Two presidents left a test score. The rest of the ranking is a statistical reconstruction

0 Quick answer: two records, forty two estimates, one hoax

No president of the United States has a published score from a modern, individually administered intelligence test, so every presidential IQ ranking in circulation was produced some other way. Two kinds of real evidence exist and they are not equally strong. Two presidents left a documented adolescent test result. One peer reviewed study published estimated figures for all 42 chief executives from George Washington to George W. Bush.

The documented pair is small and old. Daniel Seligman, writing in Fortune on 15 July 1991 in a column titled The Most Important Stat, reported that Richard Nixon's biographer Roger Morris put Nixon at 143 on a test taken at Fullerton Union High School in California, and that John F. Kennedy's biographer Thomas C. Reeves put Kennedy at 119 on a test taken shortly before he entered Choate. Both sat an Otis group test. Neither result is a modern score.

The estimates are the source of nearly every list you will meet. They come from Dean Keith Simonton's paper in Political Psychology in 2006, and they are reconstructions rather than measurements. Only eight of the 42 presidents carried any IQ figure at all before the reconstruction ran. The other thirty four numbers are model output, a point Simonton makes himself when he calls the IQ scores "more tentative and approximate" than his other two measures. A published estimate of this kind is the third of four evidence classes a famous IQ claim can fall into, and the audit that sorts those claims by what has to exist before one can be checked finds twenty five of its thirty five cases in the fourth class, where no origin can be named at all.

A third category also circulates, and it deserves naming immediately because many readers arrive having seen it. In 2001 a fabricated ranking attributed to a nonexistent body called the Lovenstein Institute spread by email, reached national newspapers, and is still quoted. It was a hoax. Section nine sets out the documented record of how it travelled, and section ten uses it as the worked example of how an unsourced figure turns into common knowledge.

1 The only two documented presidential scores

Two presidents took a group intelligence test at school and the result was later found and published by their biographers. That is the entire documented record. Kennedy's 119 comes from Thomas C. Reeves, whose 1991 biography places the test shortly before Kennedy entered Choate. Nixon's 143 comes from Roger Morris, whose Nixon biography places the test at Fullerton Union High School. Seligman collected both in his Fortune column and later folded the material into his 1992 book on the American intelligence debate.

Both figures should be read with the care their own reporter applied. Seligman noted in the same column that the standard error of the Otis was about six IQ points, which he glossed as meaning there are two chances in three that the true value lies within six points of the reported one. That is the one standard error band. Widen it to the conventional ninety five percent interval and each score becomes a range roughly twelve points either side: about 107 to 131 for Kennedy, about 131 to 155 for Nixon.

Notice what that arithmetic does. The gap between the two men is 24 points, and the two intervals fail to overlap by less than a single point. The difference is real on the face of the record and it is only barely larger than the measurement error of the instrument that produced it, before anything is said about test dates, norm groups or the conditions of a school hall administration. Anyone quoting either number as a fixed personal attribute is quoting a point estimate and discarding its band.

There are further limits worth stating plainly. Neither school record is publicly available. What is documented is a biographer's report of a school record, which is strong evidence but not the same as a published test result with a manual behind it. The instrument is identified only as an Otis, without a form or a norm year. And a score obtained in adolescence is a score obtained in adolescence, not a lifetime constant. The site sets out the general logic of a score band on the page covering how measurement quality is judged, and how norms fix meaning on how a norm table is built. Nor is a school hall sitting the kind of administration a modern individual score comes from: the clinical batteries are restricted at the point of sale to Level C purchasers, normally a doctorate with formal assessment training or a state licence, so both the person who gives the test and the person who interprets it are identified.

2 What historiometric estimation actually is

Historiometry applies quantitative methods to historical records, and in the intelligence literature it means inferring a likely ability level from what a person's documented life contains. The technique is a century old. Catharine Cox, working under Lewis Terman at Stanford, published the founding study in 1926 as the second volume of the Genetic Studies of Genius series, covering 301 eminent historical figures.

Her method was specific rather than impressionistic. Teams of independent raters built chronologies of each subject's childhood and adolescent accomplishments, then compared the age at which each accomplishment appeared against the age at which the general population would be expected to reach it. If a boy read fluently at three and composed verse at seven, the ratio of that inferred mental age to his actual age produced a figure. Cox then expressed it using the intelligence quotient formula of the period, mental age divided by chronological age, multiplied by a hundred. That formula is explained on the page covering the ratio formula that produced those early figures.

Cox also understood her method's central weakness and tried to correct for it. Subjects whose childhoods were thoroughly written down produced more evidence of precocity than subjects whose childhoods were not, regardless of ability. She therefore applied an upward statistical correction where the biographical record was thin, which is why every Cox subject has both an uncorrected and a corrected figure. The mechanics of that correction are worked through on the page about how Cox's correction for missing records worked.

The failure mode follows directly. When the primary input is the surviving archive, the output partly measures the archive. A wealthy, literate, well biographied family generates more datable childhood milestones than a poor one, and the score rises accordingly. This is why the honest description of Cox's numbers, given on the page about what Cox's numbers actually track, is that they restate how thoroughly a life was recorded. Later researchers, Simonton foremost among them, found real statistical structure in the Cox data and used it for group level research. None of them treated a Cox figure as a measurement of a person, and the same discipline is applied on the page about the Cox estimate attached to Isaac Newton.

3 Inside the 2006 study that produced the ranking

The paper behind almost every presidential IQ list is Dean Keith Simonton's Presidential IQ, Openness, Intellectual Brilliance, and Leadership: Estimates and Correlations for 42 U.S. Chief Executives, in Political Psychology, volume 27, issue 4, pages 511 to 526, published in August 2006. The full text is available as a public PDF, and every figure quoted on this page can be checked against it.

The sample is 42 rather than 43 because Grover Cleveland served two nonconsecutive terms and is one man occupying two numbered presidencies. George W. Bush was the 43rd presidency and the 42nd president. That arithmetic quietly explains why some lists say 42 and others say 43 while showing the same people.

Simonton did not administer anything. He assembled three existing sets of ratings, all made by other researchers from biographical material, and combined them. The first was his own Intellectual Brilliance factor from 1986, built by having independent judges rate anonymised biographical profiles of 39 presidents on the Gough Adjective Check List, then factor analysing the reliable descriptors. The resulting factor loaded on items such as intelligent, wise, inventive, curious and insightful, and reached an internal consistency of .90. The second was Openness to Experience for 32 presidents, taken from expert ratings compiled by Rubenzer and Faschingbauer in 2004. The third was Cox's 1926 IQ figures, available for eight presidents only.

The three sets agree with each other more than their different origins would suggest. In Simonton's published correlation matrix, Intellectual Brilliance correlates .70 with the Cox IQ figures and .69 with Openness, and it had earlier correlated .80 with Thorndike's 1950 assessments of ten chief executives. Convergence across three unrelated methods is genuine evidence that all three track something shared. It is not evidence that any of them recovers a test score. Simonton follow up work, published open access in the Journal of Intelligence in 2018 and readable at the PubMed Central copy, restates the same construct and reaches the same conclusion about which measure matters. The underlying idea of a shared general factor is set out on the general factor these measures are meant to tap.

4 Only eight of the forty two numbers began as numbers

Before Simonton's statistical procedure ran, exactly eight presidents had an IQ figure attached to them, and all eight came from Cox in 1926. They are George Washington, John Adams, Thomas Jefferson, James Madison, John Quincy Adams, Andrew Jackson, Abraham Lincoln and Ulysses S. Grant. Every other president in the table received a number that did not exist until the algorithm produced it.

That claim is checkable rather than asserted. Simonton prints his original scores in bold and his reconstructed ones in ordinary type, and he separately publishes summary statistics for the original subset. Averaging the eight bolded rows in each of his four IQ columns reproduces his published means exactly: 127.5, 142.5, 140.6 and 151.9. Doing the same for all 42 rows reproduces the completed sample means of 121.0, 134.4, 136.0 and 146.8. The eight are identifiable, and there are eight of them.

The reconstruction itself is a standard missing data technique, the expectation maximisation algorithm, run in a statistics package. It uses the observed pattern across the six variables to compute the most likely value for each gap. Simonton reports that the data passed a test for being missing completely at random, and that the six measures were intercorrelated enough to support imputation. All of that is methodologically respectable. It is also, by construction, a way of writing down what the model expects given a president's ratings, not a way of finding out what he would have scored.

The clearest illustration sits in the table itself. Three pairs of presidents receive identical IQ figures to the decimal place across all four columns: William Henry Harrison and Rutherford B. Hayes, Zachary Taylor and Andrew Johnson, and James A. Garfield and Chester A. Arthur. They are identical because their inputs were identical or near identical, and the same inputs produce the same output. No measurement of two different people returns the same four numbers to one decimal place. A model does that routinely, and it is the signature of a model rather than a measurement.

5 There is no single Simonton IQ for any president

Every president in the study has four different IQ estimates, not one, and the popular lists silently publish the highest of the four. The four columns exist because Cox produced two figures for two life periods and then corrected each of them. The method section of the paper describes period I as birth to age 17 and period II as ages 18 to 26, with an uncorrected and a corrected value in each. That gives I-U, I-C, II-U and II-C.

The differences between the columns are systematic and large. Corrected values run higher than uncorrected ones by design, because the correction compensates for thin records. Second period values run higher than first period ones. Across the completed set of 42 the column means run 121.0, 134.4, 136.0 and 146.8, so the choice of column moves the average president by nearly 26 points. Every viral ranking uses the last column, the one that is both corrected and drawn from the later period, which is to say the highest available number for each man.

For an individual president the spread is wider still. James Madison sits at 120.0 in the first column and 160.0 in the last, a range of 40 points for the same person from the same study on the same page. John Adams runs 120.0 to 155.0. Andrew Jackson runs 110.0 to 145.0. If a single figure is quoted for any of them without saying which column it came from, the quotation has thrown away the study's own uncertainty before the reader ever sees it. The general problem of a score reported without its band is covered on why a standard deviation of 15 matters.

6 The full ranking, with every column and every source

Here are all 42 presidents with all four estimates, ordered by the last column, which is the ordering that circulating lists use. The final column marks whether the row came from Cox in 1926 or from the imputation. Reading down that column is the fastest way to see how much of the ranking is model output.

Two structural facts stand out once the origins are marked. The ceiling and the floor of the whole ranking are both Cox rows. John Quincy Adams tops it and Ulysses S. Grant anchors it, and in three of the four columns not one imputed president falls outside the range those eight men set in 1926. The only exceptions are James Monroe and Warren G. Harding, who dip below the eight in the first column alone.

The ordering is also column dependent. Re computing the ranking from the first column instead of the last moves George Washington from about twelfth to thirty third, the largest single shift in the table, and moves Andrew Jackson up seventeen places. The rank correlation between the two orderings, computed here from the study's published table, is about .87, which is high but plainly not one. A reader who wants the honest version of an ordering that changes with an arbitrary choice will find the same principle applied to score bands on the exact percentile for a score.

7 The uncertainty Simonton reports, in his own terms

The most commonly dropped part of this study is the part where its author explains how uncertain the numbers are. He describes the estimates as tentative in the paper's statement of aims. He notes that the reconstruction was heaviest for the IQ columns and lightest for Intellectual Brilliance, and concludes that the IQ figures must therefore be treated as the most approximate of the three measures. He observes that the corrected Cox scores may be overestimates. None of that survives the trip into a ranked list.

The quantitative version of the warning is the shrinkage of spread. Among the eight presidents Cox actually scored, the standard deviation of the last column is 13.9. Across the completed set of 42 it falls to 8.3, and the same compression appears in every column. Simonton names the cause: when most values are imputed, they regress toward the mean and variance drops. The practical reading is that the differences between imputed presidents in the middle of the table are smaller than the true differences would be, so ranking two adjacent names against each other is close to meaningless.

There is a further point that almost no secondary account mentions. In a footnote Simonton records that these are ratio IQs inherited from Cox, computed as mental age over chronological age, and therefore carry no preset standard deviation of 15 or 16 the way a modern deviation score does. He notes the sample standard deviation tends to run between 14 and 15. A 160 in this table is not a 160 on a modern scale, and treating it as one is a category error before any question of accuracy arises.

8 The only real test of the method, and it comes out mixed

Kennedy and Nixon are the only two presidents for whom both a documented school score and a historiometric estimate exist, which makes them the single available check on the method, and the check does not come out clean. The estimates rank Kennedy well above Nixon in every one of the four columns. The school records rank Nixon well above Kennedy. The order reverses.

The size of the discrepancy varies by column and by man. On the last column, Nixon's estimate of 142.9 lands within a tenth of a point of his documented 143, which looks like a triumph, while Kennedy's estimate of 159.8 sits nearly 41 points above his documented 119. On the first column the pattern inverts: Kennedy's 138.9 is about 20 points too high and Nixon's 118.9 is about 24 points too low. No single column fits both men, and no column preserves their relative order.

This is a limitation, not a refutation, and the distinction matters. Simonton never claimed to predict any individual's test result, and a method built to correlate with expert ratings of performance at the group level is not obliged to reproduce two adolescent group test scores taken decades apart on an instrument with a six point standard error. With a sample of two, nothing about accuracy can be concluded in either direction. What can be concluded is narrower and more useful: the estimates should not be read as predictions of what any president would have scored, because on the only two occasions where that reading can be checked, it fails once badly and reverses the pair.

9 The 2001 hoax, named as a hoax

In July 2001 a document claiming to report the IQs of twelve presidents began circulating by email, credited to the Lovenstein Institute of Scranton, Pennsylvania. No such institute has ever existed. The fact checking site Snopes published its assessment on 18 July 2001 and its record of the claim preserves the original text. The Museum of Hoaxes keeps a parallel account of how it spread, and traces the email's origin to a joke posted on a humour website.

The document is worth describing because its construction explains its success. It claimed a four month study, commissioned on 13 February 2001 and released on 9 July, rating twelve presidents from Franklin D. Roosevelt onward on scholarly achievement, unaided writing and clarity of speech, scored on something called the Swanson and Crain system. It gave a margin, claiming accuracy within five percentage points. It quoted two named researchers with elaborate titles. Later that year, people continued the joke by registering a web domain in the institute's name so that anyone checking would find a website.

Every one of those components is fabricated. There is no Lovenstein Institute, the named researchers do not exist, and no intelligence ranking system by that name is known to the field. The internal evidence is worse. Snopes notes that the top figure, 182, is exactly twice the bottom figure of 91, an arithmetic coincidence that real data does not produce. The scores ran cleanly along party lines with one exception, which is the shape of an argument rather than the shape of a distribution. And a study described as covering the past fifty years included a president who died in 1945.

10 How an unsourced number becomes common knowledge

The hoax is the best available worked example of a general process, because for once the entire chain is documented with dates. The email circulated in summer 2001. The Guardian ran it as a factual item on 19 July and the Southland Times in New Zealand did the same on 7 August. The Associated Press publicised the error on 12 August, the Guardian retracted on 14 August, and U.S. News and World Report described it as a hoax on 20 August. Then, on 26 August, the Doonesbury comic strip cited it anyway, and its site later acknowledged the mistake.

Read that sequence again. The correction was in circulation twelve days before one of the most widely read features in American newspapers repeated the claim. That is the mechanism in miniature: a correction does not overtake the thing it corrects, because the original was interesting and the correction is not.

Six features made this particular fabrication effective, and they generalise to almost every unsourced figure. It carried an authoritative sounding attribution that nobody phoned. It reported values to whole numbers, which reads as measurement. It stated a margin of error, which reads as methodology. It flattered a conclusion a large audience already held, so the check felt unnecessary. It arrived through a trusted intermediary rather than from the source. And as it aged, the institute's name was stripped in retelling, which made the claim harder to check rather than less present.

The reader's defence is short and it is the same four questions used everywhere else on this site. Which instrument, administered when, by whom, and against which norm group. Any figure that cannot answer all four is an assertion. That test disposes of the hoax immediately, and it also disposes of most of what circulates about living politicians, which is examined on the page about why political IQ claims run in both directions. The wider family of durable misconceptions is catalogued under the myths that survive longest, and the general problem of unverifiable public figures appears again with the same problem applied to Albert Einstein.

11 Why none of these figures sit on the modern scale

Even if every number on this page were perfectly accurate for its own time, none of them could be read against a score produced today. Three separate problems stack, and each one is enough on its own.

The first is the scale. Cox's figures, and therefore Simonton's, are ratio quotients: mental age divided by chronological age. A modern score is a deviation score, meaning it reports how far a person sits from the mean of a reference group in standard deviation units. The two are different quantities that share a name and a rough centre of 100. Simonton says explicitly in a footnote that the ratio scores have no preset standard deviation. Without a fixed dispersion, a figure of 160 does not correspond to a fixed rarity, so it cannot be converted into a percentile. Rarity on the modern scale is set out on the calculator that works out how rare a score is and the conversion itself on the percentile converter. The two metrics are set side by side in a case where an unverified childhood 167 is quoted against a documented adult WAIS-R Full Scale of 136, where the ratio arithmetic alone turns a single credited mental age of seventeen and a half into 167 at ten and a half years old and 109 at sixteen.

The second is the norm group. A score is a statement about a person relative to a specified sample, and the samples differ. The Otis tests Kennedy and Nixon sat were normed on schoolchildren of the 1920s and 1930s. Neither source identifies which form or which norm year, which means the reference population behind those two documented figures is unknown even though the figures themselves are documented.

The third is drift over time. Test performance rose across the twentieth century in most countries where it was tracked, at a rate the research literature generally puts near three points per decade, though estimates vary by test family, period and domain. The full account is on the secular rise in test scores. The direction of the correction for a score from around 1930 is therefore known and the magnitude is not, because extrapolating a per decade rate across ninety years compounds every assumption in it. That is why this page reports the historical figures as historical figures and does not restate them in current units. What a current unit actually means is covered on what counts as a good score, and how the scale itself came to be on the history of the test itself.

12 What the estimates do support, and what they never touch

Strip out everything the estimates cannot do and a real finding remains. Simonton's purpose was to correlate intellectual measures with rated presidential performance, using a composite of twelve historian and political scientist surveys that reached an internal consistency of .99. Against that criterion, Intellectual Brilliance correlated .56, while the four IQ columns correlated between .31 and .35 and Openness .34. In a regression that also included years in office, war years, assassination, scandal and war hero status, only Intellectual Brilliance remained a significant predictor.

Two readings follow. The rated trait built from biographical descriptors outperforms the reconstructed IQ figures at the job those figures are usually quoted for, which is a strong hint that the IQ column is the weaker instrument even inside its own study. And an association near .56 in a sample of 42 leaves most of the variation in presidential reputation unexplained, which is the ordinary situation whenever ability is used to predict an individual outcome. The same pattern at the population level is described on the evidence on ability and job performance and its limits on how weakly a score predicts an individual life.

One result deserves emphasis because it is the cleanest possible separation between the study and the hoax. Simonton tested whether the measures differed by party and reported that the null hypothesis of equal expected intelligence for Democratic and Republican presidents could not be rejected, whether the window began with Jackson or with Lincoln. A real analysis found no partisan pattern. A fabricated list was built out of one. Related evidence on political orientation and measured ability is summarised on what the research on political orientation shows.

Finally, the coverage. The study stops at George W. Bush, so no peer reviewed historiometric estimate exists for Barack Obama, Donald Trump or Joe Biden, and none of the three has a documented test score. Simonton addressed the gap himself in the 2018 paper, noting in a footnote that his methods could be extended to a former president once enough biographical material has accumulated, and that they cannot be fully applied to a sitting one. Reading a figure off the 2006 table for anyone outside its sample is therefore extrapolation rather than estimation. The circulating figures attached to Donald Trump's name are examined on the evidence page for Donald Trump, and the equivalent treatment for his predecessor is on the same treatment applied to Barack Obama.

13 How to read any claim about a president's intelligence

The practical question is not which president was smartest but what a given number is entitled to claim, and that can be settled with a short comparison. The table below sets the three kinds of figure discussed on this page against a modern supervised battery and against what ACIS reports about itself, using the same five questions for each.

Two of the ACIS entries need their qualifications stated rather than implied. The standard error of measurement of about 1.60 points for Full Scale IQ, alongside a composite omega of .9886 and a g loading of .958, comes from a technical analysis set of 2,750 complete records inside an adult reference frame of 3,243 English speaking records aged 16 to 90. That frame is self selected rather than census based, administration is unsupervised, and the frame is modelled rather than drawn as a representative sample. Those limits travel with the figures wherever they are quoted, and the full derivation sits in the ACIS technical manual. ACIS is not a clinical instrument and is not appropriate for diagnosis, hiring decisions, educational accommodations or high IQ society admission. The domain structure behind the six indices is described on the CHC domain structure and enumerated on the six primary cognitive domains.

The professional framework for all of this is the Standards for Educational and Psychological Testing, published jointly in 2014 by the American Educational Research Association, the American Psychological Association and the National Council on Measurement in Education, and available as the 2014 edition with an overview from the APA testing standards page. The Standards are explicit that validity attaches to an interpretation for a specified use rather than to a test in the abstract, that a score should be reported with its error, and that a test user is responsible for the population and conditions under which a score was obtained. Judged that way, a historiometric estimate is a legitimate research variable being misused when it is quoted as a personal score, an eighty year old school result is a historical record and not a current measurement, and a fabricated list fails at the first question. If you want a figure that answers all five columns for yourself rather than for a president, the routes are compared on the comparison of supervised and online testing, with the adult path on the adult testing route, the research and screening workspace on the workspace for researchers and educators, and the difference from an untested quiz on a free quiz set against a validated battery.

14 Frequently Asked Questions