Does IQ Change With Age? What Moves and What Stays

How much can an IQ score vary from measurement error alone?

For the three example reliabilities in the table below (.95, .90 and .80), measurement error alone can move a single result by roughly 7 to 13 IQ points in either direction at an approximate 95% level.

The standard error of measurement (SEM) is the standard deviation of the errors of measurement attached to a test's scores. In Harvill's NCME teaching module (1991), it equals the standard deviation of the test scores times the square root of one minus the reliability. IQ scales use a standard deviation of 15, so SEM = 15 × √(1 − reliability). An approximate 95% range around an observed score is the score plus or minus 1.96 SEMs. It assumes normally distributed errors of the same size at every score level, and strictly speaking the interval is built around the true score, which is why the table says “about”.

So a result of 110 on a test with a reliability of .90 is consistent with a range of roughly 101 to 119. Mensa's own online test page notes that people often get different scores from the same test taken at different times. For how this applies to online tests, see how accurate IQ tests are.

How far can an IQ score drift over decades?

Over six or seven decades, people's positions moved enough that the age-11 to age-70 correlation was .67 rather than close to 1.

A correlation is easier to read in points. Take someone who scored 115 at age 11, one standard deviation above average. If both scores follow the usual bell curve on the same 15-point scale, the expected later score is 100 + r × 15, and the 95% range is that value plus or minus 1.96 × 15 × √(1 − r²). The table applies the three coefficients from the Scottish cohorts.

This is a simplified model. It assumes a straight-line relationship and the same spread at both ages, and Gow and colleagues also report corrected coefficients that are larger than these raw ones. Part of the spread is the measurement error from the section above, and a correlation alone cannot separate error from real change. Read the table as a sense of scale, not a forecast for one person.

Which mental abilities rise and fall with age?

Different abilities peak at different ages: processing speed in the late teens, working memory around 30, and vocabulary at about 50 or later.

The five knowledge-based WAIS subtests (vocabulary, information, comprehension, arithmetic and similarities) peaked significantly later than nearly all other tasks in that study. For the split between solving new problems and using stored knowledge, see what fluid intelligence is.

Follow-up of the same people shows how small the average decline is before the 60s. In the Seattle Longitudinal Study (Schaie and Willis, 2010), reliably replicable average declines did not appear before age 60, though small but significant declines showed up in some, but not all, cohorts from their 50s. Below 60 the average decline was under 0.2 standard deviations. By age 74 reliable decline showed in every ability tested. By 81 the average reached about 1 standard deviation for most variables, yet fewer than half of individuals showed a reliable decline over the previous 7 years. On the IQ scale, 0.2 standard deviations is 3 points and 1 standard deviation is 15 (our arithmetic).

What does the Flynn effect have to do with age and IQ?

The Flynn effect is the long-term rise in IQ test scores across birth years, so older and younger adults differ partly because they were born in different decades.

A formal meta-analysis by Pietschnig and Voracek (Perspectives on Psychological Science, May 2015) pooled 271 independent samples from 31 countries, almost 4 million people, covering 1909 to 2013. Gains differed by test domain, with fluid reasoning rising fastest, were stronger for adults than for children, and have slowed in recent decades.

The rise does not run in one direction forever. Bratsberg and Rogeberg (PNAS, June 2018) used Norwegian conscription scores for birth cohorts from 1962 to 1991. The rise, its turning point and the later decline all showed up in differences within families.

This matters for age because a one-time comparison of a 25-year-old with a 75-year-old mixes age with birth year. Schaie and Willis found that short-term longitudinal data gave different peaks and rates of decline than cross-sectional data, and that cross-sectional designs overestimate age change before the 60s for abilities with negative cohort gradients. Test publishers also refresh their norms, and Pearson lists updated norms for the 2024 WAIS-5.

Can your IQ go up or down in adulthood?

In healthy adults an age-normed score moves little over the years, but a single result can sit several points above or below the last one.

Part of any gap between two sittings is measurement error, and a difference between two scores carries the error of both, so it is less reliable than either score alone (Harvill, 1991). Practice on a familiar format can also lift a result without changing the ability underneath. How much a second sitting adds, and whether a repeat is worth it, is a retesting question covered in should you retake an IQ test.

If your thinking or memory changes and stays changed after an illness, a head injury or a change in medication, see a doctor or a qualified psychologist.

What does this mean for an online estimate like IQMog's?

An online estimate is a snapshot of one ability on one day. Take it again in a year and you get a second dated reading to compare with the first.

IQMog is a fixed form of 20 Raven-style visual pattern puzzles that takes about 10 minutes. It converts your correct answers to the usual IQ scale (mean 100, standard deviation 15) and shows a score range and percentile next to the estimate. Each correct answer is worth 3 points on a 70 to 130 scale: 14 correct is 70 + 3 × 14 = 112, a worked illustration with a likely range of roughly 100 to 124. IQMog uses a fixed assumed norm (a mean of 10 correct out of 20, a spread of 5) instead of age-group norms, so its estimate does not compare you with your own age group.

The test is free to take, and the full report, with a PDF you can keep, is a one-time payment, $4.99 for most visitors and €4.99 incl. VAT in euro countries, with no subscription. Take it once with a clear head and read the range against the tables above.