Verbal vs quantitative ability: math person or words person?

0 The short answer

A math person or a words person is not a type of person but a position on a continuum: tilt, the difference between a person's quantitative score and verbal score, predicts which field someone enters better than it predicts how much they achieve there. In a follow-up of 2,409 intellectually talented youth who took the SAT before age 13, the math-minus-verbal gap separated later STEM careers from humanities careers, while overall ability level did not. Critics argue that much of that signal is a statistical artifact of subtracting two correlated scores, and any gap you see on your own report carries more measurement error than either score alone.

1 What Do Verbal Ability and Quantitative Ability Mean in the CHC Model?

In the Cattell-Horn-Carroll (CHC) model, verbal ability corresponds to comprehension-knowledge (Gc) and quantitative ability to quantitative knowledge (Gq), two broad abilities that are related but separable. McGrew's review of the framework describes CHC as the consensus psychometric structure that merges the Cattell-Horn Gf-Gc model with Carroll's three-stratum model, and it recommends the CHC taxonomy as a common vocabulary for research findings. Gc is the breadth and depth of acquired knowledge of language and the culture, the kind of ability that vocabulary, word-opposite and general-information tasks sample. Gq is acquired knowledge of mathematics and facility with number, the kind of ability that arithmetic and mathematics-achievement tasks sample. The page on the CHC model lays out all the broad abilities, and the page on fluid versus crystallized intelligence explains why Gc is the crystallized side of the pair.

Two cautions keep the labels honest. First, "quantitative" is used in two ways. Tasks that ask a person to compute, or to recall mathematical procedures and facts, lean on Gq. Tasks that ask a person to reason about novel quantitative relations, such as completing a numeric series nobody has seen before, lean on fluid reasoning (Gf). The page on the quantitative reasoning test separates those two uses of one name, and the section on reading an ACIS report below returns to why it matters. Second, in the CHC framework both Gc and Gq are abilities expressed through knowledge that schooling and reading build up. That makes them sensitive to how much a person has learned and practiced, which is a different thing from the speed or power of the underlying reasoning. The page on intelligence versus knowledge develops that distinction.

An ACIS report puts the two abilities on adjacent lines. According to the ACIS home page, read on October 20, 2026, the Verbal Comprehension Index (VCI) is built from Antonyms, Vocabulary, Information, Synonyms and Similarities, and the Quantitative Reasoning Index (QRI) from Mathematical Achievement and Arithmetic. Each is reported on the standard scale with a mean of 100 and a standard deviation of 15, with a percentile and a 95 percent confidence interval. The page on what a verbal IQ test measures describes how a verbal index is built, and the pages on the Vocabulary and Mathematical Achievement subtests describe the formats of one task from each side. Those pages describe formats only, and this page does the same.

The rest of this page uses "verbal" and "quantitative" for the two scores that research compares, usually SAT or ACT sections, and says Gc and Gq when the claim is about the CHC abilities themselves. That distinction matters because most of the best evidence on the math person idea comes from college admission tests, which sample school-taught knowledge heavily.

2 Is There a Math Person and a Words Person?

There is no evidence for two kinds of people, and the data point the other way: skill in language and skill in mathematics are positively correlated, so most people who are strong in one are above average in the other. The largest synthesis is Peng and colleagues' 2020 meta-analysis of the relation between language and mathematics. It pooled 344 studies with 393 independent samples and more than 360,000 participants and found a correlation of .42, with a 95 percent confidence interval of .40 to .44. That is a moderate correlation, which means two things at once. Language and mathematics are far from the same skill, because a correlation of .42 leaves most of the variance unshared. They are also far from independent, because people who read and speak well tend to do better in mathematics than people who do not.

The same meta-analysis offers a partial explanation. Working memory and intelligence together explained over 50 percent of the variance in the language and mathematics relation, and they explained more of it for complex mathematics skills. In other words, a large part of the overlap between a words score and a numbers score is general cognitive ability, which the page on the g factor describes. In the longitudinal studies the authors pooled, language and mathematics each predicted the later development of the other even after controlling for initial performance. A person who is strong in one tends to have an easier time building the other.

So where does the "math person" idea come from? It comes from within-person comparison. Because most people have some gap between their two scores, almost everyone can say that one subject has come more easily to them. That observation is correct and does not need a theory of types. What the research asks is narrower and more testable: when the gap is large, does it predict anything about the person's later choices and outcomes, and is the gap itself a real property or an artifact of how scores are subtracted? Those are the questions the next sections take in order.

The idea of a hemispheric split, with the left brain for words and numbers and the right brain for creativity, is a separate folk theory. The page on left brain versus right brain covers why it fails, and nothing in the tilt literature depends on it. Tilt is defined from test scores, and no study on this page used a brain measure.

3 What Is Ability Tilt and How Is It Computed?

Ability tilt is the within-person difference between two ability scores, most often the math score minus the verbal score on an admission test, so that a positive value means math is the stronger of the two. Wai, Hodges and Makel state the computation plainly: tilt equals math minus verbal. For the SAT it is the mathematics section minus the verbal section. For the ACT and for the EXPLORE test, they averaged the reading and English subtests into a verbal composite and subtracted it from the mathematics score. A student who scores 640 in mathematics and 560 in verbal has a tilt of positive 80 points. A student with the reverse scores has a tilt of negative 80, which researchers call verbal tilt.

Two features of the construction are worth holding onto. The first is that tilt is orthogonal to level by design when the two scores have equal spread. In the 2007 study that popularized the approach, Park, Lubinski and Benbow analyzed ability level as the sum of the math and verbal scores and ability tilt as the difference. In the thesis version of that work, which carries the same title, the two correlated at .02 (Park, 2007). That near-zero value is partly a property of how sums and differences behave when two scores have similar variances, so it is a feature of the method and not, by itself, a discovery about people. The second feature is the one the critique in a later section turns on: a difference between two scores inherits the measurement error of both.

Tilt can be computed for any pair of abilities. Coyle's 2018 review describes tilt research built from the SAT, ACT, PSAT and the Armed Services Vocational Aptitude Battery (ASVAB), and the same logic has been applied to spatial and mechanical abilities. This page stays with the verbal and quantitative pair, because that is the pair behind the math person question. Spatial ability, the domain of the page on spatial intelligence, adds something of its own, and a later section returns to it.

Finally, tilt is a relative measure. A positive tilt does not say a person is good at mathematics. A person with a verbal score of 450 and a math score of 520 has a math tilt of 70 and still scores near the middle of many admission-test distributions. Conversely, a person with scores of 780 and 740 has a verbal tilt of 40 and is exceptionally strong at both. That is why researchers pair tilt with level, and why the best studies model both.

4 What Did 25 Years of Follow-Up Find in Gifted Youth?

In the longest follow-up of tilt, the gap between math and verbal scores at age 12 or 13 pointed toward STEM or toward the humanities 25 years later, while overall ability level did not distinguish the two groups. The study is Park, Lubinski and Benbow's 2007 paper in Psychological Science, drawn from the Study of Mathematically Precocious Youth. The detail below comes from Park's thesis at Vanderbilt, which has the same title and the same data. Children took the SAT before age 13 in talent searches, and those in the top 1 percent for their age entered the study. The combined sample was 1,569 males and 840 females, 2,409 in all. Participants were surveyed about 20 years later, and the authors then collected professional status, patents and literary publications at least 25 years after the first assessment. Response rates in the 20-year surveys ranged from 77 percent to 82 percent across cohorts.

The design had a strength that most tilt studies lack. Because SAT scores of 12-year-olds are spread much like those of college-going seniors, the able and the exceptionally able were distinguishable, with no pile-up at the ceiling. And the outcomes had high ceilings: tenure-track faculty posts, patents, published novels. The thesis reports that no statistically significant difference in ability level separated the STEM and humanities groups in the broad achievement categories, while ability tilt separated them in every comparison. Tilt differed between STEM and humanities groups by an effect size (Cohen's d) of 0.71 for terminal bachelor's and master's degrees, 0.63 for doctorates, 1.62 for tenure-track positions, and 1.67 for patents against literary publications. The table gives the group means the thesis printed.

Read the table by rows, not by columns. The STEM doctorate group scored higher in mathematics than in verbal by 143 points, and the humanities doctorate group scored higher in verbal than in mathematics by 19 points, which is our subtraction of the printed means. The humanities groups were not weak in mathematics in any absolute sense: a mean SAT mathematics score of 553 or 561 at age 12 is high for that age. Their profile was simply flatter or reversed. The authors' summary is that ability level is informative for how much creative work occurs, while tilt contributes to predicting the domain in which it occurs.

The sample also produced large numbers. Overall, the participants earned 817 patents and published 93 books, 56 of them novels and 37 nonfiction, according to the thesis. The author also flags a ceiling problem. The 18 participants who went on to tenure-track STEM positions at top-50 universities had a mean mathematics score of 697, and two participants earned the maximum of 800, so for the most able the test could no longer separate individuals, which would shrink measured tilt for that group. The thesis adds that predictions would likely improve if spatial ability were included alongside the quantitative and verbal scores, and the page on spatial intelligence covers that third dimension.

5 What Does Tilt Predict in the General Population?

In samples that are not selected for giftedness, tilt shows modest but consistent associations with the abilities, college majors and jobs that match its direction, though the strongest synthesis comes largely from one research program. The central source is the work of Thomas Coyle. His 2014 paper with Purcell, Snyder and Richmond, Ability tilt on the SAT and ACT predicts specific abilities and college majors, used the National Longitudinal Survey of Youth. Coyle's 2018 review of non-g factors, which is open access, summarizes the program and treats tilt as a non-g factor, meaning a pattern of specific abilities that remains once the general factor is set aside. Its main sample, the 1997 cohort of the National Longitudinal Survey of Youth, had 8,989 participants, with SAT, ACT and PSAT scores and the 12-subtest ASVAB, from which the review draws an estimate of general ability (g).

The review reports tilt effects on specific ASVAB abilities with a mean absolute standardized effect of about .28, and effects on STEM versus humanities majors and jobs of about .35, and it says the patterns held across sex and race groups. The direction is the intuitive one. Math tilt went with higher math ability, STEM majors and STEM jobs, and with lower verbal ability. Verbal tilt went the other way. Effects of that size are real but leave most of the variation unexplained, and the section on measurement below returns to what that means for one person.

A second line of evidence concerns what college tests are. Frey and Detterman's study correlated SAT scores with a measure of g drawn from the ASVAB in 917 participants of the 1979 National Longitudinal Survey of Youth and found a correlation of .82, or .86 after correcting for nonlinearity. In a second sample of 104 undergraduates the SAT correlated .483 with Raven's Advanced Progressive Matrices, or .72 after correcting for restricted range. The authors concluded that the SAT is mainly a test of g. This matters for tilt in an unflattering way: a tilt computed from SAT sections is a difference between two heavily g-loaded scores, and the critique below depends on that fact. The page on IQ versus the SAT covers the relation between the two tests, and the page on aptitude versus intelligence covers why an aptitude test is not a separate thing from cognitive ability.

Choices, not only abilities, show up in the data. Wang, Eccles and Kenny used a national longitudinal study with 1,490 participants, measured in 12th grade and again at age 33. Mathematically capable individuals who also had high verbal skills were less likely to pursue STEM careers than individuals with high math and only moderate verbal skills. The high-math, high-verbal group also included more females than males. The authors' reading, in the title of the paper, is "not lack of ability but more choice": people with strength on both sides have more occupations open to them, and many take a non-STEM one. That is a result about the combination of two scores, and it is difficult to describe by sorting people into math and words types.

What about workplaces? Kato and Scherbaum tested tilt against job performance in a large sample from the General Aptitude Test Battery database. Tilt related to performance in the expected direction in 27 of the 36 tilt and job combinations they examined, with a mean effect size of .04 when the tilt matched job requirements. The mean incremental validities were .007 over g and .003 over g and specific abilities, and on average tilt explained 7.1 percent of the variance in job performance. The authors describe the evidence as limited. In plain terms, a tilt that fits the job barely adds to what overall ability and the specific abilities already say. The page on IQ and job performance covers the broader meta-analytic picture.

6 Why Do Verbal and Quantitative Skills Pull Apart?

The leading explanation is investment theory, which holds that time and effort put into one domain build that ability and leave less for a competing one, but the data cannot yet separate investment from interest, opportunity or measurement. Coyle's papers frame the idea this way: investment in a specific ability boosts similar abilities and retards competing ones. The theory lines up with the observed pattern that math tilt correlates negatively with verbal ability and verbal tilt with math ability. The 2018 review adds that if investment is the mechanism, effects should strengthen with ability level, and it presents that magnification prediction as a hypothesis still to be tested. The same idea of differences growing with level connects to Spearman's law of diminishing returns, which has its own page.

Three points restrain the interpretation. First, correlation does not establish a mechanism. A student who finds mathematics rewarding will spend more time on it, and a student who spends more time on it will find it rewarding, so interest, practice and ability reinforce each other in both directions. Second, the abilities in question are Gc and Gq, which are expressed through acquired knowledge. A person who reads widely builds vocabulary and general knowledge, and a person who takes advanced mathematics builds Gq, so a tilt partly records what a person has studied. The page on whether you can raise your IQ describes what practice does and does not move. Third, even a person's investment choices happen in an environment of courses, teachers, expectations and role models.

Coyle's 2018 review notes its own limits, and they are instructive. Measurement error can make g look more predictive than non-g factors unless corrections are applied. Shrinkage and range restriction can distort effect sizes. The standardized-to-correlation conversion the review uses has been criticized by other methodologists. And several of its findings rest on the author's own studies and on a single national sample. None of this makes the tilt results wrong. It does mean that "tilt predicts majors" is better described as a replicated pattern in a few data sets than as a settled law.

Is tilt inherited? Coyle and colleagues analyzed three samples, the Georgia Twin Study in the United States, the Swedish Twin Registry and the US adult study MIDUS II, and reported that tilts are modestly heritable after controlling for age and for residual correlations with general cognitive ability. They also found comparatively large nonshared environmental influences, and they argue that non-zero heritability in two countries and several age ranges weakens the claim that tilts are mere artifacts. Whether that argument works is exactly what the critics dispute, and the next sections take up the dispute. Heritability here would mean that differences in tilt among people in those samples partly track genetic differences, and it would say nothing about whether any single person's tilt is fixed. The page on what heritability actually means covers that limit.

7 Do Men and Women Differ in Verbal and Quantitative Tilt?

Average differences in verbal and in mathematics performance between the sexes are small, but the direction of tilt differs on average, and the difference grows in the extreme upper tail of ability. Start with the two components. Lindberg and colleagues' 2010 meta-analysis of 242 studies published from 1990 to 2007, testing 1,286,350 people, found an overall mathematics effect size of d = 0.05, which indicates no gender difference, and a variance ratio of 1.08, which indicates nearly equal male and female variances. In large probability samples of US adolescents the effect sizes ranged from negative 0.15 to positive 0.22, and variance ratios from 0.88 to 1.34. For verbal ability, the earlier meta-analysis by Hyde and Linn of 165 studies found a weighted mean effect size of 0.11, a slight female advantage, which the authors described as so small that they argued gender differences in verbal ability no longer exist. That 1988 result is dated and should be read as a baseline, not a current estimate.

Tilt is the difference of two small average differences, and it behaves differently in the tail. Wai, Hodges and Makel's 35-year examination analyzed 2,053,265 academically talented students in the United States (SAT, ACT and EXPLORE) and 7,119 in India (the ASSET test), all in the top 5 percent. Across every measure and sample, tilt favored males for math over verbal and females for verbal over math. Sex differences in tilt increased as the ability cut became more selective. In the SAT regressions, the coefficient for sex was 17.58 in the top 5 percent and 29.73 in the top 1 percent, and the 95 percent interval for the top 0.01 percent was 88.23 to 106.11, all in SAT points, with positive values meaning greater male tilt toward math. The authors also report that sex differences in tilt were fairly stable over time and replicated across the two countries. They add, citing earlier work, that in the top 0.01 percent in math on the SAT and ACT there are currently about 2.5 males for every female.

The Park thesis shows the same direction in the gifted-youth sample. Males had a greater quantitative tilt than females (d = 0.72) and also scored higher on ability level (d = 0.40). In that sample 9.9 percent of men and 3.0 percent of women earned a STEM doctorate, 10.6 percent of men and 1.3 percent of women held patents, and 1.3 percent of men and 2.0 percent of women had a literary publication. The base rates are low and the sample is specialized, so those are figures about a selected group.

How much weight should this carry for an individual? Very little, for three reasons. First, the overlap between the male and female distributions is large at every level examined. Second, Wai and colleagues note, citing earlier work, that math ability in isolation, relative to factors such as interests, is likely a lesser factor in the underrepresentation of women in STEM, and they describe their result as one factor among several. Third, the Wang, Eccles and Kenny result above shows that a sizable group has both high math and high verbal ability and more choices, and that group included more women than men. A group difference in tilt describes populations, and the page on average IQ by gender covers the wider evidence on group differences in IQ with the same caution.

8 Is Tilt a Real Trait or a Statistical Artifact?

Critics argue that tilt correlations are arithmetic consequences of subtracting two related scores, and the argument is strong enough that the safest use of tilt is descriptive, not explanatory. The critique comes mainly from Kenneth Sorjonen and colleagues. In their 2022 paper, using the 1997 National Longitudinal Survey of Youth, they found that associations between an ability and an ability tilt may simply reflect more positive associations between two measures of the same or similar abilities than between measures of different abilities. That finding, they say, calls into question the theoretical reading that tilt correlations show differential investment of time and effort in one ability at the expense of another.

The core of the argument can be stated without statistics. A tilt is a score minus another score, so its correlation with any third variable depends on the difference between the two component correlations. If mathematics scores correlate more strongly with a STEM major than verbal scores do, then math minus verbal will correlate with that major, whatever the reason mathematics predicts it. There is a second, related point that is simple algebra and does not depend on any dataset: a tilt is a linear combination of its two components, so a regression that already contains both scores cannot gain anything from adding their difference. Sorjonen, Melin and Nilsonne's 2025 comment shows that pattern empirically, reporting that models with a tilt and one component gave the same explained variance as models with both components.

Their demonstration used the Georgia Twin Study: 108 monozygotic and 130 dizygotic twin couples, 476 people in all, with four Primary Mental Abilities tests, face length, nose length and two random variables. In their reanalysis, replacing age with an arbitrary twin-pair identifier produced shared-environment shares above 85 percent for every outcome and tilt, including tilts between two random variables, and they argue that tilt heritability simply tends to equal the mean of the heritabilities of its components. For example, the Verbal minus Number tilt correlated .306 in monozygotic twins and .153 in dizygotic twins. In their view, a method that yields the same pattern for random variables is not detecting a psychological mechanism specific to abilities.

The defenders have replies. Coyle and colleagues, and Woodley of Menie and colleagues in the exchange that Sorjonen's 2025 comment answers, argue that tilts are meaningful when the two components load on the same factor, and they point to a network of findings, including negative relations with competing abilities, links to majors and jobs, and modest heritability, that they say investment theory predicts and an artifact would not. Woodley of Menie and colleagues also cite Kato and Scherbaum's incremental validity result as evidence against spuriousness, and Sorjonen and colleagues answer that the incremental validity of a tilt is itself a consequence of the validity of its components. Kato and Scherbaum's own numbers are modest: .007 over g, .003 over g and specific abilities.

What can a reader keep from a debate that remains open as of 2025? First, neither side disputes that a person's two scores can differ, or that people with a large gap often differ from other people in later fields. Second, the disagreement is about interpretation: whether the gap is a trait with its own causes, or a convenient summary of two scores that each have causes. Third, for an individual the practical consequence is the same under either view. The two scores carry the information, the gap is a summary of them, and a summary adds nothing that cannot be read from the original numbers. The page on reliability and validity explains the evidence a score summary needs before it earns an interpretation of its own.

9 How Much of a Verbal and Quantitative Gap Is Measurement Noise?

A gap between two scores is less reliable than either score, and it gets less reliable the more strongly the two scores are correlated, so a modest gap on one report is weak evidence of anything. Every score has error: the score you obtained is the true score plus noise. A difference between two scores carries the noise of both and, because the shared part of the two scores cancels in the subtraction, loses some of the true signal as well. The Standards for Educational and Psychological Testing address this directly. Standard 2.4 says that when an interpretation emphasizes differences between two observed scores of an individual, reliability and precision data, including standard errors, should be provided for those differences. Its comment notes that differences between verbal and performance scores have often been used in diagnosis, and that when interpretation centers on the peaks and valleys of a profile, the reliability of score differences is critical.

Classical test theory gives a compact rule for the case of two scores with equal spread. The reliability of the difference equals the average of the two reliabilities, minus the correlation between the scores, divided by one minus that correlation. The standard error of the difference, in score units, is the standard deviation times the square root of two minus the two reliabilities added together. The numbers below are our own arithmetic with hypothetical values and are not ACIS statistics or any test's published figures. They show the shape of the problem.

The first three rows hold each score's reliability at .90 and raise the correlation between the scores. The reliability of the difference falls from .83 to .50 even though each score is equally precise. With a standard error of the difference of 6.7 points, an observed gap would need to be about 13 points before a 95 percent interval around it excluded zero (our arithmetic: 1.96 times 6.7). In the last row, two slightly less reliable scores widen the standard error to 8.2 points, and the gap needed grows to about 16 points. A 10 point gap between a verbal score and a quantitative score would be well within noise in the second and fourth rows.

Two practical corollaries follow. A score built from more tasks is usually more reliable than one built from two, so an index that rests on two subtests deserves wider caution than one that rests on five, all else equal. And the correlation between verbal and quantitative skill is not small (the meta-analytic .42 above), so the problem applies to the pair this page discusses. The page on how large an index gap has to be to count works through base rates for gaps among index scores, and the page on the standard deviation of 15 explains the unit used in the table.

10 How Should You Read a Verbal and Quantitative Gap on an ACIS Report?

Read the gap as one fact about two numbers with intervals around them, check it against the rest of the profile, and decline to turn it into a label. An ACIS report gives a Full Scale IQ and six primary indices, each on the standard scale with a percentile and a 95 percent confidence interval. According to the ACIS home page, read on October 20, 2026, the six indices are VCI (verbal comprehension), FRI (fluid reasoning), QRI (quantitative reasoning), VSI (visual spatial), WMI (working memory) and PSI (processing speed). VCI and QRI are the pair that corresponds to the verbal and quantitative question in this page. The page on the Full Scale IQ explains how the overall number sits beside them.

A short sequence keeps the reading honest.

1. Look at the intervals before the points. If the 95 percent intervals of VCI and QRI overlap heavily, the report gives you no reason to say one is stronger. The table in the section on measurement noise shows why a gap can shrink toward noise.
2. Check the other indices. The fluid reasoning index includes Visual Number Series, a numeric task, and the visual spatial index covers a third domain. A person whose QRI is high and FRI is not has a different pattern from one whose QRI and FRI are both high. The page on abstract reasoning tests describes how fluid reasoning tasks are built. Quantitative knowledge, Gq, is not the same ability as reasoning about novel relations, which is Gf, and the page on what each subtest measures maps the subtests to their domains.
3. Ask what you have studied. VCI and QRI both lean on acquired knowledge. A person who read widely but stopped taking mathematics at 16 may show a verbal lead that records a decade of habit. The page on how to prepare for an IQ test covers what preparation does and does not do to a score.
4. Consider language. The ACIS test is in English only. A person who learned English as a second language may have a VCI that understates verbal ability in the first language, and the page on average IQ by language covers why the same person's score changes with the language of the test.
5. Keep the decision separate from the score. A gap, even a reliable one, does not choose a career. Wang and colleagues found that people strong in both math and verbal were less likely to pursue STEM than people with high math and only moderate verbal skills, which is a reminder that other things matter more than the lead of one score over another.

On forms and prices, all read on October 20, 2026 and subject to change: the Quick form costs 15 dollars for 6 subtests in 3 domains and takes about 45 minutes, the Optimized form costs 30 dollars for 13 subtests in 5 domains and takes about 110 minutes, and the Full Scale form costs 50 dollars for all 20 subtests in six domains and takes about 175 minutes. All are one time payments with no subscription, breaks are allowed, a free trial needs no card, there is a 5 day quality guarantee and a customer has 30 days to complete. Only the Full Scale form is stated to cover all six domains. Check the description of a shorter form to confirm that both VCI and QRI are included before relying on a comparison.

The limits are plain. The test is online and unsupervised, it is not a clinical or diagnostic instrument, it is not for hiring, school accommodations or admission to high IQ societies, and it is in English only. The technical documentation is at the ACIS technical manual, and this page quotes no figures from it.

11 What Does the Evidence Support, Stated Narrowly?

The evidence supports a handful of narrow claims, and each of them is weaker than the popular phrase math person suggests. Stated as carefully as the sources allow:

• Language and mathematics are positively related (r = .42 in a 344-study meta-analysis), so being strong at one usually goes with being above average at the other.
• Among very able youth, the math-minus-verbal gap at age 12 was associated with STEM or humanities outcomes 25 years later, and ability level was not (2,409 participants, one research program).
• In national samples, tilt associates with matching majors and jobs with effects of roughly .28 to .35 in a review of one research group's work, and with job performance at a mean effect of .04 in a different data set.
• The average direction of tilt differs by sex, and the difference grows toward the top of the ability range, while average differences in mathematics itself are near zero in large meta-analyses.
• Critics have shown that tilts between random variables produce the same shared-environment pattern and that a tilt adds nothing beyond its two components by arithmetic, and the dispute about interpretation has not been settled.

What the evidence does not say about you is as important. It does not say you are a math person or a words person, because the studies describe groups and a gap between two scores is a continuous quantity. It does not say your gap is large enough to be real, because that depends on measurement error, which is large for differences. It does not say what you should study or do for a living. The studies in this literature are associations, mostly observational, and the gifted-youth study cannot say whether tilt caused the later choices or recorded them. And it does not say your profile is fixed. Gc and Gq are built largely through learning, and they move with it. The page on what IQ scores mean gives the general frame for reading any number on a report.

12 What Do the Standards and APA Guidance Say About Reading a Gap Between Scores?

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014) ask that any interpretation of a subscore, a score difference or a profile come with its own rationale and evidence, which is the discipline this page has tried to follow. Four of the Standards bear directly on the math person question, and each was read in the 2014 edition for this page. Standard 1.14 says that when interpretation of subscores, score differences or profiles is suggested, the rationale and relevant evidence in support of that interpretation should be provided. Standard 2.3 says that estimates of reliability and precision should be reported for each total score, subscore or combination that is to be interpreted, and its comment warns that total scores can be reliable while subscores are not. Standard 2.4, discussed above, asks for reliability data for the differences themselves. Standard 12.12, which concerns educational decisions, says that when an individual's scores from different tests are compared, the decision should take into account the overlap between the two constructs and the reliability or standard error of the difference score.

The APA Guidelines for Psychological Assessment and Evaluation, approved by the APA Council of Representatives in March 2020, add two points from the same direction. They tell psychologists to use multiple sources of relevant and reliable information, noting that performance on a test is only one piece of an assessment and that convergence and divergence across sources increase confidence in the use of test results. They also state that validity is evaluated in context, for a given purpose and group, and not as a fixed property of a test. For a reader of an online report, the translation is simple: a verbal and quantitative gap is a reasonable question to ask, a reasonable thing to track, and a poor basis for a decision about school or work, unless other evidence agrees with it.

13 Sources Behind This Page

Every figure and attribution on this page traces to one of the sources below, or is labeled as our own arithmetic. Our own arithmetic covers the row-by-row subtractions of printed means in the gifted-youth table, the 1.96 multiples used for gap thresholds, and the reliability and standard error table, which uses hypothetical values and no test's published figures. The details of the gifted-youth study (sample split, effect sizes, group means, patent and book counts, percentages by sex) come from the thesis version, which was opened in full, while the paper's own abstract was checked through its DOI record. The Standards and APA documents were read in full text on October 20, 2026. ACIS facts were read on the ACIS home page on October 20, 2026.

• Park G, Lubinski D and Benbow C P. Contrasting intellectual patterns predict creativity in the arts and sciences. Psychological Science, 2007, volume 18, issue 11, pages 948 to 952. Detail figures from Park G, Contrasting intellectual patterns predict creativity in the arts and sciences: Tracking intellectually precocious youth over 25 years, master's thesis, Vanderbilt University, December 2007.
• Coyle T R, Purcell J M, Snyder A C and Richmond M C. Ability tilt on the SAT and ACT predicts specific abilities and college majors. Intelligence, 2014, volume 46, pages 18 to 24.
• Coyle T R. Non-g factors predict educational and occupational criteria: More than g. Journal of Intelligence, 2018, volume 6, issue 3, article 43.
• Wai J, Hodges J and Makel M C. Sex differences in ability tilt in the right tail of cognitive abilities: A 35-year examination. Intelligence, 2018, volume 67, pages 76 to 83.
• Wang M T, Eccles J S and Kenny S. Not lack of ability but more choice: Individual and gender differences in choice of careers in science, technology, engineering, and mathematics. Psychological Science, 2013, volume 24, issue 5, pages 770 to 775.
• Peng P, Lin X, Unal Z, Lee K, Namkung J, Chow J and Sales A. Examining the mutual relations between language and mathematics: A meta-analysis. Psychological Bulletin, 2020, volume 146, issue 7, pages 595 to 634.
• Lindberg S M, Hyde J S, Petersen J L and Linn M C. New trends in gender and mathematics performance: A meta-analysis. Psychological Bulletin, 2010, volume 136, issue 6, pages 1123 to 1135.
• Hyde J S and Linn M C. Gender differences in verbal ability: A meta-analysis. Psychological Bulletin, 1988, volume 104, issue 1, pages 53 to 69.
• Frey M C and Detterman D K. Scholastic assessment or g? The relationship between the Scholastic Assessment Test and general cognitive ability. Psychological Science, 2004, volume 15, issue 6, pages 373 to 378.
• Kato A E and Scherbaum C A. Exploring the relationship between cognitive ability tilt and job performance. Journal of Intelligence, 2023, volume 11, issue 3, article 44.
• Coyle T R, Woodley of Menie M A, Penaherrera-Aguirre M, Madison G and Sarraf M A. The heritability of ability tilts. Personality and Individual Differences, 2023, volume 213, article 112187.
• Sorjonen K, Nilsonne G, Ingre M and Melin B. Spurious correlations in research on ability tilt. Personality and Individual Differences, 2022, volume 185, article 111268.
• Sorjonen K, Melin B and Nilsonne G. Still spurious: A comment on attempts to revive cognitive ability tilts. PLOS One, 2025, volume 20, issue 10, article e0326486.
• McGrew K S. CHC theory and the human cognitive abilities project: Standing on the shoulders of the giants of psychometric intelligence research. Intelligence, 2009, volume 37, issue 1, pages 1 to 10.
• American Educational Research Association, American Psychological Association and National Council on Measurement in Education. Standards for Educational and Psychological Testing, 2014, Standards 1.14, 2.3, 2.4 and 12.12.
• American Psychological Association, APA Task Force on Psychological Assessment and Evaluation Guidelines. APA Guidelines for Psychological Assessment and Evaluation, approved March 2020.

14 Frequently Asked Questions