The Psychology of Sunk Cost and Why Empires Fall

In the summer of 216 BCE, the Roman Republic sent the largest army it had ever assembled — approximately eighty-six thousand soldiers by most estimates — to destroy the Carthaginian force that had been ravaging southern Italy for two years. The Romans had tried twice before with smaller armies and failed, losing tens of thousands of men at the Trebia River and Lake Trasimene. The Senate responded to each defeat by escalating. The logic was explicit in the historical record: Rome had already sacrificed so much blood and treasure that retreat was politically unthinkable. At Cannae, Hannibal’s force of roughly fifty thousand encircled and killed seventy thousand Romans in a single afternoon, the worst defeat in Roman history and one of the worst single-day military catastrophes in the record of any civilization.

The Senate had committed its army to an escalating commitment because the costs already incurred made de-escalation feel like waste. This is the sunk cost fallacy operating at the scale of a republic — and it is not unique to Rome, to military contexts, or to the ancient world. The same cognitive structure appears in the decisions of empires abandoning failed provincial occupations, corporations continuing to fund failing product lines, and individual investors holding deteriorating positions. The psychology is identical across contexts. Only the stakes change.

What the Bias Actually Is

The sunk cost fallacy is the tendency to allow past unrecoverable expenditures — of money, time, life, or political capital — to influence decisions about future resource allocation. The rational economic framework is clear: past costs are gone regardless of future choices, so future choices should be made entirely on the basis of expected future costs and benefits. A project that has consumed ten million dollars and is now expected to yield five million should be abandoned, regardless of how much has been spent. The sunk ten million is not recoverable by continuing.

This is straightforward in theory and consistently violated in practice, and the violation is not irrational in the narrow sense of being internally inconsistent. It is understandable as a response to several real psychological pressures. Loss aversion — the well-documented tendency to feel losses more acutely than equivalent gains — means that acknowledging the waste of past expenditure is emotionally painful. Escalation allows decision-makers to defer that acknowledgment. Identity investment, the degree to which a person’s sense of competence and judgment is tied to a previous decision, creates strong incentives to justify past choices rather than reverse them. And social accountability — the fact that decision-makers answer to constituencies who also made the original commitment — means that reversal requires persuading others to acknowledge shared failure.

These are real pressures, not cognitive errors to be easily corrected by reminding people that sunk costs are sunk. Behavioral economists have documented that the bias persists even when people are explicitly informed about it and asked to consider only future costs. The mechanism that produces the bias is not ignorance of the economic principle but something more fundamental: the human capacity to tolerate the dissonance between a past commitment and a present abandonment is limited, and the social and psychological cost of maintaining that dissonance rises with the magnitude of the commitment. Larger past investments produce stronger resistance to abandonment, which is precisely the opposite of the rational pattern, where larger past losses should actually increase the urgency of cutting further losses.

The Roman Pattern and Its Successors

Rome after Cannae did not collapse, which is often cited as evidence of Roman resilience and governance quality. This interpretation reverses the causation. Rome survived Cannae not because its institutions corrected the escalation error but because Hannibal lacked the siege equipment and supply lines to destroy walled cities, and the Roman confederation had enough demographic and productive depth to absorb the losses. The Senate’s doubling-down strategy was not vindicated by eventual Roman victory; Hannibal was ultimately expelled not by the strategy that produced Cannae but by a strategic shift to Fabian tactics — deliberate avoidance of pitched battle — that Rome had rejected for years because it felt like cowardice to a Senate emotionally committed to offensive engagement.

The more instructive case is Rome’s later decline, where sunk cost dynamics operated over decades rather than a single campaign season. The Roman commitment to defending the Rhine-Danube frontier required ever-larger armies, ever-larger payments to buy the loyalty of those armies, and ever-higher taxation to fund both. Each component of this commitment created constituencies with strong interests in its continuation and strong incentives to frame any reduction as unacceptable loss-booking. The legions stationed on the frontiers represented past investment in infrastructure, training, and positioning that their commanders genuinely believed should not be abandoned. The result was a strategic posture that the empire’s productive base could no longer sustain, maintained past the point of sustainability because the sunk costs of the existing commitment were too visible and too large to write off without a political crisis.

This pattern appears in every imperial overextension on the historical record. The Spanish Crown’s commitment to maintaining its Italian territories in the sixteenth century required military expenditures that consumed silver revenues from the Americas faster than they arrived. The appropriate response — a negotiated reduction of Spanish military presence in Italy — was repeatedly rejected because it would have required acknowledging that previous military investments had been excessive. The Ottoman Empire maintained unprofitable frontier garrisons long past the point where the territories they defended generated sufficient revenue to justify them, because the garrisons represented past conquest that the sultans were unwilling to treat as historical sunk costs. The British commitment to the Raj was sustained into the 1940s partly by the accumulated psychological and political investment of two centuries of administrators, soldiers, and politicians whose careers and identities were bound up in the imperial project.

Why Institutions Are Especially Vulnerable

Individual sunk cost errors are psychologically understandable. Institutional sunk cost errors are structurally predictable, and the mechanisms are distinct from the individual-level psychology. Organizations generate sunk cost problems through several specific pathways that have no individual equivalents.

Career incentives create systematic bias toward commitment. An executive who initiated a project that is now underperforming faces different incentives than a new executive evaluating the same project cold. The initiating executive’s career record includes the original commitment; acknowledging the project’s failure is acknowledging a personal error. A new evaluator has no such investment. This creates a selection effect: the people with the most accurate information about a failing commitment are precisely the ones with the strongest career incentives to misrepresent it, while outsiders who might evaluate it objectively lack the internal knowledge to assess it accurately.

Budget processes compound this problem. Annual budget cycles create baseline commitments that are extraordinarily difficult to reverse because they require active political effort to eliminate, while continuing them requires only inertia. Programs that were established in response to past conditions persist long after those conditions have changed because elimination requires going through a political process that defenders of the program will actively resist. The result is institutional accumulation of commitments that no individual decision-maker would endorse if evaluating them fresh, but that collectively form the budget that every organization operates within.

The worst institutional failures occur when these mechanisms combine with genuine uncertainty about outcomes. When a project is clearly failing — revenues are falling, costs are overrunning, objectives are unmet — the case for abandonment is at least clear in its terms. The harder cases, and the ones that produce the largest historical disasters, are commitments whose failure is gradual, ambiguous, or contested. Military occupations, large infrastructure projects, and long-horizon technology investments are all characterized by genuine uncertainty that is exploitable by committed advocates of continuation. The uncertainty provides cover for what is ultimately a psychological and political resistance to loss-booking masquerading as reasonable empirical dispute.

The Organizational Response That Works

History does provide examples of sunk cost problems being successfully corrected, and the mechanism in the successful cases is remarkably consistent. The correction requires either an external shock large enough to make the existing commitment visibly untenable, or the replacement of decision-makers in a way that severs the identity and career links to the original commitment.

Robert McNamara’s memoir of the Vietnam War is instructive precisely because it is a detailed retrospective accounting of how the American government’s escalation met none of the conditions for successful correction. The decision-makers who initiated the commitment remained in their positions long enough to have their career reputations bound up in it. The failure was gradual and contestable enough that optimistic interpretations of the evidence were always available. The social and political costs of abandonment — the acknowledgment of massive waste of life and resources — were distributed across an entire administration and political party rather than concentrated in a single accountable decision-maker. The result was continued escalation from a position that independent analysis had identified as untenable as early as 1966.

The contrast case is the British withdrawal from Palestine in 1948. The commitment to the British Mandate had consumed decades of administrative effort, military investment, and political capital. The withdrawal was rapid, complete, and acknowledged as such — not framed as strategic repositioning or temporary reduction. The decisiveness was possible partly because the commitment’s failure was publicly visible in a way that made continued escalation politically impossible, and partly because the incoming Attlee government had no personal investment in the commitment it was abandoning. External accountability and leadership replacement are the two reliable mechanisms that override the institutional sunk cost dynamic.

The Decision Rule That History Endorses

The historical record endorses a single consistent decision rule for dealing with large commitments: the question of whether to continue must be answered using prospective rather than retrospective analysis, and the people conducting that analysis must have interests that are independent of the original commitment. This is simple to state and difficult to implement because it requires institutional design that creates accountability structures separating initiation from continuation decisions.

The most successful long-run organizations in history — not the most impressive in any single period, but the ones that survived across the longest timescales — are invariably those that developed mechanisms for terminating commitments before sunk costs became existential. The Venetian Republic, which survived for over a thousand years, had elaborate institutional structures for removing unsuccessful doge and senior officials, creating the leadership replacement mechanism that is the single most reliable corrective to escalating commitment. The Catholic Church survived reformations, political assaults, and doctrinal crises across two millennia partly by being willing to abandon peripheral commitments — particular political alliances, specific theological positions — when the costs of maintaining them exceeded the benefits, however slowly and painfully this recognition developed.

The empires that fell — consistently and predictably — were those that could not develop the political technology to write off past investments when those investments had ceased to generate returns. They maintained commitments until the commitments destroyed them, not because their leaders were stupid or uninformed but because the psychology of large past investment made acknowledgment of failure feel worse than the costs of continuation. Rome at Cannae, Britain in America after 1775, the United States in Vietnam, the Soviet Union in Afghanistan: each case is a large institution in which the normal human response to loss — double down to recover — was permitted to operate at the scale of a state, without the institutional correctives that might have arrested the escalation before it became catastrophic.

The lesson is not that escalation is always wrong. Sometimes reinforcing a commitment is the right decision, because the underlying investment genuinely will pay off with additional resources. The lesson is that the psychological mechanisms that make escalation feel right are not reliable guides to when escalation actually is right. The past expenditure is never, by itself, a reason to continue. It is always, and only, the expected future return that justifies continued investment. Every institution that has forgotten this distinction has eventually been destroyed by the gap between what it needed to believe and what was true.