hindsight ^foresight: the effect of outcome …hindsight *£ foresight 289 numbers," the belief...

Journal ol Experimental Psychology:Human Perception and Performance1975, Vol. 1, No. 3, 288-299

Hindsight ^Foresight: The Effect of Outcome Knowledgeon Judgment Under Uncertainty

Baruch FischhoffHebrew University of Jerusalem, Israel

One major difference between historical and nonhistorical judgment is thatthe historical judge typically knows how things turned out. In Experiment1, receipt of such outcome knowledge was found to increase the postdictedlikelihood of reported events and change the perceived relevance of event-descriptive data, regardless of the likelihood of the outcome and the truthof the report. Judges were, however, largely unaware of the effect thatoutcome knowledge had on their perceptions. As a result, they overesti-mated what they would have known without outcome knowledge (Experi-ment 2), as well as what others (Experiment 3) actually did know withoutoutcome knowledge. It is argued that this lack of awareness can seriouslyrestrict one's ability to judge or learn from the past.

Hindsight and foresight differ formally inthe information available to the observer.The hindsightful judge possesses outcomeknowledge, that is, he knows how thingsturned out. The foresightful judge doesnot. Although foresight usually implieslooking at the future, in the absence of out-come knowledge, past and future events canbe equally inscrutable.

The studies presented here ask two ques-tions about the judgmental differences be-tween hindsight and foresight: (a) Howdoes receipt of outcome knowledge affectjudgment? (b) How aware are people of theeffects that outcome knowledge has on theirperceptions? Answers to these questionswill shed light on how people do learn andmight better learn from history.

The two hypotheses explored are: (a)

Preparation of this report was supported by theAdvanced Research Projects Agency of the De-partment of Defense (ARPA Order 2449) andwas monitored by the Office of Naval Researchunder Contract No. N00014-73-C-0438 (NR 197-026). The research reported constitutes part ofa doctoral dissertation submitted to The HebrewUniversity of Jerusalem. I am deeply indebted toAmos Tversky, Daniel Kahneman, Paul Slovic,Ruth Beyth, and Sarah Lichtenstein for their con-tributions to this project. The detailed commentsof two anonymous reviewers on a previous draftare gratefully acknowledged.

Requests for reprints should be sent to BaruchFischhoff, Oregon Research Institute, P.O. Box3196, Eugene, Oregon 97403.

Reporting an outcome's occurrence increasesits perceived probability of occurrence; and(b) people who have received outcomeknowledge are largely unaware of its havingchanged their perceptions in the manner de-scribed in the first hypothesis. In combina-tion, these two hypotheses indicate that re-porting an outcome produces an unjustifiedincrease in its perceived predictability, for itseems to have appeared more likely than itactually was.

Indirect support for the first hypothesismay be found in a variety of sources. Forexample, the historian Georges Florovsky(1969) notes:The tendency toward determinism is somehow im-plied in the method of retrospection itself. Inretrospect, we seem to perceive the logic of theevents which unfold themselves in a regular orlinear fashion according to a recognizable patternwith an alleged inner necessity. So that we getthe impression that it really could not have hap-pened otherwise, (p. 369)

An apt name for this hypothesized tend-ency to perceive reported outcomes as hav-ing been relatively inevitable might be"creeping determinism"—in contrast withphilosophical determinism, which is the con-scious belief that whatever happens has tohappen.

Phenomena resembling creeping determi-nism have been noted by psychologists aswell as historians. One example is Tverskyand Kahneman's (1971) "law of small

288

HINDSIGHT *£ FORESIGHT 289

numbers," the belief that data which wereobserved more or less had to be observed. Asecond example is the tendency to reworkor reconstruct the biographies of deviantsto show that their present diagnoses (labels)are inevitable products of their life histo-ries (Lofland, 1969; Rosenhan, 1973; Schur,1971). A third is the defensive attributionof responsibility for accidents, a process inwhich people carefully scrutinize the datadescribing accidents in order to uncover orimpose a pattern that will increase their per-ceived predictability and avoidability (Wals-ter, 1967).

All of this evidence for creeping determi-nism is, however, either indirect, imprecise,unsystematic (anecdotal), or confounded bymotivational and emotional issues. Experi-ment 1 directly tested the validity of thecreeping determinism hypothesis and ex-plored some of the concomitant effects ofoutcome knowledge on judgment.

EXPERIMENT 1

Method

Design. The six subexperiments described inthis section are identical except for the stimuliused. In each, subjects were randomly assignedto one of five experimental groups, one Beforegroup and four After groups. In each subexperi-ment, the Before group read a brief (150 word)description of a historical or clinical event forwhich four possible outcomes were provided. TheAfter groups read identical passages to which afinal sentence presenting one of the possible out-comes as the "true" outcome had been added. Asthe possible outcomes were mutually exclusive,three of the four After groups received "true"outcomes that actually had not happened. Sub-jects in all groups were asked to (a) estimate thelikelihood of occurrence of each of the four pos-sible outcomes, and (b) evaluate the relevance ofeach datum in the event description. In two ofthe subexperiments subjects were also asked toindicate the relative extent to which they relied onthe passage and on outside information.

Instructions. The cover sheet of each question-naire read:

In this questionnaire we are interested inknowing how people judge the likelihood ofpossible outcomes of social events. A passagedescribing an unfamiliar historical event ap-pears below. We will ask you to evaluate theprobability of occurrence of each of the four

possible outcomes of the event [including thatwhich actually happened—for After subjects] inthe light of the information appearing in thepassage.

A typical passage, as taken from Woodward's(1938) The Age of Reform, was:

[1] For some years after the arrival of Hastingsas governor-general of India, the consolidationof British power involved serious war, [2] Thefirst of these wars took place on the northernfrontier of Bengal where the British were facedby the plundering raids of the Gurkas of Nepal.[3] Attempts had been made to stop the raidsby an exchange of lands, but the Gurkas wouldnot give up their claims to country under Britishcontrol, [4] and Hastings decided to deal withthem once and for all. [5] The campaign beganin November, 1814, It was not glorious. [6]The Gurkas were only some 12,000 strong; [7]but they were brave fighters, fighting in terri-tory well-suited to their raiding tactics. [8JThe older British commanders were used to warin the plains where the enemy ran away froma resolute attack. [9] In the mountains of Nepalit was not easy even to find the enemy. [10]The troops and transport animals suffered fromthe extremes of heat and cold, [11] and the offi-cers learned caution only after sharp reverses.[12] Major-General Sir D. Octerlony was theone commander to escape from these minor de-feats, (pp. 383-384)

The possible outcomes offered were: (a) Brit-ish victory, (b) Gurka victory, (c) military stale-mate with no peace settlement, and (d) militarystalemate with a peace settlement. For After sub-jects, the appropriate outcome was appended tothe passage in the form of an additional sentence,such as, "The two sides reached a military stale-mate, but were unable to come to a peace settle-ment."

Following the passage, subjects were asked, "Inthe light of the information appearing in the pas-sage, what was the probability of occurrence ofeach of the four possible outcomes listed below.(The probabilities should sum to 100%)." On thefollowing page, each datum appeared on a separateline followed by a seven-point scale on whichsubjects were asked to indicate "how relevant orimportant each datum in the event description wasin determining the event's outcome." The numbersin the passage above indicate the division intodatum units. They did not appear in the passagepresented to subjects.

Stimulus selection. Four different events wereused to achieve greater generality for the resultsobtained: Event A, the British-Gurka strugglecited above; Event B, the near-riot in Atlanta,Georgia in July 1967, as described in the KernerCommission Report on Civil Disorders (NationalAdvisory Commission, 1968, pp. 53-5); EventsC and D, clinical cases reported by Albert Ellis

290 BARUCH FISCHHOFF

(1966, pp. 262-264).* For Events C and D, theword "social" in the instructions was replaced by"individual" and the word "historical" was de-leted.

Several methodological considerations guided theevent selection process: (a) The event should besufficiently familiar to permit intelligent responsesand sufficiently unfamiliar to rule out the possibilityof subjects knowing what really happened—espe-cially those receiving false outcome reports, (b)Past events were used to allow provision of "true"outcomes to the After groups.2 (c) The space ofpossible outcomes had to be readily partitionable.For Events B, C, and D, the set of outcomes wasconstructed to be mutually exclusive and exhaus-tive. Although this is not the case for Event A,pretests indicated that these four outcomes con-stituted an effective partition.

Subjects. Approximately equal numbers of sub-jects participated in each group in each subexperi-ment. Event A (Gurkas) was administered twice,once to a group of 100 English-speaking studentsrecruited individually at The Hebrew Universitycampus in Jerusalem and once to a class of 80Hebrew-speaking subjects at the University of theNegev in Beer Sheba. Event B (riot) was ad-ministered to two separate classes at The HebrewUniversity, one containing 87 Hebrew-speakingpsychology majors with at least one year's study ofstatistics and one of 100 Hebrew-speaking stu-dents with no knowledge of statistics. Event C(Mrs. Dewar) was administered to the 80 Uni-versity of the Negev students; Event D (George)to the 100 Hebrew University students withoutstatistics training.

Procedure. Questionnaires for the various ex-perimental groups were distributed randomly.Subjects devoted 20-30 min to the completion ofeach questionnaire.

ResultsProbability estimates. Table 1 presents

the mean probability assigned to each out-come by subjects in each experimental groupfor each subexperiment.

Similar patterns of data emerged in thetwo subexperiments using Event A (differ-ing in subjects' language) and in the twousing Event B (differing in subjects' knowl-

1 Copies of these stimuli with the four offeredoutcomes will be supplied with all requested re-prints. The permission granted by the OxfordUniversity Press and Ronald Press to use thesecopyrighted materials is gratefully acknowledged.

2 It might be wondered whether Before subjectsmight not behave like After subjects with referenceto their predictions, which they know are actuallypostdictions. In a series of five experiments(Fischhoff, in press), we found that manipulatingthe temporal setting of possible outcomes has noeffect on their perceived likelihood.

edge of statistics). For the sake of tabularbrevity, only one subexperiment in each pairis presented.

The creeping determinism hypothesis pre-dicts that After subjects told that a particu-lar outcome has happened will assign it ahigher probability than will Before subjects.Four outcomes reported to different groupsin each of six subexperiments afford 24 op-portunities to test the hypothesis. The criti-cal comparisons are between the outlineddiagonal cells (those indicating the meanprobability assigned to an outcome by sub-jects for whom that outcome was reported tohave happened) and the Before cell in thetop row above them.

In each of the 24 cases, reporting an out-come increased its perceived likelihood ofoccurrence (p < .001; sign test). Twenty-two of these differences were individuallysignificant (p < .025; median test). Thusthe creeping determinism effect was obtainedover all variations of subject population,event description, outcome reported, andtruth of outcome reported. The differencesbetween mean Before and After probabili-ties for reported outcomes ranged from3.6% to 23.4%, with a mean of 10.8%.Slightly over 70% of After subjects assignedthe reported outcome a higher probabilitythan the mean assignment by the corre-sponding Before subjects.

No outcome was judged inevitable by anyBefore subject, whereas a small proportion(2.1%) of After subjects did assign 100%to reported outcomes. Evidently, most Aftersubjects felt that in the light of the factsgiven in the description, other (unreported)outcomes were still possible (e.g., "TheGurkas had a 70% chance of winning, butthe British still might have pulled it off").Similarly, After subjects found a higher per-centage of unreported outcomes to have beenimpossible (as indicated by a probability of0%) than did Before subjects (11.5% ver-sus 8.0%).

Another way to appraise the extent ofcreeping determinism is to translate meanBefore probabilities into the form of a prioriodds and the mean After probabilities forreported outcomes into a posteriori odds.

HINDSIGHT ̂ FORESIGHT 291

TABLE 1MEAN PROBABILITIES ASSIGNED TO EACH OUTCOME, EXPERIMENT 1

Experimental groupOutcomeprovided

Outcome evaluated

Event A: British-Gurka struggle (English-speaking subjects)

Before 20After 20

202020

none 33.8 21.3 32.3 12.31 57. 2 14.3 1S.32 30.3 38,4 20.43 25.7 17.0 48.04 33.0 15.8 24.3

13.410.59.9

27.0

Event B: Near-riot in Atlanta (subjects with knowledge of statistics)

BeforeAfter

2020202020

none1234

11.2 30.8 43.8 14.2[ 30.6

5.53.9

16.7

25.851.823.931.9

23.324.350.823.4

20.318.521.427.9

Event C: Mrs. Dewar in therapy

Before 19After 13

171617

none 261 432 263 304 21

6 15.8 23.4 34.41 13.95 23.26 14.12 10.2

17.313,434.122.6

25.836.921.346.1

Event D: George in therapy

BeforeAfter

1718182017

none 27.4 26.91 33.62 22.43 20.54 30.6

20.841.822.319.5

39.4 6.337.828.950.037.7

8.07.17.3

12.3

Note, The actual outcomes are numbers 1, 1 ,4 , and 2 for Kvents A, B, C, and D, respectively. Outlined cells are those withAfter probabilities of reported outcomes.

The ratio of prior and posterior odds foroutcome i provides a sort of average likeli-hood ratio for the impact of the datum"Outcome i did actually occur" (where thetwo hypotheses are "Outcome i occurs" and"Outcome i does not occur"). Over the24 outcomes reported, these likelihood ratiosvaried from 1.2 to 3.5 (M = 1.96). Thusin the" present sense, reporting an outcome'soccurrence approximately doubles its per-ceived likelihood of occurrence.

Because the outcomes varied considerablyin their mean Before probability (from 6.3%to 44.0%), reporting their occurrence maybe seen as confirming (or disconfirming)

subjects' expectations to varying degrees.There was a highly significant negative cor-relation (r=.435; p < .001) between theprior odds and likelihood ratios associatedwith reported outcomes (as computed in thepreceding paragraph). Thus, the more un-likely an outcome report, the greater theimpact it has.

Relevance judgments. Table 2 presentsthe mean relevance judgments for eachdatum in one subexperiment. Inspection re-veals that the relevance attributed to anydatum is highly dependent on which out-come, if any, subjects believe to be true.Some of these differences seem readily inter-


TABLE 2MEAN DATA RELEVANCE JUDGMENTS FOR EVENT A,

EXPERIMENT 1 (HEBREW-SPEAKING SUBJECTS)

Datum number

Outcome reported

NoneBritish victoryGurka victoryStalemate

with peace treatywithout peace treaty

l

4.504.783.66

4.504.94

2

5.114.444.83

4.725.50

3

4.225.283.55

4.554.39

4

5.784.834.44

5.895.11

5

4.504.615.89

5.505.33

6

6.004.445.11

4.175.11

7

5.504.614.11

4.224.78

8

5.444.564.61

5.004.39

9

4.395.723.72

4.224.17

10

4.565.335.22

5.223.72

11

4.285.784.11

4.894.50

12

5.564.114.78

4.944.61

pretable. For example, the fact that "theBritish officers learned caution only aftersharp reverses" (Datum No. 11) wasjudged most relevant by subjects told of aBritish victory, and rather irrelevant by sub-jects told of a Gurka victory.

A less impressionistic analysis on the ef-fects of outcome knowledge on relevancejudgments proceeded in the following man-ner. For each subexperiment, a two-way(outcome reported X datum evaluated) fixedmodel analysis of variance (ANOVA) wasperformed on subjects' judgments of datarelevance. To accommodate the varyingnumber of subjects in the experimentalgroups of subexperiments, the following pro-cedure was adopted: The analysis was re-peated three times to produce maximum,minimum, and middle solutions. For themaximum solution, subjects were randomlysampled from the smaller experimentalgroups and their responses duplicated, equat-ing the size of all cells. For the minimumsolution, subjects were randomly deletedfrom the larger groups until cell size wasequated. For the middle solution, a com-bination of duplication and elimination wasperformed.

The same ANOVA was performed on thethree sets of data. The results discussedhold for all three solutions, (a) In eachsubexperiment, there was a significant Out-come Reported X Datum Evaluated interac-tion reflecting the differential effect of out-come knowledge on relevance judgmentsdiscussed above and shown in Table 2. (b)Over the six subexperiments, only two weakoutcome-reported effects emerged. Thus,there is no indication of an entire set of data

having greater mean relevance for one out-come than another, (c) Datum-evaluatedeffects appear in all but one subexperiment.They reflect data perceived to be either rele-vant or irrelevant whatever happens (e.g.,Hastings' decision to deal with the Gurkas"once and for all" [Datum No. 4] was uni-versally perceived as relevant).

Reliance. Subjects in two of the sub-experiments were asked to indicate with anumber from Q% to 100% the extent towhich they had relied on the material pre-sented in the passage compared with general(outside) knowledge. In both cases, Beforesubjects indicated relying significantly (/> <.05; median test) more on the passage thandid After subjects.

Discussion

Reporting an outcome's occurrence con-sistently increases its perceived likelihoodand alters the judged relevance of data de-scribing the situation preceding the event.Hindsight-foresight differences in perceiveddata relevance have also been noted by his-torians observing the creeping determinismeffect. Consider, for example, Tawney(1961): "Historians give an appearance ofinevitability to an existing order by dragginginto prominence the forces which have tri-umphed and thrusting into the backgroundthose which they have swallowed up" (p.177). Or, consider Wohlstetter (1962):"It is much easier after the event to sortthe relevant from the irrelevant signals.After the event, of course, a signal is alwayscrystal clear. We can now see what disasterit was signaling since the disaster has oc-curred, but before the event it is obscure and

HINDSIGHT 76 FORESIGHT 293

pregnant with conflicting meanings" (p.387).

How justified are the judgmental changesaffected by receipt of outcome knowledge?It is hard to say, simply because there is nounequivocal way to assign probabilities orrelevance weights to unique events like theBritish-Gurka struggle. If, for example,someone claims that there was no chance(or a 7% chance or a 98.6% chance) of athermonuclear war during the 1960s, whocan prove him wrong? Indeed, the onlywrong estimate is that it was 100% likely.

Were such events well defined and re-occurring, the wisdom of increasing thepostdicted probability of some reported out-comes would be readily apparent. Considera judge who has previously encountered fourBritish-Gurka-type struggles, each indistin-guishable from the one used here, two ofwhich were won by the Gurkas. Uponlearning of another Gurka victory, he mayproperly update that outcome's predictive(Before) probability of 50% to a higherpostdictive (After) probability. Hearing ofa Gurka victory may also justify somechange in relevance judgments by showing,for example, the true importance of Britishsuffering from climatic extremes. It mayalso teach him something about the natureof 19th century colonialism—and thus changethe sort of "laws" or reasons he uses indrawing inferences from the event descrip-tion.

Thus, the judgmental changes which wehave called creeping determinism could con-ceivably reflect what judges learn from out-come reports. The skepticism expressed inthe anecdotal observations presented above(e.g., Florovsky, Schur, Tawney), however,suggests that this is not the case. In thelight of these comments, it appears that vvhatpasses for the wisdom of hindsight oftencontains heady doses of sophistry—that theperceived inevitability of reported outcomesis imposed upon, rather than legitimately in-ferred from, the available evidence.

As described in these accounts, postdictivelikelihood estimates are exaggerated througha largely unconscious process evoked by re-ceipt of outcome knowledge. How awarepeople are of the effect that outcome knowl-

edge has on their perceptions was examinedin Experiment 2. Aside from helping toclarify the nature of creeping determinism,these results have considerable intrinsic in-terest. Awareness is clearly crucial to know-ing what one has learned from the past(i.e., from outcome knowledge). It may benecessary for learning from the past at all.

EXPERIMENT 2Method

Design. Subjects were presented stimulus ma-terials identical to those used in the After groupsof Experiment 1, with each event description ac-companied by a "true" outcome. They were askedto respond "as they would have had they notknown the outcome." For each of the four eventsthere were four After (ignore) groups, one re-ceiving each possible outcome as true. If sub-jects are aware of the effect of outcome knowl-edge on their judgments, the responses of all ofthe After (ignore) groups should resemble thoseof that Before group in Experiment 1 which dealtwith the same event. If After (ignore) subjectsare completely unable to ignore the effect of out-come knowledge, their responses should resemblethose of the After group in Experiment 1 whichreceived the same outcome as "true."

Instructions. The cover of each test bookletread:

A number of short descriptions of real socialand personal events appear below, each with anumber of possible outcomes. On the basis ofthese data, we ask you to evaluate the likelihoodof the outcomes listed. We thank you for yourparticipation.

Each remaining page of the test booklet was identi-cal to the corresponding page of the Experiment 1booklet, except that each response section waspreceded with the instruction to "answer as youwould have had you not known what happened."

Subjects. Eighty members of an introductorystatistics class at the University of the Negevparticipated.

Procedure. Questionnaires were randomly dis-tributed to a single group of subjects. Each sub-ject received one version of each of the four dif-ferent events. In a test booklet, Events A, B, andC alternated systematically as the first three events,with Event D (the least interesting) always ap-pearing last. Order was varied to reduce thechances that subjects sitting in adjoining seatseither copied from one another or discovered theexperimental deception. All materials were inHebrew. Questionnaires were anonymous.

Results

Probability estimates. Table 3 presentsmean probability assignments by subjects in


TABLE 3MEAN PROBABILITIES ASSIGNED BY SUBJECTS RESPONDING "As IF You DID NOT

KNOW WHAT HAPPENED," EXPERIMENT 2

Kxperimental groupOutcomeprovided

Outcome evaluated

Event A: British-Gurka struggle

BeforeAfter (ignore)

1720IS1818

none1234

29.4129.8

38.022.118.1

23.S27.421.731.832.9

34.7 12.424.919.731.928.9

18.420.714.321.2

Event B: Near-riot in Atlanta


3917212020

none1234

11.3[24.6

9.06.3

13.3

29.027.041.524.520.3

43.9 16.328.336.443.536.5

19.813.125.824.0



1919191520

none 26.61 36.42 24.73 25.14 18.3

15.8 23.4 34.410.2 16.1 37.428.8 15.5 31.913.7 34.9 26.412.3 21,8 52.8



1717182018

none 261 412 243 184 21

4 26.9 39.4 6.38 16.56 35.93 20.40 21.1

35.332.457.338.4

6.57.04.0

19.6

Note. In each case the Before results are taken from the corresponding before (no outcome) group in Experiment 1 (subjectswho actually responded not knowing what happened). Outlined cells are those with After(ignore) probabilities of repo rteoutcomes.

each of the After (ignore) groups along withthe responses of the corresponding Beforegroups from Experiment 1. (The Hebrew-speaking group is used for Event A, thepooled responses of both relevant subexperi-ments for Event B.) The entries in eachrow will be called a profile. They indicatethe probabilities subjects believed theywould have assigned to the outcomes hadthey not known "what really happened."

These reconstructed probabilities indicateno more than marginal awareness of the ef-fects of outcome knowledge. In 13 of 16cases, the mean After (ignore) probability of

the reported outcome was higher than themean Before probability for the same event.For reported outcomes the mean Before-After (ignore) difference of 9.2% wasslightly but not significantly less than the10.8% mean Before-After difference in Ex-periment 1 (p > .10; Mann-Whitney Utest).

The After (ignore) profiles closely resem-bled the corresponding After profiles. For14 of 16 profiles, the mean absolute differ-ence between corresponding cells was smallerfor the After(ignore)-After comparison thanfor the relevant After (ignore)-Before com-


parison (/> < .002; sign test). The medianabsolute difference between correspondingcells was 3.7% for After (ignore)-After,and 6.4% for After (ignore)-Before (p <.001; Mann-Whitney U test). There is noapparent reason, other than sampling error,for the weaker results obtained with Event A.

Relevance judgments. If After(ignore)subjects are able to ignore outcome knowl-edge, the outcome report they receivedshould have no effect on their reconstructedrelevance judgments. Instead, however,these relevance judgments clearly reflectedthe outcomes that After (ignore) subjectsbelieved to have happened (but were in-structed to ignore). For example, in Ex-periment 1 After subjects told of a Britishvictory assigned substantially greater impor-tance to the fact that "British officers learnedcaution only after sharp reverses" (DatumNo. 11) than did Before subjects; those toldof a Gurka victory assigned it slightly lessimportance. After (ignore) subjects in Ex-periment 2 who were asked to ignore areport of British victory believed that evenwithout the report they would have per-ceived the relevance of Datum No. 11; thosetold to ignore a report of Gurka victory be-lieved that they in foresight would haveseen its irrelevance. When the relevancejudgment ANOVA of Experiment 1 is re-peated on the present data, this dependenceis reflected in highly significant (p < .0005)Outcome Reported X Datum Evaluated in-teractions.

Interestingly, for 128 of the 184 individualdatum units evaluated by subjects in thefour outcome groups of the four events Afterand After(ignore) relevance judgments wereeither both higher or both lower than thecorresponding Before judgments (as was thecase in the example, Datum No. 11, givenabove) (#=5.23; sign test). There wasno tendency for After and After (ignore)relevance judgments to be consistentlyhigher or lower than Before relevance judg-ments, which might in itself account forthis result.

Discussion

Experiment 1 showed that receipt of out-come knowledge affects subjects' judgments

in the direction predicted by the creepingdeterminism hypothesis. Experiment 2 hasshown that subjects are either unaware ofoutcome knowledge having an effect on theirperceptions or, if aware, they are unable toignore or rescind that effect. Both therelevance and the probability judgments ofAfter (ignore) subjects suggest that subjectsfail to properly reconstruct foresightful (Be-fore) judgments because they are "an-chored" in the hindsightful state of mindcreated by receipt of outcome knowledge.

It might be asked whether this failure toempathize with ourselves in a more ignorantstate is not paralleled by a failure to empa-thize with outcome-ignorant others. Howwell people manage to reconstruct the per-ceptions that others had before the occur-rence of some event is a crucial question forhistorians, and indeed for all human under-standing. The assumption that we clearlyperceive how others viewed situations be-fore receipt of outcome knowledge underliesmost second-guessing of their decisions. Ex-periment 3 examined this question.

EXPERIMENT 3

Method

Design. Subjects were presented with stimulusmaterials identical to those used in Experiments1 and 2. They were asked to respond as had otherstudent judges who had not known the true out-come. Before (others) subjects were not providedwith any outcome knowledge. After (others) sub-jects received versions of the stimulus events withone of the four possible outcomes presented asthe true outcome (what had actually happened).After(others) subjects' task was essentially to ig-nore outcome knowledge in order to respond likeBefore (others) subjects.

Instructions. The cover of each test bookletread:

Short descriptions of a number of real social andpersonal events appear below, each with severalpossible outcomes. These descriptions were pre-sented to students of social science in other uniTversities in Israel. (However, they were nottold which of the possible outcomes actuallyhappened.) We will ask you to guess the judg-ments of these students regarding the likeli-hood of possible outcomes. We thank you foryour participation.

The section in parentheses only appeared in the in-structions for After (others) subjects. Each pageof the test booklets was identical to the corre-


TABLE 4MEAN PROBABILITIES ASSIGNED BY SUBJECTS RESPONDING "As DID OTHER STUDENTS

WHO DID NOT KNOW WHAT HAPPENED," EXPERIMENT 3

Experimental groupOutcomeprovided

Outcome evaluated

Event A: British-Gurka struggle


2117172217

none 26.4 24.S1 39.42 18.83 31.14 28.2

22.442.621.221.9

29.5 19.520.320.326.623.7

18.820.020.026.2

Event B: Near-riot in Atlanta


2017181916

none 11.0 24.0 41.8 23.21 15.0 24.72 13.2 36.03 4.8 22.54 12.3 26.4

36.535.251.138.4

23.814.621.622.8



2118181619

none 19.6 15.9 24.0 40.51 20.3 20.0 28.32 31.9 23.3 14.83 30.6 12.5 26.94 12.5 20.4 22.6

31.430.030.144.4



1915161716

none 30.7 22.4 39.2 7.81 46.0 15.3 30.0 8.72 22.5 36.6 34.1 6.93 19.8 14.8 57.7 7.84 23.5 18.3 40.3 17.8

Note, Outlined cells are those with After (ignore) probabilities of reported outcomes.

spending page of the Experiment I test booklets,except for the addition of a reminder, "Answer asyou think other students (who did not know whathappened) answered" before each response section.

Subjects, Ninety-four members of an inter-mediate statistics class at the University of theNegev participated.

Results

Probability estimates. Table 4 presentsmean probability assignments by subjects ineach group. After (others) subjects' in-ability to ignore the effects of creeping de-terminism is clearly evident. For 14 of the16 reported outcomes (p < .002; sign test),they attributed higher probabilities to out-

come-ignorant others than did Before (oth-ers) subjects. As in Experiment 2, beingtold to ignore outcome knowledge slightly,but not significantly (p > .10; Mann-Whit-ney U test), reduced its impact. The meanBefore(others)-After(others) difference was8.7% compared with the mean Before-Afterdifference of 10.8% in Experiment 1.

Relevance judgments. After (ignore) sub-jects who had received different outcome re-ports attributed markedly different relevancejudgments to the outcome-ignorant others.The dependence of the relevance judgmentsthat they attributed on the outcome knowl-edge they were to ignore produced signifi-


cant (p < .01) Outcome Reported X DatumEvaluated interactions for each of the fourevents. Thus, After (ignore) subjects ex-pected other subjects to have seen in fore-sight patterns of data relevance that theythemselves only saw in hindsight.

Projection. Comparing Tables 1 and 4and Tables 3 and 4, it is apparent that theentries in corresponding Before and Before-(others) cells are quite similar, as are cor-responding After (others) and After(ignore)cells. The mean absolute difference betweenentries in corresponding cells is 3.5% for thefirst comparison, 5.1% for the latter. Thissuggests that when asked to respond likesimilar others, subjects respond as they be-lieve they themselves would have respondedin similar circumstances (i.e., by projec-tion). Both the probability and relevancejudgments of After (others) subjects moreclosely resembled those of After (ignore)and After subjects than those of Before(oth-ers) subjects.

Reasons. Some 87% of the subjects pro-vided reasons for their judgments. Al-though content analysis of these reasonsproved intractable, one interesting finding isthat After (others) subjects offered consist-ently more reasons than Before subjects (p<.05; median test). In Experiment 1,After subjects reported relying more on out-side information (as compared with the text)than did Before subjects. Perhaps in bothcases knowing what happened facilitatesknowing where to look for and what to ac-cept as reasons.

GENERAL DISCUSSION

Finding out that an outcome has occurredincreases its perceived likelihood. Judges are,however, unaware of the effect that outcomeknowledge has on their perceptions. Thus,judges tend to believe that this relative in-evitability was largely apparent in foresight,without the benefit of knowing what hap-pened.

In a fourth study (Fischhoff & Beyth,1975), subjects were asked on the eve offormer President Nixon's trips to Chinaand the USSR (in early 1972) to estimatethe probability of various possible outcomes

of the visits (e.g., Nixon's meeting Chair-man Mao, visiting Lenin's tomb, or an-nouncing that the trips were successful).From 2 wk to 6 mo after the trips' comple-tion, these same subjects were asked to re-member as best they could their own origi-nal predictions. They were also asked toindicate for each event whether or not theybelieved that it had actually happened.

Results showed that subjects rememberedhaving given higher probabilities than theyactually had to events believed to have oc-curred and lower probabilities to events thathadn't occurred. Their original predictionsshowed considerable overestimation of lowprobabilities, that is, too many events thatthey judged to be extremely unlikely or im-possible did occur. The probability judg-ments that they remembered, however, con-sistently underestimated low probabilities.Indeed, almost no events to which they re-membered assigning low probabilities wereperceived to have occurred.

Thus, undiagnosed creeping determinismnot only biases people's impressions of whatthey would have known without outcomeknowledge, but also their impressions ofwhat they themselves and others actually didknow in foresight.

Explanations

The simplest hypothesis regarding themanner in which judges process outcomeknowledge suffices to account for these re-sults. Assume that upon receipt of outcomeknowledge judges immediately assimilate itwith what they already know about the eventin question. In other words, the retrospec-tive judge attempts to make sense, or a co-herent whole, out of all that he knows aboutthe event. The changes in relevance judg-ments could reflect such assimilative mean-ing adjustment.

Assimilation of this type would tend toinduce creeping determinism for judgesusing any of the techniques for producingsubjective probability estimates appearing inTversky and Kahneman's (1974) com-pendium. Judges using the heuristic of"representativeness" perceive outcomes aslikely when they match or represent the


dominant features of the situation thatproduced them. Assimilation of outcomeknowledge should certainly increase the per-ceived "fit" between the reported outcomeand the situation that preceded it. A secondheuristic leads judges to evaluate an out-come's likelihood by the relative "avail-ability" of scenarios leading to its occur-rence and nonoccurrence. The judge whoknows what happened, and has adjusted hisperceptions in the light of that knowledge,may well find it difficult to imagine howthings could have turned out otherwise.

An alternative mode of explanation focuseson structural differences between the tasksof hindsight and foresight. Judges possessingoutcome knowledge may, for example, tendto reverse their temporal perspective andproduce scenarios that proceed backwardin time, from the outcome to the precedingsituation. Such scenario retrodiction mayeffectively obscure the ways in which eventsmight not have taken place, much as solvinga maze backward can obscure the ways inwhich one might have gotten lost enteringfrom the beginning.

Receipt of outcome knowledge may alsorestructure the task of judges using the"anchoring and adjustment" heuristic (Tver-sky & Kahneman, 1974). Judges may esti-mate the likelihood of a reported outcome byinitially assigning it 100%, the most salientpossible value, and then looking for reasonsto adjust downward from there. Adjust-ment from initial values is typically inade-quate and would produce creeping determ-inism with this task.

Of these explanations, those based onassimilation most readily account for theunderestimation of creeping determinismfound in Experiments 2 and 3 and in theFischhoff and Beyth (1975) study. Makingsense out of what one is told about the pastseems so natural and effortless a responsethat one may be unaware that outcomeknowledge has had any effect at all on him.Judges who are aware that outcome knowl-edge has affected their perceptions still facethe unenviable task of reconstructing theirforesightful state of mind. "Undiagnosedcreeping determinism" would characterizethe responses of subjects who, in reconstruc-

tion, were unable to adequately unanchorthemselves from the perspective of hind-sight.

Implications

In the short run, failure to ignore out-come knowledge holds substantial benefits.It is quite flattering to believe, or lead othersto believe, that we would have known allalong what we could only know with out-come knowledge, that is, that we possesshindsightful foresight. In the long run, how-ever, unperceived creeping determinism canseriously impair our ability to judge thepast or learn from it.

Consider a decision maker who has beencaught unprepared by some turn of eventsand who tries to see where he went wrongby recreating his preoutcome knowledgestate of mind. If, in retrospect, the eventappears to have seemed relatively likely, hecan do little more than berate himself fornot taking the action which his knowledgeseems to have dictated. He might be saidto add the insult of regret to the injuryinflicted by the event itself. When second-guessed by a hindsightful observer, his mis-fortune appears to have been incompetence,folly, or worse.

In situations where information is limited,and indeterminate, occasional surprises-—•and resulting failures—are inevitable. Itis both unfair and self-defeating to castigatedecision makers who have erred in falliblesystems, without admitting to that fallibilityand doing something to improve the system.According to historian Roberta Wohlstetter(1962), the lesson to be learned from Amer-ican surprise at Pearl Harbor is that wemust "accept the fact of uncertainty andlearn to live with it. Since no magic willprovide certainty, our plans must workwithout it" (p. 401).

When we attempt to understand pastevents, we implicitly test the hypotheses orrules we use to both interpret and anticipatethe world around us. If, in hindsight, wesystematically underestimate the surpriseswhich the past held and holds for us, we aresubjecting those hypotheses to inordinatelyweak tests and, presumably, finding littlereason to change them. Thus, the very out-

HINDSIGHT 96 FORESIGHT 299

come knowledge which gives us the feelingthat we understand what the past was allabout may prevent us from learning anythingfrom it.3

3 Elaboration on this point as well as specula-tion on how hindsight can be improved may befound'in Fischhoff (1974) and Fischhoff and Beyth(1974).

REFERENCES

Ellis, A. Psychosexual and marital problems. InL. A. Berg & L. A. Pennington (Eds.), Anintroduction to clinical psychology. New York:Ronald, 1966.

Fischhoff, B. Hindsight: Thinking backward?Oregon Research Institute Research Monograph,1974,14, No. 1.

Fischhoff, B. Temporal setting and judgmentunder uncertainty. Organisational Behaviorand Human Performance, in press.

Fischhoff/ B., & Beyth, R, Failure has manyfathers (Review of Victims of groupthlnk byI. Janis). Oregon Research Institute ResearchBulletin, 1974, 14, No. 17.

Fischhoff, B., & Beyth, R. "I knew it would hap-pen"—remembered probabilities of once-future

things. Organisational Behavior and HumanPerformance, 1975, 13, 1-16.

Florovsky, G. The study of the past. In R. H.Nash (Ed.), Ideas of history (Vol. 2). NewYork: Button, 1969.

Lofland, R. Deviance and identity. EnglewoodCliffs, N.J.: Prentice-Hall, 1969.

National Advisory Commission. Kerner Com-mission report on civil disorders. New York:Bantam, 1968.

Rosenhan, D. On being sane in insane places.Science, 1973, 79, 250-252.

Schur, E. Labelling deviant behavior. New York:Harper & Row, 1971.

Tawney, R. H. The agrarian problem in thesixteenth century. New York: Franklin, 1961.

Tversky, A., & Kahneman, D. Belief in the lawof small numbers. Psychological Bulletin, 1971,76, 105-110.

Tversky, A., & Kahneman, D. Judgment underuncertainty: Heuristics and biases. Science, 1974,185, 1124-1131.

Walster, E. "Second guessing" important events.Human Relations, 1967, 20, 239-250.

Wohlstetter, R. Pearl Harbor: Warning anddecision. Stanford, Calif.: Stanford UniversityPress, 1962.

Woodward, E. L. Age of reform. London:Oxford University Press, 1938.

(Received January 6, 1975)

hindsight ^foresight: the effect of outcome …hindsight *£ foresight 289 numbers," the belief...

Documents