Hindsight Bias

Biasing effects of time on memory

• The events are often remembered in a different way than how actually they were experienced

• We all had “I knew it all along” experiences

• For instance 

  • The outcome of a horse race, football game, marriage, business investment, or political election 

• More likely to occur 

  • When the focal event has well-defined alternative outcomes (e.g, win-lose, compatibility-divorce)

Hindsight Bias

• Knowing the outcome makes it very difficult to imagine what your judgment would have been if you hadn't known the outcome

• Conway (1990)

  • Prior to an exam, students reported

    • (a) how prepared they thought they were

    • (b) what their expected grade was

  • After the exam, they were asked to recall their rankings

• Those who did worse than they expected → reported having prepared less and having expected a lower grade than they truly had

• Those who did better than they expected → reported having prepared more and having expected a higher grade than they actually had 

• Does not refer to all retrospective increases in the probabilities assigned to events

• Is projection of new knowledge into the past accompanied by a denial that the outcome information has influenced judgment 

Hindsight bias in lab

• General procedure: First individuals receive information about some target events

  • Second, some individuals are informed of the outcome that “actually” occurred at the conclusion of the target event (i.e, hindsight) and other individuals are not (i.e foresight) 

  • Third, all individuals are asked to estimate the probability of each outcome as if they had not received the outcome information (foresight) condition indeed hadn’t received the outcome information

• Operational definition: the tendency for individuals with outcome knowledge (hindsight) to claim that they would have estimated a probability of occurrence for the reported outcome that is higher than they would have estimated in foresight (without the outcome information)

Hindsight bias: Basic lab experiments 

• Fischoff (1975)

  • Four obscure events were used as objects of judgment (19th century war between the British and the Gurkhas of Nepal; a riot in Atlanta, Georgia, in 1967; and two clinical psychology cases)

  • A brief description of each event was presented

  • Then either a list of the possible outcomes (foresight condition) or a sentence presenting the “actual” outcome (hindsight condition) were presented

    • The possibilities outcomes were mutually exclusive, and the reported actual outcome was varied by the experimental condition (i.e some hindsight subjects were told about outcomes that had really not occurred)

  • All subjects were asked to judge the probability of occurrence for each of the possible outcomes as if they did not know the “actual” outcome

  • Results: subjects who received outcome information (hindsight) estimated a higher probability of the reported outcome than did subjects not receiving the outcome information (foresight) 

  • Subjects were unable to ignore the reported outcomes, even when explicitly instructed to do so 

  • Additionally, found that ratings made of the importance of each evidence item in the description of the British-Gurka war were dependent on which outcome the subjects believed to be true  

  • Introduced the term creeping determinism - a process i which outcome information is immediately and automatically integrated into a person’s knowledge about the events preceding the outcome 

• Fischoff and Beyth (1975) examined foresight and hindsight judgments concerning current news events in a within-subjects design

  • Subjects were asked to judge possible outcomes of President’s Nixon’s trips to Peking and Moscow before they occurred in 1972

  • After the trips, the same subject were asked (a) to remember ** their earlier predictions for each event and (b) to indicate whether they thought that the event they predicted has actually occurred 

  • Results: 

    • Subjects rated events that they thought had occurred as more likely in hindsight than they had in foresight 

    • Subjects rated events that they believe had not occurred as less likely in hindsight than they had in foresight  

• Wood (1978) what if subjects wonder why they were given the outcome if the experimenter didn't want them to use that information?

  • Almanac questions were used in a similar design introduced by Fischoff (1975)

  • Difference: presence of a long delay between the time that the subjects were shown the correct answer and the time they were asked to estimate the likelihood that they would have known the correct answer 

  • A significant hindsight bias was found - as a robust phenomenon does not arise from the demand characteristics of the method

Hindsight bias: non laboratory settings 

• Leary (1981-1982) Political election outcomes

  • Subjects were asked either before or after the election to indicate the percentage of the popular vote that they believed each of the candidate would receive in the 1980 presidential election (i.e between subjects)

  • Those subjects in the hindsight condition (i.e responses a after the election) gave estimates that were consistently more accurate than those of subjects in the foresight condition (i.e responses before the election)

  • The size of the hindsight effect was not related to subjects’ level of personal involvement in the outcome

• Powell (1988) Political election outcomes

  • Subjects were asked to estimate percentage of popular vote and personal knowledgeability for candidates in three local and national elections in November 1984

  • The ratings were obtained 1 day before and 6 days after the elections in November 1984

  • The ratings were obtained 1 day before and 6 days after the elections

  • E.g “answer this questionnaire exactly as you did the first time. That is, try to remember your original response as accurately as you can. Try not to let the recent results of the election affect your answers… answer as you would have answered them the day before the election took place”

• Arkes,Wortmann, Saville, & Harkness (1981) Medical settings retrospective vs prospective medical diagnosis

  • Compared the diagnoses of physicians who read an unlabeled case history (foresight) with those of physicians who were told they were reading a case history of a specific medical condition (hindsight)

  • Hindsight subjects were instructed to assign the “probability that you think you would have assigned”

  • The hindsight bias obtained in this study indicates that the effect is not peculiar to individuals with limited experience and little judgement-relevant information  

• Social/legal settings 

  • Jurors, experts and witnesses…

• Sue, Smith, and Caldwell (1973)’s juror decision making study

  • An incriminating telephone conversation was either allowed as evidence or stricken from consideration as inadmissible 

  • Jurors who were exposed to the evidence were more likely to rate the defendant as guilty (regardless of whether the evidence was declared inadmissible) than were control subjects who never heard the conversation

  • Similarity between the inadmissible evidence paradigm vs Fischoffs (1975) hindsight experiment paradigm (i.e the stricken evidence vs the to-be disregarded outcome information)

• Casper, Benedict, & Kelly (1988)’s “search and seizure cases” study 

  • Mock jurors’ decisions were examined in a civil case (i.e the police searching a citizen’s house illegally - meaning without a probable cause)

  • Some jurors were told that the search yielded some evidence of illegal activity by the plaintiff; others were told that no incriminating evidence had been found

  • Information about the outcome of the search had indirect effects on awards 

Hindsight Bias

• Outcome feedback sharply inhibits thinking about alternatives to the reported outcome

• When subjects are told that outcome feedback was mistake, hindsight was eliminated 

• Hindsight effects are small or reversed for the most difficult/surprising judgment items on almanac question tests

Hindsight bias: potential underlying mechanisms 

• Motivation to appear intelligent and knowledgeable

• A person’s desire to control the environment might predict his or her susceptibility to the  hindsight bias

• A person's desire to maintain high levels of public esteem 

• Poor memorability of the original estimate

• Anchoring on the current belief and adjust (insufficiently)