ArticlePDF Available

Back to the future: Temporal perspective in the explanation of events

Authors:

Abstract

Prospective hindsight involves generating an explanation for a future event as if it had already happened; i.e., one goes forward in time, and then looks back. In order to examine how shifts in perspective might influence people's perceptions of events, we investigated two possible factors: temporal perspective (whether an event is set in the future or past) and uncertainty (whether the event's occurrence is certain or uncertain). In the first experiment, temporal perspective showed little influence while outcome uncertainty strongly affected the nature of explanations for events. Explanations for sure events tended to be longer, to contain a higher proportion of episodic reasons, and to be expressed in past tense. Evidence from the second experiment supports the view that uncertainty mediates not the amount of time spent explaining, but rather subjects' choice of explanation type. The implications of these findings for the use of temporal perspective in decision aiding are discussed.
Journal of Behavioral Decision Making. Vol. 2, 25-38 (1989)
Back
to the
Future: Temporal Perspective
in the
Explanation
of
Events
DEBORAH
J.
MITCHELL
The Wharton School,
University
of
Pennsylvania,
U.S.A.
J.
EDWARD RUSSO
Cornell
University,
U.S.A.
NANCY PENNINGTON
University
of
Colorado,
U.S.A.
ABSTRACT
Prospective hindsight iiivolves generating
an
explanation
for a
future event
as if it
had already happened; i.e., one goes forward
in
time,
and
then looks back.
In
order
to examine how shifts
in
perspective might influence people's perceptions
of
events,
we investigated
two
possible factors: temporal perspective (whether
an
event
is set
in
the
future
or
past)
and
uncertainty (whether
the
event's occurrence
is
certain
or
uncertain).
In the
first experiment, temporal perspective showed little influence
while outcome uncertainty strongly affected
the
nature
of
explanations
for
events.
Explanations
for
sure events tended
to be
longer,
to
contain
a
higher proportion
of
episodic reasons,
and to be
expressed
in
past tense. Evidence from
the
second
experiment supports
the
view that uncertainty mediates
not the
amount
of
time
spent explaining,
but
rather subjects' choice
of
explanation type.
The
implications
of these findings
for the use of
temporal perspective
in
decision aiding
are
discussed.
KEY
WORDS Causal explanation Prospective hindsight Scenarios
Temporal perspective Uncertainty
People make sense of their environment by observing what happens and then reasoning about why
things happen as they do. Using such causal reasoning, individuals act as amateur scientists who seek to
explain what they perceive (Ross, 1977). There is great variation in the nature of these explanations.
Much depends on what observers know of an event's circumstances, their explanatory skills, and how
complete an explanation is desired.
In many situations substantial inference is required. These inferences about an event are crucially
affected by its context. An event can be defined in terms of who is involved, what factors or actions are
relevant, where it takes place, and when a factor we call temporal perspective. Events may vary in
temporal relation to observers, specifically in the difference between the time of a target event and the
Addressee
for
correspondence: Deborah
J.
Mitchell, Department
of
Marketing,
The
Wharton School, University
of
Pennsylvania, 1450 Steinberg Hall-Dietrich Hall, Philadelphia,
PA
19104-6371, U.S.A.
0894-3257/89/0l0025-14$07.00
Received 24 Januuary 1988
© 1989
by
John Wiley
&
Sons,
Ltd.
Revised 28 July 1988
26 Journal of Behavioral Decision Making Vol. 2 Iss. No. 1
present. When people consider an event that has not yet occurred, they adopt a forward perspective. If
they look back in time to a concluded event, they adopt a backward perspective.
It is natural that past and future events are usually treated differently. Normally people explain the
past and predict the future. Looking back at events that have already occurred, the critical question is
"why.' Looking forward into the future, the question is more often 'what (will happen).' The two
questions are related. One generally hopes to learn from explaining the past and to apply this
knowledge to predicting the future. It is clear that the better one can anticipate future events, the more
successful decisions and actions can be.
Although only the backward perspective is normally associated with explanation, suppose that
people shifted their perspective on future events to think about them as if they had already occurred.
Why might they do this? People can make better decisions about a future course of action if they fully
understand all that is required to make it happen, including the perhaps-not-so-obvious. Weick (1979),
Hogarth (1983) and others have suggested that traveling forward in time to look backward on a future
event might improve decision making, by helping people see more of these necessary ingredients.
This
belief,
though never directly tested, stems in part from studies showing that temporal perspective
can influence the way people perceive and evaluate events. How subjects described past and future
events differed in terms of amout of detail, causal complexity (as measured by description length), and
level of abstraction (Bavelas, 1973; Sevon, 1984). In these studies, subjects described past events in more
details and at greater length than they did future events.
Explaining future events as past (i.e., prospective hindsight) might well improve decision making if
people do 'see' more when they look backward. Given this potential value, prospective hindsight
deserves further investigation. However, to understand this phenomenon it is important to control for a
natural confounding factor, uncertainty of outcome.
Uncertainty and explanation
In previous temporal perspective studies, future events were always presented to subjects as uncertain,
and past events always had sure outcomes. For instance, Sevon (1984) concluded that decision makers'
conceptions of past events were more causally complex then their conceptions of identical future events.
However, her experimental task compared explanations for Sweden's past high inflation/unemploy-
ment with explanations for possible but uncertain high inflation/unemployment in a future period.
Although such a comparison legitimately reflects real world conditions, it is not possible to know
whether the observed effects were due to differences in temporal perspective or to the accompanying
differences in uncertainty. Other studies also suggest that uncertainty contributes to the inferences
people make about events. For example, in the 'sealed-fate effect' subjects bet less when dice are thrown
but not yet revealed, than when dice are thrown after bets are placed (Strickland, Lewicki, and Katz,
1966;
Rothbart and Snyder, 1970).
This leads us to ask: What is the role of uncertainty in accounting for the differences in explaining
events, that have heretofore been attributed to temporal perspective? The most extreme hypothesis is
that the entire effect of temporal perspective is due to uncertainty. It is uncertainty that is 'driving' the
observed differences in explanation, and the natural confounding between these two contextual factors
has led to a misattribution of these effects to temporal perspective. At the least, it would be useful to
disentangle the two factors and assess the independent impact of each.
If certainty of outcome does influence how subjects perform an explanation task, how might this
happen? Subjects might work at explaining an event either longer (an effort hypothesis) or differently (a
processing hypothesis). The effort hypothesis presumes that subjects devote more time and, thereby,
explain more thoroughly a sure event than an uncertain one. Possibly people feel more obligated to
explain a known outcome than to waste time constructing an explanation for an outcome that may or
may not occur. In contrast, the processing hypothesis asserts that uncertainty influences the cognitive
D.
J. Mitchell, J. Edward Russo and N. Pennington Back to the Future 27
strategies subjects use to explain events. Subjects may view explanation under uncertain conditions as a
different kind of task needing a different kind of explanation than under certainty. Or different kinds of
explanations may bave different ecological values in uncertain and sure task environments. Tbe effort
bypotbesis predicts tbat people work longer, tbe processing bypotbesis claims tbat tbey work
differently.
Experimental rationale
Our initial interest was to examine and test tbe claims offered by prospective bindsigbt as a decision aid.
Can it belp people 'see' more? Paramount in understanding and evaluating any potential benefits of
sbifts in temporal perspective are assessing wbetber particular context variables influence explanation,
bow tbey migbt do so, and tbe specific nature of tbeir effects.
Our primary goal is to disentangle tbe separate effects of temporal perspective and outcome
uncertainty on tbe explanation of events. In our experiments temporal perspective is manipulated by
asking subjects to explain events tbat differ in wbetber tbey occur in tbe past or tbe future. Uncertainty
is independently manipulated by specific instructions about tbe level of certainty to be assumed.
Second, we wisb to determine wbetber systematic differences in explanations occur because subjects
work longer or tbink differently. Explanations of future and past events migbt differ in lengtb alone or
in more qualitative ways (abstract vs. concrete terms; past, present, or future tense). If different contexts
produce only differences in time invested, explanations are more likely to vary in lengtb tben quality,
wbereas if different event contexts elicit different cognitive processes, qualitative as well as quantitative
differences are likely.
In tbe two experiments reported bere, subjects were asked to explain tbe occurrence of an outcome
event in terms of underlying causes. Eacb sucb event was tbe consequence of some activity and could be
classified as eitber a success or failure. For example, playing tennis could result in winning or losing and
a sales presentation could result in a sale or no sale. Eacb event was set in eitber tbe past or tbe future
and its occurrence was depicted as eitber sure or uncertain. Tbese events provided a ricb causal
environment for explanation, permitting variability in tbe lengtb and content of subjects' explanations.
Tbe aspects of subjects' explanations tbat were analyzed included lengtb, content, and verb tense.
Because lengtb (or complexity) bas often been measured by tbe number of causes or attributions
provided for an event (Ward, Hastie, and Taylor, 1978; Sevon, 1984), we asked subjects to list as many
reasons as tbey could for eacb outcome. For content of explanations we developed a measure based on
differences in level of abstraction. Tbis content measure was derived from tbe results of pilot work and
current conceptions of bow events and knowledge about events are mentally represented (e.g., Abbott,
Black and Smitb, 1985; Abelson, 1981; Bower, Black and Turner, 1979; Galambos and Rips, 1982). In
pilot work we observed tbat subjects' explanations differed as to wbetber tbey explained an event in
terms of actions consistent with the outcome (a style we refer to as episodic) or in terms of a listing of
factors contributing to tbe outcome (a style we call abstract).
Episodic or action-based explanation is cbaracterized by listing of actions. Tbese are often stereotypic
and consistent witb a specific instantiation of a script (Abelson, 1981), altbougb temporal ordering of
events need not be preserved. Examples of episodic reasons for a party's success are:
Sbe greeted tbe guests witb a smile.
Everyone will dance until tbe wee bours.
Tbe food was served on time.
Tbe band will play great music.
Abstract or factor-based explanations contain lists of underlying factors, like fixed parameters or
enduring propositions. Sucb factors are similar to tbe devices tbat imbue scripts witb predictive
generality. Examples of abstract reasons for a party's success are:
28 Journal of Behavioral Decision Making Vol. 2 Iss. No. 1
She is friendly.
The guests are reasonable people.
She is a good cook.
Music is an excellent mood-setter.
Reasons were also coded and analyzed by their verb tense. Tense and content are partially con-
founded, due to linguistic differences in the expression of abstract and episodic reasons. Because
abstract reasons often concern enduring propositions that remain true over time, they are stated in the
neutral present tense. Episodic reasons represent specific instances or actions and occur at some point in
time;
that point may be expressed in the past, present, or future tense. Fischhoff (1976a) found that
subjects often (mistakenly) used future tense when describing past events of which they are uncertain,
responding as if they are predicting future events. We analyzed the tense of each reason to discover
whether subjects systematically alter tense according to temporal conditions.
A measure of effort was obtained (in the second experiment only) by recording the length of time
subjects spent constructing their explanations. In addition, subjects were asked to rate the difficulty of
the explanation task.
Using the dependent measures described above, we sought to determine whether previous research on
the effects of temporal perspective would be replicated when temporal perspective was not confounded
with certainty. In this regard the main issue centered on whether subjects generate more reasons for past
events than for future ones, regardless of the certainty of the event's outcome. If qualitative differences
in explanations also appear, such differences might provide information regarding whether subjects
work longer, differently, or both, and would be useful in addressing the issue oi quality of explanation.
EXPERIMENT ONE
Method
Materials
Four scenarios were developed, depicting two professional and two personal situations. Eight versions
of each scenario varied only in outcome: either future or past, e.g., an executive will complete a report
by the deadline versus the executive completed the report by the deadline; sure or uncertain, e.g., the
executive completed the report by the deadline or may have completed it; and a success or a failure, e.g.,
he completed the report by the deadline versus he did not. Appendix A presents one scenario with its
eight possible outcomes.
Temporal perspective and certainty level were between-subject factors; the four scenarios and valence
of outcome were within-subject factors. The experimental materials were distributed as booklets. Each
booklet contained four scenarios, two positive and two negative, all with the same level of certainty and
temporal perspective.
Subjects
Subjects were 114 unpaid University of Chicago MBA students enrolled in an introductory decision
making course. Random assignment to the four experimental conditions (past certain, past uncertain,
future certain, future uncertain) yielded approximately 28 subjects in each.
Procedure
Subjects were told that they would be given information about situations and then asked to answer
questions about the material. For each of the four scenarios subjects began with the background
information, and then generated as many plausible reasons as they could think of for the outcome in
D.
J. Mitchell, J. Edward Russo and N. Pennington Back to the Future 29
question. Finally, they provided a subjective probability for the outcome's occurrence. This served as a
manipulation check to establish that subjective uncertainty corresponded to the intended uncertainty
level. Subjects spent as much time as they wished, up to thirty minutes in total on the four scenarios.
Times spent working were not recorded.
Results
Subjective probabilities
,The subjective probabilities were used to check that subjects assumed the intended level of certainty.
Within the two sure conditions, only two subjects assumed less than absolute certainty; they were
dropped from further analysis. All subjects in the uncertain conditions assumed less than absolute
certainty. The mean subjective probability across uncertain conditions was .54. Within the two
uncertain conditions, the future subjects assumed greater uncertainty (M = .50) than did the past
subjects {M
=
.57). This difference is marginally reliable, t
=
1.28,/»<
.10.
Number of reasons
Our measure of length of subjects' explanations was the number of different reasons listed for a
particular scenario. In cases where reasons were either numbered or presented in an ordered list,
counting reasons was straightforward. In cases where the explanation was expressed as a paragraph,
each complete sentence was counted as a reason. (There were very few such cases, and in all of them
each individual sentence was judged to convey a separate idea.) Numbers of reasons were analyzed in a
2 (future/past) x 2 (certain/uncertain) x 2 (professional/personal scenario type) x 2 (positive/negative
outcome) mixed-model ANOVA. Exhibit
1
displays the means for temporal perspective by certainty.
There was no reliable difference in the number of reasons generated for future and past events,
F(l,108) = 1.49; but subjects did generate more reasons for sure outcomes than for uncertain ones,
/^l,108) = 19.79, p < .001. There was a marginally significant certainty x temporal perspective
interaction, /(1,108) = 3.04, p < .09. This was manifested as a negligible difference in the number of
reasons generated for future sure and past sure events, but a more substantial difference in the number
for future and past uncertain ones. There was no reliable difference in the number of reasons generated
for positive and negative outcomes, or for professional and personal scenarios (both Fs less than 1).
Past
Future
Combined
Past
Future
Combined
Experiment 1
Certain
4.43
4.57
4.50
Uncertain
3.72
2.94
3.30
Experiment 2
Certain
5.17
4.41
4.79
Uncertain
3.06
3.38
3.23
Combined
4.08
3.76
Combined
4.12
3.90
Exhibit 1. Mean number of reasons generated per
scenario by level of certainty and temporal
persective
30Journal of Behavioral Decision Making
Experiment 1
Vol. 2 Iss. No. 1
Past
Future
Combined
Past
Future
Combined
Certain
.51
.49
.50
Uncertain
.31
.23
.27
Experiment 2
Certain
.71
.75
.73
Uncertain
.31
.42
.37
Combined
.41
.36
Combined
.57
.59
Exhibit 2. Mean proportion of episodic reasons by
level of certainty and temporal perspective
Type of reasons
Each generated reason was coded as episodic (action-based) or abstract (factor-based). For each
scenario, we computed the proportion of reasons that were episodic. This measure was subjected to an
arcsin transform and then analyzed in the same 2 x 2x 2 x 2 mixed-model ANOVA. Exhibit 2 displays
the relevant proportions.
Again, there was no reliable difference between past and future outcomes, but there was a significant
effect of certainty. More episodic reasons were generated for sure outcomes (.50) than for uncertain
outcomes (.27), fl[l,108)
=
21.41,p <
.001.
A greater proportion of episodic reasons was also generated
for personal scenarios (.41) than for professional ones (.36), /=l[l,108) = 5.73, p < .02. There was no
difference for positive and negative outcomes (Fless than 1).
Tense of reasons
A final measure was the linguistic tense of each reason. A crossover in tense was defined to be either a
reason in the past tense for a future outcome, or a future tense reason for a past outcome. The
proportion of tense crossovers was negligible in all but the sure future condition, where a third of all
reasons generated was expressed in the past tense (see Exhibit
3).
In general, subjects showed a tendency
to use past tense when generating reasons for sure outcomes, regardless of the temporal perspective of
the outcomes.
Experiment 1 Experiment 2
Certain Uncertain Certain Uncertain
Past .00 .00 .00 .00
Future .33 .02 .38 .02
Exhibit 3. Mean proportion of total reasons that are tense
crossovers
D.
J. Mitchell, J. Edward Russo and N. Pennington Back to the Future 31
Discussion
In Experiment 1, subjects reacted indifferently to changes in temporal perspective but exhibited sub-
stantial effects of certainty-of-outcome. Sure outcomes yielded more reasons, and a higher proportion
of episodic reasons. In addition, subjects frequently expressed reasons for such outcomes in the past
tense, even when the outcomes were depicted as future. These results indicate that previous findings
attributed to temporal perspective may be due almost solely to the level of certainty. In other words, in
the natural confounding of certainty and temporal perspective in the real world, it is certainty that is the
more fundamental factor and that is driving the observed effects on explanation.
In addition, the results of Experiment
1
are consistent with a cognitive processing hypothesis, the idea
that subjects may think differently about sure events than about uncertain ones. The explanations
generated for these events differed in content as well as length, even though the information available to
the subject (the outcome's antecedents) was held constant. There is no a priori reason for this to occur.
In our task the set of possible explanations for an outcome should not change due to level of certainty
about the outcome. The reasons for the outcome are equally uncertain in both sure and uncertain cases.
If subjects used the same strategy to explain both types of outcomes, but spent more time generating
reasons for sure ones (the effort hypothesis), only differences in the length of explanations would be
seen. However, our findings show both length and qualitative differences.
EXPERIMENT TWO
Experiment 1 supports the cognitive processing hypothesis that subjects think differently about sure
and uncertain events. To further investigate whether subjects work differently or merely longer when
explaining sure outcomes, we conducted a second experiment in which we collected two additional
measures. First, time spent working on the entire exercise was recorded for a subset of subjects. Second,
for each scenario all subjects rated the difficulty of generating an explanation on a seven-point scale (7
signifying very difficult). This measure reveals whether subjects view the task of explaining sure and
uncertain outcomes as unequal.
Experiment 2 also attempts to replicate the findings of Experiment 1, while removing a possible
artifact. The source of outcome information in Experiment 1 differed between sure and uncertain
conditions. In uncertain conditions outcomes were suggested by scenario actors, while in sure
conditions subjects were told by the experimenter that an outcome was sure. Thus, the certainty effect
found in Experiment
1
could have been a source effect. To control for this, all outcomes in Experiment
2 were suggested by the experimenter (see Appendix B for examples of this rephrasing).
An additional change was suggested by the work of Koriat, Lichtenstein, and Fischhoff (1980) and
Hoch (1984) which shows that subjective probabilities reported after reason generation may be biased
by differential availability of
evidence.
Thus, the subjective probabilities reported in Experiment
1
may
not be representative of subjects' initial levels of uncertainty before they generated reasons. To remove
this potential artifact, subjects in Experiment 2 provided subjective probabilities before they generated
reasons instead of afterward.
Method
Materials
Two scenarios similar to those of Experiment 1 were used. However, all outcomes were suggested
'anonymously'; unlike the stimuli used in Experiment 1, no outcomes were depicted as being suggested
by scenario characters. As before, eight versions of each scenario were written in which only the
32 Journal of Behavioral Decision Making Vol. 2 Iss. No. 1
outcome statement was varied. Outcomes were either future or past, sure or uncertain, professional or
personal, and positive or negative, using the same design as Experiment 1. One scenario with its eight
possible outcomes is presented in Appendix B.
Subjects
Subjects were composed of two groups. The first group consisted of 76 University of Chicago MBA
students enrolled in an introductory marketing course, who performed the experimental task as part of
an in-class assignment. The second group was composed of 32 respondents to an ad for paid
participants. This groups consisted of undergraduate and graduate University of Chicago students, who
performed the task individually. It was this latter group for whom we unobtrusively recorded the time
spent on the entire task. Subjects in both groups were randomly assigned to the four experimental
conditions, yielding a total of approximately 27 subjects in each.
Procedure
The experimental procedure was identical to Experiment
1
except for the following. First, subjects read
and responded to two scenarios instead of four. Second, subjective probabilities were assessed before
generation of reasons instead of afterward. Lastly, subjects rated the difficulty of explanation, on a
seven-point scale, each time they completed an explanation.
Results
An initial analysis of the paid and unpaid subjects' data showed no significant differences on any
dependent measure. Consequently, all analyses are reported for the combined group. The one exception
is the time measure, which was collected only for the 32 paid subjects.
Subjective probabilities
Within the two sure conditions, only one subject assumed less than absolute certainty; that subjects was
dropped from further analysis. Subjects in uncertain conditions assigned an average probability of .46;
there was no reliable difference between the two uncertain conditions in the mean subjective probability
assigned (M
=
.46 for the past condition and M
=
.45 for the future).
Number of reasons
This measure was analyzed as before. The results replicated Experiment 1 (see Exhibit 1). Subjects
generated more reasons for sure outcomes than for uncertain ones, f{l,99)
=
19.92,p <
.001;
and there
was no reliable difference between past and future outcomes (F less than 1). Additionally, the
marginally significant certainty x temporal perspective interaction of Experiment 1 disappeared. Thus,
temporal perspective showed no influence at all on the number of reasons generated. One finding which
differed between experiments was the effect of scenario type: a greater number of reasons was generated
for outcomes related to the personal scenario than for the professional one {M - 4.35 and M
=
3.68,
respectively), F{1,99)
=
13.71,
p <
.001.
There was no reliable difference between positive and negative
outcomes (F less than 1).
Type of reasons
Content of reasons, episodic versus abstract, was analyzed as in Experiment 1. The general pattern of
results in Experiment
1
was repeated (see Exhibit 2). The proportion of episodic reasons was greater for
sure outcomes (.73) than for uncertain ones (.37), f{l,99)
=
26.63,p <
.001.
No other reliable differences
were found (all Fs less than 1).
D.
J. Mitchell, J. Edward Russo and N. Pennington Back to the Future 33
Tense of reasons
Again subjects showed a tendency to use the past tense when generating reasons for sure outcomes,
regardless of the temporal perspective of these outcomes (see Exhibit 3). The proportion of tense
crossovers was negligible in all but the sure future condition, where it was .38.
Difficulty of explanation
Subjects found if more difficult to generate explanations for uncertain outcomes than for sure ones, (M
= 3.79, and M - 3.00, respectively), F{\,99) - 13.49,p < .001. No other reliable differences in difficulty
were found {Fs less than 1).
Time spent working
No reliable differences were found in time spent working across the experimental conditions (Fs less
than 1). Notably, M - 18.3 minutes and M
=
20.9 minutes for the certain and uncertain conditions
respectively. The direction of this difference is contrary to the prediction of the effort hypothesis.
Discussion
Experiment 2 verified the results of Experiment 1, removing any reservations due to artifacts. Indeed,
any reliable effect of temporal perspective completely disappeared, while the effect of certainty was
somewhat larger.
Experiment 2 also supports the hypothesis that subjects work differently, rather than just longer,
when explaining sure outcomes. We found no differences in the amount of time subjects spent explain-
ing, yet 4.60 reasons were generated per sure outcome and only 3.28 per uncertain one. This suggests
that different, more efficient reason generation strategies were used when the outcomes of an event was
sure.
At least this is the case when the task is to generate as many reasons as possible ignoring their
content. A difference in the efficiency of sure and uncertain strategies is supported by subjects reporting
greater difficulty explaining uncertain outcomes. In sum, the results of Experiment 2 point to different
strategies being employed as certainty changes, not the same strategies being pursued longer.
Strategies for explaining sure versus uncertain outcomes
This finding naturally leads to futher questions about what specific processing strategies might be
adopted in sure and uncertain environments. Because our experiments were not designed to identify the
strategies used, the hypotheses we can offer are necessarily/»oi/ hoc. In considering plausible processing
hypotheses, we assume several things about the information available to subjects and the strategies they
adopted.
First, we assume that all subjects have conceptual knowledge concering an event, including standard
roles played by people (say, the host and guests at a party), typical props, objects and locations involved
(e.g., a party's location, refreshments and activities), a standard set of scenes and actions (planning the
party, inviting the guests, greeting them, eating, participating in activities, leaving, cleaning up, etc.),
and some normal consequences of a successful (or unsuccessful) completion of the event. Presuming
this knowledge seems reasonable for the common 'scripted'events used in our experiments. Second, we
assume that all subjects have particular episodic knowledge from instances of the described events
personally experienced or learned about. Third, we assume that subjects used information presented in
the stimulus scenario in combination with conceptual knowledge and specific experience to construct a
mental representation of the particular event and outcome.
Given these assumptions, we offer three possible cognitive strategies for constructing explanations.
They correspond to the locus of the difference between the way sure and uncertain events are tested:
different knowledge retrieved, hierarchially 'deeper' use of knowledge, or different inference processes.
34 Journal of Behavioral Decision Making Vol. 2 Iss. No. 1
1.
Retrieving different knowledge, abstract versus episodic
For sure outcomes subjects may retrieve episodic instances from personal experience and use them to
construct event-based reasons. When an outcome is uncertain, conceptual knowledge regarding which
key variables influence outcomes for the generic event under consideration may be retrieved instead.
Here an explanation would consist of a list of factors thought to affect an outcome's likelihood.
2.
Processing retrieved knowledge to a deeper, episodic level
According to this hypothesis, for both sure and uncertain outcomes the knowledge domain activated is
the same: information hierarchially organized with abstract variable information at a higher level and
episodic action-based information at a lower level (Abelson 1976). Explanations for sure outcomes may
involve a deeper level of processing within the knowledge hierarchy.
3.
Reporting retrieved knowledge or engaging in inference
This final hypothesis suggests that the same knowledge is retrieved for both sure and uncertain out-
comes, specifically abstract factor-based knowledge. For uncertain outcomes, key factors may be
selected and listed without further inference. In the case of sure outcomes, such information may be
used to construct a mental model of the situation, enabling simulations to be run (Holland, Holyoak,
Nisbett, and Thagard, 1986). Here, events seen to lead to the outcome in the simulations would
comprise an explanation for its occurrence.
Our data do not permit any conclusions about which of these classes of strategies is responsible for
the observed effects. That identification remains a task for future research.
GENERAL DISCUSSION
We set out to understand the role of temporal perspective in explaining events in two ways. First, we
tested whether the effects of temporal perspective are due to the level of outcome certainty which is
usually confounded with it. Second, we asked whether the observed differences in the number and kind
of reasons generated follow from different processing strategies or merely working longer in one
condition.
The answer to the first question is remarkably clear: most, if not all, of the effect of temporal
perspective is caused by the level of outcome certainty. Previous studies could not reveal this
mechanism because they preserved the natural confounding between temporal perspective and certainty
future outcomes are always uncertain, past are usually sure. The answer to the second question also
seems clear, though the evidence is not definitive. The number and type of reasons generated, the time
spent on the task and the rated difficulty all point to a strategy difference and not an effort difference.
Our data support the idea that cognitive differences affect explanation, but do not address why such
differences arise. The reason may be based in part on natural differences in the knowledge available for
the explanation. First, we note that people often refrain from thinking about events in abstract terms, if
they believe that concrete (event-based) reasoning is possible (Abelson, 1976; Nisbett, Borgida,
Crandall, and Reed, 1976). In day-to-day experience, the key abstract factors pertaining to an
outcome's occurrence are available for both past and future events. However, only for a past outcome
can one also draw on the specific, antecedent events that led to its occurrence. Such events are not
available, except through simulation, for uncertain future outcomes. Thus, in our natural ecology
reasoning about sure outcomes is most likely to be concrete, while reasoning about uncertain outcomes
is abstract.
As an example, consider predicting the winner of the first basketball game of a championship series.
Before the game predicting the winner is based on general factors: match-ups between key players, team
D.
J. Mitchell, J. Edward Russo and N. Pennington Back to the Future 35
strengths and weaknesses, etc. Even past games are used to identify factors relevant to the imminent
contest. After the game it's a different story. The defeat of one team is explained both by these general
factors and by specific events like Player A's early foul trouble Player B's 'off night', too much inactivity
since the team won its previous series, etc. Such events are too numerous to anticipate beforehand and
to relevant to ignore afterward.
Aiding decisions
Our initial interest in temporal perspective was prompted by claims made for the prospective hindsight
technique. Can people 'see' better by taking a backward perspective on a future event? The present
findings reveal that although prospective hindsight produces more reasons for an event, these reasons
are typically episodic in nature. This is 'seeing' more, which is not necessarily the same as 'seeing' better.
The value of a greater number of reasons must be qualified by such factors as the worth of an episodic
reason versus an abstract one. Since we cannot judge the value of each reason, it cannot be concluded
that prospective hindsight generates superior explanations. It depends on whether longer episodic
reasons for an event are more valuable than shorter abstract ones.
At the same time that our findings qualify the value of the prospective hindsight technique, they
highlight the potential benefit of manipulating outcome uncertainty. For instance, prospective
hindsight would seem to be valuable whenever one wishes to facilitate the construction of events chains
for an outcome. We can also create the natural complement to prospective hindsight, 'historical
foresight', which turns sure outcomes into uncertain ones. It might make sense to adopt historical
foresight in situations where the discovery and investigation of key underlying factors are desirable. For
example, learning from experience is based on the premise that history repeats
itself.
Unfortunately,
learning may be hampered because people tend to focus on specific details of antecedent events to
explain a past (sure) outcome.
If a new product fails, even though test markets had returned favorable results, the manufacturer
naturally wants to know what happened. The manufacturer would gain most from discovering the key
variables that infiuence such failures (e.g., a geographically limited appeal). Unfortunately, often the
producer will seek the specific actions that caused the product failure in this particular case (like the
Monday morning quarterback who concentrates mainly on Sunday's events). Using historical foresight,
i.e., generating abstract reasons for a past outcome, might be more suitable than letting episodic
reasoning dominate. This logic underlies the traditional case study approach, wherein students are
asked to analyze a past event without knowledge of its actual outcome. In this way, learning may be
enhanced.
Prospective hindsight may help in intuitive forecasting by utilizing people's natural facility to
generate event chains for sure outcomes. Decision makers can generate explanations to support many
alternative solutions or outcomes. In this way they can increase realistic awareness of uncertainty
present in a future situation, including the range of potential outcomes (Koriat et al., 1980; Hoch, 1984;
Hogarth, 1983; Weick, 1979). And the more concrete the explanation, the more likely the debiasing
effect (Fischhoff, 1980; Kahneman and Tversky, 1979).
Relatedly, the construction of scenarios (i.e., descriptions of alternative hypothetical futures) is one
method of assessing potential long-range social, economic, and technical developments (Jungermann,
1985).
There are different categories or types of scenarios one can generate. For example, a distinction
can be made between 'exploratory' and 'anticipatory' scenarios (Jungermann and Thuring, 1987). The
former begin with a known or assumed concrete event and proceed forward. Anticipatory scenarios
start from some future outcome and ask for possible events which could produce this outcome. The
methodology of scenario generation is concerned with issues such as what kinds of scenarios results
from differences in instructions or technique used, and whether higher quality scenarios can be
obtained. From a methodological point of view, event chains are likely to be more elaborated and
36 Journal of Behavioral Decision Making Vol. 2 Iss. No. 1
detailed when people are asked to generate anticipatory scenarios for future, certain outcomes than for
others. In terms of quality differences, the question of whether longer event chains result in better
forecasting remains open.
There are other closely related problems where imagining outcome certainty may help. In fault tree^
analysis we try to anticipate the cause of disastrous outcomes like a nuclear plant malfunction. The
hidden causes include both general factors like corrosion of cooling pipes and specific events like the
unorthodox testing that led to the Chernobyl disaster. In situations where one wants to reveal obscure
or rare causes of a possible outcome, it should prove useful to assume that the outcome has already
occurred and then work backward to its possible causes. The same procedure can usefully be repeated
to identify causal events earlier in the sequence. If a nuclear disaster can be caused by an unorthodox
experiment (that deUberately disables key safety systems), we can ask what events would prompt the
perceived need for such an experiment. At a more mundane level the same strategy of assuming an
outcome with certainty and then seeking its causes can be used to anticipate what can go wrong with a
European vacation or a move of one's laboratory to a new building. A backward perspective on the
outcome to convey its certainty should generate more insight into events that might, in fact lead to it.
In sum, one does not have to be a mad scientist to travel in time. By simulating different situational
perspectives, decision makers may gain new insights about the causes of events in their environment.
ACKNOWLEDGEMENTS
We thank Paul Schoemaker for introducing us to prospective hindsight and posing the initial question
that stimulated this research. We also thank Reid Hastie, Josh Klayman, Richard Thaler, and members
of the Center for Decision Research, the University of Chicago for their helpful comments on earlier
versions of this paper.
APPENDIX A
Example scenario for experiment 1
Basic scenario
Robert H. is an entry-level assistant with a limited business background. Hired for his potential, he is
hard-working and shows great willingness to learn. However, to date his boss has been unimpressed
with his performance. Robert has been assigned a new project, unrelated to anything he has done
previously. If performed well and completed by the deadline, it will provide a valuable opportunity
for him to show his ability to his superiors.
Outcome versions
Past certain: As it turned out, the project (was/was not) finished by the deadline. Please list as many
reasons as you can think of why the project (was/was not) finished by the deadline.
Past uncertain: The morning after the deadline had passed, the gossip around the office was that the
project probably (had/had not) been finished by the deadline. Please list as many reasons as you can
think of why the project may (have/have not) been finished by the deadline.
Future certain: As it will turn out, the project (will/ will not) be finished by the deadline. Please list as
many reasons as you can think of why the project (will/will not) be finished by the deadline.
Future uncertain: The gossip around the office is that the project will (probably/probably not) be
finished by the deadline. Please list as many reasons as you can think of why the project (may/may
not) be finished by the deadline.
D.
J. Mitchell, J. Edward Russo and N. Pennington Back to the Future 37
APPENDIX B
Example scenario for experiment 2
Basic scenario
Imagine that you work at a major consumer products firm. A good friend of yours, Heather, has
been employed for eight months as a brand assistant. Single and in her mid-twenties, she is
determined to combine both a successful career and a satisfying social life. Recently, for professional
and social reasons she has decided to give a large dinner party. She has spared no expense to make
the party a huge success.
Outcome Versions
Past certain: As it turned out, the party was a (great success/terrible failure). Please list as many
reasons as you can think of why the party was a (great succes/terrible failure).
Past uncertain: Please list as many reasons as you can think of why the party may have been a (great
success/terrible failure).
Future certain: As it will turn out, the party will be a (great success/terrible failure). Please list as
many reasons as you can think of why the party will be a (great success/terrible failure).
Future uncertain: Please list as many reasons as you can think of why the party may be a (great
success/terrible failure).
REFERENCES
Abbott,
V.,
Black,
J. B. and
Smith,
E. E. 'The
Representation
of
Scripts
in
Memory', Journal
of
Memory
and
Language. 24(1985), 179-199.
Abelson,
R. P.
'Script Processing
in
Attitude Formation
and
Decision Making",
in J. S.
Carroll
and J. W.
Payne
(eds.).
Cognition
and
Social Behavior, Hillsdale, NJ: Erlbaum,
1976.
Abelson,
R. P.
'Psychological Status
of
the Script Concept', American Psychologist,
36
(1981), 715-729.
Bavelas,
J.
B. 'Effects
of
the Temporal Context
of
Information', Psychological Reports, 32 (1973), 695-698.
Bower,
G. H.,
Black,
J.
B.
and
Turner,
T.
'Scripts
in
Text Comprehension and Memory', Cognitive Psychology,
11
(1979),
Ml-220.
Fischhoff,
B.
'Temporal Setting
and
Judgment Under Uncertainty', Organizational Behaviour
and
Human
Performance, 15 (1976a), 180-194.
Fischhoff,
B.
'For
Those Condemned
to
Study
the
Past: Heuristics
and
Biases
in
Hindsight',
in D.
Kahneman,
P.
Slovic
and A.
Tversky (eds.) Judgment Under Uncertainty: Heuristics
and
Biases,
New
York: Cambridge
University Press,
1980.
Galambos,
J. A. and
Rips, L. 'Memory
for
Routines', Journal
of
Verbal Learning
and
Verbal Behavior, 21(1982),
260-281.
Hoch,
S. J.
'Availability
and
Inference
in
Predictive Judgment', Journal
of
Experimental Psychology: Learning,
Memory
and
Cognition, 10 (1984), 649-662.
Hogarth,
R. M.
'Small Probabilities: Imagination
as
Experience', Discussion, TIMS/ORSA Meeting, Chicago,
April,
1983.
Holland,
J. H.,
Holyoak,
K. J.,
Nisbett,
R. E. and
Thagard,
P. R.
Induction, Cambridge, MA:
The MIT
Press,
1986.
Jungermann, H. 'Inferential Processes
in
the Construction
of
Scenarios', Journal of Forecasting,
4
(1985), 321-327.
Jungermann,
H. and
Thuring,
M. 'The Use of
Mental Models
for
Generating Scenarios',
in G.
Wright
and P.
Ayton (eds.). Judgmental Forecasting, New York: John Wiley
and
Sons, Ltd.,
1987.
Kahneman,
D. and
Tversky, A. 'Prospect Theory:
An
Analysis
of
Decision Under Risk', Econometrica, 47 (1979),
263-291.
Koriat,
A.,
Lichtenstein,
S. and
Fischhoff,
B.
'Reasons
for
Confidence', Journal
of
Experimental Psychology:
Human Learning
and
Memory,
6
(1980), 107-118.
Nisbett,
R. E.,
Borgida,
E.,
Crandall,
R. and
Reed,
H.
'Popular Induction: Information
is Not
Necessarily
38 Journal of Behavioral Decision Making Vol. 2 Iss. No. I
Informative', in J. S. Carroll and J. W. Payne (eds.). Cognition and Social Behavior, Hillsdale, NJ: Erlbaum,
1976.
Ross,
L. 'The Intuitive Psychologist and His Shortcomings: Distortions in the Attribution Process', Advances in
Experimental Social Psychology, 10 (1977), 173-220.
Rothbart, M. and Snyder, M. 'Confidence in the Prediction and Postdiction of an Uncertain Outcome', Canadian
Journal of Behavioral Science, 1 (1970),
58-43.
Sevon, G. 'Cognitive Maps of Past and Future Economic Events', Acta Psychologica, 56 (1984), 71-79.
Strickland, L. H., Lewicki, R. J. and Katz, A. M. 'Temporal Orientation and Perceived Control as Determinants of
Risk Taking', Journal of Experimental Social Psychology, 2 (1966),
143-151.
Ward, S. J., Hastie, R. and Taylor, S. E. 'Some Determinants of the Complexity of Causal Analysis: Expectation,
Prediction and Explanation', unpublished manuscript. Harvard University, 1978.
Weick, K. E.
TJie
Social Psychology of Organizations, Reading, MA: Addison-Wesley, 1979.
Authors' biographies:
Deborah Mitchell is Assistant Professor of Marketing at the Wharton School of the University of Pennsylvania.
She holds an MBA from the University of Chicago and is currently completing doctoral studies there.
J. Edward Russo is Associate Professor of Marketing and Behavioral Science at the Johnson Graduate School of
Management of Cornell University. He earned a Ph.D. in psychology from the University of Michigan in 1971. His
current research focus is decision aiding, especially that of consumers.
Nancy Pennington is Assistant Professor of Psychology at the University of Colorado. She earned her Ph.D. from
Harvard University in education and has served on the faculty at the University of Chicago. Her main research
interest is cognitive processes in decision making.
... Premortems can be used to enhance and refine a health promotion program prior to implementation. Support for this approach comes from research indicating that prospective hindsight increases the ability to correctly identify potential explanations for program failure by 30% [21]. Also, a premortem circumvents groupthink by creating a safe place for individuals to voice concerns and leverages collective intelligence by involving diverse individuals to identify program weaknesses/threats [22]. ...
... We conducted a critical assessment using a premortem approach similar to Beidas et al's example from USbased implementation scientists to examine the promises and pitfalls of the field [27]. Premortems use prospective hindsights which imagine future failures and generate potential explanations for why they occur as a way of putting in place measures that can reduce the likelihood of those failures occurring [27,28]. The guiding prompt used in the discussions was: "There has been a significant growth in the application of implementation science in low-resource settings such as ours. ...
Article
Full-text available
Background As the field of implementation science continues to grow, its key concepts are being transferred into new contexts globally, such as Low and Middle Income Countries (LMICs), and its use is constantly being reexamined and expanded. Theoretical and methodological positions commonly used in implementation research and practice have great utility in our work but in many cases are at odds with LMIC contexts. As a team of implementation scientists based in Zambia, we offer this commentary as a critical self-reflection on what has worked and what could limit us from fully utilizing the field’s promise for addressing health problems with contextual understanding. Main body We used a ‘premortem,’ an approach used to generate potential alternatives from failed assumptions about a particular phenomenon, as a way to reflect on our experiences conducting implementation research and practice. By utilizing prospectively imagined hindsights, we were able to reflect on the past, present and possible future of the field in Zambia. Six key challenges identified were: (i) epistemic injustices; (ii) simplified conceptualizations of evidence-informed interventions; (iii) limited theorization of the complexity of low-resource contexts and it impacts on implementation; (iv) persistent lags in transforming research into practice; (v) limited focus on strategic dissemination of implementation science knowledge and (vi) existing training and capacity building initiatives’ failure to engage a broad range of actors including practitioners through diverse learning models. Conclusion Implementation science offers great promise in addressing many health problems in Zambia. Through this commentary, we hope to spur discussions on how implementation scientists can reimagine the future of the field by contemplating on lessons from our experiences in LMIC settings.
... People tend to favor information that supports their views, making it challenging to consider alternatives in decisionmaking [11]. To improve self-confidence calibration, we design an intervention inspired the "pre-mortem" proposed by Klein [32] and Mitchell [54]. Participants are asked to imagine a scenario where their initial decision was actually wrong, encouraging them to think beyond their initial perspective [31]. ...
Preprint
Full-text available
In AI-assisted decision-making, it is crucial but challenging for humans to achieve appropriate reliance on AI. This paper approaches this problem from a human-centered perspective, "human self-confidence calibration". We begin by proposing an analytical framework to highlight the importance of calibrated human self-confidence. In our first study, we explore the relationship between human self-confidence appropriateness and reliance appropriateness. Then in our second study, We propose three calibration mechanisms and compare their effects on humans' self-confidence and user experience. Subsequently, our third study investigates the effects of self-confidence calibration on AI-assisted decision-making. Results show that calibrating human self-confidence enhances human-AI team performance and encourages more rational reliance on AI (in some aspects) compared to uncalibrated baselines. Finally, we discuss our main findings and provide implications for designing future AI-assisted decision-making interfaces. CCS CONCEPTS • Human-centered computing → Empirical studies in HCI.
... d Pre-gaming: If a metric is proposed, the exercise of imagining how it could be gamed, and building incentives aimed at forestalling gaming, can be useful. This idea is closely related to research about the effectiveness of such planning by Mitchell, Russo, and Pennington, 33 which Gary Klein later popularized as a pre-mortem. 34 If done well, these can be very helpful-but some care should be taken about how to actually run such exercises, as they are often carried out poorly. ...
Article
Full-text available
Metrics are useful for measuring systems and motivating behaviors in academia as well as in public policy, medicine, business, and other systems. Unfortunately, naive application of metrics to a system can distort the system and even undermine the original goal. There are two interrelated problems to overcome in building better metrics in academia and elsewhere. The first, specifying evaluable metrics that correspond to the goals, is well recognized but still often ignored. The second, minimizing perverse effects that undermine the metric or that enable people to game the rewards, is less recognized but is critical. This perspective discusses designing metrics, beginning with design considerations and processes; the presentation of specific strategies for mitigating perverse impacts, including secrecy, randomization, diversification, and post hoc specification; and continuing with important desiderata and tradeoffs involved with examples of how they can complement each other or differ. Finally, this perspective presents a comprehensive process integrating these ideas.
Article
Carbon capture and storage (CCS) requires the safe, long-term storage of CO 2 in the subsurface. This challenges subsurface practitioners to build on and adapt many of the techniques and processes developed for hydrocarbon exploration and production to create innovative approaches to assessing CO 2 storage risk and uncertainty. A wide-ranging and integrated understanding of the processes controlling CO 2 behaviour in the subsurface is required to facilitate effective risk identification and characterization. In this paper we explore the different physical processes acting within storage formations and their evolution throughout the injection project life cycle to illustrate key controls on CO 2 plume behaviour and frame risk identification and characterization. An approach applying wide-reaching premortem risk identification coupled with evidence-led uncertainty characterization is described to frame inputs for detailed risk characterization and management activity during early screening of prospects for CO 2 storage.
Article
Software design teams need methods to evaluate plans as part of agile development processes. The premortem is a cognitive, structured, analytic technique that supports team plan evaluation and re-planning. Few empirical or longitudinal studies involving premortems exist. Ten game development teams (n=68 members) conducted premortems early on during a year-long game development project. Teams provided initial ratings of their game design plan, then conducted a premortem, and revised their plans. In the premortems, teams identified 17.8 unique reasons on average for project failure and 16.7 mitigations to those failures. Reasons for project failure focused mainly on game design execution, team communication, and game complexity (e.g., too many levels, branching). While most teams identified solutions for these challenges, surprisingly few teams revised their plans to scale back the game design complexity.
Chapter
Full-text available
briefly describe the scenario concept, what technical methods exist for constructing scenarios, and what types of scenarios are to be distinguished / propose a conceptual framework for understanding the cognitive processes involved in the intuitive generation of scenarios / description of retrieval and inference processes based on mental models of causal knowledge (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Article
Full-text available
Made a comparison between the confidence of 23 undergraduates either predicting the outcome of an uncertain event or postdicting the outcome of the same event. Using a procedure in which Ss rolled a die and bet on the outcome, it was found that Ss predicting the outcome before the die was rolled bet more money and reported greater confidence in being correct than Ss postdicting the outcome after the die had already been rolled. Results are interpreted as support for the hypothesis that a form of magical thinking mediates the frequently reported positive relationship between S's desire for a particular outcome and his estimate of that outcome's likelihood of occurrence. (French summary) (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Article
Full-text available
Conducted 3 experiments to investigate the cognitive processes underlying predictive judgments about an uncertain future. A hypothesis-generation task was developed to study the joint influence on predictive judgments of the availability heuristic and of retrieval processes in associative memory. 160 graduate students generated reasons why a future event might (pro) and might not (con) occur. The order of generation was counterbalanced. The 1st set of reasons was generated in greater quantity and more quickly; previously generated reasons appeared to interfere with Ss' attempts to generate additional reasons. Moreover, this differential reason production led to systematic primacy effects in Ss' predictions. Likelihood judgments were influenced more by whatever side of the issue Ss thought about first. When a delay was introduced between the pro and con generation tasks, interference largely disappeared, but judgments were more influenced by the 2nd set of reasons—a recency effect. Two by-products of thinking that may function as cues about underlying probability of occurrence have to do with content and with metamemory. (53 ref) (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Article
Full-text available
Suggests that there has been growing interest within several subfields of psychology in the schematic nature of mental representations of real-world objects and events. One simple form of schema is the "script," embodying knowledge of stereotyped event sequences. The present author traces applications of the script concept in artificial intelligence, cognitive psychology, and social psychology. Scripts are compared and contrasted with related concepts such as habits, roles, and games. The suggested theoretic function of the script concept is to unify central notions in learning, developmental, clinical, social, and cognitive psychology. The present concept, while still incompletely articulated, offers encouragement toward such a unification. Areas of accumulating empirical evidence and of needed theoretical extension of the script concept are indicated. (76 ref) (PsycINFO Database Record (c) 2012 APA, all rights reserved)
Chapter
Attribution theory is concerned with the attempts of ordinary people to understand the causes and implications of the events they witness. It deals with the “naive psychology” of the “man in the street” as he interprets his own behaviors and the actions of others. For man—in the perspective of attribution theory—is an intuitive psychologist who seeks to explain behavior and draw inferences about actors and their environments. To better understand the perceptions and actions of this intuitive scientist, his methods must be explored. The sources of oversight, error, or bias in his assumptions and procedures may have serious consequences, both for the lay psychologist himself and for the society that he builds and perpetuates. These shortcomings, explored from the vantage point of contemporary attribution theory, are the focus of the chapter. The logical or rational schemata employed by intuitive psychologists and the sources of bias in their attempts at understanding, predicting, and controlling the events that unfold around them are considered. Attributional biases in the psychology of prediction, perseverance of social inferences and social theories, and the intuitive psychologist's illusions and insights are described.
Article
Discusses situations in which people fail to use valid but abstract information about base-rate behavior. A review of empirical studies indicates that concrete information is used heavily in judgments about self and others; evidence is also presented showing little support for the view that people use consensus information in making attributions. Methodological issues are also examined. (PsycINFO Database Record (c) 2012 APA, all rights reserved)