Targeted academic interventions for pupils with special educational needs and disabilities raise reading, writing and maths results at the end of the intervention compared with teaching as usual or active controls (about 0.3 SD for reading and 0.4 for maths on standardised tests; larger on researcher-made tests), but whether those gains last is not shown. Reading and writing gains are similar in mainstream and special settings; maths gains have been shown in mainstream settings but not in special schools.
Status
weakened
Last change
2026-11-07
Serves
Q1 (How do humans actually learn, and what helps them learn best?)
Ages
grundskola-lag, grundskola-hog, overgripande
Area
Special education and inclusion
Relates to
C31 relates-to C32 (special needs: where vs what is taught)
Bounds
Targeted, structured academic interventions; not placement (C31). Evidence on file: Van Herwegen, Masterman, Dockrell, Marshall, Gordon & Thomas 2026 (Review of Education, DOI 10.1002/rev3.70127; abstract, open access; full text read 2026-11-01, see below): overall g = 0.44 (95% CI 0.38 to 0.50); positive for reading, maths and writing; very high heterogeneity (I² ≈ 93%); reading and writing gains similar in mainstream and special settings, maths larger in mainstream. Stockard et al. 2018 (Review of Educational Research; adjacent): structured Direct Instruction curricula show positive academic effects. Weak spot: published-study bias and small samples could inflate the average. Weakened 2026-11-01 (Hall, Dahl-Leonard, Cho, Solari, Capin et al. 2023, Reading Research Quarterly 58(2), 285–312, DOI 10.1002/rrq.477; PMC full text): 53 studies (77% RCTs, N = 6,053) of reading interventions for K–5 pupils with or at risk for dyslexia, standardised norm-referenced outcomes only: g = 0.33 (95% CI 0.25 to 0.41); grades 3–5 g = 0.16 vs K–2 0.36 (not significant); no publication bias. Van Herwegen et al. 2026 full text now read (UCL Discovery): reading 0.36, maths 0.51, writing 0.43 overall; outcome type moderated reading (p < .001) and maths (p = .02) — researcher-made measures reading 0.54 / maths 0.77, standardised reading 0.27 / maths 0.40; reading high-quality studies 0.32; maths in mainstream 0.65 with no significant positive effect in special schools (46 outcomes, including pupils with Down syndrome (7 outcomes) and severe learning difficulties (4); Van Herwegen et al. 2026 p. 23); writing trim-and-fill 0.36 → 0.32; control type not a moderator (about a quarter active controls). Time bound (weakened 2026-11-07, point (b) of the closer reading): Van Herwegen et al. 2026 analysed "only short-term outcomes" (p. 6) and only 13.5% of its studies reported any follow-up (p. 32); Hall et al. 2023 excluded follow-up-only reports and reports no follow-up analysis — so the pooled gains are end-of-intervention effects; durability is not shown.
What would kill this
A large meta-analysis of randomised and quasi-experimental studies finds an average gain below about 0.2 standard deviations, or none, once study quality is accounted for. Promoted from K61 on 2026-10-29 in the bold expansion Jonatan approved after four empty attack nights (26–29 Oct): new sharp claims in thin areas, each naming a comparison, a rough size or direction and a population.
Tutor consequence
none yet
Attack record
2 attacks: 2 narrowed
2026-11-01 narrowed
Attack
A meta-analysis of 53 reading-intervention studies for primary pupils with or at risk for dyslexia (77% RCTs), using only standardised norm-referenced tests, found an average gain of g = 0.33, and the full text of C32's own source shows standardised-test gains of 0.27 for reading and 0.40 for maths, with no shown maths gain in special schools — so "roughly 0.4 SD … in mainstream and special settings alike" overstates what independent tests show.
Target
claim-text
Source
Hall, Dahl-Leonard, Cho, Solari, Capin, Conner, Henry, Cook, Hayes, Vargas, Richmond & Kehoe 2023 | Reading Research Quarterly 58(2), 285–312, DOI 10.1002/rrq.477 (PMC author manuscript, full text read 2026-11-01; Crossref-checked); with Van Herwegen et al. 2026, Review of Education, DOI 10.1002/rev3.70127 (full text read 2026-11-01, UCL Discovery)
Steelmans to
C32's 0.4 is the pooled estimate across every kind of test; researcher-made tests aligned to the intervention are known to inflate effects, and the strictest available synthesis for the largest SEND group (standardised tests only, mostly RCTs, no publication bias) lands at a third of a standard deviation, as does the reading estimate on standardised tests in the claim's own source; the "alike" clause is contradicted for maths by the same source.
Why
Nightly-attacks 2026-11-01 Europe/Stockholm. Hall et al. 2023 full text: g = 0.33 (95% CI 0.25 to 0.41), I² = 63%, modified Egger p = .89; grades 3–5 g = 0.16 vs K–2 0.36 (ns); comprehension smaller than word reading (b = −0.080, p = .024); design and sample size not significant moderators. Van Herwegen et al. 2026 full text: overall 0.44 (0.38–0.50); reading 0.36, maths 0.51, writing 0.43; outcome type moderated reading (p < .001) and maths (p = .02): experimental measures 0.54 / 0.77, standardised 0.27 / 0.40; reading high-quality 0.32; maths in mainstream 0.65, no significant positive effect in special schools (46 outcomes), mixed or clinical settings; reading (p = .21) and writing (p = .71) did not differ by setting. Kill withheld: every quality-filtered estimate is still about 0.3 or more, well above the 0.2 kill line, and the 2023 meta found no publication bias. Not a duplicate (empty ledger); not a restatement (the claim text said "alike" and "roughly 0.4" for all three subjects; the abstract-only Bounds did not split by test type). Not a preprint. Status alive → weakened. Claim field and Bounds rewritten in the same turn; brief and diagram core rewritten (C32 has no front-page rule).
Rewrite
claim-field+brief
2026-11-07 narrowed
Attack
Both of C32's own full-text sources measure results only at the end of the intervention — Van Herwegen et al. 2026 analysed "only short-term outcomes" and only 13.5% of its studies had any follow-up, and Hall et al. 2023 excluded follow-up-only reports and reports no follow-up analysis — so the claim's "raise" with no time bound overstates what is shown.
Target
claim-text
Source
Van Herwegen, Masterman, Dockrell, Marshall, Gordon & Thomas 2026 | Review of Education, DOI 10.1002/rev3.70127 (stored full text, pp. 6 and 32); Hall, Dahl-Leonard, Cho, Solari, Capin et al. 2023 | Reading Research Quarterly 58(2), DOI 10.1002/rrq.477 (stored full text; follow-up-only reports excluded)
Steelmans to
C32 asserts that targeted interventions "raise" results with no time window; every pooled figure already on the claim is an immediate post-test, and the largest SEND meta itself flags the near-absence of follow-up — so a lasting-gain reading is not supported.
Why
Nightly-attacks 2026-11-07 Europe/Stockholm. A lead from a closer reading of the study, verified tonight in the stored full texts: Van Herwegen p. 6 "only short-term outcomes were included in the analysis"; p. 32 "Only 13.5% of studies reported follow-up measures in addition to post-test"; Hall exclusion list names follow-up-only reports. Kill withheld: the end-of-intervention effect is still well above 0.2. Not a duplicate of the 2026-11-01 weaken (that split size by test type and settings; no time bound was added) and not a restatement of point (a) of the closer reading (factual Down-syndrome/SLD count, applied 2026-11-02). Not a preprint. Status stays weakened. Claim already over ~600 characters, so rewritten short in the same turn under the short-claim rule (Hall 0.27/0.33 detail already in Bounds; durability clause added). Brief and diagram core rewritten (C32 has no front-page rule).
Rewrite
claim-field+brief
Corrections to our notes
These fix a number or a fact in our notes. They are not attacks and do not count toward the floor of 3 survivals.
— C32's limits describe the 46 special-school maths outcomes as "mostly pupils with Down syndrome or severe learning difficulties"; Van Herwegen et al. 2026 names 7 Down syndrome and 4 severe-learning-difficulty outcomes (p. 23) (a closer reading of the studya).
Meta-analysis of 42 maths interventions for students with learning disabilities: most components significant, mean effects 0.21–1.56; explicit instruction and heuristics strongest; student goal-setting feedback and within-class peer tutoring not significant. Per-component, not a pooled average; abstract only.
Standardised-only reading interventions for K–5 dyslexia-risk: g = 0.33; excluded follow-up-only reports; no follow-up analysis. Used for C32 size weaken 2026-11-01 and time-bound weaken 2026-11-07.