Reading comprehension strategies: what the evidence shows.

Schools timetable reading comprehension like a subject. Tests score it like a skill. Curricula drill it like a muscle that strengthens with reps. The research behind all three habits tells a stranger story. A small toolkit of mental moves pays off almost immediately, then stops paying soon after. The long-term work belongs to something schools often treat as a separate matter entirely: what children know.

The skill that isn’t one

Start with what understanding a text actually is. In the standard cognitive account, a reader builds a mental model of the situation the text describes — who did what, where, why, what follows (Kintsch, 1998). The model is assembled partly from the words and partly from knowledge the reader already holds. The words are instructions for building the model. They are never the whole model.

That is why two children can read the same sentence and understand different amounts. “The batter checked his swing” is a complete scene to one reader and a grammar puzzle to the other. Nothing about their reading mechanics differs; their model-building materials do. Comprehension, in this account, is not one skill. It is an outcome — of decoding, language, attention and, above all, knowledge (Kintsch, 1998).

The strategy movement grew from the other half of the account: attention. Weak comprehenders often read on autopilot — eyes moving, model unbuilt, no alarm ringing. If part of the failure is unmanaged attention, then the management can be taught. That idea produced the most studied intervention in comprehension research.

The strategy toolkit

In 1984, Palincsar and Brown taught struggling seventh-grade comprehenders four moves: predict what comes next, generate questions, clarify confusions, summarize what was read. The method, reciprocal teaching, ran as a dialogue — teacher and a small group taking turns leading the discussion with the four moves, the adult handing over control as the children got better (Palincsar & Brown, 1984). In the original studies, comprehension scores rose dramatically and the gains held for weeks after training ended.

A research industry followed. By 2000 the National Reading Panel could review dozens of controlled studies across sixteen categories of comprehension instruction (National Reading Panel, 2000). Roughly half earned firm scientific support: comprehension monitoring, question generation and answering, summarization, story structure, graphic organizers, cooperative formats, and multi-strategy packages like reciprocal teaching.

How the toolkit is taught matters as much as the list. In the studies that worked, the teacher modelled each move out loud on a real text — showing what a summary sounds like, what a genuine question looks like — and then handed the moves to the group in guided practice (Palincsar & Brown, 1984). The strategy was never the lesson’s content for long. The text was. That design choice quietly foreshadows how this story ends.

Notice what the supported list has in common. Every entry is a way of making the reader do something with the text — interrogate it, compress it, map it, explain it to a peer. The toolkit works by refusing to let attention idle. The question the field took longer to ask was how much of this toolkit there is to learn.

The verdict: strategies work

The cleanest pooled numbers come from Rosenshine and Meister’s meta-analysis of reciprocal teaching — a meta-analysis pools many studies into one estimate. The pooled effect was roughly d ≈ 0.32 on standardized comprehension tests, and about 0.88 on tests the experimenters built themselves (Rosenshine & Meister, 1994).

The gap between those numbers is itself a finding. Experimenter-designed tests sit close to the training — a summarization program, measured by asking for summaries, shines. Standardized tests sample comprehension broadly, on unfamiliar topics, and the advantage thins to a third of a standard deviation (Rosenshine & Meister, 1994). Both numbers are real. The smaller one is the honest forecast for report cards, and it is still worth having — few classroom interventions clear 0.3 on standardized measures at reciprocal teaching’s cost, which is near zero.

The plateau

The strategy literature contains a finding almost nobody designed for. When the pooled reciprocal-teaching studies were examined for dosage, program length barely mattered: students taught for a handful of sessions performed about as well on outcome tests as students taught for around fifty (Rosenshine & Meister, 1994).

Daniel Willingham built the uncomfortable argument on that base. Strategy instruction, he proposed, behaves like learning a trick, not building a skill. Checking whether a paragraph made sense, asking who-did-what, pausing to summarize — these are things a reader either knows to do or doesn’t. Once known, further drilling adds little, because there is nothing incremental to strengthen (Willingham, 2006). Compare decoding or fluency, where month after month of practice keeps compounding. Strategy knowledge is a one-time unlock. A valuable one — and a cheap one, which is exactly why it should be taught quickly rather than taught forever.

The plateau reframes the classroom economics. If ten sessions capture most of the benefit, the marginal cost of strategy teaching is tiny (Willingham, 2006). And the marginal cost of a multi-year strategy curriculum is enormous, because every hour of it displaces something that does compound. The question stops being does strategy instruction work. It becomes: what was the rest of that time for?

The baseball study

What fills comprehension’s long game, if not strategies? The most famous single demonstration is small, old and unreasonably clear. Recht and Leslie took junior-high students and sorted them two ways: strong or weak on a standardized reading test, and high or low on knowledge of baseball. Everyone read the same passage describing half an inning of a game, then re-enacted it with a model field and retold and summarized it (Recht & Leslie, 1988).

Reading ability lost. Weak readers who knew baseball outperformed strong readers who did not — recalling more, re-enacting the play more accurately, summarizing the game better (Recht & Leslie, 1988). The strong low-knowledge readers had every measured skill and none of the materials. Their fluent decoding delivered words their models could not use. “The runner tagged up” identifies every word and explains nothing.

The situation-model account predicts exactly this. Knowledge is not a bonus applied after reading; it is a load-bearing part of the reading itself (Kintsch, 1998). It also explains an ugly testing consequence: a general comprehension score is partly a general knowledge score in disguise. A child scored weak on a passage about tides may simply never have met the sea — a distinction the score sheet does not record (Recht & Leslie, 1988).

Head to head: strategies vs content

If the baseball logic is right, a classroom that spends its reading hour talking about what the text means should do at least as well as one drilling how to think about texts. That comparison has been run. McKeown, Beck and Blake taught fifth-grade classrooms the same texts two ways across a school year (McKeown, Beck & Blake, 2009). One condition rehearsed the endorsed strategy toolkit. The other was built on open questions about the meaning of the passage itself.

The strategies did not win. On the comprehension outcomes, the content-focused classrooms matched the strategy classrooms and ran ahead on measures like narrative recall (McKeown, Beck & Blake, 2009). The authors’ interpretation is the parsimonious one: what strategy instruction mostly buys is attention to the text’s meaning, and direct discussion of the meaning buys the same thing without the apparatus.

One study, one program, one grade — the finding deserves its hedges. But its direction agrees with the dosage plateau and with the knowledge results, and that agreement across methods is what a mature reading of this literature stands on (Elleman & Compton, 2017).

The knowledge road

Elleman and Compton’s review of the field says the quiet part plainly (Elleman & Compton, 2017). Comprehension research spent decades refining generic strategy instruction. The next gains lie elsewhere — in building the knowledge, vocabulary and inference base that comprehension runs on.

The practical translation is curricular. If comprehension depends on knowledge, then the reading curriculum is the knowledge curriculum. Sequenced, coherent topic teaching — science, history, geography, the arts — is reading instruction, even when no reading skill is named on the worksheet. Text sets that stay on a topic for weeks let each text feed the next reader’s model of the subject (Elleman & Compton, 2017). Random passage-of-the-day anthologies, by the same logic, are knowledge diets of empty calories.

Vocabulary sits inside the same argument. Words are not a separate list to memorize; a word is a compressed piece of knowledge about the world. Children learn photosynthesis easily inside a month of plant study, and painfully as an isolated flashcard. The review reads the evidence the same way: vocabulary and inference teaching work best woven into content, where each new word attaches to a model already under construction (Elleman & Compton, 2017).

The strategy toolkit keeps its place on this road. Taught briefly and early, it hands children the habit of noticing when the model stops building (Palincsar & Brown, 1984). What it cannot do is supply the bricks. The literature’s two halves are not rivals; they are a sequencing instruction. Strategies first, briefly. Knowledge always.

What the evidence doesn’t show

Each half of this story tempts an overcorrection. The honest limits:

• The plateau is not a refutation. Strategy instruction works; d ≈ 0.32 on standardized tests is a real, cheap gain no school should skip (Rosenshine & Meister, 1994).
• The dosage finding is correlational. No trial randomized children to six versus fifty sessions; the plateau is inferred across studies of different lengths (Rosenshine & Meister, 1994).
• The baseball study is small and observational. A few dozen students, one topic, no random assignment — it illustrates the mechanism; the pooled literatures carry the weight (Recht & Leslie, 1988).
• Head-to-head trials are scarce. The content-versus-strategies comparison rests on few studies, in particular grades, with particular programs (McKeown, Beck & Blake, 2009).
• Knowledge-curriculum trials at scale are young. The long-run causal case for knowledge-building curricula is still being built; the review calls for exactly those trials (Elleman & Compton, 2017).
• Strategies were tested mostly in the middle grades. The pooled reciprocal-teaching evidence centres on grades 3–8; very early and very late applications are thinner (Rosenshine & Meister, 1994).

Where the evidence stops

1. 1The plateau is not a refutation
2. 2The dosage finding is correlational
3. 3The baseball study is small
4. 4Head-to-head trials are scarce
5. 5Knowledge-curriculum trials are young
6. 6Tested mostly in the middle grades

© 2026 FUTURE PROOF™

Comprehension by the evidence

Put together, the literature reads as a short schedule rather than a method war. It is cheap to follow and expensive to ignore, because the costs hide in displaced hours rather than failed lessons.

Teach the toolkit briefly, early, as dialogue. A small set — summarize, question, clarify, monitor — taught reciprocal-teaching style, with the adult handing over control (Palincsar & Brown, 1984). The endorsed list is short; use it (National Reading Panel, 2000).

Stop when the gain stops. Plan strategy instruction in weeks, then retire it into habits — a routine prompt before and after reading, not a permanent subject (Rosenshine & Meister, 1994). The freed hours are the point of the plateau (Willingham, 2006).

Spend the freed time on knowledge. Sequenced topics, sustained for weeks, across science, history and the arts — chosen so each text builds the model the next text needs (Elleman & Compton, 2017).

Choose texts for what they teach, not just their level. A topic-connected text set at moderate difficulty beats disconnected passages at a perfect difficulty, because the reader carries knowledge forward (Kintsch, 1998).

Read comprehension scores as partly knowledge scores. Before diagnosing a comprehension deficit, ask what the passage assumed the child knew (Recht & Leslie, 1988). The remedy for a knowledge gap is not a strategy lesson. It is the knowledge.