Effectiveness of Treatment Approaches for Children and Adolescents with Reading Disabilities: A Meta-Analysis of Randomized Controlled Trials
Galuschka, K., Ise, E., Krick, K., & Schulte-Korne, G. · 2014
grade Bmeta-analysisindependentreplicatednumbers spot-checked
Sample
22 RCTs / 49 experimental-vs-control comparisons; 1,138 participants in experimental groups and 764 in control groups (1,902 total). 18 of the 49 comparisons (10 trials) also measured spelling.
Population
Children/adolescents with reading disabilities (reading below the 25th percentile or >=1 SD/year/grade below expectation, IQ >= 85, poor reading in mother tongue). No adult studies qualified. English, Spanish, Finnish, Italian and Brazilian samples pooled; the authors flag untested cross-orthography transferability.
Design
RCT-ONLY meta-analysis comparing treatment approaches (phonics, phonemic awareness, reading fluency, comprehension, auditory training, medication, coloured overlays). Random-effects model, Hedges g, effects corrected for pre-test differences >= 0.20. Outcome measures are transfer measures only — self-constructed measures matching training content were excluded by design, and 19 of 22 trials used standardized tests. Quality caveats the authors themselves report: allocation concealment was inadequately described in ALL 22 trials, 16 did not state whether outcome assessment was blinded and 2 stated it was unblinded; only 4 trials used blind assessment. Controls were untrained groups or placebo-training groups. No follow-up measures anywhere in the meta — every estimate is at post-test, so this source says nothing about durability.
Key findings
In an RCT-only meta for reading disability, phonics was the only treatment approach whose efficacy was statistically confirmed (reading g = 0.322; spelling g = 0.336); every alternative was non-significant. Three qualifications from the full text that the abstract-level record missed, all of which cut against a strong reading. (1) The trim-and-fill publication-bias correction cuts the phonics reading effect to g = 0.198 (95% CI 0.039-0.357) — still significant, but half the headline size and barely above zero; the funnel plot was asymmetric and the authors assume the same bias affects the other approaches. (2) The formal between-approach test is null (Q = 3.164, df = 6, p = 0.788): phonics is not statistically superior to fluency, phonemic-awareness, auditory or even coloured-overlay training. Its unique "confirmed" status comes from having 29 comparisons against their 2-5, i.e. from statistical power, not from a demonstrated advantage. (3) Effects delivered by the study author (g = 0.806) are ~3x those delivered by classroom teachers (g = 0.247) — the efficacy-to-effectiveness decay, measured inside this meta. Also: no follow-up data exist anywhere in the 22 trials, so nothing here speaks to durability.
Genetic confound
RCT design controls selection/passive rGE.
Replication notes
Anti-refutation anchor: preserves the phonics verdict while capping its magnitude at small. The paper states its phonics result is consistent with McArthur et al. 2012 (Cochrane review of phonics training for poor readers) and Ehri et al. 2001 (NRP systematic-phonics meta) — hence "replicated" rather than the previous "unreplicated", which asserted more than the evidence allows. Caveat: those syntheses draw on an overlapping set of primary trials, so this is convergence between meta-analyses of the same literature, not independent replication.
DOI / URL
10.1371/journal.pone.0089900
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Phonics instruction, reading (all 22 trials) | Hedges g | 0.322 (95% CI 0.177-0.467), 29 comparisons, I2 = 0 — the only approach whose efficacy was statistically confirmed (Table 1) | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Phonics instruction, reading — publication-bias-corrected | Hedges g (trim and fill) | 0.198 (95% CI 0.039-0.357) after Duval & Tweedie trim-and-fill imputed 10 missing studies; the funnel plot was asymmetric with a gap on the left (Table 4). Per this archive's effect-size rules this corrected value, not 0.322, is the headline phonics estimate for reading-disabled children | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Phonics instruction, reading — standardized measures only (19 trials) | Hedges g | 0.424 (95% CI 0.246-0.601), 25 comparisons. Effects were HIGHER, not lower, when the 3 trials using non-standardized transfer measures were dropped — the opposite of the usual measure-type inflation pattern. Not trim-and-fill corrected | standardized | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Phonics instruction, spelling | Hedges g | 0.336 (95% CI 0.062-0.610), 10 trials, I2 = 22%; 0.376 (0.065-0.686) with standardized measures only. The only approach for which a spelling mean could be computed at all | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Differences BETWEEN treatment approaches (the comparison the headline implies) | subgroup Q test | Q = 3.164, df = 6, p = 0.788 — no statistically significant difference between any of the seven treatment approaches (Table 1). Phonics is not shown to beat the alternatives; it is the only arm with enough comparisons (29 vs 2-5) for its CI to exclude zero | mixed | end-of-treatment | active-alternative | end-of-treatment | domain-skill |
| Phonemic awareness instruction, reading | Hedges g | 0.279 (95% CI -0.244 to 0.802), 3 comparisons — not significant. Authors note PA training is effective for preschool at-risk children but does not transfer to already reading-disabled school-age children | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Reading fluency training, reading | Hedges g | 0.301 (95% CI -0.105 to 0.707), 5 comparisons — not significant; 0.280 with standardized measures only (4 comparisons). Repeated-reading practice alone is not an effective remediation for reading disability | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Reading comprehension training, reading | Hedges g | 0.177 (95% CI -0.181 to 0.535), 3 comparisons — not significant. All three comparisons came from a single author, so this null is very weakly evidenced | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Auditory training, reading | Hedges g | 0.387 (95% CI -0.065 to 0.838), 3 comparisons — not significant. Treatments aimed at putative underlying causes rather than at literacy itself were not confirmed | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Medical treatment (piracetam), reading | Hedges g | 0.125 (95% CI -0.072 to 0.322), 2 comparisons — not significant. Authors conclude the side-effect risk outweighs the benefit | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
| Coloured overlays / Irlen lenses, reading | Hedges g | 0.316 (95% CI -0.012 to 0.644), 4 comparisons — not significant. Small effects against untrained controls, negligible against placebo controls, which the authors read as a placebo effect | mixed | end-of-treatment | unclear | end-of-treatment | domain-skill |
| Who delivered the intervention (independence discount, measured) | Hedges g by conductor | reading — study author 0.806 (0.397-1.215) vs teacher 0.247 (0.046-0.449) vs special-education therapist 0.256 (0.090-0.422), p = 0.088; spelling — student 0.945 (0.417-1.474) vs teacher 0.099 (-0.412 to 0.610) vs therapist 0.148 (-0.082 to 0.378), p = 0.021. Effects delivered by ordinary teachers are roughly a third the size of those delivered by the researcher | mixed | end-of-treatment | business-as-usual | end-of-treatment | domain-skill |
Cited by
- Systematic/explicit phonics vs whole language and balanced literacystrong supportconf: highgc: low