Skip to content

Game-based Learning

Principle · All principles
Evidence · 51 claims (23 for, 28 mixed) · 40 studies (13 quant-synthesis, 10 causal, 10 review, 3 associational, 3 theoretical, 1 design), q1–q4 · 15 of 40 report an effect size · 45 claims rest on one study

Conditional relationship

For a learner who has not yet reached a stated capability, a game in which the repeated move that wins is the target reasoning itself (place the counter on the numbered square, choose the next equation step, pick the trade-off) is expected to improve performance on that capability measured outside the game, when three conditions hold: the mechanic demands the target thinking and cannot be won by a shortcut; a learner without a usable strategy is shown one (a model, a worked example, an adult naming what happens); and play is followed by a task that asks for the same reasoning without the game. Two further conditions come from the syntheses: the game is played over more than one session, and it is part of a course of instruction rather than the whole of it. The expectation is against no instruction or passive instruction (reading, lecture); against active non-game instruction on the same content, the one synthesis that split its comparisons this way found no advantage for the game. The expected change is in immediate-performance and, where it has been tested, delayed-retention and near-transfer on an aligned measure. Enjoyment, time played and in-game score are separate outcomes (motivation-affect, persistence); they are not evidence of the capability, and the one comparison here that set a game against a tutoring system found them moving in opposite directions.

This page is about learning through a full game: the game is the learning activity. It is not about adding points, badges, levels or leaderboards to ordinary instruction, which is Gamification; the gamification meta-analysis's numbers and the game meta-analyses' numbers below do not transfer to each other. The design discipline the relationship depends on, choosing a game mechanic that carries the learning without adding confounds, is set out in Learning Embedded in the Core Mechanic; this page does not repeat it. No converted pattern instantiates this principle yet; Game-Based Mastery Learning and Epistemic Games are related unconverted patterns. Studying worked examples before or between rounds is modelled in Worked Examples.

Three meta-analyses in the wiki compare games with non-game conditions as a whole: digital games for K–16 learners (57 studies, g = 0.33), games in primary and lower-secondary STEM (cognition g = .67, with a prediction interval from −0.38 to 1.76) and computer-based simulation games in workplace and university training (published studies d = .52, unpublished d = −.10). They give an average over very different games, not a forecast for one: effects vary widely, shrink with secondary-school learners, depend on the comparison condition, the number of sessions and whether play is competitive, and show signs of publication bias. They do not test the page's mechanism (whether the winning move carries the target reasoning). For that the wiki has one randomized test of a game's mechanic with young children, one randomized test of adding worked examples to a puzzle game with adults, and a pooled comparison of enhanced against standard versions of the same games (g = 0.34). The default design below now rests on these where it can; the rest is the earlier page's guidance, labelled.

Default design, while the relationship is untested

The page's earlier guidance, kept as a concrete default a designer can act on and revise. Each step says what supports it; most are untested as written.

  1. Name the capability, then make the winning move require it. Write the one decision or action the learner must get better at, and check that the game's win condition cannot be reached without it (no guessing, button-mashing or reward route around it). Evidence: in one randomized study [+M] the same board game raised preschoolers' number knowledge when its squares were numbered 1–10 in a line, and not when they differed only in colour. One study, one domain; it does not show which features of other games matter.
  2. Show a novice the strategy before or between rounds. Give one or two worked examples or a modelled round of play before the learner plays unaided, or between a first and second round. Evidence: adults who studied worked examples between two rounds of a puzzle game [~M] improved more than those who did not, but still learned a small fraction of what experts showed [~M]. Across games, versions with added scaffolding (personalised hints, agents, adaptation to the learner) beat the standard version of the same game (enhanced designs [+M], g = 0.41 for scaffolding, 9 studies). How many examples is not reported on the claim pages. Fade the support in later levels as the learner succeeds unaided: carried from non-game instruction, where fading support [+M] and example–problem sequences [+M] help novices and guidance that helps novices can become redundant for the more expert [~M]; untested inside a game.
  3. Give an immediate, informative consequence for each move. The game shows what the move did and, after an error, a hint toward the target reasoning rather than only a lost life. Keep the controls and screen free of demands the target does not need, since reducing extraneous load [+M] helps in multimedia instruction (not tested in games). Otherwise untested; from the earlier page.
  4. Let learners fail safely and try again, and let them play individually by default. Several attempts per level, with no penalty that ends play. Competition is optional: across digital-game studies, single-player play with no formal competition gave the largest effects (g = 0.45) and single-player competitive play none (g = −0.06) (player configuration [~M]; after controls only the gap to single competitive play held, same analysis [~M]). If scores are shared, see the team-score design in Gamification.
  5. Spread play over several sessions and leave the game open between them. In the digital-game meta-analysis, games played over more than one session beat non-game conditions (g = 0.44) and single-session games did not (g = 0.08), while total minutes of play were unrelated to the effect (sessions [+M]; same analysis [+M]). In workplace and university simulation games, unlimited access gave a larger advantage than limited access (d = .68 against .31; unlimited access [+M]). The tested child dose was four 15–20 minute sessions over two weeks, with an adult playing alongside (board-game study [+M]); the K–16 media comparisons averaged 11 sessions over 38 days, about 350 minutes in all. Proposed arc, untested as a whole: session 1 a modelled round and short unaided play; sessions 2–4 unaided play with hints and a worked example between rounds where play stalls; later sessions harder levels with hints faded, the game left open for voluntary replay; move on when the learner succeeds at the outside-game task in step 7, not when the levels are finished. Record the dose given.
  6. Put the game inside instruction, and debrief after play. Teach the content around the game (before, between or after sessions) rather than letting the game replace it, and ask learners to state the rule or strategy that won, why it worked and where it would fail, then apply it to one task with no game around it. Evidence is mixed: in workplace and university simulation-game studies, games used as a supplement beat the comparison (d = .51) and standalone games lost to it (d = −.12) (supplement, not standalone [+M]), but in K–16 digital-game studies adding non-game instruction made no detectable difference (b = 0.04, null moderator [~M]). The debrief's self-explanation prompts are carried from non-game instruction (self-explanation [+M]); untested in games.
  7. Judge the design outside the game, on an aligned task at the horizon the objective sets, and record enjoyment separately. A game score or completion rate is not the outcome. See the DragonBox report [~W]; and Tetris training, in a review of casual-game studies, produced gains mostly on tests resembling the game, not on general cognitive tasks (Tetris [~M]).

How far the neighbouring claims may be carried: the board-game study is low-income preschoolers learning number with an experimenter playing alongside; the puzzle-game studies are 72 university students in one session-length experiment. The session, competition and supplement findings are moderator analyses inside meta-analyses: they compare studies that differ in many ways at once, so they show which designs went with larger effects, not that changing one feature in a given game will produce the difference. The simulation-game findings come from workplace and university training, the digital-game findings from K–16.

Fitting the design to a situation

The three facts that most change the decision:

  • Whether the learner already has a strategy for the target task. A learner with none tends to flail through trial and error; one with a usable strategy can use play as practice. If unstated, ask: have they solved a task like this before, unaided, outside a game?
  • Whether an adult or peer will play alongside. The tested child gains came with an adult running the game, and the board-game claim's own Discussion names the facilitator as a possible critical ingredient. If unstated, ask: who is present while they play, and will that person talk about the content or only the rules?
  • What outcome must change, and where it will be measured. A game can raise enjoyment while a plainer tool raises test scores. If unstated, ask: what will the learner be able to do afterwards, outside the game, and how will you check?
If the brief says… Then change… Basis
Learners are novice on the target task Add one or two worked examples or a modelled round before unaided play and again between rounds; keep game controls minimal Worked examples between rounds, puzzle game [~M]; tested with university students, untested with children
Learners are preschool or early-primary children Use a simple game whose board or move is the representation (e.g. a numbered linear track); an adult plays alongside naming the numbers or quantities; four sessions of 15–20 min over two weeks Board-game study [+M]; guided play [~W] (guided play, not games specifically); in K–8 STEM games beat conventional teaching on average (STEM meta-analysis [+M], cognition g = .67)
Learners are secondary-school students Expect a smaller gain than with primary pupils; keep the game to the one move that carries the target, pair it with taught content and a debrief (step 6), and check the outside-game task before committing more time School level [~M] (K–8 STEM: secondary pupils gained less than primary, difference g = −0.67, 12 against 28 studies; why is not established)
Learners are adult or professionals in a workplace-clinical setting Build the game around one job decision with real consequences in its rules; use it as one part of a training programme, not the whole of it; leave it open for repeated voluntary play; debrief against the real procedure; keep competition optional Supplement, not standalone [+M] and unlimited access [+M] (simulation games in training, pooled); the whole-task argument behind building around one job decision is untested
The game would replace active non-game teaching of the same content (practice with feedback, discussion, problem solving) Keep the active teaching; add the game alongside it only for a reason other than a learning gain (access, practice volume, interest), and check the outside-game task Active comparison [~M] (simulation games: d = −.19 against active instruction, d = .38 against passive, d = .61 against none)
The goal is verbal-association (vocabulary, facts) A quiz or practice game is a reasonable vehicle; add glosses or supplementary word material inside it, and test recall a week later, not only at the end of play Kahoot! and outdoor games with migrant children [+W] (confounded); adventure game with supplementary material [+W] (second-hand); long-term vocabulary retention [+W] (a meta-analysis known only through a later review, no effect size)
The goal is a concept, principle or complex-skill Make the core move the reasoning itself; add a debrief and one non-game transfer task; expect slower gains than for facts untested; from the earlier page
The goal is attitude-motivation (interest, confidence) Measure enjoyment and the capability separately; do not report one as the other; a game can win on the first and lose on the second; do not buy entertainment value expecting more learning DragonBox vs Lynnette [~W]; entertainment value did not moderate learning [~M]; games raised motivation over conventional K–8 STEM teaching on average (g = .51, STEM meta-analysis [+M])
The brief wants competition, a leaderboard or head-to-head play Make individual non-competitive play the default; if competition is wanted, use team scores built from each member's learning (Gamification, step 5) rather than individual head-to-head play Player configuration [~M] (single competitive play g = −0.06 against non-game conditions); after controls [~M]
online-self-paced, no teacher present Put the modelling, hints and debrief questions inside the game (a worked example screen between levels, a "why did that work?" prompt after each level); log attempts per level to find where learners stall untested proposal; the tested child gains had an adult present
Time is a single-session Use one short game round, a worked example, a second round and a debrief; do not expect a gain over non-game teaching; if days-weeks are available, spread play over several sessions rather than lengthening one Sessions [+M] (single-session games g = 0.08, not different from non-game conditions; total minutes unrelated to effect); the single-session plan itself is an untested proposal
Stakes are high or the game feeds a grade Do not grade the in-game score unless its measurement has been designed for that (Assessment Mechanic); assess on an aligned external task untested; from the earlier page
Devices, motor demands, reading load or disability limit access Choose a mechanic with no timing or precision demands beyond the target; offer a paper or board version; read text aloud or reduce it untested; from the earlier page

Nine of the thirteen rows rest on a claim. The novice, child, vocabulary and motivation rows rest on single studies in narrow populations or on second-hand reports; the secondary, workplace, active-comparison, competition and single-session rows rest on moderator analyses in three meta-analyses, which compare studies rather than versions of one game. The other four are untested proposals, most from the earlier page. Team scoring built from individual learning is set out in Gamification and is not tested in full games here.

Observation, state and explanation

What can be observed is play under stated conditions: moves made, levels reached, attempts per level, time spent, what a learner says while playing, and performance on a task outside the game. "Understands the concept" and "is motivated" are inferred states. Winning a level is an observation about the game, not about the capability, unless the winning move is the capability. Record the game version, any help given, who was present, and the dose.

Response Plausible explanations Observation that could change the interpretation
Wins levels, fails the same idea outside the game The game can be won by a shortcut that bypasses the target reasoning; knowledge tied to the game's representation; the outside task is unfamiliar in format Ask the learner to explain the winning move; give the outside task once in the game's representation and once without it.
Enjoys the game, little gain on the test The mechanic does not demand the target reasoning; play displaced practice time; the test is misaligned with what the game trains Compare the test items with the moves the game requires; record time on the target move against total play time.
Many failed attempts, no improvement across them No strategy to learn from consequences (novice); feedback says only "wrong"; controls or reading load are the barrier Show one worked example or modelled round and see whether the next attempts change; replay with simplified controls or text read aloud. In one biology game's logs, students who restarted and failed a cell cycle more often gained less from pre- to posttest (r ≈ −.30 and −.34, 39 students), while total attempts were unrelated (Progenitor X logs [~W]): count repeated failures on the target move, not attempts in general.
Stops playing early Too hard; competition discourages a low scorer; the learner does not value the task Note where play stops; offer a non-competitive mode and an easier entry level, separately; ask what they wanted from it. Difficulty alone is a weak explanation: in Physics Playground, harder game mechanics went with slightly more persistence (3% per unit of difficulty; difficulty and persistence [~W], associational).
Improves after the game and a debrief Gain from play, from the debrief, from added time on the topic, or from the adult's talk Keep all exposures recorded. Only a comparison holding time and talk constant can credit the game.

These are proposals, untested with learners. A probe can shift confidence between explanations; it does not identify a cause, and probing itself teaches.

Evidence and qualifications

Three meta-analyses in the wiki compare games with non-game conditions on average; the remaining claims test parts of the relationship.

Taken together: on average, games beat no or passive instruction by a small-to-moderate amount, more with several sessions, inside a course, without head-to-head competition and with primary rather than secondary pupils, and not at all against active non-game instruction; the averages are inflated by publication bias and hide wide variation (three meta-analyses). Within that, a game helps when its move carries the content (one randomized child study), added scaffolding improves a game (pooled value-added comparisons), novices learn more when shown worked examples and still learn little from play alone (one adult study), and enjoyment and entertainment value are not evidence of learning (a moderator analysis and a second-hand report). The vocabulary-game evidence is confounded or second-hand. Learners, domains and outcomes differ across these claims; do not rank them by effect label.

The evidence by question

Further evidence, not yet read against this model

Claims read against this model that do not change it. Each says what it is and why it stays here.

Objective and learner-valued goal

The designer's objective names the capability the game is meant to build and how it will be checked outside the game. Elicit separately what the learner wants from playing: to win, to beat a friend, to finish quickly, to get better at something they care about. Agreement and divergence are both observations; do not infer motivation from time played.

For example, a secondary-school student playing a city-budget simulation game may value topping the class leaderboard, while the designer's objective is reasoning about trade-offs between spending choices. If the leaderboard rewards the fastest balanced budget, the student can pursue what they value by repeating one safe pattern and never reason about trade-offs. Recording both aims shows where to change the scoring (reward an explained trade-off) and keeps a high score from being read as the objective met.

What would revise this model?

The model should weaken if games whose core move carries the content, with modelling and a debrief, fail to beat a non-game version with the same content, time and talk on an aligned outside measure; or if, like the colour-only board, games with content-free moves produce the same gains, which would point to the adult's talk or the time spent rather than the game. If modelling inside games keeps helping while play alone adds little, the model should be restated as practice with modelling, delivered in a game format. If enjoyment and later persistence turn out to predict learning well, revise the separation of the two outcomes. If within-game comparisons (the same game played in one session against several, or embedded against standalone) fail to reproduce the between-study moderators, drop steps 5 and 6's reliance on them. If games keep matching but not beating active non-game instruction, restate the page's case for a game as access, practice volume or interest rather than a learning gain. Do not protect the model by calling every failure a badly designed game.

A gain in the game, a gain on an aligned outside test, delayed retention, transfer to a new task and continued voluntary practice are separate claims. The wiki's evidence does not yet establish a dose beyond one preschool schedule and the meta-analyses' averages, which game genres help which goals, why secondary pupils gain less than primary, or whether games beat active non-game practice for anyone.

Claims

Examples

  • Learning Mechanic — the unit a game's teaching is actually built from: the theory-grounded activity repeated throughout play, which a concrete game mechanic instantiates.
  • Assessment Mechanic — its measurement counterpart, designed so the game's own log yields interpretable evidence rather than only a score.
  • Scenario-based digital games: Learners manage resources, make choices, and see consequences unfold across rounds or levels.
  • Board or card games for concept practice: Structured turn-taking and rules create repetition with feedback while keeping attention high.
  • Quiz-show style review games: Fast cycles of attempt and feedback can increase practice volume when questions still align to real objectives.
  • Collaborative mission games: Teams solve a shared problem, requiring explanation, coordination, and iterative strategy updates.
  • Epistemic Games — a pattern where the embedding is total: the repeated activity is the professional community's actual work, so there is no separable "learning part" to bolt on
  • Game-Based Mastery Learning — a pattern where the tension is live, since the mechanics sustaining repetition are not the mechanics carrying the concept
  • Frame simulation lessons as a movie-director game with scaffolding, varied texts, and reflective discussion to sustain engagement

Key Sources

  • Eseryel, D., Law, V., Ifenthaler, D., Ge, X., & Miller, R. (2014). An investigation of the interrelationships between motivation, engagement, and complex problem solving in game-based learning. Educational Technology & Society, 17(1), 42-53.
  • Liu, Z. Y., Shaikh, Z., & Gazizova, F. (2020). Using the concept of game-based learning in education. International Journal of Emerging Technologies in Learning, 15(14), 53-64. https://doi.org/10.3991/ijet.v15i14.14675
  • Boghian, I., Cojocariu, V. M., Popescu, C. V., & Mâţă, L. (2019). Game-based learning - Using board games in adult education. Journal of Educational Sciences and Psychology, 9(1).
  • An, Y. (2020). Designing effective gamified learning experiences. International Journal of Technology in Education, 3(2), 62-69. https://doi.org/10.46328/ijte.v3i2.27
  • Plass, J. L., Homer, B. D., Kinzer, C., Frye, J., & Perlin, K. (2011). Learning mechanics and assessment mechanics for games for learning (G4LI White Paper #01/2011, Version 0.1). Games for Learning Institute (New York University, CUNY Graduate Center, Teachers College Columbia University). researchgate.net/publication/272815253
  • Isbister, K., Flanagan, M., & Hash, C. (2010). Designing games for learning: Insights from conversations with designers. CHI '10 Extended Abstracts on Human Factors in Computing Systems, 2041–2044. doi:10.1145/1753326.1753637
  • Kirschner, P. A., Sweller, J., & Clark, R. E. (2006). Why minimal guidance during instruction does not work: An analysis of the failure of constructivist, discovery, problem-based, experiential, and inquiry-based teaching. Educational Psychologist, 41(2), 75–86. doi:10.1207/s15326985ep4102_1
  • Schnotz, W., Böckler, J., & Grzondziel, H. (1999). Individual and co-operative learning with interactive animated pictures. European Journal of Psychology of Education, 14(2), 245–265. doi:10.1007/bf03172968
  • Swink, S. (2008). Game feel: A game designer's guide to virtual sensation. Morgan Kaufmann. doi:10.1201/9781482267334
  • Juul, J. (2003). The game, the player, the world: Looking for a heart of gameness. In M. Copier & J. Raessens (Eds.), Level up: Digital Games Research Conference Proceedings (pp. 30–45). Utrecht University. doi:10.26503/dl.v2003i1.65
  • Shen, C.-Y., & O’Neil, H. (2006). The Effectiveness of Worked Examples in a Game-Based Learning Environment. Paper presented at the annual meeting of the American Educational Research Association (AERA), San Francisco, CA. https://eric.ed.gov/?id=ED491961
  • Clark, D. B., Tanner-Smith, E. E., & Killingsworth, S. S. (2016). Digital games, design, and learning: A systematic review and meta-analysis. Review of Educational Research, 86(1), 79–122. https://doi.org/10.3102/0034654315582065
  • Arztmann, M., Hornstra, L., Jeuring, J., & Kester, L. (2023). Effects of games in STEM education: A meta-analysis on the moderating role of student background characteristics. Studies in Science Education, 59(1), 109–145. https://doi.org/10.1080/03057267.2022.2057732
  • Sitzmann, T. (2011). A meta-analytic examination of the instructional effectiveness of computer-based simulation games. Personnel Psychology, 64, 489–528. doi:10.1111/j.1744-6570.2011.01190.x
Page metadata
Field Value
type principle
id game-based-learning
aliases learning-embedded-in-the-core-mechanic, add-worked-examples-to-commercial-off-the-shelf-games-for-training
title Game-based Learning
description For a learner who has not yet reached a stated capability, a game whose winning move is the target reasoning, played over several sessions inside a course of instruction, with modelling for novices and a debrief that asks for the reasoning outside the game, is expected to improve aligned outside-game performance over passive or no instruction; meta-analyses find small-to-moderate average gains that vary widely and disappear against active non-game instruction, and enjoyment and game score are separate outcomes.
status review
canonical True
generated by claude/unspecified on 2026-10-19
trust unverified — no one has confirmed this page's content
sources 9 — liu-2020, an-2020, plass-et-al-2011, isbister-2010, kirschner-2006, schnotz-1999, shen-2006, clark-2016, arztmann-2023

Frontmatter is the page's machine-readable record — see the Schema & Guide.