The Overjustification Effect

What it is

The founding experiment is Mark Lepper, David Greene and Richard Nisbett's, published in 1973. Preschool children who already loved drawing were offered a good player award. The children promised the award in advance drew less in free play a week later than children who received a surprise award and than children who received none at all. The reward had turned play into work.

The meta-analysis that made the case is Edward Deci, Richard Koestner and Richard Ryan's, in Psychological Bulletin in 1999, covering a hundred and twenty-eight studies. Its shape is conditional rather than blanket, and the conditions are the finding.

What undermines: expected tangible rewards reduce free-choice persistence on an already-interesting task, with engagement-contingent and completion-contingent rewards doing it most strongly and performance-contingent rewards less so. What does not: task-noncontingent rewards are null, unexpected tangible rewards are flat at essentially zero, and verbal praise and positive feedback enhance intrinsic motivation on both outcome measures. The self-report effects are tiny across the board. The free-choice column is where the finding lives, and the stated-interest column is close to nothing.

One design decision defines the result and almost nobody prints it. The authors excluded uninteresting and boring tasks from the primary meta-analyses by design, on the stated ground that where there is no intrinsic motivation there is nothing to undermine. Thirteen studies manipulating task interest were analysed separately. So the headline is a finding about interesting tasks, and it is a finding about free-choice behaviour, and a sentence that drops either clause is wrong. Two further qualifications come from the paper itself: tangible rewards are more detrimental for children than for college students while verbal rewards are less enhancing for children, and the performance-contingent category is coded four ways by control group and feedback valence, because the competence a performance-contingent reward affirms can offset its controlling aspect, so the sign of that effect depends on which control group it is measured against.

In effect

The opposing tradition is a real literature rather than a footnote and deserves stating at full strength.

Judy Cameron and W. David Pierce, in the Review of Educational Research in 1994, meta-analysed ninety-six experiments and concluded that overall, reward does not decrease intrinsic motivation, that verbal praise increases it, and that the only negative effect is minimal and confined to expected tangible rewards given simply for doing the task, on the free-time measure. Robert Eisenberger and Cameron argued in the American Psychologist in 1996 that the detrimental effect is limited and easily avoided in applied settings. Cameron, Katharine Banko and Pierce, in The Behavior Analyst in 2001, reported that rewards on low-interest tasks enhance intrinsic motivation, that verbal rewards on high-interest tasks produce positive effects, that negative effects occur on high-interest tasks where rewards are tangible, expected and loosely tied to performance, and that where rewards are linked to performance level intrinsic motivation increases or does not differ from controls. Their conclusion was that there are no pervasive negative effects.

The exchange is three papers in one issue of the Review of Educational Research in 2001. Deci, Koestner and Ryan set out a detailed methodological objection to the Cameron and Pierce analysis. Cameron argued that there is no inherent negative property of reward, and that collapsing distinct reward procedures under one theory is what manufactures pervasiveness. Deci, Ryan and Koestner replied that the comment revealed no problem with the 1999 analysis. Neither side conceded and the exchange did not resolve.

The verdict is that the two camps agree on the facts and disagree on how to describe them. Both find verbal praise enhancing. Both find unexpected rewards null. Both find that an expected tangible reward for engaging with an already-interesting task reduces free-choice persistence. Two genuine empirical disagreements remain: whether performance-contingent rewards undermine, and whether dull tasks belong in the analysis at all. Everything else in the dispute is a quarrel over the word pervasive against the word limited, and that is a quarrel about framing rather than about data.

So the usable form is long and it has to be. Expected, tangible, engagement or completion contingent, on a task the person already found interesting, measured as free-choice persistence. Shorten it and you have stated something the literature does not support.

What it does not say

It does not say rewards are bad. Verbal praise enhances on both sides of the dispute, unexpected rewards are flat, and rewards on tasks nobody enjoys are not covered by the finding at all because those tasks were excluded from it by design.

It does not say this is one of the most robust findings in social science. Daniel Pink writes exactly that in Drive, and calls it also one of the most ignored. The sentence carries no note, because there is nothing to attach to it. It describes a conditional finding with a live opposing meta-analytic literature, with moderators that flatten or reverse the sign, and with modest effect sizes on one of two outcome measures. Pink names neither Cameron, Pierce nor Eisenberger anywhere in the book, so a reader of it does not know a dispute exists, and that omission is the worse failure because there is nothing on the page to check. He is owed credit in the same breath: he reports Lepper's null in the body, gives a whole chapter to the conditions under which rewards do work, and writes in his own voice that it is not that all rewards at all times are bad.

It does not transfer to pay. The literature is about small rewards for discrete tasks in short experiments, mostly with children and students, and reading it as a claim that salaries destroy motivation goes well beyond anything either camp has measured.

---