Daniel Kahneman and Amos Tversky distinguished two ways to forecast projects and decisions. Kahneman coined the labels in Thinking, Fast and Slow; Epstein and Flyvbjerg supply independent demonstrations in Range.
Inside view
The inside view builds a story from the specific details in front of you — this team's skills, this timeline, this unique challenge. It feels authoritative because you are the expert on your project. Kahneman's curriculum-writing team estimated two years; they took eight, and the client had moved on. Seymour the curriculum expert, asked for an outside comparison, knew that 40 percent of similar teams never finished and none finished in under seven years. The inside view ignored that base rate.
More internal detail makes forecasts more extreme, not more accurate. Venture capitalists predicted 50 percent higher returns for their own deals than for conceptually similar outside deals until forced to compare. wysiati explains why: the story feels complete once vivid particulars are in place.
Outside view
The outside view finds a reference class of structurally similar cases and asks how those turned out on average. It requires ignoring seductive uniqueness and hunting deep similarity — the pillar of analogical thinking at planning scale.
Dan Lovallo showed private-equity investors and movie fans could beat statistical models by generating diverse analogies. A single best analogy collapsed predictive power; a full reference class worked. Bent Flyvbjerg's tram-project consultants used the same method to flag a £320m estimate heading toward £1bn.
Reference class forecasting is the operational fix for the planning fallacy: start from the class baseline, then adjust modestly for this case — not the reverse.
Intuition vs formulas
Kahneman's overconfidence chapters add a related lesson: in low-validity environments, simple algorithms often beat expert judgment. Paul Meehl's meta-analysis (~200 studies): formulas win or tie in roughly 60% of comparisons; when humans get the formula's score and are allowed to override, they still do worse. Experts add complexity, "think outside the box," and overweight fragmentary case knowledge.
The boundary is not "never trust experts." It is validity: chess and firefighting in familiar buildings reward System 1 pattern match (kind environments). College admissions, medical prognosis, and long-horizon project outcomes sit at the wicked, low-validity end — where the outside view and mechanical rules help most.
Interviewers who "know" after forty-five minutes suffer the substitution heuristic: they answer whether the candidate interviews well, not whether she will succeed. Kahneman's Israeli Army officer-selection experience — years of confident predictions, near-zero predictive validity — is the institutional version.
When to use which
Kind environments with repeating patterns reward inside-view pattern match. Novel, high-uncertainty decisions reward outside view first, then evaluate options — the reverse of firefighter/chess-master instinct. Pair with premortem when the organization must still commit after the reference class looks ugly.
Sources
- thinking-fast-and-slow — primary: curriculum team, VC optimism, Meehl/meta-analysis, officer selection, planning fallacy mechanism
- range-why-generalists-triumph — Lovallo analogies, Flyvbjerg tram case, wicked-domain planning
- fooled-by-randomness — alternative histories, reference classes