The Evidence on Teaching

Reconsidering the Evidence That Systematic Phonics Is More Effective Than Alternative Methods of Reading Instruction (+ 2021 reply to Fletcher)

Bowers, J. S. · 2020

grade Ccritiqueindependentnot-applicablenumbers spot-checked
Sample
Review of 12 phonics meta-analyses + England national data
Population
English-speaking children ages ~4-12; England state-school cohorts.
Design
Narrative/vote-counting review of all 12 phonics meta-analyses plus England national data (Machin et al. 2018 CLLD/ERDp evaluation, PIRLS 2001-2016, PISA 2018, KS1/KS2 SATs 2006-2018, Phonics Screening Check 2012-2018). Core estimand critique: the metas' controls conflate 'no phonics' with 'unsystematic phonics', so they test phonics-vs-none, not systematic-vs-the-incidental-phonics of real classrooms. Weakness (per Fletcher, Brooks): relies on significance vote-counting rather than CIs. NOTE, verified against full text: Bowers does NOT ignore the population-scale value of small effects — he explicitly rebuts Hammill & Swanson on exactly that point (p. 694, "these small effect sizes translate into real benefits when considering an entire population of children"). Full-text verification covers the 2020 article only; the 2021 reply to Fletcher was not in the retrieved PDF.
Key findings
The anchor skeptic paper, and stronger than a summary suggests. Concedes children MUST be explicitly taught GPCs, but argues no meta cleanly shows SYSTEMATIC phonics beats alternatives already containing some phonics, that benefits are decoding/word-level and fade, and that England's mandate produced no downstream reading gains. Its steelmanned core survives; its significance-vote-counting method is a genuine flaw. Crucially, NO critic produces an alternative that beats explicit code instruction. His conclusion is explicitly symmetrical: "the reading wars ... is best characterized as a draw", and the failure to support systematic phonics is NOT an argument for whole language. The specific meta-analytic re-readings he disputes, and which the effects rows now carry: NRP/Ehri 2001 (fadeout 0.41 to 0.27, null for below-average-cognitive struggling readers, whole-language subanalysis 0.31 with no RCTs), Camilli 2003/2006 (0.24 significant then 0.12 NOT significant), Torgerson 2006 (0.20-0.21 marginal once the Umbach outlier is dropped), McArthur 2012 (word-accuracy effect driven by two Levy studies at d = 1.12/1.80; the other eight average 0.16, and the control got NO phonics at all), Galuschka 2014 (0.198 after trim-and-fill; phonics indistinguishable from coloured overlays), Suggate 2010 vs 2016 (phonics 0.07 at 11-month follow-up, the smallest of four intervention types), Hammill & Swanson 2006 (r2 = 0.04), Han 2010, Adesope 2011, Sherman 2007 and Torgerson 2018's tertiary review.
Genetic confound
Design-based critique; correctly restricts probative evidence to experimental/quasi-experimental designs.
Replication notes
Contested by Fletcher et al. 2021, Buckingham 2020, Brooks 2023; Bowers replied 2021/2023. Dispute remains live and unresolved.
DOI / URL
10.1007/s10648-019-09515-y

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Synthetic-systematic phonics vs whole language (only 4 relevant NRP studies)d0.91, 0.12, 0.07, -0.47 (Fletcher concedes 'very modest')mixedend-of-treatmentactive-alternativeend-of-treatmentdomain-skill
NRP (2000) headline systematic-phonics effect, immediate vs 4-12 month delayd0.41 immediate falling to 0.27 at 4-12 months; NRP never assessed delayed effects on spelling, text reading or comprehension (Bowers p. 687, his Table 1)mixedimmediate and 4-12 month delayunclearunder-1yrdomain-skill
NRP systematic phonics for "low achieving" poor readers (below-average or unassessed cognitive level, above grade 1)d0.15, NOT significant (8 comparison groups) — vs 0.32 for "reading disabled" readers of at least average cognitive ability (Bowers p. 686)mixedend-of-treatmentunclearend-of-treatmentdomain-skill
NRP systematic phonics vs whole language specifically (12 studies, none an RCT)d0.31 overall (significant); decoding 0.55, comprehension 0.19; only 2 studies measured a delay (Bowers p. 688)mixedend-of-treatmentactive-alternativeend-of-treatmentdomain-skill
Camilli et al. (2006) multilevel reanalysis — systematic vs UNsystematic phonicsd0.12, NOT significant (vs 0.24 significant in Camilli 2003 and 0.41 in the NRP); Bowers' single strongest number against the systematic-phonics claim (p. 689)mixedend-of-treatmentactive-alternativeend-of-treatmentdomain-skill
Torgerson et al. (2006) RCT-only word reading accuracy, excluding the confounded Umbach 1989 outlier (d = 2.69)d0.20-0.21 (p = .03 on one analysis, p = .09 on another); comprehension 0.24-0.35 ns and spelling 0.09 ns even WITH the outlier; funnel plot showed publication bias (Bowers pp. 689-690)mixedend-of-treatmentunclearend-of-treatmentdomain-skill
Suggate (2016) long-term follow-up (mean delay 11.2 months) by intervention typedphonics 0.07 — the SMALLEST of the four; fluency 0.28, phonemic awareness 0.36, comprehension 0.46 (overall 0.37 short-term falling to 0.22); contradicts Suggate 2010 (Bowers p. 694)mixedmean 11.2-month follow-upunclearunder-1yrdomain-skill
Galuschka et al. (2014) phonics effect after trim-and-fill correction for publication biasg'0.198 corrected (from 0.32 uncorrected); uncorrected phonics 0.32 was statistically indistinguishable from phonemic awareness 0.28, fluency 0.30, auditory training 0.39 and coloured overlays 0.32 (Bowers pp. 692-693)mixedend-of-treatmentunclearend-of-treatmentdomain-skill
England synthetic-phonics rollout (Machin et al. 2018) — fadeout across key stagesstandardized effectERDp cohort: 0.298 at foundation stage, 0.075 at KS1 (age 7), -0.018 at KS2 (age 11). CLLD cohort: 0.217, then 0.017, then 0.019. Persistent KS2 effects only for non-native speakers (0.068) and FSM children (0.062) in CLLD, not ERDp; advantaged native-English ERDp children read WORSE at KS2 (-0.061, p < .1) (Bowers pp. 697-698)standardizedfoundation stage, age 7, age 11business-as-usualover-2yrdomain-skill
England reading transfer (PIRLS/PISA/SATs) after synthetic-phonics mandategainnone discernible; only the Phonics Screening Check itself rose (58% meeting standard in 2012 to 82% in 2018). PIRLS England ranked 3rd in 2001 (pre-mandate) and 8th in 2016 (11th for state schools only); Northern Ireland, which does not mandate systematic phonics, ranked higherstandardizedpopulationnoneover-2yrdomain-skill
Comprehension across the whole 12-meta corpuscount of metas with a significant comprehension benefit1 of 12 (the NRP), and only at short delay; 5 of the 12 metas found a NON-phonics intervention with a larger effect than phonics (Bowers pp. 695-696)mixedend-of-treatmentunclearend-of-treatmentdomain-skill

Cited by