Can Practice Raise an IQ Score? Practice Effects, Explained

Yes, practice raises the score you get on a cognitive test — reliably, measurably and by a lot. No, that is not the same as raising the ability the test was built to estimate, and the evidence for the second claim is much weaker than the evidence for the first. Almost every argument about this subject is caused by the two being treated as one.

There are actually three things being collapsed, and separating them makes the research legible.

Three claims wearing one sentence

Claim one: your score goes up because the test is now familiar. You have seen the item formats, you know the instructions, you have stopped wasting seconds working out what is being asked. This is the practice effect, and it is a measurement artefact — the number moved, you did not.

Claim two: you got genuinely better at the task. Your arithmetic really is faster; you really can hold a longer sequence. This is a real gain and a useful one, and it is what practice is for.

Claim three: the underlying ability changed. The broad capacity the test is supposed to index is now higher, and it should show up on tasks that look nothing like the one you drilled. This is the claim that marketing makes and the evidence does not support.

Claim one is a nuisance. Claim two is worth having. Claim three is where honesty is required.

How large practice effects actually are

Jendryczko, Scharfen and Holling, writing in the Journal of Intelligence in 2019, tracked performance across repeated administrations of cognitive ability tasks. The largest retest effect appeared between the first and second sessions, with an effect size of 0.72. Gains remained positive and significantly different from zero through the fourth administration, after which mean ability slightly decreased between the fourth and fifth sessions — a difference that was not statistically significant. The authors connect the shape to the power law of practice: large early gains, diminishing quickly.

An effect of 0.72 on a standardised scale is not a rounding error. On a metric with a standard deviation of 15 points, an effect of that size is in the region of ten points, which is larger than most differences anybody cares about.

Speed follows its own trajectory. Scharfen, Blum and Holling's 2018 meta-analysis of mental speed tests pooled 36 studies, 49 samples, 212 outcomes and 21,810 participants, and found response times falling by 0.237 standard deviations from the first administration to the second, 0.367 by the third and 0.499 by the fourth. Notably, they reported that a plateau had not been reached after four administrations — response times kept improving after accuracy gains had flattened. More complex tasks showed larger reductions than simple ones, and longer test-retest intervals attenuated the effect somewhat without removing it.

The effect is big enough to distort research

If you want one demonstration of how seriously to take this, it is Salthouse's work on cognitive ageing.

Cross-sectional comparisons — different people of different ages, tested once — show performance lower at older ages from early adulthood onward. Longitudinal comparisons — the same people, retested over years — often show performance higher, particularly below 65. Salthouse's quasi-longitudinal analysis found age trends in each cognitive domain much closer to the cross-sectional pattern than the longitudinal one, and attributed the discrepancy to practice effects from repeated testing rather than to cohort differences.

In other words: a well-designed longitudinal study of thousands of adults can be pushed in the wrong direction by the same mechanism that inflates your second attempt at a web quiz.