Who benefits from homework assignments?

96,925 pupils in the naive specification; 4,316 pupils in 254 classes for the school-grade fixed-effects analysis; 254 classes for the dispersion analysis

Dutch primary school pupils in grades 2, 4, 6 and 8, PRIMA survey waves 1994-95 to 2002-03

Economics of Education Review 30(1), 55-64. NOTE: the data are DUTCH (the PRIMA survey), not Norwegian; the author is at Statistics Norway. Homework is measured only at the EXTENSIVE MARGIN - the teacher reports whether homework is given to the whole class or to nobody, with classes where only weak or only strong pupils get homework dropped - so this study cannot speak to dosage at all. Outcome is an externally graded standardized test, standardized within grade and year. The headline analysis is a difference-in-differences on within-class test-score DISPERSION with school-grade fixed effects. The author's own limitations are serious: teacher and class fixed effects 'cannot be identified in the PRIMA data', the homogeneity assumption behind the difference-in-differences is 'restrictive', and homework frequency is unobserved. Grade C rather than B for those reasons.

THE COLLAPSE IS THE FINDING. Naive OLS says 'pupils who get homework perform 12 percent of a standard deviation WORSE on average than pupils who do not get homework' (-0.1206, SE 0.0163). Add individual characteristics and it is -0.0049 (0.0115), not significant. Add class and teacher characteristics and it is -0.0016 (0.0114). The entire association is composition. Under school-grade fixed effects the average effect is +0.0520 (SE 0.0385), which the author reports as 'only statistically significant at the 18 percent level' - a null. THE INEQUALITY RESULT is what the paper is cited for: homework raises within-class dispersion (variance +0.0567, SE 0.0284; the 85th-15th percentile gap +0.1800, SE 0.0676; the 85th-50th gap +0.1646, SE 0.0545) while the 50th-15th gap is null. 'The lower part of the distribution coincide in homework and non-homework classes, implying that the weakest pupils are unaffected by homework. The upper part of the distribution is on the other hand skewed to the right in homework classes.' Splitting by mother's education: +0.0722 (SE 0.0404) where the mother has more than primary education, -0.0105 (0.0624) where she does not. The author's proposed mechanism is worth recording because it is testable and untested: homework may act as 'a substitute for classwork, i.e., teachers exert less effort in the classroom when homework is given.'

Medium to high, and it is the reason the naive estimate is worthless. Whether a class gets homework is a teacher and school decision correlated with intake; whether a child benefits is confounded with everything the family supplies, including alleles. The school-grade fixed effects handle sorting across schools but not the fact that within a school, the pupils who can work unsupervised at home differ from those who cannot in heritable ways. The mother's-education split is a description of who benefits, not an identified interaction.

The average null replicates - Jerrim and colleagues find a precisely estimated zero at the same age with a stronger design, and Cooper's synthesis puts the K-6 correlation at -.04. The INEQUALITY finding does not: Jerrim explicitly reports 'a null association for the different socioeconomic backgrounds' and names this paper as the result it fails to reproduce. Recorded as mixed for that reason - the headline that homework widens gaps in primary school rests on one study with an extensive-margin measure and no teacher fixed effects.