Syntes: Abe & Usami (2024) — teachers' written comments beat a numeric grade on a later test (meta-analysis, 9 studies)

Källa: https://doi.org/10.2333/jbhmk.51.45 (Kodo Keiryogaku 51(1), 45–54; Japanese, English abstract)

Verified content

• [full text] Inclusion (p. 46–47): the teacher writes verbal feedback on a test or piece of work and the learner receives it on paper; two groups, verbal feedback versus numeric rating; a post-test given after an interval following the feedback ("FB 後に期間を空けて実施された事後テスト"); school-education setting. 32 effect sizes from 9 studies (11 effects from 3 studies not in Koenka et al. 2021).
• [full text] Random-effects pooled g = 0.42 (95% CI 0.17 to 0.66, p = .002); τ² = .39; I² = .902; fixed-effect 0.38 (0.31 to 0.45); Bayesian 0.42 (0.17 to 0.67); Egger intercept 0.60, p = .70 (pp. 50–51).
• [full text] Meta-regression (Table 1, p. 52; reference category task-level feedback): learner age −.07 per year (95% CI −.12 to −.04, p < .001); self-level feedback (praise such as "very good", no task content) −.53 (−1.03 to −.04, p = .04); process-level .04 (ns, 3 effects); intra-individual-change comments .04 (ns, 2 effects); factual/procedural-knowledge tests −.47 (p = .06); high-stakes post-test −.41 (p = .07); number of feedback occasions .01 (ns). R*² = .78.
• [full text] Countries: USA 11 effects, Israel 10, Germany 10, Venezuela 1; ages from upper primary to university; mostly maths and languages; many comments were self-level ("very good") (p. 50).
• [OSF table, authors' coding] Studies: Butler & Nisan 1986 (0.13 quantitative task, 1.51 divergent thinking); Butler 1987 (task comments 1.76, 1.78; "very good" −0.25, 0.78); Butler 1988 (0.94, 0.68, 1.63, 1.45); Dlaska & Krekeler 2017 (university German L2 corrective feedback: 0.21, −0.09, 1.15, 0.39); Elawar & Corno 1985 (1.62); Hake 1973 (dissertation; encouraging/personal comments −0.40, −0.43, 0.37, −0.03); Krampen 1987 (age 13.9, maths, 4–8 feedback occasions over 180–360 days: subject-matter comments 0.22 and 0.00; social-comparison comments −0.24, −0.12; individual-change comments 0.50, 0.13); Lutkenhouse 1975 (dissertation; −0.54, 0.16, 0.15); Starkey 1970 (praise/encouraging comments 0.05, 0.01, −0.08, −0.02). Page 1958 is discussed but not among the 32 effects.

Limits

• [full text] Nine studies, many old (1970–1987) and from three countries; authors caution about generalising to Japan and to current curricula (pp. 53–54). Moderators other than age and self-level feedback are not significant, partly for lack of power (p. 52).
• [assessment] The large task-comment effects come mainly from Butler's Israeli grade 5–6 experiments, where the later session was within about two days; the longest classroom study (Krampen 1987, school year, maths) gives subject-matter comments over grades 0.22 and 0.00.
• [assessment] The meta compares comments with grades; it does not compare task comments with praise alone (its praise-type comments are compared with grades, not with task comments).

Role for C4

• [assessment] A kill-condition test of C4's comments-versus-grade comparison on later tests: comments came out ahead (g = 0.42), and task-level comments ahead of praise-type comments; C4 survived (nightly attacks 2026-11-14). The age moderator (−.07 per year) is recorded in C4's Bounds as a limit.

Added 2026-11-14 by the project (nightly attack run). [full text] = read in the full text; [OSF table] = the authors' own per-effect coding table on OSF; [assessment] = the project's own assessment.