top of page

Standard, Scaled, T and Stanine Scores: One Table, Four Scales

Writer: Andrea Chernin
Andrea Chernin
5 days ago
19 min read

The short version


A standard score, a scaled score, a T score and a stanine are four ways of saying the same thing, and the conversion between them is arithmetic that nobody disputes. That is not where the trouble is.


The trouble is that one of those four words is a family name rather than a metric, two of them are used interchangeably by the publisher whose glossary defines them, one of them is derived two different ways by two different publishers, and the document that governs the whole field has a single word covering all four and defines it without a number in it.


None of that shows up on a report. What shows up is a column of numbers.

You are reading three scales at once and nothing says so


Open a typical file. A cognitive battery prints subtest scores clustered around 10. An achievement battery prints subtest scores clustered around 100. An executive function rating scale prints scores clustered around 50, and on that one the high end is the bad end. Somewhere there may be a stanine, and somewhere there is certainly a percentile rank.


Four metrics, three reports, one child. Not one of the three reports tells you that the other two exist.


The arithmetic that converts between these scales is correct. The assumption that the four columns are four readings of one instrument is the part that is wrong.

This post is about that assumption. If you want the error band around any single one of these numbers, that is a different question and it has its own post: standard error of measurement covers what the confidence interval on a report is and is not. If you want the frame the number is being read against, norm-referenced and criterion-referenced interpretation covers that. This one is about the metric itself.

The table, and the one number in it that matters


Here is the conversion, stated as the publishers state it. Each line gives the metric, its mean, its standard deviation, and its usual reported range.


  • Standard score — mean 100, standard deviation 15. On the Wechsler batteries, composites run 40 to 160.

  • Scaled score — mean 10, standard deviation 3, reported range 1 to 19.

  • T score — mean 50, standard deviation 10, running roughly 20 to 80.

  • Stanine — mean 5, standard deviation 2, range 1 to 9.

  • Normal curve equivalent — mean 50, standard deviation 21.06, range 1 to 99.

  • z score — mean 0, standard deviation 1.

  • Percentile rank — range 1 to 99, and not a standard deviation unit at all.


Riverside Insights' glossary of testing, measurement and statistical terms supplies the definitions for the T score, the stanine and the normal curve equivalent in those words, and Pearson's own WISC-V interpretive report supplies the first two.


What a one-point step is worth


The means and ranges are the part everyone remembers. The standard deviations are the part that changes what you can say, because they set how much of the distribution a single reported point covers.


Expressed in standard-score points, where one point is a fifteenth of a standard deviation:


  • One scaled-score point is worth five standard-score points.

  • One T-score point is worth one and a half standard-score points.

  • One stanine is worth seven and a half standard-score points, because a stanine is half a standard deviation wide.

  • One normal curve equivalent point is worth about seven tenths of a standard-score point.


A scaled score is five times coarser than a standard score. That is not a rounding detail. It is the reason a child can move a descriptive category on the strength of one more correct item.

The two metrics you may not average


Riverside's glossary is explicit that scale scores, "unlike Percentile Ranks and Grade Equivalents, are equal-interval, a property that allows these scores to be added, subtracted, and averaged."


Percentile ranks are not equal-interval. The distance from the first percentile to the second is a much larger distance in the underlying distribution than the distance from the fiftieth to the fifty-first, so a mean of percentile ranks is not a meaningful quantity.


The normal curve equivalent exists to solve exactly that problem, and the glossary says so: its advantages are that it has "the properties of an equal interval score, which means they can be averaged, unlike Percentile Ranks," and that "all test publishers derive these scores in the same way, which leads to some comparability in reporting procedures."


Read that second clause again. The one metric in the whole set that a publisher is willing to describe as derived identically across publishers is the one almost nobody reports in a clinical file. Grade equivalents sit on the other side of the same line, and the publishers are blunter about them than most practitioners expect.

The governing document has one word for all four of them


The Standards for Educational and Psychological Testing is published jointly by the American Educational Research Association, the American Psychological Association and the National Council on Measurement in Education.


The 2014 edition has been open access since March 2021, the full text is a free download, and AERA's own page for the edition is the place to confirm you are reading the current one. Its fifth chapter is titled "Scores, Scales, Norms, Score Linking, and Cut Scores" and runs from Standard 5.0 to Standard 5.23.


That is the chapter on this exact subject, written by the three bodies that set the rules, and it does not use any of the four words.


The counts, and how they were taken


Counted across the whole document as whole words, in flow order:


  • "standard score" — zero

  • "scaled score" — zero

  • "stanine" — zero

  • "T score" and "T-score" — zero

  • "z score" — zero

  • "normal curve equivalent" — zero

  • "normalized" — zero

  • "derived score" — zero

  • "grade equivalent" and "age equivalent" — zero

  • "scale score" — 42

  • "percentile rank" — 7


Three methodological notes, because a zero is only worth printing if it survives them.


The document was extracted twice, once in flow order and once preserving the two-column page layout, and every term was counted in both.


The layout pass threw up four apparent hits the flow pass did not, three for "stanine" and one for "standard score", and all four are the same artefact: a word broken across the column gutter, so that "Stan" at the foot of one column and "dards" at the head of the next read as a match for "stanine".


The same column splitting runs the other way too. The layout pass returned 39 for "scale score" and 5 for "percentile rank" against the flow pass's 42 and 7, which is why the flow-order figures are the ones reported here.


And the counts above are whole-word counts. A loose substring search for "t-score" returns three hits, every one of them inside the compound "test-score". A loose search is how a zero becomes a three, and an unexamined three is how a finding becomes wrong.


Standard 5.1 puts the obligation on whoever hands you the number


The chapter's first substantive requirement is short:


Test users should be provided with clear explanations of the characteristics, meaning, and intended interpretation of scale scores, as well as their limitations.

Its comment adds that "All scores (raw scores or scale scores) may be subject to misinterpretation," and Standard 5.2 requires that the procedures for constructing a reporting scale "be described clearly," including a rationale for "the number of score points that are used."


So the Standards do have a view about the metric on your report. The view is that someone owes you an explanation of it, including its limits. The Standards simply decline to name the metrics themselves, and the reason becomes obvious once you read the glossary entry.


The definition, in full, with no numbers in it


The glossary of the 2014 Standards defines the one term it does use like this:


scale score: A score obtained by transforming raw scores. Scale scores are typically used to facilitate interpretation.

That is the whole entry. No mean, no standard deviation, no range. It is a superordinate term, and it has to be, because it covers the Wechsler scaled score and the Wechsler standard score and the BRIEF T score and the stanine with equal accuracy. The governing document has one word for the category. Your desk has four words for members of it.

"Standard score" is a family name


Here is where the vocabulary starts costing something. Riverside's glossary defines the term as "a type of derived score, which is a transformation of the raw score, and whose score distribution in a specified population has convenient, known values for the mean and standard deviation." It then adds that "often this term is used to specifically denote z-scores (mean =0.0 and standard deviation =1.0), and any linear transformation of z-scores."


A standard score, on that definition, is any linear transformation of a z score. A T score is one. A stanine is one. The 100-and-15 metric is one. The term names the family, not the member.


The same glossary then proves it twice over in its own entries. It defines a T score as "a normalized standard score, having a mean of 50 and a standard deviation of 10." It defines a stanine as a normalized standard score with a mean of 5 and a standard deviation of 2.


And it carries a separate entry for the developmental standard score, which "provides a continuous growth scale of achievement from kindergarten through grade 12 for such tests as the Iowa Assessments" and has no 100-and-15 metric about it at all.


Three entries, one abbreviation


The abbreviation does not save you either. In that one glossary, three different entries claim the initials SS: "Scale Score (Scaled Score, SS)", "Standard Score (SS)", and "Developmental Standard Score (DSS or SS)".


If you have ever read "SS = 105" in somebody else's report and assumed you knew which scale it was on, the glossary of the publisher who issued the test disagrees with you.

The publisher says the two commonest words are interchangeable


This is the sentence worth carrying around. It appears in Riverside's glossary twice, once under Scale Score and once under Standard Score:


The term scale score and standard score are often used interchangeably, even though these scores may be derived at by different methods, their purpose and use can be similar.

That is not a critic's complaint. It is the glossary of a major test publisher, stating that the field's two most-printed score words are swapped for each other in practice while being produced by different procedures.


Which raises the question the rest of this post is about. If the derivation differs, what does the difference buy, and when does it bite?

Same publisher, same file, two subtest metrics


Start with the cleanest case, because it is one publisher and two of its own flagship instruments.


Pearson's interpretive report for the WISC-V states its own metrics plainly:


The primary and secondary subtests are on a scaled score metric with a mean of 10 and a standard deviation (SD) of 3. These subtest scores range from 1 to 19, with scores between 8 and 12 typically considered average.

A few lines later it does the same for the composites: the primary index scores and the Full Scale IQ "are on a standard score metric with a mean of 100 and an SD of 15", the index scores run 45 to 155 and the Full Scale IQ runs 40 to 160, and for both, "scores ranging from 90 to 109 are typically considered average."


So far this is the familiar picture: subtests on one scale, composites on another. Then the same report says this about the complementary battery, and it is the sentence worth stopping on: "Both the complementary subtests and index scores are on a standard score metric with a mean of 100 and an SD of 15, with a range of 45-155."


On one report, from one administration, some subtests are reported on a mean of 10 and others on a mean of 100. The level in the hierarchy does not tell you the metric. Only knowing which battery a subtest belongs to tells you the metric.


And the achievement battery from the same publisher uses neither


The sample score report for the WIAT-4, published by Pearson as a PDF, puts subtests and composites alike on the 100-and-15 metric. Its profile chart is banded from 40 to 160, its composite rows note that the raw score for a composite is the "Sum of Subtest Standard Scores", and the phrase "scaled score" does not appear in the report at all. Neither does "stanine".


Then take a third instrument in the same cabinet. Pearson's store entry for the CTOPP-2 lists its outputs as "Subtest scaled scores, percentile ranks, age and grade equivalents, composite indexes, and developmental scores," at qualification level B. Subtests on 10, composites called indexes, and two developmental metrics besides.


Three instruments, one publisher, three different answers to "what is a subtest score on this test."


The average band is not the same band


Now do the arithmetic, because this is the practical consequence and it is checkable in two lines.


A scaled score of 8 is two thirds of a standard deviation below the mean. A scaled score of 12 is two thirds above. Converted to the 100-and-15 metric, that band runs from 90 to 110.


Pearson's stated average band on the standard-score metric is 90 to 109, and the same report describes an index score of 111 as being in the High Average range. The top of the scaled-score average band therefore lands one point past the top of the standard-score average band.


A scaled score of 12 and a standard score of 110 are the same point in the distribution. One is inside the publisher's average band and the other is the first point outside it.

This is not a publisher error. It is what happens when you set integer boundaries on two scales whose steps differ by a factor of five. It matters because reports are written in words, not z scores, and "average" and "high average" are the words.


A sentence that compares a subtest in scaled-score units with a composite in standard-score units is comparing two different rounding regimes, and basal and ceiling rules can move the raw score underneath both of them before the conversion even starts.

A T score is not one derivation


Riverside's glossary calls a T score "a normalized standard score." A normalized score is one whose distribution has been forced to match the normal curve, so its percentile is read off that curve by construction.


Now read what the publisher of the BRIEF2 says about its own T scores. The BRIEF2 interpretive report states:


These scores are linear transformations of the raw scale scores (M = 50, SD = 10).

A linear transformation is not a normalization. Both produce a number with a mean of 50 and a standard deviation of 10, and the two numbers are not the same number unless the raw distribution was already normal. The percentile printed next to a linear T score comes from the sample; the percentile implied by a normalized T score comes from the curve.


Nothing on the face of a report tells you which kind you are holding. You have to go to the manual, and the manuals use the same two words.


And the direction reverses


The second thing a T score does, in the files this audience actually reads, is run the other way.


PAR's BRIEF-2 page describes the score report as including "index T scores and percentiles." The interpretive report then sets the bands, in its own words, "for all BRIEF2 clinical scales and indexes":


  • T scores from 60 to 64 are "considered mildly elevated".

  • T scores from 65 to 69 are "considered potentially clinically elevated".

  • T scores at or above 70 are "considered clinically elevated".


Convert those thresholds to the achievement metric and the problem is visible. A T score of 60 sits exactly one standard deviation above the mean, the same place as a standard score of 115. A T score of 70 sits two standard deviations above the mean, the same place as a standard score of 130.


On the BRIEF-2, 70 is clinically elevated. On Pearson's achievement report, 130 is Extremely high. Identical distance from the mean, opposite valence, and a file that contains both will print them in adjacent paragraphs with no note that the arrows point in different directions.


One more detail from the same report, worth knowing before you look for a number that is not there: "T scores are not generated for the Inconsistency scale." A validity scale on a T-score instrument need not be on the T-score metric.

The stanine is the only one whose publisher prints the warning


You are unlikely to find a stanine on a clinical battery's output. Pearson's WIAT-4 sample report does not contain the word at all. You are quite likely to find one in a school testing report that arrives in a file, which is why it is worth knowing how wide the band is.


Riverside's glossary explains the name, a contraction of "standard-nine", and gives the structure: stanines 2 through 8 are each half a standard deviation wide, with stanine 5 defined as the quarter of a standard deviation either side of the mean. It then publishes the mapping to percentile ranks, together with the descriptor it attaches to each band and the share of the norm group it expects to find there.


  • Stanine 1 — percentile ranks 1 to 4, four per cent of examinees, "low".

  • Stanine 2 — percentile ranks 5 to 11, seven per cent, "well below average".

  • Stanine 3 — percentile ranks 12 to 23, twelve per cent, "below average".

  • Stanine 4 — percentile ranks 24 to 40, seventeen per cent, "somewhat below average".

  • Stanine 5 — percentile ranks 41 to 59, twenty per cent, "about average".

  • Stanine 6 — percentile ranks 60 to 76, seventeen per cent, "somewhat above average".

  • Stanine 7 — percentile ranks 77 to 88, twelve per cent, "above average".

  • Stanine 8 — percentile ranks 89 to 95, seven per cent, "well above average".

  • Stanine 9 — percentile ranks 96 to 99, four per cent, "high".


And then the glossary says this, about its own score type:


Due to their coarseness, stanines are less precise indicators than percentile ranks, and at times be misleading (e.g., similar PR's can be grouped into different stanines [e.g., PR=23 and PR=24] and dissimilar PR's can be grouped into the same stanine [e.g., PR=24 and PR=40]).

Set that beside the table and the practical version is stark. Percentile ranks of 24 and 40 both report as stanine 4, "somewhat below average". Percentile ranks of 40 and 41 are one point apart and report as stanine 4 and stanine 5, "somewhat below average" and "about average".


One percentile point can change the adjective in a report. Sixteen percentile points can leave it untouched.

That is the publisher's own caution, printed in the publisher's own glossary, and it is a reasonable reading of what Standard 5.3 asks for: "If there is sound reason to believe that specific misinterpretations of a score scale are likely, test users should be explicitly cautioned."


The comment on 5.3 offers an example of its own that will be familiar to anyone who has read about outdated norms: a scale point "originally defined as the mean of some reference population should no longer be interpreted as representing average performance if the scale is held constant over time and the examinee population changes."

Converting across two tests is a linking claim


There is a distinction here that the conversion table hides, and getting it right is what keeps this post honest.


Converting a scaled score to a standard score within one instrument is arithmetic on one scale built from one normative sample. It is fine. It is the same information in different units.


Converting a score on one test into the metric of a different test is not that. It is a claim that the two scores are comparable, and the Standards have a name for such claims and a bill of requirements attached.


Standard 5.17 reads: "When scores on tests that cannot be equated are linked, direct evidence of score comparability should be provided, and the examinee population for which score comparability applies should be specified clearly."


Its comment lists the cases it covers, and the list ends on the two that matter here: "tests measuring different constructs, and tests from different publishers."


Standard 5.18 then specifies what a published linking owes its users, and the list is long: "the reliability of the sets of scores being linked, the correlation between test scores, an assessment of content similarity, the conditions of measurement for each test, the data collection design, the statistical methods used, the standard errors of the linking function, evaluations of sampling stability, and assessments of score comparability."


Note the sixth item. A linking has its own standard error, over and above the standard error of each score being linked. When a report says a child's phonological processing scaled score of 7 is "consistent with" a word reading standard score of 85, that sentence is a linking claim with no published linking behind it. The numbers can be converted. The constructs were never equated.

What the professional documents require, and what they do not say


The AET Code of Ethics, Revision #9, 2025, reissued September 2025, carries the two clauses that bear on this directly. Under its instruction and assessment responsibilities, the educational therapist is to "select and use appropriate assessment instruments, recognizing their limitations with respect to reliability, validity, and bias", and to "seek interpretation of assessment data from professionals in related fields (e.g. medical, psychological, speech/language, neuropsychological)".


Knowing which metric a number is on, and whether it may be averaged, compared or converted, is squarely inside "recognizing their limitations." The clause does not have to name a score type to reach it.


What the Code does not do is name one. A read of the current file this run found no occurrence of "standard score", "scaled score", "stanine", "percentile" or "T score" anywhere in it. That is consistent with the Code's general shape, which sets duties rather than procedures, and with the AET Fact Sheet, which draws the boundary around administration rather than around metrics.


Three honest caveats on that paragraph. First, the host now serves an interstitial page to automated clients, so the document could not be machine-counted this run the way the Standards were; the absence above is reported from a read of the document rather than from a word count, and anyone for whom it is load-bearing should open the PDF and search it.


Second, the Code is served under more than one file name. The one linked above and the file named for September 2025 both return the same document, Revision #9, 2025, reissued September 2025. A newer-looking file name is not a newer revision, and it is worth checking the revision line rather than the URL.


Third, the dotted clause references this site uses, such as II.I.B and II.I.D, are a convention for pointing at a provision. The Code itself numbers by section heading and letter and prints no dotted references.

A sequence, not a decision


What to do with a file full of mixed metrics, in order. None of these steps is a clinical judgement and none of them requires a conversion table on the wall.


  • Name the metric before you read the number. For every score you intend to quote, find the sentence in the report or the manual that gives its mean and standard deviation. If you cannot find it, say so rather than assuming.

  • Check which level of the instrument you are on. Subtest and composite are often different metrics, and on at least one battery some subtests are on the composite metric.

  • Convert into one metric before you compare anything. Pick the metric you will write in and move everything into it explicitly, so that a five-point difference is visibly five points of the same size.

  • Keep percentile ranks out of any arithmetic. Report them, do not average them, and do not treat a gap between two of them as a quantity.

  • Treat a cross-test comparison as a claim, not a calculation. If the two instruments were not linked, write the two scores and their metrics and let them stand side by side.

  • Check the direction of the scale. A high score is not good news on every instrument in the file.

  • Attach the band before you attach the adjective. The error band around a score usually crosses at least one descriptor boundary, and the descriptor is what the reader will remember.

When several organisations share a vocabulary


The score-type cluster on this site is now four posts deep, and it is worth reading as a set, because the same shape turns up in each.


Grade equivalent scores found two publishers printing the same score while giving opposite advice about using it. Standard error of measurement found that the width of the band on a report is a setting chosen by the publisher rather than a property of the test.


Norm-referenced versus criterion-referenced found that these name two interpretations rather than two kinds of test, and that the criterion in a criterion-referenced cut score is very often a percentile. Outdated test norms found that a scale point defined as average stops being average when the population moves and the scale does not.


The same pattern runs through the instrument posts. Psychoeducational evaluation covers what the report in front of you will and will not contain, and executive function assessment tools covers the rating scales that produce the T scores discussed above.


Test-retest intervals covers what the manuals permit when a score is taken again, and the pillar on educational therapists, learning specialists and executive function coaches covers who in the file is entitled to do which part of this work.

What is out of scope here


This post reports what the measurement standards and the test publishers say about score metrics. It is not advice about eligibility, entitlement or placement, and it is not a statement about what any statute or regulation requires. Nothing here is legal advice. It also stops short of telling you what any particular score means for a particular student, which is interpretation and belongs with the qualified professional who administered the instrument.

Where the published record runs out


Several things a practitioner would reasonably want are not in the public record, and it is worth being plain about which.


  • Whether a given instrument's T scores are linear or normalized is usually stated only in the technical manual, which is typically not web-readable. The BRIEF2 states it in its interpretive report, which is unusually transparent.

  • The descriptive category boundaries that publishers attach to score bands are printed on sample reports and in manuals but are not standardised across publishers, and no body publishes a crosswalk of them.

  • The AET Code's silence on score metrics is reported above from a read of the document rather than a machine count, because the host blocks automated retrieval.

  • Published linkings between the common clinical batteries are, as far as the public record shows, not available for most pairs a practitioner would want. The absence is the finding: if there is no published linking, there is no conversion to rely on, however easily the arithmetic runs.


One note on the sourcing. This post carries fewer outbound citations than most on this site, and that is deliberate rather than thin. The primary-source universe for score metrics is small: one standards document and its publishers' pages, one publisher glossary, three publisher sample reports or store entries, one rating-scale publisher, and one professional code.


Every one of them is cited above. Padding the list would have meant citing other people's blog posts about score conversion, which is exactly the kind of source this site does not use.

Frequently asked questions


What is the difference between a standard score and a scaled score?


In practice, a standard score usually means a metric with a mean of 100 and a standard deviation of 15, and a scaled score usually means one with a mean of 10 and a standard deviation of 3.


In the strict sense a standard score is any linear transformation of a z score, which makes scaled scores, T scores and stanines all standard scores. Riverside's glossary states that the two terms are often used interchangeably even though the scores may be derived by different methods.


How do I convert a scaled score to a standard score?


Multiply the distance from the mean by five and add it to 100. A scaled score of 7 is three points below a mean of 10, which is one standard deviation, so it corresponds to a standard score of 85. The conversion is exact within one instrument. Across two instruments it is a comparability claim rather than a calculation.


Is a T score of 65 bad?


It depends entirely on which instrument printed it. A T score of 65 is one and a half standard deviations above the mean, which on the 100-and-15 metric is 122.5. On an achievement test that would be a strength. On the BRIEF2, where a high score means more reported difficulty, the published band from 65 to 69 is "potentially clinically elevated".


What percentile is a stanine of 4?


Riverside's published mapping puts stanine 4 at percentile ranks 24 through 40, with the descriptor "somewhat below average". The band is wide by design, and the same glossary cautions that two percentile ranks as far apart as 24 and 40 fall in the same stanine while 23 and 24 fall in different ones.


Can I average percentile ranks across subtests?


No. Percentile ranks are not an equal-interval scale, so their mean is not a meaningful quantity. If you need an average of relative standing, the normal curve equivalent was built for that purpose and has a mean of 50 and a standard deviation of 21.06.


Do the testing standards say which metric to use?


No. The 2014 Standards for Educational and Psychological Testing never use the words standard score, scaled score, stanine or T score. The chapter on scores and scales uses the single term "scale score" and defines it as a score obtained by transforming raw scores.


What the Standards do require, in Standard 5.1, is that test users be given clear explanations of the characteristics, meaning and intended interpretation of those scores, together with their limitations.

Join the Illuminate educator network


If this is the kind of thing you want to read more of, the Illuminate educator list goes out to educational therapists, learning specialists and executive function coaches, and carries new practitioner posts as they publish. Join the educator list here.

Recent Posts

See All
Standard Error of Measurement: What That Band Means

Every score report prints a confidence interval and almost no practitioner acts on it. What the standard error of measurement is, why the band's width is a software setting rather than a property of t

 
 
 

Comments


bottom of page