IQ vs GRE: Overlap Without Conversion

0 The Short Answer

The GRE General Test is not an IQ test, and no legitimate conversion table between the two exists. It is an admissions instrument published by ETS, designed to help graduate programs forecast how applicants will perform in coursework and research. Its scores are compared against other people who registered to take it, not against the general adult population. Its scale, 130 to 170 per section, has no arithmetic relationship to the mean of 100 and standard deviation of 15 that cognitive batteries use. And a candidate can prepare for it, sit it up to five times a year, and then choose which attempt the schools ever see.

Dismissing it as an empty hoop would be the opposite error. Admissions tests of this family do load heavily on general cognitive ability, and the research showing that is strong enough that anyone claiming the GRE General Test measures nothing but coaching is arguing against the evidence. A high score is genuinely informative. It is just not informative in the specific way an IQ estimate is informative, and the gap between those two kinds of information is what this page is about.

Most of what circulates online skips that gap entirely. Search for a conversion and you will find tidy tables mapping a 165 Quantitative Reasoning score onto some three digit figure, usually presented with the confidence of a currency exchange rate. Those tables are constructed, not measured. They exist because people want the number, and because nothing stops anyone from publishing a column of digits next to another column of digits. Understanding exactly why the mapping fails is more useful than the mapping would have been, because the same reasoning applies to the SAT, the LSAT, the MCAT and every other admissions instrument someone has tried to launder into an IQ figure.

1 What the GRE General Test Actually Contains

Describing the exam accurately matters, because the version most articles describe stopped existing in September 2023. ETS shortened the test that month, cutting it from close to four hours to under two, and a large amount of published commentary still refers to the old shape.

The current GRE General Test runs about 1 hour and 58 minutes and contains three measures across five scored blocks. Analytical Writing has been reduced to a single "Analyze an Issue" essay with 30 minutes allotted. Verbal Reasoning is split into two sections, 12 questions in 18 minutes followed by 15 questions in 23 minutes. Quantitative Reasoning is also split in two, 12 questions in 21 minutes followed by 15 questions in 26 minutes. That is 54 scored questions in total plus one essay. The unscored experimental section that used to pad the sitting was dropped in the same revision.

Two design features deserve attention because they shape what the score can mean. First, the test is section level adaptive rather than question level adaptive. ETS states that the second section of each measure is selected based on overall performance on the first. This is coarser than the item by item adaptation used by professional cognitive batteries, where each item is chosen against a running ability estimate.

Second, candidates can move freely inside a section, skip questions, flag them and come back. That is a deliberate concession to test taking comfort, and it makes the administration feel less like a measurement session and more like an exam. The difference is not cosmetic. Adaptive batteries hold administration conditions rigid precisely so that scores from different people mean the same thing, and every degree of freedom handed to the candidate is a degree of freedom that strategy can exploit.

2 What Admissions Committees Actually Do With the Score

The purpose of an instrument constrains what its output can mean, so it is worth stating plainly what graduate programs are trying to accomplish when they ask for scores.

A committee reading a file has a forecasting problem. It has undergraduate grades from hundreds of different institutions with incompatible grading cultures, letters of recommendation written by people with incompatible enthusiasm calibrations, and a personal statement that has usually passed through at least one editor. None of those inputs are on a common scale. An admissions test is the one number in the file that came from the same instrument, under the same conditions, for every applicant. Its job is comparability, not depth.

What the committee wants to predict is narrow: will this person survive coursework, pass qualifying exams, produce research, and finish the degree. Nothing in that list requires knowing where the applicant sits relative to the general population. A department admitting six students out of two hundred applicants only needs to rank the two hundred. Whether the whole pool sits in the top quarter of the population is irrelevant to the decision, and the test is not built to tell them.

This is the design difference that most conversion attempts never account for. A cognitive battery is built to place an individual against a representative sample of everyone. An admissions test is built to spread out a specific, self selected, already filtered group. Those are different engineering problems, and they produce different instruments even when the underlying question types look similar on the page. A quantitative comparison item and a matrix reasoning item can both tap fluid reasoning, but the surrounding measurement apparatus decides what the resulting number is licensed to say.

There is also a policy layer that has nothing to do with measurement. Many departments have made score submission optional over the past several years, some have dropped it entirely, and others use it mainly as a screen for funding decisions. The weight assigned to the number varies enormously across fields and institutions, which is another way of saying that even inside its intended use the score is not a fixed quantity of evidence.

3 The Scale Has No Relationship to the IQ Scale

Start with the arithmetic, because it disposes of a surprising fraction of the conversion tables on its own.

Verbal Reasoning and Quantitative Reasoning are each reported from 130 to 170 in one point increments. That is a 41 point range, floor to ceiling, and 130 is the score you receive if you answer nothing correctly. Analytical Writing is reported from 0 to 6 in half point steps, giving 13 possible values. The modern IQ scale, by contrast, is a deviation scale with a mean fixed at 100 and a standard deviation fixed at 15, and it extends in principle without a hard ceiling, which is why percentile tables can meaningfully distinguish 145 from 160.

These scales are not merely different in their endpoints. They are different in kind. On the IQ scale, the distance between two scores has a fixed meaning in standard deviation units, by construction. On the admissions scale, the distance between 160 and 165 depends entirely on where those scores fall in the test taker distribution, and that distribution is neither normal in shape nor stable across measures. Quantitative Reasoning in particular bunches heavily near the top, so a five point move at the ceiling covers a very different slice of candidates than the same five points near the middle.

The history makes the arbitrariness obvious. Before the August 2011 revision, the sections were reported on a 200 to 800 scale in 10 point increments, inherited from the same scaling tradition as the SAT. ETS compressed that into 130 to 170 partly to stop admissions readers from over interpreting small differences that were inside the measurement error. A candidate whose ability never changed would have carried a 700 in 2010 and something in the low 160s in 2012. If a number can be rewritten by an administrative decision without the underlying person changing at all, it is a reporting convention, not a physical quantity.

That same 2011 revision also removed antonym and analogy items from Verbal Reasoning and moved the test from question level adaptation to section level adaptation. Any conversion table built on pre 2011 data, and a striking number of the ones still circulating were, is describing an instrument that no longer exists in either its content or its scoring.

4 The Norm Group Is the Applicant Pool

This is the single most important reason the exam cannot function as an IQ estimate, and it is the reason almost no popular article mentions.

A percentile only means something once you know the comparison group. ETS publishes percentile ranks in a document called GRE General Test Interpretive Data, and the comparison group is the population of people who took the test during a recent multi year window. The company also publishes separate distributions of scores within intended broad graduate major field and within roughly 300 individual major fields, which tells you how seriously the differences between applicant subgroups are taken. For the July 2015 to June 2018 window covering around two million test takers, the reported means were about 150 on Verbal Reasoning and about 153 on Quantitative Reasoning.

Now consider who is in that group. Everyone in it completed or is completing an undergraduate degree. Everyone in it decided to pursue graduate study. Everyone in it paid a registration fee in the region of two hundred dollars and gave up a morning. That is a heavily filtered slice of the adult population, filtered on exactly the traits the test measures. A person at the 50th percentile of this group is not at the 50th percentile of adults in general. They are somewhere well above it, and the exact distance is not something ETS publishes, because ETS has no reason to measure it.

Compare that with how a cognitive battery establishes its average. The norm sample is drawn to represent the general population on age, sex, education, region and other stratification variables, specifically so that a score of 100 means the median adult. Building such a sample is expensive and slow, and it is the entire reason the resulting score can be spoken about in population terms.

So when someone converts a 90th percentile Quantitative Reasoning score into an IQ figure, they have quietly performed an illegal move: they took a percentile within graduate applicants and read it off a table built for percentiles within the general population. The two percentiles share a word and nothing else. This one substitution accounts for most of the inflation in the conversion tables you will find, and it is why those tables so often produce cheerful figures in the 140s for scores that thousands of people achieve every year.

5 What the Research Says About Admissions Tests and g

Having established what the score cannot do, honesty requires stating clearly what the evidence does show, because the relationship between admissions tests and general cognitive ability is real and substantial.

The cleanest study in this literature examines the SAT rather than the graduate exam. Frey and Detterman, writing in Psychological Science in 2004 under the title "Scholastic Assessment or g?", took 917 participants from the National Longitudinal Survey of Youth 1979, extracted a general ability factor from the Armed Services Vocational Aptitude Battery, and correlated it with SAT scores. The correlation was .82, rising to .86 after correction for nonlinearity. Their conclusion was blunt: the test is mainly a measure of general cognitive ability.

That result generalizes to the graduate exam by family resemblance rather than by direct measurement, and the distinction is worth preserving. The GRE General Test samples the same broad territory, verbal comprehension through vocabulary and reading, quantitative reasoning through mathematics and data interpretation, with heavy demands on working memory under time pressure. Nobody who understands psychometrics doubts that it loads substantially on g. But "loads substantially on g" and "is an estimate of g on a known scale" are separated by everything discussed in the previous two sections.

Notice also what the Frey and Detterman result implies about the direction of the error. If an admissions test is largely a g measure, then the reason it cannot be converted to IQ is not that it measures the wrong thing. It measures a great deal of the right thing. The obstacle is entirely in the scaling and the reference population, which is a much more interesting problem than "the test is fake" and a much harder one to wave away.

One more piece of that 2004 paper is directly relevant here, and it is the piece that leads into the next section. The same study ran a second sample, 104 undergraduates, and correlated revised SAT scores with Raven's Advanced Progressive Matrices. The correlation there was .483, not .82. Same construct, same kind of test, half the correlation. The reason is the concept that anyone arguing about admissions tests and intelligence needs to hold in their head.

6 Range Restriction, With a Worked Example

Range restriction is what happens when you measure a relationship inside a group that has already been selected on one of the variables. The relationship does not disappear, but the number you compute shrinks, sometimes dramatically, and the shrinkage is a property of your sample rather than of the world.

Take the non psychometric version first. Suppose you want to know how strongly height predicts rebounding in basketball. Measure it across a random sample of adults and you will find an enormous correlation, because the sample contains people of every height and the short ones cannot rebound at all. Now measure the same relationship among professional centers. Everyone in that sample is between about 6 foot 9 and 7 foot 2. The height variable barely varies, so it can barely explain anything, and your correlation collapses toward zero. Height did not stop mattering. You removed the variation that made it visible.

Now the psychometric version, using the numbers from the study cited above. In the broad national sample, where participants ranged across the full ability distribution, the SAT correlated .82 with a general ability factor. In the undergraduate sample, where everyone had already been admitted to a university and therefore already cleared a substantial ability bar, the correlation with Raven's matrices was .483. Correcting that second figure for the restricted range brought it back to .72. The underlying relationship was always strong. The restricted sample just could not see it.

Every study of the GRE General Test faces this problem in an especially severe form, because the samples available to researchers are usually students who were admitted, which means selected partly on the very scores being studied. That is double restriction: filtered once by who chooses to apply to graduate school, and again by who gets in. Any raw correlation computed in such a sample is an underestimate of the population relationship, and by an amount that depends on how selective the programs were.

The practical lesson cuts both ways, which is why it is the concept to take away from this page. When someone tells you an admissions test correlates only .3 with some outcome and therefore measures nothing, ask what sample that came from. When someone tells you the corrected correlation is .7 and therefore the test is an IQ measure, remember that a correction is a statistical estimate of what would have happened in a population nobody actually tested. Both the raw number and the corrected number are answers to different questions, and neither one hands you a conversion formula.

7 What the Score Predicts, and How Well

The evidence on predictive validity is unusually good for this exam, because one meta-analysis dominates the field and it is large.

Kuncel, Hezlett and Ones published "A comprehensive meta-analysis of the predictive validity of the Graduate Record Examinations" in Psychological Bulletin in 2001, drawing on 1,753 independent samples, 6,589 correlations and 82,659 graduate students. They found that scores predicted graduate grade point average, first year performance, comprehensive examination results, faculty ratings and degree attainment, with positive relationships to research productivity as well. Kuncel and Hezlett returned to the topic in Science in 2007 with a defense of standardized testing in graduate admissions that reached a similar conclusion.

The magnitudes are worth stating carefully, because both the promoters and the abolitionists misquote them. Validity coefficients against graduate grades typically land in a band from roughly .30 to .45 once corrections are applied, which means the test accounts for a meaningful minority of the variance in academic outcomes and leaves most of it to other things. That is a genuinely useful signal for an admissions committee sorting hundreds of files, and a genuinely weak basis for any statement about an individual.

Two details from that meta-analysis complicate the popular picture. The Subject Tests, which assess knowledge of a specific discipline, often predicted better than the general measures did. That is an awkward finding for the view that the general exam works because it captures raw ability, since a content specific achievement test outperforming it points toward prepared knowledge doing real predictive work. And the criteria being predicted, grades and faculty ratings, are themselves academic performance measures, which is what the instrument was built for.

None of this is a knock on the exam. An admissions test that predicts academic outcomes at those magnitudes is doing its job. But look at the shape of the claim being supported: this score forecasts performance in this environment over the next few years. That is a forward looking, context specific statement about behavior. An IQ estimate is a backward looking, context free statement about a person's standing on a construct. A tool can be excellent at the first job while being unlicensed for the second, and this one is. If you want to see how the second kind of claim is evaluated, the literature on cognitive ability and job performance is a cleaner place to look.

8 Preparation, Retakes and Score Selection

Three features of how the exam is delivered would each, on their own, disqualify a score from serving as a clean ability estimate. Together they settle the question.

The test is prepared for, intensively. An entire industry sells courses and materials, and ETS itself distributes free POWERPREP practice tests so candidates can rehearse the interface, the timing and the item formats. That is a reasonable thing for a test publisher to do when the goal is fair access to a fair exam. It is also an explicit acknowledgment that familiarity with the instrument moves the score. A measure that improves when you study the measure is contaminated as an ability estimate by exactly the amount it improves, and that amount is not reported on anybody's score sheet.

The test can be repeated. ETS permits a sitting once every 21 days, up to five times in any rolling 12 month period. Even setting aside genuine learning, repeated attempts at a timed exam produce score movement through practice effects, familiarity and simple variance. Take any measurement five times and report the best result and you have reported an upper tail of a distribution, not a central estimate. Professional cognitive assessment treats this so seriously that retest intervals and alternate forms are built into administration rules precisely to keep practice effects out of the number.

The candidate chooses what schools see. Under ETS ScoreSelect, a test taker can send scores from a single administration of their choosing, or from any subset of administrations within the five year reporting window. The score a program receives is therefore not a sample of the applicant's performance. It is the applicant's own selection from their performances, which is a maximum, not a mean.

Stack the three. A candidate studies for the format, sits the exam three times, and sends only the best result. The reported number is the top of three attempts by a prepared test taker. Compare that with a supervised battery administered once, under fixed conditions, with the profile reported in full including the parts that came out weak. The second procedure is trying to estimate something. The first is trying to present something, and that is not a criticism of applicants, it is a correct description of what the system asks them to do.

9 Why No Legitimate Conversion Table Can Exist

Pull the previous sections together and the impossibility becomes structural rather than a matter of nobody having done the work yet. The same proof runs for the GMAT, and even for the ASVAB, the one test in this family that comes closest to a cognitive battery and still does not convert.

A valid conversion between two scales requires three things. You need both instruments administered to the same people, in a sample large enough and representative enough to anchor the mapping. You need the two scales referenced to a common population, or a documented bridge between their reference populations. And you need the instruments to measure a sufficiently similar construct that a single number can travel between them without losing its meaning.

The graduate admissions exam fails the first two outright. ETS does not co-administer cognitive batteries to its test takers and does not publish general population norms for its scores, because neither activity serves its business or its stated purpose. Without a linking sample there is no empirical basis for a mapping, and without general population norms there is nothing on the other end to map onto. Any table you find has therefore been built by assumption, and the usual assumption is the one identified earlier: treat applicant pool percentiles as if they were population percentiles and read across.

The third requirement is where it gets more interesting, because this is the one the exam comes closest to satisfying. The instrument really does tap verbal comprehension and quantitative reasoning, and it really does load on general ability. But even a perfect construct match would not rescue the conversion, because scaling and reference population are separate problems from construct validity. Two thermometers can both measure temperature honestly and still disagree completely if one reports Celsius against sea level calibration and the other reports an arbitrary index calibrated against a group of people standing in a sauna.

There is a further problem that no correction can fix. A single admissions score, or even three of them, compresses several distinct abilities into one figure. Structured assessment separates verbal comprehension, fluid reasoning, visual spatial processing, working memory, processing speed and quantitative reasoning because those domains dissociate within individuals. A candidate with excellent verbal comprehension and mediocre processing speed and one with the reverse profile can post identical section scores. Converting either to a single IQ figure discards the information that actually distinguishes them, which is the same information a domain based assessment exists to produce.

10 Where the Internet Tables Come From

Knowing that the tables are invented is useful. Knowing how they get invented is what stops you from being fooled by the next one. The obsolete anchor trick runs even harder on the undergraduate exams: the ACT comparison notes that science and writing are now optional and no longer feed the Composite, so any table descended from the older four section Composite is describing an instrument that no longer exists.

The percentile swap. The most common construction takes a score, looks up its percentile among test takers, then reads that percentile off a normal curve with mean 100 and standard deviation 15. It looks rigorous because both steps use real numbers. It is invalid because the two percentiles refer to different populations, and the error is systematically in one direction: it inflates, often by a large margin.

The obsolete anchor. A second family of tables descends from work done on the pre 2011 scale, when sections ran 200 to 800 and the item mix was different. Some of these trace back further, to eras when the exam included an Analytical Ability section that was replaced by Analytical Writing in October 2002. Tables built on those foundations are describing instruments that have been redesigned twice since.

The society threshold. High IQ societies have historically accepted admissions test scores as qualifying evidence, and their published cutoffs get scraped and reversed into general conversion tables. Those cutoffs were set by the societies for their own admission purposes, often decades ago on scales that no longer exist, and they were never intended as a measurement bridge in either direction.

The single study extrapolation. The most sophisticated version takes a real finding, usually the SAT result discussed earlier, and extends it to the graduate exam by analogy. This is the most defensible of the four and still does not produce a table, because a correlation between two variables does not give you a conversion between their scales. Correlation tells you they move together. Conversion requires knowing the intercept, the slope and the reference population, and a correlation coefficient contains none of those.

A quick test for any table you encounter: does it state which edition of the exam it applies to, which normative sample the IQ side comes from, and what linking data produced the mapping? Legitimate score linkages publish all three, because those details are the work. The tables in question publish none of them, which is the tell.

11 Side by Side With a Cognitive Battery

Laid out against each other, the two instrument types differ on every dimension that determines what a score licenses you to say.

Read the table as a description of two competent tools rather than a winner and a loser. If your question is which of these two hundred applicants will finish a doctorate, the left column is the better instrument and the right column would be an odd and possibly unlawful thing to demand. If your question is where a person's reasoning sits relative to adults in general and which of their cognitive domains is relatively strongest, the right column is the only one of the two that answers, and the left column cannot be persuaded to.

The mistake worth avoiding is treating one as a cheap proxy for the other. A high score in the left column is real evidence of capability, particularly in the quantitative domain, and it deserves to be read that way. It just cannot be turned into a figure that means what people think a good IQ score means, and no amount of arithmetic performed on it afterward will change what it was built to do.

12 If You Want an Estimate of Your Cognitive Profile

People searching for a conversion are usually after something specific and reasonable. They did well or badly on an admissions exam, and they want to know what that says about them in a broader sense. The exam cannot answer that, but the question has a real answer, and getting it requires an instrument built for it.

Three properties separate a defensible estimate from a number that flatters you. The first is declared norms: the assessment should tell you who the comparison group is and how the scores were derived, so you can judge whether the reference population is one you care about. Any assessment that reports a percentile without naming its comparison group is telling you nothing, for the same reason applicant pool percentiles told you nothing about the general population.

The second is domain structure. A single number hides the pattern that makes a cognitive profile informative. Separate scores for fluid reasoning, verbal comprehension, quantitative reasoning, working memory, visual spatial processing and processing speed let you see relative strengths, which is what people are usually curious about even when they ask for one figure. It also stops a single weak subtest from silently dragging down an impression of the whole.

The third is honest uncertainty. Every score is an estimate with error around it, and an assessment that reports a bare point value is hiding the width of that band. Intervals are not a hedge, they are part of the result, and their absence from a report is a signal about how seriously the report takes measurement.

ACIS is built on those three principles. It is an online self assessment rather than a clinical or diagnostic evaluation, and it should be read that way, but it administers 20 subtests across six cognitive domains, reports a profile with intervals rather than a lone number, and states the basis of its norms. If the admissions score you were trying to convert came out strong, a domain profile will show you where that strength actually sits. If it came out weak, the profile will usually show you why, and the reason is rarely uniform across domains.

13 Frequently Asked Questions

14 Sources Behind This Page

Comparing tests means comparing what each one measures. The sources below cover the official test documentation and the research on how admissions scores track cognitive ability.

• Educational Testing Service. The official GRE documentation: what the test contains and how it is scored.
• Kuncel, N.R., Hezlett, S.A. & Ones, D.S. (2004). Academic performance, career potential, creativity, and job performance: can one construct predict them all? Journal of Personality and Social Psychology, 86(1).
• Frey, M.C. & Detterman, D.K. (2004). Scholastic Assessment or g? Psychological Science, 15(6). The study linking SAT scores to measured cognitive ability, and the reason SAT conversions are estimates rather than measurements.
• Koenig, K.A., Frey, M.C. & Detterman, D.K. (2008). ACT and general cognitive ability. Intelligence, 36(2). The companion study on how ACT scores track g.
• Study of Mathematically Precocious Youth, Vanderbilt University. A 50 year longitudinal study of five cohorts totalling over 5,000 individuals, identified between the top 3% and the top 0.01% of ability and followed for decades.
• Voncken, L., Albers, C.J. & Timmerman, M.E. (2019). Improving confidence intervals for normed test scores. Behavior Research Methods. Open access. Documents the mean 100, SD 15 metric and the uncertainty that norming from samples adds to any score.
• Crawford, J.R., Garthwaite, P.H. & Slick, D.J. (2009). On percentile norms in neuropsychology. The Clinical Neuropsychologist, 23(7), 1173-1195. Three definitions of a percentile coexist in practice and can return different results for the same score.
• Pearson Clinical Assessment Scientific Council (2023). Standardized Clinical Assessment for Practitioners: A Primer. How standard scores, percentile ranks and the standard error of measurement are meant to be read together.
• Pearson (2008). WAIS-IV Score Report sample. What a real report contains: every composite paired with a percentile rank, a 95% confidence interval and a qualitative description, never a bare number.
• Trahan, L.H., Stuebing, K.K., Hiscock, M.K. & Fletcher, J.M. (2014). The Flynn effect: A meta-analysis. Psychological Bulletin. Open access. Scores rose about 2.31 points per decade across 285 studies, which is why norms age and older scores overstate standing.