Should we Help our Children With Homework? A Meta-Analysis Using PISA Data

180 effects drawn from PISA 2009, 2012, 2015 and 2018 in every country that administered the PISA family questionnaire

15-year-olds and their families across Europe, Southeast Asia, and Latin America and the Caribbean.

Psicothema 34(1), 56-65. A meta-analysis whose primary studies are all cross-sectional national samples from the same survey programme, which makes it large, comparable and causally inert — grade D follows from the design, not the execution. Each effect is the standardized difference in PISA score between students whose families help with homework and those whose families do not, scaled by the pooled SD; PISA scores are standardized to SD = 100. The reported overall d = 0.23 favours the NO-HELP group (the sign convention in the paper is that positive values indicate benefit, and the reported result is that helped students score lower). Outliers and influential cases were identified via Cook's distance and DFFITS. The design's value is invariance testing: the same question asked four times across nine years, across three world regions, and across reading, maths and science.

Students who receive more family help with homework score LOWER, d = 0.23, 95% CI [0.21, 0.25], and the finding is invariant across subject and stable across a decade. It is NOT invariant across culture: Europe d = 0.30 [0.27, 0.33], roughly three times the Southeast Asian d = 0.09 [0.07, 0.11]. Mathematics specifically runs at d = 0.24 [0.20, 0.28], indistinguishable from reading. The authors' conclusion is the correct one to record: "family help with homework does not ensure students' academic success, and it is more important how that help is given than how much." The reverse-causation reading — struggling students get more help — is the obvious rival and the design cannot exclude it; the paper's contribution is not that helping harms, but that the volume of parental homework help is not a lever, in any country, in any subject, in any of four PISA cycles.

HIGH, plus a dominant reverse-causation channel. Parents help more when the child is doing badly, and children who do badly have parents who did badly, so the negative association is over-determined by selection running in both directions. Nothing in a cross-sectional PISA design can separate them. The value of the source is as a boundary marker on the positive claim, not as evidence of harm.

Converges with Barger et al. 2019 (448 studies, 480,830 families; homework assistance r = -.15, the only involvement type with a negative sign) and with Xu et al. 2024's three-level meta (r = -0.064 overall, with autonomy-supportive help positive and content support, parental control and sheer frequency unrelated). The consistent structure across three independent syntheses is that the AMOUNT of parental homework help is null-to-negative while the STYLE of help carries whatever signal exists — which is the same specificity gradient the family-literacy metas find, arriving from the opposite direction.