Naıve Theories of Causal Force and Compression of Elapsed Time

ATTITUDES AND SOCIAL COGNITION
Naı̈ve Theories of Causal Force and Compression of Elapsed
Time Judgments
David Faro
Ann L. McGill and Reid Hastie
London Business School
University of Chicago
Recent research has shown that when people perceive a causal relation between 2 events, they
“compress” the intervening elapsed time. The present work shows that a naı̈ve mechanical–physical
conception of causality, in which causal forces are believed to dissipate over time, underlies the estimates
of shorter elapsed time. Being primed with alternative, nondissipative causal mechanisms and having the
cognitive capacity to consider such mechanisms moderates the compression effect. The studies rule out
similarity, mnemonic association, and anchoring as alternative accounts for the effect. Taken together,
the findings support the hypothesis that causal cognition plays a major role in judgments of elapsed time.
The implications of the compression effect on the timing of future actions, persistence, and causal
learning are discussed.
Keywords: causal reasoning, causal attribution, time estimation, lay theories, persistence
causal connection (Haggard, Clark, & Kalogeras, 2002; see also
Buehner & Humphreys, 2009; Eagleman & Holcombe, 2002;
Moore, Lagnado, Deal, & Haggard, 2009; Wohlschlager, Haggard,
Gesierich, & Prinz, 2003). In both cases, on a historical and on a
short time scale, judgments of causally linked events typically
resulted in an underestimation of the actual interval (Faro et al.,
2005; Haggard et al., 2002).
The tendency to bind past causes and effects in time and, as a
result, to underestimate the time that transpired between them may
have several consequences. If people bind past causes and effects
in time, they might expect more rapid effects in the future. Suppose the parent wants to time the tough and somewhat unpleasant
lecture to the child to have maximum impact before upcoming
final exams. If the parent underestimates the amount of time the
lecture might take to have impact, he or she may delay giving it
until it is potentially too late (see also Buehler, Griffin, & Ross,
1994; Roy, Christenfeld, & McKenzie, 2005). Further, if the parent
expects a rapid effect that does not materialize, he or she might
revert to the same course of action, which the child perceives as
badgering. The tendency to bind causes and effects too closely in
time may also affect attributions of causality. Assume, as suggested, that the parent delays giving the tough lecture until it is too
late for it to really have an effect, but a short time later, the child
shows a marked improvement in schoolwork. The parent might
attribute that improvement to the lecture instead of another, perhaps more plausible cause (e.g., the child’s discovering that a
higher grade point average is rewarded by the school with offcampus privileges).
Why do people bind causes and effects in time? Is the perception of causality really the factor producing this effect, and if so,
what underlies this tendency? The empirical research on this
How much time might have elapsed between the launch of an
economic program and the emergence of an economy from recession, between joining a dating service and finding someone you
want to marry, or between giving your child a tough lecture about
trying harder in school and seeing an effect on his or her performance? Recent research has shown that people subjectively bind
such cause– effect events in time and “compress” the time elapsed
between them. Hence, for instance, if a parent believes that the
tough lecture was the reason for an improvement in the child’s
performance, the parent would estimate the interval to be shorter.
Several behavioral studies have established that people judge the
time elapsed between pairs of historical events to be shorter when
they perceive the events to be causally linked than when people do
not perceive them to be so (Faro, Leclerc, & Hastie, 2005).
Similar effects occur on a shorter time scale. For example, in
another study, when participants intentionally made a movement
that appeared to cause a sound, they thought the events were closer
together than when the two events occurred with no apparent
David Faro, London Business School, London, United Kingdom; Ann
L. McGill, and Reid Hastie, Booth School of Business, University of
Chicago.
Financial support for this work was provided by the Centre for Marketing at the London Business School and the Kilts Center for Marketing at
the Booth School of Business, University of Chicago. We thank Dan
Bartels, Simona Botti, Katherine Burson, Dan Goldstein, Monika Heller,
Joshua Klayman, France Leclerc, Ben Rottman, and George Wu for helpful
comments on this research.
Correspondence concerning this article should be addressed to David
Faro, London Business School, Regent’s Park, London NW1 4SA, United
Kingdom. E-mail: [email protected]
Journal of Personality and Social Psychology, 2010, Vol. 98, No. 5, 683–701
© 2010 American Psychological Association 0022-3514/10/$12.00
DOI: 10.1037/a0019261
683
684
FARO, MCGILL, AND HASTIE
phenomenon, for judgments of both historical events and motor
movements, has been largely limited to documenting its existence.
In this article, we explore why the effect occurs by identifying
some of its underlying cognitive mediators. First, we argue that
estimates for cause– effect intervals depend on a person’s beliefs
about specific types of causal mechanisms presumed to link the
events. Second, we propose that the general tendency to bind
causally related events in time is driven by the dominant role that
concepts of mechanical–physical causation play in human cognition (Lewin, 1935; Talmy, 1988). Mechanical–physical causality
involves people’s perceptions and inferences about whether and
how, for instance, the movement of one object leads to the movement of a second object. We focus in particular on the naı̈ve
physics concepts of force and momentum (McCloskey, 1983). We
hypothesize that when people believe two events are causally
related, they bind the two events in time because they believe most
causal forces, similar to physical forces, dissipate over time.
Why do we claim that the mechanical–physical metaphor and a
dissipative view of causal forces dominate causal thinking and drive
the shorter time estimates? Researchers have speculated that intentional control of physical objects in the immediate environment is the
root metaphor from which more sophisticated but commonsense
notions of causality develop (Leslie, 1984; Leslie & Keeble, 1987;
Piaget, 1955; White, 1988a). Thus, an infant’s first experiences of
causality involve his or her motor movements followed by movements and noises from nearby objects. An alternative interpretation,
which also supports the notion that mechanical–physical causality is
fundamental, suggests that phenomenal visual impressions of causality (Michotte, 1963) are the source of early causal intuitions (Leslie,
1988). In naı̈ve mechanical–physics, the force of actions and movements is perceived to dissipate with the passage of time (McCloskey,
1983). Belief in and experience of dissipative forces is, hence, an
integral part of early perception of causality. We suggest this fundamental belief underlies the temporal binding of cause– effect events.
When people believe one event caused another, they bind the two in
time because they feel the force imparted by the cause would otherwise dissipate before having impact.
However, although the physical–mechanical, dissipative conception of causation is primary and dominates much of causal
reasoning, it is not the only way people can think about causation
(C. A. Anderson & Lindsay, 1998). In some contexts, the salient
conceptual model of causation may involve forces that are, for
instance, perceived as building up over time (Einhorn & Hogarth,
1986; Sterman, 2002). Under such conditions, the tendency to bind
causally related events in time may be reduced or reversed. This
implies that the tendency to compress estimates of time between
causally related events will not be universal but will depend on the
salient mental model of causation.
First, we review the literature on the use of time as a cue to
causality. Then, we propose a typology of naı̈ve causal mechanisms with different trajectories of force over time, along with
predictions about how each will affect time estimates. We test
these predictions by manipulating the salience of different naı̈ve
causal mechanisms. Our findings show that differences in estimates of time depend on the type of causal process participants
attribute to the events. Compression occurs when participants
consider causal forces they believe dissipate over time. Hence, for
instance, they are more prone to underestimate the time between
events involving emotions because they tend to think emotions
lose their force over time (cf. Gilbert, Pinel, Wilson, Blumberg, &
Wheatley, 1998). The compression in estimates of cause– effect
intervals can be muted or reversed if alternative (nondissipative)
causal mechanisms are salient when time intervals are estimated.
However, the mechanical–physical, dissipative view of causation
appears to act as the default and to result in a general tendency to
provide shorter time estimates for causal intervals.
The Role of Time in Causal Inference
Time plays a central role in causal inference (Einhorn & Hogarth, 1986; Hume, 1938; Lagnado & Sloman, 2004; Lagnado,
Waldmann, Hagmayer, & Sloman, 2007). For example, when
several variables are intercorrelated, temporal order tends to dominate strength of correlation when people infer causal relation in a
system of events (Lagnado & Sloman, 2006). Beyond the basic
assumption of temporal precedence, people also expect a cause to
be proximate in time to the effect. Michotte’s (1963) studies
showed that temporal proximity plays a key role in adults’ immediate perceptions of phenomenological causality. Participants
watched an animation of a billiard ball coming in contact with
another ball and the second ball starting to move in the same
direction as the first. Their reports of the launching effect—the
perception that the first ball caused the movement of the second—
decreased as the time gap between the first ball touching the
second and the second one moving increased (Scholl & Tremoulet,
2000). Temporal proximity also affects judgments of causal
strength in the more conceptual causal induction paradigm. In this
research, people infer causality between events on the basis of
statistical contingencies. For example, in one study, participants
pressed a button on a personal computer keyboard and then saw a
picture of a flute playing on the screen (J. R. Anderson & Sheu,
1995). Ratings of causal strength decreased as the time interval
separating the events increased (J. R. Anderson & Sheu, 1995;
Shanks, Pearson, & Dickinson, 1989). In a similar vein, a person’s
own thought is perceived to be a cause of his or her action if that
thought precedes the action closely in time (Wegner & Wheatley,
1999). Finally, in legal cases, liability decreases with the temporal
interval between a negative, harmful event and the actions of a
defendant who allegedly caused the harm (Johnson & Drobny,
1985).
Although these findings show close temporal proximity is a key
component in causal inference, there are important exceptions to
this generalization. For example, if a person thinks advertising
works through a gradual diffusion process but sales go up the day
after an ad is placed, the increase in sales will not be causally
attributed to the advertisement because the time gap is too short
(Einhorn & Hogarth, 1986). Similar qualifications apply to the
findings of experimental studies of causal induction. For example,
when people have a reason to expect a delay between cause and
effect, judgments of causal strength for temporally distant events
are not diminished (Buehner & May, 2003). Further, identical
contingency information can result in different judgments of causality, depending on the temporal assumptions people have about
the cause– effect events prior to observing the events (Hagmayer &
Waldmann, 2002).
CAUSAL FORCE AND COMPRESSION OF ELAPSED TIME
A Temporal Typology of Naı̈ve Causal Processes
As the preceding review suggests, people expect certain relations between causal strength and time elapsed between the occurrence of two events. Although close temporal proximity is
typically expected for two events to be seen as causally related, the
specific inference may depend on the mechanism by which the
cause is believed to lead to the effect (Einhorn & Hogarth, 1986).
One recurring mechanism notion that features in how people
reason about the workings of physical causal processes is force
over time (McCloskey, 1983). Physical forces tend to and are
believed to diminish over time: Kinetic forces are lessened through
friction, and sound and light waves dissipate. More generally,
mechanical–physical causality is associated with temporally descending trajectories (Michotte, 1963; Morris, Sheldon, Ames, &
Young, 2007). Dissipative causes, in which the force of the cause
degrades with time, are common and, as noted, are the first causal
relations to be understood by children. It is important to note that
concepts such as dissipating force and momentum from
mechanical–physical causality metaphorically serve to understand
other relations, including some social interactions and events (Heider, 1944; Markman & Guenther, 2007; Talmy, 1988). For instance, people tend to describe downtrends in the stock market in
terms of mechanical metaphors (e.g., the stock slid, dropped,
plummeted; Andreassen, 1987; Morris et al., 2007). Similarly,
people may believe that social factors (e.g., an actor’s thoughts or
emotions, other people’s comments or actions) behave much like
physical forces, starting strong in potential impact but dissipating
with time (Markman & Guenther, 2007).
We suggest this dominant tendency to view the force of causes
as dissipative underlies the binding of causally related events in
time. To explicate this point, consider the phenomenon of displacement (Freyd & Finke, 1984). In the original studies with this
paradigm, observers see a moving target on a screen. The target
then disappears and observers are asked to indicate its last location
on the screen. They usually report it traveling farther than it
actually did (Freyd & Finke, 1984). More important for the present
research, later studies have shown that when the target’s movement is triggered by another object, it is remembered as having
traveled less far on the screen (Hubbard, Blessum, & Ruppel,
2001). This appears to occur because observers perceive the force
imparted from the object into the target to dissipate over time.
Hence, when the target’s movement is caused by another object,
the distance it is perceived to have traveled contracts.
We propose that just as the notion of dissipating force makes
people remember the visual target traveled less far in space, it may
also make people bind cause– effect events in time. The reasoning
is similar: Because the force imparted from the cause is thought to
dissipate over time, for that force to have led to the effect, only
short time must have elapsed. Hence, the binding of causally
related events in time is an extension of the mechanical–physical
theory of causation in which forces degrade over time. Much past
work suggests this mechanical–physical view is people’s default
approach to causation (Keil, Levin, Richman, & Gutheil, 1999).
Therefore, when dissipative causes are salient or when people
think of generic, unspecified causal relations or opt for the easiest
interpretation, they will bind causes and effects in time and tend to
underestimate the interval.
685
However, people also learn about other types of causes with
different temporal implications. The force of some causes builds
over time. People speak of “snowball” effects in which a small
problem grows into a larger one. “A stitch in time saves nine”
reflects the understanding that a small tear can grow to a big one
if not dealt with right way. In particular, people learn quite early in
life that many biological processes involve patterns of accumulation and growth that physical objects and artifacts do not embody.
Children as young as 3 years old are able to understand that
animals grow larger with time, and older children allow for a rather
dramatic change in size and shape (Rosengren, Gelman, Kalish, &
McCormick, 1991). Similarly, most people know how the force of
a planted seed increases as it grows, how illnesses may grow more
severe if not treated, and how small outbreaks of a disease may
swell to an epidemic (Rottman & Ahn, 2009). Further, recent
studies following Michotte’s (1963) launching effect paradigm
have revealed that when there is delayed acceleration of the second
object, observers describe the causal relation in terms of biological
animacy (Tremoulet & Feldman, 2000). In general, animacy, as
opposed to inanimate mechanics, is associated with ascending
trajectories (Heider & Simmel, 1944; Michotte, 1963; Morris et
al., 2007; Rosengren et al., 1991).
It is typically more difficult, even for well-educated adults, to
reason about accumulative causal processes (Cronin, Gonzalez, &
Sterman, 2009; Einhorn & Hogarth, 1986; Sterman, 2002; White,
1998b, 1999). Perhaps this is why people exhibit a default reliance
on close temporal proximity as a cue to causality (Sterman, 2002).
However, accumulative forces may, at least metaphorically, feature in people’s lay conceptions of causation. People speak of
others being “eaten up” by envy as if the emotion were a wasting
disease. Fashions and opinions are sometimes described in epidemic terms (e.g., viral). A small idea may gestate, grow, and burst
out. Hence, just as the metaphor of physical causation may influence judgments of social causal processes, so, too, might the
metaphor of biological causation. If a mechanical–physical, dissipative view of causes underlies the temporal binding of causally
related events then when a biological metaphor is salient— or,
more generally, when people perceive the causal process to involve accumulating strength—people are unlikely to compress
time; they may instead judge that the two events occurred with
greater temporal delay between them.
Finally, people may believe some causal factors exert a relatively constant force over time. In everyday language we often
refer to these factors as enabling conditions. Usually, they operate
in conjunction with other necessary conditions and are not perceived as precipitating events because they do not change states
before the effect is observed. (Philosophers and statisticians often
do not label stable enabling conditions as proper causes because
they prefer to attribute causal relations only to pairs of events
where a change in one event 关variable兴 precedes a change in
another event 关variable兴.) The presence of oxygen for fire, for
instance, is seen as a necessary, stable enabling condition. Again,
people may use the metaphor of such causal dynamics to account
for social events, particularly in comparing across people instead
of over time (McGill, 1989). For example, people may believe that
others’ stable personality traits render them more or less prone to
problems with alcohol or to stress-related outbursts. Similarly,
people often speak of an “accident waiting to happen.” In the case
of stable causes, time is weakly associated with causal force, if at
686
FARO, MCGILL, AND HASTIE
all. As a consequence, when people estimate a time interval on the
basis of a stable causal relation, they may not compress or expand
estimates of the time.
To summarize, we suggest causes can be broadly categorized in
terms of the force associated with each type: dissipative, accumulative, and stable. We further suggest that the theory and associated
belief a person holds about how causal force varies over time will
determine the impact of causation on time estimates. For example,
in perceiving a causal relation between giving a tough lecture to a
child and seeing an improvement in his or her schoolwork, a parent
might bind the two events in time if he or she views the lecture as
a physical force that “strikes” the child and causes a reaction that
might dissipate as time elapses. The parent might give longer
estimates if he or she thinks of the first, potentially causal event as
a biological force that grows, eventually inciting the child to
action. Finally, we suggest that the dissipative view of causes, as
rooted in mechanical–physical causation, is the dominant view of
causal relations and results in a general tendency to bind causally
related events in time.
Of course, this typology of causal processes— dissipative, accumulative, or stable—and the domain metaphors we proposed
have limitations and exceptions. For example, people may know of
physical processes that accumulate force (e.g., tectonic plates) and
of biological forces that lose momentum (e.g., an infected individual’s contagiousness). Although prior research shows that domain
metaphors such as mechanical physics and biology can affect
judgments in other contexts, we do not, for instance, propose that
when people think of causal relations in the social world, they
think only in terms of Hume’s billiard balls or Newton’s cradle.
Nonetheless, we believe our general typology and metaphors cover
many beliefs people have about causal processes and about profiles of causal force over time.
Overview of the Experiments and Predictions
We report five experiments that explore the role of participants’
beliefs about causal processes in shaping their judgments of time. The
purpose of the first study is to establish that causal beliefs per se—
and not naturally confounded variables—produce the timecompression effect. The first study tests and rules out two factors,
mnemonic association and similarity between events, as possible
sources of the effect. Once we show that causality is at the heart of
the effect, in subsequent studies we manipulate the specific contents of participants’ causal mechanism beliefs and examine the
effects on time estimates. In Studies 2 and 3, we rely on participants’ intuitions and beliefs about the progression of causal force
over time in different domains to manipulate the salient causal
process. In Study 4, we hold domain constant and manipulate
salient beliefs about causal processes by exposing participants to
different trajectories of a causal force in an unrelated task. In Study
5, we manipulate cognitive load, making it more or less difficult to
consider alternative, more complex types of causal processes.
In each study, participants estimate time elapsed between several pairs of historical events. We present two focal analyses. The
first is an analysis of time estimates across participants in different
conditions. We predict that the accumulative-cause and stablecause manipulations will result in longer time estimates than will
the dissipative-cause manipulation and a control in which no
information, priming, or other manipulation is presented to trigger
a type of causal process. The dissipative-cause manipulations
should result in shorter or qualitatively the same estimates as a
control. The second type of analyses is based on withinparticipants, across-items correlations between time estimates and
ratings of perceived causality. Dissipative causal processes imply
a negative relation between these variables—the stronger the
causal relation, the shorter the time. We predict our manipulations
of the availability of alternative causal processes will reduce or
reverse this negative relation.
Study 1: Ruling Out Similarity and Mnemonic
Association
A recent demonstration of the compression effect relied on
participants’ estimates of elapsed time between pairs of historical
events (Faro et al., 2005). Estimates were shorter when a given pair
of events was perceived to be causally linked than when not.
Similarly, causally related pairs were judged to be closer in time
than equidistant noncausal pairs. Although these findings are consistent with the assertion that time estimates depend on causal
beliefs, two other factors naturally confounded with causality—
similarity and mnemonic association— could account for these
results.
People expect causes and effects to be similar to each other on
dimensions such as size, magnitude, and appearance (Einhorn &
Hogarth, 1986; Gilovich & Savitsky, 1996; Golonka & Estes,
2009; LeBoeuf & Norton, 2007). For instance, many people have
a hard time believing the HIV virus, which led to a worldwide
epidemic of a potentially fatal disease, is itself fairly weak, much
weaker than the virus that causes the common cold (Rozin &
Nemeroff, 2002). It is possible that in previous studies showing
compression, time estimates between events perceived to be causally related were shorter, compared with causally unrelated events,
because participants thought the related events were similar to
each other (Morewedge, Preston, & Wegner, 2007). Hence, they
might also believe the events occurred at similar times. This
account predicts that increasing the perceived similarity between
two events, independent of the strength of the causal link, would
result in shorter elapsed time estimates.
A second possible confound in prior studies is mnemonic association. Causally related events are more likely to bring each other
to mind, and it could that be strong mnemonic associations drive
the shorter time estimates between such events. Recent research
has shown that ease of recall or feelings of conceptual fluency can
affect duration and elapsed time estimates (Raghubir & Menon,
2005; Whittlesea, 1993; Xu & Schwarz, 2005; see also Menon,
1993). In one study, events that were easier to recall were judged
to have occurred more recently (Xu & Schwarz, 2005). Participants seemed to reason that it is generally harder to recall events
in the distant past than to recall events in the recent past. Therefore, they inferred that events easier to remember must have
occurred more recently. Similarly, if one event easily brings another to mind, as causes and effects do, participants might feel the
two occurred close together in time. This account suggests the
shorter time estimates are driven by mnemonic association (which
results in ease of recall or fluency) rather than by the causal
relation between the events. Pairs of events should, therefore, be
judged to have occurred closer in time if their strength of association is increased. Study 1 contrasts these alternative accounts with
CAUSAL FORCE AND COMPRESSION OF ELAPSED TIME
the proposition that the perception of causal relation between the
events underlies the shorter time estimates.
Method
Participants. For $8 each, 118 University of Chicago students
participated in a 30-min experiment on knowledge of historical
events.
Procedure and design. The design was a 3 (pair type: causal
versus same domain [noncausal] versus different domain [noncausal], manipulated within subjects) ⫻ 3 (explanation task: causality explanation versus similarity explanation versus no explanation, manipulated between subjects) mixed design.
Pair type. The study had three types of pairs: causal, same
domain, and different domain. Causal pairs consisted of events
causally related to each other (e.g., Russians launch the first
satellite that circles the earth [Sputnik]—American astronauts
landing on the moon.). The same-domain pairs consisted of events
from the same subject domain but not causally related in an
obvious way (e.g., launch of first weather satellite by United States
of America—American and Soviet spacecrafts dock in orbit,
cooperating for the first time). The different-domain pairs consisted of two events selected from different domains and not
causally related (e.g., Kennedy elected, the youngest [and first
Catholic] president—Federal Express begins operations). For each
matched set of causal, same-domain pairs, and different-domain
pairs, the actual elapsed time and years of occurrence were approximately the same (see Appendix A for full list of event pairs).
Participants estimated intervals of time between events for 12 pairs
of each type. The actual elapsed time between events in each pair
ranged from 3 years to 34 years. The different pair types were
presented in a mixed order, and two different orders of pair
presentation were used.
Explanation task. The experiment also involved a written
explanation task that took place prior to time estimates. This
manipulation had three levels: causality explanation, similarity
explanation, and no explanation. Prior to estimating time between
each pair, participants in the causality-explanation condition wrote
an explanation of how the two events might be causally related.
Participants in the similarity-explanation condition provided a
written explanation of how the events might be similar to each
other. Participants in the no-explanation condition did not provide
any written explanation.
Dependent measures. Participants estimated the number of
years that elapsed between the events for each pair. The time
judgment task took place after the explanation task for each pair.
After the explanations and time judgments for all pairs were
completed, participants completed a surprise cued-recall task. This
task was intended to measure the strength of mnemonic association
for the event pairs. Participants viewed only the first event in each
pair and were asked to recall the event previously paired with this
cue event. Following the cued-recall task, participants in the
causal-explanation and no-explanation conditions rated the degree
to which the events were causally related (on a 0 –100 scale).
Participants in the similarity-explanation and no-explanation conditions rated the degree to which the events in each pair were
similar to each other (on a 1–11 scale).
687
Predictions
If similarity drives shorter time estimates, the similarityexplanation task should shorten time estimates to the extent that it
increases perceived similarity. If mnemonic association drives
shorter time estimates, manipulations increasing strength of association should result in shorter time estimates. Both the similarityand causality-explanation manipulations are expected to increase
strength of association for all pair types and should result in shorter
time estimates. If, however, causality is responsible for the shorter
time estimates, only the causality explanation task should result in
shorter time estimates, and only for pair types for which participants can construct plausible causal explanations. We expected
that finding such explanations would be easier for the samedomain and causal pairs.
Results
Causality and similarity measures. An analysis of variance
(ANOVA) on the causality ratings showed a main effect of pair
type (Mcausal ⫽ 67.45; Msame domain ⫽ 45.64; Mdifferent domain ⫽
13.22), F(2, 154) ⫽ 422.09, p ⬍ .001. The effect of the causalityexplanation task was also significant (M causality ⫽ 44.28;
Mno explanation ⫽ 39.87), F(1, 77) ⫽ 4.59, p ⬍ .05. The interaction
of pair type and explanation was not significant. However, contrasts indicated that as expected, trying to find a causal explanation
significantly increased causality ratings for the same-domain pairs
(Mcausality ⫽ 48.86; Mno explanation ⫽ 42.33), F(1, 77) ⫽ 4.09, p ⬍
.05.1 The effects of causal explanation for the causal and differentdomain pairs were not significant. We suspect that causal explanation did not significantly increase causality ratings for the causal
pairs because of a potential ceiling effect; these pairs were already
judged highly causally related. Participants likely were not able to
find convincing causal explanations for the different-domain pairs.
An ANOVA on the similarity ratings showed a main effect of
pair type (Mcausal ⫽ 7.75; Msame domain ⫽ 6.61; Mdifferent domain ⫽
3.04), F(2, 154) ⫽ 403.70, p ⬍ .001. The effect of the similarityexplanation task was marginally significant, resulting in greater
perceived similarity compared with the no-explanation condition
(Msimilarity ⫽ 6.04; Mno explanation ⫽ 5.57), F(1, 77) ⫽ 2.94, p ⬍
.10. The interaction of pair type and explanation was not significant.
Time estimates. The main dependent measure for time estimates used in the analyses of variance across the studies is signedorder-of-magnitude-error (SOME; Brown, 2002; Brown & Siegler,
1992, 1993; Nickerson, 1981). SOME was computed for each
estimate according to the following formula (where the estimate
was 0, the constant 1 was added to the estimate):
SOME ⫽ log 10 共estimate/actual 兲
Taking the ratio of estimate to actual elapsed time normalizes
the data over a range of possible responses spanning several orders
of magnitude and allows for averaging across event pairs that vary
in terms of actual time (Brown & Siegler, 1992, 1993). Taking the
log of the ratio helps to equate order-of-magnitude underestimates
with order-of-magnitude overestimates. Smaller SOMEs imply
1
All statistical tests reported in this article are two-tailed.
FARO, MCGILL, AND HASTIE
688
shorter time estimates, and SOMEs below zero imply underestimation of actual time.
In Study 1, 82 of 4,284 estimates (about 2%) were identified as
outliers and were excluded from further analysis (a similar proportion of outliers was found in subsequent studies and excluded
from the analysis). The order of presentation of the pairs did not
significantly affect estimates. Figure 1A shows the mean SOMEs
as a function of pair type and explanation. An ANOVA revealed a
main effect of pair type on time estimates, F(2, 230) ⫽ 79.67, p ⬍
.05. We did not observe a main effect of explanation manipulation,
F(2, 115) ⫽ 1.90, p ⬎ .15. The interaction of pair type and
explanation was significant, F(4, 230) ⫽ 5.12, p ⬍ .05. For the
causal-pairs, time estimates in the causality-explanation condition
were shorter than were those in the no-explanation condition, F(1,
115) ⫽ 6.86, p ⬍ .05. The same was true for the same-domain
pairs, F(1, 115) ⫽ 4.31, p ⬍ .05. By contrast, for the differentdomain pairs, time estimates in the causality-explanation condition
were not different from those in the no-explanation condition
Time Estimates (SOME)
A
0
-0.1
-0.2
-0.3
No explanation
Similarity explanation
Causality explanation
-0.4
Different-domain
Same-domain
Causal
Pair Type
B
90
80
Correct Recall %
70
60
50
40
Causality explanation
Similarity explanation
No explanation
30
(F ⬍ 1). The similarity-explanation task did not significantly affect time estimates for any pair type. The compression effect is
therefore replicated— greater perceived causality results in shorter
time estimates. Greater perceived similarity does not appear to
result in shorter time estimates.
The effects across the different pair types and explanation
treatments are the main focus of this study. However, the data also
show underestimation relative to actual time. For instance, back
transforming the mean SOME for the causal pairs in the noexplanation condition yields a (geometric) mean of .58, suggesting
participants are underestimating time for these pairs by about 40%.
The underestimation is greater on causal explanation.
Cued recall. We next examined the effects of pair type and
explanation on strength of mnemonic association as measured by
cued recall. We calculated the proportion of items recalled correctly
by each participant for each pair type in each of the explanation
conditions. Figure 1B shows the mean correct recall proportions as a
function of pair type and explanation. An ANOVA with the correct
recall proportion as the dependent variable indicated a main effect of
pair type, F(2, 230) ⫽ 143.15, p ⬍ .005. The analysis also revealed
a main effect of explanation, F(2, 115) ⫽ 11.29, p ⬍ .005. The
interaction of pair type and explanation, driven by the greater impact
of the explanation tasks on the different-domain pairs, was also
significant, F(4, 230) ⫽ 11.50, p ⬍ .005. However, there were no
significant differences in cued recall between the causality- and
similarity-explanation conditions for any pair type. Essentially, both
explanation tasks resulted in the same increase in cued recall, but as
noted earlier, only the causal explanation had an effect on time
estimates.
Correlations between time estimates and causality ratings.
For each participant, we calculated the correlation between time
estimates and causality ratings across the 36 pairs judged. To
examine the effect of the explanation manipulation, these correlations were entered to an ANOVA after a Fisher transformation (the
reported correlations are back-transformed). The mean correlation
was stronger and more negative in the causality-explanation condition than in the no-explanation condition (␳causality ⫽ ⫺.19;
␳no-explanation ⫽ ⫺.12), F(1, 77) ⫽ 4.84, p ⬍ .05.
Correlations between estimates and actual time differences.
The SOME measure reflects over- and underestimation relative to
actual time. The correlations between time estimates and actual
elapsed time, calculated within participants and across the 36 pairs,
reflect instead whether participants were able to distinguish between far-apart and close-together pairs of events. Overall, these
correlations were weak but positive. Hence, participants tended to
respond with relatively short time estimates when actual elapsed
time is short and relatively long time estimates when actual
elapsed time is long. An ANOVA on the Fisher-transformed
correlations revealed no main effect of explanation (␳causality ⫽
.13; ␳similarity ⫽ .16; ␳no explanation ⫽ .16; F ⬍ 1).
Discussion
20
Different-domain
Same-domain
Causal
Pair Type
Figure 1. A: Mean time estimates (SOME) and standard errors as a
function of pair type and explanation in Study 1. B: Mean correct recall
proportion as a function of pair type and explanation in Study 1. SOME
(signed-order-of-magnitude-error) ⫽ log10 (estimate/actual).
Study 1 replicates the time-compression effect. Greater perceived
causality resulted in shorter time estimates. There was also a negative
correlation between time estimates and causality ratings within participants, and it was stronger and more negative when participants
thought about causality. The main purpose of Study 1 was to examine
whether perceived causality truly underlies the shorter time estimates
CAUSAL FORCE AND COMPRESSION OF ELAPSED TIME
Study 2: Priming Physical Versus Biological Causal
Processes
Method
Participants. For $3 each, 74 University of Chicago students
participated in a 15-min experiment on general knowledge.
Procedure and design. The design was a 2 (causal process
primed: physics versus biology, manipulated between subjects) ⫻
2 (pair type: causal versus noncausal, manipulated between subjects) design.
Causal process priming. Prior to making any time estimates,
participants completed a physics or biology knowledge test. As
part of this task, participants were asked to elaborate on the causal
mechanisms of three physical phenomena or three biological phenomena. In the physics-prime condition, participants were asked to
provide a brief, written explanation of how a baseball that is
pitched goes over a fence, how a rock thrown into water capsizes
a toy boat, and how a driver that hits the brakes of a car on icy road
hits a tree. In the biology-prime condition, participants were asked
to provide a brief, written explanation of how a caterpillar turns
into a butterfly, how a person who starts smoking contracts lung
disease, and how a tree produces fruit. The purpose of this manipulation was to make physical or biological causality salient in
participants’ minds prior to making time estimates.
Pair type. After providing written explanations for the three
physical or biological phenomena, participants took part in a task
on history knowledge. As part of this task, they were asked to
judge the time between pairs of historical events. Eight of the
causal pairs used in Study 1 were used in this study. We also
include noncausal pairs in the design (the same-domain pairs from
Study 1). The purpose here is to ascertain that our manipulation
triggers different types of causal processes rather than merely
short- versus long-interval magnitudes on which participants might
anchor. If the effects on time estimates are driven by anchoring,
our manipulation should have an effect on both types of pairs. That
is, if considering physical versus biological processes triggers
short versus long time intervals, then this effect should persist
when judging any time intervals. If, however, the manipulation
triggers different ways of thinking about the causal processes, it
should have an effect only on the causal pairs. Two presentation
orders for the event pairs were used.
Dependent measures. As in Study 1, participants estimated
the number of years that elapsed between the events for each pair.
After participants completed time estimation task for all pairs, they
were asked to rate the strength of the causal relation between the
events in each pair (on a 0 –100 scale).
Results
Time estimates. The order of presentation did not significantly affect time estimates. As in Study 1, the analysis was
conducted on the SOME measure. Figure 2 shows the mean SOME
as a function of pair type and causal process. The ANOVA showed
a main effect of pair type, F(1, 70) ⫽ 79.87, p ⬍ .001. There was
also a significant main effect of causal process, F(1, 70) ⫽ 4.69,
p ⬍ .05. These effects were qualified by a marginally significant
interaction of pair type and causal process, F(1, 70) ⫽ 3.65, p ⫽
.06. For the causal pairs, time judgments in the physics-prime
condition were significantly shorter than were those in the biologyprime condition, F(1, 70) ⫽ 8.86, p ⬍ .01. The causal process
0.2
0.1
Time Estimates (SOME)
or whether time compression is the result of similarity or associations
between the to-be-judged events. The results show that perceived
similarity cannot account for the shorter time estimates. The
similarity-explanation task increased perceived similarity but did not
affect time estimates. In contrast, however, the causality-explanation
task did shorten time estimates, and this effect occurred only for pairs
for which a plausible causal story could be constructed, the causal and
same-domain pairs.
The results of the cued-recall task also argue against the mnemonic
association account. Both explanation tasks increased association
strength as measured by cued recall. However, strengthened association did not result in shorter time estimates. In fact, the pattern of
results for time estimates and cued recall show a dissociation of these
two variables (see Figure 1). Although the causal-explanation task
increased association strength for all pair types, it shortened time
estimates only for the causal and same-domain pairs, those that
participants could causally link. Strengthening association for pairs
that could not be causally linked did not shorten time estimates.
Taken together, the results of Study 1 strengthen the claim that
inferring a causal link is necessary to produce the time-compression
effect. We next delve more deeply into what about perceived causality
results in time compression. In the following studies, we manipulate
the specific contents of participants’ causal mechanism beliefs and
measure time estimates. In Study 2, we prime two core domains in
causal cognition, mechanical–physics, and biology, prior to a time
estimation task for historical events (Inagaki & Hatano, 2004). We
expect that being primed with mechanical–physical processes will
evoke thoughts about causes that, like kinetic force, dissipate over
time and result in shorter time estimates. In contrast, we expect that
being primed with biological processes will trigger a view of causes
that build up over time and result in longer time estimates.
689
0
-0.1
-0.2
Biology-prime
Physics-prime
-0.3
-0.4
Non-causal
Causal
Pair Type
Figure 2. Mean time estimates and standard errors (SOME) as a function
of pair type and causal process prime in Study 2. SOME (signed-order-ofmagnitude-error) ⫽ log10 (estimate/actual).
690
FARO, MCGILL, AND HASTIE
manipulation did not affect time estimates for the noncausal pairs
(F ⬍ 1).
Correlations between time estimates and causality ratings.
As in Study 1, we calculated the correlation between time estimates and causality ratings across the pairs judged by each participant. To examine the effect of the manipulations, we entered
these correlations into an ANOVA after a Fisher transformation.
This analysis yielded a marginally significant interaction effect of
pair type and causal process, F(1, 70) ⫽ 3.28, p ⫽ .07. For the
causal pairs, the mean correlation between estimates and causality
ratings was stronger and more negative in the physics condition
than in the biology condition (␳physcis ⫽ ⫺.34; ␳biology ⫽ ⫺.02),
F(1, 70) ⫽ 4.52, p ⬍ .05. The difference between the mean
correlations for the participants judging the noncausal pairs was
not significant (␳physcis ⫽ ⫺.15; ␳biology ⫽ ⫺.23; F ⬍ 1).
Correlations between estimates and actual time differences.
As before, we supplement the SOME results with the sensitivity
correlations. An ANOVA on the Fisher-transformed correlations
between estimates and actual time differences revealed main effects of pair type and causal process. The mean correlation between estimates and actual elapsed time was larger for the noncausal pairs than for the causal pairs (␳noncausal ⫽ .52; ␳causal ⫽
.08), F(1, 70) ⫽ 30.15, p ⬍ .01. Hence, estimates of participants
judging noncausal pairs were more sensitive to actual time than
were estimates of participants judging causal pairs. In addition,
participants exposed to physical processes exhibited greater sensitivity than did those exposed to biological processes (␳physics ⫽
.38; ␳biology ⫽ .23), F(1, 70) ⫽ 4.45, p ⬍ .05.
Discussion
In Study 2, we manipulated dissipative and accumulating causal
mechanisms by priming either physical or biological causal processes
prior to the time judgment task. The results indicate that when physical causal processes are salient to participants, time estimates for
subsequently judged causal intervals are shorter than when biological
processes are salient. Hence, time estimates for causal time intervals
depend on the type of causal process that is salient during judgment.
When we ran this study, we did not include a control condition in
which no causal process was primed. However, in Study 1, participants from the same university population made estimates for the
same eight causal pairs used in this study without being exposed to
any prime or to any manipulation. Comparing the mean estimates of
these participants (Mno prime ⫽ ⫺.24) with the mean estimates in
physics (Mphysics ⫽ ⫺.28) and biology (Mbiology ⫽ ⫺.14) conditions
shows that the biology prime resulted in significantly longer estimates, F(1, 76) ⫽ 4.07, p ⬍ .05. This supports our prediction that
being primed with causality can result in longer estimates of time
relative to control, providing the causal process primed is accumulative. The physics prime resulted in estimates that were statistically not
different from the control ( p ⬎ .30).
In Study 1, the mean within-participants correlations between
time estimates and causality judgments were negative across conditions, and additional focus on (nonspecific) causality resulted in
a stronger negative correlation. The correlations in Study 2 were
also negative overall. However, as predicted, being primed with
biological causal processes dampened the negative relation between estimates and causality ratings.
Finally, the manipulation of causal process type had an effect on
time estimates only for the causal pairs. This manipulation did not
have an effect on the time estimates for the noncausal pairs, suggesting that our manipulation did not prime short or long time intervals on
which participants anchored in the time estimation task.
In Study 2, we relied on participants’ preexperimental beliefs about
the progression of causal processes in two core scientific domains. In
Study 3, we manipulate participants’ beliefs concerning social causal
processes (Brickman, Ryan, & Wortman, 1975). We suggest that the
same pair of events may be linked through different aspects of the
people involved in the events. For instance, the Russians’ launch of
Sputnik may be linked to American astronauts landing on the moon
by evoking the various emotions experienced by the people involved.
Emotions such as joy, pride, and fear may be relevant in this case. On
the other hand, these same events could be linked by evoking the traits
associated with the people involved, perhaps determination, stubbornness, and courage in this case. We expect that relative to traits,
emotions would tend to be viewed as dissipating over time (an
assumption we verify with collateral ratings). Thus, if participants
believe emotions are responsible for the causal relation, they would
judge that the events occurred relatively close in time. However,
evoking trait explanations is instead expected to result in a temporally
stable view of causality because traits are typically perceived to be
relatively stable over people’s life spans (Ross, 1989). Therefore,
considering traits as responsible for the causal link should result in
longer time estimates.
Although this is our main prediction, we note that emotions, and
perhaps traits, may in some situations be associated with different
time profiles (Gilbert et al., 1998; Igou, 2004; Labroo & Mukhopadhyay, 2009; Ross, 1989). There may also be individual differences in people’s implicit theories about the progression of these
factors over time (Ross, 1989). To the extent that there is such
variation, it should be reflected in time estimates.
Study 3: Manipulating Emotion and Trait Causal
Processes
Method
Participants. In return for course credit, 121 University of
Michigan students participated in a 15-min experiment on judgment and decision making.
Procedure and design. One factor, causal process type, was
manipulated between subjects with three levels: actor emotion,
actor trait, and control.
Participants estimated time between each of 12 causally related
pairs of historical events, the same as those used in Study 1. In the first
stage of the task, participants in the actor-emotion condition were
asked to designate emotions associated with actors in the events that
would help explain how the first events could be causally related to
the second events. The list of emotions provided included pride,
passion, joy, hope, envy, anger, guilt, and terror. Participants in the
actor-trait condition were asked to choose trait characteristics associated with actors in the events that would help explain how the first
event could be causally related to the second. The list of traits
provided included courage, aggressiveness, charisma, arrogance, loyalty, open-mindedness, selfishness, and shortsightedness. Because the
historical events potentially involved multiple actors, participants
were allowed to select more than one emotion or one trait from the
CAUSAL FORCE AND COMPRESSION OF ELAPSED TIME
lists. In the control condition, participants did not make any explicit
choices related to causal relations.
Dependent measures. After each emotion or trait selection,
participants made a time estimate for that pair. They were free to
use a time unit of their choice. Finally, after the choice task and
time estimate, participants were asked to judge the degree to which
they thought the two events were causally related (on a 0 –100
scale). These tasks were repeated in this order for each of the 12
pairs. Two presentation orders for the event pairs were used.
To measure the beliefs participants held about emotions or traits
and the progression of their impact over time, we asked them to
rate the degree to which each emotion or trait in the lists dissipated,
was stable, or accumulated in force over time. Participants made
the ratings on ⫺3 (⫽ decreases) to ⫹3 (⫽ increases) scale, with
0 (⫽ stable) as the midpoint. These ratings were collected after all
choice tasks, time estimates, and causality ratings were completed.
Results
Time estimates. As before, the order of presentation did not
significantly affect time estimates. An ANOVA on the SOME measure yielded a main effect of causal process, F(2, 118) ⫽ 8.46, p ⬍
.05. Explanations involving actor emotions resulted in the shortest
time estimates, followed by the control condition and then actor-trait
explanations (Table 1). As predicted, time estimates in the actoremotion condition were significantly shorter than were those in
the actor-trait condition, F(1, 118) ⫽ 8.20, p ⬍ .005. Estimates in
the actor-emotion condition were also significantly shorter than
were those in the control condition, F(1, 118) ⫽ 5.47, p ⬍ .05.
Time estimates in the actor-trait were marginally longer than were
those in the control condition, F(1, 118) ⫽ 3.07, p ⫽ .08.
For each participant in the actor-emotion condition, we constructed an emotion-trend score by summing up his or her ratings
of the eight emotions. We did the same for the eight traits for
participants in the actor-trait condition. As expected, trend scores
were smaller in the actor-emotion condition than in the actor-trait
condition (Memotion ⫽ ⫺.02; Mtrait ⫽ 5.37; t ⫽ ⫺3.99, p ⬍ .01).
Emotion-trend scores were positively correlated with time estimates (␳ ⫽ .33, p ⬍ .05). Hence, viewing the impact of emotions
as dissipating over time was associated with shorter time estimates.
The trait-trend scores were not related to the time estimates in the
actor-trait condition (␳ ⫽ ⫺.18, p ⬎ .25).
Correlations between time estimates and causality ratings.
As before, we calculated the correlations between estimates and
causality ratings across the 12 pairs judged by each participant and
used the Fisher-transformed values in the analysis. There was not
a significant main effect of causal process (␳emotion ⫽ ⫺.35;
␳trait ⫽ ⫺.39; ␳control ⫽ ⫺.46), F(2, 118) ⫽ 1.20, ns. As in the
Table 1
Mean Time Estimates (SOME) as a Function of Causal Process
Type (Study 3)
Emotion
Trait
Control
Causal process
M
SD
M
SD
M
SD
Time estimates
⫺0.97
0.40
⫺0.63
0.33
⫺0.78
0.35
Note.
SOME ⫽ signed-order-of-magnitude-error.
691
previous two studies, the mean within-participants correlations
between time estimates and causality ratings were negative in all
conditions. Somewhat counter to our initial intuitions, the mean
correlation was weakest in the actor-emotion condition. This, we
suggest, may be due to the different lay beliefs people may hold
about the impact of emotions over time (Igou, 2004; Labroo &
Mukhopadhyay, 2009; Mukhopadhyay & Johar, 2005). Further
examination showed although emotion-trend scores were on average smaller than trait-trend scores, there was greater variability
in emotion-trend scores and the maximum and minimum values reached both ends of the scale (Minimumemotion ⫽ ⫺15;
Maximumemotion ⫽ 12; Minimumtrait ⫽ ⫺7; Maximumtrait ⫽ 16).
If participants hold different, sometimes opposing views about
the manner in which emotions impact over time, the correlation
between time estimates and causality may, too, depend on
emotion-trend scores. For participants who viewed the impact of
emotions as dissipating over time, the correlation between time
and causality should tend to be more negative. In contrast, for
participants who view the impact of an emotion as accumulating
over time, correlations should tend to be positive or not as negative. Supporting this account, in the actor-emotion condition there
was a significant, positive correlation between emotion-trend
scores and the Fisher-transformed correlations between time and
causality (␳ ⫽ .31, p ⬍ .05). Trait-trend scores in the actor-trait
condition were not significantly related to the correlations between
estimates and causality ratings (␳ ⫽ ⫺.10, p ⬎ .50).
Correlations between estimates and actual time differences.
An ANOVA on the Fisher-transformed correlations between estimates and actual time differences did not reveal a significant effect
of the causal process manipulation (␳emotion ⫽ ⫺.07; ␳trait ⫽ ⫺.06;
␳control ⫽ ⫺.02), F(2, 118) ⬍ 1, ns.
Discussion
The results of this study show considering different types of
social-causal processes can affect estimates of elapsed time between two events. Focusing on actor emotions to explain the causal
relation resulted in shorter time estimates than focusing on actor
traits. Relative to a control condition, where no specific type of
causal process was evoked, focusing on dissipative causal forces
(emotions) resulted in compression. Focusing on alternative causal
mechanisms, stable ones (traits) in this case, resulted in marginally
longer time estimates. The positive correlation between emotiontrend scores and time estimates in the actor-emotion condition
suggests that time estimates are driven by the theory people hold
about the specific causal process underlying that relation.
The within-persons negative correlation between time estimates
and causality ratings, observed in Studies 1 and 2, obtained in this
study as well. In Study 2, the negative correlation between time
and causality was significantly weaker when biological processes
were primed. Similarly, in this study, for participants who tended
to view emotions as less dissipative (or perhaps accumulative) in
impact over time, the correlations between time and causality were
less negative. Consequently, there was a positive correlation between emotion-trend scores and the within-persons correlations
between time and causality. Thus, the relation between time and
causality depends on the theory people hold about the specific
causal process underlying that relation. Unlike in Study 2, we did
not find that making alternative causes, stable in this case, more
692
FARO, MCGILL, AND HASTIE
salient significantly dampens the negative correlation between
time and causality. We suspect that rating perceived causality after
each time estimate may have cued participants to the default
negative relation between these two variables.
To manipulate dissipative, accumulative, and stable processes, in
Studies 2 and 3 we relied on participants’ beliefs about causes in
different scientific domains and about different social explanations.
By changing domains, or types of causal explanation, our manipulations might have evoked not only different trajectories of causal force,
as intended, but also different interval lengths. Perhaps events from
physics or events explained by emotions typically take a short time to
transpire, and anchoring on such short intervals resulted in shorter
time estimates relative to the other conditions. The results for the
noncausal pairs in Study 2 argued against this possibility. However, a
more direct control for this explanation would be to manipulate the
salient causal process while holding domain constant. In Study 4,
prior to making time estimates, we expose participants to a causal
event from a given domain. To manipulate the salient trajectory of
causal force, we use an event that could have dissipating, stable, or
accumulating impact over time.
Study 4: Manipulating Causal Processes Holding
Domain Constant
Method
Participants. Eighty-five University of London students participated in a 15-min experiment in return for the equivalent of
U.S.$6 each.
Procedure and design. One factor, causal process type, was
manipulated between participants and had four levels, dissipating,
accumulating, stable, and control.
Filler tasks. Prior to the focal time estimation task, participants took part in a sequence of filler analogical reasoning tasks on
a computer. In one of the filler tasks, for example, they saw related
pairs of words (e.g., poverty: money) followed by five other pairs
of words (wealth: gold; hunger: food; car: driver; cook: stove;
butcher: knife). They were asked to find the pair that best expresses a relation similar to that expressed in the original pair. In
a second filler task, for instance, they saw an argument (e.g., All
German cars are safe. Dale drives a German car, so his car is safe.).
They were then asked to choose from a list of five arguments the
one that most closely resembles the logic of the original argument.
The filler tasks were identical across conditions.
Causal process type. In the focal task, participants were asked
to identify analogous elements between two events. The first event
described a manager of a company who sent samples to key
customers. Participants were then presented with information
about the impact of this action on brand awareness (Zauberman,
Levav, Diehl, & Bhargave, 2010). In the dissipating-cause condition, a figure showed the impact of sending the samples on
customer awareness was strong at the outset but diminished over
time. In the accumulating-cause condition, the figure showed the
impact was weak at the outset but built up over time. In the stable
condition, the figure showed a relatively constant impact. Finally,
in the control condition, participants only read that this action
affected brand awareness but were not shown any figure (see
Appendix B).
Next, participants were presented with an historical event on the
computer screen (Russians launch the first satellite that circles the
earth [Sputnik]). For this event, they were asked to indicate which of
four options is analogous to customers’ brand awareness in relation to
the previous event. The four options included one that could plausibly be affected by the historical event (Russians’ determination
to make progress in social sciences; Americans’ determination to
make progress in the space race; Americans’ determination to
make progress in social sciences; Russians’ determination to make
progress in agricultural sciences.).
In the initial scenario, the effect on brand awareness was triggered
by sending the samples, and this effect followed a certain trajectory
over time. In the next stage, when asked which factor is analogous to
brand awareness, participants were therefore implicitly asked to
choose a factor that could be triggered by the historical event presented to them. Thinking about an analogous causal mechanism for
the historical event was expected to evoke the specific type of causal
process and trajectory previously considered.
Dependent measures. After making their choice, participants
were presented with a second historical event, the designated effect
of the first historical event (American astronauts landing on the
moon). They were asked to estimate the amount of time that
elapsed between the first historical event and the historical event
presented after the choice task. They provided time estimates in a
time unit of their choice. Then they were asked to rate the extent
to which they thought the two historical events were causally
related (on a 0 –100 scale). Participants went through these steps
for a second pair of historical events (“The U.S. government health
authorities determine that smoking is hazardous to health” and “In
a move coined ‘Marlboro Friday,’ the tobacco company Philip
Morris announces plans to cut the price of its flagship Marlboro
brand.”). Two orders of event presentation were used. Finally,
participants were asked to write a few sentences about why they
thought the options they chose for the first historical events were
analogous to customers’ brand awareness and why they thought
they might follow a similar temporal trajectory.
Results
Time estimates. The order of presentation of the two events
did not significantly affect time estimates. The mean time estimates (SOME) for each of the two items are given on Table 2. For
the Sputnik pair, contrasts indicated participants in the dissipatingcause condition gave significantly shorter time estimates than did
those in the accumulating-cause condition, F(1, 81) ⫽ 4.54 p ⬍
.05, and marginally shorter estimates than those in the stable-cause
condition, F(1, 81) ⫽ 3.69, p ⬍ .06. Estimates in the control
condition were marginally shorter than were those in the
accumulating-cause condition, F(1, 81) ⫽ 3.25, p ⬍ .08, and not
different from those in the dissipating-cause and stable-cause conditions. We observed a similar pattern for the Marlboro pair:
Contrasts indicated participants in the dissipating-cause condition
gave significantly shorter time estimates than did those in the
accumulating-cause condition, F(1, 81) ⫽ 5.07, p ⬍ .05; estimates
in the dissipating-cause condition were not different from those in
the stable-cause condition. Estimates in the control condition were
significantly shorter than those in the accumulating-cause condi-
CAUSAL FORCE AND COMPRESSION OF ELAPSED TIME
693
Table 2
Mean Time Estimates (SOME) as a Function of Causal Process Type (Study 4)
Dissipating
Accumulating
Stable
Control
Causal process
M
SD
M
SD
M
SD
M
SD
Sputnik
Marlboro
⫺0.70
⫺2.14
0.49
0.69
⫺0.38
⫺1.65
0.40
0.68
⫺0.41
⫺1.94
0.49
0.79
⫺0.64
⫺2.14
0.49
0.64
Note. SOME ⫽ signed-order-of-magnitude-error.
tion, F(1, 81) ⫽ 5.63, p ⬍ .05, and not different than those in the
dissipating-cause and stable-cause conditions.2
Explanations. We next report some of the participants’ responses when asked to explain why they picked the option paralleling
customers’ brand awareness. In responding about the Sputnik event,
participants in the dissipating-cause condition wrote explanations
such as, “At first the Americans would be jealous of the Russian
success, and their determination would be great. Over time they
would become less bothered, especially if the Russians were not
making much further progress”; “Initially the shock generated by the
event is likely to make it appear as a magnified threat. Over time,
however, new events will capture the imagination of politicians and
the masses, so that the Sputnik no longer appears as an immediate
danger.”
In contrast, participants in the accumulating-cause condition
wrote, “Americans would be spurred on by the achievement of the
Russians even though initial dismay/shock/negative emotion could
possibly cause hindrance in the beginning”; “Americans might
want to keep up with the progress made by the Russians when they
launched Sputnik, thus, as time goes by, there might be a heightened determination to make progress in the ‘space race.’”
Finally, participants in the stable-cause condition wrote explanations of the following type: “More research and technological advances may increase the base of knowledge, but the determination to
make progress will likely remain constant”; “Once they have entered
the space race, their determination to remain at par with the counterpart is not going to diminish over a period of time and, therefore, can
be assumed to be constant.” (See Appendix C for other excerpts.)
Discussion
In this study, we used a manipulation of causal-process type
holding domain constant. The results show again that thinking of
dissipating, stable, or accumulative causal processes during judgment affects participants’ time estimates. In Studies 2, 3, and 4, we
manipulated the salience of alternative causal-process types and
measured the effects of these manipulations on time estimates. In
the next study, we put some participants under high cognitive load
while making time estimates. If, as we suggest, dissipating causes
are the default and drive binding of causal events in time and if
making alternative causal processes more salient reduces or reverses this tendency then cognitive load should exacerbate it.
Presumably, under cognitive load, participants would be more
likely to adopt the default mechanical–physical view of causation,
which is familiar and easy to reason by (Fletcher, Danilovics,
Fernandez, Peterson, & Reeder, 1986; Schlottmann, 1999). They
would be even less likely to consider alternative causal mechanisms, such as those we experimentally manipulated in Studies 2,
3, and 4. Therefore, under limited cognitive resources we should
observe even shorter time estimates for causal intervals.
Study 5: Manipulating Cognitive Load
Method
Participants. One hundred sixteen University of London students participated in a 20-min experiment for the equivalent of
$7.50 each.
Procedure and design. The design was a 2 (pair type: causal
versus noncausal, manipulated within subjects) ⫻ 2 (cognitive load:
high versus low, manipulated between subjects) mixed design.
Participants were told that social and technological changes
have led to increased levels of multitasking (e.g., text messaging
while watching TV, driving while talking on the phone, etc.),
making it important to study how people concentrate on multiple
simultaneous tasks. They were told this study, presented on a
computer, involved memorization and multitasking. Participants
were asked to estimate in years the elapsed time between 24 pairs
of historical events while memorizing strings of letters. Half of the
events were the causal pairs, and the other half were noncausal.
The latter were included in the design to test whether cognitive
load may affect time estimates in general rather than moderate the
time-compression effect, as we suggest. Two presentation orders
for the event pairs were used.
Half the participants completed the time estimation task
while holding two letters in memory (low load), whereas the
other half completed time estimation while holding a string of
eight letters in memory (high load; Gilbert & Hixon, 1991; Shiv
& Fedorikhin, 1999). Participants were asked to memorize
strings of letters, rather than numbers, to avoid numerical
anchoring effects. Each time a participant completed six estimates, he or she was asked to report the string of letters and was
given a new string of letters to memorize. After participants
completed the time estimation task for all 24 pairs, they were
asked to rate the strength of the causal relation between the
2
Because participants in this study made judgments for only two event
pairs, we could not calculate any within-participants correlations between
time estimates and causality ratings. We examined the across-participants
correlations. They reveal patterns largely consistent with our predictions.
For the Sputnik pair, the correlations were ␳dissipating ⫽ ⫺.46, p ⬍ .05;
␳accumulating ⫽ .41, p ⬍ .06; ␳stable ⫽ ⫺.07, ns; ␳control ⫽ ⫺.49, p ⬍ .05. For
the Marlboro pair, these were ␳dissipating ⫽ ⫺.44, p ⬍ .05; ␳accumulating ⫽
.51, p ⬍ .05; ␳stable ⫽ ⫺.30, ns; ␳control ⫽ ⫺.16, ns. However, we suggest
that the within-persons correlations between time estimates and causality
ratings are easier to interpret.
FARO, MCGILL, AND HASTIE
694
events in each pair (on a 0 –100 scale). This task was not subject
to a manipulation of cognitive load.
Results and Discussion
Time estimates. Order of presentation did not affect time
estimates. Figure 3 presents the mean time estimates (SOME) as a
function of pair type and cognitive load. The data were analyzed in
an ANOVA with pair type as a within-subjects factor and cognitive load as a between-subjects factor. This analysis revealed a
main effect of pair type, F(1, 114) ⫽ 50.93, p ⬍ .005. The main
effect for cognitive load was not significant F(1, 114) ⫽ 1.76, p ⬎
.20. There was a significant interaction of pair type and cognitive
load, F(1, 114) ⫽ 4.28, p ⬍ .05. For the causal pairs, high-load
participants gave significantly shorter time estimates than did
low-load participants, F(1, 114) ⫽ 4.45, p ⬍ .05. The difference
was not significant for the noncausal pairs (F ⬍ 1). Hence, as
predicted, time estimates were even shorter for causal processes
when participants were under cognitive load. The effect of cognitive load was seen only for the causal pairs, suggesting that the
manipulation affects causality and how it interacts with time,
rather than general estimation of time.
Correlations between time estimates and causality ratings.
As before, we calculated the within-subjects correlation between
time estimates and causality ratings across the 24 pairs. The main
effect of cognitive load on the Fisher-transformed correlations was
not significant (␳high-load ⫽ ⫺.22; ␳low-load ⫽ ⫺.26; F ⬍ 1).
Hence, contrary to expectations, the correlation between estimates
and causality ratings was not affected by cognitive load.
Correlations between estimates and actual time differences.
An ANOVA on the Fisher-transformed correlations between estimates and actual time differences did not reveal a reliable effect of
cognitive load. (␳high-load ⫽ .16; ␳low-load ⫽ .15; F ⬍ 1).
General Discussion
Summary and Related Issues
We started the article by asking about the time elapsed between
events such as the launch of an economic program and the jump
Time Estimates (SOME)
0.05
-0.05
-0.15
Low-load
High-load
-0.25
Non-causal
Causal
Pair Type
Figure 3. Mean time estimates and standard errors (SOME) as a function
of pair type and cognitive load in Study 5. SOME (signed-order-ofmagnitude-error) ⫽ log10 (estimate/actual).
start of an economy in recession and events such as giving a child
a tough lecture about schoolwork and seeing an effect on his or her
performance. People tend to give shorter estimates of the time
between such events than if they had not perceived a causal link
between them. Their perception of causality binds the events
together in time. However, the degree of temporal binding appears
to be conditioned by the specific type of causal forces they believe
are connecting the two events. When people view causation
through a mechanical–physical lens, where causes are believed to
have dissipative force, the cause and effect are “pulled together.”
However, when they imagine the two events are connected by
accumulative forces such as those in the biological world, the
binding is reduced and the events may be pushed apart.
Study 1 replicates the compression effect and rules out two alternative accounts, similarity and mnemonic associations between two
events. The results support the interpretation that causality is the
effective relation producing compression in time estimates. Subsequent studies show elapsed time estimates for causally related events
depend on beliefs about the specific types of causal mechanisms
linking these events. In every study, priming dissipative causes results
in judgments of time similar to control or compressed relative to
control. In contrast, in Study 2, a control, no-mechanism-primed
condition was not included in the experimental design, but comparisons with Study 1, which had identical to-be-judged materials and
sampling from the same participant population, suggest priming biological processes increases interval estimates compared with a control. In Study 3, when we primed traits as the type of causal link
between the to-be-judged events, time estimates are again (marginally) longer than in the control condition. In Study 4, when we primed
an accumulating process, we again obtain longer time estimates than
in the control condition. Hence, there are repeated observations of
differences between putative accumulating or stable processes and a
control condition, suggesting that evoking these alternative processes
can expand time estimates. These findings support the interpretation
that the availability of alternate causal processes and the specific
contents of these processes influence the impact of causation on time
estimates.
Across the studies, the within-persons correlations between time
estimates and causality ratings tended to be negative. However,
these negative correlations were damped when accumulative
causes were made salient. One could argue that positive correlations would be expected between time and causality when accumulative causes are salient. We agree and suspect that overriding
the default, negative relation between these two variables may
have been hindered by the two specific aspects of our studies: The
judgment of a large number of events pairs consecutively and the
ratings of causality sometime right after time estimates. Both of
these factors might have cued participants into the default and
familiar (negative) relation between these variables.
Longer time estimates following the priming of nondissipative
causes and the muted negative association between time and causality
reflect the flexibility of our cognitive system to imagine mechanisms
with different temporal relations between cause and effect (Ahn,
Kalish, Medin, & Gelman, 1995; Buehner & May, 2003; Hagmayer
& Waldmann, 2002). Nonetheless, our results consistently demonstrate the dominance of mechanical–physical causality and the notion
of dissipative causes. The default effect of perceived causality is to
bind events in time. First, causal pairs are systematically judged to be
closer in time than noncausal ones. Second, considering nonspecific
CAUSAL FORCE AND COMPRESSION OF ELAPSED TIME
causal relations results in shorter time estimates and in stronger
negative time-causality correlations; the same happens when priming
dissipative causes. Finally, under cognitive load, which presumably
leads people to opt for default, familiar notions of causality, time
estimates for cause– effect events are even shorter. Overall, these
results show beliefs about causal mechanisms are the underlying
mediator of the compression effect, and the dominant belief is consistent with mechanical–physical causation.
Our explanation for the temporal binding effect concerns perceptions of causal forces and their trajectory over time rather than
perceptions of short or long delay. In Studies 2 and 4, we addressed the possibility that our manipulations evoked concepts of
short and long delay rather than dissipative, stable, and accumulative forces, as we intended. In Study 3, measures of trajectory
over time predicted time estimates and the within-persons correlations between time and causality. The notions of force over time
and delay are related (Sterman, 2002). Accumulative processes
often imply hidden processes in which there may be no visible
effect of the causal process until after a long delay. Temporal
contiguity and dissipation of force are often experienced jointly;
when one billiard ball hits another, cause and effect are in close
proximity, but the observer also witnesses dissipative force as the
second billiard ball comes to a halt (Hubbard et al., 2001). Nonetheless, when people think about causal relations, as our participants did in the studies, notions of mechanism naturally arise (Ahn
et al., 1995). We believe causal mechanisms with different force
trajectories drive the time estimates we observed in our studies.
To trigger different causal mechanisms, we have relied on the
extensive literature supporting the claim that people develop theories about different causal mechanisms in specific domains of
knowledge (Inagaki & Hatano, 2004; Tenenbaum, Griffiths, &
Kemp, 2006). The effects of specific mental models of causation
have been demonstrated in studies of causal induction (Hagmayer
& Waldmann, 2007; Tenenbaum et al., 2006; Waldmann,
Holyoak, & Fratianne, 1995; Waldmann, 1996) and in category
learning and induction tasks (Danks, 2005; Rehder, 2003; Rehder
& Hastie, 2001). Our research is thus related to other phenomena
that involve relying on causal mechanisms to make judgments
such as frequency (Kahneman, 1982), choice (Sloman & Hagmayer, 2006), and legal decisions (Hastie & Pennington, 2000;
Pennington & Hastie, 1991).
Implications
The tendency to bind causes and effect in time and, as a result,
underestimate the intervening time can affect decisions that depend on
assessment of time-to-effect. As we depicted back in the opening
example, people might be prone to underestimate how long lecturing
a child about schoolwork takes to have an effect on his or her
performance, and as study 5 suggests, they might do so, especially if
they are busy thinking about something else. As a result, people might
delay talking to the child until it is potentially too late and, in the
absence of rapid results, might repeat the same action or resort to
another one without allowing sufficient time for the original one
to have an effect. In a similar vein, people may expect politicians to
deliver quick solutions to the economic downturn, feel ready to take
part in a challenging sport competition only a short time after exercising in the gym, or lose patience in the ability of their children’s
teachers to improve reading skills—in part because they think past
695
instances must have taken only a short time. This research may thus
suggests another reason for impatience and lack of persistence, one
that is rooted in people’s memory for elapsed time between past
actions and outcomes (see also Buehler et al., 1994; Kivetz, Urminsky, & Zheng, 2006; Morwitz, 1997; Roy et al., 2005; Vohs &
Schmeichel, 2003; Zauberman & Lynch, 2005).
The tendency to bind causes and effect in time may also reinforce
the role temporal proximity plays in causal cognition (Einhorn &
Hogarth, 1986). As noted, close temporal proximity between cause
and effect is an important, if somewhat simplistic, cue for causality:
When we experience food poisoning, our default, first inclination is to
think it must be due to the last thing we ate, and when our spouse
seems upset, we tend to think it must be something we said or did
recently. There may be several reasons for the dominance of this cue
to causality. Temporally proximate causes and effects are the first we
learn. In addition, proximate causes may come to mind more easily
than more distant ones. The compression effect we examined suggests
another reason: Even when causes and effects are not very near in
time, we experience them as if they are (Haggard et al., 2002) or
remember them as if they were (Faro et al., 2005).
Marsh and Ahn (2009) made a related point in a recent article on
how people learn and interpret ambiguous causal relations. Suppose a
person tries a new energy drink and, after some time, feels invigorated. The temporal distance between these events introduces some
ambiguity. Looking back, the person might ponder the effects of the
drink but notice no change in energy because the effect was perhaps
at a large temporal distance. Similarly, the person might wonder what
caused him or her to feel so energetic but not identify the drink as a
candidate explanation, again because it was outside a plausible temporal window (Marsh & Ahn, 2009). However, if the person believes
or wants to believe that the drink gives energy, he may bind the events
in time and bring them to a subjective distance sufficiently close to
code them as cause and effect. This may occur even when there is in
fact a large temporal gap between them that potentially precludes an
actual causal relation. Marsh and Ahn (2009) illustrated this ability to
assimilate ambiguous events as causes and effects and noted that
temporal binding is one mechanism that enables such assimilation to
occur. Long elapsed time between cause and effect is typically a
limiting factor for the emergence of causal beliefs. Binding causally
related events in time allows people to form and hold causal beliefs
that might otherwise conflict with the temporal proximity cue for
causality (Einhorn & Hogarth, 1986).
The main focus of the present research is on the effects of different
naı̈ve mechanisms of causation on judgments of elapsed time. A
related question, important for practical purposes, is how accurate the
estimates are relative to actual elapsed time. Whether a primed mechanism of causation makes estimates more or less accurate depends in
part on whether they were accurate, overestimated, or underestimated
prior to any intervention. Past research on duration judgments suggests one important determinant of interval over- and underestimation
is actual interval length. Short intervals tend to be overestimated,
whereas long intervals tend to be underestimated (Loftus, Schooler,
Boone, & Kline, 1987; Roy & Christenfeld, 2008; Vierordt, 1968;
Yarmey, 1990). Another factor that may affect accuracy is response
scale. For instance, in Studies 3 and 4, many participants made
estimates in months, not years, as participants were instructed to do in
the other studies. The pattern of predictions with respect to manipulated causal mechanisms held, but participants underestimated the
actual intervals to a greater degree in those studies.
FARO, MCGILL, AND HASTIE
696
Although interval length, response scale, and other factors may
affect the accuracy of time estimates in a given setting, the present
research demonstrates underestimation is more likely when causality
is salient and when people hold stronger causal beliefs. For instance,
people are more likely to underestimate past causal intervals if they
receive evidence of covariation between the cause– effect events
(Cheng, 1997; Einhorn & Hogarth, 1986; Moore et al., 2009) or if
they do not take into account alternative causes for the effect (Kelley,
1973; McClure, 1998; Morris & Larrick, 1995; see also Chandon &
Janiszewski, 2009). Underestimation is also more likely when a
mechanical–physical view of causes is salient and when judgments
are made under cognitive load.
Our second measure of accuracy reflects the sensitivity of
estimates to actual interval length. Overall, with the exception of
Study 3, participants’ estimates are moderately correlated with the
actual to-be-judged intervals, suggesting that even though estimates are off of actual elapsed time, participants could distinguish
between the close-together and far-apart pairs. The specific causal
process available when estimates are made did not significantly
affect this measure of relative accuracy.
Conclusion
Time and causality are inextricably linked. In this article, we
studied an emerging dimension of the relation—the influence of
perceived causality on judgments of time (Buehner & Humphreys,
2009; Eagleman & Holcombe, 2002; Faro et al., 2005; Haggard et al.,
2002; Moore et al., 2009; Wohlschlager et al., 2003). Estimates of
how long causes take to have an effect depend on how people think
about causality, and the way they think about causality can markedly
change, depending on the immediate context. People’s initial and
dominant inclination is to think of a cause like a billiard ball speeding
toward another as a strong force that dissipates with time. However,
with more effort, they can think of a cause as a seed under a newly
built house, a causal force that gathers momentum and becomes
stronger over time. These are the fundamental ways people think
about causation. They develop early in life and shape how individuals
think about causation and time in other domains of their experience.
The ramifications of this habit of thought are remarkably widespread
and affect even our intuitive judgments of elapsed time.
References
Ahn, W.-K., Kalish, C. W., Medin, D. L., & Gelman, S. A. (1995). The role
of covariation versus mechanism information in causal attribution. Cognition, 54, 299 –352.
Anderson, C. A., & Lindsay, J. J. (1998). The development, perseverance,
and change of naı̈ve theories. Social Cognition, 16, 8 –30.
Anderson, J. R., & Sheu, C.-F. (1995). Causal inferences as perceptual
judgments. Memory & Cognition, 23, 510.
Andreassen, P. B. (1987). On the social psychology of the stock market:
Aggregate attributional effects and the regressiveness of prediction.
Journal of Personality and Social Psychology, 53, 490 – 496.
Brickman, P., Ryan, K., & Wortman, C. B. (1975). Causal chains: Attribution of responsibility as a function of immediate and prior causes.
Journal of Personality and Social Psychology, 32, 1060 –1067.
Brown, N. R. (2002). Real-world estimation: Estimation modes and seeding effects. In B. H. Ross (Ed.), Psychology of learning and motivation
(Vol. 41, pp. 321–360). New York, NY: Academic Press.
Brown, N. R., & Siegler, R. S. (1992). The role of availability in the
estimation of national populations. Memory & Cognition, 20, 406 – 412.
Brown, N. R., & Siegler, R. S. (1993). Metrics and mappings: A framework for understanding real-world quantitative estimation. Psychological Review, 100, 511–534.
Buehler, R., Griffin, D., & Ross, M. (1994). Exploring the “planning
fallacy”: Why people underestimate their task completion times. Journal
of Personality and Social Psychology, 67, 366 –381.
Buehner, M. J., & Humphreys, G. R. (2009). Causal binding of actions to
their Effects. Psychological Science, 20, 1221–1228.
Buehner, M. J., & May, J. (2003). Rethinking temporal contiguity and the
judgement of causality: Effects of prior knowledge, experience, and
reinforcement procedure. Quarterly Journal of Experimental Psychology: Section A, 56, 865– 890.
Chandon, E., & Janiszewski, C. (2009). The influence of causal conditional
reasoning on the acceptance of product claims. Journal of Consumer
Research, 35, 1003–1011.
Cheng, P. W. (1997). From covariation to causation: A causal power
theory. Psychological Review, 104, 367– 405.
Cronin, M. A., Gonzalez, C., & Sterman, J. D. (2009). Why don’t welleducated adults understand accumulation? A challenge to researchers,
educators, and citizens. Organizational Behavior and Human Decision
Processes, 108, 116 –130.
Danks, D. (2005). The supposed competition between theories of human
causal inference. Philosophical Psychology, 18, 259 –272.
Eagleman, D. M., & Holcombe, A. O. (2002). Causality and the perception
of time. Trends in Cognitive Sciences, 6, 323–325.
Einhorn, H. J., & Hogarth, R. M. (1986). Judging probable cause. Psychological Bulletin, 99, 3–19.
Faro, D., Leclerc, F., & Hastie, R. (2005). Perceived causality as a cue to
temporal distance. Psychological Science, 16, 673– 677.
Fletcher, G. J. O., Danilovics, P., Fernandez, G., Peterson, D., & Reeder,
G. D. (1986). Attributional complexity: An individual differences measure. Journal of Personality and Social Psychology, 51, 875– 884.
Freyd, J. J., & Finke, R. A. (1984). Representational momentum. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 10, 126 –132.
Gilbert, D. T., & Hixon, J. G. (1991). The trouble of thinking: Activation
and application of stereotypic beliefs. Journal of Personality and Social
Psychology, 60, 509 –517.
Gilbert, D. T., Pinel, E. C., Wilson, T. D., Blumberg, S. J., & Wheatley, T.
(1998). Immune neglect: A source of durability bias in affective forecasting. Journal of Personality and Social Psychology, 75, 617– 638.
Gilovich, T., & Savitsky, K. (1996). Like goes with like: The role of
representativeness in erroneous and pseudoscientific beliefs. Skeptical
Inquirer, 20, 34 – 40.
Golonka, S., & Estes, Z. (2009). Thematic relations affect similarity via
commonalities. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35, 1454 –1464.
Haggard, P., Clark, S., & Kalogeras, J. (2002). Voluntary action and
conscious awareness. Nature Neuroscience, 5, 382–385.
Hagmayer, Y., & Waldmann, M. R. (2002). How temporal assumptions
influence causal judgments. Memory & Cognition, 30, 1128.
Hagmayer, Y., & Waldmann, M. R. (2007). Inferences about unobserved
causes in human contingency learning. Quarterly Journal of Experimental Psychology, 60, 330 –355.
Hastie, R., & Pennington, N. (2000). Explanation-based decision making.
In T. Connolly, H. R. Arkes, & K. R. Hammond (Eds.), Judgment and
decision making: An interdisciplinary reader (2nd ed., pp. 212–228).
New York, NY: Cambridge University Press.
Heider, F. (1944). Social perception and phenomenal causality. Psychological Review, 51, 358 –374.
Heider, F., & Simmel, M. (1944). An experimental study of apparent
behavior. American Journal of Psychology, 57, 243–259.
Hubbard, T. L., Blessum, J. A., & Ruppel, S. E. (2001). Representational
momentum and Michotte’s (1946/1963) “launching effect” paradigm.
CAUSAL FORCE AND COMPRESSION OF ELAPSED TIME
Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 294 –301.
Hume, D. (1938). An abstract of a treatise concerning human understanding. London, England: Cambridge University Press.
Igou, E. R. (2004). Lay theories in affective forecasting: The progression
of affect. Journal of Experimental Social Psychology, 40, 528 –534.
Inagaki, K., & Hatano, G. (2004). Vitalistic causality in young children’s
naı̈ve biology. Trends in Cognitive Sciences, 8, 356 –362.
Johnson, J. T., & Drobny, J. (1985). Proximity biases in the attribution of civil
liability. Journal of Personality and Social Psychology, 48, 283–296.
Kahneman, D. T. A. (1982). The simulation heuristic. Judgment under
uncertainty: Heuristics and biases, 201–210.
Keil, F. C., Levin, D. T., Richman, B. A., & Gutheil, G. (1999). Mechanism and explanation in the development of biological thought: The case
of disease. In D. L. Medin & S. Atran (Eds.), Folkbiology (pp. 285–319).
Cambridge, MA: The MIT Press.
Kelley, H. H. (1973). The processes of causal attribution. American Psychologist, 28, 107–128.
Kivetz, R., Urminsky, O., & Zheng, Y. (2006). The goal-gradient hypothesis resurrected: Purchase acceleration, illusionary goal progress, and
customer retention. Journal of Marketing Research, 43, 39 –58.
Labroo, A. A., & Mukhopadhyay, A. (2009). Lay theories of emotion
transience and the search for happiness: A fresh perspective on affect
regulation. Journal of Consumer Reseaerch, 36, 242–254.
Lagnado, D. A., & Sloman, S. (2004). The advantage of timely intervention. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30, 856 – 876.
Lagnado, D. A., & Sloman, S. A. (2006). Time as a guide to cause. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 32, 451– 460.
Lagnado, D. A., Waldmann, M. R., Hagmayer, Y., & Sloman, S. A. (2007).
Beyond covariation: Cues to causal structure. In A. Gopnik & L. Schultz
(Eds.), Causal learning: Psychology, philosophy, and computation (pp.
154 –172). New York, NY: Oxford University Press.
LeBoeuf, R. A., & Norton, M. I. (2007, July). Effects that lead to causes:
Using an event’s outcomes to infer its causes. Paper presented at the Annual
Meeting of the Association of Consumer Research, Memphis, TN.
Leslie, A. M. (1984). Spatiotemporal continuity and the perception of
causality in infants. Perception, 1, 287–305.
Leslie, A. M. (1988). The necessity of illusion: Perception and thought in
infancy. In L. Weiskrantz (Ed.), Thought without language (pp. 185–
210). Oxford, England: Oxford University Press.
Leslie, A. M., & Keeble, S. (1987). Do six-month-old infants perceive
causality? Cognition, 25, 265–288.
Lewin, K. (1935). The conflict between Aristotelian and Galilean modes of
thought in contemporary psychology. In K. Lewin (Ed.), A dynamic
theory of personality (pp. 1– 42). New York, NY: McGraw-Hill.
Loftus, E. F., Schooler, J. W., Boone, S. M., & Kline, D. (1987). Time went
by so slowly: Overestimation of event duration by males and females.
Applied Cognitive Psychology, 1, 3–13.
Markman, K. D., & Guenther, C. L. (2007). Psychological momentum:
Intuitive physics and naı̈ve beliefs. Personality and Social Psychology
Bulletin, 33, 800 – 812.
Marsh, J. K., & Ahn, W.-K. (2009). Spontaneous assimilation of continuous
values and temporal information in causal induction. Journal of Experimental Psychology: Learning, Memory, and Cognition, 35, 334 –352.
McCloskey, M. (1983). Naı̈ve theories of motion. In D. Gentner & A. L.
Stevens (Eds.), Mental models (pp. 299 –324). Hillsdale NJ: Erlbaum.
McClure, J. (1998). Discounting causes of behavior: Are two reasons better
than one? Journal of Personality and Social Psychology, 74, 7–20.
McGill, A. L. (1989). Context effects in judgments of causation. Journal of
Personality and Social Psychology, 57, 189 –200.
Menon, G. (1993). The effects of accessibility of information in memory
on judgments of behavioral frequencies. Journal of Consumer Research,
20, 431– 440.
697
Michotte, A. (1963). The perception of causality. Oxford, England: Basic
Books.
Moore, J. W., Lagnado, D., Deal, D. C., & Haggard, P. (2009). Feelings of
control: Contingency determines experience of action. Cognition, 110,
279 –283.
Morewedge, C. K., Preston, J., & Wegner, D. M. (2007). Timescale bias in
the attribution of mind. Journal of Personality and Social Psychology,
93, 1–11.
Morris, M. W., & Larrick, R. P. (1995). When one cause casts doubt on
another: A normative analysis of discounting in causal attribution. Psychological Review, 102, 331–355.
Morris, M. W., Sheldon, O. J., Ames, D. R., & Young, M. J. (2007).
Metaphors and the market: Consequences and preconditions of agent
and object metaphors in stock market commentary. Organizational
Behavior and Human Decision Processes, 102, 174 –192.
Morwitz, V. G. (1997). It seems like only yesterday: The nature and
consequences of telescoping errors in marketing research. Journal of
Consumer Psychology, 6, 1–29.
Mukhopadhyay, A., & Johar, G. (2005). Where there is a will, is there a
way? The effects of consumers’ lay theories of self-control on setting and
keeping resolutions Journal of Consumer Research, 31, 779 –786.
Nickerson, R. (1981). Motivated retrieval from archival memory. In J. H.
Flowers (Ed.), Nebraska Symposium on Motivation 1980 (pp. 73–119).
Lincoln, NE: University of Nebraska Press.
Pennington, N., & Hastie, R. (1991). A cognitive theory of juror decision
making: The story model. Cardozo Law Review, 13, 519 –557.
Piaget, J. (1955). The child’s construction of reality. London. England:
Routledge & Kegan Paul.
Raghubir, P., & Menon, G. (2005). When and why is ease of retrieval
informative? Memory & Cognition, 33, 821– 832.
Rehder, B. (2003). Categorization as causal reasoning. Cognitive Science,
27, 709 –748.
Rehder, B., & Hastie, R. (2001). Causal knowledge and categories: The
effects of causal beliefs on categorization, induction, and similarity.
Journal of Experimental Psychology: General, 130, 323–360.
Rosengren, K. S., Gelman, S. A., Kalish, C. W., & McCormick, M. (1991).
As time goes by: Children’s early understanding of growth in animals.
Child Development, 62, 1302–1320.
Ross, M. (1989). Relation of implicit theories to the construction of
personal histories. Psychological Review, 96, 341–357.
Rottman, B. M., & Ahn, W.-K. (2009). Causal learning and tolerance and
sensitization. Psychonomic Bulletin & Review, 16, 1043–1049.
Roy, M. M., & Christenfeld, N. J. S. (2008). Effect of task length on remembered
and predicted duration. Psychonomic Bulletin & Review, 15, 202–207.
Roy, M. M., Christenfeld, N. J. S., & McKenzie, C. R. M. (2005).
Underestimating the duration of future events: Memory incorrectly used
or memory bias? Psychological Bulletin, 131, 738 –756.
Rozin, P., & Nemeroff, C. (2002). Sympathetic magical thinking: The contagion and similarity “heuristics.” In T. Gilovich, D. Griffin, & D. Kahneman
(Eds.), Heuristics and biases: The psychology of intuitive judgment (pp.
201–216). Cambridge, England: Cambridge University Press.
Schlottmann, A. (1999). Seeing it happen and knowing how it works: How
children understand the relation between perceptual causality and underlying mechanism. Developmental Psychology, 35, 303–317.
Scholl, B. J., & Tremoulet, P. D. (2000). Perceptual causality and animacy.
Trends in Cognitive Sciences, 4, 299 –310.
Shanks, D. R., Pearson, S. M., & Dickinson, A. (1989). Temporal contiguity and the judgement of causality by human subjects. The Quarterly
Journal of Experimental Psychology B: Comparative and Physiological
Psychology, 41, 139 –159.
Shiv, B., & Fedorikhin, A. (1999). Heart and mind in conflict: The
interplay of affect and cognition in consumer decision making. Journal
of Consumer Research, 26, 278 –292.
698
FARO, MCGILL, AND HASTIE
Sloman, S. A., & Hagmayer, Y. (2006). The causal psycho-logic of choice.
Trends in Cognitive Sciences, 10, 407– 412.
Sterman, J. D. (2002). All models are wrong: Reflections on becoming a
systems scientist. System Dynamics Review, 18, 501–531.
Talmy, L. (1988). Force dynamics in language and cognition. Cognitive
Science: A Multidisciplinary Journal, 12, 49 –100.
Tenenbaum, J. B., Griffiths, T. L., & Kemp, C. (2006). Theory-based
Bayesian models of inductive learning and reasoning. Trends in Cognitive Science, 10, 309 –318.
Tremoulet, P. D., & Feldman, J. (2000). Perception of animacy from the
motion of a single object. Perception, 29, 943–951.
Vierordt, K. V. (1968). Der Zeitsinn Nach Versuchen. Tubingen, Germany:
Laupp.
Vohs, K. D., & Schmeichel, B. J. (2003). Self-regulation and the extended
now: Controlling the self alters the subjective experience of time. Journal of Personality and Social Psychology, 85, 217–230.
Waldmann, M. R. (1996). Knowledge-based causal induction. In D. R.
Shanks, K. J. Holyoak, & D. L. Medin (Eds.), The psychology of
learning and motivation: Vol. 34. Causal learning (pp. 47– 88). San
Diego, CA: Academic Press.
Waldmann, M. R., Holyoak, K. J., & Fratianne, A. (1995). Causal models
and the acquisition of category structure. Journal of Experimental Psychology: General, 124, 181–206.
Wegner, D. M., & Wheatley, T. (1999). Apparent mental causation:
Sources of the experience of will. American Psychologist, 54, 480 – 492.
White, P. A. (1988a). Causal processing: Origins and development. Psychological Bulletin, 104, 36 –52.
White, P. A. (1998b). The dissipation effect: A general tendency in causal
judgments about complex physical systems. American Journal of Psychology, 111, 379 – 410.
White, P. A. (1999). The dissipation effect: A naı̈ve model of causal
interactions in complex physical systems. American Journal of Psychology, 112, 331–364.
Whittlesea, B. W. A. (1993). Illusions of familiarity. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 1235–1253.
Wohlschlager, A., Haggard, P., Gesierich, B., & Prinz, W. (2003). The
perceived onset time of self- and other-generated actions. Psychological
Science, 14, 586 –591.
Xu, J., & Schwarz, N. (2005, March). Was it long ago or unimportant?
Diverging inferences from difficulty of recall. Paper presented at the Annual
Conference of the Society for Consumer Psychology, St. Pete’s Beach, FL.
Yarmey, A. D. (1990). Accuracy and confidence of duration estimates
following questions containing marked and unmarked modifiers. Journal of Applied Social Psychology, 20, 1139 –1149.
Zauberman, G., Levav, J., Diehl, K., & Bhargave, R. (2010). 1995 feels so
close yet so far: The effect of event “markers” on the subjective feeling
of elapsed time. Psychological Science, 21, 133–139.
Zauberman, G., & Lynch, J. G., Jr. (2005). Resource slack and propensity
to discount delayed investments of time versus money. Journal of
Experimental Psychology: General, 134, 23–37.
Appendix A
Events Used in Studies 1–5
The list of event pairs in Table A1 were used in Study 1. (The
remaining studies had a subset of these pairs). Participants saw one event
pair at a time. For each matched set of causal, noncausal, and different-
domain pairs, the actual elapsed times and years of occurrence were
approximately the same. The event pairs were mixed for pair type, and
two different orders of presentation were used across the studies.
Table A1
Events in Studies 1–5
Historical event pairs
Set 1
The U.S. surgeon general determines that smoking is hazardous to health.
In a move coined “Marlboro Friday,” Philip Morris announces plans to cut the price of its flagship Marlboro
brand.
Country ad campaign begins featuring the slogan, “Come to where the flavor is. Come to Marlboro Country.”
McDonald’s bans smoking in all 11,000 of its restaurants.
The first Wal-Mart opens.
AOL goes to flat-rate pricing.
Set 2
The term AIDS (acquired immune deficiency syndrome) is used for the first time.
The popular slimming product, Ayds, using the slogan “Get Ayds and lose weight” is withdrawn from the
market.
The Gay Men’s Health Crisis is opened in New York City.
Rock Hudson, film star, dies of AIDS.
MTV debuts.
The space shuttle Challenger explosion kills its crew.
Set 3
Microsoft is chosen by IBM to create the operating system for its first personal computer.
The Federal Trade Commission begins to investigate claims Microsoft monopolizes the market for personal
computer operating systems.
Type
Year
Elapsed time in years
1
1964
29
2
3
1
2
3
1
1993
1964
1994
1962
1996
1982
1985
1982
1985
1981
1986
1980
1991
30
34
3
3
5
11
CAUSAL FORCE AND COMPRESSION OF ELAPSED TIME
699
Table A1 (continued)
Historical event pairs
Type
Year
Elapsed time in years
Wayne Ratliff develops dBase II, the first version of a personal computer database program.
Berners-Lee writes the initial prototype for the World Wide Web.
The U.S. government bails out auto manufacturer Chrysler.
The Simpsons television program debuts.
Set 4
Assembly of the ARPANET begins at University of California–Los Angeles and Stanford, marking the
beginning of the Internet.
E-mail is invented by Ray Tomlinson.
Bell Labs releases the first version of the UNIX operating system.
Nolan Bushnell founds Atari–manufacturer of the first game-oriented personal computer system.
The Woodstock Music Festival takes place.
Cohen and Boyer pioneer recombinant DNA techniques.
Set 5
“Come Alive! You’re in the Pepsi Generation” ad campaign by Pepsi kicks off the cola wars.
Coca-Cola introduces the New Coke.
Patio Diet Cola is replaced by the first Diet Pepsi-Cola.
Coca-Cola USA becomes the founding sponsor of Hands Across America, an initiative for public awareness
of hunger and homelessness.
The United States enters Vietnam.
AT&T is dismantled, and the Baby Bells are created.
Set 6
A scientific expedition discovers and photographs remains of the wreck of the Titanic at a depth of 12,460
feet on the ocean floor.
James Cameron’s film Titanic is released.
The U.S. Congress moves to make the Titanic an international memorial.
Amid widespread controversy, Americans David Leibowitz and Kimberley Miller are married inside a deep
sea submersible at the gravesite of the Titanic.
The Beatles’s John Lennon is murdered.
Netscape, a popular non-Microsoft browser, goes public.
Set 7
Mikhail Gorbachev, the leader of the Union of Soviet Socialist Republics, begins the Perestroika policy of
liberalization.
The Soviet Communist Party is disbanded, signifying the end of the Union of Soviet Socialist Republics as a
political state.
Soviet Secretary General Andropov dies; Chernenko replaces him.
Leningrad is renamed St. Petersburg.
The Chernobyl nuclear reactor disaster occurs.
Earth summit is held in Rio, Brazil, where the Convention on Protection of Species and Habitat is signed.
Set 8
Adolph Hitler’s book, Mein Kampf, in which he discusses the “Jewish problem” in Germany and Europe, is
published.
The first concentration camp is built in Germany.
Germany is admitted to the League of Nations.
Austrian chancellor Dollfuss is assassinated by the Nazis.
Charles Lindbergh flies solo across the Atlantic.
“The only thing we have to fear is fear itself”: U.S. President Franklin Delano Roosevelt launches the New
Deal.
2
1980
1991
1981
1990
11
1969
1972
1969
1972
1969
1973
3
(Appendices continue)
3
1
2
3
1
2
3
1
2
1963
1985
1964
1985
1965
1984
1985
1997
1987
9
3
4
22
21
19
12
14
3
2001
1980
1995
15
1
1985
6
2
3
1
2
3
1991
1984
1991
1986
1992
1925
1933
1926
1934
1927
1933
7
6
8
8
6
FARO, MCGILL, AND HASTIE
700
Table A1 (continued)
Historical event pairs
Set 9
Young Turks revolt in the Ottoman Empire, bringing ideas of modernism and liberalism.
Polygamy is abolished in Turkey.
The Assembly declares Turkey a republic with Kemal Ataturk as president.
Ataturk, the founder of modern Turkey, dies in Istanbul.
The San Francisco earthquake occurs.
The Empire State Building opens.
Set 10
The state of Israel is founded and is recognized by the United Nations.
Saudi Arabia, Libya, and other Arab states proclaim an embargo on oil exports to the United States.
The Dead Sea Scrolls, a source on the history of Judaism and the origins of Christianity, are discovered in a
cave in Israel.
Terrorists kill 12 Israeli athletes at the Olympic Games in Munich.
The transistor is invented at Bell Labs.
Roe v. Wade legalizes abortion in the United States.
Set 11
Szilard, Teller, and Einstein compose a letter to President Roosevelt, advising him of the prospect of creating
a weapon with uranium to start a nuclear chain reaction.
The first nuclear bomb is detonated at Trinity Site in the New Mexico desert.
The first successful helicopter flight takes place in the United States.
The first supersonic flight takes place in the United States.
The movie Casablanca is released.
The National Basketball Association (Tenenbaum et al., 2006) is formed.
Set 12
Russians launch the first satellite that circles the earth (Sputnik).
American astronaut Neil Armstrong is the first man to walk on the moon.
The first weather satellite, Tiros I, is launched by the United States.
The United States and the Union of Soviet Socialist Republics cooperate on the Apollo-Soyuz test project,
during which American and Soviet spacecrafts dock in orbit.
John F. Kennedy is elected, the youngest (and first Catholic) president.
Federal Express begins operations.
Type
Year
Elapsed time in years
1
1908
1926
1915
1938
1906
1931
18
1
1948
1973
25
2
1947
1972
1947
1973
25
1939
1945
1940
1947
1942
1949
6
1957
1969
1960
12
2
3
3
1
2
3
1
2
3
1974
1960
1973
Note. Type 1 ⫽ causal pairs; Type 2 ⫽ noncausal pairs (called same-domain pairs in Study 1); Type 3 ⫽ different-domain pairs.
Appendix B
Figure B1 Presented to Participants in Study 4
Figure B1. Figures presented to participants in the dissipating, stable, and accumulating conditions in Study 4.
23
25
7
7
14
13
CAUSAL FORCE AND COMPRESSION OF ELAPSED TIME
701
Appendix C
Sample Explanations in Study 4
Dissipating-Cause Explanations
“At first, the Americans would be jealous of the Russian success
and their determination would be great. Over time they would
become less bothered, especially if the Russians were not making
much further progress.”
“Scientists would be keen to find ways, such as ways to orbit the
earth, but if not enough funds and resources . . . research may run
cold. Citizens would like the idea initially, but with media advertising pros and cons of embarking such ideas . . . may disinterest
citizens to approve of idea and funds/resources spent elsewhere.”
“Initially the shock generated by the event is likely to make it
appear as a magnified threat. Over time, however, new events will
capture the imagination of politicians and the masses, so that the
Sputnik no longer appears as an immediate danger.”
“The initial result of the Russians launching the first satellite
during the space race of the Cold War would lead to an initial
determination in the Americans to progress and beat the Russians
in the space race as matter of national pride.”
“As with customer brand awareness, as time moves on, this
event will be less influential (as would the giving of a free sample).
The Americans would not continue to be influenced so greatly by
this historical event as other historical events took its place.”
Accumulating-Cause Explanations
“Americans would be spurred on by the achievement of the
Russians even though initial dismay/shock/negative emotion could
possibly cause a hindrance in the beginning.”
“At first, the Americans might not feel as if the Russians are a
competitor in the ‘space race.’ However after the launch of Sputnik, the Russians have shown themselves to be capable of competing in this race, thus, the pressure for the American to make
progress would have increased over time.”
“Because a sense of competitiveness between Russia and America would increase over time. In the event of Russia’s satellite,
Americans began to realize about the importance of the space race.
They didn’t want to be lagged behind.”
“The Americans would want to be better than the Russians in
everything to show that they are the bigger and better world power.
Therefore, after the Russians launched their satellite, the Americans would work harder and harder.”
“Due to the competitive nature of the cold war era, knowing that
they were behind in the space race would increase American’s
interest in it.”
“Americans might want to keep up with the progress made by
the Russians when they launched Sputnik, thus as time goes by,
there might be a heightened determination to make progress in the
‘space race’.”
“As they would be spurred by the first event and this effect will
increase over time.”
“As time went on, the Americans would want to beat the
Russians more and more in the space race.”
Stable-Cause Explanations
“Determination to make progress in the space race is constant.
More research and technological advances may increase the base
of knowledge, but the determination to make progress will likely
remain constant.”
“More research and technological advances may increase the
base of knowledge, but the determination to make progress will
likely remain constant.”
“Once they have entered the space race, their determination to
remain at par with the counterpart is not going to diminish over a
period of time and, therefore, can be assumed to be constant.”
“The Americans would want to beat the Russians in the space
race so their determination would be maintained by competition.”
Received October 17, 2007
Revision received January 7, 2010
Accepted January 9, 2010 䡲