The Evidence on Teaching

Meta-Analysis and Reading Policy: Perspectives on Teaching Children to Read

Camilli G, Wolfe PM, Smith ML · 2006

grade Ccritiqueindependentmixednumbers spot-checked
Sample
Same NRP corpus (40 studies / 225 comparisons), re-analysed with a multilevel model
Population
Same K-6 NRP corpus.
Design
Follow-up to the 2003 weighted-regression reanalysis, replacing it with a multilevel model that partitions the phonics contrast from co-occurring tutoring and language instruction.
Key findings
With a multilevel model and co-interventions controlled, the systematic-phonics effect falls to a non-significant 0.123 while individual tutoring reaches 0.49. The finding is not 'phonics does not work' — it is that the NRP's headline number was carrying tutoring and language instruction along with it.
Genetic confound
Experimental corpus; the confound at stake is instructional co-intervention, not heredity.
Replication notes
Rebutted on estimand grounds by Stuebing et al. 2008 (who argue the isolated-phonics coefficient is not the policy-relevant quantity); defended in Camilli, Kim & Vargas 2008.

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Systematic phonics net of language activities and tutoringd0.123 (NOT statistically significant)mixedend-of-treatmentactive-alternativeend-of-treatmentdomain-skill
Individual tutoring (moderator)d0.49 — larger than the phonics contrast it is usually bundled withmixedend-of-treatmentactive-alternativeend-of-treatmentdomain-skill

Cited by

Meta-Analysis and Reading Policy: Perspectives on Teaching Children to Read · The Evidence on Teaching