Chess and IQ: a modest correlation, expert knowledge and what training changes

0 The short answer

Chess skill and cognitive ability are related, but an Elo rating is not an IQ score. A corrected meta-analysis found an average correlation of about 0.22, with results varying by age, level of play and selection of participants. Knowledge, practice and experience also matter, and studies of chess instruction do not establish that playing chess reliably raises general intelligence. There is no validated universal equation for converting your rating to IQ. This page separates performance correlations, expert memory research and training experiments so that a result from one design is not mistaken for evidence from another.

1 What a Rating Measures and What an IQ Test Measures

A chess rating summarizes results against opponents, whereas a normed IQ score compares cognitive test performance with an age reference group. FIDE's rating regulations effective from March 2024 calculate changes from actual versus expected game results, scaled by a development coefficient. That coefficient varies with rating, experience and age, and the regulations include additional conditions. The title regulations separately define routes to titles. Their ordinary grandmaster norm route includes a rating of at least 2500 and qualifying norms covering at least 27 games; direct title provisions are separate. These are performance criteria within chess, not cognitive test requirements. A normed battery also separates abilities that a rating blends together; fluid reasoning, for instance, is measured with novel problems whose rule can be discovered from the item itself, as the page on testing fluid intelligence describes.

An IQ on the common deviation scale describes performance relative to an age reference group, with mean 100 and standard deviation 15, as the page on how IQ scores are normed explains. The page on the standard deviation of 15 explains the unit. Elo has no fixed population mean or universal standard deviation corresponding to an IQ norm group. Its distribution depends on the player pool and rules. A rating measures performance in rated chess; it cannot by itself locate a person on a general cognitive scale.

The two numbers also differ in what they sample. An IQ test samples several abilities, and the page on what IQ measures sets out the six domains that ACIS reports: verbal comprehension, fluid reasoning, quantitative reasoning, visual spatial ability, working memory and processing speed. Chess samples one skill, built from a very large store of learned positions and the calculation that operates on them. The question this page answers is therefore not whether chess is intelligent, which is a question about the game, but how much a person's standing on the first scale predicts their standing on the second, which is a question about people. The page on fluid versus crystallized intelligence draws the distinction the answer turns on: reasoning with novel material against knowledge acquired over years.

2 The Meta-Analysis: Burgoyne and Colleagues, 2016

The original synthesis found a modest association, and the authors' 2018 correction changes both the headline estimate and some conclusions about particular abilities. The 2016 study assembled 19 studies with 1,779 participants. Its original average correlation of 0.24 is widely repeated. The corrigendum, included with the paper in the authors' university-hosted copy, corrects the treatment of dependent effects and gives an average of 0.22, with a 95 percent interval from 0.16 to 0.28.

The distinction between cognitive models and task content matters. The page on the CHC model explains the broad-ability framework. A grouping based on numerical content is not identical to a complete quantitative index from any particular current battery. Likewise, the average across broad abilities should not be casually renamed a pooled Full Scale IQ correlation. The authors analyzed Full Scale IQ separately, with a different result and limited evidence.

The correction is also a useful lesson in how to read research. Several measurements from the same participants are not independent new samples. An analytical procedure has to account for their shared information. Revising that procedure can change uncertainty and specific significance tests even when the general conclusion remains similar. Here, the association remains positive overall, but it is inaccurate to say that every broad ability has a statistically significant relationship under the corrected analysis.

A modest correlation supports a population statement about covariation. It does not divide chess performance into a fixed percentage caused by intelligence and a complementary percentage caused by practice. Squaring a correlation is a descriptive operation within a model, not a causal allocation of all influences. Measurement error, sample selection, experience, task coverage and other factors remain relevant. Those considerations are especially important when someone wants to apply a pooled estimate to one player.

Nor does the result supply a personal conversion between the two scales. A regression slope would require appropriate means, standard deviations, sampling assumptions and a validated prediction model for the target population. Combining the pooled correlation with a conventional number for Elo variability is not enough. It would also be incorrect to reverse a regression equation and treat the result as an IQ measured from a rating. The central finding is a relationship between constructs, not a replacement for assessment.

The article therefore uses the corrected result as its main estimate and keeps the questions of selection, individual prediction and training separate. This is more informative than asking whether chess and intelligence are either perfectly connected or entirely unrelated. The evidence can support a modest association while leaving a large amount of individual variation unexplained.

3 Why Correlations Differ Across Samples: Range Restriction and Selection

Selection and restricted score ranges can change an observed correlation, but they are possible explanations to investigate rather than a complete causal answer. The corrected meta-analysis estimates the fluid reasoning association at about 0.31 in youth samples and 0.04 in adult samples, and at 0.33 in unranked samples and 0.10 in ranked samples. The latter adult and ranked intervals include zero. These comparisons should not be interpreted as proof that general ability stops mattering at a particular birthday or rating.

Age, skill level and the way chess skill was measured overlap in the included studies. Adult participants were generally rated, whereas many young participants were unranked and evaluated with chess tasks. That makes it difficult to separate the influence of age from the influence of selection or the outcome measure. A difference between groups of studies does not automatically identify the mechanism producing it.

Restriction of range is one possible contributor. A sample consisting only of strong players leaves out much of the variation found among beginners and occasional players. Selecting on performance can also change relationships among experience, ability and other variables inside the selected group. However, range restriction does not make every correlation smaller under every form of selection. Its effect depends on what is selected, how variables relate and how much variation remains.

Bilalić, McLeod and Gobet discuss selection and practice in their study of young players. Their findings illustrate why a surprising association inside a small elite subgroup should not be read as a universal law. A negative correlation in such a subgroup does not establish that lower intelligence causes stronger play or that improving cognitive performance would harm a player's chess.

Expert knowledge is another plausible contributor to differences between novices and experienced players. Years of study can change the information a player recognizes immediately and the positions they need to calculate from first principles. This provides a reason to study learning and experience, but a cross-sectional association cannot establish exactly how the balance changes for every person over time. A well designed longitudinal study would be needed to test a particular developmental explanation.

The same caution applies to reliability. Error in measurement can attenuate correlations under appropriate assumptions, but correcting a coefficient is not automatically a route to the true causal effect. The reliability and validity guide distinguishes precision from the validity of an inference. A chess rating, a short cognitive task and an estimated IQ can each introduce different measurement limitations.

For a player, the practical message is that a coefficient from one sample is not a personal ability ceiling. A club population, a group of children beginning instruction and a set of grandmasters are different targets for prediction. The evidence should be described at the level it actually supports, including uncertainty about the mechanisms behind the observed differences.

4 Young Players: Frydman and Lynn, 1992, and Bilalić, McLeod and Gobet, 2007

Studies of young players show that cognitive performance and experience can both be relevant, but selected samples do not establish a universal requirement for chess success. Frydman and Lynn tested 33 tournament level Belgian players aged 8 to 13 with the French version of the Wechsler Intelligence Scale for Children and reported in the British Journal of Psychology in 1992 a mean full scale IQ of 121, a verbal IQ of 109 and a performance IQ of 129. The authors concluded that a high level of general intelligence and of spatial ability is necessary for a high standard of play. In the Burgoyne meta-analysis this sample produced the single largest correlation in the fluid reasoning model, 0.69, a result from a small tournament sample that should not be generalized to every young player.

Bilalić, McLeod and Gobet tested 57 children from four Oxfordshire school chess clubs, mean age 10.7, with four subtests of the WISC III, Vocabulary, Block Design, Symbol Search and Digit Span, combined into an estimated IQ, and measured practice with interviews, parent questionnaires and diaries kept for about half a year. The sample's mean estimated IQ was 121.6 with a standard deviation of 16.7. IQ correlated 0.55 with a 55 item chess test, but the logarithm of practice hours correlated 0.90 with the same test. In a hierarchical regression, gender and age explained about 43 percent of chess test variance, adding IQ raised it to 57 percent, and adding practice raised it to 86 percent, after which IQ remained significant but small. Among the subtests, Digit Span and Symbol Search predicted chess skill best, and Block Design, the spatial subtest, did not show a significant association with chess skill in that sample. That finding does not establish that spatial processing is irrelevant to chess.

The study's most cited result comes from its elite subsample of 23 boys, mean age 11.4, who competed at national and international level and held official ratings averaging 1603. They had higher IQs than the rest of the sample, 133 against 114, more experience and far more practice. Within the subsample, IQ correlated minus 0.51 with rating, virtually every intelligence subtest was negatively related to rating, and the model with age, IQ and practice explained 60 percent of the variance in rating, 30 points more than the model without practice. The authors explain the negative sign by a negative correlation between IQ and practice within the elite: the boys with lower scores had practiced more. The finding influenced the pooled Full Scale IQ analysis. It illustrates how selection and relationships among predictors can complicate interpretation, but it does not prove that ability stops predicting rank in every selected group. Practice was observed rather than experimentally assigned, so its larger association is not a causal allocation of performance.

Adult Tournament Players: Grabner, Stern and Neubauer, 2007, and the Null Studies

Individual adult studies report different results, including positive associations and estimates that are not statistically distinguishable from zero. Grabner, Stern and Neubauer screened 90 adult tournament players rated from 1311 to 2387 with an intelligence structure test and personality measures and recorded their chess experience, tournament activity and practice, reporting in Acta Psychologica in 2007 a clear cut moderate relationship between general, and in particular numerical, intelligence and playing strength. The strongest predictor of the attained level was chess experience, and experience, current tournament activity, intelligence and personality together accounted for about 55 percent of the variance in rating. The authors read the result as showing that expert chess play does not stand in isolation from superior mental abilities and that the differences cannot be reduced to differences in experience, a conclusion limited to the study's design and sample.

The null results are older and smaller. Bilalić and colleagues review them in the paper above: Djakow, Petrowski and Rudik tested eight grandmasters in 1927 and found no difference from non players on general intelligence and visuospatial memory; Doll and Mayr found no reliable correlation between skill and the abilities measured by the Berlin intelligence structure test in 27 expert players in 1987; Waters, Gobet and Leyden found virtually no association between skill and visual memory ability in 36 adult players in 2002. Unterrainer and colleagues compared 25 chess players with 25 non players on the Tower of London planning task and reported in the British Journal of Psychology in 2006 that the players planned better, especially on difficult problems, but took longer, and that the two groups did not differ on Raven's matrices or on verbal and visuospatial working memory.

Read together, the studies support caution about extending a single result to all players. The participants came from selected groups, used different measures and varied in experience. A nonsignificant estimate does not establish an exact absence of association, particularly in a small sample. The visual puzzles guide explains spatial assembly tasks and why familiarity with a chessboard is not itself a measurement of spatial ability.

5 Are Chess Players Smarter Than Non Players? Sala and Colleagues, 2017

A 2017 meta-analysis found higher average cognitive performance among the studied chess players than among their comparison groups, without establishing a universal IQ average for everyone who plays. Sala and colleagues reported an overall standardized difference of 0.49, with an interval from 0.26 to 0.72, in Intelligence. The university-hosted paper examines whether cognitive differences can occur in a domain without the academic admissions tests found in some professions. Sensitivity analyses varied, including a lower estimate after excluding a study that had not controlled for education.

That result concerns a collection of cognitive measures and selected comparison groups. Multiplying 0.49 by 15 gives 7.35, but this illustrative rescaling does not turn the study into a representative IQ survey. It does not establish that chess players everywhere average 107, because neither the baseline nor the sample warrants that conclusion. A standardized group difference should retain the context of the measured outcomes.

Two other figures need the same care. Five studies in the Burgoyne synthesis reported a weighted mean Full Scale IQ of 120.5, and the Belgian and English child samples described above averaged approximately 121. These are descriptions of the participants who were tested, not a census of people who play chess. Casual online players, school club members, tournament entrants and grandmasters are overlapping but different populations. Recruiting one group does not make it representative of all the others.

Group comparisons also leave the causal direction unresolved. People may select into chess, persist for different reasons, receive different opportunities or develop particular skills through experience. The absence of an academic entrance requirement does not remove every source of selection. To test whether instruction changes a particular outcome, researchers need an intervention design with suitable comparison conditions. The later sections examine that different question.

6 Kasparov and the Reported Der Spiegel Assessment of 1987

Der Spiegel reported two IQ results for Garry Kasparov in 1987, a historical magazine assessment that should not be treated as a current clinical record. Der Spiegel published the results in issue 52 of 1987, dated December 20, 1987 in the magazine's online archive, under the title Genieblitze und Blackouts, flashes of genius and blackouts. Kasparov, then 24 and world champion since his victory over Anatoly Karpov at the end of 1985, spent three days in a suite of the Hotel Aserbaidschan in Baku with a magazine team that had brought three larger intelligence tests, by Hans Jürgen Eysenck of England, John C. Raven of Scotland and Adolf Otto Jäger of Berlin, together with a chess test and several smaller tasks. The article states that he had read no test books and practiced no test items, and that the materials were translated into Russian down to the names in the word problems.

The two IQ figures are stated in one sentence: on the English test he stood at 135 and on the Scottish test at 123. The magazine adds that such differences are not unusual and that on the basis of these tests the only thing that can be said about his intelligence is that it lies considerably above average but nowhere near that of the geniuses of the century. The Raven test described in the article is the Advanced Progressive Matrices, 48 items of one type with 50 minutes allowed. The Eysenck test was compiled for the occasion and printed for readers to try. Neither test's norm table or scoring convention is reported, which is why the page on Garry Kasparov's IQ treats both numbers as historically sourced results rather than as modern index scores. The magazine also tested 30 German Bundesliga players and states flatly that the comparison showed no correlation between IQ and playing strength.

The profile behind the numbers is the more informative part. On Jäger's Berlin intelligence structure test, which separates numerical, verbal and figural content, Kasparov was far ahead of several hundred Berlin pupils and students and of the Bundesliga players on nearly every numerical task, and behind both groups on figural and pictorial thinking. Given a sheet of 24 ellipses and three minutes to turn them into as many different drawings as possible, he sat motionless for half the time and produced three; young chess players from the Hamburger SV club given the same task averaged seven. Shown game positions of 22 or 30 pieces for five seconds each, he reproduced 117 of the 120 pieces across five positions. That result is consistent with the expertise literature, but a single performance cannot establish how much came from learned patterns, general memory or other influences.

7 The Levitt Equation: Why a Rating Cannot Be Converted to IQ

Levitt presented his equation with reservations, and the sources reviewed here do not validate it as a personal rating-to-IQ conversion. The English player and author Jonathan Levitt presented the equation in his 1997 book Genius in Chess, and the chapter is reproduced on his own site: Elo is approximately ten times IQ plus 1,000, where the approximation sign means, in his words, given many years of intense effort, will tend to equal approximately. He introduces it with the sentence that the equation is subject to a number of reservations and should not be taken too seriously, and he lists the objections himself, including that the relation is unlikely to be linear and that the numbers 10 and 1,000 recommend themselves partly because they are simple to remember. He spells out the implications: an IQ of 100 would correspond to about 2000, a strong grandmaster at 2600 or above to an IQ above 160, and a world champion at about 2800 to an IQ of about 180.

Levitt's stated precondition involves years of intense effort. That alone makes it inappropriate to rearrange the formula and present an ordinary player's inferred IQ as a measurement. A comparison with Kasparov's historical reported scores illustrates the problem, but a single journalistic assessment cannot formally test a general prediction model. Proper validation would require clearly specified measures, an appropriate sample, uncertainty estimates and independent prediction checks. The correlational research above does not provide that validation for Levitt's equation.

The formula persists because people want to know what their rating says about their intelligence. A rating is useful evidence of chess performance in its rated setting. It does not establish a precise IQ, and a group correlation does not supply a high-confidence statement about the standing of every individual player. A normed cognitive score can be interpreted within its reference system; a chess rating cannot acquire that meaning merely by passing through a memorable arithmetic formula.

8 What Chess Skill Is Made Of: Chunks, Practice and Time

Expert chess performance draws on learned patterns, but memory experiments do not allocate all of a player's skill between practice and cognitive ability. Chase and Simon showed the mechanism in Cognitive Psychology in 1973 with three players ranging from master to novice. After viewing a position for five seconds, the stronger players reproduced more pieces per trial when the position came from a real game, and the advantage was much smaller for random arrangements; subsequent work found that some expertise effects could remain with random positions. The results support an account in which familiar groups of pieces allow experts to organize information efficiently, rather than requiring each piece to be held as an unrelated item. Gobet and Simon extended the work in Memory and Cognition in 1996, comparing a computer simulation with human recall of game, distorted and random positions, and concluded that the usual estimate of some 50,000 chunks stored by a master is supported, and that each chunk represents a specific pattern of pieces in specific locations, since mirror imaging a position impairs recall. Kasparov's reported recall performance is consistent with domain expertise, without establishing its precise cognitive mechanism.

Practice research addresses a related but distinct question. Hambrick, Oswald, Altmann, Meinz, Gobet and Campitelli reanalyzed the chess studies that measured deliberate practice and reported in Intelligence in 2014 that, after correction for measurement error, deliberate practice explained 34 percent of the reliable variance in chess performance and left 66 percent unexplained. The same paper records the variability behind the average: in one Argentinian sample, one player took 26 years of serious involvement to reach master level and another less than 2 years, and four players who estimated more than 10,000 hours of deliberate practice remained intermediate. The unexplained variance does not belong automatically to ability or to any other single cause. Practice estimates are imperfect, and observational associations cannot cleanly partition every influence on expertise.

Domain-specific knowledge helps explain why excellent memory for chess positions does not automatically demonstrate a larger general memory capacity. It does not prove that chess knowledge is useless outside chess or that every possible transfer effect is zero. The page on IQ and memory distinguishes working memory measures from expertise in familiar material. The sport comparison similarly cautions against converting specialized performance into a universal cognitive ranking. Whether training transfers is an empirical question requiring an appropriate outcome and comparison group.

9 Does Chess Instruction Raise IQ or School Results? Sala and Gobet, 2016

The 2016 instruction meta-analysis reported modest pooled benefits, while limitations in the underlying designs weakened a causal interpretation. Sala and Gobet searched for studies in which pupils received chess instruction and were compared with pupils who did not, and reported in Educational Research Review in 2016 on 24 studies with 40 effect sizes, 2,788 young people in the chess conditions and 2,433 in control groups. The overall effect was a Hedges g of 0.338 with a 95 percent interval from 0.242 to 0.435. Mathematics outcomes showed g = 0.382, cognitive outcomes 0.330 and reading 0.248. Duration of treatment was a significant moderator, and the authors report that studies with 25 or more hours of chess averaged 0.427 while those with fewer hours averaged 0.303, from which they suggested a possible role for intervention length. That observational moderator does not establish a guaranteed minimum dose or show that adding lessons causes a particular gain.

The same paper explains why the pooled estimate needs qualification. Published studies showed an effect of 0.540 and unpublished studies 0.230, a significant difference consistent with publication bias, although publication status can also track other differences between studies. No included study used what the authors call an ideal design, with pre and post tests, full random allocation and both a do nothing control group and an active control group. An active control group does a comparable novel activity, such as another game, so that the excitement of a new teacher and a new subject is subtracted from the chess effect; almost none of the reviewed studies had one. The authors also place g = 0.338 below the median effect of educational interventions in general, but comparisons between unrelated intervention literatures do not establish the best use of a particular school's teaching time.

The theoretical stake is far transfer, the generalization of a skill to a loosely related domain, and the authors frame the result as weak support for the common elements view: chess shares a few elements with arithmetic, such as quantitative relations and problem solving under rules, and almost nothing with reading, and the effect sizes fall in that order. The page on IQ and academic achievement sets out how strongly measured intelligence predicts school results, which is a separate relationship from the causal effect of chess instruction, and the page on whether you can improve your IQ places chess beside the interventions with better evidence.

The Better Designs: Active Controls and the English Trial

Several stronger designs found little or no evidence of transfer to their measured outcomes, which is narrower than proving every possible effect is exactly zero. Sala and Gobet ran the experiments their meta-analysis called for and reported them in Learning and Behavior in 2017. In the first, 233 third and fourth graders were assigned to 25 hours of chess, to checkers as an active control, or to a passive control, and tested on mathematical problem solving and on a metacognition questionnaire; the three groups did not differ significantly on either. In the second, with 52 children and the game of Go as the active control, the chess and passive groups slightly outperformed the Go group on problem solving, and the difference was not significant. The authors conclude that the effects of chess instruction, if any, are modest when rigorously tested and that such interventions should not replace the mathematics curriculum.

A large English trial provides a useful comparison. The Education Endowment Foundation funded a two armed randomized controlled trial of the Chess in Primary Schools programme, delivered by tutors from the charity Chess in Schools and Communities as 30 hours of lessons during the school day to Year 5 pupils in the 2013 to 2014 academic year, and published the evaluation report in July 2016. One hundred schools across 11 local authorities took part, with 4,009 pupils. The primary outcome was the Key Stage 2 mathematics score one year after the intervention ended; 3,865 pupils contributed to the main mathematics analysis. The effect size was 0.01 with a 95 percent interval from minus 0.15 to 0.16, which the report presents as zero months of additional progress. Reading, science and the prespecified pupil subgroups also showed no clear benefit; their estimates were not all numerically identical. Jerrim, Macmillan, Micklewright, Sawtell and Wiggins published the peer reviewed analysis in the Journal of Human Resources in 2018 and reported no evidence of an effect on mathematics, reading or science.

Sala and Gobet's 2017 review in Current Directions in Psychological Science sets the chess results beside music and working memory training and finds the same gradient in each: small to moderate effects overall that fall toward zero as the design improves. In the chess literature only one study had an active control group, with an effect of 0.10 standard deviations. In music the effect was 0.25 against passive controls and 0.03 against active controls; in working memory training it was 0.18 against passive and 0.05 against active, with the one robust finding being near transfer, gains on other memory tasks of about 0.45. The review argues that far transfer is difficult to demonstrate. Its comparisons do not establish that cognitive ability is the sole cause of participation or that a transfer effect is impossible under every intervention. The pages on IQ and video games and IQ and musical ability review the parallel literatures, which require their own outcome and design qualifications.

10 The Evidence by Design

Read by design rather than by headline, the chess literature sorts into two meta-analyses of correlations, three studies of children and adults, one magazine assessment, one speculative formula, one meta-analysis of instruction, two better controlled experiments and the cognitive science of what chess skill is, and the column that matters is what each measured.

The column to read first is the outcome measure. Correlational studies concern variation among people; instruction studies concern what happened after a defined learning activity; memory experiments investigate performance under particular task conditions. A magazine assessment and a proposed equation have different evidential status again. Those designs cannot be exchanged merely because their headlines all mention intelligence. A result can be informative for its original question while being insufficient for another.

11 Chess Demands and the ACIS Subtests That Resemble Them

ACIS measures cognitive performance, not chess skill, and this publisher has no validated ACIS-to-rating conversion to offer. We sell a paid self-administered online assessment, so this section is a commercial disclosure. The technical manual describes an adult reference frame of 3,243 English speaking records and a technical analysis set of 2,750 complete records. Those counts have different documentary roles and do not independently establish a sampling method.

Several ACIS tasks sample abilities discussed in the research. Matrix Reasoning uses patterns with a missing element. Figure Weights samples quantitative balancing. Visual Puzzles requires spatial assembly, and Digit Span samples storage and manipulation of sequences. The working memory guide describes the latter construct. Similar task demands do not establish that a correlation from a different instrument transfers unchanged to ACIS.

Processing speed is sampled by tasks such as Symbol Search and Coding. The 2018 correction to the chess meta-analysis matters here: its processing-speed estimate was no longer statistically significant. It would therefore be misleading to select an earlier significant coefficient and use it to promise that measuring speed explains an individual's chess strength. A broad cognitive profile can be informative on its own terms without functioning as a chess performance analysis.

Precision figures also belong to particular forms and editions. The manual publishes coefficients and SEMs while noting the September 2026 replacement of Spatial Navigation by Layer Rotation. Historical index evidence should retain that qualification. An article should not imply that a new form has been validated merely by applying an older coefficient to its current subtest list. Read the relevant documentation and report together.

ACIS is online and self-administered; it is not a clinical or diagnostic instrument. A score can support personal interpretation only within the conditions and limitations of the assessment. It does not supply a signed clinical evaluation, establish high IQ society eligibility, or determine what rating a player could attain. Readers seeking a formal decision should check the institution's accepted assessment route before choosing a test.

A chess player interested in the cognitive domains can examine a broad profile without trying to allocate each win or loss to an index. Game analysis, coaching and actual performance history answer questions about chess more directly. A cognitive score answers a different question about sampled tasks and a reference system. Neither needs to be inflated into a substitute for the other.

What conclusions are supported?

The corrected research supports a modest association between measured cognitive abilities and chess skill. Selected samples can differ from casual players, and associations within expert groups can be difficult to estimate precisely. Expertise research supports an important role for learned knowledge. Training trials address transfer to specified outcomes; several stronger designs did not find the broad benefits sometimes claimed for chess instruction.

Those statements do not establish that practice and ability occupy mutually exclusive percentages of a person's performance. They do not prove a universal average IQ of chess players, an IQ for an untested celebrity, or a fixed rating ceiling attached to an assessment result. They support an evidence-based account of related but distinct measurements.

12 What This Does Not Say About You

The research does not provide a validated personal IQ conversion from a chess rating or a guaranteed cognitive outcome from taking lessons. A modest pooled correlation cannot specify an individual's score with useful precision without additional modeling and evidence. It also does not justify multiplying a pooled correlation by an assumed rating spread to make a personal prediction. That calculation would combine unlike samples and assumptions, then conceal the uncertainty of the prediction.

The guide to IQ score interpretation explains why even a directly measured cognitive score must retain its reference group, interval and context. An inferred number from a game rating lacks that assessment process. A strong player can have a profile that differs from the average pattern in a selected research sample. A weaker player may be inexperienced, play under different time controls, study irregularly or simply pursue chess as a casual activity. The rating does not identify which explanation applies.

A profile can show relative cognitive strengths, but the studies above do not establish one required cognitive profile for strong chess. It would be inaccurate to say that verbal comprehension never matters, or that fluid reasoning, quantitative reasoning and working memory alone determine the result. The corrected meta-analysis distinguishes broad abilities from task content, and its estimates are group findings. The Full Scale IQ guide explains why a broad assessment should be read on its own terms.

Likewise, a rating built from many games should not automatically be declared more reliable than any cognitive assessment. Reliability depends on the measurement, time window, conditions and target construct. A mismatch between a rating and a test score is not proof that the cognitive result is erroneous. The two measurements are intended to describe different performances. A player can preserve both pieces of information without forcing them to agree numerically.

The practical advice follows the goal. To improve at chess, use game analysis, instruction and practice appropriate to the weaknesses you want to address. To investigate a cognitive question, choose an assessment suited to that question. To enjoy chess, no claim about an IQ increase is required. Competition, social contact, interest and the satisfaction of learning can justify playing even when a promised transfer effect is not established.

If a training provider claims that chess will raise IQ by a stated number of points, ask which outcome was measured, what comparison group was used, whether allocation was randomized, and whether the result was retained after the intervention. A result on trained chess puzzles is not equivalent to a gain on a broad cognitive battery. A positive average in a small study is not a guaranteed personal benefit. Those questions help distinguish a useful learning activity from a promise that goes beyond the evidence.

13 Sources Behind This Page

These references distinguish correlational research, instruction experiments, measurement rules and historical reporting. Correlations, effect sizes, sample sizes and confidence intervals retain the context of the papers and reports that supply them. The 2018 correction to the principal correlational meta-analysis takes precedence over its original estimates. An illustrative effect-size rescaling is not a personal IQ conversion. ACIS descriptions retain the technical manual's edition qualifications. Historical reports are identified as such and do not have the status of a current clinical assessment.

• Burgoyne A P, Sala G, Gobet F, Macnamara B N, Campitelli G and Hambrick D Z. The relationship between cognitive ability and chess skill: A comprehensive meta-analysis. Intelligence, 2016, volume 59, pages 72 to 83, with the authors' 2018 corrigendum, Intelligence 71, 92 to 96.
• Sala G, Burgoyne A P, Macnamara B N, Hambrick D Z, Campitelli G and Gobet F. Checking the "Academic Selection" argument. Chess players outperform non-chess players in cognitive skills related to intelligence: A meta-analysis. Intelligence, 2017, volume 61, pages 130 to 139.
• Bilalić M, McLeod P and Gobet F. Does chess need intelligence? A study with young chess players. Intelligence, 2007, volume 35, issue 5, pages 457 to 470.
• Grabner R H, Stern E and Neubauer A C. Individual differences in chess expertise: A psychometric investigation. Acta Psychologica, 2007, volume 124, issue 3, pages 398 to 420.
• Frydman M and Lynn R. The general intelligence and spatial abilities of gifted young Belgian chess players. British Journal of Psychology, 1992, volume 83, issue 2, pages 233 to 235.
• Unterrainer J M, Kaller C P, Halsband U and Rahm B. Planning abilities and chess: A comparison of chess and non-chess players on the Tower of London task. British Journal of Psychology, 2006, volume 97, issue 3, pages 299 to 311.
• Der Spiegel. Genieblitze und Blackouts: Der SPIEGEL testete Intelligenz, Gedächtnis und Schachkunst Garri Kasparows. Der Spiegel, issue 52 of 1987, archive date December 20, 1987.
• Levitt J. IQ and chess strength, an extract from Genius in Chess, Batsford, 1997, ISBN 0 7134 8049 1, reproduced on the author's site.
• FIDE. FIDE Rating Regulations effective from 1 March 2024, FIDE Handbook, section B.02; FIDE Title Regulations effective from 1 January 2024, FIDE Handbook, section B.01; and the FIDE rating profile of Magnus Carlsen, regulations checked September 24, 2026.
• Sala G and Gobet F. Do the benefits of chess instruction transfer to academic and cognitive skills? A meta-analysis. Educational Research Review, 2016, volume 18, pages 46 to 57.
• Sala G and Gobet F. Does far transfer exist? Negative evidence from chess, music, and working memory training. Current Directions in Psychological Science, 2017, volume 26, issue 6, pages 515 to 520.
• Sala G and Gobet F. Does chess instruction improve mathematical problem-solving ability? Two experimental studies with an active control group. Learning and Behavior, 2017, volume 45, issue 4, pages 414 to 421.
• Jerrim J, Macmillan L, Micklewright J, Sawtell M and Wiggins M. Chess in Schools: Evaluation Report and Executive Summary. Education Endowment Foundation, July 2016; and Does Teaching Children How to Play Cognitively Demanding Games Improve Their Educational Attainment? Journal of Human Resources, 2018, volume 53, issue 4, pages 993 to 1021.
• Hambrick D Z, Oswald F L, Altmann E M, Meinz E J, Gobet F and Campitelli G. Deliberate practice: Is that all it takes to become an expert? Intelligence, 2014, volume 45, pages 34 to 45.
• Gobet F and Simon H A. Recall of random and distorted chess positions: Implications for the theory of expertise. Memory and Cognition, 1996, volume 24, issue 4, pages 493 to 503.
• Chase W G and Simon H A. Perception in chess. Cognitive Psychology, 1973, volume 4, issue 1, pages 55 to 81.

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014) and the APA Ethical Principles of Psychologists and Code of Conduct, Assessment standards provide the interpretive framework: evidence must support the use, the reference group and limitations must be stated, and assessment results require appropriate explanation. Citing these standards does not imply their authors endorse ACIS or another product.

14 Frequently Asked Questions