Do programs designed to train working memory, other executive functions, and attention benefit children with ADHD? A meta-analytic review of cognitive, academic, and behavioral outcomes.

This meta-analysis examined whether computer-based "cognitive training" programs—also called facilitative intervention training (FIT)—improve outcomes for children with ADHD. The interventions fell into three categories:

**Short-term memory training only:** Programs like Cogmed that repeatedly drill children on remembering sequences of numbers, letters, or spatial locations (e.g., remembering a sequence of lights and tapping them back in reverse order).

**Attention training only:** Programs that train sustained focus, usually through continuous performance tasks (e.g., pressing a button when a target appears but not when a non-target appears).

**Mixed executive function training:** Programs that train multiple cognitive skills simultaneously, including working memory, inhibitory control (impulse control), and set shifting (switching between tasks or rules).

The comparators were:

Active control groups (children who did a different computer activity, like watching videos or playing non-training games)

Waitlist control groups (children who received no intervention during the study period)

No-treatment control groups

The outcome measures were grouped into three categories:

**Near transfer:** Performance on tasks similar to the trained task (e.g., if you trained working memory, did working memory tests improve?)

**Far transfer:** Performance on tasks different from the trained task, including academic achievement tests (reading, math), objective cognitive tests (IQ, executive function tests), and behavioral ratings from parents and teachers

**Blinded vs. unblinded ratings:** Whether the person rating the child's behavior knew the child was in the treatment group

The meta-analysis included 25 studies with a total of 1,194 children diagnosed with ADHD (primarily combined type and predominantly inattentive type). Children ranged from approximately 5 to 16 years old. Most studies recruited from clinical referrals, outpatient clinics, and community advertisements. Approximately 75-80% of participants were male, consistent with ADHD population demographics. Studies were conducted in the United States, Europe, and Australia. Exclusion criteria across studies typically included IQ below 80, comorbid autism spectrum disorder, psychosis, or neurological disorders. Children taking psychostimulant medication were included in most studies, but medication status was not always controlled during testing sessions.

The meta-analysis extracted data from studies using a wide range of instruments:

**For near transfer (trained domain):**

Short-term memory: Digit span forward/backward (Wechsler scales), spatial span tasks, Corsi block-tapping task, letter-number sequencing

Attention: Continuous Performance Test (CPT) omission and commission errors, Test of Variables of Attention (TOVA)

Executive functions: Wisconsin Card Sorting Test (perseverative errors), Stroop test (interference score), Trail Making Test Part B

**For far transfer (untrained domains):**

Academic achievement: Woodcock-Johnson Tests of Achievement (reading, math, writing subtests), Wechsler Individual Achievement Test, grade point average, teacher-reported academic performance

Objective cognitive tests: Wechsler Intelligence Scale for Children (full-scale IQ, performance IQ), Raven's Progressive Matrices

Behavioral ratings: Conners' Parent and Teacher Rating Scales, ADHD Rating Scale-IV, Behavior Assessment System for Children (BASC), Child Behavior Checklist (CBCL)

**Blinding assessment:**

Blinded ratings: Teachers or parents who were told the child was in a "study" but not whether they received training or control condition

Unblinded ratings: Teachers or parents who knew the child was in the training group (often because they administered the training at home or were told explicitly)

**Design:** This is a meta-analysis—a statistical synthesis of 25 independent studies. The authors used random-effects models, which assume that the true effect size varies across studies (rather than being identical), and corrected for publication bias using trim-and-fill procedures and Egger's regression test.

**Search strategy:** The authors searched PsycINFO, PubMed, ERIC, and Google Scholar through March 2013, plus hand-searched reference lists of retrieved articles and contacted researchers in the field. Inclusion criteria were: (1) children with ADHD diagnosis (DSM-IV or ICD-10), (2) computer-based cognitive training intervention, (3) at least one control group (active, waitlist, or no-treatment), (4) reported sufficient statistics to calculate effect sizes, (5) published in English in a peer-reviewed journal.

**Quality assessment:** Studies were coded for methodological quality including random assignment, blinding of outcome assessors, use of active control conditions, attrition rates, and sample size.

**Statistical approach:** Effect sizes were calculated as Hedges' g (a bias-corrected version of Cohen's d). The authors computed separate meta-analyses for each outcome category (near transfer, far transfer cognitive, far transfer academic, blinded behavioral ratings, unblinded behavioral ratings). They also tested for moderator effects including type of training, age of children, duration of training, and type of control group.

**What this design can and cannot prove:**

*What it CAN prove:*

Whether cognitive training produces consistent effects across multiple studies

Whether effects differ by outcome type (near vs. far transfer)

Whether effects differ by rater blinding status

Whether study characteristics (duration, age, training type) moderate outcomes

*What it CANNOT prove:*

Causality—meta-analyses inherit the limitations of the individual studies

Long-term effects—most individual studies had short follow-up periods (typically 0-6 months post-training)

Whether specific training protocols are superior—there were too few studies per protocol to compare directly

Whether different "doses" of training matter—studies varied widely in training duration (5-25 hours total) but sample sizes were too small to test dose-response relationships

**Major methodological weaknesses flagged by the authors:**

Only 25 studies met inclusion criteria, and many had small sample sizes (median N = 24 per study)

Only 8 of 25 studies used blinded outcome assessors for behavioral ratings

Fewer than half used active control conditions (most used waitlist or no-treatment controls)

Publication bias was detected: studies with null or negative results were less likely to be published

Most studies had short follow-up periods (0-3 months post-training), making durability of effects unknown

Many studies did not report whether children were on medication during testing, and if so, whether medication status was held constant

**Near transfer (trained domain):**

Short-term memory training → short-term memory improvement: **d = 0.63** (95% CI: 0.38 to 0.88, p < .001) — moderate, statistically significant effect

Attention training → attention improvement: **d = 0.08** (95% CI: -0.22 to 0.38, p = .60) — not statistically significant

Mixed executive function training → executive function improvement: **d = 0.17** (95% CI: -0.06 to 0.40, p = .15) — not statistically significant

**Far transfer (untrained domains):**

Academic achievement: **d = 0.11** (95% CI: -0.09 to 0.31, p = .28) — not statistically significant

Objective cognitive tests (IQ, etc.): **d = 0.14** (95% CI: -0.02 to 0.30, p = .09) — not statistically significant

Blinded behavioral ratings: **d = 0.16** (95% CI: -0.06 to 0.38, p = .15) — not statistically significant

Unblinded behavioral ratings: **d = 0.48** (95% CI: 0.23 to 0.73, p < .001) — moderate, statistically significant

**Critical comparison:**

The difference between unblinded ratings (d = 0.48) and blinded ratings (d = 0.16) was statistically significant (p < .05), indicating that when raters knew children were in the training group, they reported benefits nearly three times larger than when raters were unaware of group assignment.

**Publication bias:**

Trim-and-fill analysis suggested that 4-6 studies with null or negative results were missing from the literature. After correcting for this bias, the near-transfer effect for working memory training dropped from d = 0.63 to d = 0.48 (still significant but smaller).

**Moderator analyses:**

Age did not significantly moderate outcomes

Training duration (total hours) did not significantly moderate outcomes

Type of control group (active vs. waitlist) did not significantly moderate outcomes

Studies with active control conditions showed smaller effects than studies with waitlist controls, but the difference was not statistically significant

**What d = 0.63 means for near transfer:** If you trained working memory for 5-25 hours, the average child with ADHD improved their performance on working memory tests by about 0.63 standard deviations. In practical terms, this means a child at the 50th percentile before training would move to about the 74th percentile after training—on tests that look very similar to the training tasks themselves. This is a meaningful improvement, but only on the specific skill trained.

**What d = 0.11 means for academics:** The average child showed essentially no improvement in reading, math, or writing scores after cognitive training. A child at the 50th percentile before training would move only to the 54th percentile—a trivial change that could easily be due to chance or practice effects.

**What d = 0.16 vs. d = 0.48 means for behavior:** When teachers or parents didn't know whether a child received training, they reported almost no behavioral improvement (equivalent to moving from the 50th to the 56th percentile). When they knew the child was in the training group, they reported moderate improvement (equivalent to moving from the 50th to the 68th percentile). This three-fold difference strongly suggests expectancy effects—people see what they expect to see.

**Comparison to established treatments:** For context, psychostimulant medication for ADHD produces effect sizes of d = 1.53 to 1.89 for symptom reduction (Van der Oord et al., 2008). Behavioral parent training produces d = 0.31 to 0.87. The far-transfer effects of cognitive training (d = 0.11 to 0.16) are substantially smaller than either established treatment.

**What the authors acknowledge:**

Small number of studies (25) with generally small sample sizes

Publication bias toward positive results

Most studies lacked active control conditions (children in control groups often did nothing or were on a waitlist)

Few studies used blinded raters for behavioral outcomes

Short follow-up periods (most ≤3 months)

Heterogeneity in training protocols, outcome measures, and control conditions

Many studies did not report medication status or control for it

Possible Hawthorne effects (improvement due to attention from researchers, not the intervention itself)

**What a critical reader would add:**

**Commercial funding bias:** Several studies were funded by companies that sell cognitive training programs (e.g., Cogmed), and industry-funded studies tend to show larger effects

**Training-to-test similarity:** Near-transfer effects may reflect simple practice effects—children get better at the type of task they practiced, not at the underlying cognitive ability

**No long-term follow-up:** Without data beyond 3-6 months, we cannot know if even the near-transfer effects persist

**Ecological validity:** Laboratory working memory tasks may not reflect real-world working memory demands in classrooms or social situations

**Dosage variability:** Training ranged from 5 to 25 total hours across studies, and the optimal dose is unknown

**Selection bias:** Children who enroll in cognitive training studies may be more motivated or have higher-functioning families than the general ADHD population

**No active placebo:** Most "active control" conditions (e.g., watching videos) do not control for expectations, engagement, or computer time equally

For someone running their own n=1 experiment (or a parent considering cognitive training for their child):

### What to test

**Specific intervention:** A working memory training program (e.g., Cogmed, n-back tasks, or a structured digit/visuospatial span training protocol)

**Dose:** At least 20-25 sessions of 30-45 minutes each (5-15 total hours minimum), based on the upper end of what studies used

**Frequency:** 4-5 sessions per week for 5-8 weeks

### Minimum meaningful duration

**Training phase:** At least 5 weeks (25 sessions) to match the longer studies in the meta-analysis

**Follow-up:** At least 3 months post-training to test durability (most studies only measured immediately after training)

**Total experiment:** 4-5 months minimum (baseline + training + post-test + follow-up)

### What to measure (specific metrics)

**Near transfer (primary outcome):** Digit span backward (Wechsler scales) or a spatial span task (Corsi block) — test weekly to track trajectory

**Far transfer (secondary outcomes):**

- Academic: Timed math fluency test (e.g., Woodcock-Johnson Math Fluency or a custom 2-minute addition/subtraction test)

- Attention: Continuous Performance Test (CPT) omission errors or a free online sustained attention task

- Behavior: Daily behavior log (e.g., "How often did I/my child lose focus during homework?" rated 1-5 each day)

**Blinding proxy:** Have someone who doesn't know whether training is happening (e.g., a teacher, a spouse who isn't involved) rate behavior weekly

### Key confounds to control for

**Expectancy effects:** The person rating behavior should ideally be blinded. If not possible, use objective measures (reaction time, accuracy) rather than subjective ratings

**Practice effects:** Take baseline measures 3 times over 2 weeks before starting training to establish a stable baseline and reduce practice effects on tests

**Medication:** Keep medication type and dose constant throughout the experiment. If the child takes stimulants, test at the same time of day relative to medication

**Sleep and diet:** Track sleep quality (hours, consistency) and caffeine/sugar intake daily, as these affect cognitive performance

**Computer time:** If the control condition involves no computer time, any improvement could be due to increased screen engagement rather than the specific training

**Motivation:** Use a reward system for completing training sessions (not for performance) to ensure compliance

**Seasonal effects:** Run the experiment during a stable school period (not during exam weeks, holidays, or summer break)

### What a positive result would look like

**Near transfer:** Digit span backward improves by at least 1.5-2 standard deviations (e.g., from a scaled score of 7 to 10 or higher) from baseline to post-training, AND this improvement is maintained at 3-month follow-up

**Far transfer:** Math fluency improves by at least 0.5 standard deviations (e.g., from 15 to 20 correct problems in 2 minutes) — this would exceed the meta-analytic average of d = 0.11

**Behavior:** Blinded ratings show at least a 20% reduction in ADHD symptoms (e.g., from "often" to "sometimes" on key items) — if the rater is unblinded, be skeptical of any improvement

**Pattern to watch for:** If near-transfer improves but far-transfer does not, this matches the meta-analytic findings and suggests the training is not generalizing to real-world function

**Red flag:** If only unblinded ratings improve but objective tests do not, this is likely an expectancy effect, not a real cognitive change

**Bottom line for self-experimenters:** The evidence suggests that