Willingham — Critical Thinking, Why Is It So Hard to Teach

Cognitive psychologist Daniel T. Willingham's review of decades of research on why "critical thinking" resists being taught as a general skill: thinking is intertwined with domain knowledge. Arts Education Policy Review 109(4), 21–32 (2008), reprinted from American Educator (AFT), Summer 2007. Human-added to raw/ 2026-11-08. Opens the wiki's thinking & learning domain — and is the first source to bear directly on the wiki's own method (Method Reflection).

Summary

The goal everyone agrees on (students who think critically) has been translated into calls to teach generic "critical thinking skills." After A Nation at Risk (1983) these programs boomed: by 1990 most states had initiatives, and Tactics for Thinking sold 70,000 teacher guides. They didn't work well, and 20+ years later the lament continues. Willingham's answer to "can critical thinking be taught?" is "not really" — not as a free-standing skill. The processes of thinking are intertwined with the content of thought. You can teach maxims about how to think, but without background knowledge and practice students can't implement them (Critical Thinking Is Domain-Bound).

Definition (Appendix A). To a cognitive scientist, "critical thinking" is a subset of three ordinary kinds of thinking: reasoning, judgment/decision-making, and problem solving. Each counts as critical when it has three features:

• Effective: avoids known pitfalls (one-sidedness, discounting disconfirming evidence, reasoning from passion, unsupported claims).
• Novel: not just retrieving a remembered solution. Running a familiar multi-step algorithm isn't critical thinking; devising a new algorithm is.
• Self-directed: the thinker calls the shots, not a teacher prompting each step.

Why it's hard: surface vs. deep structure

Rigorous thinking "clings to particular examples." The student who estimates in math doesn't notice their chemistry percentages sum past 100%. The one who weighs British and American views of the Revolution never asks how Germans saw WWII. Mechanism (Surface vs Deep Structure): comprehension interprets everything through prior knowledge and narrows interpretations to fit context. "The cognitive system gambles that incoming information will be related to what you've just been thinking about." That makes comprehension fast but hides deep structure. The relevant-seeming knowledge is about the scenario (bands, gardens), not the math.

• Band problem (Novick & Holyoak 1991): only 19% spontaneously used a just-studied garden problem with the same least-common-multiple solution. Told "it's similar to the garden problem," 35% solved it. Most still couldn't, even when told what to do.
• Hansel & Gretel (Chen, Mo & Honomichl 2004): ~75% of American students solved a cave/trail puzzle vs. 25% of Chinese students. The split reversed on a puzzle built on a Chinese folk tale. Deep familiarity makes deep structure perceptible: the Americans saw "leave a trail," not "sand, caves, treasure."

Two things aid transfer: (1) deep familiarity with a problem type, from long practice across many surface variants, and (2) knowing to look for deep structure. The second is a metacognitive strategy, and it "can only take you so far."

Scientific thinking: a little easier, still knowledge-bound

Most researchers treat scientific thinking as ordinary reasoning, not a different kind of thought. What makes it scientific is knowing when to deploy it, plus enough knowledge and practice to do it.

• Conditional probability: 3-year-olds infer which block is a "blicket" from patterns of covariation (Gopnik et al. 2001; Fig. 1 via Gopnik & Schulz 2004). Yet adults fail many conditional-probability problems. Physicians discount data that conflict with their working diagnosis (Groopman 2007). PhD scientists reason badly in unfamiliar contexts. "Children are not as dumb as you might think, and adults (even trained scientists) are not as smart as you might think."
• Teaching when: taught the control-of-variables strategy explicitly with springs, 2nd–4th graders applied it seven months later with new materials and a new experimenter (Chen & Klahr 1999). Older children transferred more robustly. The cue "I'm designing an experiment" triggered the strategy. But succeeding at controlling variables depends on knowing which variables matter.
• The NRC (2007): "Teaching content alone is not likely to lead to proficiency in science, nor is engaging in inquiry experiences devoid of meaningful science content."
• Knowledge shapes every step: hypothesis choice runs on plausibility, a domain judgment (Koslowski 1996: paint colour becomes worth testing once you hear the teenager drives differently in his red car). 8th-graders designed better-controlled experiments about creature survival than about pool cooling, because the variables were familiar (Friedler et al. 1990). Undergraduates with more integrated prior knowledge of circuits planned more informative experiments and interpreted results better (Schauble et al. 1991). Anomalies drive new knowledge but are only recognizable against an expectation, which is domain knowledge (Dunbar & Fugelsang 2005).
• Meta-analysis of 40 experiments on teaching science problem solving (Taconis et al. 2001): effective methods built integrated knowledge bases (e.g. concept mapping). Strategy-only methods were ineffective.

Footnote ††: for people with extensive training (PhD-level), critical thinking does acquire skill-like breadth across less-familiar content. They still make mistakes.

Generic thinking programs (Appendix B)

Programs like Feuerstein's Instrumental Enrichment, Covington's Productive Thinking, de Bono's CoRT, and Philosophy for Children run ~3 years at several hours a week, on the premise that thinking skills are domain-general. The evidence is weak:

• Typical study flaws: (1) a single post-test, so durability is unknown; (2) no control group; (3) a control with no comparison intervention, so novelty or teacher enthusiasm could explain gains; (4) no transfer measure; and (5) rarely peer-reviewed. Researchers "unconsciously bias the design and analysis" toward hoped-for conclusions.
• Philosophy for Children (Trickey & Topping 2004): 8 minimally adequate studies (1 peer-reviewed), 3 of them clouded. Of the rest, reading gains in 1 of 3 and reasoning gains in 2 of 3.
• Consensus: some modest positive effect, with two caveats. It depends on teacher skill (someone must give feedback), and it fades as test material diverges from practice material. "You learn to think critically in the ways in which you practice thinking critically." Curricula spread practice over many problem types, so no type gets enough. The modest benefit is probably the metacognitive strategies.

What to do instead (Appendix C)

1. Special programs aren't worth it: weigh the modest boost against opportunity cost.
2. Teach critical thinking inside subject matter. Model it and practise it in normal class work. Example: "consider the source" is useless on a Confederate private's post-Vicksburg letter without Civil War knowledge.
3. It's not just for advanced students. Everyone does it from early childhood. The difficulty is knowing when, and knowing enough.
4. Use student experience as an entrée when subject knowledge is thin. Would a letter against abolishing recess read differently from a principal than from a third-grader? Ice cream and crime both rise with heat: correlation ≠ causation.
5. Make strategies explicit and practise them in stages. Introduce with at least two different examples, label the strategy as reusable, show how it applies to the content at hand. Later, name the strategy and let students work out the application. Eventually students pick the strategy uncued.

Key claims

• Critical thinking is not a skill in the sense of being deployable at any time in any context. It is a type of thought that 3-year-olds can do and trained scientists can fail at.
• Teaching it "lies in small part in showing them new ways of thinking, and in large part in enabling them to deploy the right type of thinking at the right time."
• Metacognitive strategies "only tell the students what they should do — they do not provide the knowledge that students need to actually do it."
• Three-part bottom line: (1) not a skill; (2) learnable metacognitive strategies make it more likely; (3) actually doing it depends on domain knowledge and practice.

• Tension with this wiki's method: much of the wiki's engineering-craft domain consists of distilled maxims (Working Heuristics — Data, Matching, Product, Org, Tooling). Willingham predicts maxims act as cues, not capabilities. See Method Reflection (T1).

Connections

• Critical Thinking Is Domain-Bound — the central thesis, as a concept page.
• Surface vs Deep Structure — the transfer-failure mechanism (band/garden, Hansel & Gretel).
• Metacognitive Strategies — maxims that cue thinking but don't enable it; the staged teaching recipe.
• Method Reflection — this source puts the wiki's own maxim-distilling method under question.
• Working Heuristics — Data, Matching, Product, Org, Tooling — a list of exactly the kind of maxims Willingham says don't transfer without knowledge and practice.
• Falsification Is Asymmetric — Willingham's pitfalls (seeking only confirming evidence, physicians discounting disconfirming data) are what the asymmetry guards against; anomalies need prior expectations.
• The Bottleneck Shift / AI Commoditizes Expertise — gives the cognitive mechanism for why judgment is the scarce human input: judgment can't be separated from domain knowledge.
• What's Scarce When Execution Is Cheap — sharpens its open question: if knowledge is offloaded to AI, does knowledge-bound judgment erode?

Source

• Original: raw/1490.pdf (filename is the publisher's; left unrenamed because raw/ is immutable). DOI 10.3200/AEPR.109.4.21-32.
• Two pages of the PDF are publisher advertisements (ProQuest) with no article content.