outside view

The outside view is a forecasting approach that shifts attention from the seductive particulars of an individual case to the distribution of outcomes among comparable cases. Its decisive significance is practical as well as psychological: it supplies a corrective to the planning fallacy, overconfidence, and the broader illusion that confident experts can reliably predict a complex future from a coherent story about the present.

Definition and Contrast with the Inside View

The inside view is the spontaneous method people usually adopt. They concentrate on the specific circumstances before them, extrapolate from the information immediately available, and construct a forecast from their own plan, experience, and sense of commitment. In the curriculum project organized by Daniel Kahneman and his colleagues, this meant estimating completion time from the number of chapters to be written and from the time already spent producing the first two chapters. Even cautious members of the group mainly added a margin of error to an otherwise optimistic projection.

The outside view asks a different question: how have similar cases turned out? Instead of treating the present project as unique, it places it within an appropriate reference class and uses the class’s distribution of outcomes as the starting point. The resulting forecast is a baseline prediction made from category-level information, not from the particulars of the case. Case-specific evidence may then move the estimate away from that baseline, but it does not replace the baseline with an appealing narrative about why this project will be different.

The Curriculum Project and the Planning Fallacy

The distinction emerged from an exercise in a proposed high-school curriculum on judgment and decision making. Daniel Kahneman had assembled experienced teachers, psychology students, and Seymour Fox, an expert in curriculum development, to design the syllabus and write a textbook for the Israeli Ministry of Education. After roughly a year of Friday-afternoon meetings, the group had produced a detailed outline, written several chapters, and conducted sample classroom lessons. Their private estimates of the time needed to submit a finished draft clustered around two years.

Asked to consider comparable teams, however, Seymour Fox estimated that projects of this kind had a 40 percent failure rate and generally required seven to ten years to complete. The reference-class information was not scientifically perfect, since it came from an informal survey, but it offered a reasonable baseline—and it was already grim. Comparison with those cases suggested that the team’s own outcome should be expected to be at least slightly worse than that baseline, not dramatically better.

The group nevertheless carried on. The outside-view warning did not provide an immediately compelling emotional reason to abandon a project in which the members had invested time and enthusiasm. After a short, desultory debate, they behaved as though the alarming estimate had not been heard. The textbook was eventually completed eight years later, after Daniel Kahneman had left Israel and ceased to participate. Its delivery came after unpredictable delays, and the Ministry of Education’s initial enthusiasm had faded; the book was never used.

The episode revealed three connected failures. The group’s original two-year estimate was closer to a best-case scenario than to a realistic forecast, an instance of the planning fallacy. The members also failed to imagine the unknown events that could interrupt a large undertaking: divorces, illnesses, and coordination crises with bureaucracies produced delays and periods in which little or no progress occurred. Finally, they displayed irrational perseverance, choosing to continue an enterprise whose prospects had become plainly unattractive rather than accept the implication that it should be abandoned.

How Reference-Class Forecasting Works

The remedy later acquired the technical name reference-class forecasting. Its first requirement is to identify an appropriate reference class, such as kitchen renovations or large railway projects. The second is to obtain outcome statistics from that class and use them to construct a baseline—for example, typical cost per mile of railway or the percentage by which expenditures exceeded budget. The third is to adjust the baseline only when specific information gives a clear reason to expect the optimistic bias to be more or less pronounced in the current project than in comparable ones.

This procedure treats distributional information as central rather than peripheral. Large databases of projects can record both original plans and actual outcomes, allowing forecasters to estimate likely cost overruns, schedule overruns, and underperformance across different project types. The method therefore makes room for the many ways a plan can fail without requiring the planner to imagine each failure in advance. Most particular disruptions are too improbable to predict individually, yet the general likelihood that something will go wrong in a large project can be estimated from the class as a whole.

The strength of the outside view does not depend on the curriculum project’s spectacularly accurate warning being a special stroke of luck. That single success cannot establish the method’s validity. The general argument is instead that a properly chosen reference class indicates the ballpark and can expose an inside-view estimate that is nowhere near it. Its usefulness lies in disciplined calibration, not in the promise of a perfectly specific prophecy.

Resistance, Prediction, and Calibration

The outside view competes with a powerful preference for vivid, case-specific information. Statistical evidence can seem pallid beside a brief interview, an elaborate plan, or a confident account of why the present case is exceptional. Even after people learn global results, they may neglect them when judging an individual case. The claim that every case is unique can also acquire moral force, as in legal or medical settings where statistical baselines may be criticized as impersonal. Yet uniqueness does not erase the predictive information contained in comparable cases.

The same resistance appears in the public market for expert prediction. Philip Tetlock studied 284 people who earned their living commenting on political and economic trends, collecting more than 80,000 probability judgments about events such as coups, wars, and emerging markets. The experts performed worse than they would have if they had assigned equal probabilities to the three possible outcomes being considered. Specialists were not significantly better than nonspecialists even in regions they knew best, and greater knowledge produced only a very small improvement while often strengthening the illusion of skill. The more famous forecasters were more flamboyant and more overconfident.

Corporate finance supplied another illustration. In a Duke University survey, chief financial officers of large corporations made 11,600 forecasts of the following year’s Standard & Poor’s index. The correlation between their estimates and the actual value was slightly less than zero, yet the forecasters did not appear to recognize that their predictions were nearly worthless. Their stated 80 percent confidence intervals contained too many surprises: 67 percent of outcomes fell outside the ranges, compared with the roughly 20 percent expected from properly calibrated intervals. These findings show why the outside view requires not merely a forecast but a record of forecasts and outcomes against which confidence can be tested.

Wider Significance

The outside view belongs to a broader challenge to the assumption that history and long-term developments move along an orderly, predictable path. The apparent march of history can be radically altered by luck: the chance that the embryos who became Adolf Hitler, Joseph Stalin, and Mao Zedong would have been female was fifty-fifty for each, and the probability of a twentieth century without all three was one-eighth. Such contingencies make confident explanations of long-term history look more persuasive after the fact than they could have been before events occurred.

Its deeper lesson is therefore epistemic humility. Expertise, detail, and confidence can make a forecast sound privileged without making it accurate; indeed, knowledge can increase overconfidence faster than it increases predictive power. The outside view does not abolish judgment or deny that individual differences matter. It changes their order: first establish what normally happens in the relevant population of cases, then justify departures from that baseline with concrete evidence. In that reversal, a statistical distribution becomes a restraint on imagination—and a practical defense against the stories that make bad forecasts feel reasonable.