Find the Best Available Evidence

 Quick Analysis Plan Copied

 Long Analysis Plan

Evidence Base for Computerized Brain Games

The strongest claim-level evidence comes from a randomized controlled trial in which 85 high-school students completed 14 days of computerized executive-function training versus no training. Three trained tasks improved significantly across sessions — interference resolution (error rate η² = .257; RT η² = .835), task switching (RT η² = .730), and inhibition (RT η² = .576) — while goal monitoring did not improve, showing gains are task-specific, not global Training-task improvement statistics.

Far transfer is modest. Raven's Matrices scores rose more in the trained group (46.23 vs 43.67 pre/post) than in controls (44.46 vs 43.35), with a significant group × time interaction (η² = 0.143). However, the control group also improved, and the authors caution that retest effects, dropout-driven motivation differences, and the lack of an active control may account for part of the apparent benefit Far-transfer group interaction results.

Who benefits most? Lower initial IQ predicted larger IQ gains (r = -0.255, p = .018), and gains correlated with training progress on the switching task (r = 0.446, p = .01) — suggesting greater malleability in lower-ability segments Moderator: baseline IQ and gains. In a different segment — chronic acquired brain injury — a non-randomized 12-participant pilot of computer-based PreMotor Exercise Games reported significant improvements in shoulder/wrist range of motion (p = 0.01) and executive-function (EFPT) initiation/overall scores, with mixed or null strength changes, but no control group and high variability in time spent training Brain-injury computer game pilot study.

Analysis plan. (1) Extract pre/post and group × time statistics from RCTs; standardize to Cohen's d or η². (2) Test far-transfer vs near-transfer outcomes as separate effect-size strata using mixed-effects meta-regression (Python: statsmodels/pymeta) with active-vs-passive control as moderator — the passive control here is a key confound. (3) Model baseline cognitive level as a predictor of gain (the r = -0.255 compensation effect). (4) For clinical populations, restrict to controlled designs; pilot data should be weighted near zero. (5) Sensitivity analyses for dropout and retest effects (model control-group gains explicitly). Blind spots: no long-term follow-up in either study, both lacked blinding, and the ADHD/LD systematic-review record indicates VR/digital tools show promise for assessment but mixed predictive validity — transfer to real-world function remains underdemonstrated Systematic review of digital EF tools in ADHD/LD.

Confidence note: near-task skill improvement is well-supported; general-population far-transfer to fluid intelligence is modest and confounded; claims for brain-injury rehabilitation rest on uncontrolled pilot data only. Large preregistered RCTs with active controls and follow-up would most decisively change this picture.