Summary
Most people learn in the wrong ways. Rereading, highlighting, and cramming — the most popular study strategies — are among the least effective. The strategies that actually work (retrieval practice, spacing, interleaving) feel harder, slower, and less productive, which is precisely why learners avoid them. This page distils the cognitive psychology research compiled in Make It Stick (Brown, Roediger & McDaniel, 2014) into the core principles and strategies that produce durable learning.
This page is intentionally separate from the game design wiki. Where connections to game design exist (e.g., fun-as-learning, interaction-loops), they are noted but not developed here.
The core strategies
1. Retrieval practice (the testing effect)
Recalling facts, concepts, or procedures from memory — rather than rereading them — is the single most powerful learning strategy. Every act of retrieval strengthens the memory and interrupts forgetting.
Key evidence (all from Brown, Roediger & McDaniel, Make It Stick, ch. 2 “To Learn, Retrieve”, with the original studies named in the book’s chapter notes):
- Testing vs restudying prose. College students who read science passages and then took a recall test remembered more two days later than students who restudied (68% vs 54%), and the advantage held a week later (56% vs 42%) (Roediger & Karpicke, 2006, Psychological Science, as reported in Make It Stick). The book also quotes a 2010 New York Times report of this study (Carey, “Forget what you know about good study habits”) saying tested students retained “50 percent more” a week later. That headline figure is the newspaper’s summary; the single-test comparison the book itself reports (56% vs 42%) is closer to a third more.
- Cramming vs retrieval. In a 1978 study described in the same chapter, crammers did better on an immediate test, but two days later they had forgotten 50% of what they had recalled initially, while students who spent the same time practising retrieval had forgotten only 13%. The chapter notes cite Thompson, Wenger & Bartling (1978, Journal of Experimental Psychology: Human Learning and Memory) as a word-list study in which massed study beat retrieval on an immediate test but not on a delayed one. Note that this is a retrieval-versus-restudy result, not a spacing result.
- Repeated tests. After a story naming sixty objects, students tested once immediately recalled 53% then and 39% a week later; students tested three times still recalled 53% a week later, “immunised” against forgetting (Wheeler & Roediger, 1992, as described in Make It Stick). The book adds that multiple retrieval sessions are generally better than one, “especially if the test sessions are spaced out”.
- Classroom evidence. At Columbia Middle School, Illinois, eighth-grade science students averaged 92% (A−) on quizzed material vs 79% (C+) on material that had not been quizzed (Roediger, Agarwal, McDaniel & McDermott, 2011; McDaniel et al., 2011).
- Early studies. In a 1917 study, children who spent about 60% of study time reciting retained the most (Gates, 1917); in a 1939 study of over 3,000 Iowa sixth graders, once a student had taken a test, forgetting nearly stopped (Spitzer, 1939).
All of the above are as summarised in Brown, Roediger & McDaniel, Make It Stick (see source-make-it-stick).
Practical forms: Low-stakes quizzes, flashcards, self-testing, practice problems, writing from memory, explaining concepts aloud without notes.
Retrieval practice for tools and procedures. The same principle applies to software skills such as setting up a Shader Graph, a Rigidbody controller or an Animator. Rebuilding last week’s graph from a blank canvas is retrieval practice; reading the documentation or a tutorial again is rereading. A sensible sequence (an application of the evidence above, not a study of Shader Graph itself) is:
- Attempt from memory first. Open a new graph and rebuild the property setup (exposed properties, their types and defaults, how they connect) without looking anything up. Expect gaps; the effortful attempt is what strengthens memory (see generation below).
- Then use the docs as feedback. Check the Unity Manual or your notes only after the attempt, and only for the parts you got wrong or could not recall. Make It Stick reports that feedback strengthens retention more than testing alone (Butler & Roediger, 2008, as described in ch. 2).
- Repeat after a gap. Rebuild it again a few days later rather than straight away, so the second attempt requires reconstruction from long-term memory (see spaced practice below).
Why it works: Retrieval reconstructs learning from long-term memory (not short-term), strengthening neural pathways and creating multiple retrieval routes. It also provides accurate feedback about what you know and don’t know.
2. Spaced practice
Distributing study over time — with enough gap for some forgetting to set in — produces stronger and more durable learning than massed practice (cramming).
Key evidence:
- Spaced surgery lessons. Thirty-eight surgical residents took four short microsurgery lessons, either all in one day or a week apart. A month later, the spaced group outperformed the one-day group on every measure (time, hand movements and success at reattaching vessels), and 16% of the one-day group damaged the vessels beyond repair (Moulton et al., 2006, Annals of Surgery, as described in Brown, Roediger & McDaniel, Make It Stick, ch. 3, see source-make-it-stick).
- Spacing retrieval. Repeated tests help most when the test sessions themselves are spaced out (Make It Stick, ch. 2). The book’s ch. 3 notes point to Cepeda et al. (2006, Psychological Bulletin) for a quantitative review of the spacing effect.
- Sleep plays a large role in memory consolidation — practice with at least a day between sessions is beneficial
- The optimal interval: enough that retrieval requires effort, but not so much that you’re relearning from scratch
Practical implementation: the Leitner box. Four boxes of flashcards. Box 1 (most errors) is practised most frequently. When you get a card right, it moves to a less-frequent box. When you get it wrong, it moves to a more-frequent box. The better your mastery, the less frequent the practice — but it never disappears entirely.
Key insight: Cramming produces fast visible improvement (“momentary strength”) that feels like learning but is transitory. Spacing produces slower visible improvement but builds “underlying habit strength” — the kind of learning that lasts.
3. Interleaving
Mixing the practice of different skills, problem types, or subjects within a single session — rather than practising one type exhaustively before moving to the next (blocked practice).
Key evidence:
- Cal Poly baseball: players who practised on randomly interleaved pitches (fastball/curveball/changeup mixed) significantly outperformed those who practised 15 of each type in sequence — despite feeling less proficient during practice
- Interleaving the identification of bird types or oil painters improves both the ability to learn unifying attributes within a type and to discriminate between types
- Computing volumes of different geometric solids: college students were first taught all four solids, then practised either in blocks or mixed. During practice the blocked group scored higher (89% vs 60%), but on a test a week later the interleaved group averaged 63% against 20% for the blocked group (Rohrer & Taylor, 2007, Instructional Science, as described in Brown, Roediger & McDaniel, Make It Stick, ch. 3, see source-make-it-stick)
Why it works: Interleaving forces you to discriminate between problem types and select the right solution strategy — exactly the skill you need in real situations where problems arrive unpredictably. Blocked practice lets you coast on short-term memory without developing discrimination.
When interleaving helps most, and when it doesn’t
Make It Stick presents interleaving as almost always better than blocking. Later research qualifies that. The qualifications below come from the interleaving literature, not from Make It Stick.
- Similarity matters. A meta-analysis of 59 studies (238 effect sizes) found a moderate overall interleaving effect (Hedges’ g = 0.42), strongest for paintings and other visual material (g = 0.67) and smaller for mathematics tasks (g = 0.34). Interleaving helped more when the categories were similar to each other and when the material was complex; for lists of words, blocking was better (g = −0.39), and results for expository texts were unclear. The authors conclude that interleaving works but “the setting and the type of learning material must be considered” (Brunmair & Richter, “Similarity matters: A meta-analysis of interleaved learning and its moderators”, Psychological Bulletin, 2019).
- Confusable problems are the point. Rohrer argues that blocking causes errors when concepts are similar, because the learner never has to tell them apart; interleaving forces that discrimination (Rohrer, “Interleaving helps students distinguish among similar concepts”, Educational Psychology Review, 2012). He calls the evidence “limited” but concludes that “at least a portion of the exposures should be interleaved”.
- Blocking can win for very different categories. In category-learning experiments, interleaving gave better generalisation for categories with high similarity, while blocked study was better for categories with low similarity, because blocking helps learners find what members of one category have in common (Carvalho & Goldstone, “Putting category learning in order”, Memory & Cognition, 2014).
- Interleaving always contains spacing. When problems of one kind are mixed with others, repetitions of each kind are automatically spread out. Taylor and Rohrer held spacing constant and still found that mixing four kinds of maths problems doubled children’s test scores a day later; an error analysis suggested the gain came from better pairing of each problem with the right procedure (Taylor & Rohrer, “The effects of interleaved practice”, Applied Cognitive Psychology, 2010). Foster and colleagues, however, found that alternating volume problems with unrelated problems (fractions, permutations) helped about as much as alternating them with each other, which supports a spacing explanation (Foster et al., “Why does interleaving improve math learning?”, Memory & Cognition, 2019). Rohrer, Dedrick and Burgess also found an interleaving benefit for superficially dissimilar maths problems (72% vs 38%), which they attribute partly to strengthening the link between each problem type and its strategy (Psychonomic Bulletin & Review, 2014).
What this means in practice (interpretation). Arguably the safest reading is: interleave problem types that you could mistake for each other, because that is where the discrimination benefit lies. Mixing unrelated skills in one session (shaders, then rigging, then material setup) is not harmful, but most of what it buys you is spacing, because there is nothing to confuse; you could get the same benefit by spacing each skill across the week. In the classroom studies, students met each new problem type in a lesson first, and the interleaving happened in the practice that followed (Rohrer & Taylor, 2007; Taylor, The Benefits of Interleaving Different Kinds of Mathematics Practice Problems, 2008), so a short blocked introduction to a brand-new type is consistent with the evidence. Rohrer and colleagues also note there is “little or no long-term benefit from working more than a few problems of the same kind in immediate succession” (Rohrer et al., “Unanswered questions about spaced interleaved mathematics practice”, Journal of Applied Research in Memory and Cognition, 2020).
4. Desirable difficulties (Bjork & Bjork)
Short-term impediments that slow apparent learning but make it stronger, more precise, and more enduring. The term was coined by psychologists Elizabeth and Robert Bjork.
The core desirable difficulties are:
- Spacing — letting some forgetting occur between practice sessions
- Interleaving — mixing different types within a session
- Generation — attempting to solve a problem before being taught the solution
- Variation — practising in different contexts and conditions
The generation effect: Attempting to produce an answer before seeing it — even if you get it wrong — leads to better learning than passively reading the answer. Example: filling in foot-s_ _e produces better retention of “shoe” than reading foot-shoe.
The paradox: The strategies that feel most productive (massed, blocked, repetitive practice) produce the least durable learning. The strategies that feel least productive (spaced, interleaved, generative practice) produce the most.
5. Elaboration
Giving new material meaning by expressing it in your own words and connecting it with what you already know.
- There is no known limit to how much you can learn through elaboration (unlike rote memorisation, which quickly hits capacity)
- The more connections you create between new learning and prior knowledge, the more retrieval cues you have later
- Examples: relating heat transfer principles to warming your hands on a cup of cocoa (conduction), sun pooling in a room (radiation), air conditioning blast (convection)
Practical forms: Explaining concepts in your own words, finding real-world examples, connecting new material to personal experience, asking “how does this relate to what I already know?“
6. Reflection
Reflection is a form of retrieval practice combined with elaboration: What happened? What did I do? How did it work out? What would I do differently?
The neurosurgeon Mike Ebersold describes how he would go home after a difficult surgery and think through what happened, what he could improve, and mentally rehearse modifications — then try them the next day. This cycle of reflection → mental rehearsal → application built novel techniques that became reflexive under pressure.
Spacing and interleaving together (worked example: learning Shader Graph)
Students often ask whether they should use spaced repetition or interleaving, as if they were rival methods. They are not alternatives. Spacing is about when you return to a skill (with gaps, across days); interleaving is about what order you practise things in within a session (mixed rather than in blocks). Rohrer and colleagues describe the effective schedule for maths as problems of one kind “distributed across many assignments spaced weeks apart” and “problems of different kinds … interleaved within the same assignment” (Rohrer et al., “Unanswered questions about spaced interleaved mathematics practice”, 2020). Add retrieval practice (attempting from memory before looking anything up) and you have the three strategies working together.
How to combine them (interpretation of the evidence above):
- Meet a brand-new node or concept in a short block. Follow one tutorial or worked example on, say, the Fresnel Effect node and build it two or three times. This is the “lesson” that preceded interleaved practice in the studies; it is not where long-term learning happens, so keep it short. This qualification comes from the interleaving literature, not from Make It Stick.
- Interleave once each piece can be done on its own. Mix effects that are easy to confuse because they use overlapping nodes for different jobs: for example a dissolve (noise compared against a threshold, driving Alpha Clip), a UV scroll (Time feeding Tiling And Offset) and a rim glow (Fresnel Effect into Emission). Choosing which construction a brief needs is the discrimination skill interleaving trains.
- Mix in the confusable choices, not just different topics. “Toon vs PBR vs procedural” is a good interleaving set, because each asks a different answer to the same question (“how should this surface respond to light?”). “Shader vs rig vs material setup” in one session mainly gives a spacing benefit; that is fine, but do not expect it to teach you to tell shader techniques apart.
- Space the sessions. Return to each effect after a gap of a day or more, so the next attempt needs real recall.
- Start each session with retrieval. Rebuild last session’s graph from a blank canvas before opening any notes, then use the docs as feedback (see Retrieval practice for tools and procedures above).
A sample four-week plan (three 45-minute sessions a week; illustrative, not a tested programme):
| Week | Session A | Session B | Session C |
|---|---|---|---|
| 1 | Block: follow a dissolve example; rebuild it twice | Block: UV scroll example; then rebuild Monday’s dissolve from memory | Block: Fresnel rim glow; then rebuild scroll from memory |
| 2 | Interleave: three one-line briefs (“a ghost that fades at the edges”, “lava that flows”, “an enemy that burns away”); pick and build the right effect for each, no notes | Retrieval: rebuild all three from blank graphs; check the docs only for what you missed | New block: meet a toon-style stepped lighting ramp briefly |
| 3 | Interleave: mixed briefs drawn from all four effects, in random order | Variation: combine two effects in one graph (scrolling dissolve) | Spaced retrieval: rebuild the week 1 effects you have not touched for a fortnight |
| 4 | Interleave: mixed briefs including a PBR-style material set-up for contrast | Self-test: explain aloud why each brief needs its technique (elaboration) | Review: list what you could not rebuild from memory; that list becomes next month’s practice |
What to expect. Practice will feel slower and less fluent than repeating one effect ten times. That is the “momentary strength” versus “underlying habit strength” trade-off described above: the blocked approach looks better during practice and worse on a delayed test (Rohrer & Taylor, 2007, as described in Make It Stick, ch. 3). See shader-graph-overview for the nodes themselves.
How learning works: encoding → consolidation → retrieval
Encoding: Sensory perceptions are converted into mental representations (memory traces) in short-term memory. Think of notes jotted on a scratchpad.
Consolidation: The brain reorganises and stabilises memory traces for long-term storage. This involves deep processing: replaying the learning, filling in gaps, making connections to prior knowledge. Sleep plays a significant role. The process takes hours to days. An analogy: writing an essay draft, setting it aside, then revising with greater clarity.
Retrieval: Calling up learning from long-term memory. Each successful retrieval strengthens the memory traces and makes them modifiable again (reconsolidation), allowing them to connect to newer learning. This is why retrieval practice is so powerful — it doesn’t just test learning, it changes learning.
Key implication: If you practice something over and over in rapid succession, you’re leaning on short-term memory and very little consolidation or reconsolidation occurs. The visible improvement is real but temporary. When you space practice and allow some forgetting, retrieval requires reconstruction from long-term memory, which triggers reconsolidation and produces durable learning.
What doesn’t work (and why people do it anyway)
Rereading
The #1 study strategy of college students (80%+ in some surveys). Three strikes against it:
- Time-consuming
- Doesn’t result in durable memory
- Creates an illusion of knowing — growing familiarity with the text feels like mastery of the content
Massed practice (cramming)
“Practice-practice-practice” of a single skill or topic. Produces fast visible gains that are gratifying but transitory. Likened to “binge-and-purge eating — a lot goes in, but most of it comes right back out.”
Highlighting and underlining
Passive engagement with surface features of text. No retrieval, no elaboration, no generation. Feels productive; accomplishes little.
Illusions of knowing and metacognition
Metacognition is what we know about what we know. Accurate metacognition is critical for effective learning — you need to know what you don’t know in order to focus your effort.
The fluency illusion: Rereading a text creates fluency (ease of processing) that is mistaken for mastery of the underlying concepts. Students who can repeat phrases from a lecture often cannot apply the ideas in a new context.
The Dunning-Kruger effect: Incompetent performers overestimate their competence and — critically — lack the skills to recognise their own incompetence. Studies showed that students scoring at the 12th percentile believed their logical reasoning was at the 68th percentile. However, when these students were taught the skills to evaluate logical reasoning, their self-assessment became dramatically more accurate.
Kahneman’s System 1 and System 2: System 1 (fast, automatic, intuitive) is powerful but susceptible to illusion. System 2 (slow, deliberate, analytical) is needed to check System 1’s conclusions. Pilots trained to trust instruments over vestibular sensation are applying System 2 to override System 1. The China Airlines Flight 006 incident (1985) is a dramatic example of what happens when System 1 overrides instrument readings.
Calibration: The remedy for illusions of knowing is regular self-testing. Testing doesn’t just measure learning — it reveals gaps. The military concept of “shooting an azimuth” (climbing to a height to check your heading) is an apt metaphor for using practice tests to recalibrate understanding.
Learning styles: the myth
The claim that people learn better when instruction matches their preferred style (visual, auditory, kinesthetic) is not supported by empirical research. While people do have preferences, and there are indeed multiple forms of intelligence, the evidence shows that learners benefit most from “going wide” — drawing on all their aptitudes rather than limiting instruction to one modality.
What does matter is matching the nature of the content to the appropriate modality: learning geography benefits from maps (visual); learning a language benefits from speaking and listening (auditory); learning surgery benefits from practice with hands (kinesthetic). This is about the content, not the learner.
Growth mindset and neuroplasticity
The fixed mindset fallacy: Many people believe intellectual ability is hardwired from birth, and failure is an indictment of native ability.
The evidence: The brain physically changes with learning. Neurons sprout axons that connect to dendrites of other neurons, forming synapses. Myelin sheaths develop around well-used pathways, speeding signal transmission. The fine structure of neural networks is shaped by experience and is capable of substantial modification throughout life.
Implication: Failure is a “badge of effort and a source of useful information — the need to dig deeper or to try a different strategy.” Struggle is not a sign that learning is failing; it is a sign that learning is happening. This aligns with the desirable difficulties principle: when learning feels hard, you’re doing the most important work.
Connections to game design
These connections are noted briefly for cross-reference; the learning science stands on its own.
- fun-as-learning — Koster’s theory that fun = pattern recognition/learning is empirically supported by the retrieval practice and mental model research. Games that force players to recognise patterns and retrieve solutions under pressure are doing exactly what cognitive science recommends.
- interaction-loops — The LDARF loop (Learn/Decide/Action/Rules/Feedback) mirrors the retrieval-feedback cycle. Each game challenge is effectively a retrieval test with immediate feedback.
- flow — Desirable difficulties must be calibrated to the learner’s skill level — too easy and no effortful retrieval occurs; too hard and retrieval fails entirely. This is the same challenge/skill calibration that produces flow.
- meaningful-decisions — The generation effect (attempting to solve before being taught) maps directly to games that present problems before tutorials.
- Games as natural learning environments — Games inherently provide retrieval practice (challenges), spacing (between sessions), interleaving (mixed enemy types, varied puzzles), generation (puzzles before solutions), and immediate feedback on performance. This may explain why game-based learning can be effective when it works.
Evidence
Content drawn from Brown, Roediger & McDaniel, Make It Stick (Harvard, 2014). Key studies cited: Tulving (1967) on repetition and learning; Roediger & Karpicke (2006) on testing vs rereading; McDaniel, Agarwal et al. (2011) on the Columbia Middle School classroom study; Bjork & Bjork on desirable difficulties; Kahneman (2011) System 1/System 2; Dunning & Kruger (1999); Dweck on growth mindset; Doidge (2007) on neuroplasticity. The research programme was funded by the James S. McDonnell Foundation and Washington University’s Center for Integrative Research in Learning and Memory.
Figures in this page were checked against the book’s text and chapter notes on 2026-10-25. Two earlier figures were corrected: the “~50% better retention” figure is a New York Times summary that the book quotes, not a result the book reports, and the “50% vs 13% forgotten” figure compares cramming with retrieval practice (a 1978 study in ch. 2), not spaced with massed practice.
The qualifications on interleaving draw on research published after or alongside the book: Rohrer & Taylor (2007), Taylor & Rohrer (2010), Rohrer (2012), Carvalho & Goldstone (2014), Rohrer, Dedrick & Burgess (2014), Brunmair & Richter (2019), Foster et al. (2019) and Rohrer et al. (2020), as cited inline above (abstracts and, for Brunmair & Richter, the accepted manuscript were read).
Related
- fun-as-learning — Koster’s game design theory of fun as pattern learning; empirically supported by this research
- interaction-loops — The LDARF loop parallels retrieval-feedback cycles
- flow — Desirable difficulty calibration parallels challenge/skill balance
- shader-graph-overview — the nodes used in the worked study plan
- source-make-it-stick