A digitally-delivered, double-blind randomised controlled trial investigating the effect of homework length on year 7 students' attainment in mathematics

368 Year 7 pupils randomised; complete pre- and post-test data for 354 (96%)

Eleven- and twelve-year-olds in three secondary schools in Devon, England, all already using the Sparx digital maths platform

Sparx Limited white paper, 2019; no DOI; NOT PEER REVIEWED; PUBLISHED BY THE VENDOR OF THE PLATFORM UNDER TEST. Those three facts are why this grades C and carries independence developer-led despite being, as far as this archive can establish, THE ONLY GENUINE STUDENT-LEVEL RANDOMISED DOSAGE TRIAL OF HOMEWORK IN SCHOOL-AGE CHILDREN. The design itself is good: four arms, double-blind, with pupils matched within class by prior attainment (Key Stage 2 maths scaled score) into groups of four and randomly assigned within group to 0%, 50%, 100% or 150% of their normal algebra homework by time - 0, 15, 30 and 45 minutes per week. Four weeks, one algebra unit, pre- and post-assessment scored by a one-parameter IRT model. Analysis is a linear mixed model with fixed effects for prior attainment and test time and random effects for student and class.

Homework length was 'very significantly associated with the change in student attainment between the two test points', p = 0.000056, with the final model explaining 71.8% of variance; mean progress 'jump[ed] considerably between the 0 and 50% points and then increase[d] linearly with homework length'. NO STANDARDISED EFFECT SIZE IS REPORTED ANYWHERE IN THE PAPER - not d, not g, not SD units - and the tables are images from which no coefficients can be extracted, so the magnitude was not obtained. TWO SCOPE LIMITS MATTER MORE THAN THE SIGNIFICANCE. First, the largest arm was 45 MINUTES PER WEEK of algebra homework, so this trial cannot address the 60-to-90-minutes-per-DAY question that Cooper and Fernández-Alonso are arguing about; it establishes that some homework beats none at very low doses. Second, the authors' own generalisability caveat is explicit: the finding applies to 'Year 7 students getting bespoke algebra homework assigned by and delivered through the digital Sparx platform. Whether these findings are applicable to any regularly set paper-based homework, or indeed homework distributed via other digital platforms, remains to be investigated.' They also note they 'did not identify an upper limit for homework length beyond which homework does not result in better outcomes'.

Low. Pupils were matched on prior attainment and then randomised within matched quadruples inside the same class, so ability, teacher and school are all balanced by construction. This is the cleanest identification in the homework literature and it is attached to the weakest publication.

Unreplicated and vendor-produced, which under this archive's rules caps what it can establish. Its positive sign runs against the two student-randomised studies pooled by Baş and colleagues (d = -0.125) and with the class-randomised majority. Because the dose range is so narrow and the platform supplies immediate feedback and adaptive question selection, it may be measuring the same thing Roschelle and colleagues measured - the value of well-designed feedback-bearing practice - rather than the value of homework minutes.