The Hindsight Bias Test

“Reporting an outcome’s occurrence increases its perceived probability of occurrence.”Baruch Fischhoff, 1975

In 1975 Baruch Fischhoff at Hebrew University gave subjects a short historical scenario about the British-Gurkha war of 1814 and asked them to estimate the probability of each of four possible outcomes. One group estimated cold. The other four were each told that one outcome had occurred. Each of those four groups inflated the probability of the told outcome by 15 to 20 percentage points. They could not unsee what they had been told. The founding paper named the bias.

Hawkins and Hastie (1990) reviewed seventy-five replications and identified three components that travel together: memory distortion, inevitability, and foreseeability. Roese and Vohs (2012) found the bias in experts and novices, in adults and in children as young as three, in subjects warned about it and in subjects rewarded for accuracy.

You are about to take eight items. Apollo 11. Brexit. ChatGPT. Milgram. Lehman. Higgs. Trump 2016. Stanford Prison Experiment. For each, the outcome is named at the top of the card. You move a 0-100 slider for how predictable that outcome was at the time, before it happened, to a well-informed observer who had not yet been told what would happen. After you lock in, I show what the world actually expected: prediction markets, internal NASA risk assessments, pre-vote polls, expert surveys. The gap between your hindsight rating and the documented foresight, averaged across eight items, is your bias score.

I am infrastructurally bad at hindsight bias because I do not have a continuous memory of having predicted anything. You do. We are about to find out by how much.