The Impressive Effects of Tutoring on PreK-12 Learning: A Systematic Review and Meta-Analysis of the Experimental Evidence

96 randomized evaluations, 732 effect estimates; the parent-tutor cell is 11 studies / 67 estimates

PreK-12 tutoring RCTs, predominantly US, predominantly literacy

This is the NBER working-paper version (w27476, July 2020) of the meta-analysis this archive already holds in its published form as `nickow-2024-tutoring-meta`. It is recorded separately for ONE reason: its tutor-type panel (Table 3A) reports a PARENT cell, which is the only pooled estimate of parent-delivered tutoring on a common scale with other tutor types anywhere in the literature, and which the published panel recorded in this archive does not carry. The two versions are NOT independent estimates and must never be summed or averaged: the working-paper panel runs higher throughout (teacher 0.50 / paraprofessional 0.40 / nonprofessional 0.21) than the published panel as recorded here (0.39 / 0.30 / 0.17). Cite this file only for the parent cell and its conditions; cite the published file for the headline pooled effect and the dosage recipe. Read: title page, abstract, the tutor-type sections (pp. 28-33), the scale-up subsample discussion, and Table 3A in full.

Table 3A, all studies: pooled effect 0.37 [SE 0.032], NS=96, NE=732. By tutor type — teacher 0.50 [0.075] NS=18; paraprofessional 0.40 [0.040] NS=46; nonprofessional/volunteer 0.21 [0.064] NS=24; PARENT 0.23 [0.114] NS=11, NE=67, significant only at the 10% level. The parent cell is confounded with the worst delivery conditions in the whole literature: parent tutoring after school is 0.16 [0.092] across 10 of the 11 parent studies (there is only one during-school parent tutoring program in the sample), and every parent study is 1:1 because "parents usually have only a single child at a time". The authors state there were NO large-scale impact evaluations of parent tutoring programs, and that no parent tutoring study met the criteria for their scale-up subsample. Their own reading is deliberately not dismissive: "even 0.20-0.25 SD can be substantial if costs are low, as they typically are for parent tutoring programs", and the largest-sample parent study they found (Lam et al. 2013, a Hong Kong preschool paired-reading programme with just under 200 children) produced an effect close to the overall pooled estimate of slightly over a third of an SD.

Low. Randomized assignment throughout; families in the treatment and control arms of a parent-tutoring trial are equally selected and differ only in what they were asked to do. The parent's genotype is constant across arms by construction, which is exactly why this estimate is worth more than the entire observational home-schooling literature.

The `replication: unreplicated` flag refers to the finding this file is cited for — the parent-tutor estimate — not to the meta-analysis as a whole. That cell rests on 11 small studies with no large-scale impact evaluation behind it and a standard error of 0.114, and the authors say so themselves. The meta-analysis's headline pooled effect and dosage moderators are replicated and are recorded on the published-version source, `nickow-2024-tutoring-meta`. Treat this file and that one as one study in two versions, never as two studies.