Assigned homework goes with higher achievement, and in correlational studies the link is clearly stronger for pupils in grades 7–12 than in grades K–6, where it is small; but the small early-grade link has not been shown to mean a small effect: in the few experiments, classes given homework scored higher than classes without it, and in a systematic review whose homework-versus-none experiments were mostly in grades 3–6 (6 of 10; 4 in grades 7–12) the gain was about 0.45 standard deviations (mostly old studies at moderate to critical risk of bias; Guo et al. 2024).
Status
weakened
Last change
2026-11-04
Serves
Q1 (How do humans actually learn, and what helps them learn best?)
Ages
grundskola-lag, grundskola-hog, gymnasium
Area
Learning outside school
Relates to
C27 relates-to C28 (both about learning outside the lesson)
Bounds
Assigned homework and achievement; most evidence is correlational, and the size depends on how homework is measured (students' own reports gave stronger links). Evidence on file: Cooper, Robinson & Patall 2006 (Review of Educational Research, DOI 10.3102/00346543076001001; full text): generally consistent positive influence across designs; correlations stronger in grades 7–12 than K–6; five experiments with usable effects gave a mean d ≈ .60 on unit tests (95% CI about .38–.82), with design and unit-of-analysis limits. Fan, Xu, Cai, He & Fan 2017 (Educational Research Review 20, DOI 10.1016/j.edurev.2016.11.003; abstract/secondary summary only, flagged): maths and science r ≈ .22; numbers must be checked in the paper before any ledger use. Weakened 2026-11-03 (nightly attack): Guo, Li, Xu, Hu, Liu, Xing, Li, White & Yang 2024 (Campbell Systematic Reviews 20(3), e1431, DOI 10.1002/cl2.1431; open access, full text read): 11 controlled studies of homework versus no homework (2 RCTs, both at high risk of bias; 9 non-randomised, 4 moderate, 2 serious and 3 critical risk of bias; most published 1939–1995; only one with a follow-up); pooled g = 0.45 (95% CI 0.24–0.66; 14 comparisons; I² = 71%), arithmetic computation g = 0.46, problem solving g = 0.17, concepts not significant; the 14 homework-versus-none comparisons come from 10 studies published 1939–1995 in grades 3 (2 studies), 4–6 (1), 5 (1), 6 (2), 7 (1; Foran & Weber 1939), 8 (1; Ewart Anderson 1946) and senior high-school English (2; McGrath 1992, Tupesis 1972), so 6 of 10 in grades 3–6 and 4 in grades 7–12; the grade-2 randomised trial (Dolean & Lervåg 2021), the only study with a follow-up, compared more with less homework and is not in the pooled g; no grade-level subgroup analysis was possible (corrected 2026-11-04 against Table 2 and §5.3.1; the earlier "of the 8 studies that reported grade, 7 were in grades 2–6 and 1 in grade 8" was wrong). The review's RCT labels differ between §5.2.1 and Table 2; Egger's test p = 0.06. So the grade 7–12 versus K–6 contrast rests on correlations; the experimental evidence, though weak, is mostly from primary school and does not show a small effect there.
What would kill this
A meta-analysis finds no positive link between homework and achievement overall, or a link no stronger in grades 7–12 than in grades K–6, or a well-controlled trial finds that adding homework does not raise test results. Promoted from K57 on 2026-10-29 in the bold expansion Jonatan approved after four empty attack nights (26–29 Oct): new sharp claims in thin areas, each naming a comparison, a rough size or direction and a population.
Tutor consequence
none yet
Attack record
1 attack: 1 narrowed
2026-11-03 narrowed
Attack
"Where it is small" for grades K–6 rests on correlations; the controlled experiments of homework versus no homework are mostly in primary school and show a sizeable gain, so the claim's grade contrast does not hold as a statement about effects.
Target
claim-text
Source
Guo, Li, Xu, Hu, Liu, Xing, Li, White & Yang 2024 | Campbell Systematic Reviews 20(3), e1431, DOI 10.1002/cl2.1431 (open access, CC BY; full text read)
Steelmans to
The strongest version: the only causal evidence in a recent Campbell review (11 controlled studies, homework vs none, g = 0.45) comes mostly from grades 2–6, so the claim's "small" for K–6 reflects correlational designs, not what happens when homework is added.
Why
Nightly-attacks run of 2026-11-03 Europe/Stockholm. Pooled g = 0.45 (95% CI 0.24–0.66; 14 comparisons; I² = 71%); arithmetic computation g = 0.46, problem solving g = 0.17, concepts not significant. Of the 8 studies that reported grade: grade 2 (one), 3 (two), 5 (one), 6 (two), 4–6 (one), 8 (one); no grade subgroup analysis was possible. Weak evidence: 2 RCTs both at high risk of bias, 9 non-randomised studies (4 moderate, 2 serious, 3 critical risk of bias), mostly 1939–1995, one with follow-up. Not a kill: the correlational grade contrast (Cooper et al. 2006) stands, and the review does not show a link no stronger in grades 7–12. Kill-match check: the new hedge ("has not been shown to mean a small effect") does not empty the kill line; "a link no stronger in grades 7–12 than in grades K–6" and "a well-controlled trial finds adding homework does not raise test results" still name results the new text does not concede. Not a kill-mismatch.
Rewrite
claim text (the project, nightly attack; automatic apply authorised by Jonatan 2026-09-25)
Corrections to our notes
These fix a number or a fact in our notes. They are not attacks and do not count toward the floor of 3 survivals.
— C27's claim text says the review's experiments were "mostly in grades 2–6" and its limits say "of the 8 studies that reported grade, 7 were in grades 2–6 and 1 in grade 8"; the review's Table 2 and §5.3.1 show that the pooled homework-versus-none g = 0.45 comes from 10 studies, 6 in grades 3–6 and 4 in grades 7–12, and that the grade-2 RCT is not in the pooled estimate.
Weakened C27 2026-11-03: homework vs none g = 0.45 [0.24, 0.66] across 10 controlled studies published 1939–1995 (6 in grades 3–6, 4 in grades 7–12), mostly at moderate to critical risk of bias and tested only at the end; the K–6 "small" is correlational only. (Grade mix corrected against the full text 2026-11-04.)
Grade 8, 2,808 pupils, 76 Chinese schools (cross-sectional, self-report): 30–60 min of daily maths homework went with slightly higher scores than ≤30 min (β = 0.06), longer times did not (β = 0.01); 60–90 min beat ≤30 min only when perceived homework quality was low (interaction β = −0.05); the homework-time block adds 0.3% of within-school variance and the time × quality terms another 0.3%. Homework quality, not time, went with maths interest (time not significant once quality is in the model, despite the abstract). Bounds what kind of homework the 7–12 link reflects. Full text read; corrected 2026-11-02.
Review state: first-pass draft, pending human review.