Does playing action video games improve performance on tests of cognitive ability? A recent meta-analysis ( Bediou et al., 2018a ) summarized the available evidence and concluded that it can. Their analysis, however, did not adequately correct for publication bias. We re-analyzed the same set of studies with more appropriate adjustments for publication bias and found minimal evidence for transfer of training to cognitive ability measures. Instead, it is possible that there are little or no benefits, just publication bias — the exclusion of non-significant results from the published literature. That bias may be the cause of a lab effect reported in the original meta-analysis. The meta-analysis showed that studies from the Bavelier lab (the senior author of the meta-analysis) reported larger effects than other labs. We show that many of these original studies distributed different outcomes from the same or highly overlapping sets of participants across publications without noting the overlap. This salami-slicing might contribute to the extent of publication bias in the literature. More compelling, independent, and transparent evidence is needed before concluding that action video game training transfers to performance on other cognitive tasks.

Theory building in science requires replication and integration of findings regarding a particular research question. Second-order meta-analysis (i.e., a meta-analysis of meta-analyses) offers a powerful tool for achieving this aim, and we use this technique to illuminate the controversial field of cognitive training. Recent replication attempts and large meta-analytic investigations have shown that the benefits of cognitive-training programs hardly go beyond the trained task and similar tasks. However, it is yet to be established whether the effects differ across cognitive-training programs and populations (children, adults, and older adults). We addressed this issue by using second-order meta-analysis. In Models 1 ( k = 99) and 2 ( k = 119), we investigated the impact of working-memory training on near-transfer (i.e., memory) and far-transfer (e.g., reasoning, speed, and language) measures, respectively, and whether it is mediated by the type of population. Model 3 ( k = 233) extended Model 2 by adding six meta-analyses assessing the far-transfer effects of other cognitive-training programs (video-games, music, chess, and exergames). Model 1 showed that working-memory training does induce near transfer, and that the size of this effect is moderated by the type of population. By contrast, Models 2 and 3 highlighted that far-transfer effects are small or null. Crucially, when placebo effects and publication bias were controlled for, the overall effect size and true variance equaled zero. That is, no impact on far-transfer measures was observed regardless of the type of population and cognitive-training program. The lack of generalization of skills acquired by training is thus an invariant of human cognition.

How do speakers avoid producing verb overgeneralization errors such as * She covered paint onto the wall or * She poured the cup with water? Five previous papers have found seemingly contradictory results concerning the role of statistical preemption (competition from acceptable alternatives such as She covered the wall with paint or She poured water into the cup ) and entrenchment (a mechanism sensitive to all uses of the relevant verb). Here, we use more appropriate measures of preemption and entrenchment (attraction measures based on the chi-square statistic, as opposed to using only the frequency of occurrence in favoured constructions) as well as more appropriate statistical analyses and, in one case, a larger corpus to reanalyse the data from these studies. We find that for errors of verb argument structure overgeneralization (as in the examples above), preemption/entrenchment effects are almost always observed in single-predictor models, but are rarely dissociable, due to collinearity. Fortunately, this problem is much less acute for errors of reversative un - prefixation (e.g., * unsqueeze ; * uncome ), which could in principle be blocked by (a) non-reversative uses of the same verb root (e.g., squeeze, come; entrenchment), and/or (b) lexically-unrelated verbs with similar meanings to the relevant un - forms (e.g., release, go ; preemption). Across a reanalysis of two previous studies of un - prefixation, and a new extended replication with adults, we find dissociable effects of both preemption and entrenchment. A meta-analytic synthesis revealed that, across the studies, both effects are reliable, though preemption appears to increase with age. We conclude that a successful account of the retreat from verb overgeneralization is likely to be one that yields preemption and entrenchment as effects that fall naturally out of the learner’s attempts to communicate meaning, rather than one that treats these effects as mechanisms in their own right, and discuss current accounts that potentially meet this criterion. Finally, we set out some methodological recommendations that can be profitably applied not only to corpus-based experimental studies, but studies of child language acquisition in general.