Same failing division, same numbers, two very different answers
The case was called A&S Financial. Adams & Smith Company, a technology firm in decline, ten years of sales and earnings on the page, and one job: you are the financial vice-president, allocate ten million dollars of R&D funding between the Consumer Products and Industrial Products divisions. Five years later — in the case's time, minutes in the room — a second decision: another twenty million from capital reserve.
Barry Staw ran 240 undergraduates from the University of Illinois through it in a two-by-two design, published as Knee-deep in the Big Muddy in Organizational Behavior and Human Performance in 1976. Half the participants made both allocations themselves; the other half were told a predecessor had made the first one. Independently, the chosen division either improved or declined.
Three of the four groups behaved sensibly, allocating somewhere around nine million to the division they had watched struggle. The fourth group — the people who had picked that division themselves and then watched it fail — allocated an average of $13.07 million. The interaction was significant (F(1,235) = 5.56, p < .019), and it is the whole point. Bad news alone did not produce throwing good money after bad. Bad news about your own decision did.
The small yes that buys the large one
The mechanism has a well-documented smaller sibling. In 1966 Jonathan Freedman and Scott Fraser published Compliance without pressure: the foot-in-the-door technique in the Journal of Personality and Social Psychology. In their second experiment, researchers asked Californian homeowners to do something trivial — sign a petition, or display a small sign about safe driving. Two weeks later a different person asked the same households to install a large, deliberately ugly DRIVE CAREFULLY billboard on the front lawn.
The control group, asked only once, agreed at 16.7 per cent. The households that had earlier accepted the small, same-issue request agreed at 76 per cent. In the authors' own summary of the pattern:
Whereas fewer than 20% of the controls agreed to put the large sign on their lawn, over 55% of the experimental subjects agreed, with over 45% being the lowest degree of compliance for any experimental condition.Note the honest framing: 76 per cent is the best cell, not the average — the reliable headline is that a prior small commitment roughly tripled compliance. Now scale it. A pilot budget is a small sign. The steering committee that approved it has taken a public position, and the next request does not arrive to a neutral audience.
— Freedman & Fraser, Journal of Personality and Social Psychology (1966)
What it looks like when it is not a lab
The best-documented organisational case is a world's fair. Jeffrey Ross and Barry Staw studied Expo 86 in Vancouver and published it in Administrative Science Quarterly in 1986 as an escalation prototype. The projected loss did not creep; it multiplied — and the commitment held anyway.
Despite rapidly increasing deficit projections (from a 6-million projected loss in 1978 to over a 300-million projected loss in 1985), the provincial government remained steadfast in its plans to hold Expo.
— Ross & Staw, Administrative Science Quarterly (1986)
Fifty-fold worse, and still going. That is not stupidity, and reading it as stupidity is why organisations keep reproducing it. Every individual decision along that path had a defensible case attached — and each one was being made by people who had already publicly backed the last one.
But isn't this just the sunk-cost fallacy?
This is the objection worth taking seriously, because the two are constantly used as synonyms and they are not. Staw drew the line himself in the Academy of Management Review in 1981:
A salient feature of the preceding case examples is that a series of decisions is associated with a course of action rather than an isolated choice.
— Barry M. Staw, Academy of Management Review (1981)
Sunk cost is a single-decision error: counting money you cannot recover. Escalation is an organisational pattern across a sequence of decisions, in which sunk cost is one driver among several — self-justification, ego threat, agency problems, group cohesion. And the evidence supports that distinction in an uncomfortable way. The 2012 meta-analysis by Sleesman, Conlon, McNamara and Miles in the Academy of Management Journal, covering 166 independent samples drawn from an initial pool of 917 articles, found personal responsibility (ρ = .258) and sunk costs (ρ = .243) to be comparable in size — and the authors noted that the prominence of sunk costs was lower than expected.
"Ignore sunk costs" is aimed at an accounting error. But the strongest pull in the data is not the money already spent — it is the person who spent it needing the decision to have been right. A review process that fixes the arithmetic and leaves the authorship untouched has fixed the smaller half.
The technique: separate the decision from its author
None of this is solved by telling people to be more objective about their own record. It is solved by arranging the review so the record is not on trial.
- ✓Write the exit criteria before the money moves. "We stop if X by date Y" is cheap to agree while nobody owns the outcome, and nearly impossible to agree once someone does. This is the single highest-leverage move available.
- ✓Review the case, not the champion. Re-read the original argument tree as written — the claims, the evidence, the assumptions — and ask whether it would be approved today by someone seeing it for the first time. The decision record makes that a document review instead of a personality contest.
- ✓Rotate who presents the continue-or-kill case. If the person who owns the project also frames its review, the interaction Staw isolated is built into your process by design.
- ✓Make opportunity cost visible. The meta-analysis identifies de-escalating conditions — among them anticipated regret, explicit opportunity-cost information, and actively acquiring new information. "What else could this quarter's budget buy?" is not rhetoric; it is a documented lever.
- ✓Separate the person from the position on the record. Rating arguments rather than people means a champion can concede a claim without conceding their judgment — which is the concession escalation actually blocks.
Knee-deep, and the water is rising
Staw took his title from a Pete Seeger song about a captain who marches his platoon deeper into a river, insisting they push on. It is a good image precisely because the captain is not stupid. He has said out loud that this is the way, other people are watching, and turning back means the first announcement was wrong.
Your organisation will fund a failing project again. The question is whether that decision gets made by a room reading the case, or by the room's memory of who signed it — because in Staw's data, the same numbers produced two completely different answers depending on nothing more than that.
Sources & further reading
Every named source in this post, with a link where one exists.
- •Freedman, J. L., & Fraser, S. C. (1966). Compliance without pressure: The foot-in-the-door technique. Journal of Personality and Social Psychology, 4(2), 195–202. — the small-request-first effect; the lawn-billboard experiment quoted above.
- •Staw, B. M. (1976). Knee-deep in the Big Muddy: A study of escalating commitment to a chosen course of action. Organizational Behavior and Human Performance, 16, 27–44. — the A&S Financial experiment, 240 participants, and the responsibility-by-consequences interaction.
- •Staw, B. M. (1981). The Escalation of Commitment to a Course of Action. Academy of Management Review, 6(4), 577–587. — the definition by structure: a series of decisions, not an isolated choice.
- •Ross, J., & Staw, B. M. (1986). Expo 86: An Escalation Prototype. Administrative Science Quarterly, 31(2), 274–297. — the projected-loss trajectory quoted above.
- •Sleesman, D. J., Conlon, D. E., McNamara, G., & Miles, J. E. (2012). Cleaning Up the Big Muddy: A Meta-Analytic Review of the Determinants of Escalation of Commitment. Academy of Management Journal, 55(3), 541–562. — 166 independent samples; personal responsibility and sunk costs as comparable drivers.
- •Ross, J., & Staw, B. M. (1993). Organizational Escalation and Exit: Lessons from the Shoreham Nuclear Power Plant. Academy of Management Journal, 36(4), 701–732. — the other canonical longitudinal case; cited here for completeness, its figures not reproduced.