What selective schooling and test preparation demonstrably change, and what they don't. 22 verified findings — quoted verbatim, fully cited, independently checked (register v1.56.2, verified 2026-09-16).
← All domains, the protocol, and the register’s through-lines
Start with the finding that's easiest to skip past: once you account for who gets in, attending a selective school adds little or nothing to a student's achievement. This isn't a fringe result. The strongest causal evidence internationally comes from regression-discontinuity studies of Boston and New York exam schools, which exploit the sharp admissions cutoff to compare students just above and just below the line — students who are, for all practical purposes, identical except for which school they attend. The result: "little effect of exam school offers on most students' achievement," and the authors conclude the intense competition for seats "does not appear to be justified by improved learning for a broad set of students" [TST-01]. Australia's own research points the same way. The first regression-discontinuity and matching study of an Australian selective system, using Victorian data, found only "small positive effects at best" on university entrance results once students' existing high achievement and aspiration were accounted for [TST-02]. The first such study for New South Wales — the country's oldest and largest selective system — found offers had "only scattered and mostly insignificant impacts on overall student achievement," though low-SES students who did get in seemed to benefit more, while remaining underrepresented in the intake to begin with [TST-03]. The common thread across Boston, New York, Victoria and NSW is that the huge gap in average achievement between selective and non-selective schools is mostly explained by who walks through the door, not by anything the school subsequently does. One UK study, using a historical grammar-school assignment formula, did find a genuine elite-school effect on years of completed education — but even there the downstream effects were uneven and not a straightforward achievement bump: gains in income for women, none for men, and a fertility cost for women, which is a different kind of finding altogether, not evidence that the school made students better at exams [TST-04]. In Victoria today, roughly 1,000 Year 9 places are contested across four selective schools each year [TST-05], and nationally, as of 2015, 25 selective high schools and a further 70 select-entry programs operated across four states [TST-09] — a lot of family effort and money chasing a prize that, on the best available evidence, is substantially an illusion of the sorting process itself.
If the school itself doesn't do much, does preparing for the entrance test do anything? Here the evidence is more encouraging, but modest, and it discriminates sharply between good and bad preparation. A 2025 meta-analysis of 28 experimental and quasi-experimental studies found that structured test preparation does significantly improve scores on large-scale educational tests, but the effect is small to moderate — an effect size of g=.26 — and, crucially, it is not driven by simply doing lots of practice items. The gains came from workbook use, training in socio-affective strategies (managing anxiety, pacing, confidence), and explicit instruction in test-taking skills; the study found "little evidence of practice effect" from sample items and practice tests alone [TST-06]. A separate, older but well-established meta-analysis in the personnel-selection literature — 50 studies, over 134,000 participants, on retesting with cognitive-ability-style tests, the kind of reasoning test that selective-entry and scholarship exams most resemble — found essentially the same quarter-standard-deviation gain (g=.26) from taking a similar test twice, with larger gains when practice was combined with actual coaching or when identical test forms were reused [TST-07]. US research on SAT coaching, using propensity-score matching to correct for the fact that families who seek coaching differ systematically from those who don't, found real but modest gains — roughly 11 to 15 points on maths, 6 to 9 on verbal — and, tellingly, those gains were substantially larger for higher-SES students than for lower-SES students using the same kind of preparation (15 vs 5 points on maths; 9 vs 2 on verbal) [TST-08]. Put together: coaching works, a bit, mostly through quality of preparation rather than sheer volume of drilling, and the families with more resources tend to extract more benefit from it than families with less — which is worth sitting with rather than glossing over.
The third thread is what high-stakes testing does to teaching itself, and here the Australian evidence is direct and uncomfortable. A widely cited qualitative synthesis of 49 studies found the dominant, most common effect of high-stakes testing on classrooms is that curriculum narrows to what's tested, knowledge gets fragmented into test-sized pieces, and pedagogy shifts toward teacher-centred instruction — though a significant minority of tests, depending on how they're structured, produced the opposite: curriculum expansion and more student-centred teaching [TST-10]. Australia's own 2020 independent NAPLAN Review reached a strikingly similar and specific conclusion about writing: restricting the writing test to narrative and persuasive genres, with the genre often announced in advance, "led to very formulaic writing in students' responses to the prompt and, as a further unintended consequence, to very formulaic teaching of writing in some schools as they seek to prepare students for the NAPLAN writing test" — a finding serious enough that the review recommended withdrawing the writing test from census testing altogether [TST-11]. The review's stakeholder testimony makes the mechanism concrete: teachers described being coached into "gaming" the marking criteria, producing what one respondent called "cookie cutter" writing, and the panel's own analysis linked this directly to test preparation displacing the planned curriculum in the lead-up to testing — NAPLAN prep becoming, in the report's words, "the proxy curriculum" [TST-12]. This is the washback effect in its plainest form: when a test rewards a formula, schools teach the formula, and something real is lost in the process.
Put the three threads together and the honest picture for a family weighing selective-entry or scholarship testing is this. The school itself is very unlikely to be the transformative factor everyone assumes it is — the achievement gap between selective and comprehensive schools is mostly the students who were already there, not something the building or the label adds. Coaching for the entrance exam can move the needle, but only modestly, and only when it's built around real skill development — strategy, structure, confidence, exam craft — rather than repetition for its own sake; volume of practice items is close to worthless on its own. And the surrounding culture of high-stakes testing has a well-documented tendency to narrow teaching toward formula, which is precisely the trap a family should want their child to avoid, not walk into. None of this is an argument against preparing. Confidence going into an exam matters, access to selective pathways matters for families who want them, and a quarter-standard-deviation gain is real value when a single test is the gate. But it is an argument for being honest about what preparation actually buys: not a guaranteed changed trajectory, and not best pursued through drilling more of the same formulaic material a high-stakes test tends to reward. The durable asset a student carries out of a genuine test-preparation period is the literacy and reasoning skill built along the way — the actual capacity to read closely, argue clearly, and write well under pressure — which persists long after the specific exam is behind them, whether or not they get the placement. That is the case for teaching craft rather than test tricks: not because test tricks don't work at all, but because the evidence says they barely do, while the underlying skill is what was worth building regardless of the outcome on the day.
US regression-discontinuity research on elite exam-school admissions (Boston and New York) finds little to no achievement gain from attending, despite the schools' large peer-achievement gap — the perceived superiority is largely explained by intake selection, not school effects.
Our estimates show little effect of exam school offers on most students' achievement. ... the intense competition for exam school seats does not appear to be justified by improved learning for a broad set of students.
The first regression-discontinuity and matching-based estimates of the causal effect of attending an academically selective high school in an Australian state (using Victorian and other state data) find only small positive effects at best on university entrance results, once students' pre-existing high achievement and aspiration are accounted for.
The estimates point to small positive effects at best on university entrance results from attending the selective schools. Overall, the small selective school effect appears to reflect the high levels of educational aspiration of both selective students as well as applicants who attended other schools.
UK evidence using an instrumental-variables design around a historical elite/non-elite secondary school assignment formula finds attending an elite school produced large gains in completed education, but the pattern of effects on later-life labour market outcomes was uneven by sex (positive income effects for women, none for men) and came with a significant fertility cost for women — i.e. the causal gains were not simply general achievement gains.
Using instrumental variables methods that exploit the school assignment formula, we find that elite school attendance had large impacts on completed education. For women, we find that elite school attendance generated positive effects on labor market outcomes and significant decreases in fertility; for men, we find no elite school impacts on any of these later-life outcomes.
Victoria has four academically selective-entry government high schools (Melbourne High School, The Mac.Robertson Girls' High School, Nossal High School, Suzanne Cory High School), with approximately 1,000 Year 9 places offered across the four schools each year.
Victoria has four selective entry high schools that provide a specialist environment for academically high-achieving students in Years 9 to 12. ... Each year approximately 1,000 places are available for Year 9 entry across the four selective entry high schools (SEHS).
A 2025 meta-analysis of 28 experimental/quasi-experimental studies finds that structured test-preparation interventions produce a small-to-moderate, statistically significant improvement in students' scores on large-scale educational tests, with the effect driven mainly by workbook use, socio-affective strategy training, and explicit test-taking-skills instruction — not by mere repeated exposure to practice items.
Results from 28 included studies suggested that students' test performance can be significantly improved by test preparation (g = .26, 95% CI = .10-.42). Using workbooks, developing students' socio-affective strategies, and teaching test-taking skills were found to be significant moderators of the test preparation effect. Little evidence of practice effect (i.e., using sample items and practice tests) on test preparation was found in the current study.
A meta-analysis of 50 studies (107 samples, ~134,000 participants) on retesting with cognitive-ability-style tests finds an average score gain of about a quarter of a standard deviation (g=.26) simply from taking the same style of test twice, and that this gain is larger when retesting is combined with explicit coaching or when identical test forms are reused — evidence directly relevant to ability-style selective-entry and scholarship exams, which resemble cognitive ability tests more than curriculum-based achievement tests.
This study uses meta-analysis to summarize the results of 50 studies of practice effects for tests of cognitive ability. Results from 107 samples and 134,436 participants revealed an adjusted overall effect size of .26. Moderator analyses indicated that effects were larger when practice was accompanied by test coaching, and when identical forms were used.
As of 2015 there were 25 academically selective high schools across four Australian states (New South Wales, Queensland, Victoria and Western Australia), plus a further 70 schools offering select-entry programs for a subset of classes rather than the whole school.
In 2015, there were 25 selective high schools across the states of New South Wales (NSW), Queensland, Victoria and Western Australia. Most of these schools admit students based on entrance exams or tests... There were also 70 other high schools in these same states offering select-entry programs, where one or more classes of students (as compared with all students) are admitted in a similar process involving tests and interviews.
Wayne Au's qualitative meta-synthesis of 49 studies on high-stakes testing finds the dominant effect is that curriculum content narrows to tested subjects, knowledge gets fragmented into test-related pieces, and pedagogy shifts toward teacher-centred instruction — though a significant minority of cases showed the opposite (curriculum expansion, knowledge integration, more student-centred pedagogy), with the direction of the effect depending on how the test itself is structured.
the primary effect of high-stakes testing is that curricular content is narrowed to tested subjects, subject area knowledge is fragmented into test-related pieces, and teachers increase the use of teacher-centered pedagogies. However, in a significant minority of cases, certain types of high-stakes tests have led to curricular content expansion, the integration of knowledge, and more student-centered, cooperative pedagogies.
Australia's 2020 independent NAPLAN Review found that the restriction of the NAPLAN writing test to narrative and persuasive genres (with the genre often announced in advance) led to very formulaic student writing and, as an unintended consequence, formulaic teaching of writing in some schools as they prepared students for the test — one of the review's central criticisms, leading to a recommendation that the writing test be withdrawn from census testing and redeveloped.
The restriction of the writing genres to narrative and persuasive, with the specific genre being announced in advance in the early years of NAPLAN, has led to very formulaic writing in students' responses to the prompt and, as a further unintended consequence, to very formulaic teaching of writing in some schools as they seek to prepare students for the NAPLAN writing test.
The 2020 NAPLAN Review documented direct classroom testimony that teachers were coached into 'gaming' the writing marking criteria to maximise scores, describing the practice as reducing writing pedagogy to a formula ('cookie-cutter' writing) rather than genuine composition, and linked this to a broader shift where NAPLAN test preparation itself displaced planned curriculum in the weeks before testing.
Where this displacement of curriculum occurred, NAPLAN test preparation became the proxy curriculum and teaching writing was reduced to a dominant focus on the structure of the writing, the formula for producing the narrative or the persuasive form, with some commentary about maximising the score on writing through 'gaming' the criteria. In commenting on classroom preparation of students for NAPLAN writing, one respondent said "You're forensically taking out different things, and you can formulaically produce a response that maximises the result" (Member of the NAPLAN Review Practitioners' Reference Group)... "The richness of writing has been lost. 'Cookie cutter' writing is being produced" (Subject association).
Across all three tracked-cohort growth transitions (Year 3-5, 5-7, 7-9) and all five NAPLAN domains -- 15 comparisons in total -- Victoria's average score gain confidence interval overlaps the national average's confidence interval every single time in the most recent cohort-gain data (2026, base 2024). Victoria's point-estimate gain is often numerically higher than the national average (e.g. Numeracy Year 7-9: Victoria 38.08 vs Australia 34.05), but the published confidence intervals mean none of these differences can be distinguished from the national average.
Cohort gain tab, 2026 calendar year, base 2024, All/All subgroup, computed overlap test on published Difference and Difference confidence interval columns for all 3 base-target year-level pairs x 5 domains = 15 comparisons: Reading 3-5 VIC=79.95+/-5.75 AUS=83.41+/-5.33 (overlap); Writing 3-5 VIC=73.87+/-8.34 AUS=65.86+/-8.13 (overlap); Spelling 3-5 VIC=87.35+/-4.20 AUS=87.43+/-3.78 (overlap); Grammar 3-5 VIC=90.93+/-7.58 AUS=88.28+/-7.14 (overlap); Numeracy 3-5 VIC=91.07+/-4.93 AUS=89.92+/-4.41 (overlap) -- and the same overlap holds for all five domains at both the 5-7 and 7-9 transitions.
NAPLAN national results dataset (Cohort gain tab) · ACARA · 2026
The NAPLAN Individual Student Report sent to Australian parents and carers reports a child's result as a position on a proficiency scale and as one of four proficiency levels — Exceeding, Strong, Developing, or Needs additional support — not as a numeric scale score. ACARA's own guidance for parents describes the report as showing "a proficiency scale for each assessment area (numeracy, reading, writing, spelling, and grammar and punctuation) with a student's result shown as a black dot on the scale", positioned against the national average (a black triangle) and the middle 60% of students at that year level (a shaded band). The two-page parent FAQ contains no numeric score of any kind and does not use the term "scale score".
The report shows a proficiency scale for each assessment area (numeracy, reading, writing, spelling, and grammar and punctuation) with a student's result shown as a black dot on the scale. The report shows each child's achievement against the national average for their year (shown as a black triangle) and the range of achievement for the middle 60% of students in their year level / The proficiency standards include 4 proficiency levels: Exceeding: The student's result exceeds expectations at the time of testing. Strong: The student's result meets challenging but reasonable expectations at the time of testing.
The first regression-discontinuity study of NSW selective high schools (the oldest and largest selective system in Australia) finds that receiving an offer to attend a selective school has only scattered, mostly statistically insignificant effects on overall student achievement and participation in advanced coursework — while low-SES students who do get in appear to benefit more, yet remain underrepresented in the selective intake.
I find that receiving an offer to attend a selective school has only scattered and mostly insignificant impacts on overall student achievement and participation in advanced coursework. I do find suggestive evidence that selective schools benefit low socioeconomic status students, but that such students are typically underrepresented in selective schools, which has implications for Gifted and Talented education policy.
US SAT test-preparation research using propensity-score matching to correct for self-selection bias finds coaching produces modest score gains (roughly 11-15 points on math and 6-9 points on verbal), and that gains are larger for higher-SES students who use private tutoring or commercial coaching than for lower-SES students using the same preparation types — evidence that coaching effects on ability-style admissions tests are real but modest, and unevenly distributed by family resources.
For those students that have taken both the PSAT and SAT, effect estimates of roughly 11 to 15 points on the math section and 6 to 9 points on the verbal are found. Only the math effects are statistically significant. ... When coached students in the top SES quartile are compared to uncoached students in the top SES quartile, we find a mean difference of 15 points on SAT math scores. In contrast, amongst students in lower SES quartiles, coaching has only a 5 point effect. ... On the verbal section... the effect is 9 points for high SES students but just 2 points for lower SES students.
In ACARA's 2026 national NAPLAN results (all students, Australia, twenty domain-year cells), 'Strong' is the most common proficiency level in every cell, 'Strong or above' spans 53.9% to 73.9% of students, and 'Developing' alone holds between one in five and nearly three in ten students (17.3%-28.6% per cell) - a Developing result is common, not an outlier.
Computed from the dataset's per-level percentage columns for the Australia/All/2026 rows: e.g. Year 5 Reading - Needs additional support 7.7%, Developing 19.5%, Strong 53.4%, Exceeding 17.6%, Exempt 1.8%. Strong-or-above minimum 53.9% (Grammar and Punctuation, Year 3), maximum 73.9% (Writing, Year 3); Developing maximum 28.6% (Grammar and Punctuation, Year 9).
NAPLAN national results dataset (2026 release) · ACARA · 2026
Year 9 is where the 2026 national NAPLAN picture is weakest: the only two cells in the grid where 'Strong' falls below 40% are Year 9 grammar and punctuation (38.1%) and Year 9 writing (38.2%) - and writing pairs that with the grid's fastest-growing spread (national writing score standard deviation 72.7 at Year 3, 80.5 at Year 5, 90.9 at Year 7, 100.7 at Year 9) while participation declines from about 94% at Year 3 to about 89% at Year 9.
Dataset values, Australia/All/2026: Writing Year 9 - NAS 12.8%, Developing 25.6%, Strong 38.2%, Exceeding 21.6%; Grammar and Punctuation Year 9 - NAS 15.6%, Developing 28.6%, Strong 38.1%, Exceeding 15.9%; Writing SDs by year level 72.7/80.5/90.9/100.7; participation (Reading) 94.5/94.9/94.0/89.8.
NAPLAN national results dataset (2026 release) · ACARA · 2026
Checking the same CI-overlap test behind TST-15 against Indigenous vs non-Indigenous tracked-cohort growth in Victoria (15 comparisons: 3 transitions x 5 domains, All-students excluded), 14 of 15 overlap -- but Year 5 to 7 Numeracy does not: Indigenous students gained 46.5 +/- 8.6 points against non-Indigenous students' 61.4 +/- 5.9, a gap whose confidence intervals do not touch (55.1 vs 55.5).
Computed from the Cohort gain tab, Victoria, Indigeneity subgroup, 2026 calendar year base 2024: Year 5-7, Numeracy, Indigenous: Value=46.5 CI=8.6; Non-Indigenous: Value=61.4 CI=5.9. Upper bound of Indigenous range (55.1) sits below the lower bound of non-Indigenous range (55.5).
NAPLAN national results dataset (Cohort gain tab) · ACARA · 2026
A 2020 review commissioned for ACARA found that NAPLAN writing markers commonly apply the same scoring patterns across criteria -- 'rating tendencies' -- which produce superfluous information for distinguishing levels of writing capability, and recommended reducing the number of criteria where that occurs.
Evidence suggests that markers commonly use scoring patterns across criteria, which tend to produce superfluous information for the purposes of differentiating among levels of writing capability. A recommendation is made to reduce the criteria where there is evidence of superfluous information due to pronounced rating tendencies. ... Due to severity of rating tendencies for high-performing students, it is recommended that holistic score criteria are used for such performances on the authorial criteria (Audience, Text Structure, Ideas, Persuasive Devices and Vocabulary).
The 2020 review commissioned for ACARA states that sampling a single essay in one text type limits the generalisability of NAPLAN writing results and does not span the full range of text types in the Australian Curriculum: English -- and reports that every analytic approach it applied pointed strongly to local dependence in the data that none of its analyses could remediate.
Sampling a single essay in one text type limits the generalisability of the assessment results, and does not span the full range of text types in the Australian Curriculum: English. ... The quantitative analyses pointed clearly to two major findings. First, the frequency and use of score categories of the criteria is inconsistent, with infrequent categories occurring in nearly every criterion and year level. And second, every analytic approach consistently and strongly pointed to significant local dependence issues in the data that were not remediated by any analysis completed herein.
ACARA's NAPLAN Assessment Framework states that students do not all sit the same writing prompt: Years 3 and 5 respond to one of a small number of prompts for the same text type, Years 7 and 9 to one from a different small group, and different prompts may be used both within a year level and across year levels.
Students in Years 3 and 5 respond to one of a small number of prompts for the same text type. Students in Years 7 and 9 also respond to one prompt from a different small group of prompts. The prompts within each group are of similar difficulty. Different prompts may be used both within a year level and across year levels.
ACARA states that a NAPLAN scale score depends on both the number of items answered correctly AND the average difficulty of the items the student was assigned, so two students who answer the same number of items correctly receive different scale scores if one followed a less complex pathway through the adaptive test.
NAPLAN results for each student are based on both the number of items the student answers correctly and the average difficulty of the items that were assigned to the student. A student who completes a more complex set of items is likely to achieve a higher scale score. A student who answers the same number of items correctly, but follows a less complex pathway, is likely to achieve a lower scale score.
ACARA states that the NAPLAN scale was reset in 2023 and that results can be compared over time from 2023 onwards but not to years prior to 2023 -- a break in the scale itself, not only in the reporting categories. The Student and School Summary Report that carries this statement is explicitly not for distribution to parents and carers.
The NAPLAN scale was reset in 2023. Results can be compared over time from 2023 onwards, but not to years prior to 2023. ... The purpose of a Student and School Summary Report (SSSR) is to provide feedback on how students and cohorts of students within schools performed in NAPLAN tests. It is intended to be used by schools to inform teaching and learning programs, and is not for distribution to parents and carers.
Cite any finding by its anchor — for example #TST-01 — and the link resolves to the claim, quote and source above. The verification protocol lives on the register’s main page.