The Evidence on Teaching

Breakfast in the Classroom Initiative Does Not Improve Attendance or Standardized Test Scores among Urban Students: A Cluster Randomized Trial

Luan D, Foster GD, Fisher JO, Weeks HM, Polonsky HM, Davey A, Sherman S, Abel ML, Bauer KW · 2022

grade Brctindependentfailed
Sample
16 K-8 public schools randomized; 1,362 grade 4-6 students with district administrative outcome data
Population
Low-income urban US schoolchildren in Philadelphia, PA; all schools >=50% free/reduced-price eligible. Both arms received free universal breakfast.
Design
Cluster-randomized trial (2013-2016) of Breakfast in the Classroom versus free universal breakfast served in the cafeteria before school. This is the randomized version of exactly the contrast Imberman & Kugler estimated with difference-in-differences. Outcomes measured at 1.5 and 2.5 years using district standardized exams and administrative attendance - independent, not researcher-designed. Weighted GEE analysis. Secondary analysis of a trial powered primarily for obesity outcomes, so precision on achievement is a limitation.
Key findings
Null across the board after 2.5 years. Attendance beta = 0.004 (SE 0.06), P = 0.94; reading beta = 0.02 (SE 0.06), P = 0.79. Math was significantly WORSE in BIC schools: beta = -0.20 (SE 0.07), P = 0.005. The randomized estimate flatly contradicts the ~+0.10 SD difference-in-differences estimate for the same intervention contrast.
Genetic confound
None - randomized at the school level, so genes cannot differ systematically between arms.

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Standardized reading exam (domain skills)regression coefficient, SD units0.02 (SE 0.06), P = 0.79 - nullstandardized2.5 years post-randomizationactive-alternativeend-of-treatmentdomain-skill
Standardized math exam (domain skills)regression coefficient, SD units-0.20 (SE 0.07), P = 0.005 - significantly negativestandardized2.5 years post-randomizationactive-alternativeend-of-treatmentdomain-skill
School attendance (behavioral)regression coefficient0.004 (SE 0.06), P = 0.94 - nullstandardized2.5 years post-randomizationactive-alternativeend-of-treatmentbehaviour

Cited by