Promising people a tangible reward (such as money or a prize) simply for doing an activity they already find interesting makes them choose to do it less on their own once the reward stops, by roughly 0.2 to 0.4 standard deviations compared with no reward, and children are hit at least as hard as adults.
Status
alive
Last change
2026-11-06
Serves
Q1 (How do humans actually learn, and what helps them learn best?)
Ages
grundskola-lag, grundskola-hog, vuxen-livslangt
Area
Motivation and curiosity
Bounds
Expected tangible rewards for activities people already find interesting, with later free-choice engagement as the outcome; not grades or task-process feedback (C4) and not short growth-mindset lessons (C13). Evidence on file: Deci, Koestner & Ryan 1999 (Psychological Bulletin 125(6), 627–668, DOI 10.1037/0033-2909.125.6.627; abstract): meta of 128 experiments; free-choice intrinsic motivation undermined by engagement-contingent (d = −0.40), completion-contingent (d = −0.36) and performance-contingent rewards (d = −0.28); tangible and expected rewards harmed; worse for children than college students; positive feedback enhanced free-choice (d = 0.33). Cameron & Pierce 1994 (Review of Educational Research 64(3), 363–423, DOI 10.3102/00346543064003363; abstract): 96 between-groups experiments; overall reward did not decrease intrinsic motivation; the only negative effect was expected tangible rewards given simply for doing a task (minimal); verbal praise increased motivation. Kill-test survived 2026-10-27: Cameron, Banko & Pierce 2001 (The Behavior Analyst 24(1), 1–44, DOI 10.1007/BF03392017; full text): even this counter-meta found expected tangible rewards on high-interest tasks lowered free-choice (d1 = −0.18, CI −0.23 to −0.13) and rewards offered for doing the task lowered it further (all 57 studies d1 = −0.35, CI −0.43 to −0.27, not homogeneous; with outliers removed d+ = −0.30, CI −0.38 to −0.22, K = 50, N = 2,642; Figure 1, p. 17; precision corrected 2026-10-30); free-choice was unchanged or higher when rewards were tied to meeting a score or exceeding others — a contingency carve-out, not a kill of the claim's "for doing" wording. The same meta's carve-out is narrower than it looks for pay per unit solved (corrected 2026-10-28; full text pp. 19–20): when people could earn the maximum, free-choice was unchanged (K = 6, N = 345, d1 = −0.03, CI −0.25 to 0.18), but when they earned less than the maximum it fell (K = 14, N = 749, d1 = −0.22, CI −0.37 to −0.07); the authors conclude the negative effect of pay per unit goes with receiving less than the maximum reward. Murayama, Matsumoto, Izuma & Matsumoto 2010 (PNAS 107(49), 20911–20916, DOI 10.1073/pnas.1013305107; PMC full text; n = 28) is consistent: ¥200 per successful trial on an interesting stopwatch task, against yoked equal pay, cut voluntary plays in 3-minute free-choice periods after the reward stopped (F(1,26) = 6.59, P = .016) — a small, non-preregistered lab test, support only, not a kill-condition trial. Durability (added 2026-10-30): of seven experiments with rewards for doing the task that measured free choice again a few weeks later, only two still showed a significant drop (Cameron et al. 2001, pp. 23–24); the drop C22 describes is measured in a free-choice period soon after the reward stops, and whether it lasts is open (open question 19). Bitter, Wondra, McCrea, Darzi & Novak 2022 (Technology, Mind, and Behavior 3(1), 39–56, DOI 10.1037/tmb0000056; full text; added 2026-10-30): two preregistered online 2 × 2 experiments with US crowd-workers (N = 441 and 1,105; mean age about 40) on a location-memory task; removing pay of 1–3 cents per correct response lowered free-choice attempts and performance minutes later (pooled over both studies d = .19, 95% CI .05 to .34, and d = .22, .08 to .36; Study 2 alone .17 and .16) — consistent with the bottom of the range for adults, but the pay was per correct response (performance-contingent), not simply for doing, so it is support and not a kill-condition test. Child–adult prong (corrected 2026-10-30; contingency enrichment 2026-11-06): now read in full text in Deci, Koestner & Ryan 2001 (Review of Educational Research 71(1), 1–27, DOI 10.3102/00346543071001001; Table 1, the same meta-analytic data as Deci et al. 1999): rewards given simply for engaging in the task lowered free choice more in children (d = −0.43, k = 39) than in college students (d = −0.21, k = 12), and the authors report the difference as significant; for tangible rewards overall the undermining was significantly greater for children on both free-choice and self-report measures. The same Table 1: completion-contingent free-choice d = −0.44 (k = 19); performance-contingent free-choice d = −0.28 (k = 32), of which maximal reward d = −0.15 (k = 18) and not-maximum reward d = −0.88 (k = 6). Cameron et al. 2001 tried age as a moderator for rewards for doing the task and could not resolve it (p. 17). Deci 1999 and Cameron & Pierce 1994 remain abstract-only; Deci et al. 2001, Cameron et al. 2001 and Bitter et al. 2022 are full text.
What would kill this
A large preregistered experiment or meta-analysis finds that expected tangible rewards for doing an interesting task leave later free-choice engagement unchanged or higher than with no reward. Sharpened 2026-10-29 (bold expansion): the claim also dies if such a test finds the drop smaller than about 0.1 standard deviations, or larger in adults than in children. Promoted from K48 on 2026-10-27 under the project's promotion rule when C19 reached the survival floor of 3; standing policy Jonatan approved 2026-10-14). Area: Motivation and curiosity (second claim after C13).
Tutor consequence
none yet
Attack record
1 attack: 1 survived
2026-10-27 survived
Attack
A large meta-analysis would meet C22's kill-condition if expected tangible rewards for doing an interesting task left later free-choice engagement unchanged or higher than with no reward; Cameron, Banko & Pierce 2001 — the main counter-meta arguing that pervasive reward harm is a myth — still found free-choice d1 = −0.18 (95% CI −0.23, −0.13) for expected tangible rewards on high-interest tasks, and d1 = −0.35 (CI −0.43, −0.27; K = 57) when the reward was offered for doing the task.
Target
kill-condition
Source
Cameron, Banko & Pierce 2001 | The Behavior Analyst 24(1), 1–44, DOI 10.1007/BF03392017 (PMC2731358; behavior.org OA PDF)
Steelmans to
C22 dies on a large meta that finds expected tangible rewards for interesting tasks leave later free-choice unchanged or higher; this is the strongest published counter-analysis to Deci et al. 1999, and the kill fails on its own free-choice numbers for expected tangible / doing.
Why
Extra12 attack pass 2026-10-27 Europe/Stockholm (parent: C22 never-targeted first honest OA attacks; C14 only if NEW OA; leave C19/C21). Full text read and stored (behavior.org OA PDF of Behavior Analyst 24:1–44; same as PMC2731358). Table 2 free-choice high-interest: expected tangible d1 = −0.18*; for doing a task (Figure 1 / text) d1 = −0.35* (K = 57, N = 2,910). Kill fails. Authors also report ns/positive free-choice for performance-criterion contingencies (meet a score; exceed others) — logged in Bounds as a contingency carve-out, not a weaken of the claim's "for doing an activity" wording. Not a restatement of Cameron & Pierce 1994 abstract in Bounds (different paper, full-text contingency split, now a kill-test). Not a preprint. Claim sentence and Status unchanged (alive). C22 now 1 of 3 survivals.
Rewrite
none (bounds + practical rule only)
Corrections to our notes
These fix a number or a fact in our notes. They are not attacks and do not count toward the floor of 3 survivals.
— C22's limits present Cameron, Banko & Pierce 2001's contingency carve-out (score- or competition-tied rewards leave free-choice unchanged or higher) without the same meta's per-unit result: when people were paid per unit solved and earned less than the maximum, free-choice fell (d1 = −0.22, K = 14), and Murayama et al. 2010 shows the same in a small randomised lab test.
— Wording change only, not an attack: a vague claim sentence was made sharper, so that real counter-evidence can kill it.
— Checked against the full text of the study: precision of the Cameron et al. 2001 "for doing the task" estimate.
— Checked against the full text of the study: contingency detail in Deci et al. 2001 Table 1 beyond the child–adult engagement split already on file.
Two preregistered online experiments (N = 441 and 1,105 adults): removing pay per correct answer lowered later free-choice attempts and performance (pooled d = .19 and .22). Per-correct pay, not pay for doing; adults; minutes later.
Full text, Table 1: rewards for simply doing the task lowered free choice more in children (d = −0.43, 39 studies) than in college students (d = −0.21, 12 studies), significantly so. Same data as Deci et al. 1999.
OA fMRI free-choice: performance-based ¥200/success cut voluntary SW plays vs yoked control (F(1,26)=6.59, P=.016; n=28). Supports C22 force; too small for kill-condition bar — not a survival.
Durability: of seven experiments with rewards for doing the task that measured free choice again a few weeks later, only two still showed a significant drop (pp. 23–24); the homogeneous estimate for rewards for doing the task is d+ = −0.30 (K = 50, Figure 1). C22 gives no time bound.