Paying pupils to read
A Socratic walk-through of paying pupils to read — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why can rewarding children for reading leave them reading less once the rewards stop?
A school offers a pound a book. Reading goes up — it nearly always does, and the scheme is judged a success on that fact. Then, a term later, the money stops, and reading falls not to where it was but below it. The children who were paid now read less than the children who never were.
That is a strange shape for an incentive to have. A reward is supposed to be additive: you get the behaviour while you pay for it, and afterwards you are back where you started, poorer but no worse off. Something here is being subtracted as well as added. What could a payment take away?
Reasoning it through
REASONING #Two candidates, and they are worth keeping apart, because only one of them is surprising.
The first is that the reward changes which behaviour happens. Pay per book and you have not paid for reading; you have paid for finishing. Five forty-page books is two hundred pages for five pounds; one three-hundred-page book is three hundred pages for one. The child who reads more has earned less, so the scheme quietly selects for thin, easy texts — the reading that builds least. This is ordinary measure-gaming, and teaching-to-the-test.md works through the general form. It explains why the gains are smaller than they look. It does not explain the fall below baseline.
For that, look at what a reward tells a child about the activity. Lepper, Greene and Nisbett ran the demonstration in 1973 with preschoolers who already liked drawing. One group was promised a certificate for drawing, one got an unexpected one afterwards, one got nothing. Weeks later, in free play, the children who had drawn for a promised award chose to draw less than the others. The unexpected reward did no harm. The promise did.
Why should the promise be the harmful part? Try the inference from the child's side. Before, she had no particular account of why she was doing this; she simply was. Now there is a visible reason attached to it, and the visible reason is external. When the reason is withdrawn, so is the answer to "why am I doing this" — and she never had to construct another one, because the school supplied it for her. The reward did not merely fail to build a habit; it displaced the account that a habit runs on.
Notice the condition that argument needs: there has to be something to displace. If a child found reading unrewarding to begin with, no amount of paying can crowd out a motive she did not have — and there the reward is straightforwardly good, because the reading itself may become rewarding through practice. So the effect should be concentrated among the children who liked reading already, and absent or reversed among those who did not. That asymmetry is the model's most testable consequence.
It also matters what the money is attached to. Roland Fryer's incentive experiments across several American districts found, as I recall them, that paying for test scores moved almost nothing, while paying second-graders a small sum per book read did raise reading. A child asked for a higher score does not know what to do differently; a child asked to read a book does. Rewards work on actions people know how to take.
The analogy
THE ANALOGY #Think of a stake driven in to hold a sapling upright. While it is there the tree stands beautifully straight — better than it would alone. But a tree thickens its trunk in response to being pushed by wind, and a well-staked one does not get pushed. Take the stake out after a season and it stands less well than the unstaked tree beside it, which spent the season making wood.
the sapling's loss is purely mechanical and unavoidable, whereas a child's is about the account she has of her own behaviour — which means it can be pre-empted by how the reward is framed and named, and the staked-tree picture offers no equivalent of a reward that recognises rather than purchases.
Clarifying the model
THE MODEL #The honest state of this claim needs saying plainly, because it is one of the more contested in the field. Deci, Koestner and Ryan's 1999 meta-analysis concluded that tangible, expected, task-contingent rewards undermine free-choice persistence at an activity people already enjoy. Cameron and Pierce had argued the opposite from overlapping literature, and the exchange was sharp. What survived is narrower than the slogan "rewards destroy motivation": the undermining is reliably found for tangible and expected rewards on already-interesting tasks, measured as free-choice behaviour after the reward ends; verbal praise does not show it, and rewards for tasks nobody enjoys anyway do not show it. Most of the studies are short laboratory sessions, and I would not carry a number out of them. The better-supported version of the claim is that a promised payment for something a child already chose to do is the specific case at risk, not that incentives are generally corrosive.
pricing-the-sacred.md shares the crowding-out mechanism, and the fixed point of difference is what the good is: there the value is one people hold as untradeable, and the price is an insult — offering more makes it worse. Reading is not sacred. Nobody is affronted by a book token. Here the damage is not expressive at all but developmental: what the reward interrupts is the process by which an activity becomes its own reason.
A picture of it
THE PICTURE #How to readThe horizontal axis is what the child brought to the scheme — reluctant on the left, already a reader on the right. The vertical axis is what the money is attached to: an outcome she does not know how to produce at the bottom, an action she can take today at the top. Read the top-left corner as the case with most to gain and nothing to crowd out, and the top-right as the one where the behaviour rises during the scheme and can fall below its own baseline after it. The two lower points sit near the middle horizontally on purpose: paying for a score is weak for everyone, so where the child started barely matters. The placements express the argument, not measurements.
What became clearer
WHAT CLEARED #Two different things are going on and they are usually reported as one. A reward reshapes the behaviour toward whatever is being counted, which is why the reading that gets bought is thinner than the reading that was hoped for. And a promised reward supplies an external account of why the child is doing it, which is only a problem where an internal account already existed — and is exactly the case where the scheme looks most successful while it runs.
The whole model turns on that second claim, so here is what would refute it: run three phases — baseline, reward, withdrawal — against a never-rewarded control, and split the children by how much they read before anything began. Crowding-out predicts the previously-avid rewarded group falls below its own baseline after withdrawal while the reluctant group does not. If the post-reward reading simply returns to baseline in both groups, then nothing was displaced and the only real story was measure-gaming.
Where to go next
ONWARD #- Whether an unexpected or purely recognitional reward avoids the effect, as the 1973 design hints.
- What happens when the reward is attached to minutes read rather than books finished, and what that measure gets gamed into instead.
Key terms
TERMS #| Term | What it means |
|---|---|
| Overjustification effect | the reduction in later voluntary engagement after an activity has been performed for a promised external reward. |
| Motivation crowding out | displacement of an internal or social motive by an introduced external incentive. |
| Task-contingent reward | a reward given for performing an activity at all, as distinct from one given for how well it was done. |
Every term the collection defines is gathered in the glossary.