A meta-analysis of writing instruction for adolescent students.

123 documents yielding 154 effect sizes for quality of writing; 127 effect sizes across the 15 treatments with 4+ studies, drawn from 107 different studies

Students in Grades 4-12 (roughly ages 9-18) in regular public or private schools; studies in special schools excluded

Meta-analysis of experimental and quasi-experimental writing-intervention studies, weighted fixed-effects model. The single outcome is QUALITY OF WRITING - "readers' judgment of the overall merit of a composition" scored on a rating scale. That is a researcher-scored, holistic, treatment-proximal measure in essentially every included study, so METHODOLOGY's ~2x measure-type inflation applies to the whole table. Study quality was scored on nine indicators. Homogeneity was significant (i.e. rejected) for grammar, strategy instruction, word processing and process approach, so the pooled numbers conceal large between-study variation. Studies where students served as their own controls were excluded. Handwriting and spelling were deliberately excluded from the review because their effect on writing quality had only been studied in primary grades.

The headline table (weighted mean ES, 95% CI): grammar instruction -0.32 (-0.43, -0.21), k=11 - the ONLY negative entry in the entire meta-analysis; sentence combining +0.50 (0.30, 0.70), k=5; strategy instruction +0.82 (0.69, 0.95), k=20, of which SRSD +1.14 (0.89, 1.39), k=8 and non-SRSD strategy instruction +0.62 (0.44, 0.80), k=12; summarization +0.82, k=4; peer assistance +0.75, k=7; setting product goals +0.70, k=5; word processing +0.55, k=18; prewriting +0.32, k=5; inquiry +0.32, k=5; study of models +0.25, k=6; process writing approach +0.32, k=21. The process approach decomposes decisively: WITH professional development +0.46 (0.37, 0.56), k=6; WITHOUT professional development +0.03 (-0.07, 0.13), k=15, and -0.05 (-0.18, 0.07) in Grades 7-12. The authors also recalculated Hillocks (1986): grammar -0.29, sentence combining +0.35, study of models +0.22, evaluation scales +0.36, free writing +0.16, inquiry +0.56 - an independent replication of the grammar/sentence-combining sign split two decades earlier.

Low for the direction of effects - these are experimental and quasi-experimental comparisons, so treatment groups do not differ systematically in genotype. The residual risk is in the quasi-experimental half of the corpus, where selection into treatment classrooms is not controlled, and in the outcome, which is a subjective quality rating by scorers who may know the condition.