WORLD’S FIRST TITANIUM CHASENFollow the tradition — pass on the ware.Take the 1-question survey →
Follow Studio Fungwa

Your IQ Score Is a Range, Not a Number

IQ test score shown as a range rather than a single number

Someone scores 112 on a professionally administered IQ test and walks away thinking a fixed fact about their brain has been measured, the way a scale measures weight. It has not. What the test produced is an estimate with a known margin of error around it. On a full adult battery, that margin is typically about five points either side at the level psychologists usually report. A score of 112 is best read as roughly 107 to 117 — a band, not a point. Understanding that band changes how almost every IQ number should be read, including cutoffs, comparisons, and the score you got from an online quiz.

What the margin of error actually is

Psychologists describe every test result with a simple model: the score you observed equals a hypothetical true score plus error. The true score is the average you would get if you could take equivalent versions of the test many times, with no learning between sittings and no change in you. You can never observe it directly. What you can estimate is how much scores bounce around it.

That bounce is quantified as the standard error of measurement, usually shortened to SEM. It is calculated from the test’s reliability — how consistently it measures — and the spread of scores in the population. Higher reliability means a smaller SEM. For the Full Scale IQ on major batteries such as the Wechsler scales, reported SEMs are commonly around 2 to 3 points. Multiply by roughly two and you get the 95 percent confidence interval printed in professional reports: about plus or minus 5 points. Our IQ score chart shows the bands and labels that those intervals can straddle.

Two details matter here. First, the interval is not uniform across a test. Shorter scores are less precise than longer ones. In the Stanford-Binet 5 technical summary, for example, the average standard error reported for the Full Scale score is much smaller than for the abbreviated score from the same battery — roughly 2.3 points versus roughly 4.6 points in the summaries reviewed for this article. A brief screening number therefore carries about twice the uncertainty of the full score, even when both come from a reputable test. Second, index scores and single subtests are built from fewer tasks than the Full Scale, so their intervals are typically wider. A five-point gap between two index scores is often well inside measurement noise.

Why the same person gets different numbers

Measurement error is only one reason scores move. Three others show up again and again in the research.

State on the day. Sleep, illness, pain, medication, anxiety and simple distraction all shift performance, especially on timed tasks. That is random error in practice, and it is one reason a single sitting should never be treated as exact. Our guide to how IQ tests work explains how examiners try to standardise conditions for exactly this reason.

Practice. Taking a test makes you better at taking that test. A widely cited meta-analysis by Hausknecht and colleagues (2007) found that retaking a cognitive ability test raised scores by about a quarter of a standard deviation on average — roughly 4 IQ points — with larger gains when the identical form was repeated. Our article on whether practice raises your IQ score separates that score gain from any change in underlying ability. Practice gain sits on top of measurement error, which is why a retest a few points higher proves very little on its own.

Regression to the mean. An extreme score usually contains some luck, good or bad. On retesting, luck averages out and the score tends to drift back toward 100. A meta-analysis of longitudinal studies reported that adult rank order was stable at about .76 across five years (Breit and colleagues, 2024) — high, but far from perfect. In plain terms: people largely keep their rough position, while individual numbers still move by several points.

On top of all three, different tests are not interchangeable. They sample different abilities, use different norm groups collected in different years, and sometimes use different scales. That is why the WAIS-5 article on this site warns against averaging scores across editions: a few points between a WAIS-IV report and a WAIS-5 report can come from the edition and its fresh 2023–24 norms, not from any change in the person.

What the 2026 research adds

Recent work has tightened, rather than overturned, this picture. In January 2026, Robison and colleagues published a comprehensive psychometric study in Behavior Research Methods: 255 young adults completed 24 cognitive measures twice, two weeks apart, covering seven constructs including working memory, processing speed and fluid intelligence. Individual measures showed adequate to high test-retest reliability, combined construct-level estimates were more reliable still, and the underlying measurement structure held steady across the two sittings. The authors’ latent state-trait analysis found that most systematic variance reflected stable traits rather than the occasion or the specific task. Read it at its real size: this was a sample of young adults taking research tasks, not patients taking a clinical IQ battery. But it supports the central point of this article — single tasks wobble more than well-built composites do.

A second 2026 line of work, on retest effects in cognitive ability tests, reinforces that repeated sittings do not just add noise; they can add a systematic upward drift as people learn the format. Noise plus drift is exactly why professionals set minimum intervals before re-administering the same battery and why a rising sequence of online scores is weak evidence of rising ability.

There is also a live interpretive caution for the newest adult test. A 2026 analysis of the WAIS-5 standardization summary statistics, reviewed in an October 2026 guide to professional tests, favoured four group factors over the publisher’s five-index separation, with a dominant general factor, and urged emphasis on the broad total score. That is a dispute about how finely a profile should be sliced, not evidence that the test is meaningless. Its practical lesson fits this article: the Full Scale number, with its narrower interval, is usually on firmer ground than a story built from a small gap between two index scores, each with a wider interval of its own.

How to read your own score

  • Ask for the interval, not just the number. A professional report should print a confidence interval or an SEM. If it only gives a single number, ask which test, which edition, which norms, and what interval the examiner would put around it.
  • Treat small differences as no difference. As a working rule, a gap of 2 or 3 points between two sittings, two people, or two index scores is inside the noise for most batteries. Do not build a conclusion — gifted or not, improved or declined, stronger or weaker — on a gap that size.
  • Be careful near cutoffs. Cutoffs for gifted programmes, high-IQ societies and some clinical decisions sit at round numbers such as 130. A measured 128 with a 95 percent interval of roughly 123 to 133 and a measured 132 with an interval of roughly 127 to 137 overlap heavily. Sensible programmes retest, use multiple criteria, or accept a range of evidence. A single point either side of a line is measurement, not destiny.
  • Discount short online scores the most. A ten-question quiz has few items, often no proper norm sample, and no published SEM. Reading its result as exact is the largest version of the mistake this article describes. A margin of about ten points either side is a more honest way to hold a well-built short online estimate, and many quizzes deserve a wider one.
  • Compare like with like. Same test, same edition, similar conditions, enough time between sittings, and intervals checked for overlap. If the intervals overlap substantially, the honest conclusion is that no clear change has been shown.

None of this makes IQ scores useless. At group level, and for broad individual differences, well-built tests are among the most reliable measures psychology has. The error is not in using the number. The error is in using it with more precision than it carries.

Go deeper with AI

This article is the starting point. Copy any of these prompts into your favorite AI assistant to learn more about IQ score margins of error:

🔎 Turn your single score into an honest range

I scored [your score] on [name of test]. Assuming a standard error of measurement of about 3 points for a full battery (or tell me if this test is shorter), calculate my approximate 68% and 95% confidence intervals, convert the ends of the 95% interval to percentiles, and explain in plain language what decisions should and should not be based on a score this close to [any cutoff that matters to you, e.g. 130].

🧭 Check whether two of your scores really differ

I have two IQ results: [score 1] on [test and year] and [score 2] on [test and year]. Walk through every reason they might differ without my ability changing: measurement error and overlapping confidence intervals, practice effects, regression to the mean, different norm years, and different test content. End with a verdict: is this gap larger than measurement noise, and what extra evidence would settle it?

💡 Learn to spot an over-precise IQ claim

Here is an IQ-related claim I saw: [paste the claim, advert or social post]. Audit it like a psychometrician: does it name the test, edition and norm sample, publish a reliability coefficient or standard error, report a range instead of a single number, and avoid comparing scores across different tests? List what is missing and rewrite the claim so it is honest.

Tip: replace the bracketed parts with your own situation — the more specific your prompt, the more useful the answer.

The short summary

Every IQ score is an estimate built from one sitting, a finite set of tasks, and a comparison group. On major full batteries the standard error of measurement is commonly around 2 to 3 points, giving a 95 percent confidence interval of about 5 points either side — so 112 is best read as roughly 107 to 117. Index scores, abbreviated scores and short online quizzes carry wider intervals. Practice (roughly 4 points on a typical retest in the Hausknecht meta-analysis), day-to-day state, regression to the mean, and differences between tests and norm years move scores further. January 2026 retest research (255 adults, 24 measures, two weeks apart) found composites more reliable than single tasks, which is why broad totals deserve more weight than small profile gaps.

The single most important point: never act on an IQ score as if it were exact — read the range around it first, and treat any difference that fits inside that range as no proven difference at all.

Want better prompts, in your inbox?

One sharper AI prompt + one new brain training and IQ topic every week. Free forever, unsubscribe anytime.

We never share your email. One email a week, that's the deal.

Try the IQ Challenge

Take the free 25-question timed quiz and get your estimated IQ and percentile instantly.

Start the Challenge


Keep exploring: see how IQ tests actually work, check the IQ score chart, and read about the WAIS-5 standard.

For entertainment only — this guide is for fun and general information, not a clinical or professional assessment.

© FUNGWA 峯華 LLC (fēng huá — “peaks of splendor”)