Impacts of Academic Recovery Interventions on Student Achievement, 2022-23 (Road to Recovery, 8 districts)
Carbonari, M. V., DeArmond, M., Dewey, D., Goldhaber, D., Kane, T. J., Staiger, D. O., et al. · 2026
grade Cquasi-experimentindependentreplicated
Sample
8 districts / 27 subject-program evaluations / >12,000 tutored students
Population
K-8 students in mid-to-large US districts; ESSER-funded recovery programs.
Design
Value-added models on NWEA MAP + fuzzy RD in two districts. Predecessor wave (4 districts, 2021-22): all effects <0.04 SD.
Key findings
The definitive record of US pandemic-recovery tutoring spending: as actually implemented, mostly null. The only programs that worked were tiny and high-dosage. Implementation — reaching students and delivering hours — is the binding constraint, not the tutoring model.
Genetic confound
VAM identification weaker than RCT; missingness likely conservative.
Replication notes
Consistent across both project waves, Nashville, and the UK NTP.
DOI / URL
10.3102/01623737251400377
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Tutoring/small-group pooled | SD | ns pooled (~0.03-0.07); robust positives in only 5 of 27 program-subject pairs; one program significantly NEGATIVE in math | standardized | spring of treatment year | business-as-usual | end-of-treatment | domain-skill |
| The outliers that worked | SD | +0.22 to +0.33 — programs serving 1-2% of eligible students with >30 delivered hours | standardized | spring of treatment year | business-as-usual | end-of-treatment | domain-skill |
| Implementation | delivery | most programs reached <=51% of targeted students; only 4 of 27 met intended minimum dosage | standardized | during year | none | end-of-treatment | behaviour |
Cited by
- Tutoring — the honest effect, the Bloom 2-sigma myth, and what survives scalestrong supportconf: highgc: low