The abstract reasoning test and IQ: one ability, not the whole score

0 The short answer

An abstract reasoning test is not an IQ test, and it is not a minor piece of one either: it samples fluid reasoning (Gf), one broad ability among several that a full battery combines into an IQ. In a peer reviewed analysis, Raven's Progressive Matrices, the best known test of this type, shared about 50 percent of its variance with general intelligence, and the same paper pointed to a four subtest Wechsler combination with a g validity coefficient of .93 and a 14 minute administration. Practice and coaching also raise these scores, and in 2025 the UCAT withdrew its Abstract Reasoning section for that reason and for lower predictive validity. Read any one score as a single domain, measured once, under one set of conditions.

1 What Is an Abstract Reasoning Test?

"Abstract reasoning test" is the name of a genre of timed, mostly figural puzzles used in hiring and admissions, and in psychometric terms it is a test of fluid reasoning, not a test of everything that intelligence includes. The format is familiar. The candidate sees shapes, patterns or symbols that follow a hidden rule, works out the rule, and chooses the figure or symbol that obeys it. Three layouts dominate: a grid of figures with one cell missing, a series in which the next element must be predicted, and a set in which one element breaks the rule that the others share. The page on the task families a fluid reasoning test uses describes them in detail, so this page does not repeat that material and reproduces no item from any real test.

The same task family travels under several names. A publisher that sells to employers may call it abstract reasoning, inductive reasoning or nonverbal reasoning, and the label says more about the market than about the construct. SHL, a large supplier of employer assessments, calls its version Inductive Reasoning. SHL's fact sheet, which this page read on October 20, 2026, says the assessment measures a candidate's ability to identify patterns within pieces of information and generate innovative solutions, and that the test measures the ability to infer rules, patterns or sequences and then use that information to predict outcomes. It says the task presents information as shapes and alphanumeric sequences.

The fact sheet also shows how the genre is run. SHL lists an allowed time of 18 minutes, a maximum of 15 questions, one sitting, an adaptive and activity based question format, and a design for unsupervised use. It states that the assessment is for job candidates applying to mid to senior level positions that require inductive reasoning ability, and that the suite can be deployed on any device type, including mobile phones. Those are facts about one product on one date. Other employers use other products with other lengths, and the invitation that a candidate receives is the authority for that candidate's test.

In the Cattell-Horn-Carroll framework, which the page on the CHC model explains, inductive reasoning is one of the narrow abilities that sit under the broad ability called fluid reasoning, abbreviated Gf. Gf is the capacity to solve novel problems when prior knowledge cannot supply the answer. That is the mapping this page uses throughout. It never treats "abstract reasoning" as a name for intelligence in general and never as the name of an ACIS index.

Readers arrive from three directions: an invitation to sit one for a job, an admissions exam that included one, or a free puzzle site that returned a number. The aim is the same in each case, to say what the score can support and what it cannot. ACIS does not sell preparation for these tests and does not screen candidates, and nothing here is advice on how to beat a particular employer's test.

2 Is an Abstract Reasoning Test an IQ Test?

No: an IQ is a composite built from several different tasks and ranked against age norms, so one abstract reasoning test yields a score on one ability, unless a publisher combines it with other tasks. The page on what IQ measures sets out the definition, and the page on how IQ scores are normed explains the ranking. A Full Scale IQ is a deviation score: the person is compared with people of the same age on a scale with a mean of 100 and a standard deviation of 15, and the number comes from several domains at once.

The clearest evidence of how the field treats a matrix task comes from the shortest Wechsler instrument. Pearson's brochure for the Wechsler Abbreviated Scale of Intelligence, Second Edition, which this page read on October 20, 2026, describes a four subtest form (Vocabulary, Similarities, Block Design and Matrix Reasoning) that takes 30 minutes and yields a four subtest Full Scale IQ, a Verbal Comprehension Index and a Perceptual Reasoning Index. It describes a two subtest form, Vocabulary and Matrix Reasoning, that gives a two subtest Full Scale IQ in about 15 minutes. The brochure lists the age range as 6 years 0 months to 90 years 11 months and describes the Perceptual Reasoning score as a measure of nonverbal fluid abilities and visuomotor and coordination skills, and the Verbal Comprehension score as a measure of crystallized abilities.

Two things follow. The first is that a matrix task is a real and respected component of a Wechsler IQ estimate. The second is that even the briefest Wechsler estimate does not rest on a matrix task alone. Pearson pairs it with a vocabulary task, which samples a different broad ability, crystallized knowledge (Gc). The brochure lists no one subtest Full Scale IQ.

What does the typical employer report show? The SHL fact sheet read for this page describes what the test measures and how it is run. It does not describe an IQ score as an output, and nothing in it presents the result as a measure of general intelligence. That is consistent with how the genre is used. The score ranks candidates on one skill, against a comparison group the employer or publisher chooses, for one decision.

So the answer to the search question has two parts. A stand alone abstract reasoning test is not an IQ test, because it samples one ability and is not ranked against a general population by age in the way an IQ battery is. And it is part of one, because Gf is among the abilities a full battery measures, and matrix tasks are among the tasks that measure it. The rest of this page asks how large that part is, what lies outside it, and how reliable a single reading of it can be.

3 How Much of General Intelligence Does One Abstract Reasoning Test Capture?

In the most direct peer reviewed estimate this page found, Raven's Progressive Matrices shared about 50 percent of its variance with general intelligence, which is a substantial share and is also far from the whole. Gignac (2015) set out to test the claim that Raven's is a pure indicator of g, which implies that it has a remarkably high association with g, shares no variance with a group level factor, and has virtually no test specific variance. Using a bifactor model across several large samples, he reported that Raven's shared approximately 50 percent of its variance with g, approximately 10 percent with a fluid intelligence group factor orthogonal to g, and had approximately 25 percent test specific reliable variance.

A bifactor model splits a test score into three kinds of variance. One part is shared with g, which is the general factor the page on the g factor describes. One part is shared with a narrower group factor, here fluid intelligence, that is independent of g. One part belongs to the test alone, such as the particular skills the format rewards. The three figures Gignac gives add to about 85 percent. The remainder, about 15 percent, is by subtraction variance that is not reliable, which is our arithmetic and not a figure he reports.

Gignac's reading was blunt. He wrote that the results suggest Raven's is not a particularly remarkable test with respect to g, and he discouraged researchers from including only Raven's when a valid estimate of g is the goal. As one alternative he suggested a four subtest combination from the Wechsler scales, with a g validity coefficient of .93 and a 14 minute administration time. That pairing of numbers is the sharpest answer to the question "how much of IQ is in the matrix test": about half of a matrix test is general intelligence, and a short multi domain Wechsler set captures nearly all of g in about the same time.

A caution about generalizing. Gignac analyzed Raven's. Employer products, free puzzle sites and the abstract section of an admissions test are different instruments with different items, lengths and time limits, and none of them inherits his percentages automatically. What the paper establishes is a limit on one famous matrix test, together with a method that any publisher could apply to its own.

A different literature tells a seemingly opposite story, and the two fit together. Valentin Kvist and Gustafsson (2008) fitted a second order factor model to a battery of 17 tests covering four broad abilities, in 2,358 Swedes, 620 European immigrants and 591 non European immigrants. According to the abstract of the working paper version, which this page read on October 20, 2026, the relationship between g and Gf was 0.83 in the total group and close to unity within each of the three groups. The difference between the two findings is the unit of analysis. Kvist and Gustafsson describe the fluid factor, which is estimated from many tests at once and so is free of the quirks of any one. Gignac describes one test. A factor can sit very close to g while a single indicator of it, with its own format and its own noise, sits well below.

One more number deserves a direct answer because it dominates the search results: the claim that abstract reasoning correlates 0.7 to 0.8 with IQ. Test preparation pages repeat it, and this page could not open a primary source for it. If the figure is meant as a loading on g, Gignac's 50 percent corresponds to about .71, the square root of .50, which is our arithmetic and sits at the bottom of that range. If it is meant as the correlation between a matrix test and a full battery score, the value depends on which battery and which sample, and this page does not state one. The page on Spearman's law of diminishing returns takes up a related contested claim, that the share of variance g explains is not the same at every ability level, which is another reason a single quoted correlation should be treated as an estimate for a particular sample.

4 What Does a Matrix Problem Ask the Mind to Do?

A matrix problem asks for two things at once, finding abstract relations and keeping several goals active while doing so, which is why an abstract reasoning score is partly a working memory score. The most detailed account comes from Carpenter, Just and Shell (1990), who analyzed the processing in the Raven Progressive Matrices Test using verbal protocols, eye fixation patterns and errors. They expressed their theory as a pair of computer simulation models that performed like the median or best college students in their sample.

Their findings separate what everyone does from what differs between people. The processing characteristic common to all subjects, they reported, was an incremental, reiterative strategy for encoding and inducing the regularities in each problem. The processes that distinguished among individuals were primarily the ability to induce abstract relations and the ability to dynamically manage a large set of problem solving goals in working memory. In plain terms, the problem is solved one regularity at a time, and the people who do better find the relations more readily and lose fewer of the partial results they must hold while they work.

That second ingredient explains why working memory keeps appearing in discussions of abstract reasoning. Ackerman, Beier and Boyle (2005) meta-analyzed 86 samples relating working memory to intelligence and reported an average correlation between true score estimates of working memory and g of .479, which they described as substantially less than unity. Their conclusion was that the two constructs are related and not identical. Other researchers have argued for a stronger link, and the page on fluid intelligence tests presents that dispute as a dispute. For this page the safe reading is narrow: a matrix score draws on more than one ability, and working memory is one of the contributors, not a rounding error.

Two other contributors deserve mention, and both are implications of the format and not findings this page can quantify. Figural items present their content visually, so some visual processing is involved, which is the territory of visual spatial ability (Gv). And any time limit gives processing speed (Gs) a chance to matter. The SHL fact sheet allows 18 minutes for a maximum of 15 questions, which is about 72 seconds a question by our arithmetic. How much of the final score that speed accounts for varies by test, and this page found no primary estimate for the products employers use. The page on processing speed explains the construct, and the page on working memory tests explains what a direct measure of working memory looks like.

The practical consequence is a plain one. If two people score differently on an abstract reasoning test, the gap can come from rule induction, from holding goals in mind, from visual processing, from speed under a deadline, or from familiarity with the format. A single score cannot say which. A profile that measures working memory, speed and spatial ability on their own tasks can begin to separate them, which is the case for looking at a domain next to its neighbors.

5 What an Abstract Reasoning Test Leaves Out

An abstract reasoning test samples one of the broad abilities in the CHC model directly and leaves the others to chance, so it cannot say anything about verbal knowledge, quantitative skill or working memory that was measured on its own. The CHC model lays out the broad abilities and the page on the six cognitive domains shows how ACIS groups its subtests under them. The table below sets the typical coverage of an abstract reasoning test beside the domain structure the ACIS report uses.

The ACIS column is taken from the structure ACIS publishes for its 20 subtests and six indices, read on October 20, 2026. The middle column is a description of the genre and not a count of any particular product.

The table carries a point that the search results rarely make. In the ACIS structure, fluid reasoning is not only figural. The Fluid Reasoning Index includes a verbal analogy format, a number pattern format and a logic puzzle format next to the matrix and balance scale formats, because Gf is a capacity, and a capacity can be sampled through several kinds of content. An employer's abstract reasoning test usually samples it through one kind, which is part of why it is a narrower measure.

Why does the omission matter? Because a person can be strong in one of these domains and not in another, and the difference has consequences that a single puzzle score hides. Someone with a high matrix score and a modest vocabulary may do well on novel problems and less well on tasks that depend on accumulated knowledge. Someone with the opposite pattern may excel at professions that reward knowledge and find unfamiliar puzzles tiring. The page on index score differences explains why spreads between indices are common and what they do and do not mean, and the page on verbal versus quantitative ability takes up the tilt between two other domains. The page on the Stroop test and IQ covers a well known task whose interference score is a different construct from reasoning, and it shows how a task name can suggest a link to IQ that the evidence supports only modestly.

None of this is an argument against abstract reasoning tests. It is an argument about the size of the claim. A score that samples one domain supports a claim about that domain. When a web page, a recruiter or a puzzle app turns it into a claim about "how smart" a person is, the claim has outrun the measurement.

6 Can You Practice for an Abstract Reasoning Test, and What Does a Higher Second Score Mean?

Scores on tests of cognitive ability rise when people retest or are coached, and the evidence shows that part of the rise is specific to the test, so a higher second score is not proof of a higher underlying ability. The best summary of how large the rise is comes from Hausknecht, Halpert, Di Paolo and Moriarty Gerrard (2007). They report that, according to previous studies, as many as 25 percent to 50 percent of applicants in organizational and educational settings are retested with measures of cognitive ability. Their meta-analysis of 50 studies covered 107 samples and 134,436 participants and found an adjusted overall effect size of .26. The effects were larger when practice was accompanied by test coaching and when identical forms were used.

The effect size is a standardized mean difference, so .26 means about a quarter of a standard deviation. On a scale with a standard deviation of 15, that would be roughly 4 points, which is our arithmetic and rests on reading the effect size in standard deviation units. The meta-analysis covers tests of cognitive ability in general and not abstract reasoning tests alone, so it sets an expected size for the family and does not predict any one person's change.

Why do scores rise? A study of matrix performance supplies a mechanism. Hayes, Petrov and Sederberg (2015) recorded eye movements while 35 university students solved two short forms of Raven's Advanced Progressive Matrices, Set II, in two sessions about a week apart. They reported that about one third of the variance in score gains could be attributed to changes in test taking strategy alone, and that after strategy was partialled out the residual score gains were no longer significant. Their reading was that gains on such tests can reflect skill acquisition, the learning of procedures that are then applied at the second sitting, and not necessarily a change in the ability the test was built to estimate. It is one small study with one test, so it is an illustration of a mechanism and not a measure of how much any particular person gains.

The professional standards address the question directly. The Standards for Educational and Psychological Testing (American Educational Research Association, American Psychological Association and National Council on Measurement in Education, 2014) include Standard 1.7. It says that if test performance, or a decision made from it, is claimed to be essentially unaffected by practice and coaching, then the propensity for performance to change with these forms of instruction should be documented. Its comment adds that materials for test takers should give practical guidance about the value of test preparation, including coaching. The standard places the burden on the publisher, which is the right place for it.

What follows for a reader? Three points. First, familiarity with a format is a legitimate advantage and employers know it, which is why some publishers use alternate forms and why the meta-analysis found larger effects with identical forms. Second, a score obtained after heavy practice on a specific format describes that performance, and a reader should not compare it with a first attempt as if the two measured the same thing. Third, none of this means ability cannot change. The page on whether you can improve your IQ reviews that question on its own terms, and the page on preparing for an IQ test covers what is sensible before sitting any assessment. This page does not sell preparation, and it offers no tactics for a particular employer's test.

7 Why Did the UCAT Drop Its Abstract Reasoning Section?

The UCAT Consortium withdrew Abstract Reasoning from the test in 2025 and gave two reasons, lower predictive validity than the other sections and high coachability, which are the same two tensions this page documents. The consortium's own page, UCAT 2025, read on October 20, 2026, states that Abstract Reasoning would be withdrawn from the test in 2025. It says the section has lower predictive validity than the other sections, and that the test also has high coachability. It adds that removing the section lets the organizers allocate additional time to the remaining subtests, and in Decision Making to increase the number of questions.

The page also describes the result. The test has three cognitive subtests plus the Situational Judgement Test. It lists Verbal Reasoning at 44 questions in 22 minutes, Decision Making at 35 questions in 37 minutes and Quantitative Reasoning at 36 questions in 26 minutes. Each cognitive subtest is scored from 300 to 900, which gives a cognitive total of 900 to 2700, and Situational Judgement is reported in bands from 1 to 4. The page says the total exam takes just under 2 hours, and that the scores for each subtest will remain broadly comparable to previous years.

Two limits on the evidence are worth stating. The page gives the reasons without publishing the data behind them, so the claims about predictive validity and coachability are the consortium's statements and cannot be checked from that page. And the decision concerns one use of one format. It says that, for the purpose the UCAT serves, the section was judged to add less than the others and to be easier to coach. It does not say that matrix style tasks fail to measure fluid reasoning, and it would be a mistake to read it that way. Gignac's finding that Raven's is a solid but unremarkable indicator of g and the consortium's decision are compatible: a task can measure a real ability and still be a poor use of limited test time in a high stakes selection setting.

A reader weighing an abstract reasoning result from any source can ask the question the consortium implicitly asked: how much of this score is the ability, and how much is exposure to the format?

The page on how IQ and the UCAT relate covers the exam in full, including how its sections compare with the abilities an IQ battery samples, and it is the right place to continue for a reader who came here from medical school admissions.

8 Does an Abstract Reasoning Test Predict Job Performance?

Cognitive ability tests do predict job performance, but the most recent re-estimates are lower than the older ones, this page found no estimate specific to abstract reasoning tests, and a score is only as useful as the validation behind the particular use. The standard for judging that validation comes from the testing profession. Standard 11.1 of the Standards for Educational and Psychological Testing says that before an employment or credentialing test is developed and implemented, a clear statement of the intended interpretations of test scores for specified uses should be made, and that the subsequent validation effort should be designed to determine how well this has been achieved for all relevant subgroups.

The best recent evidence on how well selection procedures predict job performance is a reassessment by Sackett and colleagues (2022). They argued that the range restriction corrections applied in earlier meta-analyses often produced substantial overcorrection, so that the validity of many selection procedures had been substantially overestimated. In their revised estimates, most of the same procedures that ranked high in prior summaries remained high in rank, with mean validity estimates reduced by .10 to .20 points, and structured interviews emerged as the top ranked procedure. They concluded that selection procedures remain useful but that the predictor and criterion relationships are considerably lower than previously thought. This page read the abstract and not the tables, so it does not quote a revised figure for cognitive ability tests, and it does not state one for abstract reasoning tests specifically.

Three consequences follow for the way abstract reasoning tests are used. First, the older, larger validity figures still in circulation were produced with corrections that the more recent analysis criticizes, so a figure quoted without a date or a method deserves a question. Second, the validity of a general cognitive test is not automatically the validity of a short figural test used for one role, and the Standards place the burden on the user to show that the interpretation for the intended use is supported. Third, the same reassessment paired its validity estimates with information about mean subgroup differences for each procedure, which treats validity and diversity as a tradeoff to be examined together and not separately.

What does a vendor say about prediction? SHL's fact sheet describes the Verify Interactive suite as offering an interactive, business relevant approach to predict workplace performance, and it names sample tasks such as troubleshooting, creative problem solving and innovative idea generation as work that may require inductive reasoning. The fact sheet read for this page gives no validity coefficient, no criterion and no sample, so the claim is a statement of purpose and not a result a reader can check. A fact sheet is a product description, and technical documentation, where it exists, is a separate document that an employer can ask for.

For a candidate, the practical reading is modest. The result will be one input among several, it measures one ability in a short time, and employers who rely on it alone are relying on less than the research supports. For an employer, the reading is stricter. Ask the supplier for criterion validity evidence for roles like yours, for subgroup data, for retest rules, and for the alternate forms policy, and compare any figure with the 2022 reassessment before building a cut score on it. The page on pre-employment cognitive tests covers the employer side in more detail, and the page on IQ and job performance reviews the broader research on general mental ability and work. ACIS is not designed for hiring decisions and says so plainly later on this page.

9 Is an Abstract Reasoning Test Fair, or Culture Free?

A test with no words is culture reduced, not culture free, and score differences between groups on tests of this family reflect a mix of causes that a single score cannot separate. The strongest piece of evidence for the first half of that sentence is historical. Flynn (1987) collected data from 14 nations and found IQ gains ranging from 5 to 25 points in a single generation. He reported that some of the largest gains occurred on culturally reduced tests and tests of fluid intelligence. He also reported that the Dutch data showed unknown environmental factors so potent that they accounted for 15 of the 20 points gained. Flynn drew a strong conclusion from this, that IQ tests measure a correlate with a weak causal link to intelligence and not intelligence itself, and this page reports that as his hypothesis without endorsing it. What the finding establishes more narrowly is that scores on fluid and culture reduced tests moved a great deal within a single generation, a span too short for changes in the gene pool to account for them, so performance on them depends in part on the environment in which people grew up. That is our inference from the finding, not a sentence from the paper. The page on the Flynn effect covers the later literature.

Group differences are a related but separate matter, and they need careful handling. Roth and colleagues (2001) reviewed ethnic group differences in cognitive ability in employment and educational settings. They reported that a difference of about one standard deviation summarizes Black and White differences for college application tests and for overall analyses of tests of g for job applicants in corporate settings. They also reported that the one standard deviation summary fails to capture many complexities: job complexity, within job versus across job design, applicant versus incumbent samples and the exact construct of interest were important moderators, and in many instances the standardized differences were smaller than one standard deviation. Hispanic and White differences were somewhat smaller than Black and White differences.

Those are statements about averages of groups in particular samples. They are not predictions about any person, because the distributions of the groups overlap heavily and an average difference says nothing about where one individual falls. They are also not statements about causes. The Flynn evidence above is one reason to expect that environment contributes, and the Sackett reassessment is one reason employers track subgroup differences next to validity. Valentin Kvist and Gustafsson (2008) found that the relationship between g and Gf was close to unity within each of their three groups, which says that in that battery fluid reasoning and general intelligence moved together within each group. It does not say that the groups scored alike, and it should not be stretched to do so.

There is a more ordinary fairness issue, which is the setting. SHL's fact sheet says its assessment can be taken unsupervised on any device, including a phone. A matrix puzzle on a small screen in a noisy room is a different task from the same puzzle on a desktop in a quiet one, and a score obtained under one condition is not interchangeable with a score obtained under the other. That observation is our inference from the facts the fact sheet states, not a finding the fact sheet makes. Language is a further layer, because the instructions of a wordless test are still written in a language. The page on IQ tests for non-native English speakers covers that, the page on bias in IQ tests covers how bias is investigated, and the pages on culture fair IQ tests and nonverbal IQ tests cover what the labels do and do not promise.

10 How Do the ACIS Fluid Reasoning Subtests Differ From a Hiring Style Abstract Test?

ACIS samples fluid reasoning with five subtests in five formats and reports it as one index beside five other domains, which answers a different question from a single timed puzzle. ACIS sells an online assessment, so this section is a disclosure as much as a description, and the reader should check it against the rest of the page. The ACIS facts below were read on October 20, 2026, and prices can change.

The Fluid Reasoning Index contains Matrix Reasoning, Figure Weights, Visual Number Series, Logic Grid and Complex Relations. According to the public subtest pages, Matrix Reasoning presents visual pattern problems, Figure Weights presents balance scale problems, Visual Number Series presents number pattern puzzles, Logic Grid presents deductive logic puzzles, and Complex Relations presents verbal analogies. The pages describe the task families and do not disclose items, and neither does this page.

The ACIS report gives a Full Scale IQ and six primary indices, which are the Verbal Comprehension Index, the Fluid Reasoning Index, the Quantitative Reasoning Index, the Visual Spatial Index, the Working Memory Index and the Processing Speed Index. Indices and the Full Scale IQ are on the standard scale with a mean of 100 and a standard deviation of 15, and subtest scores are scaled scores from 1 to 19 with a mean of 10 and a standard deviation of 3. The report gives percentiles and a 95 percent confidence interval. Adult norms cover ages 16 to 90. The technical manual holds the documentation of how the instrument was built and scored, which this page does not summarize.

There are three forms, each a one time payment with no subscription. Quick costs 15 dollars, covers 6 subtests in 3 domains and takes about 45 minutes. Optimized costs 30 dollars, covers 13 subtests in 5 domains and takes about 110 minutes. Full Scale costs 50 dollars, covers all 20 subtests in six domains and takes about 175 minutes. Breaks are allowed, a free trial needs no card, there is a 5 day quality guarantee, and a purchaser has 30 days to complete the assessment.

The limits matter as much as the features. ACIS is online and unsupervised, so every caution in the sections above about setting, device and effort applies to it. It is not a clinical or diagnostic instrument. It is not intended for hiring decisions, for school accommodations, or for admission to high IQ societies. It is in English only. Someone who needs a score that an employer, a school or a society will accept needs a different route, and the page on where to take an IQ test lays out the options.

What the ACIS structure offers, in relation to the question on this page, is a way to see fluid reasoning next to the other domains: knowledge, quantitative skill, spatial ability, working memory and speed. That comparison is the thing a stand alone puzzle cannot do.

11 How Should You Read an Abstract Reasoning Score You Already Have?

Before comparing an abstract reasoning score with anything, establish four things: what it was ranked against, how long you had, whether it was a first attempt, and whether the number is an IQ, a percentile or a band. Those four facts decide almost everything about what the score can support. The table below applies them to the situations readers most often describe.

The last two columns are our synthesis of the evidence above and are not quotations from any publisher.

A second layer concerns the number itself. An IQ, a percentile and a band are different things, and the page on the difference between an IQ score and a percentile explains why a percentile of 90 and an IQ of about 119 are statements about nearly the same position on different scales. A percentile calculator converts one to the other for a score on the standard scale, and a rarity calculator shows what a position means in a population. Those conversions only apply when the original score was on a documented scale with a documented norm group. A raw count of puzzles solved does not convert to anything.

The setting matters too. A deadline can change what a test measures. If the original sitting was timed tightly, speed contributed. If it was a repeat, the evidence in the practice section applies. If it was unsupervised on a phone, the conditions were not those of a standardized administration, which is the situation the APA guidance in the next section addresses. A logic puzzle that circulates online is a different thing again: the page on Einstein's riddle shows how a puzzle genre can acquire an "IQ test" reputation that no data supports.

If the number came from an employer, the useful step is to ask about retest rules and the format of any later stage. If it came from curiosity, look at a second domain, such as a verbal or working memory measure, before concluding anything about overall ability. If it matters for a clinical or educational decision, the route is a qualified psychologist who can administer a supervised battery. The page on the reliability and validity of IQ tests explains the evidence a score should come with.

12 What the Standards and the APA Guidance Say About Reading One Score

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014) and the APA assessment guidelines both treat a score as evidence for a specific interpretation in a specific setting, which makes a single abstract reasoning score thin support for any broader claim. The Standards are the joint document of the American Educational Research Association, the American Psychological Association and the National Council on Measurement in Education. Two of their standards have already appeared on this page. Standard 1.7 asks that the propensity of performance to change with practice and coaching be documented when a score is claimed to be essentially unaffected by them. Standard 11.1 asks for a clear statement of the intended interpretations of scores for specified uses, and for validation designed to show how well those interpretations hold for all relevant subgroups.

The American Psychological Association's Guidelines for Psychological Assessment and Evaluation, approved by its Council of Representatives in March 2020, add four points that bear on this topic. Guideline 4 notes that validity is not a unitary property of the test instrument. Guideline 6 asks psychologists to select assessment tools that show sufficient validity evidence for their uses and sufficient score reliability. Guideline 7 asks them to use multiple sources of relevant and reliable clinical information. Guideline 8 asks them to attend to the appropriate normative comparison, situational influences, effort and standardized administration. The guidelines are addressed to psychologists and are aspirational, not enforceable rules, so this page borrows their logic and does not claim they bind an employer or a puzzle site.

Applied to one abstract reasoning score, that logic produces four rules a reader can use. State the interpretation before looking at the number: ranking for a job is not the same as estimating intelligence. Ask what validity evidence exists for that specific interpretation. Use more than one source, which in practice means other domains, other occasions or other methods. And record the conditions: supervised or not, device, time limit, effort, and whether it was a first attempt.

What does the evidence support, stated narrowly? A matrix style abstract reasoning test samples fluid reasoning, and in the one peer reviewed analysis this page found for Raven's, about half of the variance in that test was shared with general intelligence. It also draws on working memory and may draw on speed and visual processing. Its scores rise with retesting and coaching, and part of that rise is tied to test taking strategy. Its use in selection is supported by validity evidence that has been revised downward, with no estimate specific to abstract reasoning tests in the sources read here. Its status as culture free is contradicted by the evidence of large environmental effects on fluid and culture reduced tests. What the evidence does not support is any reading of one such score as a person's IQ, a measure of how smart someone is, or a verdict on anyone's potential. The page on what IQ measures is the place to continue.

13 Sources Behind This Page

Every figure above traces to one of the following, and each is linked at the point where it is used. Statements marked as our arithmetic are calculations on published figures and are not attributed to the authors.

• American Educational Research Association, American Psychological Association and National Council on Measurement in Education. Standards for Educational and Psychological Testing. American Educational Research Association, 2014. Standards 1.7 and 11.1.
• American Psychological Association, APA Task Force on Psychological Assessment and Evaluation Guidelines. APA Guidelines for Psychological Assessment and Evaluation. American Psychological Association, approved March 2020.
• Ackerman PL, Beier ME, Boyle MO. Working Memory and Intelligence: The Same or Different Constructs? Psychological Bulletin, 2005, volume 131, issue 1, pages 30 to 60.
• Carpenter PA, Just MA, Shell P. What one intelligence test measures: A theoretical account of the processing in the Raven Progressive Matrices Test. Psychological Review, 1990, volume 97, issue 3, pages 404 to 431.
• Flynn JR. Massive IQ gains in 14 nations: What IQ tests really measure. Psychological Bulletin, 1987, volume 101, issue 2, pages 171 to 191.
• Gignac GE. Raven's is not a pure measure of general intelligence: Implications for g factor theory and the brief measurement of g. Intelligence, 2015, volume 52, pages 71 to 79.
• Hausknecht JP, Halpert JA, Di Paolo NT, Moriarty Gerrard MO. Retesting in selection: A meta-analysis of coaching and practice effects for tests of cognitive ability. Journal of Applied Psychology, 2007, volume 92, issue 2, pages 373 to 385.
• Hayes TR, Petrov AA, Sederberg PB. Do we really become smarter when our fluid-intelligence test scores improve? Intelligence, 2015, volume 48, pages 1 to 14.
• Roth PL, Bevier CA, Bobko P, Switzer FS, Tyler P. Ethnic group differences in cognitive ability in employment and educational settings: A meta-analysis. Personnel Psychology, 2001, volume 54, issue 2, pages 297 to 330.
• Sackett PR and colleagues. Revisiting meta-analytic estimates of validity in personnel selection: Addressing systematic overcorrection for restriction of range. Journal of Applied Psychology, 2022, volume 107, issue 11, pages 2040 to 2068.
• Valentin Kvist A, Gustafsson JE. The relation between fluid intelligence and the general factor as a function of cultural background: A test of Cattell's Investment theory. Intelligence, 2008, volume 36, issue 5, pages 422 to 436. Abstract read in the IFAU working paper version.
• UCAT Consortium. UCAT 2025. Page read October 20, 2026.
• SHL. Verify Interactive Inductive Reasoning, Assessment Fact Sheet. SHL and its affiliates, 2022. Page read October 20, 2026.
• Pearson. Wechsler Abbreviated Scale of Intelligence, Second Edition, brochure. Pearson Education, 2019. Page read October 20, 2026.

14 Frequently Asked Questions