Composite Scores and Subtest Scatter: The Rule Nobody Published
The short version
You were taught a rule about composite scores and subtest scatter. Most of us can even quote a number: if the subtests inside a composite sit too far apart, the composite is not interpretable and you report the parts instead.
Here is the awkward part. That number is not in the manuals.
We read the governing standards document and nine publisher documents for this piece, counting words in two extraction modes rather than trusting memory. Across nine of those ten documents — 229,781 words of them — the word scatter appears zero times. It appears twice in the tenth, and that tenth document is the one that offers you a second composite rather than taking the first one away.
What the publishers actually publish is four different architectures for the same problem. Not one of them is a prohibition on interpreting the composite.
That does not mean the rule is wrong. It means it is a convention from the interpretation literature that has been absorbed into practice as though it were a publisher requirement, and if you cite it in a report as a publisher requirement you are citing something that is not there.
Where the rule you were taught actually comes from
It comes from the interpretation textbooks, and you can watch a publisher hand the question over to one.
Pearson's own WISC-IV Technical Report #4, the document that introduced the General Ability Index, lists the clinical situations in which a practitioner might consider using the GAI instead of the Full Scale IQ. One of them is intersubtest scatter. Then, when it comes to saying how much scatter counts, it writes this:
Sattler (2001) suggests that differences between scores that occur in less than 10% to 15% of the standardization sample should be judged as unusual.
That is the publisher citing a textbook author for the threshold. Not a norm table, not a technical finding, not a rule the publisher states in its own voice — a suggestion, attributed, from outside the test.
This matters for how you write. "Pearson recommends the GAI when intersubtest scatter within working memory or processing speed is significant and unusual" is a true sentence you can source. "Pearson says a composite is uninterpretable when the subtests differ by more than X points" is not a sentence anyone can source, because Pearson does not say it.
The governing document never uses the word
The 2014 Standards for Educational and Psychological Testing are the joint standards of AERA, APA and NCME, and they are a free download. We extracted the full text twice, in flow order and in layout order, because a two-column PDF breaks words across the gutter and a single extraction will lie to you about counts.
Both extractions agree. As whole words, across 132,962 words of flow-order text: scatter, zero. uninterpretable, zero. subtest, two. profile, two. unitary, one. composite, twenty.
The one instance of "unitary" is the famous sentence that validity is a unitary concept. It has nothing to do with composites.
The same two-mode count returns zero for "scatter" in eight publisher documents as well: Pearson's WISC-V interpretive report, its WAIS-5, WIAT-4 and CELF-5 sample score reports, Riverside's glossary and its WJ IV bulletins numbered 2 and 11, and PAR's BRIEF2 interpretive sample report. The only document of the ten in which the word appears at all is the GAI technical report quoted above, where it appears twice.
So the vocabulary practitioners use to discuss this question — scatter, unitary, uninterpretable — is not the vocabulary of the document that governs the tests. That is worth knowing before you put any of those words in a report alongside a reference to the Standards.
What the Standards do require, and it is not a gate
The twenty instances of "composite" are not nothing. They are just a different shape of requirement from the one practitioners remember.
Standard 1.14 puts the duty on the developer, not the user:
When interpretation of subscores, score differences, or profiles is suggested, the rationale and relevant evidence in support of such interpretation should be provided. Where composite scores are developed, the basis and rationale for arriving at the composites should be given.
Its comment adds the sentence that does the most work in this whole area, and it cuts against reading differences freely rather than against reading composites:
Moreover, evidence for the validity of interpretations of two or more separate scores would not necessarily justify a statistical or substantive interpretation of the difference between them.
Read that twice. The caution is aimed at the difference, not at the total. The Standards are more worried about you interpreting the gap than about you interpreting the composite.
Standard 2.4's comment is the Standards' entire treatment of what we call scatter, written without the word:
Psychodiagnostic inferences are frequently drawn from the differences between subtest scores… When the interpretation of test scores centers on the peaks and valleys in the examinee's test score profile, the reliability of score differences is critical.
The requirement that follows is a reliability requirement on the publisher: provide standard errors for differences. It is not a threshold for the user.
Standard 2.5's comment goes further, and it is the sentence most directly at odds with the folk rule:
A single total score can be computed on tests that are multidimensional. The total score on a test that is substantially multidimensional should be treated as a composite score.
Multidimensionality, in the Standards' handling, is a reason to classify the total score as a composite and to report its reliability accordingly. It is not a reason to withhold it.
And the glossary definition is as spare as you would expect from a document that defines its terms without numbers in them: a composite score is "a score that combines several scores according to a specified formula." That is the whole entry. The formula has to be specified. Nothing is said about what the parts must look like.
Pearson offers a second composite, not a prohibition
Across three Pearson documents we could read in full, the pattern is consistent: give the practitioner another number and the base rates to judge it by, and never tell them the first number is void.
The GAI technical report's list of situations in which to consider the GAI includes a discrepancy between verbal comprehension and working memory, between perceptual reasoning and processing speed, between working memory and processing speed, and, in its own bullet, "significant and unusual intersubtest scatter exists within WMI and/or PSI." It points you to the critical-value and base-rate tables in the Administration and Scoring Manual, and to Table B.6 for subtest scatter specifically.
Then it describes what to do with the finding:
When unusual variability is observed within the set of subtests that comprise the FSIQ, clinical interpretation should characterize this diversity of abilities in order to be most useful for parents, teachers, and other professionals.
Characterize. Not suppress, not withhold, not declare invalid. The instruction is to describe the diversity, and the alternative composite exists so that an ability-achievement comparison is not distorted by the two indexes most likely to be depressed.
Two scope notes, because this is exactly where a careless citation goes wrong. That report is a WISC-IV document, last updated December 2008, and its tables are WISC-IV tables keyed to the WISC-IV standardization sample. Do not transplant its numbers to a later edition.
The later editions have their own documents, and they do not add a gate. Pearson's WAIS-5 sample score report, dated 2024, still prints a General Ability Index as an ancillary composite alongside the Full Scale IQ, under the heading "Composite Score Summary" — and the word "scatter" appears nowhere in it. Pearson's WISC-V interpretive report frames differences as additive information rather than as a disqualification:
It is common for children to exhibit score differences across areas of performance. Comparing the score differences in relation to three separate benchmarks may yield a richer portrait of a child's strengths and weaknesses.
A richer portrait. That is a publisher saying the differences make the picture better, in the same document where every composite is reported with a ninety-five per cent confidence interval.
Riverside builds its clusters from different abilities on purpose
The Woodcock-Johnson family answers the question in a different frame entirely, and the design intent is stated in print.
Assessment Service Bulletin Number 2, the WJ IV technical abstract and a free download, says this about how the clusters were built:
…clusters were constructed to subsume two or more qualitatively different narrow abilities. The principle of cluster interpretation was adopted to improve the content validity of measures for broad abilities such as reading, fluid reasoning, and general intelligence.
On this battery, the tests inside a cluster are supposed to be measuring different things. Qualitative difference between the parts is the reason the cluster is trustworthy as a measure of a broad ability, not evidence that something has gone wrong. A practitioner carrying over a Wechsler habit of reading within-composite difference as a warning sign will systematically misread a WJ profile.
What Riverside supplies instead of a gate is a normed difference apparatus: *variations*, which compare one score against the average of the others, and *comparisons*, which compare one score against a predictor such as the General Intellectual Ability or Gf-Gc Composite cluster. Both are built on the co-normed sample, which is the point:
Because all tests in the WJ IV are co-normed, the variation and comparison difference scores do not contain error that is inherent in measures based on different samples.
And then the threshold question is explicitly devolved. The bulletin says the apparatus "allows a professional, school district, or state to define a criterion of significance in terms of either the discrepancy SD or the discrepancy PR." The publisher supplies the distribution. Somebody else supplies the cut.
Riverside's definitions match that restraint. Its free Glossary of Testing, Measurement and Statistical Terms defines a composite score as "a score that is derived by combining one or more scores according to a specified formula," typically by averaging or summing contributing scores "which are often weighted according to their relative importance."
Formula and weighting — nothing about what the parts must look like. And Assessment Service Bulletin Number 11, which sets out a full selective-evaluation method across 23,000 words and three worked case reports, never uses the word "scatter" once.
PAR's software applies a base rate and hedges the composite
Of the four publishers here, exactly one gates a composite in the report narrative — and it does it in software, using a base rate, without ever showing the practitioner a point threshold.
The BRIEF2 parent-form interpretive sample report prints this:
The Global Executive Composite (GEC) is an overarching summary score that incorporates all of the BRIEF2 clinical scales. Although review of the BRI, ERI, CRI, and individual scale scores is strongly recommended for all BRIEF2 profiles, the GEC can sometimes be useful as a summary measure. In this case, the three summary indexes are not substantially different from each other, with differences between T scores for each seen in 90% of the standardization sample. Thus, the GEC adequately captures the elevation or severity of the overall profile.
Three things in one paragraph. The composite "can sometimes be useful," which is a markedly weaker endorsement than Pearson gives the Full Scale IQ. The gate is a base rate — the differences are common enough in the standardization sample to be unremarkable. And the conclusion is conditional: *in this case*, the composite captures the profile, with the clear implication that in another case it would not.
That is the closest thing to the rule you were taught that any of these documents contains, and note what it is not. It is not a number of points. It is not stated in the manual as a user-applied threshold.
It is applied for you by the scoring platform, and the practitioner's job is to read the sentence and understand that it is a conditional. PAR's own product page confirms the score report carries the index T scores and percentiles; the interpretive narrative is where the conditional lives.
One publisher does not use the word at all
It is worth seeing how differently the same object can be presented.
Pearson's CELF-5 individual sample score report contains the word "composite" zero times. Its overall score is the Core Language Score, and it sits in a table headed "Core Language Score and Index Scores" — a score beside other scores, with no hierarchy marked in the vocabulary and no interpretability condition printed anywhere.
Pearson's WIAT-4 sample score report does the opposite and names the formula on the page: "For composites, Raw score refers to Sum of Subtest Standard Scores." That single line satisfies Standard 1.14's requirement that the basis for arriving at a composite be given, and it attaches no condition whatever to interpreting the result.
So within one publisher's catalogue you have a report that never says "composite," a report that defines its composites arithmetically and stops there, and a technical report that offers a second composite and borrows its threshold from a textbook. If you work across instruments — and you do — there is no single rule to carry between them.
Four architectures, one problem
What the published record actually offers, stated as plainly as we can manage:
An alternative composite plus base-rate tables — the Wechsler approach. Significant and unusual difference is a reason to consider a different summary score, and to describe the variability, not a reason to withhold the Full Scale score. The threshold for "unusual" is attributed to a textbook author.
A normed difference distribution with the cut devolved — the Woodcock-Johnson approach. Within-cluster difference is a design feature; the apparatus quantifies difference against the co-normed sample and leaves the criterion of significance to the professional, district or state.
A software-applied base-rate conditional — the BRIEF2 approach. The narrative tells you whether the composite captures the profile in this case, in words, with no point threshold shown.
Formula disclosed, no condition stated — the WIAT-4 and CELF-5 approach. Here is how the score was made; interpret it as a score.
Four defensible designs. No shared vocabulary. And the folk rule — a point spread above which a composite becomes uninterpretable — belongs to none of them.
What you may and may not write
The sourcing discipline here is narrow and it matters, because a report is a document somebody else will rely on.
You can write that a publisher offers an alternative composite and name the conditions under which it suggests considering one, with the citation. You can report a base rate from the publisher's tables. You can describe the variability you observed and what it limits, in your own voice, as a clinical observation.
You can say that a difference reached statistical significance and separately whether it was common or uncommon in the standardization sample, because those are two different facts and the Wechsler documents are explicit that they answer different questions.
You cannot attribute a point threshold to a publisher that does not state one. If the number you are using comes from the interpretation literature, name the author, not the test.
You cannot write that a composite is invalid, or uninterpretable, on the strength of subtest variability alone. Nothing in the Standards or in any of these publisher documents supports that sentence, and Standard 2.5's comment points the other way.
And you cannot use any of this to identify a disorder. A pattern of scores is not a diagnosis. What a profile means clinically sits outside an educational therapist's scope unless you hold another qualification that covers it.
The AET Code of Ethics (Revision #9, 2025, reissued September 2025) is explicit on both halves of that. Members use only those assessment instruments for which they have been adequately trained (II.I.C). They select instruments while "recognizing their limitations with respect to reliability, validity, and bias" (II.I.B). And they "seek interpretation of assessment data from professionals in related fields" (II.I.D).
We should say plainly how we read that Code. Its host serves a challenge page to automated clients, so the Code was read through a fetch-and-summarize route rather than extracted and word-counted as the Standards were.
On that read, the words composite, subtest, scatter and index score do not appear in it. Treat that as a characterization of the document and not as a machine count — the distinction is the difference between a claim we can prove and a claim we can only report.
A sequence, not a decision
Find out which composite you are holding, and what formula made it. A Core Language Score, a cluster, a Global Executive Composite and a Full Scale IQ are four different objects with four different design intents. The report usually tells you the formula if you look.
Read the differences with their own reliability data, not by eye. Standard 2.4 exists because a difference between two reliable scores can be far less reliable than either. The critical-value tables are in the manual for this reason.
Ask the base-rate question separately from the significance question. Significance says the difference is probably real. The base rate says whether it is unusual. Only the second one licenses the word "unusual" in a report.
Check whether your battery publishes an alternative composite. If it does, the publisher has already answered the question you are asking, and the answer is a different number rather than silence.
If a threshold is load-bearing for what you are about to write, name its source in the report. If it is a textbook rule, say so. That costs you a clause and protects the whole document.
Write the differences; do not suppress the composite. Characterize the diversity, state what the composite does and does not summarize, and let the reader see both.
When several organisations share a vocabulary
This is the fourth piece on this site to find the same thing in a different corner of the assessment literature, and the cluster is now deep enough to be useful on its own.
Our comparison of standard, scaled, T and stanine scores found the Standards using one superordinate word for four incompatible metrics and defining it with no number in it — the same spareness you have just seen in the glossary entry for "composite score."
The piece on the standard error of measurement found that the width of the band on a report is a setting rather than a finding, which is the reliability-of-differences problem in its single-score form.
Norm-referenced versus criterion-referenced found that the Standards define two interpretations rather than two kinds of test, and that a publisher's own glossary allows one test to be both. Basal and ceiling rules found the same devolution of a numeric rule to the manual rather than the standard.
Three more are directly adjacent. Grade equivalent scores are the clearest case of a metric whose publishers warn against the interpretation practitioners actually make of it. Outdated test norms is the question of what a composite is a composite *of*, once the sample is old. And test-retest intervals and practice effects is where the difference between two administrations gets the same treatment the difference between two subtests gets here.
If the instrument itself is the open question, achievement test qualification levels covers who may purchase and administer what, psychoeducational evaluation for educational therapists covers what an educational therapist may and may not do with a report somebody else wrote, and test accommodations and what the manual permits covers the administration side.
For executive function specifically, where the BRIEF2 lives, executive function assessment tools, compared sets out the instruments. And if you are still working out where your own role sits in all of this, the educational therapist, learning specialist and EF coach comparison is the place to start.
What is out of scope here
This is not guidance on eligibility determination, and it is not legal advice. Whether a particular score pattern supports a specific educational classification is a question for the team making that decision under the rules of its own jurisdiction, and those rules vary. Nothing here describes what any law requires.
It is also not diagnostic guidance. We have deliberately not said which composite to report for any described presentation, or what a given profile means clinically, because those are judgements that belong to a qualified evaluator working with the full record.
And it is not a substitute for the manual on your shelf. Every numeric threshold that genuinely exists for your instrument is in that manual, keyed to that standardization sample.
Where the published record runs out
Honesty about the limits of this piece, because they are substantial.
The technical manuals are the documents that would settle this, and they are not web-readable. The WISC-IV Administration and Scoring Manual's Table B.6, which is where Pearson sends you for subtest scatter base rates, is a print table in a kit. So are the WISC-V and WAIS-5 Technical and Interpretive Manuals, the BRIEF2 Professional Manual, and the WJ IV Technical Manual.
If a scatter threshold with a number attached exists for your instrument, that is where it is, and we have not seen it. What we can say is bounded and we mean it narrowly: across one standards document and eight publisher documents that are free to read, no numeric interpretability threshold for a composite is stated. That is a claim about the readable record, not about every page a publisher has printed.
The source count here is twelve external citations across nine domains, and that is the whole universe rather than a shortfall. The readable primary record for this question is one standards document, five Pearson sample and technical reports, two Riverside bulletins plus its glossary, one PAR interpretive report and its product page, and one professional code.
Every one of them is cited. Padding past that would mean citing competitors' blogs or the interpretation textbooks second-hand, and we do neither.
The AET Code was characterized, not counted — see the note above. A host that blocks automated extraction gets a weaker claim from us, and we say which we did.
And the Sattler attribution is as far as we can take the threshold question. We can show you a publisher citing a textbook author for what counts as unusual. We have not read that textbook for this piece, so we are not reporting what it says beyond the sentence Pearson quotes.
Frequently asked questions
Is it true that the Full Scale IQ should not be interpreted when the index scores differ significantly?
Not as a publisher rule. Pearson's published guidance in the GAI technical report is that significant and unusual difference is a reason to *consider* the General Ability Index and to characterize the variability in the clinical interpretation.
The word "uninterpretable" does not appear in any of the Pearson documents read for this piece, and the 2014 Standards do not contain a threshold of any kind. The rule as commonly stated is a convention from the interpretation literature.
What counts as "unusual" scatter, then?
The only threshold we found stated in a publisher document is a base rate, and the publisher attributes it: Technical Report #4 reports Sattler's suggestion that differences occurring in less than 10% to 15% of the standardization sample be judged unusual. That is a WISC-IV-era citation in a WISC-IV document. If you use it, cite it as what it is.
Does the same logic apply to a Woodcock-Johnson cluster?
No, and this is the most important cross-instrument point here. Riverside states that WJ IV clusters were deliberately constructed from two or more qualitatively different narrow abilities, to improve content validity for broad abilities. Difference between the tests in a cluster is by design. The WJ apparatus for reading difference is the variation and comparison procedures, and the criterion of significance is explicitly left to the professional, district or state.
Why does my BRIEF2 report comment on whether the composite is useful?
Because PAR's interpretive narrative applies a base-rate check and writes the result into the sentence. The report says the Global Executive Composite "can sometimes be useful as a summary measure" and then states, for that profile, whether the three summary indexes are substantially different, using the frequency of those differences in the standardization sample. It is a conditional applied by the software, not a threshold for you to apply.
Can I use subtest scatter to identify a specific learning disability or a processing disorder?
No. A score pattern is not a diagnosis. Standard 1.14's comment warns explicitly that validity evidence for interpreting two separate scores does not by itself justify interpreting the difference between them, and Standard 1.15's comment says test manuals and score reports should discourage overinterpretation of information subject to considerable error, "especially … if interpretation of performance on isolated items, small subsets of items, or subtest scores is suggested."
For educational therapists, the AET Code's assessment clauses limit you to instruments you are trained on and direct you to seek interpretation from professionals in related fields.
What should I actually write in a report when the subtests are far apart?
Report the composite with its confidence interval, report the differences with their significance and their base rates as separate facts, describe in your own clinical voice what the variability limits about the summary score, and if you are relying on a threshold, name its source.
If your battery publishes an alternative composite, report that too and say why. What you should not do is leave the composite out and let the reader assume it was never computed.
Join the Illuminate educator network
We write one of these a day for educational therapists, learning specialists and executive-function coaches — assessment literacy, program comparisons, and what the issuing bodies actually say, sourced every time.
Join the Illuminate educator list and we will send new pieces as they publish.


Comments