Escalation of Commitment: Why a Failing Plan Gets More Money, Not Less

A decision maker who learns that a chosen course is failing has two ways to make the news bearable: stop, and accept that the first decision was wrong, or continue, and keep open the possibility that it was right. Escalation of commitment is the second route, and the explanation Staw proposed for it is self-justification. Having chosen publicly, and often having argued for the choice, the person is motivated to show that it was sound, and the clearest way to show that is to make it succeed. More resources go in, and the stake that needs justifying grows with them.

Staw's design separated this motive from the plain economics of the case. Every student in the 1976 experiment saw the same financial data for the second decision; what varied was whether the first decision had been theirs. Those who owned a failing choice gave it the most money. Those who inherited the same failing choice from a predecessor did not. The numbers alone could not produce that gap, because the numbers were the same.

Later accounts added other engines. Brockner's 1992 review set self-justification alongside three more: the way prospect theory predicts people will gamble to avoid a sure loss once they are in the domain of losses, the pull of a decision rule that says to finish what has been started, and the fear of how others will judge a public reversal. Staw's own 1981 review added the organizational layer, where a project acquires defenders whose reputations depend on it and where stopping becomes a political act. Sleesman and colleagues' 2012 meta-analysis of the laboratory literature found that no single factor explains escalation; the sunk costs already incurred, how close a project is to completion, and the decision maker's personal responsibility for the original choice all push in the same direction, among many others.

How large the responsibility effect is on its own is a fair question. Staw's result rested on one cell of a two-by-two design, and later studies have found it in some settings and not others. The meta-analytic picture is that personal responsibility is one real determinant among several, and that features of the project often matter as much as who chose it.

Barry Staw published "Knee-deep in the big muddy: A study of escalating commitment to a chosen course of action" in Organizational Behavior and Human Performance in 1976.

The participants were 240 undergraduate business students at the University of Illinois. Each worked on the "A and S Financial Decision Case," which described the Adams and Smith Company in 1967, a large technology firm whose profits had been slipping. The case asked the student, as financial vice president, to put 10 million dollars of extra research funding into one of two divisions, Consumer Products or Industrial Products, and to write a paragraph defending the choice. A second part of the case then moved to 1972 and gave five years of sales and earnings figures for both divisions. This time 20 million dollars was available and could be split in any proportion. The amount given to the previously chosen division was the measure of commitment.

Two things were manipulated. Half of the students saw figures showing that the chosen division had deteriorated while the other recovered; half saw the reverse. And half of the students made the first decision themselves, while the other half received only the 1972 part of the case and were told that an earlier financial officer had made the 1967 choice. That produced four groups of 60.

Students who had made the choice themselves gave the previously chosen division 11.08 million on average; students who inherited the choice gave it 8.89 million. Negative results drew more money than positive ones, 11.20 million against 8.77. The important finding was the interaction: students who had made the choice personally and then watched it fail allocated 13.07 million to the failing division, and that group was the only one that differed significantly from the other three. Owning a bad decision, not merely seeing one, was what moved the money.

Escalation needs three things: a decision somebody owns, bad news about it, and a chance to put more in. The rows below are illustrations, not findings from the studies cited on this page.

The strongest counter follows directly from Staw's design: the people who inherited a decision did not escalate it. Separating the person who judges a project from the person who chose it removes the motive to justify.

• Write the stopping rule down before the project starts: the date, the milestone or the budget at which it will be reviewed as if new, and who will do the review.
• Rotate the reviewer. A manager who did not pick the vendor, the hire or the strategy can read the numbers without needing them to come out one way.
• Ask the predecessor question from the demo. If you had just taken over and found this project on your desk, with these results and this much money left, what would you do?
• Reframe the sunk costs as gone. Money and time already spent do not come back under either option, so they belong in neither column of the decision.
• Make reversal cheap in advance. Teams that are praised for stopping a failing project early have less to defend than teams for whom a cancellation is a public verdict.
• Notice the words. Turnaround, momentum, so close, and too much invested to quit are all sentences about the past dressed as sentences about the future.

Escalation of commitment is the sequence: a decision, bad news, and a further commitment that defends the decision. Its neighbors share pieces of that sequence.

• It is not the sunk cost fallacy, which is one decision made badly because costs already spent are counted as if they could be recovered. Sunk cost reasoning is one cause of escalation; escalation also runs on reputation, public commitment and the wish not to have been wrong, and it unfolds over repeated decisions.
• It is not irrational escalation, the competitive version that the dollar auction illustrates, where each bid is made to avoid losing the last one to a rival. Staw's students had no opponent; they were defending themselves against their own earlier choice.
• It is not status quo bias, the general preference for the current state over a change. Escalation is more active than inertia: the person does not merely keep a project, they feed it.
• It is not choice-supportive bias, the tendency to remember a past choice as better than it was. That is a distortion of memory; escalation is a distortion of the next allocation.