Use reference classes to ground personal estimates in base rates

Before estimating how your situation will unfold, find similar past situations and check what happened.

Why it works

People generate predictions primarily from inside view — their own model of the specific case — which is systematically optimistic because it ignores how often similar plans fail in general. Reference class forecasting corrects this by grounding the prior in outside view: the statistical distribution of outcomes for similar cases. The two views are then combined rather than the inside view dominating.

How to do it

1. Define your situation type: "This is a home renovation project / a software product launch / a behavior change attempt."
2. Find data on how that type of project typically goes: timeline, cost overruns, success rates.
3. Use that base rate as your starting probability, then adjust for features that make your case better or worse than average.
4. Do not let the inside view increase your probability more than the specific distinguishing features actually justify.

Evidence

Reference class forecasting was developed by Daniel Kahneman and Amos Tversky and later formalized by Bent Flyvbjerg in infrastructure project planning. Studies show it reliably reduces planning fallacy — especially cost and time overruns — compared to inside-view estimates. (observational)

Reference class forecasting works best when good historical base rate data exist; for novel situations, the reference class itself is uncertain.

Sources

• Flyvbjerg (2006), from Nobel Prize to project management: getting risks right, Project Management Journal

Common mistake

Choosing the reference class that makes your project look most distinctive (therefore most likely to be exceptional) rather than the class that most honestly resembles it.