Hindsight bias
Claim assessed: Once we know how something turned out, it seems to have been predictable.
Published · how we work
In short
Hindsight bias, also called the “I knew it all along” effect, means that once we learn how a situation turned out, we overestimate how predictable that outcome was. We misremember what we believed before, or imagine that anyone could have seen what was coming. The effect makes the past seem easier to foresee than it was and can stop us from learning from it.
What it says
The classic study. Fischhoff (1975) showed that people who learn the outcome of an event judge it more likely than those who don’t know it, and that they misremember their own predictions, moving them closer to the outcome. We haven’t read it; we describe it from Bernstein et al. (2011).
From 3 to 95 years. Bernstein, Erdfelder, Meltzoff, Peria and Loftus (2011) tested 194 people, from 3-year-old children to 95-year-olds. Every age group showed the effect, on all three tasks: general-knowledge questions and blurry pictures that gradually became clear. The effect was larger in preschoolers and older adults. Preschoolers simply replaced their original answer with the correct one. Older adults more often forgot their original answer and reconstructed it closer to the correct one. The authors note that the effect has been documented across cultures and in medical, legal, sports and election judgements.
How much the estimate rises. Salmen, Ermark and Fiedler (2023) asked 90 participants to estimate how likely an outcome was in scenarios about fictional experiments. Those who knew the outcome put it at 59% on average, against 50% for those who didn’t. Those who first estimated without knowing went up to 63% after finding out. But those who knew the outcome and were then asked to judge as if they didn’t went down to 49%, managing an undistorted estimate. Two weeks later, people mostly remembered the last answer they had given.
In experts. Banham-Hall and Stevens (2019) invited 93 physicians to assess three cases. For each case, some learned that the patient made a full recovery, others that the patient died shortly after discharge. In two of the three cases, the same care was rated as poor when the patient had died and as good or adequate when the patient recovered. Seniority made no difference. Schill, Gray and Brady (2023) showed that visual perception is affected too: 40 experienced radiologists recognised an abnormality in a blurry mammogram better if they had first seen the clear image.
Example
After a financial crisis, commentators explain why it was “obvious” it would come. If you look at what the same people wrote before the crisis, few predicted it. The example is an illustration, not a study cited here.
How to apply it
The steps below are a practical way to apply it that we propose, based on the studies cited.
- Write down your predictions beforehand. It’s the only sure way to know what you really believed, not what you now think you believed.
- When you judge a decision by its outcome, ask what could have been known at the time. See also the entry on outcome bias, a close relative.
- Explicitly ask for the “as before” judgement. Salmen et al. show that people can correct their estimate when clearly asked to ignore the outcome.
- In case reviews, hide the outcome when possible. Physicians judged the same care much more harshly when they knew the patient had died.
Limits and nuances
- The effect is large sometimes, small at other times. In the physicians’ study, one case in three showed no effect at all. The size depends on the situation.
- It isn’t only a memory error. Salmen et al. show that how the question is asked matters too: people answer according to what they think is being asked, and what they said last then becomes their memory.
- It overlaps with other effects. Outcome bias is about judging the decision; hindsight bias is about how predictable the outcome seemed. In practice they often appear together.
- We haven’t read the meta-analyses. There are meta-analyses from 1991 and 2004, available only behind a paywall. The claim that the effect is robust and documented across cultures comes from Bernstein et al.
- We excluded Lawrence Sanna’s work. He published several studies on hindsight bias but has had papers retracted for fabricated data.
- What we read. Bernstein et al. (2011), Salmen et al. (2023) and Schill et al. (2023), from PubMed Central, we read in full. For Banham-Hall and Stevens (2019) we read the abstract. We haven’t read Fischhoff (1975); we describe it from Bernstein et al.
Sources
- Baruch Fischhoff (1975). Hindsight is not equal to foresight: The effect of outcome knowledge on judgment under uncertainty
- Daniel M. Bernstein, Edgar Erdfelder, Andrew N. Meltzoff, William Peria, Geoffrey R. Loftus (2011). Hindsight bias from 3 to 95 years of age
- Edward Banham-Hall, Sian Stevens (2019). Hindsight bias critically impacts on clinicians' assessment of care quality in retrospective case note review
- Karolin Salmen, Florian K. G. Ermark, Klaus Fiedler (2023). Pragmatic, constructive, and reconstructive memory influences on the hindsight bias
- Hayden M. Schill, Samantha M. Gray, Timothy F. Brady (2023). Visual hindsight bias for abnormal mammograms in radiologists
See also: Curse of knowledge, Outcome bias