The General Ability Index, and what it leaves out

0 Quick Answer

The General Ability Index is a composite score built from the reasoning and knowledge subtests of a cognitive battery, excluding working memory and processing speed, and it is reported alongside the Full Scale score rather than instead of it. On the WAIS-IV it draws on six subtests. On ACIS it draws on 15. In every case the arithmetic is the same as any other composite: scaled scores are summed and the sum is converted onto a mean of 100 with a standard deviation of 15.

The reason it exists is that the two excluded domains behave differently from the rest. They correlate less with the general factor, they are the domains most affected by attention and by conditions that slow output, and on a battery with an uneven profile they can move the composite by several points. When a person's reasoning scores and their efficiency scores tell different stories, one number cannot tell both.

What the GAI is not is a corrected score. Pearson's own materials list the idea that it can routinely substitute for the Full Scale score as a myth, on the grounds that working memory and processing speed are integral aspects of general intelligence rather than noise sitting on top of it. Excluding them does not remove error. It changes the question the composite is answering, which is a different and much smaller claim than the one the number is usually asked to carry.

1 What the GAI Is Actually Made Of

The composition changes from one instrument to the next, so the phrase General Ability Index does not name a fixed set of tasks. What is fixed is the exclusion. No published version of the GAI includes a working memory subtest or a processing speed subtest, and every version is built from verbal comprehension plus some form of reasoning.

The differences beyond that are substantial and they matter when comparing scores across instruments. A six subtest adult GAI samples far less of the reasoning space than a fifteen subtest one, and a version that includes Arithmetic is drawing on a task most batteries assign to working memory.

Read the rows against each other and the pattern is clear. The Wechsler versions are compact by design, because the GAI has to be computable from subtests the examiner already administered. The ACIS version is broad because the Full Scale form runs 20 subtests anyway, so the reasoning and knowledge set left after removing the five efficiency tasks is still 15 indicators wide.

Breadth is not a free improvement. A six subtest composite takes an hour less to administer, and in a clinic where time is the binding constraint that is the whole argument. What breadth buys is precision, and the arithmetic behind that trade is the same for every composite: more indicators, smaller standard error. The general mechanics of turning subtest performance into a composite are set out on the page on how an IQ score is calculated.

2 Where the Composite Came From

The GAI was not invented as a marketing option. It emerged from a specific clinical complaint about the third generation Wechsler scales. When the WISC-III and WAIS-III added working memory and processing speed as separate indexes, practitioners noticed that the Full Scale score for some referred children now sat below every reasoning subtest that fed it.

Prifitera, Weiss and Saklofske proposed a general ability composite for the WISC-III in 1998, in the opening chapter of the WISC-III clinical use and interpretation volume. Tulsky, Saklofske, Wilkins and Weiss did the equivalent for the adult scale three years later, publishing the development of a general ability index for the WAIS-III in Psychological Assessment in 2001, in volume 13 at pages 566 to 571. Saklofske, Prifitera, Weiss, Rolfhus and Zhu then wrote the clinical interpretation chapter for the WISC-IV version in 2005, at pages 33 to 65 of the WISC-IV clinical use and interpretation volume from Academic Press.

Pearson published the operational tables the same year. Raiford, Weiss, Rolfhus and Coalson issued WISC-IV Technical Report #4 in January 2005, updated in December 2008, and it remains the clearest public statement of what the composite is for.

That footnote is worth holding onto. The composite is only as trustworthy as the conversion table behind it, and the table is only as trustworthy as the sample it was built on. This is the same dependency that governs every composite in every battery, and it is why the norm sample section of a technical manual matters more than the composite list. The general case is made on the page on how scores are normed.

3 The Publisher's Decision Rule for Reporting It

Pearson does not tell clinicians to report the GAI whenever it is higher. It gives four specific situations, and each one requires a difference that is both statistically significant and statistically unusual. Those are two separate tests and the second one is the one people skip.

Technical Report #4 lists the conditions under the heading of when to use the GAI. A significant and unusual discrepancy between the Verbal Comprehension index and the Working Memory index. A significant and unusual discrepancy between the Perceptual Reasoning index and the Processing Speed index. A significant and unusual discrepancy between Working Memory and Processing Speed. Or significant and unusual scatter among the subtests inside Working Memory, Processing Speed, or both.

The report then defines unusual by base rate rather than by significance. Statistical significance only says a difference is unlikely to be zero. It says nothing about how common the difference is, and in a large standardization sample many statistically significant differences are entirely ordinary. Pearson directs the reader to the base rate tables in the Administration and Scoring Manual, citing Sattler's 2001 criterion that differences occurring in fewer than 10 to 15 percent of the standardization sample should be judged unusual.

• Significance is the first gate. Technical Report #4 gives critical values for the Full Scale to GAI difference itself: 2.86 points at p of .15 and 3.89 points at p of .05, across all ages.
• Base rate is the second gate. A difference can clear the critical value and still occur in a quarter of the population, in which case it describes nothing distinctive about the person.
• Scatter inside an index counts too. Two subtests that disagree sharply within Working Memory can make that index uninterpretable as a unit, which is a reason to look elsewhere even when the index means agree.
• The Full Scale score does not go away. The same report states that working memory and processing speed must still be reported and interpreted even when the GAI is used as the ability estimate.

Pearson's public WAIS-IV sample score report shows what the arithmetic looks like on a real profile. The sample examinee has a Verbal Comprehension index of 145 and a Processing Speed index of 122, a 23 point spread that clears its critical value comfortably. His Full Scale IQ is 139 and his GAI is 139. The difference is zero, against a critical value of 3.61. A large index spread does not guarantee a Full Scale to GAI gap, because the gap depends on where the spread sits rather than on how big it is.

4 How Common a GAI to Full Scale Gap Actually Is

Pearson published the base rates, and they show that a modest gap is the normal condition rather than a finding. Technical Report #4 gives the cumulative percentages for the WISC-IV standardization sample of 2,200, in both directions, at every discrepancy size.

Read the top row first. Nearly half the standardization sample has a Full Scale score at least one point below their GAI, and another 43 percent have it at least one point above. A gap exists for almost everyone, because two composites built from overlapping but different subtest sets will rarely land on exactly the same number. The mean discrepancy in the direction of Full Scale below GAI was 4.4 points, with a standard deviation of 3.0 and a median of 4.0.

The interesting result is not in the total sample column. It is in the breakdown by ability level, which Pearson also publishes. Among children whose GAI fell at 120 or above, a Full Scale score five or more points below the GAI occurred in 29.6 percent of cases against 6.7 percent in the opposite direction, and a gap of ten or more points occurred in 9.4 percent against 0.4 percent. Among children whose GAI fell at 79 or below the pattern reverses: a Full Scale score five or more points below the GAI occurred in 13.5 percent of cases, against 31.0 percent the other way.

The practical consequence is that a gap has to be read against the base rate for that ability level, not against zero. A ten point gap at a GAI of 130 is common. A ten point gap at a GAI of 100 is not. The same score difference carries different information depending on where on the scale it occurs, which is the same nonlinearity that governs the relationship between a score and a percentile.

5 The Groups Where the Gap Is Systematic

In several clinical and gifted samples, a Full Scale score below the GAI is the majority pattern rather than a minority one. Technical Report #4 reports the rates group by group, and the specific figure worth knowing is that in every group listed the mean gap is between four and seven points.

Two things are true of this table at once. The direction is consistent and the magnitude is small. A mean gap of five or six points is real at the group level and is smaller than the width of a 95 percent confidence interval on most single indexes, which means it is not something to read off one person's report as a fact about them. In the gifted sample of 63, only 13.3 percent had a Full Scale score ten or more points below their GAI.

The group means from Pearson's Wechsler GAI symposium point the same way and are equally modest. In a WISC-IV ADHD sample of 89 children the mean GAI was 100.3 against a mean Full Scale of 97.6. In a WAIS-IV ADHD sample of 44 adults the means were 99.8 and 96.9. In a WISC-IV Asperger sample of 27 they were 104.7 and 99.2, and in a WAIS-IV Asperger sample of 40 they were 102.5 and 97.5. The differences between those published means run from under one point to about five.

That is worth saying plainly because the number that circulates is larger. The often repeated claim that the GAI runs five to ten points above the Full Scale score in ADHD is at the upper edge of what these published group means support, and the samples behind them are small. What the evidence supports is a consistent direction with a modest average size and wide individual variation. The clinical picture in attention conditions is developed on the ADHD and IQ page.

6 The GAI Is Not the Truer Number

Pearson's own training materials list the belief that the GAI can routinely replace the Full Scale score as a myth, and the correction is blunt: working memory and processing speed are integral aspects of general intelligence. The publisher that built the composite is the source arguing hardest against the popular reading of it.

The Wechsler GAI symposium sets out three myths. The first is that the GAI is the Full Scale score with working memory and processing speed removed. The correction is that the GAI still loads on both domains, just less heavily, because no measure of one cognitive domain is pure. The second is that the GAI can substitute for the Full Scale score as a matter of routine. The correction states that the GAI is not a substitute except in rare situations such as motor impairment or an invalid Full Scale score. The third is that significant differences between indexes invalidate the Full Scale score. The correction is that a Full Scale score is invalid only when valid subtests are missing.

The number attached to that third myth is the most useful figure on this page. In the WAIS-IV standardization sample, 73.5 percent of people have at least one index that differs significantly from the mean of their own indexes. An uneven profile is what almost three quarters of the population has. Treating unevenness as the trigger for switching composites would mean switching composites for most people who ever take the test.

Technical Report #4 makes the same point in the language of validity. It states that the GAI is not necessarily a more valid estimate of overall cognitive ability than the Full Scale score, that working memory and processing speed are vital to a comprehensive evaluation, and that excluding them can be misleading. It offers the concrete case: two children with the same GAI but different working memory and processing speed will perform differently in a classroom, and only the second pair of numbers tells you which is which.

One qualification belongs in the same breath, because Pearson also writes the opposite sentence in a narrower context. Technical Report #4 states on its second page that the GAI can be used as a substitute for the Full Scale score when determining eligibility for special education services and placement classification. The two statements are reconcilable: the GAI is an admissible substitute for a particular administrative calculation where a regulation demands an ability figure, and it is not a more valid description of the person. Reading the first sentence without the second is how the myth propagates.

7 What Happens When Somebody Tests Which One Predicts Better

The direct comparison has been run, on exactly the population where the GAI is recommended, and the Full Scale score won. This is the part of the literature that rarely appears in discussions of the composite, and it is the reason this page does not recommend reporting the GAI on its own.

Rowe, Kingsley and Thompson published the study in School Psychology Quarterly in 2010, in volume 25 at pages 119 to 128, under the title of the predictive ability of the General Ability Index versus the Full Scale IQ among gifted referrals. The sample was 88 children referred for gifted evaluation, every one of whom showed significant variability between indexes, which is precisely the condition under which Pearson's decision rule points to the GAI. Both composites significantly predicted reading and mathematics achievement. The Full Scale score explained more variance than the GAI did. In regression, working memory and verbal comprehension each predicted uniquely, while processing speed and perceptual reasoning added nothing beyond them. The authors attributed the difference to the inclusion of working memory in the Full Scale composite.

The mechanism is not mysterious. Working memory is heavily involved in reading and in arithmetic, so a composite that samples it carries information about school achievement that a composite excluding it does not. Removing a domain that predicts the outcome you care about will reduce prediction of that outcome. That is arithmetic rather than a surprise.

The broader literature on index level prediction points the same direction. Technical Report #6 cites Konold's 1999 analysis in the Journal of Psychoeducational Assessment, at volume 17 pages 24 to 35, in which the four WISC-III indexes taken together accounted for 61 percent of the variance in reading, 65 percent in mathematics, and 48 percent in writing. It also cites Mayes and Calhoun in School Psychology Quarterly in 2007, at volume 22 pages 234 to 249, reporting that the four WISC-IV indexes together accounted for 68 percent of basic reading, 70 percent of reading comprehension, 77 percent of numerical operations, and 58 percent of written expression. Both results are about using all four indexes, not about dropping two of them.

None of this makes the GAI useless. It makes the case for it a different case: the GAI describes reasoning and knowledge standing without efficiency mixed in, which is a legitimate thing to want to know and a reason to report it next to the Full Scale figure. It is not a case that the GAI is what the person's intelligence really is. The general question of what a composite predicts is covered on the academic achievement page.

8 The Cognitive Proficiency Index, the Other Half

Reporting a GAI without its complement discards the information the exclusion created. The Cognitive Proficiency Index is the composite built from exactly the subtests the GAI leaves out, and the pair together says more than either alone.

Weiss and Gabel set out the rationale in WISC-IV Technical Report #6 in 2008. The CPI represents functions whose common element is the efficiency with which certain kinds of cognitive information are processed, and the argument, attributed there to Weiss, Saklofske, Prifitera and Holdnack in 2006, is that efficient processing facilitates fluid reasoning and the acquisition of new material by freeing cognitive resources for higher order tasks. On the WISC-IV the CPI is built from Digit Span, Letter-Number Sequencing, Coding, and Symbol Search. On the WAIS-IV it uses Digit Span, Arithmetic, Coding, and Symbol Search.

The discrepancy between the two composites has been studied as a marker, and Technical Report #6 is careful about what it can support. In a non clinical sample, 9 percent of people show a CPI below their GAI without any ability to achievement discrepancy. Among 516 non clinical cases tested on the WISC-IV with the WIAT-II, 21 percent had a large ability to achievement difference with no GAI to CPI difference at all. Pearson's conclusion is that the discrepancy on its own cannot serve as a diagnostic marker.

Where the pattern does carry signal is in combination. Only 2 percent of children in the normal sample show both a large CPI below GAI gap and a large achievement below GAI gap, while in the reading and written expression disorder samples that combination reaches 45 to 50 percent. Even there the classification statistics are modest: with a cut of five points or more, sensitivity and specificity for reading and writing learning disability were 66 and 63 percent respectively. For closed head injury the figures were 65 and 61 percent at a cut of four points, and for Asperger disorder 68 and 63 percent at a cut of eleven.

Sensitivity in the sixties is better than chance and nowhere near a test. A marker that misses a third of cases and falsely flags close to four in ten non cases is a reason to look further, not a basis for a conclusion. That is the honest description of what a GAI to CPI split contributes, and it is why the split belongs in a report alongside everything else rather than in place of it. The efficiency side of the profile is covered in more depth on the processing speed page and the memory page.

9 Gifted Identification and the Twice-Exceptional Case

The National Association for Gifted Children does not recommend the GAI. It recommends that no single score be required, and its list of acceptable scores includes the Full Scale IQ. That distinction gets lost constantly, and it changes what the position statement can be cited for.

The 2018 position statement on use of the WISC-V for gifted and twice exceptional identification, approved in August of that year, sets out three problems with requiring a Full Scale score. Large discrepancies between indexes can leave the composite without a unitary interpretation. An overemphasis on processing skills, and on processing speed in particular, confounds identification. And structural and administration changes between editions can lower scores. The statement says directly that weaknesses in areas less relevant to advanced academic programming, such as slowness on timed paper and pencil tasks, may lower a Full Scale score below identification cutoffs.

The remedy it proposes is plural rather than substitutive. The statement lists six scores any of which should be accepted for identification: the Verbal Expanded Crystallized Index, the Nonverbal Index, the Expanded Fluid Index, the General Ability Index, the Expanded General Ability Index, and the Full Scale IQ itself. It adds a condition that is easy to miss and psychometrically serious: the score qualifies if it falls within the confidence interval of the score required for admission. The evidence base named in the document is WISC-V data from 390 gifted children across seven United States sites.

• The argument is about cutoffs, not about truth. A rule that admits on any of six scores admits more children than a rule that admits on one, whichever score is most valid.
• The confidence interval clause does the heavy lifting. Accepting a score whose interval reaches the cutoff is an acknowledgment that a point estimate should not decide an admission.
• The Full Scale score stays on the list. The statement is not an argument that the composite is wrong. It is an argument against a single gate.
• Twice-exceptional identification is the sharpest case. A child whose reasoning sits two standard deviations above the mean and whose output speed sits at the mean can miss both the gifted cutoff and the disability cutoff at once.

For adults the same logic applies with none of the institutional machinery, since there is no admission committee and no cutoff to clear. What an adult usually wants from a general ability estimate is a read on reasoning that is not diluted by how fast they work under a clock, which is a reasonable thing to want and a poor thing to convert into a claim about their real intelligence. The adult case is set out on the gifted testing page for adults, and the profile pattern itself on the twice-exceptional page.

10 The Expanded GAI and What Adding Subtests Buys

Pearson widened the composite in 2019, and the technical report attached to it is the cleanest published demonstration of what breadth does to precision. The Expanded General Ability Index runs eight subtests where the standard WISC-V GAI runs five.

Raiford, Silverman, Gilman and Courville published WISC-V Technical Report #5 in July 2019. The expanded composite draws Similarities, Vocabulary, Information and Comprehension from verbal comprehension, Block Design from visual spatial, and Matrix Reasoning, Figure Weights and Arithmetic from fluid reasoning. Working memory and processing speed remain excluded, so the expansion widens the sampling of reasoning and knowledge rather than reintroducing efficiency.

The published reliability is .97 overall with a standard error of measurement of 2.62 IQ points, and by age band the standard error runs from 2.14 to 3.02. The report puts those figures next to the existing composites from the WISC-V technical manual: the Full Scale IQ at .96, the Nonverbal Index at .95, and the standard GAI at .96. The expanded composite is marginally the most reliable of the four.

The report is explicit about status. The Expanded GAI does not replace any existing WISC-V composite score. It is an additional option for situations where a broader estimate of higher order reasoning is wanted without working memory and processing speed in it. That sentence is the same disclaimer the original GAI carries, written a second time about its own successor.

Read the three figures as one sequence. A six subtest adult GAI carries a standard error of about 2.6 points, an eight subtest child version carries about the same, and a fifteen subtest version carries about 1.6. Adding indicators narrows the interval and it does so with diminishing returns, which is the ordinary behavior of a composite. What no number of additional reasoning subtests can do is put back the information that excluding working memory and processing speed removed.

11 How ACIS Builds and Reports the GAI

ACIS computes the GAI from the 15 non speeded reasoning and knowledge subtests and reports it alongside the Full Scale score and, on the complete form, alongside a Cognitive Proficiency Index built from the five it left out. The composite is defined before any result is seen, from a fixed subtest set, which is the same discipline a published manual applies.

The published figures are these. The GAI has a composite omega of .9885, a hierarchical omega of .9079, a standard error of measurement of 1.61 IQ points, and a general factor loading of .954. The Full Scale composite across all 20 subtests has a composite omega of .9886, a hierarchical omega of .9164, a standard error of about 1.60, and a g loading of .958. The two composites are close on every coefficient because 15 of the 20 subtests are shared, which is exactly why they usually land within a few points of each other.

The reason they can diverge sits in the domain level numbers. Among the six primary ACIS indexes, the published g loadings run .922 for Fluid Reasoning, .906 for Visual Spatial, .882 for Quantitative Reasoning, and .864 for Verbal Comprehension, against .788 for Working Memory and .648 for Processing Speed. The two excluded domains are the two least saturated by the general factor. Their intercorrelation with the rest of the profile is correspondingly loose: Working Memory correlates .475 with Processing Speed and Verbal Comprehension correlates .577 with Processing Speed, against .801 between Verbal Comprehension and Fluid Reasoning.

The ACIS CPI is built from Digit Span, Alphanumeric Sequencing, Visual Sequence, Coding, and Symbol Search. It is a five indicator composite rather than the four indicator Wechsler version, because the ACIS working memory index includes a visual sequence task alongside the two auditory ones. Its standard error of 3.82 is more than twice the GAI's, so a GAI to CPI gap has to be substantially larger before it means anything, and the report should be read with that asymmetry in mind.

Availability follows from composition. The GAI needs verbal, fluid, visual spatial, and quantitative indicators, so the Quick form at 15 dollars and six subtests cannot produce it. The Optimized form at 30 dollars and 13 subtests can, which makes it the first tier that reports a general ability estimate. The Full Scale form at 50 dollars and 20 subtests reports the GAI, the Full Scale IQ, the CPI, and every other composite. Which subtests feed which composite is documented in the technical manual, and the domain structure on the cognitive domains page.

12 How to Read a GAI and a Full Scale Score Together

The two numbers answer different questions, and the useful move is to ask which question is in front of you rather than to pick a winner. Everything above supports that one operating rule and very little beyond it.

The Full Scale score answers what this person's overall cognitive standing is, sampling every domain the battery measures, including how efficiently they hold information and how quickly they produce output. That is the number to use when the question is broad standing, when the score will be compared to a published distribution, or when an outcome that depends on efficiency is at stake. The GAI answers what this person's reasoning and knowledge standing is when efficiency is set aside. That is the number to use when the question is specifically about reasoning capacity, or when a documented constraint makes the efficiency domains a poor sample of what the person can do.

• Report both, always. Technical Report #4 requires that working memory and processing speed still be interpreted even when the GAI is used as the ability estimate.
• Check the gap against the base rate for that ability level. A five point gap at a GAI of 130 occurred in 29.6 percent of the high ability band of the WISC-IV sample. The same gap in the middle of the distribution is less ordinary.
• Compare the intervals, not the points. If the two confidence intervals overlap heavily, the difference between the composites is not something to build an interpretation on.
• Look at the CPI before concluding anything. A high GAI with a mid range CPI and a high GAI with a low CPI describe different working situations and the GAI alone cannot separate them.
• Do not read a gap as a diagnosis. Pearson's own classification statistics put sensitivity for learning disability in the sixties at the recommended cut, which is a signal to investigate rather than a finding.

One boundary belongs here rather than in a footer. ACIS is an unsupervised online assessment. It is not a clinical instrument, it is not diagnostic, and it is not appropriate for hiring decisions, accommodation requests, educational placement, or high IQ society admission. A GAI to CPI split on an ACIS report is a description of a profile, and any use of it that requires a diagnosis needs a proctored administration by a qualified examiner. What that route involves is on the professional versus online comparison.

What an unsupervised battery can honestly offer is the shape. A profile with a 15 subtest reasoning and knowledge composite, a five subtest efficiency composite, six domain indexes and 20 subtest scores tells a person which parts of their cognition run ahead of the rest, with a published standard error attached to each figure. That is a more useful object than any single number, and it is the reason the composite exists at all. The reasoning side of the profile in particular is developed on the fluid intelligence page, and the interpretation of the whole profile on the page on what scores mean.

13 Sources Behind This Page

Every figure above comes from a document that can be opened, and the places where the evidence runs out are named in the sentences that use it. The Pearson technical reports are free public PDFs, which is unusual for materials of this kind and is the reason this page can quote decision rules rather than paraphrase them.

• Raiford, S. E., Weiss, L. G., Rolfhus, E., and Coalson, D. (2005). WISC-IV Technical Report #4: General Ability Index. Pearson, January 2005, updated December 2008. Source for the composition, the four part decision rule, the significant and unusual criterion, the Sattler base rate threshold, the Full Scale to GAI critical values, the standardization base rates in Table 3, and the special group rates in Table 4. Published PDF.
• Zhu, J., Raiford, S. E., and Coalson, D., with Saklofske, D. H. as discussant. The What, When, and How of the Wechsler General Ability Index. Pearson symposium slides, copyright 2007 and citing literature through 2009. Source for the WAIS-IV GAI and CPI composition, the GAI standard error of 2.57, the three GAI myths, the 73.5 percent dispersion figure, the index intercorrelations, and the clinical group means. Published PDF. The exact presentation date is ambiguous in the document itself.
• Raiford, S. E., Silverman, L., Gilman, B. J., and Courville, T. (2019). WISC-V Technical Report #5: Expanded General Ability Index. NCS Pearson, July 2019. Source for the eight subtest composition, the reliability of .97 and standard error of 2.62, the comparison with the Full Scale and Nonverbal indexes, and the statement that it replaces nothing. Published PDF.
• Weiss, L. G., and Gabel, A. D. (2008). WISC-IV Technical Report #6: Using the Cognitive Proficiency Index in Psychoeducational Assessment. Pearson. Source for the CPI composition and rationale, the 9 percent and 21 percent non clinical figures, the 2 percent against 45 to 50 percent combination rates, and the sensitivity and specificity values. Published PDF.
• National Association for Gifted Children (2018). Position Statement: Use of the WISC-V for Gifted and Twice Exceptional Identification, approved August 2018. Source for the six acceptable scores, the confidence interval condition, the processing speed argument, and the 390 child evidence base. Full text via ERIC.
• Rowe, E. W., Kingsley, J. M., and Thompson, D. F. (2010). Predictive ability of the General Ability Index versus the Full Scale IQ among gifted referrals. School Psychology Quarterly, 25(2), 119 to 128. N of 88 gifted referrals, all with significant index variability. The full text is behind a paywall, so no variance figures are quoted here.
• Raiford, S. E., Holdnack, J., Drozdick, L., and Zhang, O. (2014). Q-interactive Technical Report 9: The WISC-V and Children with Intellectual Giftedness and Intellectual Disability. Pearson, November 2014. Source for the gifted comparison of 24 children against 24 matched controls, with a GAI of 127.1 against 107.4 and an effect size of 1.72. Published PDF. This report does not publish Full Scale or Processing Speed figures for the gifted group, so it cannot support a GAI to Full Scale comparison.
• Bremner, D., McTaggart, B., Saklofske, D. H., and Janzen, T. (2011). WISC-IV GAI and CPI in psychoeducational assessment. Canadian Journal of School Psychology, 26(3), 209 to 219. Open PDF. Small samples and case material, cited here for the rationale rather than for figures.
• Foundational chapters without public text. Prifitera, Weiss and Saklofske (1998) in WISC-III clinical use and interpretation, pages 1 to 38. Tulsky, Saklofske, Wilkins and Weiss (2001), Development of a general ability index for the WAIS-III, Psychological Assessment, 13, 566 to 571. Saklofske, Prifitera, Weiss, Rolfhus and Zhu (2005) in WISC-IV clinical use and interpretation, pages 33 to 65. Referenced through Technical Report #4's bibliography, not quoted.
• ACIS technical manual. Every ACIS figure on this page: composite compositions, omegas, hierarchical omegas, standard errors, g loadings, index intercorrelations, and the score availability matrix by form. Published on this site.

The professional framework governing the choice between these composites is written down. The Standards for Educational and Psychological Testing, published jointly in 2014 by the American Educational Research Association, the American Psychological Association and the National Council on Measurement in Education, require that the interpretation of any score be supported by evidence for that specific use, that measurement error be reported with the score, and that the basis for selecting one composite over another be documented rather than assumed. The International Test Commission guidelines on test use place the same obligation on the administrator. APA standards on test use add the requirement that whoever reports a composite state what it supports and what it does not, which on this topic means saying out loud that a higher GAI is a different measurement rather than a better one.

14 Frequently Asked Questions