When IQ scores are described, the reference distribution is usually the general population of the country where the test was normed. Mean 100, standard deviation 15, distributed approximately normally. This is the framework most discussions of cognitive testing assume. But for someone trying to understand a specific result, the question of which population the percentiles reference matters more than people usually appreciate. The "84th percentile" against the general population is a different statement than "84th percentile" within a graduate-level engineering program or within an applicant pool for a competitive job.
This piece is about how IQ distributions actually shift across populations, why this matters for reading any specific result, and what the appropriate cautions are.
Every standardized cognitive test is normed against some reference population. The norms are derived from administering the test to a representative sample, computing the score distribution, and then setting up the scoring system so that the average in that sample equals 100 with standard deviation 15.
The choice of reference population matters considerably. Common reference populations include:
For most reported results, the comparison is to a national general-population norm. A reasonable guide to interpreting IQ scores should specify what population the percentiles reference. The standardization process is documented in detail in published test manuals.
Mean IQ scores differ across various sub-populations, often in ways that reflect what the populations are selected on. Some of these differences are large enough to matter for interpretation.
The practical upshot: an IQ percentile result depends not just on the score but on which population the percentile is calculated against. Tests typically disclose this; users typically don't notice.
A few situations where reading the reference population matters:
You're comparing scores from different tests. Two online tests can produce different percentile ranks for the same person if their norm groups differ. This is normal and not evidence that either test is wrong. The same raw cognitive performance can sit at different percentiles in different reference populations.
You're using a score to predict something. If a test was normed on the general population but you're interested in performance within a competitive context (say, a competitive graduate program), the percentile against the general norm overstates your position within the relevant comparison group. A score at the general-population 90th percentile may be at the 50th percentile within an admitted cohort.
You're comparing yourself to others. When someone reports their IQ percentile, the implicit reference is usually the general population. Two people in a competitive context can both be at "the 95th percentile" against the general population while having quite different positions within their specific peer group.
You're interpreting an online result. Online tests using self-selected samples for norming will produce different percentile interpretations than tests normed against representative national samples. The numbers can look similar but mean different things.
Even within a defined population, the variance in cognitive scores is substantial. Knowing the mean of a group tells you what to expect on average; it doesn't tell you much about any specific individual within the group. Some illustrations:
This is the standard statistical observation that between-group differences are often smaller than within-group differences. It applies to cognitive scores across most population comparisons, and it's a useful corrective to the assumption that group membership tells you much about an individual's score.
When you encounter an IQ result — yours or someone else's — the honest reading practice involves a few quick questions:
These questions don't take long to ask, and they prevent most of the common misreadings of cognitive scores. They also make conversations about scores less confused, because the implicit comparison group is made explicit. Resources on intelligence theory and measurement provide deeper context for users who want to understand the technical basis.
IQ distributions vary meaningfully across populations, and the reference group behind any reported percentile matters considerably for what the score actually says. The same raw cognitive performance sits at different percentiles in different reference populations, and within any defined population the variance is substantial. The honest reading practice involves checking what comparison group the percentile references, recognizing that group means don't predict individual scores, and treating a single percentile as a partial summary rather than a complete cognitive description. Once you internalize that all percentile ranks are conditional on a comparison group, most of the common over-readings of cognitive scores quietly resolve.
Yes. Most major tests are normed against the country where they're administered. Norms shift over time and across populations, which means cross-country and cross-cohort comparisons require careful attention to the specific norm group being used. The percentile interpretation is conditional on the reference population.
Several reasons can contribute: different norm groups, different reliability properties, different item content emphasis, and different testing conditions. Most well-designed instruments will produce results within several points of each other for the same person, but exact agreement isn't expected, and the gap doesn't necessarily mean either test is wrong.
Substantially. Professional groups vary in mean IQ from around 100 for occupations with no specific cognitive selection up to around 125-130 for highly selective intellectual professions. Within-group variance is typically large — there's substantial overlap between the score distributions of different occupations.
In raw cognitive performance terms, mean scores rose substantially through the twentieth century in many countries. In percentile terms, the same person retains roughly the same position because norms get updated. The raw performance behind "IQ 100" today represents higher absolute cognitive performance than "IQ 100" represented in 1960, though both correspond to the population median at their respective times.