The association between homework and primary school children's academic achievement. International evidence from PIRLS and TIMSS

24 countries; fourth-grade children who sat PIRLS and TIMSS in the same 2011 round

Nine- and ten-year-olds in 24 countries, each observed in reading, mathematics and science

European Journal of Education 55, 248-260; full text read from the University of Malaga repository accepted manuscript. THE BEST-IDENTIFIED HOMEWORK STUDY IN PRIMARY SCHOOL, and it works because of an accident of test administration: in 2011 the SAME fourth-grade children sat PIRLS (reading) and TIMSS (mathematics and science), with teacher-reported homework time recorded for each subject. Student fixed effects across three subjects therefore difference out gender, socio-economic status, family, school, ability and motivation - every time-invariant, subject-invariant confound in one step. Homework time is reported by the TEACHER, not the child, which removes the self-report artefact that Cooper's synthesis shows carries the entire correlational association. Robustness checks restrict to children reporting equal enjoyment and equal confidence across the three subjects, add homework frequency and three teacher homework-practice variables with interactions, and split by sex and by tertiles of home learning resources. Limits the authors state: no prior-achievement data so no value-added model, no homework-motivation data, and a design that by construction cannot detect a uniform across-the-board homework effect.

A precisely estimated zero. 'The average association across the 24 countries is just 0.01 standard deviations for every extra ten-minute increase, with the results only statistically significant in Georgia and Lithuania. However, even in these countries, the effects are small (0.022 and 0.024 standard deviations for every extra 10-minute increase in homework time respectively). Moreover, given the large sample sizes, one can see that these small effects are estimated quite precisely, providing further evidence of a genuine null association (i.e., it is not a lack of power as a result of poor sample sizes that drives this result).' Every robustness check returns the same answer: 'homework time presents zero correlation with students' academic achievement.' TWO FINDINGS CUT AGAINST RECEIVED WISDOM IN OPPOSITE DIRECTIONS. The equity claim fails: 'we obtained a null association for the different socioeconomic backgrounds. Hence, it seems that homework may not prejudice poor students nor benefit rich ones (unlike the results of Rønning, 2011).' And the policy conclusion is blunt: 'we conclude that the homework teachers assign in primary school does not produce a meaningful positive influence on achievement. Homework needs to be improved if this time-consuming activity is to become a worthwhile investment.'

Low, and by construction - this is the only paper in the homework literature that removes it. Student fixed effects difference out every stable child and family trait, including the conscientiousness, prior attainment and parental supervision that make ordinary homework correlations uninterpretable. What remains is subject-specific confounding: a teacher might set more homework in the subject the class is weakest at, which biases toward zero, or the subject she teaches best, which biases away.

No direct replication of the within-student design exists in this literature. It converges with the correlational record at the same age - Cooper's grades K-6 correlation of -.04 (fixed-error) and Ozyildirim's fourth-grade TIMSS estimate of d = -0.057 - and with Rønning's null average effect in Dutch elementary schools. It directly contradicts Rønning's inequality finding, using the stronger design.