Does Weed Lower IQ? What the Long-Term Studies Actually Show

Does weed lower IQ? The strongest long-term evidence says heavy, persistent cannabis use that starts in the teens is linked to lower test scores, but how large the effect is, and whether cannabis causes it, are still argued over. The best-known study, from New Zealand, found an average drop of about 6 IQ points in people with repeated cannabis dependence. Twin studies, adjusted analyses and pooled cohorts found smaller effects, or none once family background and cigarette smoking were accounted for, though a 2022 follow-up of the New Zealand group to age 45 found a 5.5-point decline in long-term users. Here are the numbers from each, and what none of them has shown.

What did the largest long-term study find?

Meier and colleagues (Proceedings of the National Academy of Sciences, 2012) followed 1,037 members of a New Zealand birth cohort from childhood to age 38. Cannabis use was asked about at ages 18, 21, 26, 32 and 38, and IQ was measured in childhood and again at 38. The change in IQ, in standard deviations (SD) so that it can be compared across tests, grew with the number of assessments at which a person met the criteria for cannabis dependence:

• Never used cannabis (242 people): +0.05 SD
• Dependent at one assessment (80 people): -0.11 SD
• Dependent at two assessments (35 people): -0.17 SD
• Dependent at three or more assessments (38 people): -0.38 SD, about 6 IQ points

On a scale where one SD is 15 IQ points, -0.38 SD is about 6 points. For the 23 people who became dependent before age 18 and stayed dependent, the drop was -0.55 SD, about 8 points, while people who became dependent only as adults showed no apparent decline. The authors called the pattern suggestive of a neurotoxic effect on the adolescent brain, and named the limits: causality is unproven, cannabis use was self-reported with no biological check, and it is a single cohort from New Zealand. To turn SD changes into points yourself, see our guide to IQ z-scores.

Why do critics say cannabis may not be the cause?

Ole Rogeberg’s PNAS reanalysis (2013) built a simulation with no cannabis effect at all. Socioeconomic status predicts cannabis use, and children from low-status homes can get a temporary schooling boost of about 4 IQ points (0.25 SD) that fades; his simulation reproduced the New Zealand pattern from that alone. He concluded that the true effect could be zero and that the causal claim was premature, though his work was a simulation and added no new data. Meier’s team replied (Moffitt et al., PNAS, 2013) that low status did not predict adolescent-onset dependence (P = 0.56) and was unrelated to IQ change (r = -0.006), and that the association barely moved with status controlled (β from -0.152 to -0.158), among middle-class members only (β = -0.155) or with childhood self-control controlled (β = -0.151). They accepted that observational data cannot prove causation.

Twin studies test the family-background explanation directly, because twins share a home and, in identical pairs, their genes. Jackson and colleagues (PNAS, 2016) followed two cohorts, 789 twins in Los Angeles and 2,277 in Minnesota, measuring IQ at ages 9 to 12 and again at 17 to 20. Users’ Vocabulary scores fell 4.0 points further than non-users’ in Los Angeles and 3.4 further in Minnesota. But within twin pairs where one used cannabis and the other did not, there was no significant difference in IQ change, and heavier use (30 or more occasions, or daily use) showed no extra decline. The authors concluded the declines likely reflect familial factors rather than direct exposure. Their limits were modest numbers of discordant pairs and attrition.

The Avon Longitudinal Study of Parents and Children (ALSPAC) in England followed 2,235 teenagers, and 24% had tried cannabis by age 15. Mokrysz and colleagues (Journal of Psychopharmacology, 2016) compared people who had used it 50 or more times with never-users. After adjusting for IQ at age 8, users scored 2.9 points lower at 15 (p < 0.001), but with cigarette use and other factors in the model the gap disappeared (p = 0.959). Exam scores at 16 were 11.6 percentage points lower unadjusted and 11.0 after adjusting for earlier results, and the fully adjusted model erased that gap too. The authors named young age at outcome, no measure of THC dose and a short IQ test as limits.

What do the pooled analyses say?

Scott and colleagues (JAMA Psychiatry, 2018) pooled 69 studies covering 2,152 users and 6,575 comparison people, mean age about 21. Most were single-time cognitive tests across ten domains, not IQ. Frequent or heavy use was linked to lower performance with an effect size of d = -0.25 (95% confidence interval -0.32 to -0.17), and age of onset did not change it. In the 15 studies that required more than 72 hours of abstinence (928 people) the effect was d = -0.08 (-0.22 to 0.07), not different from zero, against d = -0.30 in 54 studies with looser abstinence rules. The authors read the effects as small and possibly reflecting residual acute or withdrawal effects.

Power and colleagues (Psychological Medicine, 2021) pooled seven longitudinal cohorts with IQ measured before and after (808 users, 5,308 comparison people). Frequent or dependent use in youth was linked to a decline of d = -0.132 (95% CI -0.198 to -0.066), about 2 IQ points on an SD-15 scale, and the authors called for longer follow-up. The 2022 follow-up of the New Zealand cohort to age 45 (Meier and colleagues, American Journal of Psychiatry) found that long-term users (86 people) lost 5.5 IQ points against a gain of 0.70 for never-users, with tobacco users losing 1.5 and alcohol users 0.5, and that childhood status, self-control and other substances did not explain it. That paper’s title uses the term cognitive reserve, which our piece on brain wiring and cognitive reserve explains.

How do the studies compare on one scale?

The studies report effects in different units, so it helps to put them on one yardstick, an IQ scale where one SD is 15 points. This is a conversion for scale only, not a like-for-like comparison, because the samples, tests and measures differ:

• Meta-analysis, after more than 72 hours of abstinence: d = -0.08, about 1.2 points
• Seven pooled cohorts, frequent or dependent youth use: d = -0.132, about 2 points
• Meta-analysis, heavy use without an abstinence rule: d = -0.25, about 3.8 points of test performance, not IQ
• New Zealand, dependent at three or more assessments: -0.38 SD, about 5.7 points
• New Zealand to age 45, long-term users: 5.5 points, about -0.37 SD
• New Zealand, dependent before age 18: -0.55 SD, about 8.3 points

What did the 2026 teen study add?

Wade and colleagues (Neuropsychopharmacology, 2026) analysed the Adolescent Brain Cognitive Development (ABCD) Study, 11,036 US children followed from ages 9 to 10 to 16 to 17, with hair testing in a subsample of 645. Cannabis-by-age interactions appeared across memory, processing speed, inhibitory control, language and other domains, and hair-confirmed THC use predicted worse episodic memory over time (β = -0.60, p = .007), while CBD showed no difference. Users appeared to start with better scores and then gained less than non-users. No IQ was measured, and the authors note that causation is unproven and the CBD group was small. The University of California San Diego release that accompanied it was covered by ScienceDaily on 7 August 2026.

Does IQ come back after you quit?

No study reviewed here follows the same people’s IQ rising after they stop. In the original New Zealand study, adolescent-onset persistent users still showed a decline at 38 whether they had used rarely (median 14 days) or often (median 365 days) in the year before testing, and the authors said quitting did not fully restore functioning, though it may prevent further impairment. In the age-45 follow-up, people who had quit (60 people, no use at 45) changed by -3.3 points, not significantly different from long-term users’ -5.5. The meta-analysis shows something different: the gap with non-users shrinks after more than 72 hours of abstinence, which compares studies rather than tracking people.

Which numbers get misquoted?

• “8 IQ points” applies to the 23 adolescent-onset persistent users. All persistent dependent users lost about 6, and regular users (four or more days a week at three or more assessments) about 5.
• The twin study’s “4 points” is one Vocabulary subtest, and it did not appear within twin pairs.
• The meta-analysis figure of -0.25 is test performance across ten cognitive domains, not a change in IQ.
• The 2026 ABCD study measured no IQ at all.

What can and cannot be said?

Taken together, heavy, persistent use that starts in adolescence is the pattern most consistently linked to lower scores. The size shrinks, or vanishes, when family background, tobacco use and earlier IQ are adjusted for, and nobody has shown that scores recover. An average shift of about 2 points is smaller than the margin of error on any one person’s score, which is typically several points; our piece on the IQ margin of error explains how to read it. For another “does X lower IQ” question read the same way, see our fluoride article.

An IQ score is one snapshot on a norm group. To see where a score sits, take the IQ test and check the result on the percentile calculator.

Common questions