The SAT is the most widely taken standardised cognitive test in the United States, and its results are interpreted by college admissions offices as if they reveal something specific about applicants’ cognitive abilities. The psychometric question is sharper than the everyday interpretation suggests: when a student obtains a higher SAT-Math score than SAT-Verbal score, does this reflect a domain-specific advantage in quantitative reasoning, an artefact of educational background, or noise around their general cognitive ability? The cognitive-measures-of-reasoning-and-language literature has answered this with progressively more refined factor analyses, and the picture that emerges is more nuanced than either “the SAT measures intelligence” or “the SAT measures specific subjects” admits.
The CHC framework: where reasoning and language fit
Modern psychometric theory organises cognitive abilities hierarchically through the Cattell-Horn-Carroll (CHC) framework, summarised by McGrew (2009) in Intelligence. CHC distinguishes a general factor (g) at the top, broad abilities at the second stratum, and narrow abilities at the third. The broad abilities most relevant to reasoning and language tests are crystallised intelligence (Gc: vocabulary, general knowledge, language comprehension), fluid intelligence (Gf: reasoning in novel situations, induction, pattern detection), and quantitative knowledge (Gq: numerical and mathematical concepts).
The framework is empirical, not theoretical: it is a summary of what factor analyses of large cognitive-ability batteries have repeatedly found about how individual differences cluster. Gf and Gc are the two most heavily loaded broad abilities and the most relevant for understanding what reasoning-and-language tests actually measure. The cognitive distinction between them maps roughly onto the everyday distinction between “reasoning” and “knowledge” — reasoning ability is about manipulating relationships, while crystallised ability is about what one has learned and stored, including language. This is the broader framing of the fluid versus crystallised intelligence distinction.
The SAT as a measure of g
The SAT was originally designed as an aptitude test, rebranded as a knowledge-and-reasoning test, and is now formally framed as a measure of college readiness. Frey and Detterman’s (2004) analysis in Psychological Science reported that SAT scores correlate with general cognitive ability at r ≈ 0.82 in college-bound samples (corrected for restriction of range). That is large enough that the SAT can reasonably be described as a heavily g-loaded test — most of what it measures is general cognitive ability, with subscale-specific variance riding on top.
Coyle and Pillow (2008) asked whether the SAT predicts academic outcomes only through g or whether subscale-specific variance carries independent predictive value. Using structural-equation modelling to remove the g component, they found residual SAT-Math and SAT-Verbal scores still predicted college GPA — small but real, indicating the SAT subscales measure crystallised verbal knowledge and quantitative-reasoning skill that distinguish performance even between equally generally-able students.
Reasoning and language as separable dimensions
The within-test factor question is whether SAT-Math and SAT-Verbal reflect distinct cognitive processes or just different surface content built on the same underlying ability. Jouve (2010) approached it from outside the SAT, by adding a third measure: 106 test-takers completed the Jouve-Cerebrals Test of Induction (JCTI) — an untimed, figural inductive-reasoning test with no verbal content at all — and reported their SAT section scores. A content-free reasoning test that aligns with one SAT section and not the other tells you something the SAT’s own internal structure cannot.
The split was sharp. The JCTI correlated .80 with SAT-Mathematics and .27 with SAT-Verbal — a difference of .53, 95% CI [.37, .71], positive in every one of 10,000 bootstrap resamples. Adjusting for SAT-Mathematics left no detectable JCTI–Verbal association at all (partial r = −.10, 95% CI [−.28, .10]), which is consistent with the modest zero-order verbal correlation arising indirectly, through the overlap between the two SAT sections, rather than through a direct link between figural reasoning and verbal ability. What SAT-Math measures is therefore largely Gf — reasoning capacity applied to quantitative content — rather than mathematical knowledge as such, and whatever SAT-Verbal adds, a pure reasoning test does not reach it.
Two cautions come with that reading, and the study states both. The absence of a detectable association is not the absence of an association: with 106 respondents the data can exclude a residual link larger than about ±.30 and no tighter bound, so a small one may well be there. And conditioning on SAT-Mathematics removes general reasoning variance along with quantitative content, so the design establishes a differential association — the JCTI tracks one section far more closely than the other — not the mechanism behind it. With three measures and no independent index of g, no second, language-specific factor can be identified here; that interpretation comes from the wider CHC literature below, not from these data.
What the factor structure means in practice
Three implications follow. A substantial Math-Verbal gap on the SAT is psychometrically interpretable, reflecting a real difference in how a student’s general reasoning interacts with language-specific crystallised resources. A content-free reasoning test and SAT-Math measure heavily overlapping ability — they share about two-thirds of their variance — though sharing most of a construct is not the same as being interchangeable, and the JCTI is not a substitute for a section score. And SAT-Verbal carries a language-specific component that schooling and reading exposure shape directly, which is why preparation effects are larger and more reliable on SAT-Verbal than on SAT-Math at equivalent dose.
The broader principle — that complex cognitive tests decompose into a strong general factor plus modest specific factors — is consistent with the structural finding from analyses of spatial vs. abstract reasoning: a single dominant factor accounts for most of the variance in nonverbal reasoning batteries, with smaller but real specific-factor structure.
The hierarchical picture
Putting the results together yields a layered interpretation. At the top is g: the dominant source of variance on any cognitively demanding test. Below g sit two consequential broad abilities for the SAT — a fluid-reasoning factor (Gf) that JCTI and SAT-Math both load heavily on, and a crystallised-intelligence factor (Gc) that SAT-Verbal taps preferentially. At the third stratum sit narrow abilities (inductive reasoning, vocabulary, reading comprehension) contributing task-specific variance. This hierarchy is not unique to the SAT — Niileksela and Reynolds (2019) found the same pattern across the Wechsler family: g dominates, broad abilities matter for prediction, narrow abilities account for task-specific variance.
Limitations and what remains open
The Jouve (2010) JCTI+SAT analysis used SAT scores that respondents reported themselves, unverified, in a self-selected sample of 106 whose section means sat well above national norms — a mild restriction of range, correcting for which leaves the contrast intact. Larger samples would sharpen the estimates: the adjusted verbal association is pinned only to within about ±.30, so the study can show that the JCTI tracks SAT-Math far more closely than SAT-Verbal without settling whether the residual verbal link is exactly zero. The design is also correlational and cross-sectional: whether SAT-Verbal reflects long-term language exposure or test-day reading speed is a question it cannot resolve. How these findings should be weighted in admissions decisions is also wider than the psychometric question; Frey-Detterman’s high SAT-g correlation supports treating the SAT as a general-ability proxy, Coyle-Pillow’s residual-prediction finding supports treating subscale scores as additionally informative.
Frequently asked questions
Does the SAT measure intelligence?
Substantially yes. Frey and Detterman (2004) reported that SAT scores correlate with general cognitive ability at r ≈ 0.82 in college-bound samples (corrected for restriction of range). This is large enough that the SAT can reasonably be interpreted as a heavily g-loaded test, with smaller specific contributions from quantitative reasoning and language ability.
What’s the difference between what SAT-Math and SAT-Verbal measure?
SAT-Math draws primarily on fluid reasoning (Gf) applied to quantitative content; SAT-Verbal draws on fluid reasoning plus crystallised intelligence (Gc), with the latter reflecting language-specific resources built through reading and education. Jouve’s (2010) study bears this out from the reasoning side: a purely figural inductive-reasoning test correlated .80 with SAT-Math but only .27 with SAT-Verbal, and once SAT-Math was adjusted for, no independent link with SAT-Verbal remained detectable.
Why is fluid reasoning treated as separable from crystallised knowledge?
Decades of factor-analytic research summarised in McGrew’s (2009) review of Cattell-Horn-Carroll theory show that performance on novel reasoning tasks (where prior knowledge is minimised) and performance on knowledge-loaded tasks (vocabulary, general information) cluster onto separable broad factors. The two remain correlated through the underlying g factor but are reliably distinguishable in well-designed batteries.
Does SAT prep change what the SAT measures?
It changes the score but not what the test measures. The SAT-Verbal responds to long-term language exposure (reading, vocabulary instruction), which is why preparation effects are largest for students with weaker baseline language background. Short-term test-prep produces smaller and inconsistent gains, and does not turn one cognitive ability into another. The factor structure of the test — what it indexes — is invariant across preparation levels.
Is the JCTI a substitute for the SAT?
For inductive-reasoning measurement, the two share enough variance to be used interchangeably as Gf proxies. The JCTI does not measure the language-specific crystallised abilities SAT-Verbal taps, so the two are complementary rather than equivalent for general academic-aptitude assessment. The JCTI’s relative advantage is content-lightness: it minimises confounding with educational background and language exposure.
References
- Coyle, T. R., & Pillow, D. R. (2008). SAT and ACT predict college GPA after removing g. Intelligence, 36(6), 719–729. https://doi.org/10.1016/j.intell.2008.05.001
- Frey, M. C., & Detterman, D. K. (2004). Scholastic Assessment or g? The relationship between the Scholastic Assessment Test and general cognitive ability. Psychological Science, 15(6), 373–378. https://doi.org/10.1111/j.0956-7976.2004.00687.x
- Jouve, X. (2010). Inductive reasoning and SAT performance: The JCTI relates strongly to SAT-Mathematics but shows no detectable independent association with SAT-Verbal. Cogn-IQ Research Papers. https://pubscience.org/ps-1ml2g-552578-YLVi
- McGrew, K. S. (2009). CHC theory and the human cognitive abilities project: Standing on the shoulders of the giants of psychometric intelligence research. Intelligence, 37(1), 1–10. https://doi.org/10.1016/j.intell.2008.08.004
- Niileksela, C. R., & Reynolds, M. R. (2019). Enduring the tests of age and time: Wechsler constructs across versions and revisions. Intelligence, 77, 101403. https://doi.org/10.1016/j.intell.2019.101403