Does Mental Math Actually Improve Cognition? The Research
Parents, teachers, and brain-training apps keep repeating the same line: mental math makes you smarter overall. The peer-reviewed record tells a narrower story. Four decades of randomised trials, neuroimaging studies, and meta-analyses point to real gains in a few well-defined skills, no change at all in general IQ, and a dose curve that most casual practice never reaches. This post pulls apart what the evidence supports, what it rejects, and what a sensible training target looks like for an adult.
What The Meta-Analyses Found
The three largest meta-analyses on arithmetic training and cognition pool roughly 180 trials and 12,000 participants. The pooled effect on trained arithmetic accuracy is large, around d = 0.80 after 8 to 12 weeks of practice. The pooled effect on working-memory span is moderate, around d = 0.35, and shows up mostly on digit-span and n-back tasks that share structure with the training. The pooled effect on fluid intelligence tests like Raven's matrices sits at d = 0.05 and loses significance when researchers correct for small-study bias. That gap between trained-task gains and far-transfer gains is the single most replicated finding in the field.
The same pattern appears in children, adults, and older learners aged 60 to 80. Older adults show the largest relative gains on processing speed, which starts from a lower baseline. Children show the largest gains on mathematical reasoning tests that resemble school curricula. Neither group shows reliable transfer to unrelated reasoning, memory for faces, or problem solving in non-numeric domains. If a study claims otherwise, the effect usually shrinks or disappears when a second lab tries to replicate it with a larger sample.
The Transfer Problem: Near Versus Far
Cognitive scientists split training effects into near transfer and far transfer. Near transfer means the gain shows up on tasks that share surface features with the training. Far transfer means the gain shows up on tasks that share only abstract features. Mental arithmetic training produces strong near transfer to any task involving number manipulation, calculation under time pressure, or working-memory updates with digits. It produces weak-to-zero far transfer to verbal reasoning, spatial rotation, or executive function tests like the Stroop task.
This matters because most popular claims about mental math (sharper focus, better memory for names, faster decision-making at work) are far-transfer claims. The research does not support them at the effect sizes brands advertise. It does support the claim that daily mental arithmetic makes you faster and more accurate at mental arithmetic, and that the gain persists for months after practice stops. For a documented example of near-transfer training in a related skill, our breakdown of working memory and mental math walks through the RCT evidence in more detail.
Where The Gains Hold Up
Four specific gains survive across replications. First, speed of basic arithmetic: after 20 minutes a day for 6 weeks, median response time on single-digit multiplication drops 30 to 45 percent, with gains stable at 3-month follow-up. Second, digit span: forward span gains 0.6 to 0.9 digits on average after 8 weeks of mental-math practice that includes carry operations. Third, mathematical self-efficacy: self-reported anxiety on calculation tasks drops by about 20 percent, which predicts real-world willingness to tackle numeric problems at work. Fourth, in older adults only, processing-speed scores on the Digit Symbol Substitution Test rise by about 0.3 standard deviations after 12 weeks.
- Basic arithmetic speed: 30 to 45 percent faster after 6 weeks, stable at 3 months.
- Digit span: +0.6 to +0.9 digits after 8 weeks of carry-heavy drills.
- Math anxiety: about 20 percent drop in self-reported dread of numeric tasks.
- Older-adult processing speed: +0.3 SD on Digit Symbol after 12 weeks, age 60+.
These effects are small-to-moderate in absolute terms, but they are the parts of the training-and-cognition literature that replicate. If a research summary promises more than these four gains, it is either reporting a single unreplicated study or inflating a statistically fragile effect.
Where The Claims Collapse
Several widely marketed claims fail to survive replication. Mental math does not raise full-scale IQ: the two large preregistered trials that tested this (Jaeggi 2014 replication, Melby-Lervag 2016 meta) found no effect once publication bias was corrected. It does not improve academic performance in non-math subjects: a 2019 synthesis of 24 school trials found a d = 0.04 effect on reading comprehension, indistinguishable from zero. It does not reliably slow age-related cognitive decline beyond the processing-speed gain noted above: the ACTIVE trial and its 10-year follow-up found no protection against dementia incidence from reasoning or speed training.
Brain-imaging studies add a sharper caveat. fMRI work from the past decade shows that trained calculators recruit different neural circuits than novices (more parietal, less frontal), but this shift sits inside the arithmetic network itself. It does not propagate to regions handling language, planning, or episodic memory. The neural picture matches the behavioural one: practice rewires the circuit you train, not the brain at large.
How Much Practice Moves The Needle
Dose matters more than method. The threshold for measurable near-transfer gains in adults is roughly 15 minutes a day, 5 days a week, for 6 weeks. That totals around 7.5 hours of focused practice. Below that threshold, effects are noisy and do not separate from placebo. Above 30 minutes a day, gains plateau in most studies, which argues against marathon sessions. Spacing beats bulking: two 10-minute sessions beat one 20-minute session on retention tests measured 4 weeks later.
Difficulty calibration is the second lever. Adaptive training that keeps the learner at roughly 75 to 85 percent accuracy produces larger gains than fixed-difficulty drills, because it forces working-memory recruitment on every trial. A quick way to hit that zone is a timed single-digit round on /daily followed by a harder carry-heavy set on /menu; the first round warms up retrieval, the second pushes into the carry-and-hold region where gains concentrate. Techniques like Chisanbop finger-counting add a physical-anchor layer that helps children sustain the 75 to 85 percent zone without losing track.
What This Means For Your Training
The honest verdict: mental math is a narrow but real cognitive investment. It will make you faster at calculation, slightly expand your working-memory span, and reduce the dread that stops most adults from attempting numeric problems in public. It will not raise your IQ, fix your attention span, or prevent cognitive decline. The ROI case is strongest for anyone who handles numbers under social pressure (teachers, founders, parents helping with homework, shopkeepers, poker players) and weakest for someone hoping for a general cognitive upgrade.
If you want to turn this into a training block, pair a daily 10-minute session on /daily with a weekly harder round on a named method such as the Doomsday algorithm. The daily round builds retrieval speed in the four gains that replicate, and the weekly method builds a second memory anchor that keeps the practice from becoming rote. Six weeks in, measure yourself on a timed single-digit round and a seven-digit forward-span test. Those are the metrics the research moves.
