Cognitive dissonance

Why does being paid too little to argue a case you disbelieve change your mind more than being paid well?

Ask someone to argue publicly for a position they privately reject, and pay them for it. Common sense, and any straightforward account of reward, says the bigger the payment the more the person should come round — reward what they said and they will say it more, and eventually believe it. What is actually found is the reverse: the person paid a trifle shifts toward the position, and the person paid handsomely does not. A reward that works better when it is smaller is a peculiar thing. What would have to be true of the mind for that to make sense?

Set aside reward for a moment and look at what the person now holds. Two things that do not sit together: I believe not-X, and I stood up and argued X. Notice that only one of them is still open. The speech is over; it cannot be un-said. So if the discomfort of holding both is to be relieved, it must be relieved by changing something else.

What else is available? A third thought would do it — any cognition that makes the two consistent without either being false. "I did it because they paid me well" is exactly such a thought. It leaves the belief intact and the act explained. Notice what follows immediately: that thought is only available to the well-paid person. Pay a trifle and there is nothing there to say; a token payment does not explain why a reasonable person would argue something they reject.

So the well-paid person has a cheap repair and takes it. The underpaid person has no cheap repair, and the least resistant remaining element is the belief itself — so the belief drifts toward the speech. The odd result is not really about reward at all. It is about what else was on hand to absorb the inconsistency.

That reframing suggests a way to break the effect, which is the useful test of any account. If the mechanism is inconsistency needing somewhere to go, then hand the person any other route out and the attitude should stop moving. Supply an obvious external reason, and it does. Take away their sense of having chosen — order them to give the speech — and it does: the thought "I had no choice" is itself a perfectly good third cognition. The effect turns out to require perceived choice, a public and irrevocable act, and consequences the person could foresee. Strip any of those and it weakens or disappears. A theory that could be broken in four specified ways and is, in each of them, is doing real work.

But there is a rival worth taking seriously, because it explains the same data without any inner discomfort. On the self-perception account you do not have privileged access to your own attitudes; you infer them as an observer would, from your behaviour and the visible circumstances. Watching yourself argue X for a trifle, you conclude you must have half-believed it. No tension, no repair — just inference. Both theories predict the same reversal.

How would you separate them? Only dissonance says the driver is an actual unpleasant state. So: create the inconsistency, then give the person something plausible to blame the resulting arousal on — a pill they were told has side-effects. If discomfort is doing the work, attributing it elsewhere should abolish the attitude change; if the person is merely inferring, the pill is irrelevant. Studies of that design found the change abolished, which refutes a pure inference account.

The core is a claim about resistance, not about honesty. Dissonance is reduced wherever it is cheapest, and a completed public action is expensive to revise while a private attitude is not. That is why the theory predicts self-justification generally: effort already spent makes the thing it bought look better. It also fixes the boundary with Conformity, its neighbour here: there the pressure is other people's stated answers and the change may stay public, whereas dissonance needs no audience and moves the private belief.

Now the honest accounting, because this is a literature whose fame outruns its evidence in specific and documented ways.

The founding underpayment experiment is small — of the order of twenty people per condition, from memory — and its numbers should not be treated as an effect size by anyone. Cite the direction, not the magnitude.

More seriously, one whole branch of the evidence has a demonstrated flaw. In the free-choice paradigm, people rate a set of items, choose between two they rated equally, and then re-rate: the chosen item rises, the rejected one falls, which was read for decades as dissonance reduction. It was later shown that the design produces this spreading even when no attitude changes at all, because the choice itself reveals a genuine pre-existing preference that the noisy first rating had failed to capture. The proper control — comparing against choices made blind — was rarely run. So findings resting on free choice are much weaker than their citation count suggests, including some widely-repeated demonstrations in children and monkeys.

The induced-compliance design this question is about has held up better, and its boundary conditions — choice, commitment, foreseeable consequences, measurable arousal — have been shown repeatedly. That is the narrower, better-supported version worth carrying. Meanwhile the popular extension, that dissonance makes people dig in when corrected, has fared badly: large pre-registered studies across many issues found corrections generally moved people toward accuracy, with backfire rare rather than typical.

• Why effort spent on something makes it look more valuable afterwards, and how cleanly that separates from simple sunk-cost reasoning.
• How the arousal-misattribution result has fared under modern replication standards, given how much rests on it.

Every term the collection defines is gathered in the glossary.