Contribution
Two things. A standard vocabulary for reward contingencies, under which studies that had seemed to conflict line up (pp. 736–744). And the first experiment in which the same performance-contingent reward is given in two interpersonal styles, each with a no-reward control receiving the same feedback, so that the effect of the reward can be separated from the effect of the information it carries and from the style it is given in (pp. 744–747).
Key insight
Performance-contingent rewards make both the informational and the controlling aspect of a reward more salient, so their effect is not fixed by the contingency: “performance-contingent rewards can be perceived as either informational or controlling, depending on how they are administered” (p. 742). But “like all other rewards”, they “tend to lower intrinsic motivation relative to no rewards if there is identical feedback within the same interpersonal context” (pp. 748–749). The comparison group is what had been confusing the literature.
Assumptions
- Expected rewards only. All four reward types are announced before the task; unexpected rewards “are not relevant to the present discussion” (footnote 1, p. 736).
- Interesting tasks only. Dull-task conditions are excluded as “not germane” (p. 738).
- Free-choice time as the measure of intrinsic motivation: seconds spent on the target puzzles in a 6-minute unobserved period, out of 360 (pp. 745–746).
- Initial interest as covariate (ANCOVA); sex dropped after showing no main effect or interaction (p. 746).
Key results
The review (pp. 736–744)
- Task-non-contingent rewards (paid for attending) do not undermine: three studies (p. 738).
- Task-contingent rewards, given without feedback, undermine relative to no rewards; with positive feedback added, Deci 1972b, Harackiewicz 1979 and Swann and Pittman 1977 found no difference from no-reward, no-feedback groups (p. 740). The evidence that they undermine more than task-non-contingent rewards is “weak” (p. 740).
- Performance-contingent vs no reward with positive feedback: Karniol and Ross 1977 and Rosenfield et al. 1980, no difference; Harackiewicz 1979, rewarded subjects less intrinsically motivated (p. 741). “The issue … remains unresolved” (p. 741).
- Performance- vs task-contingent without feedback: mixed (Greene and Lepper 1974 none; Luyten and Lens 1981 n.s.; Boggiano and Ruble 1979 and Enzle and Ross 1978 performance-contingent higher, the latter on an attitude measure) (p. 742).
- Harackiewicz reinterpreted (p. 743): her performance-contingent group without norms is read as controlling (mean −.96), the one with norms as informational (+.07), a task-contingent group −.64; the no-reward, no-feedback group fell between. “the results were not significant”.
- Competitively contingent rewards undermine (Pritchard et al. 1977; Deci et al. 1981), and Greene and Lepper’s “performance-contingent” good-player award is reclassified as competitive (p. 744).
The experiment (pp. 744–748)
- Design. 96 introductory-psychology students, equal men and women, randomly assigned, n = 16 per cell. A 2 × 2 of reward ($3, performance-contingent, “if you do well”) by feedback style (informational: “Do as well as you can”; controlling: “should try as hard as possible because I expect you to perform up to standards”, feedback with “just as you should”), plus a no-reward, no-feedback group and a no-feedback, task-contingent reward group (“for doing the puzzles”). Hidden-figures puzzles; the free-choice observer was blind to condition (p. 745).
- Table 1, free-choice seconds, mean (SD) (p. 746):
| no reward | performance-contingent | task-contingent | |
|---|---|---|---|
| informational, feedback | 282.6 (134.7) | 229.4 (139.9) | |
| controlling, feedback | 195.8 (133.6) | 117.3 (169.3) | |
| neutral, no feedback | 164.1 (150.1) | 113.4 (129.4) |
- H1, reward vs same feedback without reward: F(1, 89) = 4.34, p < .05. All three rewarded vs all three unrewarded: F(1, 89) = 6.36, p < .02.
- H2, controlling vs informational (the four feedback cells): F(1, 89) = 6.11, p < .02.
- The two performance-contingent groups: F(1, 89) = 4.25, p < .05; informational performance-contingent vs task-contingent: F(1, 89) = 4.58, p < .04. Controlling performance-contingent and task-contingent do not differ (117.3 vs 113.4) (p. 747).
- Feedback vs none: marginal, F(1, 89) = 3.51, p = .06; informational feedback clearly above no feedback, controlling feedback barely (p. 747).
- Self-report. An 11-item interest factor correlated .42 with free choice (p < .001); its ANOVA “failed to reach significance, although the pattern of results paralleled” (p. 748).
- Pressure–tension (3 items): controlling 3.61 vs informational 2.91, F(1, 90) = 9.19, p < .005; no-feedback 3.14 between (p. 748).
- Recomputed without the covariate, from Table 1 (pooled SD 143.5, df 90): H1 F ≈ 3.37 (p ≈ .07); H2 ≈ 7.69; the two performance-contingent groups ≈ 4.88; informational performance-contingent vs task-contingent ≈ 5.23; rewarded vs unrewarded ≈ 4.31; feedback vs none ≈ 4.73. Standardized differences: informational vs controlling performance-contingent d ≈ 0.78; reward vs its feedback-matched control d ≈ −0.37 (informational) and −0.55 (controlling). The reported Fs are in the same range, so the table and the tests agree; H1 depends on the covariate for p < .05.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Classified by a common terminology, the reward-contingency studies mostly agree | moderate | narrative review, pp. 736–744; no effect sizes |
| C2 | The same performance-contingent reward undermines free choice more when given controllingly than informationally | moderate | one experiment, F(1, 89) = 4.25, n = 16 per cell, p. 747 |
| C3 | Performance-contingent rewards lower free choice relative to no-reward groups given identical feedback | moderate | F(1, 89) = 4.34 with the covariate; ≈ 3.37 (p ≈ .07) without |
| C4 | Controlling administration raises felt pressure and tension | moderate | F(1, 90) = 9.19, three items, p. 748 |
| C5 | The informational/controlling salience of a reward is what mediates the contingency effects | weak | inferred from the pattern of means and the pressure manipulation check; no mediation analysis |
| C6 | Task-contingent rewards undermine relative to no rewards | strong (as reviewed) | many studies, p. 740; here 113.4 vs 164.1, within the combined contrast |
Concepts
- Task-non-contingent reward — given for being in the session, independent of the activity (p. 736).
- Task-contingent reward — given for doing or completing the task, without respect to quality (p. 737).
- Performance-contingent reward — given “for meeting a set criterion, norm, or level of competence” (p. 737).
- Competitively contingent reward — winners of a limited number of rewards outperform others (p. 737).
- Informational vs controlling aspect — feedback that signifies competence within self-determined performance, vs pressure “toward specified outcomes” (pp. 737–738).
Connections
It applies Ryan 1982’s informational/controlling manipulation of feedback (the “should” locution) to rewards, and Deci and Ryan 1980’s cognitive-evaluation framework. Deci, Koestner and Ryan 1999 (LIT-557) use its contingency taxonomy; Ryan et al. 2022 (LIT-1967) cite it as the taxonomy’s source. Cameron, Banko and Pierce (LIT-562) re-code the same literature by what participants were told.
Bearing on the record
- THEORY-049: the primary study its promote_when names, read. Its promote_when asks for experiments that “manipulate only the interpersonal framing of the same reward (informational versus controlling, as in Ryan, Mims and Koestner 1983) and find the predicted difference mediated by felt control”. This study manipulates the framing and finds the difference (C2), and finds felt pressure higher under controlling framing (C4), but runs no mediation analysis, and the pressure effect is reported for the information–control dimension as a whole, not for the rewarded cells. So it is the single study the clause describes, not a mediation result.
- THEORY-049’s text, a correction. THEORY-049 reports, from LIT-557 (p. 653), that this study “found the same pattern, not significantly”. The primary paper reports the informational and controlling performance-contingent groups as significantly different, F(1, 89) = 4.25, p < .05 (p. 747). Either LIT-557 computed a different contrast (for example, each rewarded group against its own control), or the record’s paraphrase is wrong; NOTE-438 should be checked against p. 653.
- THEORY-049’s refutation clause. The clause says the account is refuted “if rewards tied to a performance standard raise free choice against a control group given the same performance feedback”. This experiment has exactly that control in both styles, and the rewards lowered free choice (229.4 vs 282.6; 117.3 vs 195.8; C3). One study of n = 16 per cell; it does not meet the clause and counts against the refuting outcome.
- THEORY-154 (Cameron, Banko and Pierce). The paper’s review is the earlier, CET-side reading of the same contingency literature. Its own experiment’s rewards were for doing “well” with no stated score or comparison, so it does not test THEORY-154’s distinction between rewards for surpassing a score and for exceeding others.
- QUESTION-066. The free-choice period measures persistence once the experimenter’s support is removed, and initial interest is a covariate, so it approaches the question’s “valuation held fixed”, for one 6-minute period.
- No instruction for machine-learning practice.
Limitations
- Small cells (n = 16) and high within-cell variance (SDs 130–170 s on a 0–360 s measure; the controlling performance-contingent SD of 169.3 exceeds its mean of 117.3, so the distribution is likely skewed or floored).
- The covariate carries H1. Without it the reward contrast is at p ≈ .07 (recomputed); the adjusted means are not reported.
- The self-report measure did not reach significance (p. 748).
- Mediation is inferred, not tested. The “salience of controlling and informational aspects” that “mediate the contingency effects” (abstract) is a reading of the means.
- The review is narrative and some reclassifications (Harackiewicz’s groups, Greene and Lepper’s award) are post hoc.
- College students, one task, one session.
Open questions
- Does felt control, measured between the reward and the free-choice period, statistically mediate the framing effect on the same reward?
- Does the framing effect replicate at adequate power, and with rewards tied to a stated score, where THEORY-154 predicts no undermining?