Feature

Psychologists have long been intrigued by the aptly named "hindsight bias"--the way the memory of our judgments changes when we learn the outcome of an event.

Research finds, for example, that a physician is apt to be far more confident in a diagnosis once he or she knows the ultimate outcome of a case than when making the actual diagnosis with no benefit of hindsight.

Researchers at the Center for Adaptive Behavior and Cognition (ABC) at the Max Planck Institute for Human Development in Berlin, Germany, now have a theory about what, at least in part, may cause hindsight bias: They speculate that it's a cognitive mechanism that allows us to unclutter our minds by tossing out inaccurate information and embracing the right answers--a kind of merge/purge for the mind.

The Max Planck team calls its theory of hindsight bias Reconstruction After Feedback with Take the Best (RAFT). And, in a new series of studies, they find that it accurately predicts up to 80 percent of the instances when hindsight bias would occur, not occur or even reverse.

According to the model, hindsight bias occurs during "reconstruction"--when people attempt to reconstruct their previous judgment about an event's outcome. The idea is that any feedback or correct information they receive after they have conducted their initial analysis changes the knowledge base underlying the original judgment and causes a bias toward the new information, says Ulrich Hoffrage, PhD, who conducted the studies with Ralph Hertwig, PhD, and Gerd Gigerenzer, PhD.

And rather than thinking of the phenomenon as a flaw of human cognition, as previous research has done, the researchers argue that it's a by-product of an adaptive mechanism.

Explains Hoffrage, "Hindsight bias is a cheap price we have to pay for a much larger gain: a well-functioning memory that is able to forget what we do not need--such as outdated knowledge--and that constantly updates our knowledge, thereby increasing the accuracy of our inferences."

The studies of the RAFT model are published in an article appearing in the May Journal of Experimental Psychology: Learning, Memory and Cognition (JEP:LMC) (Vol. 26, No. 3).

A little background

Traditionally, psychologists have described two types of hindsight bias. One type--and the one that the JEP:LMC studies focus on--occurs when people make a judgment or choice based on their current knowledge and are later asked to recall their judgment. If, in the interim, they're told what the correct judgment would have been, their memory of their own judgment becomes biased toward the new information. For example, if a man is asked to estimate the length of the Rhine River and he says 2,500 km but then learns that the correct answer is 1,320 km, he may later recall that his answer was 1,800 km.

The second type occurs when people are told about a situation and its outcome and are then asked to reconstruct the outcome based on the facts but ignoring what they know about the actual outcome. For example, economics students may be given a hypothetical scenario involving the U.S. economy, including economic indicators and the ultimate outcome--say a recession. If they're then asked to reconstruct what they would have predicted given the economic indicators, they're more likely to predict a recession and will be more confident in their decision than if they hadn't been told the true outcome.

Hindsight bias crops up during the reconstruction of an outcome, Hoffrage, Hertwig and Gigerenzer believe. And the reason can be explained by their RAFT model, which makes three general assumptions:

  • If people cannot remember the original judgment--which, of course will always be the case for hypothetical situations--they will reconstruct the judgment based on what they know about the situation.

  • Feedback about an event's outcome automatically updates a person's knowledge about the situation.

  • People will reconstruct their original judgment using the updated know-ledge rather than the knowledge they originally had. So while feedback does not directly affect a person's memory for the original response, it indirectly affects the memory by updating the knowledge used to reconstruct the response.

The RAFT model predicts that hindsight bias should be stronger in hypothetical situations because people must always reconstruct the outcome--and indeed, this is what several researchers have often found in their studies.

What mechanisms underlie the original judgment and the reconstruction process? The RAFT model assumes that, in both cases, the brain uses the same "fast and frugal" strategy--what the researchers call "take the best," which assumes that people base their judgments on one critical cue and ignore the rest. For example, a woman trying to decide whether to take preventive action against breast cancer is told that family history, age and diet are predictors of breast-cancer risk. Rather than trying to weigh and combine all these and myriad other predictors, she may just focus on what she believes is the most important one--family history. If it provides an obvious choice between prevention or not, she will make a decision based solely on that cue.

Testing the model

To determine whether the RAFT model accurately predicts hindsight bias, Hoffrage, Hertwig and Gigerenzer conducted two studies. In both, the researchers taught study participants about a topic and then asked them to make decisions based on what they learned. In the first study, they taught participants that fat content, number of calories and protein content can be used as cues to predict a food's cholesterol level and gave them these cue values for a set of foods. The researchers then showed participants a list of the same foods split into pairs and asked them to decide which item in the pair had more cholesterol and then say how confident they were in their choice.

Either a day or a week after the study, participants returned to the laboratory and had to remember the decisions they made about the food-item pairs and how confident they had been in their decisions. Some participants simply had to remember. Others first got to see the actual cholesterol content of each food item. And others not only got feedback on the cholesterol contents, but also got to see again the fat, calorie and protein values for each food item.

The RAFT model predicts, and the study supports, that this last condition--which reminds participants of the cues they used to make their original decision and thereby facilitates reconstruction--will actually reduce hindsight bias.

The second study was similar to the first, but rather than cholesterol content in food, participants had to make judgments about a person's blood pressure based on how they scored on three risk factors related to hypertension. As expected, overall, participants who received feedback on what the correct decision should have been--either the true cholesterol values of the foods or the applicants' blood pressures--were more likely to show hindsight bias than were participants who received no feedback.

In addition, Hoffrage and his colleagues found that when study participants received feedback, their know-ledge of the cues was updated to be more consistent with the feedback. In contrast, cue knowledge was unchanged in people who received no feedback. By plugging people's updated knowledge into the RAFT model, the researchers were able to make precise predictions about when hindsight bias occurs.

"RAFT is the first process model [of hindsight bias] that is able to predict for an individual item of an individual parti-cipant whether hindsight bias will occur, disappear or even reverse," says Hoffrage.

In fact, the researchers found that the model's predictions were accurate up to 80 percent of the time.

These studies are part of a larger program of research by the ABC Research Group into "fast and frugal heuristics"--the idea that the mind tends to use simple inference strategies, that require little information, for deciphering a problem rather than complex strategies that process lots of information. A report of their progress can be found in the recent book by Gigerenzer, Peter Todd, PhD, and the ABC Research Group, "Simple Heuristics That Make Us Smart" (Oxford University Press, 1999).


This article is part of the Monitor's "Science Watch" series, which reports news from APA's journals.