A meta-analysis of single subject design writing intervention research.

88 single-subject design studies from which an effect size could be calculated

School-age students, predominantly struggling writers and students with disabilities

Meta-analysis of SINGLE-SUBJECT (n-of-1, multiple-baseline) writing intervention research - a design class that METHODOLOGY grades D because participants serve as their own controls with no randomisation, no comparison group, and typically no blinding of the person scoring the writing. Average effect sizes (PND-family metrics) were computed only for treatments tested in 4+ studies with a comparable outcome, identifying 9 supported treatments. Recorded here specifically because it is the ONE synthesis in the field that finds positive effects for grammar instruction, and an honest grammar topic has to say so and then say why it does not overturn the experimental record.

Nine writing treatments were supported as effective: strategy instruction for planning/drafting, teaching grammar and usage, goal setting for productivity, strategy instruction for editing, writing with a word processor, reinforcing specific writing outcomes, use of prewriting activities, teaching sentence construction skills, and strategy instruction for paragraph writing. The grammar-and-usage result is the outlier against Hillocks 1986, Graham & Perin 2007, Graham et al. 2012, Koster et al. 2015, Andrews et al. 2004/2006 and the three EEF/Englicious RCTs. Note also what "teaching grammar and usage" means in a single-subject study of struggling writers - usually targeted correction of a specific error class in a specific pupil, measured on that same error class, which is nearer to mechanics remediation than to a grammar curriculum.

Not applicable in the usual sense - within-subject designs net out stable individual differences by construction. The problem is the opposite one: with no control group these designs cannot separate treatment from maturation, practice, or regression to the mean, and effect metrics for single-subject designs are not commensurable with between-group standardised mean differences.