effect of some sort, we propose that this effect is an experimental design artifact, and is consistent with a placebo effect interpretation.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The effects observed in passive-control studies reflect something other than a training effect.

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

    Premisecitation

    One set of guidelines for interpreting Bayes factor magnitudes comes from Jeffreys's (1961) labeling scheme.

    “Jeffrey's (1961) labeling scheme provides one set of guidelines for interpreting the magnitude of BFs.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Jeffreys (1961)

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

  • May combine two claims.

    Interpretationown data

    Neither the present meta-analysis nor that of Au et al. (2015) provides evidence that n-back training improves general cognitive functions.

    “the meta-analysis presented here and in Au et al. (2015) on n-back training does not provide such evidence.”
    Quote in the paperSubstantial 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).

    Interpretationnone given

    A Bayes factor can be read numerically as how strongly the evidence favors one model over another.

    “the beauty of the BF is that it can be interpreted numerically as the strength of evidence for a particular model relative to an alternative.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    These weaknesses include the ongoing use of passive controls and inappropriate analyses that correlate training gains with transfer gains.

    “One of these areas of methodological deficiencies entails the continued use of passive controls and the other the use of inappropriate analysis techniques that entails correlating training gains with transfer gains (see Tidwell et al., 2014).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Tidwell et al., 2014 source

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

    Interpretationreasoning

    Comparing control groups without the training groups says little about training's effect, because training effects must be measured against a control.

    “The comparison of control groups while ignoring the training groups isn't particularly informative regarding effect of training, since the effects of training can only be assessed relative to the control.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    Masking which condition participants are in is especially hard in cognitive training studies, since participants directly experience the intervention.

    “This is particularly challenging for cognitive training studies because participants literally see and engage in the intervention, making it difficult to mask what condition participants believe they have been assigned to.”
    Quote in the paperNot a substantial claim
35 of 35

  • The authors state this as an assumption.

    The effects observed in passive-control studies reflect something other than a training effect.

    “Because the passivecontrol studies show a substantial positive effect while the active control studies do not, it seems reasonable to assume that the effects observed in the passive-control studies reflects something other than a training effect.”

  • The authors state this as an assumption.

    A one-sided interval placing prior mass on effects greater than zero is reasonable under the hypothesis that training improves Gf.

    “The one-sided test is a reasonable assumption under the hypothesis that training should lead to improvements in Gf.”

  • The paper relies on this without support.

    NHST tends to overstate evidence for the alternative hypothesis and does not allow evaluating the relative probability that the null hypothesis is true.

    “two well-known limitations of the NHST framework are that it tends to overstate evidence for the alternative hypothesis and does not permit one to evaluate the relative probability that the null hypothesis is in fact true.”

  • The authors state this as an assumption.

    Studies that use active controls may also use other procedures that reduce expectancy or placebo effects.

    “it is reasonable to assume that studies that adopt active controls might also adopt other procedures that minimize expectancy or placebo effects.”
4 of 4

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?

  • Subsequent meta-analyses did not find any significant far transfer effects, especially when the effects of WM training were compared with an active control group.

    “However, subsequent meta-analyses did not find any significant far transfer effects [5, 13, 17], especially when the effects of WM training were compared with an active control group, that is, a group practicing tasks with little WM demands.”

    From Adaptive working memory training does not produce transfer effects in cognition and neuroimaging (2022) · cites this paper as [17]

    The Meta Analysis Bayes Factor section reports that, across 12 active-control studies, the estimated effects of n-back training on fluid-intelligence transfer were not different from zero and the meta-analytic Bayes factor favored the null. The claim is broader than this analysis: it refers to WM training and far transfer generally, whereas this paper reanalyzes n-back training and transfer to fluid intelligence. The Abstract also notes that the earlier meta-analysis found an overall positive effect. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • 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 [17]

  • The lack of follow-up far-transfer effects seems to reject the hypothesis that some time is needed for generalized results reported in most recent meta-analyses and systematic reviews in the broader field of WM training.

    “Interestingly, the lack of follow-up far-transfer effects also seems to reject the hypothesis that some time is needed in order for generalized results reported in most recent meta-analyses and systematic reviews in the broader field of WM training”

    From The cognitive and academic benefits of Cogmed: A meta-analysis (2019) · cites this paper as Dougherty, Hamovits, & Tidwell, 2016

  • A meta-analysis found that working-memory training studies with passive control groups were more likely to favor the alternative hypothesis, whereas those with active control groups were more likely to support the null.

    “Dougherty et al. (2016) conducted a meta-analysis of working-memory training studies using a Bayesian approach and found that studies that using a passive control group were more likely to favor the alternative hypothesis that training was effective, while studies employing active control groups were more likely to support the null.”

    From A simultaneous examination of two forms of working memory training: Evidence for near transfer only (2016) · cites this paper as Dougherty et al. (2016)

  • The effectiveness of n-back training in eliciting transfer to other different tasks remains unclear.

    “The effectiveness of training based on the n-back task in eliciting transfer to other different tasks thus remains unclear.”

    From Nearest transfer effects of working memory training: A comparison of two programs focused on working memory updating (2019) · cites this paper as [20]

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.

  • Stated effect-size-to-t formula contradicts Table 1

    Minor

    For the cited row, the stated formula gives 1.109 × √(1/8 + 1/8) = 0.5545, not the tabulated t of 2.218. Dividing g by the square-root term instead gives the tabulated value, so the conversion formula appears to be reversed.

1 of 1

Cites

Flagged first, then by citations

1 of the 54 references couldn’t be matched to a known paper.

Research works printed but not linked: 4

  1. Jaeggi, S. M., Buschkuehl, M., & Jonides, J. (2009). Working memory training and transfer. Arlington: Presented at the annual ONR Contractor’s meeting.
  2. Jeffreys, H. (1961). Theory of probability. New York: Oxford University Press.
  3. Katz, B., Jaeggi, S. M., Buschkuehl, M., Shah, P., & Jonides, J. (2015). Money can’t buy you fluid intelligence (but it might not hurt either): The effect of compensation on transfer following a working memory intervention. Manuscript under review.
  4. Schwarb, H. (2012). Optimized cognitive training: Investigating the limits of brain training on generalized cognitive function (Unpublished doctoral dissertation). Atlanta: Georgia Institute of Technology.

Cited by

Sorted by citations

Related

Authors
Dougherty, M. R., Hamovitz, T., & Tidwell, J. W.
Affiliations
University of Maryland, College Park (US)
Corresponding author
Dougherty, Michael R.
ORCID
Dougherty, Michael R.
Venue
Psychonomic Bulletin & Review
Publisher
Springer Science and Business Media LLC
Published
2015-06-17
Volume
23
Issue
1
Pages
306-316
ISSN
  • 1069-9384 (ISSN-L)
  • 1531-5320
DOI
10.3758/s13423-015-0865-9
OpenAlex
W808339630
Semantic Scholar
2058f86b82fec07fce805d1e617f1088fb8d4bd5
Cited by
106
References
54
Citations by year
2026: 2 · 2024: 4 · 2023: 4 · 2022: 7 · 2021: 12 · 2020: 10 · 2019: 17 · 2018: 17 · 2017: 13 · 2016: 13 · 2015: 3
FWCI
23.50
Type
journal article
Language
en
Open access
Licence
www.springer.com/tdm (tdm)
Locations
  • Psychonomic Bulletin & Review (publishedVersion)
  • PubMed (publishedVersion)
Topics
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Cognitive Abilities and Testing
  • Physical Sciences › Mathematics › Statistics and Probability › Cognitive and developmental aspects of mathematical skills
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Cognitive Functions and Memory
Keywords
working memory training, fluid intelligence, transfer of training, Bayesian hypothesis testing, null hypothesis, placebo effect, active control, effect size
MeSH terms
Bayes Theorem, Data Interpretation, Statistical, Humans, Intelligence, Memory, Short-Term, Transfer, Psychology, Meta-Analysis as Topic

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