Average IQ by State
Where the Numbers Actually Come From

0 Quick Answer

Updated August 16, 2026 by Structural. There is no direct measurement of average IQ by US state. Nobody has administered a standardised intelligence battery to a representative sample of every state, and no agency collects such data.

Direct answer: the state IQ tables that circulate are estimates, derived by taking state-level school achievement results, most often from the National Assessment of Educational Progress, and converting them onto an IQ-like scale. The best known academic attempt at this produced estimates spanning roughly the mid-90s to the mid-100s, with New England states at the top and several southern states at the bottom.

The important point is what that spread means. A range of about ten points across all fifty states is smaller than the confidence interval on many individual test scores, and it is dwarfed by the variation within any single state. Knowing which state somebody lives in tells you close to nothing about their cognitive ability.

1 Why No Direct Data Exists

The absence is not an oversight. Producing genuine state-level IQ averages would require something nobody has attempted and nobody is likely to fund.

You would need a representative probability sample within each state, large enough to estimate a mean with useful precision, individually administered by qualified examiners using a standardised battery, under controlled conditions, across all fifty states simultaneously. That is fifty separate standardisation studies, each comparable in scale to what a test publisher conducts once per revision for an entire country.

Test publishers do standardise nationally, and their samples are stratified to represent the population by region among other characteristics. But regional stratification is a design feature to make the national sample representative, not a mechanism for producing regional estimates. The number of cases in any one state is far too small to support a state mean, and publishers do not report them because the estimates would be meaningless.

Nor does any government agency collect it. Cognitive ability testing at population scale is politically fraught, expensive, and serves no administrative purpose that achievement testing does not already serve better. The data that does exist at state level is educational achievement data, collected because education is administered at state level and requires evaluation.

So every state IQ figure in circulation is a conversion of something else. That is not automatically illegitimate, and it does mean the figures inherit every limitation of the source data plus the assumptions of the conversion.

2 How the Estimates Are Actually Built

The standard method, and the one behind the most cited academic attempt, works from the National Assessment of Educational Progress.

NAEP is a federally administered assessment given to representative samples of students in each state, in reading and mathematics, at several grade levels. It is the only nationally comparable achievement measure collected consistently at state level, which is why it is the input to every serious attempt at this question. Its results are public, its sampling is documented, and its methodology is scrutinised.

The conversion works by treating state-level NAEP performance as an indicator of the underlying cognitive ability distribution in that state, standardising the state means, and rescaling them onto a distribution with a mean of 100 and a standard deviation of 15. The most cited published version of this exercise appeared in the journal Intelligence and produced state estimates from that transformation.

Three assumptions are doing the work, and each is contestable.

The first is that achievement measured in schoolchildren indexes cognitive ability in the state's adult population. These are related but distinct, and the relationship is mediated by everything schools do.

The second is that the relationship between achievement and ability is the same in every state. If schooling quality varies, and it demonstrably does, then equal ability produces unequal achievement and the conversion attributes a schooling difference to ability.

The third is that state populations are stable enough for a measure of current schoolchildren to describe the state. Interstate migration is substantial and selective, particularly among the educated, which means the children tested in a state are not a fixed population.

3 What the Estimates Show

The pattern is consistent across published attempts, which is worth stating because consistency is sometimes mistaken for validity.

The highest estimates cluster in New England and the upper Midwest, with Massachusetts, New Hampshire, Vermont, Minnesota, and North Dakota appearing near the top across analyses. The lowest cluster in the Deep South, with Mississippi, Louisiana, and Alabama appearing near the bottom.

The total range across all fifty states is approximately ten points on an IQ-like scale, running from the mid-90s to the mid-100s. Most states sit within a few points of 100, and the distinctions between adjacent states in a ranking are far smaller than the uncertainty attached to either estimate.

The consistency across analyses is real and is easily over-interpreted. Different researchers using the same NAEP data with similar transformations will naturally produce similar orderings. That is a property of the input data, not independent confirmation, and the same ordering appears in straightforward NAEP achievement rankings without any conversion to an IQ scale at all.

What the ordering actually tracks is visible from a different direction. State rankings on NAEP correlate strongly with state median income, educational attainment of the adult population, per-pupil spending, and poverty rates. Any of these would produce approximately the same ordering, which is the central interpretive problem covered in section 5.

4 Within-State Variation Dwarfs Between-State

This is the single most important fact about state IQ tables and the one most consistently omitted from the articles that publish them.

Cognitive ability within any large population has a standard deviation of 15 points by construction. That means the middle two thirds of any state's population spans thirty points, and the range from the 2nd to the 98th percentile spans sixty.

The entire between-state range is about ten points. So the difference between the highest and lowest estimated state is roughly two thirds of one standard deviation, while the spread inside any single state is several standard deviations.

Put concretely: pick two people at random, one from the highest ranked state and one from the lowest. The probability that the person from the lower ranked state has the higher score is close to a coin flip, and that stays true no matter which two states you pick. Group means that differ by a fraction of a standard deviation provide almost no information about individuals drawn from those groups.

This is a general property of overlapping distributions rather than a special pleading about states. It applies equally to any comparison of group averages where the between-group difference is small relative to within-group variation, and it is the reason population-level statistics are informative about populations and close to useless about people.

Anybody using a state ranking to make an inference about a person they have met has made an error of the same kind as reading a percentile as a percentage correct, and the magnitude of the error is larger.

The point holds even at the extremes of the ranking, which is where intuition resists it hardest. Take the highest and lowest estimated states and consider the top few percent of each. Because the distributions overlap almost entirely, the number of very high scoring people in the lower ranked state is not far off the number in the higher ranked one, adjusted for population size. A small shift in a mean barely moves the count in the tail when the standard deviation is unchanged, which is why every state contains large numbers of people at every level of ability and why the ranking predicts nothing about who you will encounter.

5 What the Differences Are Confounded With

Even taking the estimates at face value, attributing them to anything requires disentangling variables that move together across states.

Every one of these correlates with the state ranking, and they correlate heavily with each other. That makes statistical separation extremely difficult with fifty data points, which is the sample size when the unit of analysis is the state.

The selective migration point deserves emphasis because it inverts a causal story people find intuitive. If educated adults move toward economic opportunity over decades, then state differences in measured ability are partly an effect of economic conditions rather than a cause of them. The direction of causation between a state's economy and its measured cognitive profile cannot be established from correlational data at all, and any article implying otherwise is asserting rather than demonstrating.

With fifty observations and a dozen intercorrelated candidate explanations, the honest conclusion is that the ranking is real and its causes are not identifiable from this data.

6 Achievement Is Not Ability

The conversion at the heart of these estimates treats an achievement measure as a proxy for an ability measure. The two are related and they are not the same thing, and the difference matters most exactly where these tables are used.

An achievement test measures what has been learned, and it is designed to be sensitive to instruction, because its purpose is evaluating whether instruction worked. An ability measure is designed to be relatively insensitive to specific instruction, which is why fluid reasoning items use novel material rather than curriculum content.

This design difference means the two respond differently to the same intervention. Improving a school system raises achievement scores substantially, which is the point of doing it. The effect on measures of fluid reasoning is much smaller. A state that improves its schools will rise in a NAEP-derived IQ estimate while the underlying construct the estimate claims to measure moves far less.

The relationship also varies by what is being measured. Achievement tests correlate strongly with crystallised knowledge, which is unsurprising since both are built on acquired information. Their correlation with fluid reasoning is weaker. So a NAEP-derived estimate is closer to a state-level measure of Gc than of general cognitive ability, and treating it as the latter overstates what it captures. The distinction between these abilities is set out in Cognitive Domains.

There is a further asymmetry worth naming. Achievement can be raised deliberately and reasonably quickly, which is the entire premise of educational policy, while measured ability responds far less to the same interventions over the same timescale. A conversion that maps one onto the other therefore produces a metric that policy can move without the thing it claims to measure moving, and one that will drift with schooling changes rather than tracking any stable characteristic of the population. That is not a subtle limitation, it means the converted metric behaves like the achievement data it came from and not like the construct it is named after.

None of this means NAEP is a poor measure. It is an excellent measure of what it measures, which is student achievement in reading and mathematics, evaluated at state level for the purpose of educational policy. The problem arises entirely in the conversion, where a well designed instrument is repurposed to answer a question it was not built for.

7 What NAEP Actually Measures, in Detail

Since every state IQ estimate rests entirely on this one instrument, its specifics determine what the estimates can support. Several of them are not widely known.

NAEP assesses students at grades 4, 8, and 12, in reading and mathematics for the state-level component. It is administered to samples of students within sampled schools rather than to every student, and the state estimates are population estimates from those samples with their own sampling error attached. Those error bars are published alongside the results and are routinely dropped when the data is repackaged.

It is a curriculum-referenced assessment. The frameworks specify what content is assessed at each grade, developed through a governing board process, and items are written to those frameworks. That is the correct design for measuring whether instruction achieved its objectives, and it is why performance responds to what schools teach.

A technical detail with substantial consequences for state comparisons concerns which students are assessed. Students with disabilities and English language learners may be excluded from the assessment or tested with accommodations, and the rates at which states exclude vary. A state excluding a larger share of lower-performing students produces a higher mean than a state excluding fewer, from the same underlying population. NCES has documented this variation and worked to reduce it precisely because it distorts state comparisons.

That single fact undermines fine-grained state rankings more than any of the conceptual objections. When adjacent states differ by less than the combined effect of sampling error and exclusion rate differences, the ordering between them is not measuring anything about the students.

None of this is hidden. NCES publishes the sampling methodology, the standard errors, the exclusion rates, and the guidance on which comparisons are statistically defensible. The information loss happens downstream, when the results are converted, ranked, and republished without any of it.

8 The Regional Pattern Is Older Than the Data

The geography of the ranking is not a recent discovery, and its persistence has explanations that do not require anything about the populations involved.

Public education developed unevenly across the United States. New England established tax-supported common schooling and compulsory attendance decades before much of the country, building an institutional base and a culture of educational participation that compounded across generations. Southern states came to universal public education substantially later and with far less funding, and for a long period operated segregated systems in which resources were distributed radically unequally.

Economic history reinforced the same geography. Industrialisation concentrated in the Northeast and upper Midwest, creating demand for educated labour and the tax base to supply it. Agricultural economies produced neither, and the relationship is self-reinforcing over time because educated adults raise children in better-funded school districts.

Migration then amplified it. The movement of population across the twentieth century was substantial and selective in both directions, and continues today as educated adults move toward economic opportunity. A century of this produces measurable differences in the adult composition of states without any difference in the underlying capability of the people born there.

The reason this matters for interpretation is that it supplies a complete and well documented explanation for the observed pattern that requires no reference to ability at all. Educational investment, economic structure, and selective migration operating across a century and a half will produce exactly the achievement geography that NAEP records. When a sufficient explanation exists and is independently documented, additional explanations are not required by the data.

It also predicts something the ability interpretation does not: that the gap should respond to policy. States that have substantially increased educational investment have moved in these rankings within a generation, which is far too fast for any explanation operating at the level of populations rather than institutions.

9 The Comparison That Is Actually Defensible

Given all of the above, there is a version of this question that the data answers well, and it is worth setting out because it is more useful than the version people search for.

State achievement differences are real, substantial, consequential, and measured properly by an instrument built for the purpose. A student in a low-performing state receives measurably less from their schooling than a student in a high-performing one, and that difference affects their opportunities. This is a genuine finding with policy implications and it requires no conversion to an IQ scale to be important.

The defensible comparisons are the ones NCES itself supports: state against national average, state against itself over time, and state against state where the difference exceeds the published standard errors. The last of these is the crucial constraint and the one that rankings ignore, because it means most adjacent pairs in any ordered list are not distinguishable.

Change over time is the most informative comparison of all and the one least often reported. A state whose scores have risen over two decades has done something, and identifying what is a question with an actionable answer. A state's rank position in a single year is close to uninformative by comparison, because it reflects a century of accumulated conditions rather than anything anybody can currently act on.

The conversion to an IQ scale adds nothing to any of this. It does not improve precision, it does not enable a comparison that was otherwise unavailable, and it does not answer a question the achievement data left open. What it does is change how the numbers are read, from a statement about school systems into a statement about people, and that change is the entire reason the converted version circulates more widely than the original.

Anybody genuinely interested in the underlying question is better served by the NAEP results directly, with the standard errors attached, which are free, public, and considerably more informative than any table derived from them.

10 How These Tables Get Misused

State IQ rankings circulate widely and are used for things the underlying data cannot support. Naming the specific errors is more useful than a general warning.

Ranking states as if the order were meaningful. Adjacent states in these tables differ by fractions of a point, far below the precision of the estimates. Presenting them as an ordered list implies a resolution that does not exist, and a list ordered by NAEP score with no IQ conversion would be more honest and identical in content.

Treating the estimate as a property of individuals. Covered in section 4. The overlap between state distributions is so nearly complete that the state tells you almost nothing about a person.

Inferring causes from the ordering. The confounds in section 5 are inseparable with fifty observations. Any claim that the ranking demonstrates the effect of a particular policy, demographic, or cultural factor is unsupported by the data being cited.

Comparing across sources without checking the method. Different analyses use different NAEP years, grade levels, subject weightings, and transformations. Two tables giving different numbers for the same state are usually not disagreeing about a fact, they are reporting different calculations.

Presenting estimates as measurements. The most common failure. An article headed with a table of state IQ figures, without stating that no state was tested and that the numbers are transformed achievement scores, has misrepresented the entire content of what it published.

The reasonable use is narrow. State-level achievement differences are real, they are worth understanding, and they are best examined as achievement differences using the actual NAEP results rather than through a conversion that adds an unfalsifiable layer and a misleading label.

11 The Same Problem at International Scale

Everything above applies with greater force to the national IQ estimates that circulate even more widely, and the parallel is worth drawing because the two literatures share a method and a set of criticisms.

National IQ datasets are compiled from published studies of varying quality, conducted at different times, on samples of wildly differing representativeness, using different instruments with different norms. Where no study exists for a country, values have been estimated from neighbouring countries. The resulting tables present a single number per country with no indication that the number for one country came from a large representative sample and the number for another was inferred.

The methodological criticisms are extensive and come from within the field. Sample representativeness varies from national probability samples to convenience samples of schoolchildren in one city. Instruments differ in what they measure and how their norms were built. Data collection years span decades during which population performance was drifting, as covered in Average IQ by Generation. And the imputation of missing values from geographic neighbours builds the assumed conclusion into the data.

The state-level version at least uses one consistent instrument administered under one protocol in the same year, which makes it considerably better than the international version. It inherits the achievement-versus-ability problem and the confounding problem regardless.

The general lesson from both literatures is that group averages built by combining heterogeneous sources answer questions much less precisely than a single number implies, and that the confidence with which such tables are presented is inversely related to the strength of what lies behind them.

12 What This Means for You Personally

The practical takeaway is short and worth stating plainly for anybody who arrived here after seeing their state near the bottom or the top of a list.

Your state's estimate says nothing about you. It was derived from schoolchildren's achievement scores, transformed by assumptions you did not participate in, and the between-state spread is a fraction of the spread within your own neighbourhood.

If you want to know something about your own cognitive profile, the way to find out is to be measured. An individual assessment compares you against an age-matched reference sample and reports where you sit with a confidence interval attached, which is a fundamentally different kind of statement from a group average.

The more interesting output of an individual assessment is not the composite anyway. It is the profile across domains, showing where your abilities diverge from each other, which is information no group statistic contains at any level of aggregation.

ACIS reports six domains with percentiles and confidence intervals, against a stated reference group, with its norms documented in technical materials. Administration is unsupervised, which is a real limitation stated in the report rather than around it, and it answers a question about you rather than about where you happen to live.

One further thing worth saying to anybody who found this page while feeling defensive about where they are from. The achievement differences behind these tables are about school systems, funding, and a century and a half of uneven institutional investment. They are not a statement about the people who live in a place, and treating them as one gets both the statistics and the history wrong.

13 FAQ: State IQ Estimates

14 Best Next Step

State IQ tables are transformed achievement data presented with a label the data does not support. The achievement differences behind them are real and worth understanding, and they are best examined as what they are.

If you want to know what the average actually means as a statistical concept, read Average IQ. For why group averages say so little about individuals, read IQ Score vs Percentile. For the generational version of the same question, read Average IQ by Generation. For something about you rather than your postcode, take the assessment.

15 Sources Behind This Page

The underlying achievement data is public. The conversion to IQ-like scales comes from published academic work, and the criticisms of that conversion come from the same literature.

• National Center for Education Statistics. The Nation's Report Card, National Assessment of Educational Progress. The state-level achievement data that every state IQ estimate is derived from, with published sampling methodology.
• McDaniel, M.A. (2006). Estimating state IQ: measurement challenges and preliminary correlational results. Intelligence, 34(6), 607-619. The most cited attempt to derive state IQ estimates from NAEP data, including its own account of the measurement challenges.
• National Center for Education Statistics. Understanding NAEP scale scores. What the achievement scale represents and why it is not an ability metric.
• American Educational Research Association, American Psychological Association & National Council on Measurement in Education. Standards for Educational and Psychological Testing. Requirements governing what inferences a test score supports and the distinction between achievement and ability measures.
• McGrew, K.S. (2009). CHC theory and the human cognitive abilities project. Intelligence, 37(1), 1-10. The distinction between crystallised knowledge and fluid reasoning that determines what an achievement-derived estimate actually captures.
• Trahan, L.H., Stuebing, K.K., Fletcher, J.M. & Hiscock, M. (2014). The Flynn effect: a meta-analysis. Psychological Bulletin, 140(5), 1332-1360. Population-level drift, which complicates any comparison of estimates built from data collected in different years.
• Ritchie, S.J. & Tucker-Drob, E.M. (2018). How much does education improve intelligence? A meta-analysis. Psychological Science, 29(8), 1358-1369. Quantifies the effect of schooling on cognitive test performance, which is the mechanism confounding achievement-derived ability estimates.
• Voncken, L., Albers, C.J. & Timmerman, M.E. (2019). Improving confidence intervals for normed test scores. Behavior Research Methods. Open access. Why the between-state spread is smaller than the uncertainty on many individual scores.
• United States Census Bureau. Educational attainment data by state. One of the confounds that tracks the state ranking closely and cannot be separated from it with fifty observations.
• United States Census Bureau. Income and poverty data by state. The economic variables that correlate with state achievement rankings and complicate any causal reading.
• Crawford, J.R., Garthwaite, P.H. & Slick, D.J. (2009). On percentile norms in neuropsychology: proposed reporting standards. The Clinical Neuropsychologist, 23(7), 1173-1195. Why a point estimate without an interval overstates precision, which applies to group means as much as to individual scores.
• Buros Center for Testing. Mental Measurements Yearbook. Independent evaluation of what published instruments do and do not support, including the boundary between achievement and ability measurement.