What can lower your IQ, and what only lowers a single test score

0 The short answer

Only a few exposures have evidence that they lower measured IQ in a lasting way, the clearest being lead in childhood and severe early deprivation, while sleep loss, anxiety and scarcity mainly lower performance on the day of a test. In pooled cohorts of 1,333 children, a rise in blood lead from 2.4 to 30 micrograms per deciliter went with a 6.9 point lower IQ, and a randomized trial of foster care for institutionalized children found a 9 point advantage at age 18. For alcohol disorders, fluoride, cannabis, breastfeeding, fitness and screen time the best designs give smaller, disputed or null answers, and the sizes come in different units, so they cannot be added into one list of points lost.

1 What Does It Mean for Something to Lower IQ?

An exposure lowers IQ in the sense that matters only if it reduces the underlying ability that a score estimates, for a lasting period, in people who started at the same level. That is a stricter claim than saying that people exposed to something score lower. Three different things hide inside the phrase "lowers IQ", and a headline usually blurs them.

The first is a lasting loss of ability: a person measured before the exposure and measured again years afterward scores lower than a comparable person who was not exposed. The second is a temporary drop in performance: the ability is intact, but attention, working memory or speed is worse on the day, so the score is worse on the day. The third is a difference that was already there: people who drink heavily, use cannabis or grow up in poor households may have differed from other people before the exposure began, so a gap measured afterward is partly or wholly a starting gap. The page on what a low IQ is covers the score ranges below 85 and 70, and this page does not repeat them.

Measured IQ here means a score on a standardized test with a mean of 100 and a standard deviation of 15, which the page on how IQ scores are normed and the page on the standard deviation of 15 explain. Any single score also carries measurement error, and the page on reliability and validity shows why two sittings of the same person rarely match exactly. A change of a point or two between two tests is inside that noise. A difference of several points in a group, found again in other samples, is a different kind of signal.

Many of the studies below do not measure a full IQ at all. They measure one narrower ability, usually working memory, attention or processing speed, which the CHC model places in the families called Gwm and Gs. A narrow result is reported as such, and the page on what IQ measures shows how those abilities sit inside a complete battery.

This page does not give medical advice, does not tell any reader what caused their own score, and does not describe any person. It covers 11 exposures that readers ask about. Each has its own page on this site with the full literature, and each is linked where it appears.

2 How Is the Evidence Ranked, and Why Are the Numbers Not Added Up?

The ranking asks five questions of the best study for each exposure, and it is an editorial judgment made for this page, not a score assigned by any of the authors. The questions are listed here so that a reader can rerun the ranking with a different weighting.

1. Was ability measured before the exposure began, so that a starting gap can be seen?
2. Was the exposure assigned at random, or was family background separated out by comparing twins or siblings?
3. Does the effect grow with the dose of the exposure?
4. Do independent samples agree?
5. Was the effect still there months or years later?

An exposure that earns a yes on most of these is ranked above one that earns a yes on few. Ties are broken by the size and independence of the sample. The five study designs that appear below separate different things, as the next table shows.

The sizes cannot be added up. The lead estimate is in IQ points across a range of blood lead, the sleep estimate is a standardized difference on attention tasks, the anxiety estimate is a standardized difference on working memory tasks, and the fluoride estimate is points per unit of urinary fluoride. Some outcomes are a full IQ, some are a single subtest, and some are a change over decades. A column of numbers drawn from them would look precise and would mean nothing, and a total of "points lost" across the 11 exposures would be an invention. The table in the next section therefore reports each effect in the unit its own authors used, and says what that unit is. Our arithmetic is flagged wherever it appears.

3 The 11 Exposures at a Glance: Design, Sample, Size and Whether It Lasts

The table ranks the exposures by the strength of the best study for each, and the two top rows are the only ones where the effect on a measured IQ is sizable and shown in a design that could separate it from background. Read it left to right: the best design, the sample, the result as the authors published it, and whether the study can say the effect persists. The final column is our evidence grade for the question "does this lower measured IQ in a lasting way", with the reasoning in the sections that follow.

4 What Is the Evidence That Childhood Lead Lowers IQ?

Lead has the strongest evidence of the 11 exposures because independent cohorts, measured blood levels and a dose response all point the same way, though the effect shown is in childhood IQ and for groups. The pooled analysis by Lanphear and colleagues combined data from seven international population based cohorts, 1,333 children in all, who were followed from birth or infancy until 5 to 10 years of age. The outcome was Full Scale IQ. The geometric mean blood lead of the children peaked at 17.8 micrograms per deciliter and fell to 9.4 by ages 5 to 7. Of the children, 244 (18 percent) never exceeded 10 micrograms per deciliter at their maximum measurement.

After adjusting for covariates the authors found an inverse relationship between blood lead and IQ. Using a log-linear model, an increase in concurrent blood lead from 2.4 to 30 micrograms per deciliter was associated with a 6.9 point decrement (95 percent confidence interval 4.2 to 9.4). The decrements by step were 3.9 points from 2.4 to 10, 1.9 points from 10 to 20, and 1.1 points from 20 to 30. Those three steps add to 6.9 by our arithmetic, and they describe a curve that is steepest where the exposure is lowest. For a given increase in lead, the decrement among children whose maximum blood lead stayed below 7.5 micrograms per deciliter was significantly larger than among those who went higher (p = 0.015).

Three limits belong next to those figures. First, the design is observational: the authors adjusted for covariates, but no one assigned the lead, and the lead was a concurrent measure, so the study cannot show how long the deficit lasts. Second, the number is a group estimate across a dose range. It does not say that a child with a given blood lead lost 6.9 points, and it says nothing about an adult whose exposure is unknown. Third, observational cohorts can be affected by confounders that no one measured, which is why the ranking places this design below a randomized trial even though a dose response is a strength. The page on environmental toxins and IQ sets lead beside mercury, pesticides, air pollution and the other contaminants, with the units each study used.

Why does lead rank first when its estimate is smaller than the 9 points in the next row? Because its evidence meets four of the five questions: ability was tracked in cohorts from early life, the exposure was measured in blood rather than recalled, the result has a dose response, and independent samples agree on its direction. The fifth question, persistence into adult life, is not answered by this study, and the ranking says so.

5 What Does the Randomized Trial of Early Deprivation Show?

Severe early deprivation has the only randomized evidence in the set that an environment can change IQ by about 9 points, but the exposure is institutional care, which is far from ordinary family stress. The Bucharest Early Intervention Project randomized young children who had been exposed to severe psychosocial deprivation to foster care or to care as usual in an institution. Humphreys and colleagues report data from 135 participants assessed at age 18, 16 years after randomization. Those who had been randomized to high quality foster care had significantly higher IQ scores, 9 points or 0.6 standard deviations, than those randomized to care as usual. The authors report that mediation analyses suggest the effect was explained in part by higher quality caregiving and attachment security, and that exploratory analyses link stable placements through childhood with the largest long-term gains.

The arithmetic is consistent with the scale. A difference of 0.6 standard deviations on a scale with a standard deviation of 15 is 9 points, which is our check and matches the authors' pairing of 9 points with 0.6.

What the result supports is narrow. It shows that the quality of caregiving in the first years of life can change measured IQ by a large amount in children who start from severe deprivation, and that the change is still present at 18. It does not show that ordinary stress, a hard childhood or a difficult year lowers IQ by anything like that. The page on stress and IQ sorts the studies of acute stress, chronic stress and childhood adversity by design, and the estimates there depend heavily on what was compared. It also matters which way the trial runs. It shows the gain from foster care, so it speaks to what a deprived environment held back, not to what any single hard event takes away.

Because it is a randomized study, this row earns a strong grade on the first four questions, with a follow-up of 16 years for the fifth. It ranks second rather than first only because the sample is small, 135 people, and the exposure is an unusual one, so its reach is narrower than lead's.

6 What About Alcohol and Fluoride, Where the Evidence Is Moderate?

For alcohol related disorders and for fluoride above 1.5 milligrams per liter the evidence is moderate, because each has a real signal and each has a design problem that keeps it out of the top two rows. Alcohol is a case where a retest exists and shows that most of the gap was there before the diagnosis. Fluoride is a case where many studies exist and most are rated at high risk of bias.

Alcohol related disorders

Grønkjær and colleagues used a follow-up of middle aged Danish men who sat the same intelligence test again that they had taken at military conscription. The analytic sample was 2,499. Alcohol related hospital diagnoses, obtained from national registries, were present in 8 percent. Those men had a significantly lower baseline IQ than the others (95.0 against 100.5) and a larger decline from baseline to follow-up (minus 8.5 against minus 4.8). The larger decline stayed significant after adjustment for year of birth, retest interval, baseline IQ, education, smoking, alcohol consumption and psychiatric and somatic comorbidity. The authors conclude that the lower mean ability seen in people with alcohol related disorders is probably a result of both lower premorbid intelligence and more intellectual decline.

Our subtraction of the two declines gives 3.7 points of extra decline, and the baseline gap of 5.5 points was present before any diagnosis. Following those means, the gap at follow-up is about 9.2 points, so the baseline gap is more than half of it. That is why the page on alcohol and IQ argues that cross-sectional comparisons of drinkers and non-drinkers overstate what drinking did, while the retest shows a genuine extra decline. The sample is male, Danish and defined by hospital diagnoses or self-reported treatment, so the findings do not describe moderate drinking.

Fluoride in children

The largest synthesis is a systematic review and meta-analysis by Taylor and colleagues of 74 studies, 64 cross-sectional and 10 cohort, most of them in China. Fifty-two were rated at high risk of bias and 22 at low risk. Across 59 studies with group level measures (20,932 children) the pooled standardized mean difference was minus 0.45 (95 percent CI minus 0.57 to minus 0.33). In 31 studies of fluoride measured in drinking water the dose response difference was minus 0.15, but the association was null at less than 1.5 milligrams per liter, and the authors state that data are limited and the dose response uncertain at those lower concentrations. In 13 studies with individual level measures, each 1 milligram per liter increase in urinary fluoride went with a 1.63 point lower IQ (95 percent CI minus 2.33 to minus 0.93), and 1.14 points among low risk of bias studies.

The National Toxicology Program, on its page for the fluoride assessment, which we opened on October 13, 2026, states with moderate confidence that exposure above 1.5 milligrams of fluoride per liter of drinking water is associated with lower IQ in children, and that data are insufficient to determine an effect at 0.7 milligrams per liter, the level recommended for United States water supplies. The same page says the program found no evidence of adverse effects on adult cognition. This page takes no position on water policy. The page on fluoride and IQ walks through the dose response and the disputes about study quality.

The ranking logic is the same for both: a real association, no randomization, and a question about direction or dose that has not closed. Fluoride sits below alcohol because most of its studies are cross-sectional, while the alcohol result comes from retesting the same men over decades.

7 Do Short-Term Sleep Loss, Anxiety and Scarcity Lower IQ, or Only a Score?

Short-term sleep loss, anxiety and scarcity have solid evidence that they lower performance on cognitive tasks, but the evidence is about attention and working memory on the day, and none of it shows a loss of ability. This group matters most to a reader worried about a score, because these are the exposures that can change a result from one sitting to the next.

Sleep loss

Lim and Dinges meta-analyzed 70 articles with 147 cognitive tests on short-term total sleep deprivation, defined as under 48 hours. They separated six cognitive categories: simple attention, complex attention, working memory, processing speed, short-term memory and reasoning. Effect sizes ranged from small and nonsignificant (reasoning accuracy, g of minus 0.125, 95 percent CI minus 0.27 to 0.02) to large (lapses in simple attention, g of minus 0.776, 95 percent CI minus 0.96 to minus 0.60). The two extremes are the useful part: a night without sleep shows up first in lapses of attention, and reasoning accuracy barely moved. The page on sleep and IQ separates the night before a test from chronic sleep habits.

Anxiety

Moran's meta-analysis of 177 samples (22,061 people) found that self-reported anxiety was reliably related to poorer working memory capacity, with g of minus 0.334. The estimate was minus 0.342 for complex span tasks, minus 0.318 for simple span tasks such as digit span, and minus 0.437 for dynamic span tasks such as the N-back. The review also reports that anxiety, whether self-reported or experimentally induced, is related to poorer performance across a wide variety of tasks. Two cautions follow. A correlation between a questionnaire and a task does not say that anxiety lowered the ability, and the outcome is working memory, not IQ. The page on anxiety and IQ covers test anxiety, and the subtest page for Digit Span shows how working memory span is measured inside a battery.

Poverty and scarcity

Mani and colleagues report two studies. In the first, experimentally induced thoughts about finances reduced cognitive performance among poor but not well-off participants. In the second, the same farmers showed diminished cognitive performance before harvest, when poor, compared with after harvest, when richer. The authors state that this cannot be explained by differences in time available, nutrition or work effort, or by stress, and they propose that poverty related concerns consume mental resources. The abstract gives no IQ point figure, and this page does not convert one. The within-person design is the strength of the second study, because the same person is compared with themselves. What it shows is a state that changes with circumstance, not a fixed trait. The page on poverty and IQ covers the cash transfer trials and the wider literature on income and child outcomes.

The common thread is that these are effects on how well a person performs under conditions, which is the reason the conditions at the time of testing matter. They fall below the lasting effects in the ranking because none of the three studies tests persistence, and a transient dip in performance is not a lower IQ.

8 Does Cannabis Lower IQ? One Exposure, Two Designs, Two Answers

Cannabis is the clearest case on this list of a result that depends on the design, because a birth cohort found a decline and two twin studies found that the decline was not specific to the user. The cohort study is Meier and colleagues, who followed the Dunedin birth cohort of 1,037 people from birth in 1972 and 1973 to age 38. Cannabis use was ascertained at ages 18, 21, 26, 32 and 38, and neuropsychological testing was done at age 13, before cannabis was used, and again at 38. Persistent cannabis use was associated with decline broadly across domains, even after controlling for years of education. The impairment was concentrated among adolescent-onset users, and stopping did not fully restore functioning among them. The authors read this as suggestive of a neurotoxic effect on the adolescent brain.

Jackson and colleagues tested the same question in two longitudinal studies of adolescent twins (789 and 2,277 participants), with standardized intelligence measures at ages 9 to 12, before cannabis involvement, and again at ages 17 to 20. Users had lower scores than nonusers and showed a significant decline in crystallized intelligence. But there was no evidence of a dose response between frequency of use and IQ change, and marijuana-using twins did not show a significantly greater IQ decline than their abstinent siblings. The authors suggest that the decline may be attributable to familial factors that underlie both marijuana initiation and low intellectual attainment.

The designs differ in what they control. The cohort has a decades-long follow-up and a measure taken before use, and the twins have family background built into the comparison, but they follow adolescents only from preadolescence to late adolescence. The sibling comparison is the sharper test of cause for the same reason: it holds fixed what the siblings share. A dose response is the other test, and the twin studies did not find one. That is why this row is graded contested rather than moderate. The page on cannabis and IQ walks through the replication record, including the critique of the cohort result and the cohort's later report. For a reader the lesson is not that cannabis is harmless, which neither study claims, but that "users score lower" and "cannabis lowered the score" are different statements.

9 What Happened to Breastfeeding, Fitness and Screen Time When the Design Got Stronger?

Breastfeeding, fitness and screen time each carried a popular claim about IQ, and in each case the answer changed or shrank when the design separated the exposure from family background. The three have little else in common, so they are taken one at a time.

Breastfeeding

The cluster randomized trial by Kramer and colleagues was a cluster randomized trial in 31 Belarussian maternity hospitals. A total of 17,046 healthy breastfeeding infants were enrolled and 13,889 (81.5 percent) were followed at age 6.5. The intervention raised exclusive breastfeeding at 3 months to 43.3 percent against 6.4 percent in the control group. The experimental group had higher means on all Wechsler Abbreviated Scales of Intelligence measures, with cluster-adjusted mean differences of plus 7.5 for verbal IQ (95 percent CI plus 0.8 to plus 14.3), plus 2.9 for performance IQ (CI minus 3.3 to plus 9.1) and plus 5.9 for full-scale IQ (CI minus 1.0 to plus 12.8). The full-scale interval includes zero.

Der, Batty and Deary analyzed 5,475 children of 3,161 mothers in a United States national survey. Maternal IQ predicted whether a mother breastfed better than race, education, age, poverty status, smoking, the home environment or the child's birth weight or birth order, and one standard deviation of maternal IQ more than doubled the odds of breastfeeding. Before adjustment, breastfeeding went with about 4 points of mental ability. Fully adjusted, the effect was 0.52 (95 percent CI minus 0.19 to 1.23), and the sibling comparisons and meta-analysis corroborated that. The two studies disagree, and the disagreement is informative: the trial randomized a promotion program, not breastfeeding itself, and its precision on full-scale IQ is low. Both are why this row is graded weak and conflicting. The page on breastfeeding and IQ sets out the confounding. None of this speaks to any other benefit of breastfeeding.

Fitness and physical activity

Åberg and colleagues analyzed all Swedish men born from 1950 to 1976 who enlisted for military service at age 18 (1,221,727), including 268,496 full-sibling pairs, 3,147 twin pairs and 1,432 monozygotic twin pairs. Cardiovascular fitness, measured by ergometer cycling, was positively associated with intelligence after adjustment for confounders (regression coefficient 0.172, 95 percent CI 0.168 to 0.176), with similar results within monozygotic twin pairs, while muscle strength was not associated with cognitive performance. The authors' cross-twin analyses attributed the association mainly to individual-specific, non-shared environmental influences. Changes in fitness between ages 15 and 18 predicted cognitive performance at 18. The sample is very large and the twin analysis is a strength, but the study ties fitness to intelligence at one age in young men and does not show a lowering of IQ by inactivity later in life. The page on exercise and IQ covers the randomized trials.

Screen time and social media

Sauce and colleagues analyzed 9,855 children from the Adolescent Brain Cognitive Development dataset in the United States, with intelligence measured at baseline (ages 9 to 10) and after two years, controlling for genetic differences in cognition and for socioeconomic status. At baseline, time watching (r of minus 0.12) and socializing (r of minus 0.10) were negatively correlated with intelligence, and gaming was not. Two years later gaming had a positive effect (standardized beta of plus 0.17), socializing had no effect, and watching videos had a small positive effect (plus 0.12), though the last was not significant in a post hoc analysis that controlled parental education instead of socioeconomic status. The study measures change over two years in children, not adults, and it does not address attention, sleep or well-being. The page on social media and IQ sets this beside the smartphone presence experiments and the observational literature.

10 What Raises Measured IQ, and What Is Outside This List?

Schooling is the best documented exposure that raises measured IQ, and it is on this page only as a contrast, because it shows that the same scale moves upward by the same logic that the list above moves it downward. Ritchie and Tucker-Drob meta-analyzed three categories of quasiexperimental studies of educational effects on intelligence: studies controlling for earlier intelligence, studies using compulsory schooling changes as instrumental variables, and regression-discontinuity studies using school-entry age cutoffs. Across 142 effect sizes from 42 data sets with over 600,000 participants they found consistent evidence of beneficial effects of approximately 1 to 5 IQ points for an additional year of education. They report that the effects persisted across the life span and were present on all broad categories of cognitive ability studied.

The meta-analysis measures gains from additional education. It does not measure losses from missing school, and this page does not turn one into the other. It is a reminder that a measured IQ is not a fixed property, which the page on whether you can improve your IQ develops.

Several causes of lower scores are not on this list because they are conditions with their own literatures, not exposures with a table of studies here, and this page states no figures for them. Genetic conditions that cause intellectual disability, brain injury, birth complications and prematurity, dementia and other diseases of aging, and medical conditions are covered, where the site covers them, by the pages on what a low IQ is, how IQ changes with age and whether IQ is genetic. A reader who has a concern about a specific condition should ask a clinician, not a ranking of group studies.

11 What the Evidence Supports, Stated Narrowly

The evidence supports four statements, and each is narrower than the headlines that circulate about it.

1. Lead in childhood is associated with lower IQ in children, with a dose response. In 1,333 children from seven cohorts the estimate was 6.9 points across a range of blood lead, with the steepest slope at the lowest levels. That is a group estimate from an observational design, in children, and persistence into adulthood is not tested by that study.
2. Severe early deprivation lowered IQ and good foster care raised it. In a randomized trial the difference was 9 points, 0.6 standard deviations, at age 18. The exposure is institutional care and does not extend to ordinary family stress.
3. Alcohol related disorders and fluoride above 1.5 milligrams per liter are associated with lower scores, with caveats of different kinds. In the Danish retest, the men diagnosed with an alcohol related disorder declined 3.7 points more over the same decades (our subtraction), and more than half of their final gap was already present at conscription. For fluoride, the meta-analysis found inverse associations, most studies were rated at high risk of bias, and the drinking water result was null below 1.5 milligrams per liter.
4. Sleep loss, anxiety and scarcity lower performance on tasks of attention and working memory at the time, and the studies do not show a lasting loss of ability. The effects are standardized differences, not IQ points.

The other four rows do not support a harm to IQ as popularly described. The cannabis twin studies found no greater decline than abstinent siblings, the breastfeeding trial has a full-scale interval that includes zero and the adjusted survey estimate is 0.52, the fitness study is an association at age 18, and the screen time study found no effect of socializing and small positive effects of watching and gaming. All of these are statements about groups. None is a statement about any person's cause of a score.

12 What Does This Say About Your Own Score, and How Should You Read It?

Nothing on this page tells you why your own score is what it is, because every estimate here describes a group, a dose range or a design, and a score is read against conditions, error and a reference group. The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014), described on the Standards' own site, which we opened on October 13, 2026, is the joint publication of the American Educational Research Association, the American Psychological Association and the National Council on Measurement in Education that sets expectations for how test scores should be interpreted and used. The Standards treat the validity of a score as a matter of the interpretation and use it is put to, and the circumstances of testing, including the state of the person, belong in that interpretation. APA is one of the three publishers.

Four practical consequences follow. A single score is an estimate with error, and the page on reliability and validity explains why a range is more honest than a point. A score taken after a short night or in a state of high anxiety can understate working memory and speed, which is the finding of rows 5 and 6. Two scores a few points apart are not evidence that anything changed. And a group result such as the lead estimate cannot be turned into a statement about what happened to one child, one adult or one exam.

ACIS is one way to take a normed test at home. It has 20 subtests in six CHC domains and reports a Full Scale IQ, six primary indices, percentiles and a 95 percent confidence interval, with scaled subtest scores from 1 to 19. On the prices read on October 13, 2026, the Quick form is 15 dollars, the Optimized form 30 dollars and the Full Scale form 50 dollars, each a one time payment with no subscription, breaks are allowed and the 30 days to finish let a person take the test rested, and there is a free trial with no card. The test is online and unsupervised, not a clinical or diagnostic instrument, and not suitable for hiring, school accommodations or admission to high IQ societies. It is in English only. Anyone worried about a change in their thinking, after an illness, an injury or a long period of heavy substance use, should see a clinician and not rely on an online score. The page on how IQ scores are normed and the technical manual document how the score is produced.

13 Sources Behind This Page

Journal records were checked through Crossref and PubMed on October 13, 2026, and the two web pages were opened on that date. Every figure on this page is given as its source prints it, and where a figure is our subtraction or arithmetic the body says so.

• Lanphear B, Hornung R, Khoury J, Yolton K, Baghurst P, Bellinger D, Canfield R, Dietrich K and colleagues. Low-Level Environmental Lead Exposure and Children's Intellectual Function: An International Pooled Analysis. Environmental Health Perspectives, 2005, volume 113, issue 7, pages 894 to 899.
• Humphreys K, King L, Guyon-Harris K, Sheridan M, McLaughlin K, Radulescu A, Nelson C, Fox N and Zeanah C. Foster care leads to sustained cognitive gains following severe early deprivation. Proceedings of the National Academy of Sciences, 2022, volume 119, issue 38, article e2119318119.
• Grønkjær M, Flensborg-Madsen T, Osler M, Sørensen H, Becker U and Mortensen E. Intelligence Test Scores Before and After Alcohol-Related Disorders: A Longitudinal Study of Danish Male Conscripts. Alcoholism: Clinical and Experimental Research, 2019, volume 43, issue 10, pages 2187 to 2195.
• Taylor K, Eftim S, Sibrizzi C, Blain R, Magnuson K, Hartman P, Rooney A and Bucher J. Fluoride Exposure and Children's IQ Scores: A Systematic Review and Meta-Analysis. JAMA Pediatrics, 2025, volume 179, issue 3, pages 282 to 292.
• National Toxicology Program, National Institute of Environmental Health Sciences. Fluoride assessment page (noncancer assessments, completed), opened October 13, 2026.
• Lim J and Dinges D. A meta-analysis of the impact of short-term sleep deprivation on cognitive variables. Psychological Bulletin, 2010, volume 136, issue 3, pages 375 to 389.
• Moran T. Anxiety and working memory capacity: A meta-analysis and narrative review. Psychological Bulletin, 2016, volume 142, issue 8, pages 831 to 864.
• Mani A, Mullainathan S, Shafir E and Zhao J. Poverty Impedes Cognitive Function. Science, 2013, volume 341, issue 6149, pages 976 to 980.
• Meier M, Caspi A, Ambler A, Harrington H, Houts R, Keefe R, McDonald K, Ward A, Poulton R and Moffitt T. Persistent cannabis users show neuropsychological decline from childhood to midlife. Proceedings of the National Academy of Sciences, 2012, volume 109, issue 40, pages E2657 to E2664.
• Jackson N, Isen J, Khoddam R, Irons D, Tuvblad C, Iacono W, McGue M, Raine A and Baker L. Impact of adolescent marijuana use on intelligence: Results from two longitudinal twin studies. Proceedings of the National Academy of Sciences, 2016, volume 113, issue 5, pages E500 to E508.
• Kramer M, Aboud F, Mironova E, Vanilovich I, Platt R, Matush L, Igumnov S, Fombonne E and colleagues. Breastfeeding and child cognitive development: new evidence from a large randomized trial. Archives of General Psychiatry, 2008, volume 65, issue 5, pages 578 to 584.
• Der G, Batty G and Deary I. Effect of breast feeding on intelligence in children: prospective study, sibling pairs analysis, and meta-analysis. BMJ, 2006, volume 333, issue 7575, page 945.
• Åberg M, Pedersen N, Torén K, Svartengren M, Bäckstrand B, Johnsson T, Cooper-Kuhn C, Åberg N, Nilsson M and Kuhn H. Cardiovascular fitness is associated with cognition in young adulthood. Proceedings of the National Academy of Sciences, 2009, volume 106, issue 49, pages 20906 to 20911.
• Sauce B, Liebherr M, Judd N and Klingberg T. The impact of digital media on children's intelligence while controlling for genetic differences in cognition and socioeconomic background. Scientific Reports, 2022, volume 12, article 7720.
• Ritchie S and Tucker-Drob E. How Much Does Education Improve Intelligence? A Meta-Analysis. Psychological Science, 2018, volume 29, issue 8, pages 1358 to 1369.
• American Educational Research Association, American Psychological Association and National Council on Measurement in Education. Standards for Educational and Psychological Testing, 2014, site opened October 13, 2026.

14 Frequently Asked Questions