Fluoride and IQ: what the dose response shows

0 The Short Answer

Above roughly 1.5 milligrams of fluoride per litre of drinking water, the association between exposure and lower childhood IQ is consistent across dozens of studies and two federal reviews. Below that level, which is where community water fluoridation operates, the evidence is thin, the estimates lose statistical significance in several analyses, and the two best designed cohorts in fluoridated countries found nothing. Those two sentences are not a compromise between two camps. They are what the primary sources say when each one is read at the exposure level it measured.

The disagreement that remains is narrow and real. It is not about whether high fluoride exposure is associated with lower scores, which almost nobody now contests. It is about whether the dose response curve continues downward through the range used in fluoridated water supplies, and about whether the studies that suggest it does are strong enough to carry that weight. Serious researchers read the same tables and reach different answers, and this page names where they diverge instead of pretending one side has already won.

The second thing worth stating early is a measurement point that gets lost in the argument. The effects under discussion, at the low end, are on the order of one to three IQ points at population level. A shift that size is genuinely important for a population, because it moves the whole distribution, and it is completely invisible in any individual person. It is smaller than the standard error of measurement on most index scores, including the ones this site publishes. Both halves of that sentence are true at once, and most coverage of this topic drops one of them.

1 The Two Numbers That Decide Most of the Argument

Almost every disagreement about fluoride and intelligence resolves into a question about which of two concentrations a study was measuring. Get those two numbers in place first and most of the literature sorts itself.

The first is 1.5 milligrams per litre. That is the guideline value in the World Health Organization's drinking water guidelines, and it is the line the United States National Toxicology Program used to define higher exposure in its 2024 monograph. It is not a fluoridation target anywhere. It is a ceiling, set with dental fluorosis in mind, above which water is considered to carry too much fluoride.

The second is 0.7 milligrams per litre. That is the concentration the United States Public Health Service has recommended for community water fluoridation since 2015, when it replaced the older 1962 standard of 0.7 to 1.2 milligrams per litre that varied with local air temperature. The change was published in the Federal Register on May 1, 2015, and the stated reason was that Americans now receive fluoride from toothpaste and rinses that did not exist when fluoridation began, so a lower water concentration keeps the benefit while reducing fluorosis.

Now the consequence. A large share of the studies reporting an association between fluoride and lower IQ were conducted in regions of China, India and Iran where groundwater carries naturally high fluoride, often several times 1.5 milligrams per litre. Those studies are evidence about high natural exposure. Whether they are evidence about a fluoridated municipal supply at 0.7 is exactly the question, and it cannot be settled by counting how many of them found an effect. Pooling results obtained with different instruments in different languages carries its own problem, discussed on the page on culture reduced testing.

One further complication belongs here rather than later. Water is not the only route. Fluoridated toothpaste, dental products, tea, some foods and in some countries fluoridated salt all contribute, which is why several of the better studies measure fluoride in urine rather than in the tap. Urinary measurement captures total intake, at the cost of a different problem discussed further down: a single spot sample is a snapshot of a substance with a short half life.

2 What the National Toxicology Program Concluded, and What It Did Not

The 2024 monograph is the document most often cited on this subject, and its conclusion is more limited than the headlines built on it. The National Toxicology Program, part of the National Institute of Environmental Health Sciences, published Monograph 08 in August 2024, titled a systematic review of the state of the science concerning fluoride exposure and neurodevelopment and cognition.

The finding, stated with moderate confidence, is that higher estimated fluoride exposures are consistently associated with lower IQ in children. Higher is defined in the document itself as drinking water fluoride above the World Health Organization guideline of 1.5 milligrams per litre. The evidence base was 72 human studies of fluoride exposure and children's IQ. Of these, 19 were judged high quality, and 18 of those 19 reported an inverse association.

Two things the monograph does not say are as load bearing as the thing it does. It does not conclude that water fluoridated at United States levels lowers IQ, and it states directly that more studies are needed to understand the potential for lower fluoride exposure to affect children's IQ. It also assigned only low confidence to conclusions about cognitive effects in adults, because only two high quality cross sectional studies were available for that population, which is a striking gap given how much of the public argument concerns lifelong exposure. High quality in this context means low risk of bias on a pre-specified checklist, not that every study used the same instrument or measured the same construct.

The document's review history is worth recording, because it is often described as either bulletproof or suppressed depending on who is describing it. Taylor and colleagues, writing in Annals of Global Health in 2025, set out the sequence: a committee of the National Academies of Sciences, Engineering, and Medicine reviewed earlier drafts, the standard National Toxicology Program peer review followed with five external experts, other National Institutes of Health institutes and Centers for Disease Control staff reviewed it, and an ad hoc group convened by the Program's Board of Scientific Counselors reviewed the authors' responses. The monograph took years and several rounds. That is a fact about process, not a guarantee about substance.

3 The 2025 Meta-Analysis Read at Full Resolution

Taylor and colleagues published the quantitative synthesis in JAMA Pediatrics on January 6, 2025, and it appeared in the March issue at volume 179, issue 3, pages 282 to 292. The full text is open in PubMed Central, and reading past the abstract changes the impression the abstract leaves.

The review covered 74 studies, 64 of them cross sectional and 10 prospective cohorts. Their geography is the first thing to notice: 45 were conducted in China, 12 in India, four each in Iran and Mexico, three in Canada, two in Pakistan, and one each in Denmark, New Zealand, Spain and Taiwan. There were no studies from the United States. The authors classified 52 of the 74 as high risk of bias and 22 as low risk.

The headline figure comes from 59 studies with group level exposure measures, covering 20,932 children. The pooled standardized mean difference was minus 0.45, with a 95 percent confidence interval from minus 0.57 to minus 0.33. Heterogeneity was extreme, at an I squared of 94 percent, which means the studies were not estimating a single common quantity. Restricting to the 12 low risk of bias studies in that set cut the estimate to minus 0.19, with a confidence interval from minus 0.35 to minus 0.04, and heterogeneity still at 87 percent.

Those are standardized units, so a conversion helps. Multiplying by 15, the standard deviation of the familiar index scale explained on the page on the 15 point standard deviation, a standardized difference of 0.45 corresponds to roughly 6.8 points and one of 0.19 to roughly 2.9 points. That arithmetic is ours, not theirs, and it assumes the individual studies used scales with comparable spread, which is an approximation. For where those point values sit relative to the classification bands, see the IQ score scale.

The single most interpretable number in the paper is the individual level regression. Across 13 studies with individual data, each 1 milligram per litre increase in urinary fluoride was associated with a decrease of 1.63 IQ points, with a 95 percent confidence interval from minus 2.33 to minus 0.93. Restricted to the 11 low risk of bias studies, the estimate was 1.14 points, with an interval from minus 1.68 to minus 0.61. Those are the figures worth carrying away, because they are already in the units people argue about.

4 The Dose Response Table Is Where the Argument Actually Lives

The meta-analysis reported its dose response results split by exposure ceiling, and those splits contain the whole disagreement in a single table. Reproduced below from the paper's own reported estimates, they show what happens to the association as the analysis is restricted to progressively lower exposures.

Read the water fluoride rows in order. The association is clear when exposures up to 4 milligrams per litre are included. It loses significance below 2. Below 1.5 the point estimate flips sign and the interval is wide enough to contain anything from a substantial harm to a substantial benefit, which is what an uninformative estimate looks like. Restricting to low risk of bias studies below 1.5 returns a negative point estimate of minus 0.32, but the interval still crosses zero.

Now read the urinary rows. Below 1.5 milligrams per litre the estimate is minus 0.08 and the interval excludes zero, barely, with an upper bound of minus 0.003. That is a statistically significant association at low exposure, and it is the single strongest piece of evidence that the effect continues into the fluoridation range. It is also small: 0.08 standardized units is about 1.2 points on a 15 point scale.

The authors' own summary in the paper is that they found inverse associations and an inverse dose response, while stating that there are limited data and uncertainty in the dose response association at concentrations less than 1.5 milligrams per litre. Both clauses are in the same conclusion. Anyone quoting only one of them is quoting selectively, and both sides do it.

5 The Critiques Published Alongside It, and the Reply

JAMA Pediatrics ran two editorials with the meta-analysis, one supportive and one critical, which is unusual and tells you how the journal read the evidence. Steven M. Levy of the University of Iowa wrote the critical commentary, published the same day, arguing that caution is needed in interpreting the evidence base.

His objections are concrete rather than rhetorical. Fifty-two of the 74 studies were rated high risk of bias by the review's own authors. The effect in the low risk subgroup is much smaller than the headline. Spot urine samples are a weak measure of an individual's fluoride exposure, since urinary fluoride responds to recent intake and dietary factors and fluoride has a short half life in the body. He also identified what he described as a factual error in the abstract.

Bruce Lanphear of Simon Fraser University wrote the supportive editorial in the same issue, arguing the evidence is sufficient to warrant regulatory attention. Reporting by STAT on the day of publication collected further objections from researchers not involved in the paper. Jennifer Meyer of the University of Alaska Anchorage rejected spot urine as a basis for causal claims. Kevin Klatt of the University of California, Berkeley pointed to the absence of United States data. Loc Do of the University of Queensland asked why countries with fluoridation programs comparable to the United States, including Australia, Sweden, Denmark and Spain, were largely absent from the analysis.

That last objection is the sharpest one on the list, and it is worth restating plainly. If the question is what happens at fluoridation levels, the informative studies are the ones conducted in countries that fluoridate at those levels. Forty-five of the 74 studies came from China. The evidence base is heavily weighted toward exposures and settings that do not resemble the policy under discussion.

A parallel line of criticism targets something different: the way individual cohort papers were analysed. Guichon and colleagues published a detailed critique in Community Dentistry and Oral Epidemiology in 2024, examining six papers drawn from one Canadian cohort. Their objections and the reply to them are covered further down, in the section on that cohort.

The review authors answered their critics in print, and the exchange narrows the disagreement to one specific claim. Taylor and colleagues published a point by point response in Annals of Global Health, volume 91, issue 1, in 2025, addressing twelve recurring critiques.

On including low quality studies, they argue that standard systematic review practice is to include all studies, apply pre-specified risk of bias criteria, and analyse by quality tier, which is what they did. On urinary measurement, they note that associations held across multiple exposure metrics, drinking water and estimated intake as well as urine, so the result does not rest on spot samples alone. On heterogeneity, they distinguish heterogeneity in methods from heterogeneity in direction, reporting that 19 of the 22 high quality studies found inverse associations. On confounding, they report that 16 of 19 studies in the monograph were rated low risk of bias for confounding, and that sensitivity analyses did not make the associations disappear.

The claim that matters most is their statement that inverse associations were observed when exposure was restricted below 1.5 milligrams per litre, for both urinary and drinking water measures. Set that against the table in the previous section and the disagreement becomes visible. For urinary fluoride, the low exposure estimate does exclude zero. For drinking water below 1.5, the reported intervals cross zero in both the full set and the low risk of bias subset. Whether that counts as an inverse association observed below 1.5 depends on whether you require the interval to exclude zero.

There is a structural reason the low exposure question is hard, and it is not anybody's fault. Estimating a one to two point effect requires either a very large sample or very precise exposure measurement, and studies in fluoridated countries have neither high contrast in exposure nor huge cognitive testing samples. A study comparing 0.3 to 0.7 milligrams per litre is trying to detect a small effect across a small dose gradient. That is the hardest measurement problem in the whole field, and it is made harder by the fact that the studies did not use a common instrument, a variation catalogued in the inventory of test types in current use.

6 The Two Prospective Cohorts at Moderate Exposure

Bashash and colleagues produced the study that changed how seriously this question was taken, because it had a prenatal exposure measure and a prospective design rather than a cross sectional comparison of two villages. The paper appeared in Environmental Health Perspectives, volume 125, issue 9, article 097017, in 2017.

The data came from the Early Life Exposures in Mexico to Environmental Toxicants project, known as ELEMENT, based in Mexico City. Fluoride was measured in archived maternal urine samples collected during pregnancy and adjusted for creatinine. Cognitive outcomes were the General Cognitive Index of the McCarthy Scales of Children's Abilities at age 4 and Full Scale IQ from the Wechsler Abbreviated Scale of Intelligence at ages 6 to 12.

Complete data were available for 299 mother and child pairs, of whom 287 contributed to the age 4 analysis and 211 to the older analysis. Per 0.5 milligram per litre increase in maternal urinary fluoride, the General Cognitive Index fell by 3.15 points, with a 95 percent confidence interval from minus 5.42 to minus 0.87, and Full Scale IQ at 6 to 12 fell by 2.50 points, with an interval from minus 4.12 to minus 0.59. Mean maternal urinary fluoride was 0.90 milligrams per litre with a standard deviation of 0.35. Adjustments covered gestational age, birthweight, sex, birth order, age at testing, maternal smoking, marital status, maternal age, maternal IQ and education.

Two features of this study deserve emphasis because they cut in opposite directions. In its favour, prenatal exposure measured before the outcome removes the most obvious form of reverse causation, and maternal IQ was among the covariates, which many earlier studies lacked. Against it, contemporaneous childhood urinary fluoride showed no clear association with IQ in the same children, which is a puzzle rather than a confirmation, and Mexico's exposure environment includes fluoridated salt at 250 parts per million alongside naturally variable water between 0.15 and 1.38 milligrams per litre. That is a mix no fluoridating country reproduces.

A follow up from the same cohort, Goodman and colleagues in Environmental Research, volume 211, article 112993, in 2022, reported domain specific results across 348 mother and child pairs. Per 0.5 milligram per litre, the full scale estimate was minus 2.12 points with an interval from minus 3.49 to minus 0.75, performance IQ was minus 2.63 with an interval from minus 3.87 to minus 1.40, and verbal IQ was minus 1.29 with an interval from minus 2.60 to 0.01. The non-verbal side carried the effect, which is a testable pattern rather than a general claim. Whether that maps onto the domains described in the cognitive domains framework in any deeper sense is not established by these data.

The second cohort moved the question into a country with ordinary community water fluoridation, and the argument about its analysis has never stopped. Green and colleagues reported in JAMA Pediatrics in 2019 that maternal fluoride exposure was associated with lower IQ in boys in a Canadian cohort. The study appeared at volume 173, issue 10, pages 940 to 948, and is open in PubMed Central.

The cohort was the Maternal-Infant Research on Environmental Chemicals program, which recruited 2,001 pregnant women between 2008 and 2011 across ten Canadian cities, seven of which fluoridate. Children were assessed at ages 3 to 4 with the Wechsler Preschool and Primary Scale of Intelligence, third edition, administered by a trained assistant under psychologist supervision. Maternal urinary fluoride, adjusted for specific gravity, averaged 0.69 milligrams per litre in fluoridated communities and 0.40 in non-fluoridated ones.

The results split by sex. In the analysis of 512 mother and child pairs, each 1 milligram per litre increase in maternal urinary fluoride was associated with a decrease of 4.49 points in boys, with a 95 percent confidence interval from minus 8.38 to minus 0.60, and an increase of 2.40 points in girls, with an interval from minus 2.53 to 7.33. The interaction test gave a p value of .02. In a separate analysis of 400 pairs using estimated fluoride intake from water and beverages, each additional milligram per day was associated with a decrease of 3.66 points across the whole sample, with an interval from minus 7.16 to minus 0.15, and no significant sex interaction, with a p value of .66.

The journal's editor published an accompanying note saying the decision to publish was not easy and that the paper had been subjected to additional scrutiny of its methods and presentation. That is an unusual disclosure and worth recording as part of the document rather than as gossip about it.

Those objections were published in Community Dentistry and Oral Epidemiology in 2024, in a paper examining six publications from the same cohort. The review authors' general answer, in their 2025 response, is that associations persisted across water and intake measures and not only urine. Readers can weigh both. What cannot honestly be done is to cite the four point figure for boys without noting that it comes from a sex stratified analysis in which the girls' estimate ran in the opposite direction.

7 The Two Studies From Fluoridated Countries That Found Nothing

Two studies stand apart from the rest of this literature, because each was conducted in a developed country at fluoridation range exposures with a design that removes the usual objections, and each found no cognitive effect. They are the strongest evidence on the null side and they deserve the same scrutiny as the positive findings.

The first is Broadbent and colleagues in the American Journal of Public Health, volume 105, issue 1, pages 72 to 76, in 2015, using the Dunedin Multidisciplinary Health and Development Study, the New Zealand birth cohort of 1,037 children born in 1972 and 1973 with 95.4 percent retention at the age 38 assessment. Childhood IQ was available for 992 members and adult IQ for 942. Fluoridated areas of Dunedin ran at 0.7 to 1.0 parts per million, non-fluoridated areas at 0.0 to 0.3.

Unadjusted childhood IQ across ages 7 to 13 was 100.0 with a standard deviation of 15.1 for the 891 members in fluoridated areas, against 99.8 with a standard deviation of 14.5 for the 99 members outside them, with a p value of .92. At age 38 the figures were 100.2 for 847 members against 98.1 for 93, with a p value of .18. In adjusted models controlling for sex, socioeconomic status, birth weight and breastfeeding, the childhood coefficient was minus 0.14 with a 95 percent confidence interval from minus 3.49 to 3.20. For adult IQ, adding educational attainment, the coefficient was plus 3.00 with an interval from 0.02 to 5.98, running in favour of the fluoridated group. The obvious weakness is the comparison group: fewer than 100 people, which leaves the confidence intervals wide. The cohort is large and long enough to serve as its own reference frame, which sidesteps much of the machinery described in how norms are built, but no amount of retention fixes a comparison group that small.

The second study fixes exactly that weakness. Aggeborn and Öhman published The Effects of Fluoride in Drinking Water in the Journal of Political Economy, volume 129, issue 2, pages 465 to 491, in 2021. It uses Swedish population registers for cohorts born between 1985 and 1992, with cognitive ability taken from the military conscription test administered to men at 18 to 20, standardized within cohort to a mean of 0 and a standard deviation of 1.

Sweden does not fluoridate, which is the point. Natural fluoride varies with the bedrock beneath each water source, and because many municipalities draw on several sources, fluoride varies within municipalities in a way that is not a policy choice and is not bundled with other geological exposure. The authors exploit that variation, along with individual moving patterns, to estimate long term effects. Their sample exceeds 80,000 individuals without covariates and about 47,000 with fixed effects and covariates. Mean fluoride was 0.35 milligrams per litre, the mass of the distribution sat between 0 and 1.5, and the maximum was 4.1.

The estimates are close to zero and sometimes positive. Taking their largest negative point estimate together with the largest standard error for that specification, they report a 95 percent confidence interval that lets them rule out negative effects larger than 0.14 standard deviations per 1 milligram per litre. Expressed on a 15 point scale, that is a bound of roughly 2 points, which is our conversion of their figure. A separate analysis of the observations above 1.5 milligrams per litre showed no evidence of a negative effect up to at least 3. The same paper confirms the dental benefit and reports a positive effect on later labour income concentrated among people from lower socioeconomic backgrounds. The authors note their confidence intervals are tighter than those in the earlier literature, and that Broadbent's intervals are much wider than theirs.

8 Reading the Designs Side by Side

Laid out by design rather than by conclusion, the literature stops looking like a fight and starts looking like a set of studies answering slightly different questions. Each row below names what a study could establish and what it could not.

Three patterns fall out of the table. The studies with the largest effects have the highest exposures. The studies conducted at fluoridation range exposures in developed countries are the two that found nothing, and one of them is by an enormous margin the largest study in the field. And the two prospective cohorts that did find effects at moderate exposure, in Mexico and Canada, are the ones whose analyses have drawn the most detailed methodological criticism.

None of that is a refutation. Sweden's exposure contrast is small, which limits what a null there can rule out at higher doses. Dunedin's comparison group is tiny. The Mexican and Canadian cohorts have prenatal measurement that the register studies lack entirely. The reason this argument persists among serious people is that every design in it has a real weakness, and the weaknesses do not overlap.

What would settle it is not mysterious. A prospective cohort in a fluoridating country, with repeated exposure measurement across pregnancy rather than a single spot sample, a pre-specified primary analysis that is not sex stratified after the fact, cognitive assessment at school age rather than at 3, and a sample large enough to resolve a two point effect. Nobody has run it. Until somebody does, this stays a question about inference from imperfect data, which is the same discipline described in the page on claims about testing that survived without evidence.

9 What a Two Point Population Shift Actually Is

The effects being argued about at the low end are real at population scale and undetectable in any individual, and holding both of those facts at once is the hardest part of reading this literature honestly. This is the point where measurement expertise is worth more than another round of epidemiology.

Start with why a small population shift matters. Moving an entire distribution down by two points does not change any one person's life visibly. It changes the tails. If a population mean drops by two points with the spread unchanged, the number of people below any low threshold rises and the number above any high threshold falls, and at population scale those counts are large. That is the correct argument for taking a small effect seriously, and it is the argument public health people are making. The same logic underlies the population level reasoning in the account of the secular rise in scores, where a shift of a few points per decade reshaped how many people cleared fixed cutoffs.

Now the other half. An individual score is an estimate carrying measurement error, and that error is published. On the ACIS technical manual's figures, the standard error of measurement is 2.40 points for the Verbal Comprehension index, 2.48 for Fluid Reasoning, 3.39 for Visual Spatial, 4.02 for Quantitative Reasoning, 4.12 for Working Memory and 5.22 for Processing Speed. The composites are tighter: 1.61 for the General Ability Index, 1.92 for the reduced verbal composite and 1.99 for the abbreviated composite.

A 95 percent confidence interval runs about 1.96 standard errors either side of the observed score. On those published figures that is roughly plus or minus 4.7 points for Verbal Comprehension, 6.6 for Visual Spatial, 8.1 for Working Memory and 10.2 for Processing Speed. Even the General Ability Index, built from 15 subtests with an omega reliability of .9885, carries an interval about 6 points wide. A one to three point exposure effect disappears inside every one of those intervals. That is not a criticism of the research and it is not a defence of anything. It is what a confidence interval is for, and it is the reason the page on how a score is calculated insists that a result is a band rather than a point.

One further consequence follows for anyone tempted to test themselves after reading about this. Retesting produces score movement from practice, sleep, motivation and ordinary measurement error, and a difference of a few points between two administrations tells you nothing about any exposure. The same reasoning appears in the evidence on how scores change across the lifespan and in the discussion of environmental influence on whether scores can be raised.

10 The Other Side of the Same Decision

A fluoridation decision is a trade, and the benefit side of the ledger has its own evidence quality problem that is rarely reported alongside the harm side. Anyone assessing this question honestly has to look at both columns with the same scepticism.

The Cochrane Collaboration updated its review of water fluoridation for the prevention of dental caries in October 2024, led by Iheozor-Ejiofor and colleagues and published in the Cochrane Database of Systematic Reviews, issue 10. The review found that fluoridation reduces caries in both primary and permanent teeth in children. It also found that confidence in the effect estimates was limited by the observational nature of the study designs, high risk of bias within studies, and applicability to current lifestyles, and that contemporary studies suggest a slightly smaller benefit than the historical literature.

The reason the contemporary benefit is smaller is not controversial: most of the classic caries evidence predates the widespread use of fluoride toothpaste. When everybody already gets fluoride twice a day from a tube, the marginal contribution of the water supply is smaller than it was in 1955. That is the same reason the Public Health Service lowered its recommended concentration in 2015.

Aggeborn and Öhman's Swedish register study is unusual in measuring both sides in one design. They reconfirmed the dental benefit and found a positive effect of fluoride on later labour income, concentrated among individuals from lower socioeconomic backgrounds, alongside their zero estimate for cognitive ability. Whatever one concludes about the cognitive question, the distributional point in that paper deserves attention, because it locates the benefit in the group least likely to have consistent access to dental care. That the mechanism runs through dental health to earnings rather than through ability is consistent with the broader pattern described on the page on measured ability and income.

The public record since 2024 is a matter of documented fact rather than interpretation. In Food and Water Watch v. EPA, decided on September 24, 2024, Judge Edward Chen of the Northern District of California found that fluoridation at 0.7 milligrams per litre poses an unreasonable risk of reduced IQ in children under the Toxic Substances Control Act, and ordered the Environmental Protection Agency to initiate rulemaking. The same ruling states explicitly that it does not conclude with certainty that fluoridated water is injurious to public health, and it does not dictate what the regulatory response must be. The agency appealed. In 2025 Utah became the first state to prohibit community water fluoridation, under legislation effective in May, and Florida followed with a law effective in July.

Those are policy events, not evidence, and they belong on this page only because readers arrive having heard about them. A court applying a statutory risk standard is answering a legal question about regulatory duty, not settling a scientific one, and a state legislature is not a peer review process. This page takes no position on what any jurisdiction should do.

11 How to Read the Next Study You See About This

The transferable output of this page is a short checklist, because more studies are coming and the headlines will keep collapsing the distinctions that matter. Five questions handle most of it.

What exposure level did it actually measure? This is the first question and it disqualifies most coverage immediately. A study at 3 milligrams per litre is not evidence about a supply at 0.7. If the exposure range is not in the summary you are reading, the summary is not usable.

Was exposure measured before the outcome? Cross sectional comparisons of high and low fluoride villages carry every confound that separates those villages, including iodine status, arsenic, lead, income and schooling. Sixty-four of the 74 studies in the 2025 meta-analysis were cross sectional. Prospective measurement is a different class of evidence.

Was the reported result the primary analysis? A result that emerges from a subgroup, particularly a sex stratified subgroup where the other sex ran the other way, is a hypothesis rather than a finding. That is the core of the criticism of the Canadian cohort's most quoted number, and it is a general rule, not a fluoride specific one.

Does the confidence interval exclude zero, and how wide is it? An interval from minus 0.91 to 0.26 is compatible with a large harm and with a modest benefit. Reporting only its midpoint is the single most common distortion in this area, in both directions.

How was intelligence measured? Preschool cognitive assessment is noisier than school age assessment. Some studies used well established batteries administered individually by trained examiners, and others used brief group tests. A result is only as good as the outcome measure, which is the argument made at length on the reliability and validity page and applied to instrument choice in the comparison of free and validated tests.

Two further habits help. Distinguish an association that has not been detected from one that has been shown to be absent, since a null in a small sample bounds an effect rather than refuting it. And notice when a claim about a population is being converted into a claim about a person, which happens constantly in coverage of environmental exposures and almost never survives contact with the measurement error. The same failure mode shows up in reading about what heritability estimates do and do not say, where a population statistic is routinely read as a statement about an individual.

12 What the Evidence Supports, Stated Narrowly

The defensible summary has three parts, and each one is smaller than the claims made on either side of this argument. Stating them narrowly is the only way to keep the page useful when the next study lands.

First, at exposures above 1.5 milligrams per litre, the association between fluoride and lower childhood IQ is consistent enough that a federal systematic review assigned moderate confidence to it after multiple rounds of external review. Eighteen of 19 high quality studies in that review pointed the same direction. That finding is about high natural fluoride, and it is a legitimate reason to be concerned about groundwater in regions that carry it.

Second, at fluoridation range exposures, the evidence does not resolve. The pooled water fluoride estimate below 1.5 milligrams per litre has a confidence interval spanning both harm and benefit. The urinary estimate at that range is statistically significant and small, at about 1.2 points on a 15 point scale by our conversion of the reported standardized figure. The two studies conducted in developed countries at those exposures found nothing, and the larger of the two, covering more than 80,000 people, can rule out effects greater than about 2 points per milligram per litre.

Third, the size of the effect at issue in the low exposure range is smaller than the measurement error attached to an individual index score. That is not an argument for ignoring it, because population shifts and individual measurements are different objects. It is an argument against any individual concluding anything about themselves or their child from a test result, and against any page on this site implying that they could.

What remains unknown is specific and worth naming. Nobody has run a prospective cohort in a fluoridating country with repeated exposure measurement across pregnancy, a pre-specified primary analysis, school age cognitive assessment and the sample size required to resolve two points. Until that exists, the low exposure question stays open, and the honest description of the state of knowledge is that a real effect and no effect are both consistent with the published intervals. Readers looking at the broader question of what environmental factors move measured ability at all will find the same pattern of small, contested and hard to isolate effects in the material on schooling and measured ability and in the sibling literature covered on whether birth timing relates to measured ability, where a large apparent effect turned out to be an artefact of when the test was taken.

13 Sources Behind This Page

Every figure above is traceable to one of the following, and each is linked at the point where it is used. Where a conversion from standardized units to a 15 point scale appears in the text, it is our arithmetic on a published figure and is labelled as such.

• National Toxicology Program. NTP Monograph on the State of the Science Concerning Fluoride Exposure and Neurodevelopment and Cognition: A Systematic Review. Monograph 08, August 2024. National Institute of Environmental Health Sciences.
• Taylor KW and colleagues. Fluoride Exposure and Children's IQ Scores: A Systematic Review and Meta-Analysis. JAMA Pediatrics, 2025, volume 179, issue 3, pages 282 to 292. Published online January 6, 2025.
• Levy SM. Caution Needed in Interpreting the Evidence Base on Fluoride and IQ. JAMA Pediatrics, 2025. Published with the meta-analysis alongside a supportive editorial by Bruce Lanphear.
• Taylor KW and colleagues. Addressing Critiques of the Evidence Linking Fluoride and Children's IQ. Annals of Global Health, 2025, volume 91, issue 1.
• Bashash M and colleagues. Prenatal Fluoride Exposure and Cognitive Outcomes in Children at 4 and 6 to 12 Years of Age in Mexico. Environmental Health Perspectives, 2017, volume 125, issue 9, article 097017.
• Goodman CV and colleagues. Domain specific effects of prenatal fluoride exposure on child IQ at 4, 5 and 6 to 12 years in the ELEMENT cohort. Environmental Research, 2022, volume 211, article 112993.
• Green R and colleagues. Association Between Maternal Fluoride Exposure During Pregnancy and IQ Scores in Offspring in Canada. JAMA Pediatrics, 2019, volume 173, issue 10, pages 940 to 948.
• Guichon JR and colleagues. Flawed MIREC fluoride and intelligence quotient publications. Community Dentistry and Oral Epidemiology, 2024.
• Broadbent JM and colleagues. Community Water Fluoridation and Intelligence: Prospective Study in New Zealand. American Journal of Public Health, 2015, volume 105, issue 1, pages 72 to 76.
• Aggeborn L and Öhman M. The Effects of Fluoride in Drinking Water. Journal of Political Economy, 2021, volume 129, issue 2, pages 465 to 491.
• Iheozor-Ejiofor Z and colleagues. Water fluoridation for the prevention of dental caries. Cochrane Database of Systematic Reviews, October 2024, issue 10.
• United States Public Health Service. Recommendation for Fluoride Concentration in Drinking Water for the Prevention of Dental Caries. Federal Register, May 1, 2015.
• ACIS reliability, standard error of measurement and g loading figures are from the published ACIS technical manual. The manual documents the adult English speaking reference frame.

Reporting by STAT on January 6, 2025 was used for the researcher comments attributed to Jennifer Meyer, Kevin Klatt and Loc Do, and is linked in the section on critiques. The 2024 federal court decision described in this page is Food and Water Watch v. Environmental Protection Agency, Northern District of California, decided September 24, 2024, and the state legislation referenced is Utah's 2025 law effective in May and Florida's 2025 law effective in July.

The professional framework governing every interpretive claim on this site is explicit. The Standards for Educational and Psychological Testing (2014), published jointly by the American Educational Research Association, the American Psychological Association and the National Council on Measurement in Education, require that score interpretations be supported by evidence for the specific use proposed, that reliability and measurement error be reported alongside every score, and that the limits of the reference sample be disclosed. APA standards on test use and the International Test Commission guidelines on test use say the same in different words. Under those standards, an unsupervised online score describes a person's current standing on a set of tasks. It is not evidence about what any environmental exposure did to them, and this site will not claim otherwise. Researchers who need verified administrations with participant level export can use the research workspace or administer the assessment to participants through private links.

14 Frequently Asked Questions