The Evidence on Teaching

Background knowledge and vocabulary as drivers of reading comprehension

Comprehension is knowledge — but vocabulary teaching moves standardized comprehension only d≈0.10. The big content-knowledge bet (Core Knowledge lottery, 0.24) is real and unreplicated.

moderate supportconf: mediumgc: low

reading · ages 514

Effect summary

Once decoding is fluent, comprehension tracks language, vocabulary, and knowledge — but these transfer weakly as instructional levers. Vocabulary instruction moves standardized comprehension only d≈0.10 (vs 0.50 on taught-word tests). The one strong, genetic-confound-resistant result is a Core Knowledge kindergarten lottery: ITT 0.24 / TOT 0.47 SD on independent state reading tests — but it is a single unreplicated bundled study.

Practical takeaway

Build broad vocabulary and content knowledge cumulatively over years (a content-rich curriculum) rather than expecting quick comprehension fixes. Teaching specific words reliably helps comprehension OF those topics; it transfers weakly to general reading.

Who this applies to

Not yet assessed. Nobody has recorded the group size, dose, delivery, or boundary conditions for this decision, so it should not be recommended for a specific situation yet — only read. That is a gap in this record, not a claim that it applies everywhere.

Verdict

The "simple view of reading" is correct that once decoding is adequate, comprehension is governed by language comprehension — vocabulary breadth, background knowledge, and inference. The hard part is causal: these are exactly the variables most confounded by heritable ability and family selection, so the correlational case (which is strong and consistent) cannot by itself establish a teachable lever. When you restrict to causal, genetic-confound-resistant evidence, the picture is: vocabulary instruction transfers weakly to general comprehension, and content-knowledge building looks like the most promising durable lever but rests on essentially one strong study. Hence moderate-support at medium confidence.

What the evidence shows

Source Design Grade Key effect
Grissmer 2023 K lottery, Core Knowledge charters B Grade 3-6 reading on independent state tests ITT 0.24 / TOT 0.47 SD; up to ITT 0.94 at a low-income CK charter
Elleman 2009 37-study vocabulary meta C Comprehension: custom d=0.50 vs standardized d=0.10 (~5x gap); ~3x larger for reading-disabled
Elleman & Oslund 2019 narrative review D Predictor hierarchy: vocabulary > inference > knowledge (correlational)
Recht & Leslie 1988 "baseball study" (n=64) D Domain knowledge can outweigh reading ability on topic-matched recall — but unreplicated, restricted-range

Two things are true at once. Teaching vocabulary reliably raises comprehension of passages using the taught words (d=0.50) but barely moves standardized comprehension (d=0.10) — a ~5× gap, and the study-level correlation between vocabulary gains and comprehension gains is only r≈0.43. So vocabulary is a teachable domain skill, not a general far-transfer lever. The famous Recht & Leslie "baseball study" (high-knowledge poor readers out-recall low-knowledge good readers) is an existence proof of near-transfer within a known domain, not evidence that teaching knowledge raises general reading — and it is grade D (n=64, single custom text, unreplicated in 30+ years, key interaction possibly non-significant).

The strong result is Grissmer's Core Knowledge kindergarten lottery: random assignment to a content-rich (E.D. Hirsch) curriculum raised grade 3-6 reading on independent state tests by ITT 0.24 / TOT 0.47 SD, with a gap-closing effect at a low-income charter. Because it is lottery-identified and uses standardized outcomes, it resists both the measure-inflation and the genetic-confound critiques — its central strength. Caveats keep it from "strong-support": it is a single unreplicated working paper, on a whole-school bundle (curriculum + peers + culture, so the knowledge-specific share is unisolated), skewed to middle-income families. The Kim MORE content-literacy RCT converges more weakly (~0.18, near > far).

Hereditarian-lens assessment

Risk: low for the verdict, because it is anchored on the lottery (Grissmer) and the experimental vocabulary meta, both of which defeat passive gene-environment correlation. But this is the topic where the lens matters most for what we exclude: the entire correlational predictor literature (vocabulary/knowledge/inference → comprehension) is grade D and presumptively confounded — the children with bigger vocabularies and more knowledge differ genetically and by family environment from those without. That literature sets instructional targets but cannot establish causal levers. The database deliberately does not let it drive the verdict; the modest, honest causal signal is what remains.

Fadeout & durability

Notably, the two most durable signals in the whole reading domain are on the language/knowledge side, not the decoding side: comprehension-oriented interventions maintain better than phonics in Suggate 2016, and Grissmer's effects are measured years out (grades 3-6) precisely because knowledge accumulates slowly and cumulatively. This is the flip side of decoding's fast fadeout: knowledge is a slow lever, but what it builds appears to stick better.

What critics say / limits

  • Enthusiasts over-cite Recht & Leslie and the correlational predictor hierarchy as if they were causal; they are not.
  • Grissmer is one study; the knowledge-building hypothesis needs replication before it can carry a strong verdict, however theoretically attractive.
  • Vocabulary's weak standardized transfer (0.10) is a caution against "teach more words → better readers" as a general strategy.

Practical guidance

  • Favor a knowledge-rich, coherent, cumulative curriculum over generic skill practice — this is a cheap curriculum-choice lever with the best (if thin) causal support for durable comprehension.
  • Teach vocabulary in service of the content being read; expect it to help comprehension of that material more than reading in general.
  • Play the long game: comprehension is built over years of broad knowledge and language exposure, not by a comprehension unit. Budget for breadth of content across the curriculum.

Open questions

  • The isolated causal effect of knowledge (vs peers/culture/selection in the CK bundle) is unidentified; replication of Grissmer is the single most valuable next study in this area.
  • Whether explicitly teaching general knowledge (vs it accruing from a rich curriculum) is the operative mechanism is unresolved.

Evidence (5 sources)

Export all: BibTeX · RIS

Related decisions

← Back to explore