The g factor test is a battery, not a single task

0 The short answer

There is no single g factor test, because g is the variance that many different tasks share, and the closest practical measure is a broad, normed battery that reports a Full Scale IQ. In an independent 2026 analysis of the WAIS-5 standardization sample (2,020 people), the best single subtest, Figure Weights, loaded .742 on the general factor, which the authors report as 55.1 percent of that subtest's variability, so even the strongest task is mostly something other than g. ACIS offers one concrete option: Quick (25 dollars, 6 subtests, about 45 minutes), Optimized (35 dollars, 13 subtests, about 110 minutes) or Full Scale (59 dollars, all 20 subtests, about 175 minutes), with prices read on November 4, 2026 and subject to change. The decisive limit is that ACIS is online and unsupervised, not a clinical instrument, and no online page can hand you a measured g, only the score of a battery.

1 Why Does No Single Task Count as the g Factor Test?

No single task is the g factor test, because g is not something a task contains: it is the variance that many different tasks share, and it can only be estimated by watching how scores on those tasks move together. Charles Spearman introduced the idea in his 1904 paper, "General Intelligence, Objectively Determined and Measured," which argued from correlations among laboratory discrimination tasks and school marks that every intellectual task draws on one common function and also on an element specific to itself. The chapters we read on November 4, 2026 in the open text of the paper rank school subjects such as Classics, French, English and Mathematics in what Spearman called a hierarchy, ordered by how strongly each shared the common function. The page on what the g factor is covers the concept itself, so this page stays with the measurement question that follows from it: if g is inferred and never observed, what does it take to measure it?

Every score a person earns on a task is a mixture. Modern factor analyses split a subtest score into a share that belongs to the general factor, a share that belongs to a broader group of abilities such as verbal comprehension or processing speed, a share that belongs to the task alone, and measurement error. The 2026 WAIS-5 analysis that anchors this page prints that split for every subtest, with columns for the general factor, each group factor, communality, uniqueness and specificity. A task that was nothing but g would show a loading of 1.0 and nothing else, and none of the 20 subtests in that analysis comes close. That is the first reason there is no g test in the sense of one task: every available task is partly something else.

The second reason is how the broad abilities are organized. John Carroll's 1993 book Human Cognitive Abilities, which Cambridge University Press describes as a reanalysis of more than 460 data sets from the factor analytic literature, presents a hierarchical, three-stratum theory of cognitive ability. The page on the CHC model explains the broad abilities that grew out of that tradition (Gf, Gc, Gq, Gv, Gwm and Gs), and the page on what IQ measures explains how g and those broad abilities fit together. A task belongs to one or more of those abilities, so choosing a task is choosing a window onto one part of the structure, not the top of it.

When someone sells a "g factor test," the phrase can mean three different things. It can mean a battery whose subtest scores are combined into one total that is mostly g, which is something a person can take. It can mean a single task marketed as highly g loaded, which is a claim that needs evidence. Or it can mean the factor analysis that researchers run on the scores of many people, which is a method and not a test. The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014), which we read in their open access edition, say in Standard 1.13 that when a score interpretation rests on premises about how the parts of a test relate, evidence about the internal structure should be provided. The comment to that standard gives the example of a claim that a test is essentially unidimensional, supported by a factor analysis showing that one major dimension is much larger than any other. In plain terms, a test that says it measures g should be able to show that analysis for its own scores.

Three neighbouring questions have their own pages and stay out of this one. The page on the Full Scale IQ explains the total score, the page on the General Ability Index explains the composite that sets working memory and speed aside, and the page on Spearman's law of diminishing returns covers the claim that g explains less of the differences among people at higher ability levels. What follows is narrower: how big real g loadings are, which tasks carry them, and what that means for anyone choosing a test.

2 What Is a g Loading, and How Large Are Real Ones?

A g loading is the correlation between a subtest and the general factor, and in a published 20 subtest analysis of the WAIS-5 the 20 loadings ran from .417 to .742, so the strongest task shares a little more than half of its variance with g and the weakest shares about one sixth. The analysis is Canivez, Watkins, McGill and Dombrowski (2026) in Assessment, an independent factor analysis of the WAIS-5 standardization sample (N = 2,020, ages 16 to 90, 13 age groups). The authors report that the sample was stratified on sex, education level, race and ethnicity, and geographic region, and that it closely matched 2022 U.S. census estimates on those variables. They used the 20 primary and secondary subtests and a Schmid and Leiman procedure, which converts a hierarchical model into loadings of each subtest on the general factor and on the group factors at the same time.

How to read the number: the table in the paper lists a standardized loading for each subtest and the variance it explains. A loading of .742 corresponds to the 55.1 percent of individual subtest variability that the authors report for Figure Weights, and the lowest loading, .417 for Naming Speed Quantity, corresponds to their 17.4 percent. The general factor accounted for 40.4 percent of the total variance across all subtests and 72.0 percent of the common variance, according to the same paper. Those two figures are the quantitative answer to "how much of a Wechsler battery is g": a large part of the shared variance, and a smaller part of everything that makes subtest scores differ.

The table below reproduces the general factor loadings the authors report for all 20 subtests, ordered from highest to lowest. The task family column is our own grouping for readability and is not the publisher's. We have not reproduced any item, only the names of the subtests, which the paper lists.

3 Which Kinds of Subtests Load Highest in a Published Battery?

In the WAIS-5 analysis the highest loadings came from reasoning tasks, quantitative and visual spatial as well as fluid, and the lowest came from the speeded and simple memory tasks, but the spread in loading among the top eleven subtests was less than one tenth, so no single family of tasks owns g. The top four are Figure Weights (.742), Arithmetic (.717), Visual Puzzles (.707) and Matrix Reasoning (.700). The next group, from .655 to .681, mixes Block Design, Spatial Addition, Vocabulary, Comprehension, Similarities, Set Relations and Information. The gap between first place and eleventh place is .742 minus .655, which is .087 by our arithmetic. A test that picked only the very top loaders would gain little over one that sampled reasoning, verbal and spatial tasks together, and it would lose the breadth that the next section shows matters more.

Reasoning tasks sit at the top because they ask the person to work out a rule or relation that has not been taught. The page on Matrix Reasoning describes that task as rule discovery in unfamiliar visual patterns and places it in fluid reasoning (Gf), the ability to solve problems without relying mainly on learned facts. Its .700 loading in the WAIS-5 data is high but does not set it apart: Figure Weights, which its page describes as discovering the quantitative rule that balances a visual arrangement, a task between Gf and Gq, loaded higher. The popular idea that matrices are "the" g test fits the data only loosely.

Verbal knowledge tasks land in the middle of the high group, which is worth stating because Gc, crystallized knowledge, is a learned ability. Vocabulary loaded .676, Comprehension .664, Similarities .663 and Information .655. The Vocabulary page describes the task as depth and precision of word knowledge and notes that it reflects long term reading, conversation and education. That a learned-knowledge task loads about as high as a reasoning task is not a contradiction: g is defined by what tasks share, not by whether a task looks like reasoning. The cost is that verbal scores also carry education and language exposure, which is why a verbal subtest on its own is a weaker g estimate than the same subtest inside a battery.

Memory and speed tasks show the widest range. Working memory subtests span from .428 for Running Digits to .653 for Letter-Number Sequencing, and the pattern inside that range is informative. The tasks that ask the person to reorder what they hold (Digits Backward at .615, Digit Sequencing at .632, Letter-Number Sequencing at .653) loaded higher than simple forward repetition (Digits Forward at .525). The Digit Span page describes the same split: immediate auditory-verbal storage in the simple versions and active manipulation in the demanding ones. The processing speed subtests loaded lower still, with Symbol Search at .578, Coding at .564 and Naming Speed Quantity at .417. In the paper's best fitting bifactor model, the authors note, most subtests drew larger portions of their variance from the general factor, and the exceptions were Running Digits, Coding and Naming Speed Quantity. The page on processing speed covers what the speed index measures, which is a separate question from how much of g a speed task carries.

4 Why Does a Broad Battery Estimate g Better Than Any One Subtest?

A broad battery estimates g better than any single subtest because each subtest carries task specific variance and error that average out across many different tasks, and the published studies of battery size and composition point the same way. The WAIS-5 analysis puts a number on how much is left over. The general and group factors together measured 56.1 percent of the total variance in WAIS-5 scores, which leaves 43.9 percent as unique variance, the combination of specific and error variance, according to Canivez and colleagues. A single subtest score therefore blends g, a group ability, a task specific skill and chance. Add a second and a third subtest from different families and the task specific parts do not line up, while the shared part does.

One study tested the idea directly by building batteries of different sizes and compositions from the same data. Floyd, Shands, Rafael, Bergeron and McGrew (2009), writing in Intelligence, placed the scores of a large sample of young adults into independent and overlapping batteries of different sizes and extracted general-factor loadings by three methods. In the abstract we read at the University of Memphis repository, test characteristics contributed the largest share of variance in the loadings, and the extraction method mattered only when principal components analysis was included. Psychometric sampling error, the error that comes from which tests happen to be in the battery, was sizeable in some analyses and grew as batteries got smaller. With principal components excluded and sampling error reduced, the loadings were highly dependable.

A second result shows what a narrow battery does to the estimate of g itself. Johnson, te Nijenhuis and Bouchard (2008) compared g factors from five test batteries given to 500 Dutch seamen. In the abstract as held at the University of Amsterdam repository, all the correlations between the g factors were at least .95 except those involving the Cattell Culture Fair Test, which the abstract describes as four matrix reasoning tasks, and the lowest g correlation was .77. In other words, four of the batteries produced nearly the same g, while the battery made of one task family tracked the others less closely. This is one study of men of similar age and a somewhat restricted ability range, so the exact figures should not be generalized, but the direction matches the case for breadth.

What counts as broad? The practical answer is a battery that samples the major abilities instead of one: verbal knowledge and reasoning, fluid and quantitative reasoning, visual spatial processing, working memory and processing speed. The page on the cognitive domains in IQ testing lists them, and the page on what separates an intelligence test from a quiz explains why norms matter alongside breadth. Length is a separate question: the page on whether a long IQ test pays off covers where extra hours stop adding precision.

5 Do Different Batteries and Different Populations Yield the Same g?

Batteries built from different tasks, and samples drawn from very different countries, tend to converge on a general factor, which is the strongest empirical argument that g is a stable feature of how cognitive performance is organized and not an artifact of one test. The Dutch seamen study just described calls itself a constructive replication of an earlier paper by Johnson, Bouchard, Krueger, McGue and Gottesman (2004) that used three batteries. The 2008 authors chose batteries that emphasized perceptual ability and dexterity over verbal tests, a different set of tasks from the usual academic mix, and still found the g factors highly correlated. They read the result as support for a general intelligence whose measurement does not depend on the specific abilities assessed.

The question of whether g belongs only to Western, schooled, industrial populations has a direct answer in Warne and Burningham (2019) in Psychological Bulletin. The authors collected correlation matrices and data files with cognitive variables from non-Western, nonindustrialized nations. Across 97 samples from 31 countries, totaling 52,340 individuals, a single factor emerged unambiguously from 71 samples (73.2 percent), and 23 of the remaining 26 samples (88.5 percent) produced a single second-order factor. The first factor explained an average of 45.9 percent of observed variance (standard deviation 12.9 percent), which the authors describe as similar to what is seen in Western samples. They concluded that g appears in many cultures and is likely a universal phenomenon in humans.

Those figures are not directly comparable with the WAIS-5 numbers earlier on this page. Warne and Burningham report the variance explained by the first extracted factor in an exploratory analysis, while the WAIS-5 analysis reports the variance attributed to the general factor after a Schmid and Leiman transformation. The shared message is qualitative: when many different cognitive tasks are given to many people, the scores correlate positively, and a general factor accounts for a large share of the shared variance. That regularity is called the positive manifold.

The regularity does not settle what g is. In a 2016 article in Psychological Inquiry, Kovacs and Conway call the positive manifold the most replicated result in the field of intelligence, and they note that identifying g with general intelligence is contradicted by a number of results. Their process overlap theory proposes instead that tests tap domain-general executive processes, identified mainly in working memory research, as well as more domain-specific processes, and that executive processes are required more often than specific ones. You do not need to choose between accounts to use a battery well. What the evidence supports is that a score built from many tasks tracks something stable across tests and samples, and that the explanation of that something is still debated.

6 Is the Full Scale IQ the Same as g?

The Full Scale IQ is not g, but in the independent WAIS-5 analysis it behaved mostly like an estimate of g, and the group indices added little beyond it, which is why the total score is the number to read first. Canivez and colleagues found that a four-factor structure with a dominant general factor was supported, similar to the WAIS-IV, and that the publisher's five-factor structure, which separates Visual Spatial and Fluid Reasoning, was not. In their Schmid and Leiman solution the general factor accounted for 40.4 percent of the total variance, while the four group factors added 2.5 percent (Perceptual Reasoning), 4.4 percent (Verbal Comprehension), 4.5 percent (Working Memory) and 4.2 percent (Processing Speed). The omega-hierarchical coefficient for the general factor was .877, which the authors call high and sufficient for interpreting a unit weighted composite, while the coefficients for the four group factors ran from .102 to .374, below the .50 minimum they apply.

On that basis the authors conclude that the WAIS-5 results replicate independent findings on the WISC-V and support primary, if not exclusive, interpretation of the Full Scale IQ as an estimate of psychometric g. Two cautions belong beside that conclusion. It is one independent team's reading of one battery, and the publisher's own manual presents a different structure, based on confirmatory analyses. And the conclusion concerns the interpretation of index scores, not the value of measuring several abilities: 43.9 percent of the variance in subtest scores was specific or error, so a profile can still show a real strength or gap that the total hides.

For a reader choosing a test, the practical reading is a division of labor. The total score is the best available single summary of general ability, and the page on the Full Scale IQ explains how it is built and read, while the page on taking a real Full Scale IQ test explains what a battery has to contain to deserve the name. The index scores describe where the person is stronger or weaker, and they carry more error. The Standards say in Standard 1.14 that when interpretation of subscores, score differences or profiles is suggested, the rationale and evidence for that interpretation should be provided, and Standard 2.4 asks for reliability data, including standard errors, when interpretation emphasizes differences between two scores. The page on index score differences shows how common sizable gaps are among people, which is the context a profile needs before anyone reads meaning into one.

A related point is what the total does not claim. A person can score high on g while having a weak area, and a person with an average total can have one strong domain. Neither fact weakens the case for measuring g; both are reasons to read the profile as well as the number. The page on Spearman's law addresses a separate claim, that the general factor explains less at high ability levels, which this page does not test.

7 How Do You Judge a Test Sold as a g Factor Test?

Judge a g factor test by five questions about its tasks, norms, structure evidence and scores, because the words g-loaded, general intelligence and pure g are marketing terms unless the publisher can show what stands behind them. The questions follow from the evidence above and from the Standards.

1. Does it use several different kinds of task? One format, such as matrices only, tracked the other batteries least closely in the Dutch seamen study, and small batteries gave less dependable loadings in the Floyd design.
2. Is the total compared with a defined reference group? An IQ is a position within a group, and the page on how IQ scores are normed explains why a number without a stated group means little.
3. Does the publisher show internal structure evidence for any claim about g? Standard 1.13 asks for it when a score interpretation depends on how the parts of a test relate.
4. Does it report precision for each score it asks you to interpret? Standard 2.3 requires reliability and precision estimates for each total score, subscore or combination of scores that is to be interpreted, and the page on reliability versus validity shows how to read them.
5. Does it stay inside its evidence? Standard 1.5 says that when a recommended interpretation is said to lead to a specific outcome, the basis for expecting that outcome should be presented.

Three kinds of product deserve a closer look against these checks. A free single task that returns a number meets question 1 poorly, while a free trial of a real battery is a different thing, and the page on free IQ tests covers the free route. Tests that advertise measurement at the very top of the range raise a further problem: the loading evidence on this page comes from a standardization sample of 2,020 people aged 16 to 90 that closely matched U.S. census estimates, and it says nothing about the extreme tail, so a claim about g at one in a million needs its own evidence. The pages on the Mega Society and high range IQ tests cover that route. And a self-rating is not a measurement at all: the page on the how smart am I question compares self-ratings with measured scores.

Employers use a different kind of test. Schmidt and Hunter (1998), reviewing 85 years of personnel selection research in Psychological Bulletin, present the validity of 19 selection procedures for predicting job and training performance and report that the combinations with the highest multivariate validity for job performance were general mental ability plus a work sample test (.63), plus an integrity test (.65) and plus a structured interview (.63). Their subject is prediction of job performance by tests of general mental ability, which is a different purpose from reporting a norm referenced IQ for personal insight. A practice test for a job screen is not an IQ test, and ACIS states plainly that it is not intended for hiring decisions.

8 Which Tests Actually Estimate g, and What Do They Cost?

Only a battery of several different tasks with adult norms, a stated reference group and reported score precision estimates g in a way worth acting on, and the options that do this range from an individually given WAIS-5 to a normed online battery. The table lists the realistic routes, with prices read on November 4, 2026 from the pages named in the text. Prices change, so check each page before paying.

The Pearson figures come from Pearson's page for the WAIS-5, which lists the age range as 16:0 to 90:11, a publication date of 2024, administration on paper or Q-interactive and Qualification Level C. The kit price is what a qualified buyer pays for materials and not what a client pays for an assessment, and the 45 and 60 minute times are administration times, not the whole appointment. The ICAR row rests on the ICAR project page, which describes a public domain assessment tool for research, says the sample test requires registering by email, and says the site is no longer updated and has moved. We treat it only as an example of free research material, not as a recommendation or a test of g for personal use.

The ACIS rows come from the published forms: one time payment, no subscription, breaks allowed, progress saved between subtests and 30 days to complete. The page on where to take an IQ test compares every route in person and online, and the page on IQ tests for adults covers testing at adult ages. Neither the psychologist route nor the online route is a substitute for the other when a clinical question is involved: ACIS is not a clinical instrument.

9 What Does the ACIS Battery Sample, Domain by Domain?

ACIS samples the abilities a broad g estimate draws on, with 20 subtests in six CHC domains reported as a Full Scale IQ and six primary indices, and this page gives no g loading for any ACIS subtest because the loadings above belong to the WAIS-5 sample. The documentation for the ACIS battery sits in the technical manual, and nothing on this page should be read as a statement about how an ACIS subtest loads. What can be said here is which abilities each domain samples, using the descriptions on the ACIS subtest pages and the manual.

The Verbal Comprehension Index (VCI) covers comprehension-knowledge, Gc, with Antonyms, Vocabulary, Information, Synonyms and Similarities. The Fluid Reasoning Index (FRI) covers Gf with Matrix Reasoning, Figure Weights, Visual Number Series, Logic Grid and Complex Relations. The Quantitative Reasoning Index (QRI), Gq, has Mathematical Achievement and Arithmetic. The Visual Spatial Index (VSI), Gv, has Visual Puzzles, Layer Rotation and Spatial Comprehension. The Working Memory Index (WMI), Gwm, has Digit Span, Alphanumeric Sequencing and Visual Sequence, and the Processing Speed Index (PSI), Gs, has Symbol Search and Coding. The page on the CHC model defines these abilities and the page on the six domains lists them.

Three subtest pages are worth reading before comparing any of this with the WAIS-5 table. The Matrix Reasoning page describes the task as inferring the missing element of a visual pattern from its underlying rule structure, a Gf task. The Vocabulary page describes it as depth and precision of word knowledge, a Gc task. The Digit Span page describes immediate auditory-verbal storage and, in harder versions, active manipulation of ordered material, a working memory task. Digit Span, Alphanumeric Sequencing and Arithmetic present their items as audio and need working sound, and the manual says Layer Rotation carries no words at all.

A comparison of structure, not of statistics, is still fair. The task families that loaded highest in the WAIS-5 analysis, reasoning, quantitative and visual spatial, each have a domain in the ACIS battery, and so do the verbal, working memory and speed families that loaded lower. The manual also lists composites built from predefined subtest sets when their subtests are available. For example, it describes the CPI as combining working memory and processing speed as cognitive proficiency, and the NVI as emphasizing nonverbal reasoning, visual spatial ability, working memory and speed with minimal reliance on verbal content. These are alternative ways to group the same subtests and are described by the manual, not tested on this page.

The forms differ in how many families they sample. Quick covers 3 domains with 6 subtests, Optimized covers 5 domains with 13, and Full Scale covers all six domains with 20. As a general principle drawn from the Floyd design, estimates from fewer and narrower tasks are less dependable, and that is the trade you accept for a shorter session. The free trial, which needs no card, lets a person try the format first, and finished trial subtests are kept if the person later unlocks a form that includes them.

10 How Do You Read a Score That Estimates g?

Read a battery score as a position with a margin: the Full Scale IQ on a scale with a mean of 100 and a standard deviation of 15, its percentile and its 95 percent confidence interval, and treat any single subtest as less precise than the total. Subtest scores on ACIS are scaled scores from 1 to 19 with a mean of 10 and a standard deviation of 3, while indices and the Full Scale IQ use the mean 100 and standard deviation 15 scale. The page on why the standard deviation is 15 explains that scale, and the page on the difference between an IQ score and a percentile explains why percentiles cannot be averaged.

Two calculators put a score into context. The IQ percentile calculator converts a score into the share of the reference group below it, and the IQ rarity calculator expresses it as one in some number of people. Both depend on the reference group, which is why the first question for any score is who it was compared with.

The confidence interval is the honest part of a score. The Standards define reliability and precision through how much scores vary across replications of a testing procedure, as the APA guidelines paraphrase them, so a 95 percent interval is a statement about that variation and not about the person's worth. Standard 2.3 adds an important caution in its comment: the form-to-form and day-to-day consistency of total scores may be acceptably high while subscores have unacceptably low reliability, depending on how they are defined and used, so users need reliability data for every score they plan to interpret. In plain terms, a Full Scale IQ rests on more evidence than any one index, and an index rests on more than any one subtest.

The g loading gives a second reason for caution. If a subtest has a loading of .70, then .70 squared, which is .49 by our arithmetic, is the share of its variance that the general factor accounts for, and the other half is group ability, task specific skill and error. Reading a single subtest as a verdict on general ability would attribute to g a share that belongs to everything else. Reading it as a clue about one ability, then looking for the same pattern in neighbouring subtests, is how profile interpretation is meant to work. The page on index score differences is the right place to see how often sizable gaps occur before treating one as meaningful.

11 What Do You Need for a Fair Unsupervised Session?

An unsupervised session gives you a valid picture only if you recreate the conditions a standardized test assumes: a quiet room, working sound, uninterrupted time on each subtest and no outside help. The APA's Guidelines for Psychological Assessment and Evaluation (2020) ask psychologists to be aware of the need for test selection, scoring and administration to reflect the appropriate normative comparison, situational influences, effort and standardized administration. Those guidelines are written for psychologists, but the principle applies to anyone taking an online battery: the norm group was compared under standard conditions, and your result is only comparable if yours resembles them.

For ACIS, the practical list is short. Use a device where sound works, because Digit Span, Alphanumeric Sequencing and Arithmetic present their items as audio. Choose a time when you can focus on each subtest from start to finish, and take your breaks between subtests, which ACIS allows. ACIS allows breaks, saves progress between subtests and gives 30 days to complete a form, so the full session can be split across days. The manual says verbal and knowledge tasks are affected by English proficiency and educational exposure, and ACIS is available in English only, so a score from a person still learning English is more informative about the language than about g.

Two further limits come from the format. An unsupervised online test cannot confirm who took it or under what conditions, and it cannot observe effort or distraction the way an examiner does. That is why the section that follows treats the result as an adult-normed estimate and not as a clinical finding.

12 Where ACIS Sits, and What a Buyer Should Do

ACIS is an online, unsupervised, English only battery with adult norms that reports a Full Scale IQ and six indices, it is not a clinical instrument, and it is one concrete option for an adult who wants a broad estimate of general ability rather than a single task. The ACIS limits are plain: it is not for clinical or diagnostic use, for hiring, for school accommodations or for admission to high IQ societies, and it cannot replace an individual assessment by a psychologist when a clinical question is involved. Its strengths are the ones this page has argued matter: several kinds of tasks, adult norms for ages 16 to 90, a Full Scale IQ with a 95 percent confidence interval, one time prices of 25, 35 and 59 dollars read on November 4, 2026, and a free trial with no card. The refund policy is stated as "Full refund within 5 days if the report does not deliver the features described at checkout or a technical issue prevents you from accessing your results."

The Standards for Educational and Psychological Testing (AERA, APA and NCME, 2014) set out what any such claim should be able to show. Standard 1.13 expects evidence about internal structure when an interpretation depends on how the parts of a test relate, Standard 1.14 expects a rationale and evidence for any interpretation of subscores, differences or profiles, and Standard 2.3 expects reliability and precision estimates for each score that is to be interpreted. The APA's 2020 guidelines add that psychologists select tools with sufficient validity evidence and score reliability (Guideline 6) and use multiple sources of information (Guideline 7), because a test result is only one piece of an assessment. A buyer can apply the same logic. Read the documentation for the battery you are considering, in the case of ACIS the technical manual, and decide for yourself whether the structure, precision and norms it reports fit the use you have in mind.

What to do next depends on the question. If you want to see what a broad battery reports before paying, start the free trial and open the example report, which show the format without a card. If you want the broadest estimate this page describes, the Full Scale form measures all 20 subtests in about 175 minutes for 59 dollars. If you are weighing several online options, the page on the best online IQ tests compares them. And if your question is clinical, educational or legal, a qualified psychologist giving an individual battery is the right route, and no online score, ACIS included, should replace it.

13 Sources Behind This Page

Every figure above is traceable to one of the following, and each is linked at the point where it is used. Two figures are our own arithmetic on published numbers and are marked as such in the text: the .087 gap between the first and eleventh loadings in the WAIS-5 table and the .49 that results from squaring a loading of .70. Web pages were opened on November 4, 2026, and prices on them can change.

• Spearman C. "General Intelligence," Objectively Determined and Measured. The American Journal of Psychology, 1904, volume 15, issue 2, starting on page 201. Open text read at Classics in the History of Psychology.
• Carroll JB. Human Cognitive Abilities. Cambridge University Press, 1993. Description read on the Cambridge Core page.
• Johnson W, te Nijenhuis J, Bouchard TJ. Still just 1 g: Consistent results from five test batteries. Intelligence, 2008, volume 36, issue 1, pages 81 to 95.
• Floyd RG, Shands EI, Rafael FA, Bergeron R, McGrew KS. The dependability of general-factor loadings: The effects of factor-extraction methods, test battery composition, test battery size, and their interactions. Intelligence, 2009, volume 37, issue 5, pages 453 to 465.
• Canivez GL, Watkins MW, McGill RJ, Dombrowski SC. Construct Validity of the WAIS-5: Complementary Exploratory and Confirmatory Factor Analyses of the 20 Primary and Secondary Subtests. Assessment, 2026, online first, pages 1 to 27.
• Warne RT, Burningham C. Spearman's g found in 31 non-Western nations: Strong evidence that g is a universal phenomenon. Psychological Bulletin, 2019, volume 145, issue 3, pages 237 to 272.
• Kovacs K, Conway ARA. Process Overlap Theory: A Unified Account of the General Factor of Intelligence. Psychological Inquiry, 2016, volume 27, issue 3, pages 151 to 177.
• Schmidt FL, Hunter JE. The validity and utility of selection methods in personnel psychology: Practical and theoretical implications of 85 years of research findings. Psychological Bulletin, 1998, volume 124, issue 2, pages 262 to 274.
• American Educational Research Association, American Psychological Association and National Council on Measurement in Education. Standards for Educational and Psychological Testing, 2014. Standards 1.5, 1.13, 1.14, 2.3 and 2.4 read in the open access edition.
• American Psychological Association, APA Task Force on Psychological Assessment and Evaluation Guidelines. APA Guidelines for Psychological Assessment and Evaluation, 2020. Guidelines 6, 7 and 8 read.
• Pearson. Wechsler Adult Intelligence Scale, Fifth Edition product page, 2024 edition, age range, administration time, qualification level and kit price read November 4, 2026.
• ICAR Project. The International Cognitive Ability Resource project page, read November 4, 2026.

14 Frequently Asked Questions