Reading-comprehension strategy instruction (reciprocal teaching, summarizing, questioning)

Verdict

Comprehension-strategy instruction (reciprocal teaching, summarizing, self-questioning, comprehension monitoring) is best understood as a low-cost, one-time "trick" — teaching a student to remember to monitor and summarize — rather than a skill that grows with practice. In tightly controlled efficacy trials on standardized measures it produces a small real effect (d≈0.32); but the effect is ~2.75× larger on the researcher's own tests than on independent ones, it does not increase with more instruction (a handful of sessions works as well as dozens), and under realistic whole-class delivery at scale it is ~null. That combination — real in some conditions, absent in others, inflated by measurement — is why the verdict is mixed rather than supportive.

What the evidence shows

The NRP 2000 comprehension chapter is often cited here with d=0.88/0.32 — but those are literally the Rosenshine & Meister medians reused; the NRP chapter computed no independent pooled effect, so it is not a second confirming source.

The two robust regularities: (1) the effect is ~2.5-2.75× larger on experimenter-built tests than on standardized ones — the standardized ~0.19-0.32 is the honest number; (2) dose-response is flat, which is the empirical basis for calling strategies a saturating trick rather than a trainable skill. Modern at-scale trials (EEF whole-class d=0.00; O'Hare 2025 definitive high-school null) push the realistic-delivery estimate down toward zero, with a small effect surviving only for targeted small-group delivery.

Hereditarian-lens assessment

Risk: low for the causal effect (the trials are experimental). But this topic is where the correlational comprehension literature is most seductive and most confounded: "good readers use more strategies" is cross-sectional and presumptively driven by heritable/g-linked comprehension ability. The predictor hierarchy (vocabulary > inference > knowledge; see Elleman & Oslund) tells you what covaries with comprehension, not what teaching moves it. The experimental record says: strategy teaching gives a small, saturating boost, and general comprehension — which tracks language and g — is not durably raised by it.

Fadeout & durability

Durability is rarely measured in this literature (effects are reported end-of-treatment). Given the saturating dose-response and the general pattern that comprehension-adjacent gains on standardized measures are small, the end-of-treatment effects should be read as upper bounds. Interestingly, Suggate 2016 finds comprehension-oriented interventions maintain better than decoding ones — but that subgroup is age-confounded (given later) and shows publication-bias signals.

What critics say / limits

• Willingham's "brief instruction suffices" framing is contested by Shanahan (who argues for more sustained work), but the flat dose-response it rests on is consistent across the literature.
• The whole-class null (EEF, O'Hare) is the most policy-relevant finding: what works in a researcher-run small group does not survive translation to ordinary classrooms.

Practical guidance

• Teach a small set of comprehension strategies briefly — a handful of sessions to install the habit of monitoring/summarizing/self-questioning. More time yields little.
• Prefer targeted small-group delivery for struggling comprehenders; do not expect a whole-class strategy program to move standardized comprehension.
• Don't substitute strategy drilling for content. The durable comprehension levers are vocabulary and background knowledge built over years (see knowledge/vocabulary). Strategies are the cheap add-on, not the engine.
• Skip it in K-2: with working memory occupied by decoding, strategy instruction is weak before ~grade 3-4.

Open questions

• Whether the small targeted-delivery effect (0.14-0.18) is durable is unmeasured.
• The exact boundary between "installing a habit" (works fast) and "building comprehension skill" (doesn't seem to happen) is conceptually important and empirically fuzzy.