The Effect of Teacher Coaching on Instruction and Achievement: A Meta-Analysis of the Causal Evidence
Kraft MA, Blazar D, Hogan D · 2018
grade Bmeta-analysisindependentunreplicatednumbers spot-checked
Sample
60 causal studies of teacher coaching programmes
Population
Teachers in coaching programmes, mostly US, K-12.
Design
Meta-analysis restricted to causal designs, with explicit tests for scale and for publication bias. The scale analysis is what makes it usable: effects are reported separately for trials with 100 or more teachers.
Key findings
Content-specific instructional coaching is the one form of teacher development with credible causal support that survives at scale (+0.10 SD in large trials), and it is mostly a reading effect — pooled math is null. This is where a professional-development budget should go.
Genetic confound
Teacher-level; student selection is not the mechanism at issue.
Replication notes
The coaching result survives both the scale test and the publication-bias correction, which is rare in this literature.
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Student achievement, pooled | SD | +0.18 | standardized | end of year | business-as-usual | end-of-treatment | domain-skill |
| Large trials (>=100 teachers) | SD | +0.10 (p < .001) — shrinks with scale but survives | standardized | end of year | business-as-usual | end-of-treatment | domain-skill |
| Publication-bias correction | SD | trim-and-fill 0.14 | standardized | end of year | business-as-usual | end-of-treatment | domain-skill |
| Subject and dosage | pattern | reading-dominated (math pooled 0.04, ns); dosage effects null; ~$3.3-5.2k per teacher | standardized | end of year | business-as-usual | end-of-treatment | domain-skill |
Cited by
- Teacher quality — selection over credentials and workshopsstrong supportconf: highgc: low