Depression Scales and High IQ: What a 2025 Study of Two US Cohorts Found

Not equally well, according to a 2025 study in Intelligence. It tested two widely used mental-health questionnaires on 14,946 Americans and found that the questions say less about depression the higher a person's cognitive-ability score is. The authors conclude that scores on these scales are not comparable across intelligence levels, so claims either way about whether very bright people are more or less prone to depression cannot be settled with them. Scientific American reported on the paper in May 2026. This guide explains what the researchers did, what the numbers show, what they do not show, and how to read the next claim that a high IQ goes with depression, or with better mental health.

What question did the study ask?

Large studies mostly find that higher intelligence goes with better mental health on average. The authors cite work showing this, and also studies that used nonlinear models and suggested the link turns negative at the very top: psychotic symptoms in more than 6,000 UK children, bipolar disorder in more than a million Swedish men, and a range of measures in almost 8,000 British adults in the team's own earlier work. Comparisons of gifted and non-gifted groups generally find no difference or better mental health in the gifted group, and the Study of Mathematically Precocious Youth, which follows people with exceptional mathematical reasoning, reports good adjustment and wellbeing (see our guide to what SMPY found). The new study set out to test whether the relationship is U-shaped, with some worsening at the top, and whether it differs between men and women.

What data and tests did it use?

It used two US cohort studies run through the Bureau of Labor Statistics. The National Longitudinal Survey of Youth 1979 (NLSY79) supplied 8,474 participants, who answered the 7-item version of the Center for Epidemiological Studies-Depression scale (CES-D) in 1992. The 1997 cohort (NLSY97), born 1980 to 1984 and 14.9 years old on average at the start, supplied 6,472 participants, who answered the 5-item Mental Health Inventory (MHI-5) up to eight times between 2000 and 2017: three items on psychological distress and two on well-being.

Intelligence was the Armed Forces Qualification Test (AFQT), built from four ASVAB subtests: arithmetic reasoning, mathematics knowledge, word knowledge and paragraph comprehension. Scores arrive as percentile ranks, which the authors converted to z-scores, and for interpretation they restated the breakpoints they found on an IQ scale with a mean of 100 and a standard deviation of 15. They note that the AFQT correlates highly with the Wechsler Adult Intelligence Scale and is a common proxy for intelligence, and that it was taken under low-stakes conditions. Our guide to the ASVAB and AFQT score to IQ conversion explains the test and its limits.

What did the regression results show?

The standard analysis found a U shape. In the older cohort, higher intelligence went with lower depression scores up to a breakpoint, then a slight rise. Below the breakpoint the standardised coefficient was -.20 for men and -.22 for women; above it, +.03 and +.05. The breakpoints were 2.10 standard deviations above the mean for men (an IQ of 132) and 1.70 for women (126). On a normal curve those points are the 98.2nd and 95.5th percentiles, roughly 1 person in 56 and 1 in 22, which you can check with our IQ percentile calculator. In the younger cohort, distress followed a similar U shape with breakpoints at IQ 115 for men and 123 for women; above the breakpoint the upturn was statistically significant for women (+.05) but not for men (+.04, p = .063), and well-being showed no clear link to intelligence at the start.

These effects are small. The intelligence terms together explained 3.8% of the variance in depression for men and 4.3% for women, and the curved term added 0.3 and 0.4 percentage points. The U shape is a statistically detectable bend in a mostly downward slope, not a cliff at the top.

Why did the authors stop short?

Because a second test failed. Local structural equation modelling checks whether a questionnaire's items relate to the underlying trait in the same way at different values of a moderator, here intelligence. The key quantity is the factor loading, the strength of the link between one item (say, feeling sad) and the trait it is meant to indicate. If loadings change with intelligence, the same total score does not mean the same thing at different intelligence levels, a failure called a lack of metric measurement invariance. In the older cohort, five CES-D items for men and six for women, out of seven, showed significant changes in loading; in the younger cohort, three of five MHI-5 items did for both sexes. The general trend was that loadings fell as intelligence rose, so the items looked less like depression indicators for higher-scoring people.

Model fit also dropped sharply at the highest intelligence values, although the effective sample at those points was small: in the older cohort it ranged from about 60 to 85 people at the thinnest points up to about 1,900 at the thickest, so the authors urge caution on that specific result and note that the general decline in loadings was stable across the range. They conclude that any finding of nonlinearity, or of its absence, cannot be treated as reliable evidence with these measures, and that the study's original hypothesis could not be addressed. They did not run their longitudinal models for the MHI-5 because the results could not be told apart from statistical artefacts.

What might explain the pattern?

The authors suggest that answering a questionnaire is itself a cognitive task, so people with different abilities may read, interpret or answer items differently. They point to a study of the Rosenberg Self-Esteem Scale (Gnambs and Schroeders, 2017) in which the scale became more unidimensional at higher ability, which was attributed to a better grasp of negatively worded items. They also note the resemblance to Spearman's law of diminishing returns, the finding that the general factor of intelligence explains less of test performance as ability rises; see our guide to Spearman's law of diminishing returns. And they raise a possibility they cannot test here: that the structure of depression itself differs at higher intelligence, or that behaviours treated as symptoms are ordinary traits of some gifted people. They call for scales that are invariant across intelligence, or for physiological and behavioural indicators of mental health.

What does the study not show?

• It does not show that very bright people are depressed, or that they are not; with these two scales the authors say they could not test it
• It does not show the scales are useless: they passed invariance checks across gender and time, and they still work within a group
• It does not use a Wechsler test or clinical diagnoses: intelligence is an AFQT proxy taken under low-stakes conditions and mental health is a symptom questionnaire
• It covers two US cohorts, so culture and education systems may matter elsewhere
• It is one study, and Scientific American's report did not name the two questionnaires; the paper names the 7-item CES-D and the MHI-5

How should you read the next headline about high IQ and depression?

• Ask which questionnaire was used and whether anyone tested it at the top of the range
• Look at effect size as well as significance: here the bend adds under half a percentage point of explained variance
• Check how many people sit at the extremes: the ends of a curve can rest on dozens of people, not thousands
• Remember a questionnaire measures reported symptoms, not a diagnosis
• For the clinical literature see our guides to depression and IQ, anxiety and IQ, bipolar disorder and IQ and high IQ and mental health

Our IQ Metrics test reports on the standard scale, and the IQ bell curve shows where a result sits. For related questions see our guides on whether smart people are lonely, how rare a high IQ score is and the margin of error on an IQ score.

Common questions