Hindsight Bias: Why Everything Looks Obvious Once You Know

Knowledge of an outcome cannot be set aside on request. It is absorbed into everything else you know about the situation, and the result is a story in which that outcome follows from the facts. Baruch Fischhoff called this creeping determinism: the past reorganizes itself around what happened, the signals that pointed the other way lose their weight, and what was one of several live possibilities becomes the only one that makes sense.

Neal Roese and Kathleen Vohs, reviewing the field in 2012, separate three things people mean by the term. There is memory distortion, where someone misremembers their own earlier prediction. There is a belief about the event, that it was bound to happen. And there is a belief about oneself, that you would have called it. A single study may find one and not the others, which is why the effect is measured in different ways in different literatures.

They also sort the causes into three levels. Cognitive: recall is selective and sense-making is active, so the facts that fit the outcome are the ones retrieved. Metacognitive: an outcome that is easy to understand feels like an outcome that was easy to predict, and the fluency gets misread as evidence. Motivational: a world in which events are orderly and foreseeable is more comfortable than one in which they are not, and there is also an interest in not having been the person who missed the warning.

Mark Pezzo's work puts a limit on the pattern. His sense-making model, supported by four experiments in 2003, holds that an unexpected outcome triggers more sense-making, which is what produces the bias; when the sense-making fails and the outcome stays baffling, the bias shrinks or reverses, and people say they never would have guessed. So hindsight bias is not a fixed tax on every surprise, and the bias and its reversal come from the same process at different degrees of success.

The costs are concrete. A decision reviewed after a bad outcome looks worse than it was, which makes blame too easy and makes people in high-risk jobs defensive rather than open. Hindsight also hides what a near miss should have taught: if the outcome was obvious, there was nothing to learn. And it feeds overconfidence about the future, because a past that looks predictable implies a future that should be.

Baruch Fischhoff's "Hindsight is not equal to foresight: the effect of outcome knowledge on judgment under uncertainty" appeared in the Journal of Experimental Psychology: Human Perception and Performance in 1975. Participants read a short passage of real but obscure history, including the 1814 campaign between British forces and the Gurkas of Nepal, and then rated the probability of each of four possible endings: a British victory, a Gurka victory, a stalemate with a peace settlement, and a stalemate without one.

One group of twenty rated the four endings with no outcome information. Each of the other groups was told that one particular ending was what had happened. Participants who had been told an ending rated that ending as substantially more probable than the uninformed group did, and they also reweighted the facts in the passage, treating the details that pointed toward the known ending as the important ones.

The second experiment closed the obvious escape route. Participants were told the outcome and then asked to respond as they would have if they had not known it. The instruction did not help. That is the design most later work uses, and it was repeated at scale in 2021 by Jieying Chen, Gilad Feldman and colleagues, whose close replication with 890 participants found hindsight bias in retrospective judgments with a mean effect size of 0.60 standard deviations; a second study replicated the prospective version with 608 participants, at 0.40.

Fischhoff and Ruth Beyth took the same question into real events the same year. Before President Nixon's 1972 trips to China and the Soviet Union, they asked people to put probabilities on a list of possible happenings. Weeks and months afterwards, the same people were asked to recall the probabilities they had given. The recalled figures had moved toward what had actually occurred, which is the memory-distortion form of the bias rather than the probability-rating form.

Hindsight arrives wherever a decision is reviewed after its result is in. The rows below are illustrations, not findings from the studies cited here.

The instruction to ignore what you know does not work, so the counters either create a record that hindsight cannot edit or force the alternatives back into view.

• Write predictions down, with the probability attached. A dated written forecast gives you a record against which to check your recollection.
• List reasons the other outcomes could have happened. Hal Arkes and colleagues gave 194 neuropsychologists a case history and told some of them which of three diagnoses was correct; the hindsight groups asked to write one reason why each diagnosis might be right succumbed less often than the hindsight groups who were not.
• Run a pre-mortem. Before the decision, write the story of how it failed. Afterwards that document competes with the story the outcome suggests.
• Review the record by date. Read what was known on day one before reading day two, and stop at each point to ask what a reasonable person would have concluded there.
• Judge the process, not the result. Ask whether the decision used the information available and the right method, which is a question the outcome cannot answer.
• In reviews of other people's decisions, state the outcome last. Whoever hears it first will struggle to unhear it.

Hindsight bias is about how predictable the past looks. Its neighbors are about how the past is judged and about what you can imagine not knowing.

• It is not outcome bias, which judges the quality of a decision by how it turned out. Outcome bias can operate even when someone fully accepts that the outcome was unlikely; hindsight bias changes the estimate of how unlikely it was.
• It is not the curse of knowledge, the difficulty of imagining what someone else does not know. That is about communication and other minds; hindsight bias is about your own earlier mind.
• It is not rosy retrospection, which remembers the past more fondly than it was. Both edit memory, but rosy retrospection changes how a past event felt rather than how foreseeable it was.