ArticlePDF Available

Contrasting predictive and causal values of predictors and causes


Abstract and Figures

Three experiments examined human processing of stimuli as predictors and causes. In Experiments 1A and 1B, two serial events that both preceded a third were assessed as predictors and as causes of the third event. Instructions successfully provided scenarios in which one of the serial (target) stimuli was viewed as a strong predictor but as a weak cause of the third event. In Experiment 2, participants' preexperimental knowledge was drawn upon in such a way that two simultaneous antecedent events were processed as predictors or causes, which strongly influenced the occurrence of overshadowing between the antecedent events. Although a tendency toward overshadowing was found between predictors, reliable overshadowing was observed only between causes, and then only when the test question was causal. Together with other evidence in the human learning literature, the present results suggest that predictive and causal learning obey similar laws, but there is a greater susceptibility to cue competition in causal than predictive attribution.
Content may be subject to copyright.
Copyright 2005 Psychonomic Society, Inc. 184
Learning & Behavior
2005, 33 (2), 184-196
This paper examines differences between predictive
and causal learning in humans. Events often occur in our
environment according to a consistent temporal distrib-
ution. Some events occur simultaneously (e.g., the sound
and sight of water running out of the tap), whereas other
events occur sequentially (e.g., hunger dissipates after
the intake of food). When the events repeatedly take
place following a sequential distribution in time, the first
event (i.e., the antecedent event) can become a signal for
the occurrence of the second event (i.e., the subsequent
event). Learning to predict the occurrence of an event on
the basis of the occurrence of another event is essential
for survival. For instance, we can learn to anticipate
when and where food or water will be available, as well
as which situations can potentially result in physical in-
juries or social reward or punishment. By learning to an-
ticipate the occurrence of significant events, we can
modify our behavior to maximize or minimize the con-
sequences of appetitive or aversive events, respectively.
But, under some conditions, we can learn that the an-
tecedent event is not only a predictor, but also a cause of
the subsequent event. Causal learning can be even more
important than predictive learning. First, a cause is al-
ways a predictor of the event, so that causal learning
might be expected to include and perhaps depend on pre-
dictive learning. Second, by knowing the causes of the
event, we can not only anticipate the occurrence of the
event, but also potentially control its occurrence. For ex-
ample, we can learn to establish a comfortable tempera-
ture in a room by manipulating the thermostat, or we can
learn that ingestion of a certain food produces an aller-
gic reaction, in which case we will try to avoid that food
in the future. Therefore, whereas predictive learning al-
O.P. was supported by a postdoctoral fellowship from the Spanish
Ministry of Education (Ref. EX2002-0739). H.M. was supported by
Grant PI-2000-12 from the Department of Education, Universities, and
Research of the Basque Government. T.B. was supported by a postdoc-
toral fellow by the Fund for Scientific Research Flanders (Belgium).
We thank Jeffrey C. Amundson, Gonzalo P. Urcelay, Kouji Urushihara,
Miguel A. Vadillo, and Daniel S. Wheeler for their comments on an
earlier version of this manuscript. Correspondence should be addressed
to R. R. Miller, Department of Psychology, SUNY-Binghamton, Bing-
hamton, NY 13902-6000 (e-mail:
Contrasting predictive and causal values of
predictors and of causes
State University of New York, Binghamton, New York
State University of New York, Binghamton, New York
and Appalachian State University, Boone, North Carolina
State University of New York, Binghamton, New York
and University of Leuven, Leuven, Belgium
University of Deusto, Bilbao, Spain
State University of New York, Binghamton, New York
Three experiments examined human processing of stimuli as predictors and causes. In Experi-
ments 1A and 1B, two serial events that both preceded a third were assessed as predictors and as
causes of the third event. Instructions successfully provided scenarios in which one of the serial (tar-
get) stimuli was viewed as a strong predictor but as a weak cause of the third event. In Experiment 2,
participants’ preexperimental knowledge was drawn upon in such a way that two simultaneous an-
tecedent events were processed as predictors or causes, which strongly influenced the occurrence of
overshadowing between the antecedent events. Although a tendency toward overshadowing was found
between predictors, reliable overshadowing was observed only between causes, and then only when
the test question was causal. Together with other evidence in the human learning literature, the pres-
ent results suggest that predictive and causal learning obey similar laws, but there is a greater suscep-
tibility to cue competition in causal than predictive attribution.
lows us to modify our behavior in order to adapt to the
impending presence or absence of an event, causal learn-
ing also allows us to use our own behavior to produce or
prevent the occurrence of an event.
The systematic study of causal learning by humans
within an associative framework began two decades ago
with a study by Dickinson, Shanks, and Evenden (1984).
They reported blocking, an effect first found in classical
conditioning with nonhuman animals (Kamin, 1968).
Conventionally, in a blocking design, a conditioned stim-
ulus (CS), A, is first paired with an unconditioned stimu-
lus (US) and then CS A is further paired with the US in
the presence of an added CS, X. At test, CS X elicits a
weak response, relative to a control condition in which
CS A was not paired with the US prior to the AX–US
pairings. In the experiments by Dickinson et al., no ac-
tual CSs or USs were presented. Rather, the participants
were exposed to neutral stimuli in a computer-based task
resembling a video game and were instructed to rate the
causal efficacy of an artillery shell being fired (analo-
gous to a blocked CS X) in destroying a tank (analogous
to a US). Additionally, a mine exploding (analogous to
blocking a CS A) could also destroy the tank.
The relevance of the Dickinson et al. (1984) study to
the area of human associative learning was twofold.
First, as previously mentioned, these authors replicated
the blocking effect with humans in a preparation in
which (1) the fictitious stimuli were affectively neutral,
and (2) the dependent variable used to assess the strength
of the association between the candidate blocked cause,
and the effect consisted of a verbal judgment (i.e., a nu-
merical rating; see also Allan & Jenkins, 1983; Alloy &
Abramson, 1979). Thus, this study showed that causal
learning by humans, as in classical conditioning of non-
human animals, is subject to the blocking effect. Second,
this observation suggests that, regardless of the differ-
ences in species and preparations, the same mechanisms
may be involved in conditioning and causal learning. As
a logical consequence of this correspondence between
conditioning and causal learning, Dickinson et al. (see
also Gluck & Bower, 1988; Shanks & Dickinson, 1987)
proposed that the associative theories developed to ex-
plain classical conditioning in nonhuman animals (e.g.,
Rescorla & Wagner, 1972) could also be used to account
for causal learning by humans.
Dickinson et al.s (1984) study greatly stimulated the
study of stimulus competition in humans within an as-
sociative framework. Stimulus competition effects, such
as the above-mentioned blocking effect (Kamin, 1968),
overshadowing (Pavlov, 1927), and relative stimulus va-
lidity effect (Wagner, Logan, Haberlandt, & Price, 1968)
were initially found in experiments using nonhuman an-
imals and then successfully replicated using predictive
or causal judgment preparations with humans (for a
demonstration of overshadowing, see Waldmann, 2001;
for demonstrations of relative stimulus validity effect,
see Kao & Wasserman, 1993; Matute, Arcediano, &
Miller, 1996; Van Hamme & Wasserman, 1994). These
studies of stimulus competition in human contingency
learning demonstrated that, as in conditioning experi-
ments with animals, responding to a target stimulus does
not depend only on the number and quality of pairings
with the outcome or US (i.e., contiguity), but also on the
associative status of any other stimuli that were present
during training of the target stimulus. According to differ-
ent theoretical approaches, stimuli compete either to enter
into an association with the outcome (e.g., Rescorla &
Wagner, 1972) or to express their previously acquired as-
sociation with the outcome (e.g., Miller & Matzel, 1988).
Thus, by replicating effects previously found in nonhuman
animals, the study of stimulus competition became a criti-
cal issue in the human associative learning literature.
However, transfer between the animal conditioning and
human associative learning literatures has not been uni-
directional (see Miller & Matute, 1996, for a detailed
discussion). Some effects originally found in humans
were also replicated in classical conditioning with non-
human animals. This is the case for backward blocking,
an effect akin to forward blocking but with the order of
the training phases reversed. That is, AX–US pairings
are followed by A–US pairings, and then responding to
X is assessed. This effect was originally reported by
Shanks (1985) in humans and then replicated in rats by
Denniston, Miller, and Matute (1996).
The study of stimulus competition has attracted a large
number of researchers and theorists in the area of human
contingency learning, especially over the last two decades
(see De Houwer & Beckers, 2002, for a recent review).
Two main paradigms have been commonly used indis-
criminately in these studies: predictive and causal learn-
ing. In the predictive learning paradigm, participants are
asked to rate the predictive relationship between a target
antecedent event, normally referred to as cue or signal (or
even predictor or indicator), and a subsequent event, re-
ferred to as the outcome (O). In the causal learning par-
adigm, participants are asked to rate the causal relation-
ship between the antecedent and subsequent events,
which are normally referred to as cause and effect, re-
spectively. Evidence of stimulus competition, mostly for-
ward and backward blocking, has been reported in both
the predictive and causal learning paradigms.
Although the study of predictive and causal learning
in humans has received extensive attention, researchers
have largely ignored possible differences between these
two kinds of learning. Certainly there were almost no sys-
tematic comparisons of these two learning paradigms.
The only exception to this was provided by De Houwer,
Beckers, and Glautier (2002), who studied predictive
and causal learning in both forward and backward block-
ing paradigms. In the causal conditions of their experi-
ments (i.e., Condition Weapons), the participants were
asked to rate the likelihood that firing a weapon would
be followed by destruction of a tank (i.e., the outcome),
whereas in the predictive conditions (i.e., Condition In-
dicators), they were asked to rate the likelihood of the
destruction of the tank when an abstract visual figure
186 PINE
was present. De Houwer et al. found both forward and
backward blocking in the causal condition, but no stim-
ulus competition in the predictive condition, which they
interpreted as suggesting that stimulus competition de-
pends on the causal nature of the stimuli.
Other studies have systematically manipulated the type
of test questions (i.e., predictive or causal) in experiments
in which the task is consistently embedded in what is
clearly a causal scenario. For example, Matute, Vegas, and
De Marez (2002; see also Vadillo, Miller, & Matute,
2005) used a variation of the allergy task (Gluck & Bower,
1988; Wasserman, 1990), in which the task of the partic-
ipants was to rate the relationship between the ingestion of
some substances (foods or medicines) and the develop-
ment of fictitious allergies. Therefore, Matute et al. (2002)
did not directly contrast predictors and causes per se, but
rather the effect of different test questions concerning a
potential cause. That is, in contrast to De Houwer et al.
(2002), who contrasted predictors and causes but did not
manipulate the type of test question, Matute et al. (2002)
exclusively focused on the study of causal learning and
manipulated the type of test question. Because these prior
studies assessed the impact of either the type of stimuli
(De Houwer et al., 2002) or the type of test question (Ma-
tute et al., 2002), an integrative assessment of the inter-
action of the type of stimuli (i.e., predictors vs. causes) and
type of test question (i.e., predictive vs. causal) in a single
study seemed necessary. Note that there are several dif-
ferent types of predictive questions that can be asked (see
Vadillo et al., 2005). For brevity and because we used only
one type of predictive question, we will refer to it simply
as the predictive question, although it is highly similar to
the predictive-value question of Vadillo et al. (2005) as
opposed to what they called their prediction question.
In order to compare predictive and causal learning in
a single experiment, at least three approaches could be
followed. The first approach consists of using different
scenarios (provided by the instructions) in the predictive
and causal conditions. For example, the stock market
task used by Chapman and Robbins (1990; see also Chap-
man, 1991), in which the participants are required to pre-
dict the behavior of the real estate market on the basis of
changes in the price of stock for a fictitious electronics
company, could be used in the predictive condition,
whereas the allergy task (Wasserman, 1990) could be used
in the causal condition. The problem with this approach is
that it relies upon comparisons among groups differing
appreciably in procedural aspects (e.g., instructions, set
of stimuli, test questions, and rating scales). A second
approach consists of using different instructions with the
same scenario to encourage the processing of the rela-
tions as either predictive or causal, an approach that was
followed by De Houwer et al. (2002). The third strategy
consists of taking advantage of the participant’s preex-
perimental knowledge of predictive and causal relations.
For example, in the allergy task, participants are expected
to process the intake of a medicine as a cause of the al-
lergic reaction, whereas the color of a flask containing a
medicine would be processed as a potential predictor.
Our present experiments used either different wording
of instructions (Experiments 1A and 1B) or participants’
preexperimental knowledge (Experiment 2), in order to
study and contrast predictive and causal learning.
Specifically, in Experiments 1A and 1B, the scenario had
two antecedent events occur serially in time and then fol-
lowed by the outcome (i.e., Stimulus 1 [S1] was followed
by Stimulus 2 [S2], and S2 was followed by O). How-
ever, the instructions were worded so that both S2 and O
were seemingly effects of S1. The purpose of Experi-
ment 1 was simply to demonstrate that a single stimulus
(S2 in this case) could simultaneously assume different
values as a predictor and as a cause of O. In Experi-
ment 2, we used the training scenario to make two an-
tecedent events either predictive or causal and then stud-
ied overshadowing between these events. The central
issue in Experiment 2 was whether predictors and causes
were differentially sensitive to overshadowing. The par-
ticipants’ preexperimental knowledge of the causal
structure of the events was used to determine whether the
training scenario was predictive or causal.
Experiments 1A and 1B examined the influence of the
cover story provided within the instructions on the pro-
cessing of stimuli as either predictors or causes of a sub-
sequent stimulus. Although the antecedent stimuli (S1
and S2) and the subsequent stimulus (O) used in these
experiments were simultaneously presented on the com-
puter screen during each treatment trial, the instructions
clearly indicated that the stimuli occurred serially.
Specifically, according to the instructions, S1 was fol-
lowed by S2, and then S2 was followed by O (see the top
panel of Figure 1). In addition to the temporal priority of
S1 over S2, the instructions encouraged processing of S1
as a common cause of both S2 and O (i.e., the presenta-
tion of S1 first S2 and then O; see bottom panel of Fig-
ure 1). On the basis of these causal relations suggested
by the instructions, we anticipated the S1 would be per-
ceived as both a predictor and a cause of O. But how
would S2 be perceived with respect to O? Cohen, Run-
dell, Spellman, and Cashon (1999) demonstrated that, in
causal-chain sequences (i.e., S1S2O), the first
causal agent in the sequence (i.e., S1) is usually regarded
as the only cause of O. Although in Experiments 1A and
1B the instructions were not designed to suggest a causal
chain (i.e., in our experiments, the instructions suggested
independent S1S2 and S1O causal relations, instead
of a single S1S2O causal chain), on the basis of
their findings, we expected (1) high predictive and
causal ratings for S1 with respect to O, and (2) high pre-
dictive ratings and low causal ratings for S2 with respect
to O. Although this result would be unsurprising, it
would provide, if observed, a clear demonstration of the
impact of a cover story in producing divergent predictive
and causal processing of a single stimulus.
Participants and Apparatus
. Twenty-two undergraduate stu-
dents from the State University of New York at Binghamton partic-
ipated in Experiment 1A and 16 participated in Experiment 1B. The
participants received either credits in partial fulfillment of a course
requirement or $5 for participating in these experiments. The ex-
periments were conducted using personal computers in individual
. The instructions for each scenario are provided in the
Appendix. At the beginning of the experiment, all the participants
were reminded of the difference between a predictor and a cause
(see first screen of instructions in the Appendix). Four different sce-
narios were used in Experiments 1A and 1B. All participants were
exposed to all four scenarios, counterbalanced for order.
In Sce-
nario 1, S1 consisted of “the gun being fired,” S2 consisted of “the
shell being ejected,” and O was “the target being hit.” In Scenario 2,
S1 consisted of “imperfect steel sheets,” S2 was “the alarm being
sounded,” and O was “defective bumpers being produced.” In Sce-
nario 3, S1 consisted of “the switch being turned,” S2 consisted of
“the floor being wet,” and O was “the flowers blooming.” In Sce-
nario 4, S1 consisted of “workers loading strawberries,” S2 con-
sisted of “red stained shoes,” and O was “workers wheezing.” In
each of these text-based scenarios, the participants were requested
to read the instructions, pay attention to the training trials, and rate
the predictive and causal value of the S1-O and S2-O relations at the
end of training for that scenario. For each scenario, the participants
were given 20 training trials in which S1, S2, and O could be either
present or absent, with the restriction that no presentation of S1 oc-
curred without the presentation of S2, and vice versa. Thus, the con-
tingencies of O with respect to S1 and S2 were identical. Following
training and testing with each of the first three scenarios, the par-
ticipants were instructed: “You will now be learning about a new
situation. What you previously learned is unrelated to this new sit-
uation. Please press the enter key to begin.
In Experiment 1A, 14 S1-S2-O trials, 0 S1-S2-noO trials, 2
noS1-noS2-O trials, and 4 noS1-noS2-noO trials were presented
during training. This resulted in a contingency (i.e., as calculated by
the unconditional ΔP, Allan, 1980) of .67 for both the S1-O and S2-
O relations. In Experiment 1B, the same number of each trial type
as in Experiment 1A was administered during training in two of the
four scenarios (i.e., high contingency). In the other two scenarios
(i.e., zero contingency), the participants received 8 S1-S2-O trials,
2 S1-S2-noO trials, 8 noS1-noS2-O trials, and 2 noS1-noS2-noO
trials, resulting in a contingency of .00 for both the S1-O and S2-O
relations. The two types of contingencies were presented in counter-
balanced order and also were counterbalanced across the different
cover stories. The use of two contingencies allowed us to determine
whether participants were sensitive to the training contingencies or
merely responding on the basis of the instructions. The different trial
types were pseudorandomly distributed across training, with the re-
striction that the first trial was always an S1-S2-O trial, and that no
more than three trials of the same type occurred consecutively. A 1-
sec pause was introduced between trials, during which the computer
screen remained blank. On each trial, the stimuli (S1, S2, O) were
presented simultaneously in a horizontal sequence on the screen.
Figure 1. Experiments 1A and 1B. Top panel: temporal distri-
bution of the stimuli, according to the instructions. Bottom panel:
causal relations provided by the instructions. Time flows from left
to right.
Figure 2. Results of Experiment 1A. A fully within-subjects design was used in this exper-
iment. See text for details.
188 PINE
Each stimulus was followed by a question mark and the presence or
absence of each stimulus was denoted by the presentation of “yes”
or “no” below the corresponding stimulus. Participants were able to
advance to the next trial at their own pace by pressing the space bar,
but only after the stimuli had been presented for at least 2 sec.
At the end of the training phase of each scenario, the participants
were asked to rate the predictive and causal relationships between
S1 and O, and between S2 and O. Importantly, the predictive and
causal questions differed only in the words predictor and cause,
presented in uppercase. The predictive/causal test questions were as
follows: “To what degree do you think that [S1 or S2] was a [PRE-
DICTOR or CAUSE] of [O]?” The four test questions were pre-
sented together on the computer screen, counterbalanced for verti-
cal order across participants. The participants could read all four of
the questions before answering any of them, as well as answer them
in any order. A rating scale was displayed below each test question.
This scale ranged from 0 (not at all) to 100 (very high). An alpha
level of p .05 was adopted for all statistical analyses.
Results and Discussion
A preliminary 4 (scenario) 2 (stimulus: S1 vs. S2)
2 (test question: predictive vs. causal) analysis of vari-
ance (ANOVA) on the test ratings of Experiment 1A
showed no main effect of scenario, nor any interaction
involving scenario (all ps .19). Therefore, the results
of Experiment 1A were pooled across scenarios for each
participant. These results are depicted in Figure 2. As
can be seen, S1 was highly rated, as expected, as both a
predictor and a cause of O. S2 was given a low rating as
a cause of O; however, the predictive ratings of S2 were
higher than its causal ratings and similar to the predictive
ratings of S1. These impressions were confirmed by a 2
(stimulus: S1 vs. S2) 2 (test question: predictive vs.
causal) ANOVA on the test ratings pooled from all four
scenarios, which yielded main effects of stimulus
[F(1,21) 41.05, MS
276.80, p .001] and test
question [F(1,21) 31.19, MS
364.60, p .001], as
well as a stimulus test question interaction [F(1,21)
76.04, MS
178.85, p .001]. The source of this inter-
action was further examined using the Tukey test, which
showed that the predictive and causal ratings of S1 did
not differ. However, the causal ratings of S2 were lower
than its predictive ratings. Also, the predictive ratings of
S1 and S2 did not differ, but the causal rating of S2 was
Figure 3. Results of Experiment 1B. The top and bottom panels depict the results
of Conditions High and Zero, respectively. A fully within-subjects design was used in
this experiment. See text for details.
lower than that of S1. Thus, whereas S1 was rated as both
a reliable predictor and a cause of O, S2 was highly rated
as a predictor, but not as a cause of O.
In Experiment 1B, again there was no main effect or
interaction with scenario; hence, data were pooled across
scenarios for otherwise equivalent conditions. The re-
sults of Experiment 1B are depicted in Figure 3. As can
be seen, the predictive and causal ratings of S1 and S2 in
Condition High replicated those of Experiment 1A (i.e.,
high predictive and causal ratings of S1, and predictive
ratings of S2 higher than its causal ratings). Also, the rat-
ings were apparently sensitive to the contingencies. The
ratings were lower in Condition Zero (bottom panel) than
in Condition High (top panel). A pattern of results sim-
ilar to that observed in Condition High was also found in
Condition Zero. Most of these impressions were con-
firmed by a 2 (contingency: high vs. zero) 2 (stimulus:
S1 vs. S2) 2 (test question: predictive vs. causal)
ANOVA on the test ratings, which showed main effects of
contingency [F(1,15) 36.16, MS
660.37, p .001],
stimulus [F(1,15) 29.60, MS
408.02, p .001], and
test question [F(1,15) 17.28, MS
344.90, p .001].
More important, this ANOVA also revealed a stimulus
test question interaction [F(1,15) 22.90, MS
306.79, p .001] and a three-way interaction [F(1,15)
5.45, MS
304.78, p .05]. The remaining interac-
tions were not significant (ps .15). Post hoc compar-
isons using the Tukey test on the data from Condition High
revealed results similar to those of Experiment 1A. These
comparisons showed that the predictive and causal ratings
of S1 did not differ, and that the causal ratings of S2 were
lower than its predictive ratings. Also, the predictive rat-
ings of S1 and S2 did not differ, but the causal rating of S2
was lower than the causal rating of S1. In Condition Zero,
post hoc comparisons using the Tukey test showed that, as
in Condition High, the predictive and causal ratings of S1
did not differ. However, contrary to Condition High, in
Condition Zero the causal and predictive ratings of S2 did
not statistically differ, despite the tendency being in the
same direction as in Condition High. Also, the causal rat-
ings of S2 were lower than those of S1, whereas the pre-
dictive ratings of S2 and S1 did not differ.
Overall, these results show that, with a moderately high
positive contingency (i.e., ΔP .67, Experiment 1A and
Condition High in Experiment 1B), S1 was highly rated as
both a predictor and cause of Stimulus O, whereas S2 was
highly rated as a predictor, but not as a cause of O. In other
words, the instructions concerning causal mechanisms,
which encouraged the processing of S2 as a mere predic-
tor and S1 as a cause of O, yielded differential predictive
and causal value of the stimuli despite their having iden-
tical contingencies with respect to O.
Although the results of Experiments 1A and 1B sug-
gest that the causal mechanisms implicit in the scenarios
favored processing of stimuli as either a predictor or a
cause, overshadowing of S2 by S1 based on the tempo-
ral priority of S1 could provide an alternative explana-
tion of the observed results. That is, the serial presenta-
tion of S1 and S2 engendered in the instructions could
have allowed for these stimuli to compete with one an-
other for the acquisition of the predictive and causal
value. Because Experiments 1A and 1B did not include
a control condition for overshadowing (i.e., a condition
in which S2 was trained with the outcome in the absence
of S1), these experiments do not permit distinguishing
between causal mechanism and temporally based stimu-
lus competition as the basis for S2 being rated as a strong
predictor and a weak cause. Therefore, although the
cover stories provided by the instructions in Experi-
ments 1A and 1B seemingly produced differential pro-
cessing of S2 as a predictor and as a cause, the role
played in the present results by participants’ preexperi-
mental knowledge concerning predictive and causal re-
lationships in the real world is unclear. Experiment 2 was
designed to eliminate the serial relationship of the an-
tecedent events, among other things.
The results of Experiments 1A and 1B suggested that
either temporal priority or the participants’ preexperi-
mental knowledge of the predictive and causal relations
established by the fictitious scenarios favored processing
of the target stimulus, S2, more as a predictor than as
cause. Experiment 2 also assessed predictive and causal
values of a stimulus, but this time we assessed over-
shadowing between simultaneous antecedent stimuli.
Overshadowing was separately assessed between predic-
tors (i.e., Group P) and between causes (i.e., Group C).
Unlike Experiments 1A and 1B, in Experiment 2 the si-
multaneous occurrence of the two antecedent events pre-
cluded temporal order from possibly contributing to dif-
ferentiation between predictors and causes. Instead,
differentiation between predictors and causes was pro-
vided exclusively by the participants’ preexperimental
knowledge of predictive and causal relations in the real
world. Thus, any differences in overshadowing observed
between predictors and between causes would presum-
ably be due to differences in the sensitivity of predictors
and causes to associative competition.
In Experiment 2 the instructions contained no explicit
information concerning the role of the stimuli as predic-
tors or causes. All participants were given a common set
of instructions that was neutral regarding the causal/
predictive nature of the task in order to minimize the in-
fluence of instructions on ratings as a function of their
being predictive or causal. The participants’ task was to
take a fictitious test for a position in the National
Weather Service. For all the participants, the outcome
consisted of the occurrence of a thunderstorm. However,
different antecedent stimuli were presented to Groups P
and C. In Group P the stimuli consisted of instances of
animal behavior (e.g., birds flying in circles), whereas in
Group C the stimuli consisted of variations in the cli-
matic conditions (e.g., sudden change of temperature).
Although both types of stimuli were clearly correlates of
190 PINE
the subsequent occurrence of a thunderstorm, the stim-
uli were expected to be processed as mere predictors of
the thunderstorm in Group P, and as causes of the thun-
derstorm in Group C.
In Experiments 1A and 1B, S1 (an implicit cause of O)
might have overshadowed the causal value of S2 (the tar-
get stimulus), but there was no control condition to as-
sess overshadowing. Experiment 2 investigated whether
the causal value and predictive value of stimuli are sub-
ject to overshadowing by other causes and by other pre-
dictors, respectively, relative to appropriate overshadow-
ing control conditions. In Experiment 2, participants
received training with two implicit causes or two im-
plicit predictors (i.e., Stimuli A and X) presented in
compound, as well as a control condition for overshad-
owing (i.e., Stimulus Y). All participants were asked
both predictive and causal test questions in order to as-
certain whether overshadowing of the predictive or
causal value of a predictor (i.e., Group P) or of a cause
(i.e., Group C) could be observed. As in Experiments 1A
and 1B, the predictive and causal test questions differed
only in one word (i.e., predictor vs. cause).
Participants and Apparatus
. The participants were 74 students
in an introductory psychology course at the State University of
New York at Binghamton, who participated in this experiment in
partial fulfillment of a course requirement. The participants were
randomly assigned to one of two groups (ns 37).
. The task used in Experiment 2 simulated a computer-
based test that the participants had to take as part of a fictitious ap-
plication for the National Weather Service.
At the beginning of the
experiment, all participants were presented with a screen of in-
structions (see the Appendix). Immediately after they had read the
instructions, training commenced.
During training, the participants were given 5 AX-O and 5 Y-O
trials, in pseudorandom order. In Group P, Stimuli A, X, and Y con-
sisted of the presentation of the text “Birds flying in circles,” “Dogs
barking,” and “Squirrels returning to nests,” counterbalanced. In
Group C, these stimuli consisted of the presentation of the text “Fall
of atmospheric pressure,” “Sudden change of temperature,” and
“High humidity,” counterbalanced. The O consisted of the presen-
tation of the text “Thunderstorm: Intensity 10 out of 20.
After each trial and before the interval separating it from the sub-
sequent trial, the participants were required to indicate whether the
stimuli (i.e., A, X, Y, and O) were present or absent on that specific
trial. In this screen, the following instructions were presented:
“Please, indicate whether the following events were present or ab-
sent on this occasion.” All of the stimuli were presented in a verti-
cal list format on each trial, counterbalanced for order, together
with a box labeled “This event was . . . ,” containing the following
two options: “Present” and “Absent.” The participants were asked
to use the computer mouse to indicate the presence or absence of
each stimulus before they were allowed to proceed to the next trial;
they received no feedback concerning their answers. The purpose
of this posttrial test was twofold. First, it encouraged the partici-
pants to pay close attention to the events. Second, it provided a cri-
terion to eliminate those participants who did not pay sufficient at-
tention during training. In order for a participant’s data to be
retained in the analyses, she/he had to correctly indicate the pres-
ence or absence of all the stimuli on at least two trials among Tri-
als 2–5 of AX-O and two trials among Trials 2–5 of Y-O (i.e., they
were always allowed to fail in the first AX-O and Y-O trials because
this first test was expected to increase attention to stimuli on sub-
sequent trials). The application of this criterion eliminated 17 par-
ticipants (8 from Group P and 9 from Group C).
During testing, the participants were presented with predictive
and causal questions concerning the A-O, X-O, and Y-O associa-
tions. The predictive and causal questions read as follows: “To what
degree do you think that [A or X or Y] is a [PREDICTOR or
CAUSE] of thunderstorms?” All six of the test questions were pre-
sented on a single screen. The order of presentation of the test ques-
tions corresponding to a physical stimulus (e.g., dogs barking) was
identical for all participants. However, because of the complete
counterbalancing of the assignment of these text labels to the dif-
ferent stimuli (i.e., A, X, and Y), the order of presentation of the test
questions of the stimuli was counterbalanced. Also, the predictive
and causal test questions were presented in a counterbalanced order
(i.e., all predictive first or all causal first). The participants were re-
quested to answer each question by rating the different predictive
and causal relations using a scale ranging from 10 to 10. A
brief note at the top of the screen explained the meaning of the ex-
treme and middle values of the scale as well as how to indicate the
rating: “In order to answer, use any number in the scale from 10
to 10, where 10 Not at all,0 Not sure, and 10 Very
high. Click on the vertical arrows to indicate your rating. Please
read ALL of the questions before you answer any of them.” The cur-
rent rating to each question could be viewed in a small window to
the right of the corresponding question. This rating, which was ini-
tially set at 0, could be changed by clicking on one of two arrows
(up or down). Clicking on the up or down arrows yielded a 1 in-
crement or decrement, respectively, in the rating. Once the partici-
pants had rated the different predictive and causal relations, they
had to click on a button labeled “Save my ratings” in order to pro-
ceed to the next screen.
Following testing, the participants were given a short quiz in
order to assess their ability to differentiate prediction from causal-
ity. This quiz consisted of three multiple choice questions, which
read as follows: Question 1: Firing a gun is _____ by the pulling
of its trigger”; “Question 2: A train coming is _____ by a flashing
red light at a railroad crossing”; and “Question 3: High tides are
_____ by singing pop songs.” For each question, the participants
had to choose among the following three answers: “predicted,
“caused,” “neither predicted nor caused.” The correct answers for
Questions 1, 2, and 3 were “caused,” “predicted,” and “neither pre-
dicted nor caused,” respectively. Participants who failed to answer
all the questions correctly were removed from the analyses. Among
the participants who passed the attentional criterion (i.e., correctly
indicating the presence or absence of all the stimuli on at least two
of each type of training trial), 5 participants (i.e., 2 and 3 partici-
pants from Groups P and C, respectively) failed to pass the present
criterion. Thus, after the participants that did not pass both criteria
were eliminated, the resulting sample of participants in each group
was 27 for Group P and 25 for Group C.
Prior to analysis, the ratings of Stimuli A and X were averaged
to obtain a single rating for stimuli trained in compound (i.e.,
Comp). Pooling the ratings of Stimuli A and X was justified on the
basis of the complete counterbalancing of the stimuli and was per-
formed in order to reduce variability of the ratings. The ratings of
Stimulus Y directly represented the ratings of the elementally
trained stimulus (i.e., Elem). Therefore, Stimuli Comp and Elem
hereafter will represent the different training conditions received
by Stimuli A and X (i.e., compound training) and Stimulus Y (i.e.,
elemental training).
Results and Discussion
The order of presentation of the predictive and causal
questions at test was found to influence the test results.
Thus, in order to minimize the influence of the order of
presentation of the predictive and causal test questions
on the test results, we performed the analyses exclusively
on the ratings corresponding to the first test question
asked concerning each stimulus. After the second ratings
were eliminated, the sample for predictive ratings con-
sisted of 14 participants for each of Groups P and C and
the sample for causal ratings consisted of 13 participants
for Group P and 11 participants for Group C.
Figure 4 depicts the results of Experiment 2. As can be
seen, overshadowing (i.e., lower ratings of Stimulus Comp
than of Stimulus Elem) was suggested in Group P (top
panel) with the predictive ratings and in Group C (bottom
panel) with both the predictive and the causal ratings.
Also, overshadowing was apparently stronger in Group C
(as assessed by the causal ratings) than in Group P (as as-
sessed by the predictive ratings). These impressions were
partially supported by a 2 (type of stimuli: P vs. C) 2
(test question: predictive vs. causal) 2 (training con-
dition: Comp vs. Elem) ANOVA on the judgmental rat-
ings. This ANOVA yielded main effects of type of stim-
uli [F(1,48) 5.40, MS
32.71, p .05], test question
[F(1,48) 34.10, MS
32.71, p .01], and training
condition [F(1,48) 20.79, MS
5.55, p .01]. More
importantly, there were significant interactions: type of
stimuli test question [F(1,48) 4.62, MS
p .05], type of stimuli training condition [F(1,48)
17.43, MS
5.55, p .01], and a three-way interaction
[F(1,48) 18.41, MS
5.55, p .01]. Post hoc com-
parisons using the Tukey test indicated overshadowing
(i.e., lower ratings of Comp than Elem) in Group C, as
assessed by the causal ratings. Despite the tendency to-
ward overshadowing in the predictive ratings in both
Groups P and C suggested by Figure 4, post hoc com-
parisons showed no significant difference between the
predictive ratings of Comp and Elem in either Group P
or Group C. It is a little difficult to appreciate the criti-
Figure 4. Results of Experiment 2. The top and bottom panels depict the
mean ratings in Groups P (predictors) and C (causes), respectively. For
Group P, participants’ preexperimental knowledge was expected to encourage
purely predictive processing of the relation between the antecedent and subse-
quent paired events. For Group C, it was expected to encourage causal pro-
cessing of the relation between the antecedent and subsequent paired events.
See text for details.
192 PINE
cal differences in Figure 4. Consequently, we converted
the ratings of Stimuli Comp and Elem into overshadow-
ing scores (i.e., Overshadowing Scores Rating Elem
Rating Comp). Figure 5 depicts these overshadowing
scores for the predictive and causal test questions in
Groups P and C. Consistent with the previous statistical
analysis, overshadowing is seen to be stronger in the
causal ratings of Group C than in the predictive ratings
of both Groups C and P.
The present series of experiments used two techniques
to assess the processing of stimuli as predictors or
causes. In Experiments 1A and 1B, two stimuli (i.e., S1
and S2) identically covaried with the presentation of a
third stimulus (i.e., O). However, the instructions used in
these experiments promoted the processing of S1 as a
cause of both O and S2 (see Figure 1). The results of
these experiments showed, perhaps unsurprisingly but
now in a controlled manner, that although S2 was rated
as a poor cause of O, its predictive value was high. How-
ever, Experiments 1A and 1B raised the question of
whether the impaired causal value of S2 was exclusively
due to the instructions having created a bias against the
processing of S2 as a cause of O or, instead, Stimulus S1
having overshadowed S2 as a cause but not as a predic-
tor, or a combination of these two factors.
Experiment 2 assessed whether overshadowing could
occur in a similar preparation, but one in which the in-
structions did not identify one antecedent event as a
cause and the other as a predictor. Instead, the role of the
stimuli as predictors or causes was manipulated using
participants’ preexperimental knowledge and, orthogo-
nally, both the predictive and the causal value of the stim-
uli were assessed during testing. This experiment used a
single scenario and a common O for both the predictive
and causal conditions, and manipulated the predictive and
causal relations by varying the kind of antecedent stimuli
presented in each condition. The participants’ preexperi-
mental knowledge concerning the different stimulus–O
relations as either predictive or causal made the scenario
predictive or causal, respectively. This approach allowed
us to discount alternative explanations of differences in
processing of predictors and causes based on the wording
of the instructions (e.g., De Houwer et al., 2002). Con-
sistent with many other reports in the literature (e.g.,
Waldmann, 2001), Experiment 2 showed overshadowing
between causes (i.e., Group C) as assessed by the causal
test question, but not as assessed by the predictive test
question. Moreover, overshadowing was not found be-
tween predictors (i.e., Group P), as assessed by either the
predictive or causal test questions. However, a non-
significant tendency toward overshadowing was ob-
served in the predictive ratings of predictors and causes.
Analogously, a nonsignificant difference in the opposite
direction was found in the causal ratings of predictors.
But we hesitate to even speculate concerning what this
might suggest because the meaning of negative causal
ratings is unclear in our preparation.
The present results were observed despite our using
nearly identical wording of the predictive and causal
questions in Experiments 1A and 1B, as well as in Ex-
periment 2, in order to avoid unnecessary differences be-
tween test questions. Because of this, the predictive ques-
tion might have been expected to resemble the causal
question in many aspects, such as for example its stimu-
lating a comparison between the predictive value of the
target stimulus and any other candidate predictor, thereby
heightening its sensitivity to the impact of the predictive
value of other stimuli (i.e., stimulus competition). In
summary, the results of Experiment 2 suggest that re-
sponding based on causal learning is a more selective
process than is responding based on predictive learning.
This higher sensitivity of causes to competition does not
imply that predictors do not compete. Although Experi-
ment 2 revealed no significant overshadowing between
predictors (i.e., Group P, predictive ratings), a tendency
Figure 5. Results of Experiment 2. Mean ratings converted into overshadowing
scores (i.e., overshadowing scores rating of element
rating of compound). See text
for details.
toward overshadowing was found. Also, it is important to
remember that stimulus competition effects have been
reported in the literature, not only among causes, but
also among predictors. This higher sensitivity of causal
learning compared with predictive learning with respect
to stimulus competition cannot be explained by any con-
temporary model in its current form. According to asso-
ciative models of learning (e.g., Dickinson & Burke,
1996; Mackintosh, 1975; Miller & Matzel, 1988; Pearce
& Hall, 1980; Rescorla & Wagner, 1972; Van Hamme &
Wasserman, 1994; Wagner, 1981), learning consists of the
formation and strengthening of associations between the
representations of antecedent and subsequent events.
These models do not differentiate between processing of
these relationships as cue–O (predictive learning) or
cause–effect (causal learning) relationships; therefore,
they are silent concerning any difference between predic-
tive and causal learning.
The present results also pose problems for statistical
models of human learning. Most of these models (e.g.,
Allan’s, 1980, unconditional ΔP; Cheng & Novick’s,
1992, focal-set theory) do not try to differentially ac-
count for predictive and causal learning; rather, they are
designed to explain covariation learning (i.e., learning
whether the occurrence of one event covaries with the
occurrence of another event) without differentiating it
from either prediction or causation. Other models that
have been explicitly developed in order to explain causal
learning in humans (e.g., Cheng, 1997; Spellman, 1996)
are also silent concerning predictive learning. Accord-
ing to these models, learning the covariation (i.e., as
measured by the unconditional Δ P) among the stimuli is
a necessary condition in order to learn different causal
relations. Causation results exclusively from the history
of training with certain covarying events and from con-
trolling for alternative causes of the outcome. Control-
ling for the alternative causes takes the form of adjusting
ΔP relative to the base-rate occurrence of the outcome
[i.e., 1 P(O|noC)] in Cheng’s Power-PC model, and of
the computation of a conditional contingency (i.e., the
contingency of the target candidate cause relative to the
contingency of the alternative candidate cause or causes)
in Spellman’s model. Importantly, predictive learning is
not equated with covariation learning in these models.
Thus, these models are also silent on the difference be-
tween predictive and causal learning.
Waldmann’s (1996, 2000, 2001) causal-model theory
explicitly differentiates between causes and effects, but
not between predictors and causes. In Waldmann’s
causal-model view, the participants’ intrinsic knowledge
concerning causal relationships (i.e., causes influence
effects, but not vice versa) results in competition be-
tween causes, but not between effects. This directional-
ity, independent of the order in which the events are pre-
sented, is the basis of the causal model. Although
published versions of Waldmann’s model do not address
the nature of predictors, it could explain the present re-
sults by assuming that, in contrast to causes (which are
necessarily antecedent events for the outcome), predic-
tors are effects that can be used to diagnose a cause. This
possibility can be applied to our Experiment 2, in which
several examples of animal behavior were used as pre-
dictors of a thunderstorm. If participants inferred that
these patterns of animal behavior were indeed elicited by
atmospheric conditions that preceded the thunderstorm
(e.g., dogs barked in response to the impending thunder-
storm), then no overshadowing should be observed in the
predictive condition because, according to Waldmann’s
causal model, only causes (but not effects), compete.
In sum, most current models of human learning can-
not explain the observed greater stimulus competition
between causes than between predictors (but see De
Houwer et al., 2002, for an account based on inferential
reasoning). However, some current models could be
adapted post hoc to explain these results. For example,
statistical models (e.g., Cheng, 1997; Cheng & Novick,
1992) could be adapted by using a weighted Δ P (see,
e.g., Kao & Wasserman, 1993), in which each cell in the
traditional 2 2 contingency matrix is multiplied by a
given value ranging from 0 to 1, and making the weights
for the cells c (i.e., cue absent and outcome present) and
d (i.e., both cue and outcome absent) higher for causal
learning than for predictive learning. If the trials during
which the target cue is absent are more heavily weighted
for causal judgments than for predictive judgments,
stimulus competition would be stronger in the former
than in the latter case. Although this adaptation of Δ P
could explain stronger blocking in causality than in pre-
dictive value, it still would face problems in explaining
stronger overshadowing in causality than in predictive
value because in a between-subjects overshadowing de-
sign there are no trials on which the target cue is absent.
However, this is the case only with a between-groups
methodology; with a within-subjects design such as the
one we used in Experiment 2, the participants receive
both AX-O and Y-O trials, and the Y-O trials constitute
O trials without X present.
Another model that could be adapted to account for
these results is the comparator hypothesis (Miller &
Matzel, 1988). According to this theory, overshadowing
and other stimulus competition effects consist of im-
paired responding elicited by the target stimulus due to
its previous training in compound with a competing
stimulus. Specifically, this theory assumes that the pre-
sentation of the target stimulus, X, directly activates the
representation of O (i.e., through the X–O association),
as well as the representation of the competing stimulus,
A (i.e., through an X–A within-compound association),
which in turn indirectly activates the representation of O
(i.e., through an A-O association). According to the
comparator hypothesis, responding to X is based on a
comparison between the strength of the direct activation
of O (i.e., proportional to the strength of the X–O asso-
ciation) and the strength of the indirect activation of O
(i.e., proportional to the product of the strengths of the
X–A and A–O associations). The behavioral effect of the
194 PINE
directly activated representation of the outcome is down-
modulated by the indirectly activated representation of
the outcome. Hence, a weak response to the target stim-
ulus is observed in stimulus competition situations. In
order to explain the results of the present experiments,
this model could be adapted by assuming that the com-
parison process has greater impact on causal than on pre-
dictive responding. For example, if the value of the indi-
rect activation of O is weighted so that it is larger in
causal than in predictive learning (i.e., analogous to the
case of cells c and d in the previous adaptation of ΔP),
then the interference produced by the competing associ-
ation on the expression of the target association would be
stronger in causal judgment than in predictive judgment.
More important, this adaptation of the comparator hy-
pothesis, as well as the adapted ΔP, does not preclude
the possibility of stimulus competition being observed
between predictors.
It must be pointed out that this post hoc adaptation of
both the ΔP model (e.g., Allan, 1980) and the compara-
tor hypothesis (Miller & Matzel, 1988) to explain the
present results does not merely assume that causes are,
per se, more susceptible to competition than are predic-
tors. As indicated by the findings of Experiment 2, only
the causal value of causes (i.e., as assessed by the causal
test question) was strongly affected by a stimulus com-
petition treatment (i.e., overshadowing in our experi-
ment). This interaction between the nature of the stimuli
(as predictors or causes) and the test question (as pre-
dictive or causal) suggests that causal responding, in-
stead of causal learning, is highly susceptible to compe-
tition. This emphasis on responding rather than learning
is consistent with our proposed modifications of these
models. Obviously, because mere predictors lack a
causal value, the weaker stimulus competition found be-
tween predictors than between causes can be explained
by the intrinsic inability of predictors to support a causal
response or rating.
Of course, the Δ P model and the comparator hypoth-
esis are not the only models that could be extended to ac-
count for the higher susceptibility of causal judgments,
relative to predictive judgments, to stimulus competi-
tion. Models are flexible and they can encompass differ-
ent post hoc adaptations to explain the present results.
For example, traditional associative models (e.g., Rescorla
& Wagner, 1972) could assume independent causal and
predictive learning, with causal learning proceeding more
rapidly than predictive learning (e.g., learning-rate param-
eters may have higher values in causal associative learning
than in predictive associative learning). Applied to Ex-
periment 2, this principle anticipates greater overshad-
owing between causes than between predictors, at least
after a few pairings of the AX compound with O, which
is what was observed. In this framework, predictive test
questions assess predictive learning and causal questions
assess causal learning.
In Experiment 2, overshadowing was observed only
when the causal status of an antecedent event was as-
sessed at test. Because participants received the predic-
tive and causal test questions for the first time during the
test phase, the test questions could not affect predictive
or causal learning during the overshadowing treatment.
Moreover, differences in competition between causes
and between predictors have been observed for the same
stimuli (e.g., Experiment 2), which means that the basis
for the differential competition is present independently
of the test question. This suggests (but certainly does not
prove) that participants have different learning functions
during training for predictive value and for causal value.
However, even if this is correct, the actual competition
could occur during training (i.e., an acquisition deficit)
or during testing (i.e., a performance deficit).
The general assumption that causal learning is a more
selective process regarding stimulus competition than is
predictive learning, although speculative, is interesting.
As humans, we need to both predict and control the
events that occur in our environment. In both predictive
and causal learning, selecting the most reliable an-
tecedent stimulus (i.e., either predictor or cause) of an
event can be of critical importance for adaptative pur-
poses. However, causal learning might be more sensitive
to stimulus selection effects than predictive learning be-
cause, in order to efficiently influence the occurrence
and/or intensity of an event, we must exactly identify and
then manipulate its cause and avoid response competition
arising from manipulating other candidate causes. In con-
trast, attending to both unique and redundant predictors
to anticipate outcomes seemingly would not be as injuri-
ous. Alternatively worded, it is likely more difficult to si-
multaneously manipulate many causes than it is to si-
multaneously attend to many predictors. This would
result in selection among candidate causes being a more
rigorous process than selection among candidate predic-
tors. As previously stated, this does not mean that no
stimulus competition will ever occur among predictors.
In fact, Experiment 2 revealed a tendency to stimulus
competition among predictors. It simply implies that
stimulus competition would be stronger in causal judg-
ment than in predictive judgment.
The evolutionary question is, thus, why has causal
learning evolved as a more selective process than pre-
dictive learning? The answer to this question likely orig-
inates in predictors usually coming from the environment
and causes often originating in our own instrumental be-
havior. More generally, learning about the relations be-
tween external events can be a form of either predictive
or causal learning, whereas learning about the relations
between our own responses and their consequences is
causal by nature (see Dickinson, 2001). A corollary of
this view is that competition among responses (e.g., re-
sponse–outcome relations in instrumental conditioning)
should be more readily observed than competition among
stimuli. But such a comparison in a well-controlled sit-
uation has not yet been performed. The present study
should be considered a first attempt to contrast two
closely related types of learning, predictive and causal.
Allan, L. G. (1980). A note on measurement of contingency between
two binary variables in judgment tasks. Bulletin of the Psychonomic
Society, 15, 147-149.
Allan, L. G., & Jenkins, H. M. (1983). The effect of representations
of binary variables on judgment of influence. Learning & Motiva-
tion, 14, 381-405.
Alloy, L. B., & Abramson, L. Y. (1979). Judgment of contingency in
depressed and nondepressed students: Sadder but wiser? Journal of
Experimental Psychology: General, 108, 441-485.
Chapman, G. B. (1991). Trial order affects cue interaction in contin-
gency judgment. Journal of Experimental Psychology: Learning,
Memory, & Cognition, 17, 837-854.
Chapman, G. B., & Robbins, S. J. (1990). Cue interaction in human
contingency judgment. Memory & Cognition, 18, 537-545.
Cheng, P. W. (1997). From covariation to causation: A causal power
theory. Psychological Review, 104, 367-405.
Cheng, P. W., & Novick, L. R. (1992). Covariation in natural causal in-
duction. Psychological Review, 99, 365-382.
Cohen, L. B., Rundell, L. J., Spellman, B. A., & Cashon, C. H.
(1999). Infant’s perception of causal chains. Psychological Science,
10, 412-418.
De Houwer, J., & Beckers, T. (2002). A review of recent develop-
ments in research and theories on human contingency learning.
Quarterly Journal of Experimental Psychology, 55B, 289-310.
De Houwer, J., Beckers, T., & Glautier, S. (2002). Outcome and cue
properties modulate blocking. Quarterly Journal of Experimental
Psychology, 55A, 965-985.
Denniston, J. C., Miller, R. R., & Matute, H. (1996). Biological sig-
nificance as a determinant of cue competition. Psychological Sci-
ence, 7, 325-331.
Dickinson, A. (2001). The Twenty-Eighth Bartlett Memorial Lecture.
Causal learning: An associative analysis. Quarterly Journal of Ex-
perimental Psychology, 54B, 3-25.
Dickinson, A., & Burke, J. (1996). Within-compound associations
mediate the retrospective revaluation of causality judgements. Quar-
terly Journal of Experimental Psychology, 49B, 60-80.
Dickinson, A., Shanks, D. R., & Evenden, J. (1984). Judgement of
act–outcome contingency: The role of selective attribution. Quarterly
Journal of Experimental Psychology, 36A, 29-50.
Gluck, M., & Bower, G. H. (1988). From conditioning to category
learning: An adaptive network model. Journal of Experimental Psy-
chology: General, 117, 227-247.
Kamin, L. J. (1968). “Attention-like” processes in classical condition-
ing. In M. R. Jones (Ed.), Miami symposium on the prediction of be-
havior: Aversive stimulation (pp. 9-31). Miami: University of Miami
Kao, S. F., & Wasserman, E. A. (1993). Assessment of an information
integration account of contingency judgment with examination of
subjective cell importance and method of information presentation.
Journal of Experimental Psychology: Learning, Memory, & Cogni-
tion, 19, 1363-1386.
Mackintosh, N. J. (1975). A theory of attention: Variations in the as-
sociability of stimuli with reinforcement. Psychological Review, 82,
Matute, H., Arcediano, F., & Miller, R. R. (1996). Test questions
modulates cue competition between causes and between effects.
Journal of Experimental Psychology: Learning, Memory, & Cogni-
tion, 22, 182-196.
Matute, H., Vegas, S., & De Marez, P. J. (2002). Flexible use of recent
information in causal and predictive judgments. Journal of Experi-
mental Psychology: Learning, Memory, & Cognition, 28, 714-725.
Miller, R. R., & Matute, H. (1996). Animal analogues of causal judg-
ment. In D. R. Shanks, K. J. Holyoak, & D. L. Medin (Eds.), The psy-
chology of learning and motivation: Vol. 34. Causal learning
(pp. 133-166). San Diego: Academic Press.
Miller, R. R., & Matzel, L. D. (1988). The comparator hypothesis:
A response rule for the expression of associations. In G. H. Bower
(Ed.), The psychology of learning and motivation (Vol. 22, pp. 51-
92). San Diego: Academic Press.
Pavlov, I. P. (1927). Conditioned reflexes (G. V. Anrep, Trans.). Lon-
don: Clarendon Press.
Pearce, J. M., & Hall, G. (1980). A model for Pavlovian learning:
Variations in the effectiveness of conditioned but not of uncondi-
tioned stimuli. Psychological Review, 87, 532-552.
Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian con-
ditioning: Variations in the effectiveness of reinforcement and non-
reinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical con-
ditioning II: Current research and theory (pp. 64-99). New York:
Shanks, D. R. (1985). Forward and backward blocking in human con-
tingency judgment. Quarterly Journal of Experimental Psychology,
37B, 1-21.
Shanks, D. R., & Dickinson, A. (1987). Associative accounts of causal-
ity judgment. In G. H. Bower (Ed.), The psychology of learning and
motivation (Vol. 21, pp. 229-261). San Diego: Academic Press.
Spellman, B. A. (1996). Acting as intuitive scientists: Contingency
judgments are made while controlling for alternative potential
causes. Psychological Science, 7, 337-342.
Vadillo, M. A., Miller, R. R., & Matute, H. (2005) Causal and
predictive-value judgments, but not predictions, are based on cue–
outcome contingency. Learning & Behavior, 33, 172-183.
Van Hamme, L. J., & Wasserman, E. A. (1993). Cue competition in
causality judgments: The role of manner in information presentation.
Bulletin of the Psychonomic Society, 31, 457-460.
Van Hamme, L. J., & Wasserman, E. A. (1994). Cue competition in
causality judgments: The role of nonpresentation of compound stim-
ulus elements. Learning & Motivation, 25, 127-151.
Wagner, A. R. (1981). SOP: A model of automatic memory process-
ing in animal behavior. In N. E. Spear & R. R. Miller (Eds.), Infor-
mation processing in animals: Memory mechanisms (pp. 5-47). Hills-
dale, NJ: Erlbaum.
Wagner, A. R., Logan, F. A., Haberlandt, K., & Price, T. (1968).
Stimulus selection in animal discrimination learning. Journal of Ex-
perimental Psychology, 76, 171-180.
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: Academic Press.
Waldmann, M. R. (2000). Competition among causes but not effects
in predictive and diagnostic learning. Journal of Experimental Psy-
chology: Learning, Memory, & Cognition, 26, 53-76.
Waldmann, M. R. (2001). Predictive versus diagnostic causal learn-
ing: Evidence from an overshadowing paradigm. Psychonomic Bul-
letin & Review, 8, 600-608.
Wasserman, E. A. (1990). Detecting response-outcome relations: To-
ward an understanding of the causal texture of the environment. In
G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 26,
pp. 27-82). San Diego: Academic Press.
1. The four scenarios used in Experiments 1A and 1B were counter-
balanced for order according to the following four different sequences:
(1) Scenario 1, Scenario 2, Scenario 3, Scenario 4; (2) Scenario 2, Sce-
nario 1, Scenario 4, Scenario 3; (3) Scenario 3, Scenario 4, Scenario 1,
Scenario 2; (4) Scenario 4, Scenario 3, Scenario 2, Scenario 1.
2. The preparation used in Experiment 2 is available at http://www.
3. A submaximal intensity of the outcome (i.e., 10/20) was used in
Experiment 2 to facilitate the observation of cue competition between
A and X relative to Y (see De Houwer et al., 2002, for a rationale).
196 PINE
Instructions Used in the Experiments
First Screen
In this portion of the experiment, you will be required to learn about various causes and predictors. A cause
is something that PRODUCES or brings about an effect, whereas a predictor is something that SIGNALS a
future event. Thus, a predictor may or may not also be a cause of the effect. Please read all of the material on
each screen carefully before you proceed to the next screen. Periodically, you will be asked to answer several
questions, using the mouse to indicate your answer. Simply move the mouse along the rating scale and click
on your chosen rating (you can use either the left or the right mouse button to select your rating).
Scenario Screens
Scenario 1: Naval artillery
. Pretend that you are on-board a Naval destroyer in order to observe an ar-
tillery demonstration. For this demonstration the gunner is firing shells at a distant target. Each time that the
gunner fires a shell, you will observe the shell casing being ejected from the gun and you can also look through
your binoculars to observe whether the target has been hit. Other ships are also performing similar demon-
strations. Although they are firing at different targets, sometimes their shells will go off course and hit your
ship’s target. Please try to determine the relationships between the events.
Scenario 2: Auto factory. Pretend that you have recently been hired to work in the quality control division
of an auto factory. As part of your job, you are asked to investigate an increase in the rate of defective car
bumpers. The car bumpers are manufactured from steel sheets, some of which might be imperfect. As the
sheets proceed through the assembly line, a machine designed to detect flaws in the steel sheets sounds an
alarm to alert the assembly line workers that an imperfect sheet exists. If the alarm sounds, the workers are
supposed to remove the imperfect sheet to prevent a defective bumper from being produced. The workers’ su-
pervisor blames the workers for not removing the defective steel sheets when the alarm is sounded, whereas
the workers claim that faulty wiring in the bumper stamping machine, which results in power fluctuations, is
the real cause of the defective bumpers. Please try to determine the relationships between the events.
Scenario 3: Greenhouse. Pretend that you own a floral supply house that distributes flowers to several local
florists. Recently, you hired a new employee to help you in the greenhouse and have been trying to determine
if the employee has been doing his job. Each day the employee is supposed to water the flowers for one hour.
If the employee waters the flowers as instructed, the flowers should bloom the next day. To water the flowers,
the employee turns on a switch that automatically waters them for one hour. Although you are not always
present when the employee waters the flowers, you can tell if he has done so by whether or not the floor is wet.
However, another employee occasionally fertilizes the flowers, which can also make the flowers bloom the
next day. Please try to determine the relationships between the events.
Scenario 4: Strawberries. Pretend that you are a doctor who has been asked to investigate an increase in
allergic asthma cases on a produce farm. As you tour the farm, you will see the workers loading crates of fruit
onto trucks. You suspect that several of the fruits are responsible for the workers’ asthma attacks. As you ex-
amine each worker for wheezing symptomatic of asthma, you notice that the shoes of some of the workers are
stained from the fruits. Some workers shoes are stained red from strawberries and others are stained blue from
blueberries. Please try to determine the relationships between the events.
The National Weather Service (NWS) has opened a number of training positions for research in meteorol-
ogy. Since you are very interested in enrolling in the NWS (meteorology is your life and their salaries are very
appealing), you are going to apply for a position. As a part of this process, you are required to take a computer-
based test, which consists of determining the relationships among different events presented on the screen of
the computer. Therefore, you have to pay attention to those events so you can then rate their relations . . . and
obtain a good grade in your exam.
Before taking the exam, it is important for you to remember the difference between causes and predictors.
A cause is something that produces or brings about an effect, whereas a predictor is something that only sig-
nals a future event.
Remember: You will have to rate the relations among the different events based on your experience with
them during the computer-based test.
Good luck!
... Alternatively, it could be due to the fact that different questions (i.e., causal vs. predictive questions) give rise to differential judgments (Matute, Vegas & De Marez, 2002;White, 2003). In addition, the difference observed by Alloy et al. could be due to their using causes in one group and predictors in the other, as causes and predictors have also been shown to produce different judgments (PineÇo, Denniston, Beckers, Matute, & Miller, 2005). Moreover, Alloy et al. did not report the participants' number of actions. ...
... The present results do not support their conclusions. Instead, the differences observed by Alloy and her colleagues could be due, as mentioned in the Introduction, to the different assessment question that they used in each case (Matute et al., 2002;White, 2003), or to the fact that they used causes in one group and predictors in the other (see PineÇo et al., 2005, for differences between them). In addition, Alloy et al. did not report the number of attempts (i.e., actions) performed by participants in the active condition, nor the value of p(C) presented to passive participants. ...
Full-text available
The illusion of control consists of overestimating the influence that our behavior exerts over uncontrollable outcomes. Available evidence suggests that an important factor in development of this illusion is the personal involvement of participants who are trying to obtain the outcome. The dominant view assumes that this is due to social motivations and self-esteem protection. We propose that this may be due to a bias in contingency detection which occurs when the probability of the action (i.e., of the potential cause) is high. Indeed, personal involvement might have been often confounded with the probability of acting, as participants who are more involved tend to act more frequently than those for whom the outcome is irrelevant and therefore become mere observers. We tested these two variables separately. In two experiments, the outcome was always uncontrollable and we used a yoked design in which the participants of one condition were actively involved in obtaining it and the participants in the other condition observed the adventitious cause-effect pairs. The results support the latter approach: Those acting more often to obtain the outcome developed stronger illusions, and so did their yoked counterparts.
... In addition to active vs. passive roles of participants, there are many other variants that can be introduced in this task and that have been shown to affect the participants' estimations of causality. Examples include changing the wording of questions asked at the end of the experiment about the causal relationship (Crocker, 1982;Vadillo et al., 2005Vadillo et al., , 2011Collins and Shanks, 2006;De Houwer et al., 2007;Blanco et al., 2010;Shou and Smithson, 2015), the order in which the different trial types are presented (Langer and Roth, 1975;López et al., 1998), the frequency with which judgments are requested (Collins and Shanks, 2002;Matute et al., 2002), the description of the relevant events as causes, predictors, or effects (Waldmann and Holyoak, 1992;Cobos et al., 2002;Pineño et al., 2005), the temporal contiguity between the two events (e.g., Shanks et al., 1989;Wasserman, 1990;Lagnado and Sloman, 2006;Lagnado et al., 2007), and many other variables that fortunately are becoming well known. In the following sections, we will focus on the variables that seem to affect the illusion most critically in cases of null contingency. ...
Full-text available
Illusions of causality occur when people develop the belief that there is a causal connection between two events that are actually unrelated. Such illusions have been proposed to underlie pseudoscience and superstitious thinking, sometimes leading to disastrous consequences in relation to critical life areas, such as health, finances, and wellbeing. Like optical illusions, they can occur for anyone under well-known conditions. Scientific thinking is the best possible safeguard against them, but it does not come intuitively and needs to be taught. Teaching how to think scientifically should benefit from better understanding of the illusion of causality. In this article, we review experiments that our group has conducted on the illusion of causality during the last 20 years. We discuss how research on the illusion of causality can contribute to the teaching of scientific thinking and how scientific thinking can reduce illusion.
... In a predictive learning procedure, participants are asked to rate a predictive relation between a cue and an outcome, whereas in a causal learning procedure, participants are asked to rate the causal relation between a cue and an outcome. De Pineño et al., 2005a) found blocking in a causal learning procedure, but not in a predictive learning procedure. Similarly, Waldmann and Holyoak (1992), Waldmann (2000) showed that blocking is obtained more readily if cues A and X are described as causes of the outcome than when cues A and X are described as effects of the outcome. ...
Full-text available
BLOCKING IS THE MOST IMPORTANT PHENOMENON IN THE HISTORY OF ASSOCIATIVE LEARNING THEORY: for over 40 years, blocking has inspired a whole generation of learning models. Blocking is part of a family of effects that are typically termed "cue competition" effects. Common amongst all cue competition effects is that a cue-outcome relation is poorly learned or poorly expressed because the cue is trained in the presence of an alternative predictor or cause of the outcome. We provide an overview of the cognitive processes involved in cue competition effects in humans and propose a stage framework that brings these processes together. The framework contends that the behavioral display of cue competition is cognitively construed following three stages that include (1) an encoding stage, (2) a retention stage, and (3) a performance stage. We argue that the stage framework supports a comprehensive understanding of cue competition effects.
... Sin embargo, Matute, Arcediano, y Miller (1996) han puesto de manifiesto que el modo en que se formulan los juicios en la tarea experimental (por ejemplo, cuando se pide un juicio predictivo frente a un juicio de causalidad) tiene una influencia crucial sobre los procesos involucrados en la emisión de los mismos (véase también Pineño, Denniston, Beckers, Matute, & Miller, 2005, & Vadillo, Miller, & Matute, 2005. Asimismo, también se ha comprobado que el formato de presentación estimular (ensayo a ensayo versus resumido) influye considerablemente en las respuestas de las personas (Allan, 1993). ...
Full-text available
Our research focuses on learning about causal relationships between events when a candidate cause is a compound integrated by several individual causes. In two experiments, we compared the predictions of the Associative Models of Rescorla and Wagner (1972) and Pearce (1994), the Inductive Models of Cheng and Novick (1992) and Novick and Cheng (2004). In contrast with previous research about this topic, in these experiments, a causality judgments task was used in which the information about the presence/absence of the causes and the effect was presented through small samples of cases. Our results showed that the learning mechanisms involved in compound cue processing could be associative in origin.
... Sin embargo, Matute, Arcediano, y Miller (1996) han puesto de manifiesto que el modo en que se formulan los juicios en la tarea experimental (por ejemplo, cuando se pide un juicio predictivo frente a un juicio de causalidad) tiene una influencia crucial sobre los procesos involucrados en la emisión de los mismos (véase también Pineño, Denniston, Beckers, Matute, & Miller, 2005, & Vadillo, Miller, & Matute, 2005. Asimismo, también se ha comprobado que el formato de presentación estimular (ensayo a ensayo versus resumido) influye considerablemente en las respuestas de las personas (Allan, 1993). ...
Full-text available
Nuestra investigación se centra en el estudio del aprendizaje de relaciones causales en las que un compuesto estimular es una causa fiable de un efecto. En dos experimentos, hemos contrastado las predicciones derivadas de los Modelos Asociativos de Rescorla y Wagner (1972) y Pearce (1994), y los Modelos Inductivos de Cheng y Novick (1992) y Novick y Cheng (2004). En claro contraste con la investigación previa sobre este tema, en nuestros experimentos, hemos utilizado una tarea de juicios de causalidad en la que la información sobre la presencia/ausencia de las causas potenciales y el efecto se presentó mediante muestras o agrupaciones de casos. Los resultados ponen de manifiesto que los mecanismos de aprendizaje involucrados en el procesamiento de compuestos podrían ser de origen asociativo.
... Experiment 4 complemented Experiment 3 in addressing whether the reduced leverpressing observed in Experiment 1 was due to instrumental intervention or competition with leverpressing by a Pavlovian fear response. Differentiation between these two possibilities in Experiment 4 was based on evidence in the human causal learning literature that causes are more sensitive to cue competition than are Pavlovian signals (e.g., overshadowing, Pineño, Denniston, Beckers, Matute, & Miller, 2005;and blocking, De Houwer, Beckers, & Glautier, 2002). In Experiment 3, we found that a Pavlovian inhibitor for footshock had a greater effect on a Pavlovian signal for an aversive noise than on an intervention to avoid the aversive clicks. ...
Full-text available
Are humans unique in their ability to interpret exogenous events as causes? We addressed this question by observing the behavior of rats for indications of causal learning. Within an operant motor-sensory preconditioning paradigm, associative surgical techniques revealed that rats attempted to control an outcome (i.e., a potential effect) by manipulating a potential exogenous cause (i.e., an intervention). Rats were able to generate an innocuous auditory stimulus. This stimulus was then paired with an aversive stimulus. The animals subsequently avoided potential generation of the predictive cue, but not if the aversive stimulus was subsequently devalued or the predictive cue was extinguished (Exp. 1). In Experiment 2, we demonstrated that the aversive stimulus we used was in fact aversive, that it was subject to devaluation, that the cue-aversive stimulus pairings did make the cue a conditioned stimulus, and that the cue was subject to extinction. In Experiments 3 and 4, we established that the decrease in leverpressing observed in Experiment 1 was goal-directed instrumental behavior rather than purely a product of Pavlovian conditioning. To the extent that interventions suggest causal reasoning, it appears that causal reasoning can be based on associations between contiguous exogenous events. Thus, contiguity appears capable of establishing causal relationships between exogenous events. Our results challenge the widely held view that causal learning is uniquely human, and suggest that causal learning is explicable in an associative framework.
Two experiments were conducted using a picture–picture procedure to investigate the functional characteristics of evaluative conditioning. Images of foods were used as cue stimuli while images of body shapes were used as reinforcing stimuli. Foods paired with obese body shapes were rated more negatively than foods that had been paired with normal body shapes and the size of this evaluative conditioning effect did not differ between men and women. In Experiment 1 repeated presentation of the foods alone after training and before test did not reduce the size of the evaluative conditioning effect. In Experiment 2 there was no difference in the size of the evaluative conditioning effect between cues that were trained alone or in compound with another cue. The absence of extinction and overshadowing accords with the idea that in some cases evaluative conditioning is based on a learning mechanism sensitive only to the co-occurrence of stimuli and not to the contingency between them.
Full-text available
We present a model aimed at accounting for learning of predictive and causal relationships involving stimulus compounds, by means of a mechanism based on a normative-methodological analysis of causality that goes beyond the traditional associative/rule-based controversy. According to the model, causal learning is attained by computing the validity of each stimulus in a given learning situation. The situation is determined by the assumptions, objectives, and aims held by the learner or demanded by the learning context. Hence, validity computation depends on task demands: causal, predictive, or diagnostic according to a general principle of normative contextualization that allows learners to adapt a between-cues competition principle in a flexible way. Validity is computed using the Relevance Relativization mechanism, a linear model, based on the balance between the probability of stimulus combinations and the probability of each cue. Thus, cue interactions occur mainly when the combination of stimuli shows predictive changes in relation to the same cues considered individually. This model makes novel predictions concerning variations of the competition principle as a function of the type of procedure, including blocking, simultaneous blocking, and relative validity. In addition, our model also integrates top-down and bottom-up processing levels, including individuals' assumptions or previous beliefs.
Associative theories of learning have been used to explain human contingency learning since the 1980's. Recent findings have led several authors to claim that there is no evidence clearly showing the engagement of associative processes of acquisition or representation in human contingency learning, and to propose non-associative accounts. Prim-ing techniques can detect associative representations when the right parameters are employed. The present paper reviews evidence available of associative representations created after human contingency learning obtained using priming techniques. The evidence reviewed supports associative theories of learning and the assumption of spreading activation and associations between representations. OVERVIEW Human contingency learning (HCL) theories try to ex-plain how humans learn the relations between the presence and absence of some cues and the presence or absence of relevant outcomes, and how this learning guides their later behaviour. There are a large number of models and theories that try to account for HCL and propose very different learn-ing mechanisms and representations of the knowledge ac-quired. Until recently, the most accepted point of view was that both associative and higher level processes, like reason-ing, could be involved in HCL. But this has now been ques-tioned by several authors (for reviews see [1-3]). Most of the discussion has focused on the nature of the processes en-gaged in acquisition during HCL. The purpose of the present work is to review the evidence available of the existence of associative representations in HCL, and its theoretical relevance.
Full-text available
Causal asymmetry is one of the most fundamental features of the physical world: Causes produce effects, but not vice versa. This article is part of a debate between the view that, in principle, people are sensitive to causal directionality during learning (causal-model theory) and the view that learning primarily involves acquiring associations between cues and outcomes irrespective of their causal role (associative theories). Four experiments are presented that use asymmetries of cue competition to discriminate between these views. These experiments show that, contrary to associative accounts, cue competition interacts with causal status and that people are capable of differentiating between predictive and diagnostic inferences. Additional implications of causal-model theory are elaborated and empirically tested against alternative accounts. The results uniformly favor causal-model theory.
Associative and statistical theories of causal and predictive learning make opposite predictions for situations in which the most recent information contradicts the information provided by older trials (e.g., acquisition followed by extinction). Associative theories predict that people will rely on the most recent information to best adapt their behavior to the changing environment. Statistical theories predict that people will integrate what they have learned in the two phases. The results of this study showed one or the other effect as a function of response mode (trial by trial vs. global), type of question (contiguity, causality, or predictiveness), and postacquisition instructions. That is, participants are able to give either an integrative judgment, or a judgment that relies on recent information as a function of test demands. The authors concluded that any model must allow for flexible use of information once it has been acquired.
This chapter describes the potential explanatory power of a specific response rule and its implications for models of acquisition. This response rule is called the “comparator hypothesis.” It was originally inspired by Rescorla's contingency theory. Rescorla noted that if the number and frequency of conditioned stimulus–unconditioned stimulus (CS–US) pairings are held constant, unsignaled presentations of the US during training attenuate conditioned responding. This observation complemented the long recognized fact that the delivery of nonreinforced presentations of the CS during training also attenuates conditioned responding. The symmetry of the two findings prompted Rescorla to propose that during training, subjects inferred both the probability of the US in the presence of the CS and the probability of the US in the absence of the CS and they then established a CS–US association based upon a comparison of these quantities. The comparator hypothesis is a qualitative response rule, which, in principle, can complement any model of acquisition.
Abstract Three experiments examined,infants’ and adults’ perception of causal sequences of events. In the causal chain sequence, the first action causes a second action that then causes a final outcome; in the temporal chain sequence, the first two actions are independent and the second action causes a final outcome. Infants and adults were shown the same event sequences; infants were tested using a visual habituation paradigm, whereas adults were given a questionnaire.Experiment 1 indicated that 15-month-old infants perceive the primary cause of the final outcome,to be the first action in a causal chain but the second action in a temporal chain. Experiment 2 showed that adults interpret the causal sequences similarly to the 15-month-olds. Finally, Experiment 3 showed that 10-month-old infants do not yet perceive causal sequences in the same manner,as the 15-month-olds and the adults. These results are interpreted both in terms of infants’ developing knowledge,of causal events and adults’ attributions of causality in complex events. Infants’ Perception 3 Infants’ Perception of Causal Chains It is considered a hallmark of sophisticatedreasoning, across both development and culture, to be able to look deep into the past to find the causes of events. Such reasoning skills obviously may be of great value, and as many researchers are fond of noting, understanding causal relations is what allows us to predict and control our world (e.g., Alloy & Tabachnick, 1984; Cheng & Novick, 1990; Crocker, 1981; Young, 1995). Often several factors contribute to an outcome, but how do we make causal attributions
In the first experiment subjects were presented with a number of sets of trials on each of which they could perform a particular action and observe the occurrence of an outcome in the context of a video game. The contingency between the action and outcome was varied across the different sets of trials. When required to judge the effectiveness of the action in controlling the outcome during a set of trials, subjects assigned positive ratings for a positive contingency and negative ratings for a negative contingency. Furthermore, the magnitude of the ratings was related systematically to the strength of the actual contingency. With a fixed probability of an outcome given the action, judgements of positive contingencies decreased as the likelihood that the outcome would occur without the action was raised. Correspondingly, the absolute value of ratings of negative contingencies was increased both by an increment in the probability of the outcome in the absence of the action and by a decrement in the probability of the outcome following the action. A systematic bias was observed, however, in that positive judgements were given under a non-contingent relationship when the outcome frequency was relatively high. However, this bias could be reduced by giving extended exposure to the non-contingent schedule (Experiment 2). This pattern of contingency judgements can be explained if it is assumed that a process of selective attribution operates, whereby people are less likely to attribute an outcome to some potential target cause if another effective cause is present. Experiments 2 and 3 demonstrated the operation of this process by showing that initially establishing another agent as an effective cause of the outcome subsequently reduced or blocked the extent to which the subjects attributed the outcome to the action. Finally, we argue that the pattern and bias in contingency judgements based upon interactions with a causal process can be explained in terms of contemporary conditioning models of associative learning.
Many researchers have noted the similarities between causal judgment in humans and Pavlovian conditioning in animals. One recently noted discrepancy between these two forms of learning is the absence of backward blocking in animals, in contrast with its occurrence in human causality judgment. Here we report two experiments that investigated the role of biological significance in backward blocking as a potential explanation of this discrepancy. With rats as subjects, we used sensory preconditioning and second-order conditioning procedures, which allowed the to-be-blocked cue to retain low biological significance during training for some animals, but not for others. Backward blocking was observed only when the tar get cue was of low biological significance during training. These results suggest that the apparent discrepancy between human causal judgment and animal Pavlovian conditioning arises not because of a species difference, but because human causality studies ordinarily use stimuli of low biological significance, whereas animal Pavlovian studies ordinarily use stimuli of high biological significance, which are apparently protected against cue competition.
This chapter discusses that experimental psychology is no longer a unified field of scholarship. The most obvious sign of disintegration is the division of the Journal of Experimental Psychology into specialized periodicals. Many forces propel this fractionation. First, the explosion of interest in many small spheres of inquiry has made it extremely difficult for an individual to master more than one. Second, the recent popularity of interdisciplinary research has lured many workers away from the central issues of experimental psychology. Third, there is a growing division between researchers of human and animal behavior; this division has been primarily driven by contemporary cognitive psychologists, who see little reason to refer to the behavior of animals or to inquire into the generality of behavioral principles. The chapter considers the study of causal perception. This area is certainly at the core of experimental psychology. Although recent research in animal cognition has taken the tack of bringing human paradigms into the animal laboratory, the experimental research is described has adopted the reverse strategy of bringing animal paradigms into the human laboratory. A further unfortunate fact is that today's experimental psychologists are receiving little or no training in the history and philosophy of psychology. This neglected aspect means that investigations of a problem area are often undertaken without a full understanding of the analytical issues that would help guide empirical inquiry.
College students rated the causal efficacy of Elements X, A, and B of food compounds AX and BX in producing the allergic reaction of a hypothetical patient. The results of a 16-day allergy test were presented to subjects in a serial, trial-by-trial manner. The response format used was a running estimate, in which subjects were asked to rate all of the three foods after each of the 16 trials. Ratings of distinctive Elements A and B diverged and ratings of common Element X decreased as the difference in the correlation of AX and BX with the occurrence and nonoccurrence of the allergic reaction increased. These human causal judgments closely correspond with stimulus selection effects observed in the conditioned responses of animals in associative learning studies. The experiment also directly demonstrated the fact that significant changes in the causal ratings of a stimulus occur on trials in which the cue is not presented. Associative theories such as that of Rescorla and Wagner (1972) predict changes in associative strength only for those stimulus elements that are presented on a particular trial. A modification of the Rescorla-Wagner model is described that correctly predicts immediate changes in the associative strengths of all relevant cues on each trial—whether presented or not.