The Evidence on Teaching

Systematic Phonics Instruction Helps Students Learn to Read: Evidence from the National Reading Panel's Meta-Analysis

Ehri, L. C., Nunes, S. R., Stahl, S. A., & Willows, D. M. · 2001

grade Cmeta-analysisindependentmixednumbers spot-checked
Sample
38 studies / 66 treatment-control comparisons (~7,000+ children)
Population
K-6 English readers; normal, at-risk (K-1), low-achieving (older), and reading-disabled subgroups; US/UK/NZ.
Design
Peer-reviewed RER version of the NRP (2000) phonics chapter. Systematic phonics vs the least-phonics available control (whole language, whole word, basal, regular curriculum). Mixes RCTs and quasi-experiments; post-1970 published-only; heavy use of researcher-designed proximal decoding measures. Estimand caution (Stuebing 2008 decomposition): 0.41 answers systematic-vs-no/unsystematic phonics, NOT systematic-vs-some-phonics.
Key findings
Systematic phonics beats unsystematic/no-phonics instruction at end-of-treatment, overall d=0.41. Effects are word-level/decoding-specific (regular words 0.98, pseudowords 0.67 in K-1) and concentrated in K-1 (0.55); they fall to 0.27 for grades 2-6 and are NOT significant for comprehension/spelling among older readers, nor for low-achieving older readers. No synthetic-vs-analytic advantage; tutoring = classroom. The NRP's own single delayed-follow-up estimate (d=0.27, k=6) already signals fadeout.
Genetic confound
Experimental controls largely address selection/passive rGE. Residual: comprehension (language/g-loaded) barely moves while trainable decoding does.
Replication notes
Corpus re-analyzed by Camilli 2003/2006 and Stuebing 2008; magnitude reproducible but contested by estimand. Load-bearing values verified against Ehri (2003) DfES open-access mirror.
Grade note
Grade C: mostly small, older, non-preregistered studies mixing RCTs with quasi-experiments; many researcher-designed proximal measures; published-only.
DOI / URL
10.3102/00346543071003393

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Overall reading (end of training)d0.41 (65 comparisons)mixedend-of-treatmentbusiness-as-usualend-of-treatmentdomain-skill
Overall reading (end of training or end of first year)d0.44 (62 comparisons)mixedend-of-treatmentbusiness-as-usualend-of-treatmentdomain-skill
Reading (delayed follow-up, 4mo-1yr post)d0.27 (only 6 comparisons; significant)mixed1yr follow-upbusiness-as-usualunder-1yrdomain-skill
Decoding regular words / pseudowords (K-1)dregular 0.98; pseudoword 0.67 (largest outcomes)researcher-designedend-of-treatmentbusiness-as-usualend-of-treatmentdomain-skill
Reading comprehension (K-1 vs grades 2-6)d0.51 (K-1); 0.12 grades 2-6 (ns)mixedend-of-treatmentbusiness-as-usualend-of-treatmentdomain-skill
Spelling (K-1 vs grades 2-6)d0.67 (K-1); 0.09 grades 2-6 (ns); ~0.35 pooledmixedend-of-treatmentbusiness-as-usualend-of-treatmentdomain-skill
By grade phonics beginsdK 0.55 / grade1 0.54 / grades 2-6 0.27 (all sig)mixedend-of-treatmentbusiness-as-usualend-of-treatmentdomain-skill
Low-achieving (non-RD) older readersdnot significant (only null subgroup)mixedend-of-treatmentbusiness-as-usualend-of-treatmentdomain-skill
Synthetic vs larger-unit phonics; tutoring vs group vs classdno significant difference on either dimensionmixedend-of-treatmentactive-alternativeend-of-treatmentdomain-skill

Cited by