top of page

Test-Retest Intervals and Practice Effects: What the Manuals Permit

Writer: Andrea Chernin
Andrea Chernin
Nov 7
24 min read

A student finished sixteen weeks of intervention three weeks ago and the family wants a number. You own the battery. You administered it in March. Can you administer it again now?


The honest answer is that the test-retest interval is a decision the field has handed to test publishers, and most publishers have not published one. What they have published instead is a practice effect figure, a reliability coefficient, and in a few cases a second form of the test. None of those three things is a minimum interval, and treating any of them as one is the mistake this piece is about.


This is a survey of what the published documentation actually says: the governing standard, the technical abstracts, the product pages, and the progress-monitoring guides. The pattern is the same one that shows up every time this site reads several publishers side by side. They are not disagreeing about the answer. They are answering different questions in incompatible units.

The short version


  • The Standards for Educational and Psychological Testing do not set a re-test interval. They assign the decision to the test developer or test user, and require that any interval used in a reliability study be reported.

  • The WJ V publishes the most detailed practice-effect data in the field, and nearly all of it comes from administrations separated by a median of 0.98 hours.

  • The WIAT-4's test-retest study used an interval of 12 to 87 days, mean 31.5 days, reported by an independent reviewer rather than on any public page.

  • PRO-ED publishes test-retest coefficients for the TOWRE-2 and GORT-5 with no interval attached on either product page, which is the shape Standard 2.19 explicitly calls insufficient.

  • The only organisation that sets a number sets it at twenty alternate forms, and it applies only to progress-monitoring tools, not to the batteries you re-administer.

  • DIBELS 8 is the one publisher that states an interval in weeks, and the interval it states is two weeks — for a different kind of instrument than the one on your shelf.

Two different things get called an interval


Before any of this: the word does double duty, and the two senses point in opposite directions.


A confidence interval is a band around a single score, expressing how much that one number could move if you measured again under identical conditions. This site has a whole piece on what the band on the report actually means, and the short version is that the band is a setting rather than a finding.


A re-test interval is elapsed time between two administrations. It does not express uncertainty. It is a scheduling fact, and it is the thing that determines whether the second score means anything.


A confidence interval tells you how much one score could have moved by chance. A re-test interval tells you how much of the second score is memory.

The two interact. If the gain you are looking at is smaller than the band, the interval is beside the point. If the gain is larger than the band but the interval was two weeks, the band is beside the point. You need both, and the manuals give you one of them.

What the Standards require, and who they hand the decision to


The Standards for Educational and Psychological Testing — the joint product of AERA, the APA and NCME, and the closest thing this field has to a governing document — went open access, and the 2014 edition is a free PDF. It is worth reading what it does and does not do here.


It defines the term, in the glossary, without any interval at all:


retesting: A repeat administration of a test, using either the same test or an alternate form, sometimes with additional training or education between administrations.

It then hands the policy to somebody else. The comment accompanying Standard 6.1, on following standardised procedures, is explicit:


Policies regarding retesting should be established by the test developer or user. The test user and administrator should follow the established policy.

That is the whole mechanism. The Standards do not say how long to wait. They say that whoever built the test, or whoever is using it, should decide and write it down — and that the person administering should follow what was written. For a practitioner holding a kit whose manual says nothing on the subject, that sentence transfers the decision to them.


Two further provisions matter, and both cut against how reliability figures are usually quoted.


Standard 2.6 says a coefficient addressing one kind of variability should not be read as interchangeable with one addressing another, and its comment names the three: "Internal-consistency, alternate-form, and test-retest coefficients should not be considered equivalent, as each incorporates a unique definition of measurement error." A coefficient above .90 therefore tells you nothing about re-testing unless you know which of the three it is.


Standard 2.19, on documenting reliability, is the one that bears directly on product pages. Its comment states the requirement and then gives a worked example of what not to write:


If the test-retest or alternate-form approach is used, the interval between administrations should be indicated.

The same comment then gives the worked example of the error. Because there are many ways of estimating reliability, each influenced by different sources of measurement error, the Standards call it "unacceptable to say simply, 'The reliability/precision of scores on test X is .90.'" Hold that sentence. Several of the pages below say almost exactly that.


Chapter 2 also names the specific hazard of the same-form re-test, in the section on factors affecting reliability: "Another limitation of test-retest coefficients is that, when the same form of the test is used, the correlation between the first and second scores could be inflated by the test taker's recall of initial responses."


And Standard 13.2, written for programme evaluation but applicable to any change score, requires that the time periods between administrations be reported and that "care should be taken to avoid practice effects."


Its comment adds that where the same test is used on multiple occasions, "the possibility of practice effects (i.e., improved performance due to familiarity with the test items) should be examined." The Standards describe the obligation thoroughly and name no number anywhere.


One more, because it concerns what you tell the family before you start. The comment to Standard 10.7 lists what a test taker should understand in advance, and the list includes "whether the test taker will have an opportunity to retest, and under what circumstances retesting could occur." That is a disclosure obligation attached to a policy most kits do not contain, and it belongs in the same conversation as informed consent.

The WJ V publishes the most detailed record, and it is almost all same-day


Riverside's WJ V Technical Abstract is a free download and it is the most useful document in this whole area. It is also the one that most clearly demonstrates the gap between what a publisher measures and what a practitioner needs.


Start with what exists. The WJ V provides parallel second forms of 27 tests — five in the Cognitive battery, seventeen in the Achievement battery, and five in the Virtual Test Library.


The abstract names two uses for them, and the second is exactly our question: second forms can be used "if the administration of a speeded test is spoiled due to an unexpected interruption during the timed portion of the test" or "when an examinee who was administered the first form of a test in a prior evaluation is being retested."


So the publisher anticipates the re-test and supplies the instrument for it. Now look at the interval in the study that validated it. The alternate-form coefficients come from the calibration/alternate-form/test-retest study that ran alongside norming, and the abstract reports its timing plainly: "Most (95.7%) of the CAR examinees were administered all the tests in their assigned administration set within the same testing session."


For the small number scheduled across two sessions, "the time elapsed between the administration of the two forms of each test was generally less than 24 hours." Table 16 reports alternate-form coefficients for 24 tests, running from 0.81 to 0.99 at ages 4 to 19.


Of the 27 second forms, three are identical to the first form, so for those three the abstract reports test-retest reliability instead, in Table 17 — 0.92 to 0.95 at ages 4 to 19.


Those are good numbers. They are also numbers about content sampling measured inside a single afternoon, which is a different thing from stability across a school term.


The practice-effect table, and the one hour behind it


The abstract's treatment of practice is better than almost anything else in print. It names the mechanism, calls it by both of its names, and quantifies it.


Prior exposure to a test task, it says, "can impact an individual's performance on subsequent administrations of the test — even when the content of the individual items is different between the two forms", and it names the phenomenon twice over, as a carryover effect and as a practice effect.


Then it reports the elapsed time, and this is the figure to carry away:


Most (96.9%) of the examinees in the CAR study (described earlier in this document) were administered the first form of the tests followed by the second form of the tests within the same day (minimum = 0.03 hours, maximum = 43 days, median = 0.98 hours).

A median of 0.98 hours. Roughly fifty-nine minutes. The footnote adds that about three per cent of the sample took the second form more than a day later, "due to unexpected scheduling conflicts" — so even the upper end of that range is an accident of scheduling rather than a design.


Table 18 reports the resulting effect sizes as Cohen's d, with positive values meaning higher scores on the second administration. Most are negligible: Number-Pattern Matching at minus 0.02, Letter-Pattern Matching at minus 0.02, Magnitude Comparison at 0.01, Rapid Letter Naming at minus 0.02 for ages 4 to 19. A few stand out at the younger age band:


  • Rapid Phoneme Naming, d = 0.57 — a medium effect, the largest in the table

  • Word Reading Fluency, d = 0.23

  • Sentence Writing Fluency, d = 0.13 and Letter Writing Fluency, d = 0.13


At ages 20 to 80 and above, Rapid Phoneme Naming is 0.52, Rapid Letter Naming 0.27 and Word Reading Fluency 0.27.


That the largest practice effect in a major battery sits on a rapid-naming task is worth knowing if you use those measures, and it lines up with what the rapid automatized naming literature already says about what those tasks are doing.


The abstract itself predicts it, naming the processing-speed and RAN tests as the ones where prior exposure matters most, and adding that retrieval-fluency tasks may improve because examinees "reflect on the strategy they used during the first administration."


Here is the part that matters for your Thursday. A d of 0.57 measured across one hour tells you that the task is sensitive to exposure. It does not tell you what the effect is at six weeks, or at four months, because the study did not measure those intervals.


Nobody should read a near-zero practice effect in Table 18 as a licence to re-administer at any interval, and nobody should read 0.57 as the correction to apply at six months. The table is a warning label, not a conversion factor.

What the WIAT-4's stability study actually sampled


Pearson's WIAT-4 does not publish its reliability study design on a public page. The figures are in the Technical Manual, and the most useful public route to them is the independent test review by Stefan Dombrowski and Corinne Casey in the Journal of Psychoeducational Assessment (2022, vol. 40, no. 8, 1033–1038), DOI 10.1177/07342829221116808. That review reports the interval:


Test-Retest reliability estimates (interval 12–87 days; mean = 31.5 days) are provided in Tables 2.5 and 2.6 of the Technical Manual.

Twelve to eighty-seven days, averaging about a month. The same review reports that coefficients ran from acceptable to excellent "except for Listening Comprehension (.60s–.70s) and Alphabet Fluency (.64) which were low."


That is a far more usable interval than a median of one hour, and it still leaves the practitioner's question open, because a range of 12 to 87 days means the figure is an average across intervals that differ by a factor of seven. A coefficient computed over 12 days and a coefficient computed over 87 days are not the same quantity, and a mean of 31.5 days does not separate them.


A second independent source adds a caution about which composites to trust across time. The evaluation of the WIAT-4 as a measurement instrument by Alexander Beaujean and Jason Parkin, Journal of Intelligence 10:30, DOI 10.3390/jintelligence10020030 and also at PMC, quotes the manual's stability estimates directly: greater than .90 for the reading composites and both mathematics composites, .85 for Written Expression, and .60 for Writing Fluency.


Their comment on the last of those is blunt — that even if the authors had made a measurement claim for that score, "the claims would be difficult to support because of the low stability estimate."


Two things to keep straight. These authors are quoting the manual's own tables, not re-running the study; and the stability figures they cite at pages 22 and 23 of the manual are not necessarily the same analysis the test review cites at Tables 2.5 and 2.6.


Read them as two independent readings of the same technical manual rather than as one finding. What survives either way: a composite with a stability estimate of .60 should not be the number you report a gain on, whatever the interval.

The PRO-ED pattern: a coefficient with no interval attached


Now the instruments most educational therapists actually own, and the exact shape Standard 2.19 names.


PRO-ED's TOWRE-2 page states its reliability evidence like this:


The average alternate forms reliability coefficients (content sampling) for the subtests exceed .90. The average test-retest (time sampling) coefficients for the same form exceed .90. The average test-retest (time sampling) coefficients for different forms of the subtests are .87.

Three coefficients, correctly distinguished by the source of error — content sampling against time sampling, which is better than most pages manage. And no interval. The page does not say whether "time sampling" meant a fortnight or a term.


PAR's GORT-5 page does the same thing for a test with two equivalent forms of sixteen passages each. Its average test-retest coefficients for the Oral Reading Index "for the same form (i.e., Form A to A, Form B to B) exceed .85", and the average coefficient "for different forms (i.e., Form A to B, Form B to A) is .85."


Again: correctly split by form, no interval. The PRO-ED GORT-5 kit page carries no reliability detail at all.


Bound this properly, because it is a claim about pages and not about manuals. The examiner's manuals for both instruments report their reliability studies in full, almost certainly including the intervals; the TOWRE-2 page says so, referring to "the numerous reliability and validity studies presented in the manual."


The finding is narrower and still useful: the only version of this information most purchasers see before they buy carries the figure without the design. A practitioner choosing an instrument because it has a published test-retest coefficient above .90 is choosing on the basis of a number whose meaning depends on a fact the page omits.

The one publisher that names a cadence


One page in this set does state a re-test frequency, and it deserves credit for it. The same PRO-ED TOWRE-2 page:


Although the TOWRE-2 has not been developed to provide weekly or monthly assessments, it certainly can be used to monitor progress in growth of word-level reading skills three or four times per year.

That is the clearest statement of re-administration cadence on any comprehensive or targeted achievement instrument this engine has read. It is a frequency rather than a minimum interval — three or four times a year implies roughly a quarter between administrations without saying so — and it comes with the limiting clause stated first.


It is also made possible by the form count. The TOWRE-2 has four alternate forms, A through D, for each of its two subtests, and PRO-ED sells a separate booklet containing all four. Four forms and three or four administrations a year fit together: you can do a year without repeating an item set.

Count the forms you actually own


Form count is the practical constraint behind every interval question, and it varies more than most practitioners realise.


  • TOWRE-2 (PRO-ED) — four alternate forms per subtest, A through D

  • GORT-5 (PRO-ED and PAR) — two equivalent forms, A and B

  • KTEA-3 (Pearson) — two parallel forms, A and B

  • WJ V (Riverside) — parallel second forms for 27 tests, three of which are identical rather than parallel second forms

  • CTOPP-2 (PRO-ED and PAR) — no alternate form described on either published page


Pearson's KTEA-3 page leads with the form count as a selling point, describing a battery "including two parallel forms, two measures of rapid automatized naming (RAN), story-based Written Expression measurement, and norm-referenced error analysis", and the store sells Form A and Form B kits separately.


The CTOPP-2 is the instructive case. Neither PRO-ED's page nor PAR's page mentions an alternate form, a test-retest coefficient or a stability estimate anywhere. PAR's technical paragraph reports internal consistency only: reliability "was demonstrated by average internal consistency coefficients that exceed .80 for all subtests except Nonword Repetition, with an average alpha of .77", and composite coefficients "all .85 or higher."


Internal consistency is the one coefficient that carries no information whatever about re-testing, and Standard 2.6 says so. Again, bound it: the examiner's manual reports more than the page does. But if your plan is to re-administer the CTOPP-2 to show change in phonological awareness, nothing on either seller's page tells you what to expect, and there is no second form named to administer.


One terminology trap in the same instrument, because it will catch somebody. The CTOPP-2 offers an Alternate Phonological Awareness Composite Score for ages 7 to 24, which PAR describes as measuring "phonological awareness exclusively with nonwords." That is a different item type, not a second form of the test. The word "alternate" there has nothing to do with re-testing.

The only number anybody sets is twenty


There is exactly one place in this field where a quantitative standard for re-testing material exists, and it is not a publisher.


The National Center on Intensive Intervention's Academic Progress Monitoring chart rates tools on several dimensions, one of which is Alternate Forms. Its rating rubric sets the threshold:


Full Bubble: There are at least 20 alternate forms and evidence is strong for comparability of alternate forms, and from a sample of students in need of intensive intervention.

A half bubble requires the same twenty forms with moderate comparability evidence. Anything less is an empty bubble.


Twenty. That is the number the field's own clearinghouse treats as the minimum for an instrument designed to be re-administered. Set it beside the table above: four forms is the best-provisioned instrument in the practitioner's kit, and a great many have two or one.


The point is not that the TOWRE-2 fails an NCII standard — it is not a progress-monitoring tool and NCII does not rate it as one. The point is the reverse, and it is the thing to take from this whole piece: the instruments built for repeated administration have twenty forms, and the instruments you re-administer have two. That is the gap the re-test interval is being asked to cover.


How NCII's bubbles are actually derived is covered in the piece on what the chart ratings mean for mastery decisions, and the tools themselves are compared in progress monitoring tools, compared.

Progress monitoring runs on a different clock entirely


Which brings us to the one publisher that states intervals in weeks. The University of Oregon's DIBELS 8 Administration and Scoring Guide is a free PDF and its Chapter 4 does what no achievement-battery manual page does: it gives a schedule, with reasons.


In general, the more foundational the skill and the more intensive the intervention, the more frequent progress monitoring should be. However, it is rarely advisable to progress monitor more than once a week. In fact, to avoid excessive assessment, we recommend progress monitoring every two weeks in kindergarten through third grade.

For grades 4 and up it recommends every second or third week depending on intervention intensity, citing slower oral reading fluency growth in the upper grades — the same growth-rate question behind oral reading fluency norms.


For students at some risk receiving strategic support, "progress monitoring every four weeks (or four to five weeks in Grade 4 and up) is generally adequate." Maze is the exception and gets a monthly ceiling, because "even under intensive intervention, reading comprehension is difficult to improve rapidly."


That schedule is supported by material. DIBELS 8 selected twenty progress-monitoring forms per grade for several of its subtests and, since 2021, ten Maze forms per grade. The schedule and the form count are the same decision described twice.


It also publishes the rule for what to do when you have run out, which no achievement battery does. Where a student "has already undergone progress monitoring to any extent with a subtest", the guide says to "choose a form that the student has not yet seen", and if the student "has seen all the progress monitoring forms, go back to the first progress monitoring form."


That is a publisher admitting in print that re-use is inevitable and telling you how to do it in the least damaging order.

A week is the published memory threshold for one task


The single most concrete interval statement in everything read for this piece is a sentence about letter naming. DIBELS 8's Letter Naming Fluency has no dedicated progress-monitoring forms, so the guide explains what to do when an administration is invalidated:


If a week or more passes between the invalidated administration and the new one, the identical form can be used.

A week. For that task, on that instrument, from that publisher, with a stated reason — the guide goes on to note that letter naming is not a meaning-laden task, unlike reading connected text.


That is the shape of the sentence practitioners have been looking for in their achievement-battery manuals, and it exists here because the publisher is a university reporting on a one-minute measure whose item pool is twenty-six letters.


It does not transfer. Nobody should read "a week is enough for letter naming" as "a week is enough", and the guide does not invite that reading. But it is worth seeing once, because it shows what the answer looks like when somebody is willing to give one.

Form effects are not practice effects, and DIBELS says so


One more distinction from the same guide, because it is the error most likely to be made in the opposite direction. Switching to a fresh form removes the memory problem and introduces a different one.


Despite "strenuous efforts to create equivalent forms", the guide says, "differences in difficulty between forms still occur", and while those form effects are "generally quite mild for many subtests as a result of the constrained item pool, they become more apparent in subtests involving connected text."


The guide then gives the worked example, and it is the clearest statement of the problem anywhere in this literature:


For example, a student who scores 100 words-correct-per-minute (WCPM) in the beginning of year and 120 WCPM in the middle of year has indeed read the middle of year passage at a faster rate, but whether the difference in 20 WCPM is due to actual growth or the middle of year passage simply being easier to read remains unclear.

Twenty words per minute of apparent growth that might be the passage. That is a publisher, in its own administration guide, telling you that its own equated forms can move a score by an amount most practitioners would report as progress.


It is the reason a second form is not a free solution to a re-test problem, and the reason the decision rules you apply to a progress-monitoring graph matter more than any single data point on it.


A related scoring note, from a different publisher of a similar measure. Acadience Learning's published rule for its nonsense word fluency measure counts words read "without first being sounded out", where DIBELS 8 credits a word "whether sounded out initially or not."


Two one-minute measures of the same construct, scoring the same behaviour differently, which the piece on pseudoword decoding measures works through in full. Change across administrations means nothing if you switched publishers between them.

What the research says about the interval itself, with its boundary stated


The interval's effect on practice gains has been studied, and the evidence is adjacent to our question rather than on it.


The largest synthesis is Calamia, Markon and Tranel, "Scoring higher the second time around: Meta-analyses of practice effects in neuropsychological assessment", The Clinical Neuropsychologist 26(4), 543–570, DOI 10.1080/13854046.2012.680913, with records at LSU and the University of Minnesota. It covers nearly 1,600 individual effect sizes, and reports that the length of the test-retest interval was associated with the magnitude of change in many cases.


Read the boundary in the same breath as the finding. That is a meta-analysis of neuropsychological measures — memory, attention, executive and intellectual tasks — administered largely to adults in clinical and research samples.


It is not a study of academic achievement batteries administered to school-age students by educational therapists. What it licenses is a general expectation: the interval is one of the things that determines the size of a practice gain, so an interval you do not record is a variable you have not controlled.


What it does not license is a correction. There is no table in this literature that converts "re-tested at eleven weeks" into points to subtract from a standard score, and anybody offering you one on an instrument-specific basis should be asked where it came from. The same discipline applies here as in the broader question of whether educational therapy itself is evidence-based: an adjacent finding stays adjacent.

What the federal practice guide says about any of this


The method this site uses on every governing document: count the words. The result here is a clean zero, and it took one correction to get right.



  • "practice effect" and "practice effects" — zero

  • "interval" and "intervals" — zero

  • "retest" — zero

  • "alternate forms", "equivalent forms" — zero

  • "reassessment", "reevaluation" — zero


Against that, "progress monitoring" appears three times and the word "twice" fourteen times.


The correction is the interesting part, and it is a reminder about method. A naive substring search returns three hits for "retest" and three for "alternate form", which looks like a finding.


Reading the matching lines shows that all three "retest" hits are the word pretest, inside the guide's methodological appendix on quasi-experimental designs, and all three "alternate form" hits are Alternate Formats, from the accessibility notice on the copyright page offering the publication in Braille or large print. Both would have been reported as findings by anybody who did not look.


So: the federal guide that makes recommendations about teaching foundational reading, across nearly sixty thousand words, says nothing about re-testing intervals or practice effects. That is not a criticism — it is an instructional guide, not a measurement manual. It is simply worth knowing that the document practitioners are most often pointed to for authority on reading instruction contains no guidance on the measurement decision they make every term.


The same guide's silences on other subjects are catalogued in the pieces on reading programme evidence ratings and decodable text criteria.

AET's Code is silent again, for the ninth post running


The Association of Educational Therapists' Code of Ethics runs to 2,978 words. Counted as whole words, it contains:


  • "retest" and "re-test" — zero

  • "interval" — zero

  • "reassessment", "reevaluation", "re-evaluation" — zero

  • "test", "tests" and "testing" — zero


Against "assessment", which appears seven times. The AET Fact Sheet runs to roughly 761 words of readable page text and carries "retest", "interval" and "reassessment" at zero, with "tests" appearing once inside its list of activities an educational therapist does not perform unless otherwise qualified.


This is now the ninth consecutive post on this site to find the Code silent on the subject the post is about. The pattern is consistent enough to be worth stating as a general fact about the credential rather than as a complaint: AET's Code governs conduct, confidentiality, records, scope and the professional relationship, and it does not govern measurement practice.


If you want to know whether re-administering an instrument after six weeks is defensible, the Code will not tell you, and the people who wrote it did not set out to. What the Code does govern, clause by clause, is covered in the pieces on informed consent and what belongs in a session note.

The question none of this answers


Every source above addresses the inputs. None addresses the output, which is the only thing the family asked about: is this gain real?


That question is not answered by an interval, a practice-effect figure or a form count. It is answered, as far as it can be, by comparing the size of the change against the measurement error of the instrument — the band, not the interval. The standard-error piece on this site works through why that band is wider than practitioners expect and why it is a setting rather than a finding.


And there is a boundary here that matters more than any of the technical detail. Nothing in this piece supports a claim about whether a particular student has or does not have a disorder, and a change score is not diagnostic evidence either way. Qualification level governs what you may administer; scope of practice governs what you may conclude; and the two are not the same question.


The boundary between administering an instrument and interpreting it diagnostically is the subject of the piece on psychoeducational evaluation for educational therapists, and it does not move because a second administration has happened.

A sequence for deciding, not a decision


What follows is an order of operations. It deliberately does not tell you how long to wait, because no source read for this piece does.


  1. Read your own manual's reliability chapter before you schedule anything. Not the product page. The chapter, for the interval its own study used. That interval is the only one the publisher has evidence about.

  2. Count your forms. If the instrument has one form, a re-test is a same-form re-test and the Standards say the correlation can be inflated by recall. If it has two, you get one clean re-administration and then you are back to the same-form problem.

  3. Decide whether you need the battery at all. If the question is "is the intervention working", a measure built for repeated administration — twenty forms, published decision rules — answers it better than a battery built for one-off comprehensive assessment. The two instruments are not substitutes.

  4. Write the interval down, in days, in the file. Standard 13.2 requires reporting the time period between administrations for change scores. The same obligation is sensible for an individual case even where no standard compels it.

  5. Compare the change to the band before you compare it to anything else. If the gain does not exceed the measurement error, the interval question never arises.

  6. Do not report a corrected score. There is no published correction factor for practice effects on these instruments at these intervals. Reporting the obtained score, with the interval and the form stated, is honest. Adjusting it is not.

  7. Say which form you used, both times. A change from Form A to Form B carries a form effect. A change from Form A to Form A carries a memory effect. The reader of your report cannot tell which without being told.

What to document


Short list, and all of it is the kind of thing that costs nothing at the time and is impossible to reconstruct later:


  • date of each administration, and the elapsed interval in days

  • form administered on each occasion, by letter

  • which edition and norms, since norms age and a re-test against different norms is not a re-test

  • the interval the manual's own reliability study used, and the page it is on

  • whether any intervening instruction targeted the specific items or task format

  • the obtained scores and their confidence bands, not a derived gain alone

What is out of scope here


This is a survey of what test publishers and the Standards publish about re-test intervals and practice effects. It is not legal advice and nothing here should be read as describing a legal obligation.


Re-evaluation timelines for students receiving special education services are set by statute and regulation, not by test manuals, and they are a separate question from the psychometric one — a different clock, run by different people, for different reasons. If your question is about a required re-evaluation cycle, the answer is with the school district and, where it matters, an attorney. Nothing above addresses it.


Score reporting to third parties, and what may leave your office, is also out of scope and is partly governed by publisher agreements rather than by professional codes.

Where the published record runs out


Stated plainly, because these gaps shaped what this piece could say:


  • No publisher read for this piece states a minimum re-test interval for a comprehensive achievement battery. Not Pearson, not Riverside, not PRO-ED. The closest thing is the TOWRE-2's three-or-four-times-a-year frequency.

  • The product pages that publish test-retest coefficients do not publish the intervals behind them. That is true of the TOWRE-2 and the GORT-5 pages, and Standard 2.19 says it should not be.

  • The WIAT-4 and KTEA-3 technical manuals are not publicly readable, so the WIAT-4 interval quoted here comes from an independent review rather than from Pearson, and the KTEA-3's interval could not be established at all.

  • The CTOPP-2's test-retest evidence is not on either seller's page. It may be strong. This piece cannot say.

  • There is no instrument-specific practice-effect correction in the published literature for school-age achievement measures at practitioner-relevant intervals, and the nearest large synthesis is in a different population and a different domain.

  • Riverside's own practice-effect data has a median interval of under an hour, which is reported transparently and still does not answer the question a re-test raises.


If any of these matters to a decision you are making, the route is the manual on your own shelf and, failing that, an email to the publisher. That is a slower answer than a blog post and it is the only one with your instrument's name on it.

Frequently asked questions


How long should I wait before re-administering an achievement test?


No publisher read for this piece states a minimum. The Standards for Educational and Psychological Testing assign that policy to the test developer or test user and require that whatever interval is used in a reliability study be reported. In practice this means reading your own manual's reliability chapter for the interval its study used, and treating anything shorter than that as unevidenced.


What is a practice effect, exactly?


Riverside's WJ V Technical Abstract defines it as the impact of prior exposure to a test task on later performance, "even when the content of the individual items is different between the two forms", and notes it is also called a carryover effect. It is reported as Cohen's d, where a positive value means a higher mean score on the second administration.


Does using the alternate form remove the problem?


It removes one problem and introduces another. A second form reduces item recall, but the WJ V abstract reports a practice effect of 0.57 on Rapid Phoneme Naming across different forms, so exposure to the task format still matters. And DIBELS 8 warns that equated forms still differ in difficulty, giving its own example of a 20 words-per-minute difference that might be the passage rather than the student.


How many alternate forms should an instrument have?


The only published threshold comes from the National Center on Intensive Intervention, whose academic progress monitoring rubric requires at least twenty alternate forms with strong comparability evidence for a full bubble. That standard applies to progress-monitoring tools. Most achievement batteries have two forms, and the TOWRE-2's four is the highest count among the instruments surveyed here.


Can I subtract the practice effect from the second score?


No published correction factor exists for these instruments at the intervals practitioners use. The practice-effect figures in the WJ V abstract come from administrations separated by a median of under an hour and cannot be applied to a re-test months later. Report the obtained score with the interval and form stated rather than an adjusted one.


Is a gain larger than the practice effect a real gain?


Not necessarily, and this is the wrong comparison to make first. The first comparison is against the instrument's measurement error at that score level, which is what a confidence band expresses. A change that does not exceed the band is not interpretable as change regardless of how long you waited.


Does AET's Code of Ethics say anything about re-testing?


No. The Code contains no instance of "retest", "interval", "reassessment" or even the word "test" as a whole word. It governs conduct, confidentiality, records and scope, not measurement practice.


Does any of this change for remote administration?


The interval question is unchanged, but the administration conditions are not, and publishers have published separately on what they permit remotely. That record is surveyed in the piece on remote assessment administration. A score obtained in person and a score obtained remotely differ in more than the interval between them.

Related reading


Join the Illuminate educator network


This piece is part of a practitioner library built for educational therapists, learning specialists and executive function coaches — instrument surveys, credential comparisons and practice operations, sourced to the issuing bodies and the publishers themselves.


Join the Illuminate educator list and new pieces arrive as they publish.

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