Reference

Average IQ by country

This is the most-cited and least-reliable dataset in the field. We publish it because people search for it, and we publish the caveats because nobody else seems to.

Read this before the table

Figures below come from the National IQ dataset compiled by David Becker (the continuation of the Lynn and Vanhanen data). We publish it because people search for it, not because we treat it as settled: the underlying samples differ enormously in size, age range and representativeness, some national estimates rest on a few hundred schoolchildren, and several are extrapolated from neighbouring countries rather than measured. Read the column of caveats before you read the column of numbers.

• Sample sizes differ by orders of magnitude. Some national figures rest on tens of thousands of conscripts tested on a standard instrument. Others rest on a few hundred schoolchildren in one region.
• Some figures are not measured at all. Where no study existed, estimates were imputed from neighbouring countries — a defensible modelling choice that becomes indefensible once the number is quoted as a measurement.
• Age ranges are inconsistent. Studies of schoolchildren and studies of adults are combined into single national figures.
• Instruments differ. Raven's matrices, Wechsler batteries and local tests are pooled despite measuring somewhat different things.
• Health and schooling confound everything. Malnutrition, parasite burden, iodine deficiency, lead exposure and years of schooling all affect measured cognitive performance and all vary systematically with national income.
• The scale itself is relative. If IQ is standardised so the mean is 100, a national figure of 85 is a comparison against a specified reference population — typically a Western standardisation sample — not a property of a country.

The dataset

A representative selection, sorted by reported figure. The caveat column is the part worth reading.

The full dataset covers around two hundred countries and territories. We have not reproduced all of it, because the tail of the list consists largely of imputed values, and reproducing an imputed value in a clean table is how a modelling assumption becomes a fact.

The moving target underneath

Any national figure is a snapshot of a moving quantity, and the movement is large.

Raw IQ performance rose by roughly three points per decade through the twentieth century in every country with long-running data. Countries measured in the 1970s and countries measured in the 2010s are therefore not directly comparable, and a large share of the apparent gap between developing and developed countries in this dataset is a gap in measurement date.

Countries that have industrialised rapidly show correspondingly rapid gains. This is the strongest single argument that the figures track development rather than anything fixed. The full story of the Flynn effect.

What the data is actually evidence for

Used carefully, the dataset does support some conclusions. They are just not the ones it usually gets quoted for.

It documents that measured cognitive performance correlates strongly with national development indicators — schooling years, child nutrition, disease burden, infant mortality. That correlation is robust and uninteresting as a causal claim in either direction on its own.

It documents that gains follow development. Countries that improved child nutrition and expanded schooling show rising figures over successive studies.

And it documents the size of the environmental component in measured intelligence, which is the finding most at odds with how the dataset is normally deployed.

How to handle a national IQ figure when you see one

Four questions, and the answers are usually unavailable, which is itself informative.

1. When was it measured? A figure from 1985 describes a different country.
2. On whom? Schoolchildren in one city are not a nation.
3. With what? Different instruments, pooled, produce a composite of uncertain meaning.
4. Was it measured at all, or imputed from a neighbour?