Bias · what we remember · 7 min read

Hindsight bias, and why the past looks predictable

Fischhoff's original outcome-knowledge studies, Roese and Vohs's later review of the three layers of the effect, and a write-it-down check on yourself.

In short

Hindsight bias is the tendency, once you know how something turned out, to feel that outcome was more predictable than it actually was beforehand, and sometimes to misremember your own earlier prediction as closer to the truth than it really was. Baruch Fischhoff demonstrated it in 1975 with a historical passage and a set of possible outcomes: people told which outcome had actually occurred rated that outcome as more probable in advance than people who were not told, even when asked to answer as though they did not know it. Susan Roese and Kathleen Vohs's 2012 review split the phenomenon into three related but separable parts, which is part of why it can be hard to argue someone out of it.

An everyday example

After a sports upset, commentators often describe the winning team's victory as though it had been building all along, pointing to signs that, before the game, nobody was treating as decisive. The same signs were sitting there before kickoff, available to everyone, and were not enough to make most people predict the result. Once the result is known, those same signs are read as though they had obviously pointed the right way from the start.

This is not usually dishonesty. It is a genuine shift in how the earlier evidence feels, once a piece of information, the actual outcome, has been added to the picture.

The classic experiments

Baruch Fischhoff's 1975 study gave participants a short passage describing an ambiguous historical military conflict between British and Gurkha forces, with several possible outcomes: a clear victory for one side, a stalemate that led to a peace settlement, or a stalemate with no settlement. One group of participants was told which outcome had actually happened and then asked to judge, in hindsight, how probable each of the possible outcomes had seemed in advance; a second group made the same judgement with no outcome information at all. Participants who were told the actual outcome rated it as having been substantially more probable in advance than participants who were not told, a gap that held up across several different historical and clinical passages Fischhoff tested the same way.

A related study that same year, by Fischhoff together with Ruth Beyth, pushed the same idea into memory itself. Before President Nixon's 1972 trips to China and the Soviet Union, Fischhoff and Beyth asked people to estimate the probability of a list of possible outcomes of the trips. After the trips were over, the same people were unexpectedly asked to recall, as accurately as they could, the probability they had originally given each outcome. People remembered having assigned a higher probability to outcomes that had actually happened, and a lower probability to outcomes that had not, than the estimates they had actually given beforehand. Knowing what happened had reached backward and reshaped the memory of what they had expected to happen.

Three different effects, not one

Roese and Vohs's 2012 review, drawing on decades of hindsight-bias research since Fischhoff's original study, split the phenomenon into three related but distinct components. The first is memory distortion: misremembering an earlier prediction as closer to the true outcome than it actually was, the pattern Fischhoff and Beyth's recall study demonstrated directly. The second is a sense of inevitability: judging, once an outcome is known, that it had to happen, almost regardless of how uncertain things genuinely looked beforehand. The third is foreseeability: the judgement that the outcome could, and should, have been predicted in advance, which is the component behind phrases like "we should have seen it coming." The three usually travel together, but they are not the same claim, and evidence, or an argument, that addresses one does not automatically address the others; someone can accept that an outcome was not inevitable while still overstating how foreseeable it was.

Does it replicate?

Replication grade: Strong: documented across many domains for decades; which of its three parts dominates varies by case

Hindsight bias is one of the more thoroughly and repeatedly documented effects in judgement research. Roese and Vohs's review surveys its appearance across a wide range of domains well beyond Fischhoff's original historical-outcome task: medical diagnosis, legal and jury judgements, evaluations of political and economic decisions, and everyday personal forecasts, generally finding the same shape, that known outcomes get read back into earlier expectations. The effect has proven robust across this range of settings and across the decades since it was first demonstrated.

What varies by domain and by study is which of the three components, memory distortion, inevitability or foreseeability, is doing most of the work, and how large the effect runs; some manipulations that reduce one component, such as asking people to list reasons the outcome might not have happened, do less to reduce the others. The overall pattern is not in serious dispute; which mechanism to blame in a given case, and how to counter it, is where the more specific questions sit.

The branches that fade once one is known

A schematic, not a real forecast. Before the fork, several outcomes looked roughly as plausible as each other, drawn as thin lines. After one of them happens, drawn bold, the other branches fade from memory, and the path taken can start to look like it was the only one that was ever likely.

The diagram on this page is a schematic, not a real forecast. It draws several branches fanning out from a decision point, roughly equally weighted, and then fades every branch except the one that actually happened, once it is known. The point of drawing it that way is that the other branches did not become less plausible at the moment the outcome occurred; they only stopped being remembered as plausible.

Try it: three scenarios

Three short situations that ask what a forecast looked like before an outcome was known, compared with how it tends to be remembered afterward.

1. Before a close election, a person tells a friend they think either candidate could plausibly win. After their preferred candidate wins, they recall having been fairly confident of the result all along. What does this recollection most likely reflect?

A genuine change of mind that happened during the campaignA memory distortion: the outcome is being read back into a prediction that was actually much less certainNew information that only became available after the result

Why

This is the memory-distortion piece of hindsight bias: once an outcome is known, people often recall their earlier uncertainty as smaller than it actually was, without any new evidence justifying the change.

2. After a company's product fails, an analyst who never publicly predicted the failure says the warning signs were obvious and the failure was bound to happen. What is this closest to?

A prediction the analyst is choosing not to mentionA sense of inevitability, reading an outcome as though it had to happen, once it is already known that it didA correct account of how predictable the failure genuinely was in advance

Why

Treating an outcome as though it were always bound to happen, once it is known that it did happen, is the inevitability component of hindsight bias, separate from whether the failure was actually foreseeable at the time.

3. A manager reviewing a project that went wrong says the team "should have seen it coming," without having raised any concerns before the project started. What question best checks whether this claim is fair?

What information was actually available to the team before the outcome was known, not afterHow the project eventually turned outWhether the manager is generally an optimistic or pessimistic person

Why

The foreseeability component of hindsight bias is judging that something should have been predicted using knowledge that was only available after the fact; the fair test is what was actually knowable at the time, not what is obvious once the outcome is known.

Nothing you type on this page leaves your browser tab: no answer, guess or score is sent anywhere or stored.

How to catch it

Because the distortion is in memory itself, the most reliable fixes involve writing things down before the fact, so there is a record hindsight cannot quietly edit.

Write predictions down, with a number if possible, before an outcome is known, and keep the record rather than relying on memory afterward.
When reviewing a decision that went wrong, ask what information was actually available at the time, not what is obvious now that the outcome is known.
Before concluding an outcome was inevitable, list at least one plausible way events could have gone differently, using only information available before the outcome.
Separate the three questions explicitly: did I predict this, was this bound to happen, and could this have been foreseen; they can have three different honest answers.
Be cautious of any post-outcome review, including your own, that does not reference a prediction made and recorded before the fact.

Check yourself: three questions

1. In Fischhoff's 1975 study, what happened to participants who were told the actual outcome of the historical conflict before judging how probable it had seemed in advance?

They rated the actual outcome as having been more probable in advance than participants who were not toldThey judged all outcomes as equally probable, regardless of what actually happenedThey refused to make a judgement once they knew the outcome

Why

Knowing the actual outcome raised participants' rated probability for that outcome, compared with participants who made the same judgement without knowing what had actually happened.

2. In Fischhoff and Beyth's related 1975 study of predictions about Nixon's trips to China and the Soviet Union, what happened when participants were later asked to recall their own original probability estimates?

They accurately remembered their earlier predictionsThey recalled having predicted the actual outcomes more strongly than they really hadThey could not remember making any earlier prediction at all

Why

Participants recalling their own genuine earlier estimate misremembered it as closer to the outcome that later occurred, showing the distortion reaches into memory of one's own past judgement, not just judgements made fresh in hindsight.

3. According to Roese and Vohs's 2012 review, which three components make up hindsight bias?

Memory distortion, a sense of inevitability, and a judgement of foreseeabilityOptimism, pessimism, and neutralitySpeed, accuracy, and confidence

Why

Roese and Vohs distinguished misremembering a past prediction, believing an outcome was bound to happen, and believing it could have been predicted, as three related but separable pieces of the overall effect.

Sources

Text on this page is original to MyTestAtlas, written from the studies listed. The diagram is drawn by this site and is not a copy of any published figure.