ArticlePDF Available

Positive and Negative Mediation as a Function of Whether the Absent Cue was Previously Associated with the Outcome

Authors:

Abstract and Figures

After presenting two cues, A and B, together, later pairings of one of the cues alone with an outcome can generate changes in the associative value of the absent cue. These changes can be in the same direction as the present cue (i.e., positive mediation) or in the opposite direction to the present cue (i.e., negative mediation). We found both mediational effects in a human contingency task. In addition, we found that the direction of the change was determined by the existence of a prior association between the absent cue and the outcome. When a prior association exists, the absent cue tends to change its value in the opposite direction to the present cue, whereas when there is no prior association, the absent cue tends to change its value in the same direction as the present cue. Recent associative models (Stout & Miller, 2007) can explain our results.
Content may be subject to copyright.
Positive and negative mediation as a function
of whether the absent cue was previously associated
with the outcome
Leyre Castro
University of Iowa, Iowa City, IA, USA
Helena Matute
Universidad de Deusto, Bilbao, Spain
After presenting two cues, A and B, together, later pairings of one of the cues alone with an outcome
can generate changes in the associative value of the absent cue. These changes can be in the same
direction as the present cue (i.e., positive mediation) or in the opposite direction to the present cue
(i.e., negative mediation). We found both mediational effects in a human contingency task. In
addition, we found that the direction of the change was determined by the existence of a prior associ-
ation between the absent cue and the outcome. When a prior association exists, the absent cue tends to
change its value in the opposite direction to the present cue, whereas when there is no prior associ-
ation, the absent cue tends to change its value in the same direction as the present cue. Recent associ-
ative models (Stout & Miller, 2007) can explain our results.
Keywords: Associative learning; Causal judgements; Causal learning.
There are many occasions in which absent events
become the content of learning. If two events
always occur together but then one of them is
absent we might nonetheless learn something
about the absent cue. For example, if we always
add milk to our coffee but one day we run out of
milk, and we enjoy the coffee even more, our
opinion of the goodness of milk in coffee will
surely change. If Sarah and Mike always agree
with each other, and one day Sarah enthusiastically
praises our new canvas, we will probably think that
absent Mike will like it too. These two examples
show that we can learn about absent events and,
also, that what we learn about the absent event
can be opposite or similar to what we learn about
the event that is actually present.
In an experimental situation, if two stimuli,
Cue A and Cue B, are always presented together,
an associative link will be established between
their mental representations; once this connection
Correspondence should be addressed to Leyre Castro, Department of Psychology, E11 Seashore Hall, The University of Iowa,
Iowa City, IA-52242, USA. E-mail: leyre-castroruiz@uiowa.edu
Support for this research was provided by Direccio
´n General de Investigacio
´n of the Spanish Government (Grant SEJ2007
63691/PSIC), Departamento de Educacio
´n, Universidades e Investigacio
´n of the Basque Government (Grant PI2008–9), and
Direccio
´n General de Investigacio
´n, Tecnologi
´a y Empresa of the Junta de Andaluci
´a (Grant SEJ-406). We would like to thank
Miguel A. Vadillo and Ed Wasserman for illuminating discussions concerning this research.
#2010 The Experimental Psychology Society 2359
http://www.psypress.com/qjep DOI:10.1080/17470218.2010.493614
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY
2010, 63 (12), 23592375
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
has been formed, the representation of one cue can
be activated by the associative link it has forged
with the other cue. When Cue A is later presented
alone, the representation of Cue B will be activated
by means of this associative link; because its
representation is activated, the absent cue
can engage in the learning process (e.g.,
Dickinson & Burke, 1996; Van Hamme &
Wasserman, 1994).
But, what do we learn about that absent event?
The rules that govern the formation of associations
involving absent events are far from being fully
understood. As shown by the examples above,
two possibilities are: (a) The representation of an
absent event enters into associations in the same
way as that of a present event, and (b) the rep-
resentation of an absent event enters into associ-
ations in the opposite way to that of a present
event. Evidence from animal and human associat-
ive learning studies suggests that both results are
possible. We refer to the former case as positive
mediation and to the latter case as negative
mediation.
For example, human retrospective revaluation
studies have consistently found that, after pairing
a compound cue (e.g., a pair of foods, such as
chicken and strawberries) with an outcome (e.g.,
a fictitious allergy), subsequent pairings of one of
these cues alone with the outcome (e.g., chicken
with allergy) leads to weakened causal attribution
to the absent cue (e.g., strawberries). Presumably,
the single cue presented in the second stage
(chicken) activates a representation of the absent
cue (strawberries), and the value of the absent
cue changes in the opposite direction to that of
the presented cue (see Chapman, 1991;
Dickinson & Burke, 1996; Wasserman &
Berglan, 1998; Williams, Sagness, & McPhee,
1994). Similar results have been obtained with
animals (see Harris & Westbrook, 1998;
Kaufman & Bolles, 1981; Miller & Matute, 1996).
The case in which responding to an absent cue
changes in the same direction as that to a present
cue has been reported mostly in animal studies.
For example, Brogden (1939) presented dogs
with a bell and a light together several times;
later, the bell alone was paired with shock.
Although the light had never been paired with
shock, it elicited a conditioned response at
testing; that is, the bell shock pairings led the
animals to change their responding to the absent
cue—the light (this is the so-called sensory pre-
conditioning effect). Presumably, during the
bellshock pairings, the bell activated the rep-
resentation of its associate, the light, so that the
light became associated with the shock as well.
1
Very little evidence for this mediational effect
has been reported with human participants,
except for studies using electrodermal condition-
ing paradigms (Brogden, 1947, and, more recently,
Vansteenwegen, Crombez, Baeyens, Hermans, &
Eelen, 2000). To our knowledge, there is no evi-
dence for this effect in human contingency learn-
ing, a domain in which retrospective revaluation
is often observed.
Thus, in sensory preconditioning, the associ-
ative strength of the absent cue changes in the
same direction as that of the present cue that
is activating its representation—positive
mediation. But, in the case of retrospective reva-
luation, the associative strength of the absent cue
changes in the opposite direction to that of the
present cue that is activating its represen-
tation—negative mediation. One possible crucial
factor is the presence of the outcome during
compound training.
1
Brogden (1939) presented the bell and the light (let us call them Cue A and Cue B, respectively) simultaneously. If the cues are
presented sequentially (that is, A B), the standard explanation is that, at the time of testing, Cue B is able to elicit a conditioned
response by means of the associative chain A Boutcome. Although this explanation seems quite straightforward when A and
B are presented sequentially, it has problems in accounting for the case in which A and B are presented simultaneously (as in
Brogden’s study and as it is in our experiment) or using a backward sensory preconditioning procedure (see Hall, 1996, for a complete
explanation of these different cases). So, even when the associative chain might have a role at the time of testing in the case of sequen-
tial forward sensory preconditioning, it also seems likely that the associatively activated representation of a cue can acquire associative
strength during training in the simultaneous and backward procedures. Because we present the cues in our compounds simul-
taneously, we assume, for the time being, that learning processes are taking place when the cues are absent during training.
2360 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12)
CASTRO AND MATUTE
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
Presence of the outcome during compound
training
In the studies that show positive mediation, the
outcome is never presented during the compound
training phase (e.g., Brogden, 1939), whereas the
outcome is presented during the compound train-
ing phase in the studies that show negative
mediation (e.g., Chapman, 1991; Dickinson &
Burke, 1996; Wasserman & Berglan, 1998). It
thus seems plausible to consider the occurrence
or nonoccurrence of the outcome when the com-
pound cue is presented as a relevant factor.
When the outcome is present during the com-
pound phase (as is the case when retrospective reva-
luation is observed), the target cue forms a within-
compound association with the other cue as well as
an association with the outcome. When later its
associate cue is paired alone with the outcome,
the absent target cue would be retrieved in
memory by means of the within-compound asso-
ciation with the present cue. In this way, the asso-
ciation of the absent cue with the outcome has the
opportunity to be modified. Learning about the
absent cue in this case involves the modification of
a previously established association. On the other
hand, when the outcome is absent during the com-
pound phase, the target cue cannot enter into an
association with the outcome. When later its
associate cue is paired with the outcome, the
absent target cue would be retrieved in memory
and will have the opportunity to develop an associ-
ation with the outcome. In this case, learning
entails the formation of a new association. Thus, it
seems plausible that, when a prior association
exists, the absent cue tends to change its value in
the opposite direction to that of the present cue—
negative mediation—whereas, when there is no
prior association, the absent cue forms a new associ-
ation that is in the same direction as that of the
present cue—positive mediation.
Cueno-outcome association versus no
association
At this point, we should clarify what it means to say
that a cue has no association with the outcome.
Although it seems a simple matter, the answer is
far from being straightforward, especially in human
studies, where the cue can be presented without
any mention of the outcome or the cue can be expli-
citly paired with the absence of the outcome.
When no information at all is provided about
the occurrence or nonoccurrence of the outcome,
the cue is likely to remain “neutral”, in the sense
that nothing can be learned about its value as a
predictor of the occurrence or nonoccurrence of
the outcome. In the case of explicit pairing of
the cue with no outcome, the cue is likely to
become a predictor of the absence of the
outcome; so, it does not remain neutral. For
example, Karazinov and Boakes (2004) reported
that a cue paired with no outcome yielded
ratings as negative as those for a cue involved in
conditioned inhibition training. The reason for
this result seems to be the knowledge that partici-
pants have about the existence of the outcome.
The mere expectation of the occurrence of the
outcome might endow the context with some
excitatory value that might cause the cues paired
with no outcome to become inhibitory. Indeed,
when inhibition studies are conducted with
animals, the target (inhibitory) cue needs be pre-
sented in an excitatory context (Miller, Hallam,
Hong, & Dufore, 1991).
Let us examine now the case of backward con-
ditioned inhibition. In human contingency
studies, pairings of the compound AB with no
outcome in a first phase, followed by pairings of
Cue A with the outcome, leads to a decrease in
ratings of Cue B—backward conditioned inhi-
bition, another negative mediation effect (e.g.,
Chapman, 1991; Larkin, Aitken, & Dickinson,
1998; Melchers, Lachnit, & Shanks, 2004;
Wasserman, Kao, Van Hamme, Katagiri, &
Young, 1996; Williams & Docking, 1995).
In a typical backward conditioned inhibition
experiment, participants are informed about the
absence of the outcome during the compound
phase by explicitly pairing the cues with no
outcome. Thus, at the end of the compound
phase, Cue B will have an association with the
outcome—in this case, an inhibitory association.
Changes to Cue B in Phase 2 will therefore
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12) 2361
LEARNING ABOUT ABSENT CUES
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
imply a modification of its prior association. As we
hypothesized before, when a prior association
exists the absent cue tends to change its value in
the opposite direction to that of the present cue.
This is exactly what it is observed in backward con-
ditioned inhibition: When Cue A is paired with
the outcome in Phase 2, and its value consequently
increases, the value of Cue B decreases even more,
compared to Phase 1.
For positive mediation to occur, it is probably
necessary that the target cue remains in a neutral
state at the end of the first (compound) phase;
that is, it may be necessary that the outcome has
been absent, and no information about it has
been provided, so that the target cue does not
develop any association with the outcome,
instead of developing an inhibitory association
with the outcome—or an association with no
outcome (Konorski, 1967).
Thus, we hypothesize that, when a cue has
already developed an association with the
outcome, be this excitatory or inhibitory, the pres-
entation of its associate paired with the outcome
(or with no outcome) will lead to changes in the
value of the absent cue in the opposite direction
as the present cue—negative mediation—whereas
when the cue has no prior association with the
outcome, the presentation of its associate paired
with the outcome (or with no outcome) will lead
to an association of the absent cue with the
outcome in the same direction as the present
cue—positive mediation. In order to test this
hypothesis and to illuminate the difference
between absence of the outcome and explicit pres-
ence of no outcome, in the present study we com-
pared three different compound training
possibilities—namely, compounds paired with
outcome, compounds paired with no outcome,
and simple exposure to the compounds.
EXPERIMENT
In our experiment, we presented participants with
three compounds, AB, CD, and EF, in the first
training phase, and then, in Phase 2, we presented
Cue A alone followed by the outcome and Cue E
alone followed by no outcome (fillers GH and G
were presented as well). We studied the influence
of presenting the compound cues in Phase 1
paired with: (a) presence of the outcome, (b) pres-
ence of no outcome, and (c) absence of the
outcome (mere exposure to the cues). We
wanted to see whether later, when one of the
cues of the compound is paired with the
outcome (or no outcome), the value of the absent
cue changes in the same or in the opposite direc-
tion to that of the present cue depending on the
type of training given in Phase 1.
In addition, we thought that the specific cover
story used could be critical (see Melchers, U
¨ngo
¨r,
& Lachnit, 2005b, for different results depending
on the experimental task). Some authors have
classified predictive relationships into two types:
causal and structural (see, for example,
J. H. Holland, Holyoak, Nisbett, & Thagard,
1986; Shanks, 1995). When a causal relationship
exists, organisms learn that an event generates
another event after a temporal interval, whereas
when a structural relationship exists, organisms
learn to predict a feature of an event from other
features that occur at the same time (as in category
learning, for example).
Common scenarios in retrospective revaluation
studies, like the food allergy task (e.g.,
Wasserman, 1990), entail a causal relationship.
In this case, people might be trying to find
which event is the cause of the effect, so that the
causal value of one cue is discounted in favour of
a stronger cue, a process that leads to negative
mediation. But other scenarios, like the symp-
tomsdisease task (e.g., Shanks, 1991) might
reflect a structural relationship. That is, the symp-
toms can be considered to be the constituents of
the disease, so that when one of the symptoms
becomes associated with the disease, the others
may be assumed to become associated with the
disease as well, a process that leads to positive
mediation. If the task scenario plays a role, then
a structural cover story should help promote posi-
tive mediation, whereas a causal cover story should
promote negative mediation.
At the same time, the distinction between the
foodallergy task and the symptoms disease
2362 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12)
CASTRO AND MATUTE
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
task could be conceptualized as a distinction
between a predictive and a diagnostic task. In a
predictive task, participants are shown a cause
(e.g., food) and have to predict the occurrence of
the effect (e.g., allergy). In a diagnostic task, the
presentation of the causal information is reversed,
so that participants are shown an effect (e.g.,
symptom) and have to predict the occurrence of
the cause (e.g., disease). Although many research-
ers have found no differences between predictive
and diagnostic scenarios (e.g., Arcediano,
Matute, Escobar, & Miller, 2005; Baker,
Murphy, Mehta, & Baetu, 2005; Cobos, Lo
´pez,
Can
˜o, Almaraz, & Shanks, 2002; Matute,
Arcediano, & Miller, 1996; Price & Yates, 1995;
Shanks & Lo
´pez, 1996), others have found nega-
tive mediation with predictive scenarios but not
with diagnostic scenarios (e.g., Tangen & Allan,
2004; Waldmann, 2000; Waldmann & Holyoak,
1992). Thus, regardless of the symptoms disease
task being understood as a structural or as a diag-
nostic task, the current evidence suggests that it is
more likely to obtain negative mediation with the
foodallergy task; the symptoms disease task
might be more likely to yield either no effect (if
it is understood as a diagnostic task) or positive
mediation (if it is understood as a structural task).
In the present experiment, we used a task that
permits identical materials to be described in
different ways. All of the participants had to
learn the relationship between different substances
in the blood and Hamkaoman disease—cues and
outcome, respectively. Half of the participants
(causal groups) were told that the substances in
the blood may be the cause of Hamkaoman
disease, so that a causal relationship between the
cues and the outcome was described. The other
half (structural groups) were told that the sub-
stances in the blood may be constituents of
Hamkaoman disease, so that a structural relation-
ship between the cues and the outcome was
described.
We expected to find negative mediation when
either outcome or no outcome was presented in
Phase 1 (O and no-O groups) and to find positive
mediation when there was no information about
the outcome (exposure groups). In addition, we
expected the structural task to facilitate positive
mediation and the causal task to facilitate negative
mediation. Thus, we expected to find the largest
negative mediation effect in the O and no-O
groups with a causal cover story and the largest
positive mediation effect in the exposure group
with a structural cover story.
Method
Participants and apparatus
A total of 150 students at the University of Iowa
took part in this experiment and received course
credit for their participation. Participants were
randomly assigned to six groups: O causal, no-
Ocausal, exposure causal, Ostructural, no-
Ostructural, and exposure structural, yielding
25 participants in each of the groups. Between 1
and 4 participants were studied concurrently on
each of four identically configured computer
workstations.
Stimuli
Eight different pictures of chemical substances of
different colours (red, orange, lilac, green, blue,
yellow, pink, and white) served as cues, and
Hamkaoman disease served as the outcome. The
cues were counterbalanced following a partial
Latin square design, which ensured that every sub-
stance was equally often assigned to each cue role.
When two cues were presented, the picture of each
specific substance appeared half of the trials on the
left side of the screen and half of the trials on the
right side of the screen. When only one cue was
presented, the picture of the substance appeared
centred on the screen.
Procedure
Participants sat in front of a workstation and were
introduced to a scenario in which they played the
role of a doctor trying to discover the relation
between a number of substances in blood and
Hamkaoman disease.
We included two between-group factors:
training in Phase 1 (outcome vs. no outcome vs.
exposure), and causal versus structural scenario.
The combination of these two factors yielded an
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12) 2363
LEARNING ABOUT ABSENT CUES
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
experimental design with six groups. The design of
the experiment is summarized in Table 1.
During Phase 1, all groups observed four com-
pounds, consisting of two substances each: AB,
CD, EF, and GH. Each of the compound cues
was presented 20 times in random order. In this
first phase, the two exposure groups were pre-
sented with just the compound cues, and they
were not informed about the occurrence or nonoc-
currence of the outcome during this phase. The
following instructions appeared on the screen at
the beginning of the first phase for exposure
groups:
We would like you to imagine that you are a specialist who tries
to discover the relationship between different substances in the
blood and diseases. In an attempt to accomplish this task, you
will first observe different patients’ files which show different
substances in the blood. You will see a separate screen for
each patient. First, you will see files with the substances that
each patient has, but you will not know yet whether or not
the patients have Hamkaoman disease. Nonetheless, to fam-
iliarize yourself with the substances at this point will help you
to decide later on. In order to help you to remember the sub-
stances, you will be asked to press the initials of each substance
when it appears on the screen.
The sentence “you will not know yet whether or
not the patients have Hamkaoman disease,”
adapted from Graham (1999), intended to main-
tain the cues in a neutral or ambiguous state.
Participants were asked to press the initials of
each of the substances in the order that they
appeared on the screen; this was done to make
sure they were reading the screen and paying
attention to the task. After entering the correct
initials, the participants could proceed to the
next trial. On completion of Phase 1, the following
instructions appeared on the screen:
Now you will be presented with patients’ files in which some of
them will have Hamkaoman disease, and some of them will not.
You will be shown the name of only one substance and will be
asked to indicate whether or not you believe the patient will
have Hamkaoman disease. After you make your prediction,
the computer will inform you of the correct answer.
On each trial of Phase 2, participants were
required to answer whether or not Hamkaoman
disease would appear. Once participants clicked
on the “Yes” or “No” button, the actual outcome
(Hamkaoman disease or no Hamkaoman disease)
was presented. Then, participants could proceed
to the next trial.
The instructions for the rest of the groups:
Ocausal, O structural, no-O causal, and no-
Ostructural, were similar, except that they
were told, from the beginning of Phase 1, that
they would see patients who could have or not
have Hamkaoman disease, and they would have
to predict on every trial whether or not each of
the patients had the disease.
In groups Ocausal and O structural, com-
pounds AB, CD, and EF were always followed
by Hamkaoman disease, whereas the filler
compound, GH, was always followed by no
Hamkaoman disease. We included compound
Table 1. Experimental design
Training
Group Phase 1 Phase 2 Testing
Exposure– structural AB, CD, EF, GH
Exposure– causal
AO,
O–structural AB O, CDO, Eno-O, B, C/D, F
O–causal EF O, GHno-O Gno-O
No-O–structural AB no-O, Dno-O,
No-O–causal EF no-O, GHO
Note: Cues B, C/D, and F are the target cues, and Cues A and E are their associates. Cues G and H are filler cues (shown in italic).
“O” indicates that the outcome was presented; “no-O” indicates that no outcome was presented.
2364 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12)
CASTRO AND MATUTE
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
GH in order to prevent participants from expect-
ing the outcome to follow all of the compound
cues. In groups no-Ocausal and no-
Ostructural, compounds AB, CD, and EF were
always followed by no Hamkaoman disease,
whereas the filler compound GH was always
followed by Hamkaoman disease; here, the
inclusion of compound GH was intended to
prevent participants from expecting no outcome
to follow all of the compound cues.
With regard to the description of the relation-
ship between the cues and the outcome, the causal
groups were told in the instructions at the begin-
ning that they had to “discover to what extent
certain new types of substances in the blood cause
Hamkaoman disease”, and that they would have
to rate the likelihood that each substance is cause of
Hamkaoman disease”. In addition, on each trial
they had to answer the question “Do you think
this patient will have Hamkaoman disease?
On the other hand, the structural groups were
told that they had to “discover to what extent
certain new types of substances in the blood are
related to Hamkaoman disease”, and that they
would have “to rate the likelihood that each
substance is indicative of Hamkaoman disease”. In
addition, on each trial they had to answer the
question “Do you think this patient has
Hamkaoman disease?
In Phase 2, Cues A, E, and G, were presented.
In all six groups, Cue A was always followed by the
outcome, whereas Cues E and G were always fol-
lowed by no outcome. Each of the cues was pre-
sented 20 times each in random order. Once
Phase 2 was completed, all groups proceeded to
the testing phase.
In testing, the eight chemical substances were
presented one by one, in a randomized order.
Each of the testing screens was headed by: Please
indicate to what extent this substance is indicative
(or cause) of Hamkaoman disease. Each of the
eight substances was presented along with a
rating scale ranging from 0 (definitively no)to8
(definitively yes). The middle point of the scale,
4, was labelled possibly. Participants had to give
their ratings by moving a slider along this scale.
Once the rating for one cue was entered,
participants advanced to the next screen, so that
they did not have the opportunity to alter their
prior ratings.
Preanalysis of the data
It would make no sense to expect any effect in
those participants who had not learned the contin-
gencies between the cues and the outcome during
Phase 2. Therefore, we eliminated from the
analysis the data of participants who did not give
a correct answer on the last trial of Cue A and
on the last trial of Cue E in Phase 2. Using this
criterion, the data from 1 participant in group
Ocausal was excluded. An alpha level of .05
was adopted for tests of statistical significance.
In addition, 95% confidence intervals (CIs) are
reported for the planned comparisons between
individual means.
Results and discussion
Ratings of all of the cues in the six groups are
shown in Table 2. Because Cues C and D received
the same training, their ratings were averaged for
statistical analyses. We refer to this average cue
as C/D. Ratings of Cue A, which had been pre-
sented followed by the outcome, were high;
ratings of Cues E and G, which had been pre-
sented followed by the occurrence of no
outcome, were low. The critical results, ratings of
the target cues B, C/D, and F, are displayed in
Figure 1 as well. The order of the cues, B .C/
D.F, suggests positive mediation in groups
exposurestructural and exposure causal. In con-
trast, the order of the cues in groups Ostructural
and Ocausal, B C/D,F, suggests negative
mediation. In groups no-O, a slight tendency to
negative mediation can be observed in group no-
Ocausal and no effect in group no-O – structural.
A 3 (outcome in Phase 1: exposure vs.
outcome vs. no outcome) ×2 (scenario: struc-
tural vs. causal) ×3 (target cue: B vs. C/D vs.
F) analysis of variance (ANOVA) revealed a sig-
nificant main effect of outcome, F(2, 143) ¼
158.44, MSE ¼4.41, p,.001, a significant
main effect of scenario, F(1, 143) ¼5.36, MSE
¼4.41, p¼.02, and a significant main effect
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12) 2365
LEARNING ABOUT ABSENT CUES
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
of cue, F(2, 286) ¼3.20, MSE ¼2.21, p¼.04.
Importantly, the Outcome ×Target Cue inter-
action was significant, F(2, 286) ¼20.00, MSE
¼2.21, p,.001, revealing that ratings of the
target cues changed differently depending on
the status of the outcome during Phase 1. The
Scenario ×Target Cue interaction was signifi-
cant as well, F(2, 286) ¼5.12, MSE ¼2.21, p
,.01, revealing that task scenario also affected
ratings of the target cues. The triple interaction
was not significant.
In group exposurestructural, the order of the
target cues, B .C/D.F, suggested positive
mediation. Planned comparisons confirmed that
ratings of Cue B were higher than ratings of Cue
F, F(1, 143) ¼26.46, MSE ¼3.49, p,.001,
CI 1.[69, 3.74]. In order to explore whether this
positive mediation was due to an increase in Cue
B or to a decrease in Cue F, we performed
additional comparisons. Ratings of Cue B were
higher than ratings of control cue C/D, F(1,
143) ¼30.30, MSE ¼1.36, p,.001, CI [1.03,
2.60], and ratings of Cue F were lower than
ratings of control cue C/D, F(1, 143) ¼5.67,
MSE ¼1.78, p¼.01, CI [0.08, 1.71]. Thus,
both an increase in the ratings of Cue B and a
decrease in the ratings of Cue F took place.
In group exposurecausal, the order of the
target cues, B .C/D.F, suggested positive
mediation as well. Planned comparisons confirmed
that ratings of Cue B were higher than ratings of
Cue F, F(1, 143) ¼7.41, MSE ¼3.49, p,
.001, CI [0.18, 3.06], so that positive mediation
did take place. As in group exposure structural,
ratings of Cue B were higher than ratings of
control cue C/D, F(1, 143) ¼10.27, MSE ¼
1.36, p¼.001, CI [0.95, 2.02], indicating an
increase in the value of Cue B. Ratings of Cue F
were not significantly lower than ratings of
control cue C/D, F(1, 143) ¼1.01, MSE ¼
1.78, p¼.3, CI [0.62, 1.38].
Both groups exposure structural and
exposurecausal exhibited positive mediation.
The specific scenario did not reverse the ordering
of the target cues, but it seems to have weakened
the effect in group exposure causal.
Figure 1. Mean final ratings of target cues B, C/D, and F for the
six groups. Error bars indicate the standard error of the means. Exp
¼exposure. Struct ¼structural. O ¼outcome.
Table 2. Mean ratings for all cues in all groups
Group
Exposure Outcome No outcome
Cue Structural Causal Structural Causal Structural Causal
A 7.96 (0.04) 7.64 (0.32) 7.76 (0.17) 7.54 (0.21) 6.76 (0.33) 5.56 (0.37)
B 5.32 (0.32) 3.96 (0.49) 5.04 (0.35) 3.91 (0.32) 1.12 (0.34) 0.16 (0.09)
C 3.48 (0.39) 2.96 (0.36) 5.00 (0.41) 4.54 (0.31) 0.84 (0.25) 0.56 (0.23)
D 3.52 (0.33) 2.84 (0.33) 4.36 (0.39) 4.66 (0.29) 1.08 (0.35) 0.44 (0.19)
E 0.68 (0.23) 0.64 (0.22) 1.64 (0.44) 0.62 (0.30) 0.76 (0.33) 0.52 (0.30)
F 2.60 (0.39) 2.52 (0.50) 5.88 (0.52) 6.29 (0.33) 1.00 (0.35) 0.96 (0.35)
G 0.04 (0.08) 0.04 (0.04) 0.76 (0.44) 0.04 (0.04) 2.96 (0.54) 2.24 (0.50)
H 3.60 (0.46) 2.28 (0.48) 0.60 (0.35) 0.54 (0.26) 5.76 (0.42) 6.84 (0.39)
Note: Means; standard errors in parentheses. Ratings of the target cues are shown in italics.
2366 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12)
CASTRO AND MATUTE
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
The direction of the changes was different for
groups Ostructural and O causal. In group
Ostructural, the target cues were ordered B ¼
C/D,F. Planned comparisons revealed that
the difference between ratings of Cue B and Cue
F was nearly significant, F(1, 143) ¼3.72, MSE
¼3.49, p¼.05, CI [2.18, 0.50]. The difference
between ratings of Cue B and Cue C/D was not
significant, F(1, 143) ¼0.54, MSE ¼1.36, p¼
.45, CI [0.33, 1.05], but the difference between
ratings of Cue C/D and Cue F was significant,
F(1, 143) ¼11.21, MSE ¼1.78, p¼.001, CI
[2.23, –0.16].
In group Ocausal, the order of the target cues,
B,C/D,F, suggested the existence of nega-
tive mediation as well. Planned comparisons con-
firmed that ratings of Cue B were lower than
ratings of Cue F, F(1, 143) ¼16.07, MSE ¼
3.49, p,.001, CI [3.18, 1.56], so negative
mediation took place. The difference between
ratings of Cue B and Cue C/D, although in the
right direction, was not significant, F(1, 143) ¼
2.66, MSE ¼1.36, p¼.1, CI [1.33, 0.44].
The difference between ratings of Cue C/D and
Cue F was significant, F(1, 143) ¼17.48, MSE
¼1.78, p,.001, CI [2.35, 1.02].
Both groups O structural and O causal
exhibited negative mediation. Again, the specific
scenario did not reverse the ordering of the
target cues, but it seems to have weakened the
effect in group O structural.
In groups no-Ostructural and no-O causal,
the target cues, B, C/D, and F were all given
very low ratings. We found a single effect of cue
in group no-Ocausal, F(2, 23) ¼2.89, MSE ¼
1.39, p,.05, suggesting backward conditioned
inhibition, but none of the planned comparisons
were significant.
As we predicted, the mere exposure to the cues
without the outcome (and with instructions that
maintain the neutral relationship between the
cues and the outcome) generates very different
results compared to pairing the cues with no
outcome. On the one hand, the ratings of target
cues in groups no-O were much lower than the
ratings of target cues in groups exposure,
suggesting that pairing the cues with no outcome
leads participants to consider them as predictors
of the nonoccurrence of the outcome, instead of
considering them as not having an association
with the outcome. On the other hand, we observed
negative mediation—albeit small—in group no-
Ocausal, but positive mediation in group
exposurecausal, suggesting that only when the
target cues are maintained in a neutral or ambigu-
ous state does training of their associates lead to a
change in the value of the target cues in the same
direction as the change of their associates. If the
target cues are predictors of the nonoccurrence of
the outcome, training of their associates leads to
a change in the value of the target cues in the
opposite direction to the change of their associates.
Thus, it seems that what is crucial to generate
positive mediation is the lack of an association
between the cues and the outcome; if an associ-
ation has been previously established, then nega-
tive mediation is the most likely result.
GENERAL DISCUSSION
We investigated whether the value of absent cues
changes in the same (e.g., positive mediation) or
in the opposite direction (e.g., negative mediation)
as the value of present cues. We found both posi-
tive and negative mediation effects. After present-
ing Cues A and B together, later pairings of Cue A
alone with the outcome generated high ratings,
not only of the A outcome association, but of
the Boutcome association as well; and, after pre-
senting Cues E and F together, later presentations
of Cue E alone with no outcome generated low
ratings, not only of the E outcome association,
but of the F outcome association as well. Thus,
the value of the absent cues B and F changed in
the same direction as the value of the present
cues A and E that were activating their represen-
tation—positive mediation. This effect has never
been previously reported in a human contingency
study.
We also found negative mediation, a result fre-
quently reported (e.g., Chapman, 1991; Dickinson
& Burke, 1996; Wasserman & Berglan, 1998). In
this case, after presentation of Cues A and B
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12) 2367
LEARNING ABOUT ABSENT CUES
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
together followed by the outcome, later pairings of
Cue A alone with the outcome generated low
ratings of the Boutcome association; and, after
presentation of Cues E and F together followed
by the outcome, later presentations of Cue E
alone with no outcome generated high ratings of
the Foutcome association. In this case, the
value of the absent cues B and F changed in the
opposite direction to the value of the present cues
A and E that were activating their representation.
Most important, we found that the status of the
outcome during early compound training influ-
enced later changes in the value of the absent
cues: Explicit information about the occurrence
or nonoccurrence of the outcome led to negative
mediation, whereas no information led to positive
mediation. This pattern of results suggests that,
when a prior association exists between the cue
and the outcome—either the cue as predictor of
outcome or the cue as predictor of no outcome—
the value of the absent cue changes in the opposite
direction to the value of the present cue, and, when
no prior association exists, the value of the absent
cue changes in the same direction as the value of
the present cue.
Excitatory association versus inhibitory
association versus no association
After developing an excitatory association with the
outcome during the earlier compound phase, the
value of the absent cue could still, at least in prin-
ciple, continue increasing during Phase 2, as it did
not reach asymptotic value during Phase 1 (in fact,
the value of its companion does increase during
Phase 2); however, no studies in human contin-
gency learning or in animal conditioning have
found an increase in the value of the absent cue
when the associate cue is trained with the
outcome after earlier training of the compound
cue with the outcome. This fact suggests that,
once a cue has acquired an excitatory association
with an outcome, it is unlikely that further training
of the associate cue alone will produce a change in
the absent cue that mimics the change in the
associate cue; that is, positive mediation is unlikely
under these conditions.
The case in which the compound is initially fol-
lowed by the absence of the outcome turns out to
be more complex because, now, either an inhibi-
tory association or no association might develop
between the cues and the outcome. Finding one
or the other result will depend on the extent to
which the occurrence of the outcome is expected
in that context (e.g., Miller et al., 1991).
Although controlling for the expectation of the
outcome is quite straightforward with animals, it
is far from simple with humans, because the pres-
entation of cues followed by no outcome, the most
usual procedure, can generate an explicit expec-
tation for the outcome in the training context
and thus an inhibitory association between the
target cue and the outcome (e.g., Karazinov &
Boakes, 2004) even when, in theory, no association
between the cue and the outcome should be
formed.
Our results show that explicit pairing of the
cues with no outcome generates a very different
result from that obtained by simply exposing the
cues and maintaining them in a neutral state in
relation to the outcome. We observed negative
mediation in group no-O causal, but positive
mediation in group exposure causal. This
pattern of results suggests that the target cue
should have no association with the outcome—
that is, it should neither predict the presence of
the outcome nor predict the absence of the
outcome, in order to later change its value in the
same direction as its associate cue. Once a cue
has acquired positive or negative value, it seems
that it will change in the opposite direction to its
associate cue.
The absence of reports of backward conditioned
inhibition with animals supports our view. Only
when another cue or the context elicits an expec-
tation of the occurrence of the outcome will a
cue that is presented without the outcome
become inhibitory (e.g., Miller et al., 1991).
Therefore, it seems that animals exhibit positive
mediation to Cue B after presentations of the
compound AB in Phase 1 followed by pairings of
Cue A with the outcome in Phase 2 (e.g.,
sensory preconditioning), because during Phase 1
there is nothing that can elicit the expectation of
2368 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12)
CASTRO AND MATUTE
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
the occurrence of the outcome. Without an expec-
tation of the outcome in Phase 1, no backward
conditioned inhibition occurs.
In most backward conditioned inhibition
studies with humans, the absence of the outcome
is made explicit (the cues are paired with no
outcome) or the possibility of occurrence of the
outcome is mentioned in the initial instructions
(thus creating an excitatory context). These pro-
cedures lead to the cues becoming inhibitory
after compound training in Phase 1. The only
exception seems to be Urcelay, Perelmuter, and
Miller’s (2008) study, who did not give any infor-
mation about the outcome until the beginning of
the second training phase. Urcelay et al. were con-
cerned about the common practice of pairing the
cues with no outcome in backward conditioned
inhibition studies and, in general, about the
effect of informing the participants, before the
start of training, about the outcome. So, Urcelay
et al. presented their participants with a contin-
gency task in which participants were told simply
to observe the events appearing on the screen. In
the first phase, two geometrical shapes of different
colour, A and X, appeared side by side; in the
second phase, only one of these shapes, A, was pre-
sented, and immediately after its disappearance, a
cross-eyed human baby (the outcome) appeared.
Later, the results showed that X had become a
conditioned inhibitor. That is, its value had
changed in the direction opposite to that of the
present cue. According to our argument, X
should not develop any association with the
outcome in Phase 1 (and thus, should not
become an inhibitor later on), because the cues
are presented on their own, and no information
is provided about the nonoccurrence of the
outcome.
However, we believe that the finding by Urcelay
et al. (2008) can possibly be understood in terms of
the participants’ later rehearsal (and revaluation) of
the compound trials presented in Phase
1. Participants could mentally present themselves
with AX and A outcome episodes while they
actually experienced only the A outcome trials
of Phase 2, so that their training would be close
to real training with AX and A outcome trials
intermixed (Chapman, 1991; see also Melchers
et al., 2004). Urcelay et al.’s simple design (training
consisted of only three types of trial) may have
rendered rehearsing particularly easy. This process
may thus have prompted participants to infer
during Phase 2 that it must have been the target
cue X that was preventing the outcome from
happening during Phase 1; thus, the formation of
an inhibitory association between X and the
outcome could be readily explained.
Structural versus causal and elemental versus
configural
Although we did not find that our cover story
could switch ratings of the target cues from posi-
tive mediation to negative mediation, we observed
that the positive mediation effect was largest when
the structural task was used, whereas negative
mediation was largest when the causal task was
used. Although, as we mentioned in the
Introduction, the symptoms disease task could
be understood as a diagnostic task, this possibility
seems unlikely in our study. On the one hand,
there are no reports in the contingency judgement
literature that suggest the possibility of positive
mediation in a diagnostic setting (just failures or
difficulties to obtain negative mediation; Tangen
& Allan, 2004; Waldmann, 2000; Waldmann &
Holyoak, 1992). On the other hand, our instruc-
tions emphasized a structural relationship
between the cues and the outcome. Participants
had to learn how the cues and the outcome were
“related” and to rate the likelihood of each cue
being “indicative” of the outcome; cause effect
terms were never used. Thus, under these
circumstances, it seems unlikely that the task was
understood as a diagnostic task; the structural
conceptualization seems to be more adequate.
In a structural relationship, because all of the
events are part of the same entity, it makes sense
to assume that what happens to one element of
the entity can be extended to the other elements;
if that is the case, then positive mediation should
be more likely to be observed. On the other
hand, when the scenario entails a causal relation-
ship, several authors have pointed out that
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12) 2369
LEARNING ABOUT ABSENT CUES
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
people normally select one among many potential
causes as the cause of the event (Hesslow, 1983;
Hilton & Slugoski, 1986; Mackie, 1974; McGill,
1989). Discounting of the causal value of one cue
in favour of a stronger cue leads to negative
mediation. But, this process can happen only if
cues within the compound are processed as indi-
vidual and independent elements.
Actually, although negative mediation effects
have been repeatedly observed in human contin-
gency judgements (i.e., Chapman, 1991;
Dickinson & Burke, 1996; Larkin et al., 1998;
Melchers et al., 2004; Shanks, 1985; Wasserman
& Berglan, 1998; Wasserman & Castro, 2005;
Wasserman et al., 1996), there are several studies
that have failed to show this effect (i.e., De
Houwer, Beckers, & Glautier, 2002; Waldmann,
2000; Waldmann & Holyoak, 1992; Williams
et al., 1994). One of the explanations for these
diverse results is that, sometimes, compound cues
can be processed as elements and, other times, as
configurations. Compound cues that are processed
configurally are less subject to negative mediation
than compound cues that are processed elemen-
tally (Williams et al., 1994). In order to find nega-
tive mediation, elemental processing is necessary,
because if the cues are perceived as part of a con-
figuration, then their values might not be
assumed to change independently of each other.
2
Several factors can influence whether two cues
that are presented in compound will be encoded
as two elements or as a unique configuration (for
a review, see Melchers, Shanks, & Lachnit,
2008). If participants are pretrained with elemen-
tally solvable tasks (Mehta & Williams, 2002;
Melchers, Lachnit, U
¨ngo
¨r, & Shanks, 2005a;
Williams et al., 1994) or are given instructions
that encourage them to encode the cues as separate
entities (Williams et al., 1994), elemental encod-
ing, and therefore negative mediation, is more
likely. Grouping and spatial location of the cues
can generate a difference as well. Glautier (2002)
used colours and symbols on cards as cues and
found negative mediation only when the cues
were spatially separated and ungrouped, but not
if they were spatially close together (see also
Livesey & Boakes, 2004). All of these studies
found that an elemental strategy encourages nega-
tive mediation, but none of them reported any
sign of positive mediation when participants
were using a configural strategy. However, a
more recent study by Liljeholm and Balleine
(2009) did report opposite effects. They found
that high similarity and no spatial separation of
the elements of the compound promoted general-
ization between them, so positive mediation was
observed; on the other hand, low similarity and
spatial separation of the elements of the compound
led to negative mediation.
In relation to our study, it is likely that this
difference between elemental and configural
encoding could underlie the effect that the use of
a structural and a causal scenario had: The structural
scenario should probably prompt participants to
process cues configurally whereas the causal scen-
ario should possibly favour an elemental strategy.
Theoretical explanations
Opposite changes in the value of absent cues pose a
critical challenge to learning theories because, in
general, models that can explain one directional
effect cannot explain the opposite one. By way of
example, consider Wagner’s (1981) SOP model.
According to Wagner’s model, cues and outcomes
are represented by nodes that consist of several
elements. The elements in a node can be in an
inactive state (I) or in one of two possible acti-
vation states: a primary active state (A1), when a
stimulus is present, or a secondary active state
(A2), when a stimulus is not present but its
representation is activated by means of another
stimulus with which it was previously paired. An
excitatory association between a cue and an
outcome will be formed when both are present—
that is, when their elements are in the A1 state.
2
It has to be noted that Pearce’s (1987, 1994) configural theory does predict negative mediation but, even when this theory
assumes that stimuli are processed configurally, it uses elemental information to determine the level of generalization between
two configurations.
2370 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12)
CASTRO AND MATUTE
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
But, a stimulus in the A2 state cannot enter into an
association with another stimulus. Thus, an absent
cue activated in the A2 state is not allowed to
change its associative strength.
Dickinson and Burke (1996) extended Wagner’s
(1981) SOP model so that it could explain negative
mediation effects. According to Dickinson and
Burke, an excitatory association may develop from
a stimulus in A1 or A2 to another concurrent
stimulus in the same activation state, and an inhibi-
tory association may develop when two concurrent
stimuli are in different states (see also Aitken &
Dickinson, 2005). However, sometimes some
elements of a stimulus may be in A1 whereas
other elements are in A2, in which case whether
an excitatory or an inhibitory association develops
will depend on which of the processes is stronger.
In our case, after pairing the AB compound with
the outcome, when Cue A is presented in Phase 2
paired with the outcome as well, it should activate
both Cue B and the outcome in A2. Because the
outcome is present, some elements of its represen-
tation will be also in A1. Then, absent Cue B will be
involved in two learning processes: one excitatory,
consequence of Cue B’s activation in A2 and the
activation in A2 of some elements of the
outcome, and another inhibitory, consequence of
Cue B’s activation in A2 and the activation in A1
of some elements of the outcome. Thus, according
to this model, a weak negative mediation effect may
or may not occur (and, potentially, a positive
mediation effect as well) depending on which of
these two processes is stronger (Aitken &
Dickinson, 2005; Larkin et al., 1998). Indeed, we
found a weak backward blocking effect (ratings to
Cue B were lower than ratings to Cue C/Din
group Ocausal, although this difference fell
short of statistical significance), a result that sup-
ports this view, although robust backward blocking
has been reported as well (e.g., Shanks, 1985;
Wasserman & Berglan, 1998; Wasserman &
Castro, 2005).
But, Dickinson and Burke’s (1996) modified
SOP cannot predict the formation of an excitatory
association between Cue B and the outcome when
Cues A and B have been paired together without
the outcome. In this case, when Cue A is paired
with the outcome in Phase 2, Cue A will activate
the representation of Cue B in A2; because all of
the elements of the outcome are in A1, an inhibi-
tory association between B and the outcome
should form. So, Dickinson and Burke’s modified
SOP cannot explain positive mediation under
these conditions.
P. C. Holland (1983) proposed a different
modification of the SOP model. According to
Holland, when the elements of a representation
are in A2, an excitatory association should form
with another representation in A1; on the other
hand, when two concurrent representations are in
A2, an inhibitory association should develop.
This modification can readily explain the opposite
effects due to the presence or absence of the
outcome in the first phase. When Cues A and B
have been paired together without the outcome,
and Cue A is later paired with the outcome, Cue
A will activate the representation of Cue B in
A2; because all of the elements of the outcome
are in A1, an excitatory B outcome association
will be formed (albeit weaker than if the elements
of Cue B were also in A1). Thus, positive
mediation should be observed. When Cues A
and B have been paired together with the
outcome, later presentations of Cue A alone will
activate the representation of Cue B in A2, but
also the representation of the outcome in A2. In
this case, the associative value of Cue B could
change in two ways because the outcome will
have elements in both A2 and A1. Parameters of
the task will then determine which association
will be stronger.
However, P. C. Holland’s (1983) revision of the
SOP model fails to explain why Cue F increases its
associative value when the EF compound is paired
with the outcome, and, later, Cue E is presented
without the outcome. In this condition, Cue E
alone is presented in Phase 2, and both Cue F
and the outcome will be activated in A2.
According to Holland’s revision, an inhibitory
association will always develop, because the
outcome is not present, and its elements do not
move into the A1 state. An increment in the
value of Cue F (as we observed in Ocausal and
Ostructural groups) would never be expected.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12) 2371
LEARNING ABOUT ABSENT CUES
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
Thus, neither the original SOP nor any of its revi-
sions can explain our entire pattern of results.
More promising turns out to be the sometimes-
competing retrieval (SOCR) model recently pro-
posed by Stout and Miller (2007), a formalized
version of Miller and his colleagues’ comparator
hypothesis (Denniston, Savastano, & Miller,
2001; Miller & Matzel, 1988). According to the
SOCR model, cues can activate the outcome
directly (when they have been paired with the
outcome) or indirectly (when they activate other
associate cues, which in turn do activate the
outcome directly). Organisms need to learn to
differentiate these two types of activation. If the
target cue B has at some point been paired with
the outcome, then participants can readily dis-
tinguish that the representation of the outcome is
being directly activated by Cue B. But, if the
target cue B has never been paired with the
outcome, and its associate Cue A has, then partici-
pants might not distinguish whether Cue B acti-
vates the outcome directly or indirectly via the
association of its associate Cue A with the
outcome. This situation is more likely to happen
when participants have little experience with Cue
B. As training proceeds, participants learn when
the representation of the outcome is being directly
and indirectly activated.
In the case of our experiments, when Cues A and
B are paired together with the outcome, and later
Cue A is paired with the outcome, the SOCR
model predicts that Cue B will activate the
outcome directly when presented later at testing.
Because the direct association between B and the
outcome is not as strong as the association
between A and the outcome (that has been paired
alone with the outcome), low responding to B
(that is, negative mediation) will be observed at
testing. In the case of Cues A and B being paired
together without the outcome, and later Cue A
being paired with the outcome, Cue B has never
been paired with the outcome; thus, when Cue B
is presented at testing it can activate the represen-
tation of the outcome only indirectly (because of
its association with Cue A). Because of the lack
of experience with Cue B alone, participants
cannot distinguish what exactly is causing the
activation of the outcome, and high responding to
B (that is, positive mediation) will be observed at
testing. In short, if the target cue has been pre-
viously paired with the outcome, then negative
mediation should be observed; if the target cue
has never been paired with the outcome, then
either negative mediation or positive mediation
may be observed, depending on the amount of
experience with the target cue alone.
Interestingly, the SOCR’s means of explaining
opposite effects may even embrace the differences
generated by elemental and configural processing.
At the beginning of training, organisms might
tend to perceive the stimuli presented in com-
pound as a unique entity because they do not
have much experience with the components that
combine to create the compound, so cue facili-
tation should be favoured, but, with further
training, organisms can learn to discriminate the
elements within the compound, so that negative
mediation could be favoured.
The SOCR model is a performance model; it
focuses on information processing at the testing
stage, as compared to the SOP model and its
modifications (and models like Mackintosh,
1975; Rescorla & Wagner, 1972; Van Hamme &
Wasserman, 1994), which emphasize information
processing during the learning stages. The fact
that the SOCR seems to be the most adequate
model to explain our results suggests, therefore,
that maybe the positive and negative mediation
effects are better understood as performance
rather than acquisition effects. If that is the case,
we should conclude that the response to absent
cues can change both in the same and in the oppo-
site direction as the response to present cues, and
the direction of this change seems to depend on
whether or not the absent cue has a prior associ-
ation with the outcome. Our experiment did not
try to address whether the effect takes place
during learning or subsequent responding, but it
will be something certainly worth studying in the
future.
Original manuscript received 18 May 2009
Accepted revision received 23 February 2010
First published online 5 July 2010
2372 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12)
CASTRO AND MATUTE
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
REFERENCES
Aitken, M. R. F., & Dickinson, A. (2005). Simulations
of a modified SOP model applied to retrospective
revaluation of human causal learning. Learning &
Behavior,33, 147–159.
Arcediano, F., Matute, H., Escobar, M., & Miller, R. R.
(2005). Competition between antecedent and sub-
sequent stimuli. Journal of Experimental Psychology:
Learning, Memory, and Cognition,31, 228–237.
Baker, A. G., Murphy, R. A., Mehta, R., & Baetu, I.
(2005). Mental models of causation: A comparative
view. In A.J. Wills (Ed.), New directions in human
associative learning (pp. 11–40). Mahwah, NJ:
Lawrence Erlbaum Associates.
Brogden, W. J. (1939). Sensory pre-conditioning.
Journal of Experimental Psychology,25, 323–332.
Brogden, W. J. (1947). Sensory preconditioning of
human subjects. Journal of Experimental Psychology,
37, 527–539.
Chapman, G. B. (1991). Trial order affects cue inter-
action in contingency judgment. Journal of
Experimental Psychology: Learning, Memory, and
Cognition,17, 837–854.
Cobos, P. L., Lo
´pez, F. J., Can
˜o, A., Almaraz, J., &
Shanks, D. R. (2002). Mechanisms of predictive
and diagnostic causal induction. Journal of
Experimental Psychology: Animal Behavior Processes,
28, 331–346.
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., Savastano, H. I., & Miller, R. R.
(2001). The extended comparator hypothesis:
Learning by contiguity, responding by relative
strength. In R. R. Mowrer & S. B. Klein (Eds.),
Handbook of contemporary learning theories
(pp. 65–117). Hillsdale, NJ: Lawrence Erlbaum
Associates.
Dickinson, A., & Burke, J. (1996). Within-compound
associations mediate the retrospective revaluation of
causality judgements. Quarterly Journal of
Experimental Psychology,49B, 60–80.
Glautier, S. (2002). Spatial separation of target and
competitor cues enhances blocking of human causal-
ity judgements. Quarterly Journal of Experimental
Psychology,55B, 121–135.
Graham, S. (1999). Retrospective revaluation and
inhibitory associations: Does perceptual learning
modulate our perception of the contingencies
between events? Quarterly Journal of Experimental
Psychology,52B, 159–185.
Hall, G. (1996). Learning about associatively activated
stimulus representations: Implications for acquired
equivalence and perceptual learning. Animal
Learning & Behavior,24, 233–255.
Harris, J. A., & Westbrook, R. F. (1998). Retroactive
revaluation of an odortaste association. Animal
Learning and Behavior,26, 326–335.
Hesslow, G. (1983). Explaining differences and weight-
ing causes. Theoria,49, 87–111.
Hilton, D. J., & Slugoski, B. R. (1986). Knowledge-
based causal attributions: The abnormal conditions
focus model. Psychological Review,93, 75–88.
Holland, J. H., Holyoak, K., Nisbett, R. E., & Thagard,
P. R. (1986). Induction: Processes of inference, learning,
and discovery. Cambridge, MA: MIT Press.
Holland, P. C. (1983). Representation mediated over-
shadowing and potentiation of conditioned aver-
sions. Journal of Experimental Psychology: Animal
Behavior Processes,9, 1–13.
Karazinov, D. M., & Boakes, R. A. (2004). Learning
about cues that prevent an outcome: Conditioned
inhibition and differential inhibition in human pre-
dictive learning. Quarterly Journal of Experimental
Psychology,57B, 153–178.
Kaufman, M. A., & Bolles, R. C. (1981). A nonassocia-
tive aspect of overshadowing. Bulletin of the
Psychonomic Society,18, 318–320.
Konorski, J. (1967). Integrative activity of the brain: An
interdisciplinary approach. Chicago: University of
Chicago Press.
Larkin, M. J. W., Aitken, M. R. F., & Dickinson, A.
(1998). Retrospective revaluation under positive
and negative contingencies. Journal of Experimental
Psychology: Learning, Memory, and Cognition,24,
1331–1352.
Liljeholm, M., & Balleine, B. W. (2009). Mediated
conditioning vs. retrospective revaluation in
humans: The influence of physical and functional
similarity of cues. Quarterly Journal of Experimental
Psychology,62, 470–482.
Livesey, E. J., & Boakes, R. A. (2004). Outcome addi-
tivity, elemental processing and blocking in human
causality judgements. Quarterly Journal of
Experimental Psychology,57B, 361–379.
Mackie, J. L. (1974). The cement of the universe: A study of
causation. Oxford, UK: Clarendon Press.
Mackintosh, N. J. (1975). A theory of attention:
Variations in the associability of stimuli with
reinforcement. Psychological Review,82, 276–298.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12) 2373
LEARNING ABOUT ABSENT CUES
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
Matute, H., Arcediano, F., & Miller, R. R. (1996). Test
question modulates cue competition between causes
and between effects. Journal of Experimental
Psychology: Learning, Memory, and Cognition,22,
182–196.
McGill, A. L. (1989). Context effects in judgments of
causation. Journal of Personality & Social Psychology,
57, 189–200.
Mehta, R., & Williams, D. A. (2002). Elemental and
configural processing of novel cues in deterministic
and probabilistic tasks. Learning and Motivation,
33, 456–484.
Melchers, K. G., Lachnit, H., & Shanks, D. R. (2004).
Within-compound associations in retrospective reva-
luation and in direct learning: A challenge for com-
parator theory. Quarterly Journal of Experimental
Psychology,57B, 25–53.
Melchers, K. G., Lachnit, H., U
¨ngo
¨r, M., & Shanks, D.
R. (2005a). Prior experience can influence whether
the whole is different from the sum of its parts.
Learning and Motivation,36, 20–41.
Melchers, K. G., Shanks, D. R., & Lachnit, H. (2008).
Stimulus coding in human associative learning:
Flexible representations of parts and wholes.
Behavioural Processes,77, 413–427.
Melchers, K. G., U
¨ngo
¨r, M., & Lachnit, H. (2005b).
The experimental task influences cue competition
in human causal learning. Journal of Experimental
Psychology: Animal Behavior Processes,31, 477–483.
Miller, R. R., Hallam, S. C., Hong, J. Y., & Dufore, D.
S. (1991). Associative structure of differential inhi-
bition: Implications for models of conditioned inhi-
bition. Journal of Experimental Psychology: Animal
Behavior Processes,17, 141–150.
Miller, R. R., & Matute, H. (1996). Biological signifi-
cance in forward and backward blocking:
Resolution of a discrepancy between animal con-
ditioning and human causal judgment. Journal of
Experimental Psychology: General,125, 370–386.
Miller, R. R., & Matzel, L. D. (1988). The comparator
hypothesis: A response rule for the expression of
associations. In G. Bower H. (Ed.), The psychology
of learning and motivation (Vol. 22, pp. 51–92).
San Diego, CA: Academic Press.
Pearce, J. M. (1987). A model for stimulus generaliz-
ation in Pavlovian conditioning. Psychological
Review,94, 61–73.
Pearce, J. M. (1994). Similarity and discrimination: A
selective review and a connectionist model.
Psychological Review,101, 587–607.
Price, P. C., & Yates, J. F. (1995). Associative and rule-
based accounts of cue interaction in contingency
judgment. Journal of Experimental Psychology:
Learning, Memory, and Cognition,21, 1639–1655.
Rescorla, R. A., & Wagner, A. R. (1972). A theory of
Pavlovian conditioning: Variations in the effective-
ness of reinforcement and nonreinforcement. In
A. H. Black & W. Prokasy F. (Eds.), Classical con-
ditioning II: Current research and theory (pp. 64–99).
New York: Appleton-Century-Crofts.
Shanks, D. R. (1985). Forward and backward blocking
in human contingency judgement. Quarterly Journal
of Experimental Psychology,37B, 1–21.
Shanks, D. R. (1991). Categorization by a connectionist
network. Journal of Experimental Psychology:
Learning, Memory, and Cognition,17, 433–443.
Shanks, D. R. (1995). The psychology of associative
learning. Cambridge, UK: Cambridge University
Press.
Shanks, D. R., & Lo
´pez, F. J. (1996). Causal order does
not affect cue selection in human associative learn-
ing. Memory & Cognition,24, 511–522.
Stout, S. C., & Miller, R. R. (2007). Sometimes-
competing retrieval (SOCR): A formalization of
the comparator hypothesis. Psychological Review,
114, 759–783.
Tangen, J. M., & Allan, L. G. (2004). Cue interaction
and judgments of causality: Contributions of causal
and associative processes. Memory & Cognition,32,
107–124.
Urcelay, G. P., Perelmuter, O., & Miller, R. R. (2008).
Pavlovian backward conditioned inhibition in
humans: Summation and retardation tests.
Behavioural Processes,77, 299–305.
Van Hamme, L. J., & Wasserman, E. A. (1994). Cue
competition in causality judgments: The role of
nonpresentation of compound stimulus elements.
Learning and Motivation,25, 127–151.
Vansteenwegen, D., Crombez, G., Baeyens, F.,
Hermans, D., & Eelen, P. (2000). Pre-extinction
of sensory preconditioned electrodermal activity.
Quarterly Journal of Experimental Psychology,53B,
359–371.
Wagner, A. R. (1981). SOP: A model of automatic
memory processing in animal behavior. In N. E.
Spear & R. Miller R. (Eds.), Information processing
in animals: Memory mechanisms (pp. 5–47).
Hillsdale, NJ: Lawrence Erlbaum Associates.
Waldmann, M. R. (2000). Competition among causes
but not effects in predictive and diagnostic learning.
2374 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12)
CASTRO AND MATUTE
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
Journal of Experimental Psychology: Learning, Memory,
and Cognition,26, 53–76.
Waldmann, M. R., & Holyoak, K. J. (1992). Predictive
and diagnostic learning within causal models:
Asymmetries in cue competition. Journal of
Experimental Psychology: General,121, 222–236.
Wasserman, E. A. (1990). Attribution of causality to
common and distinctive elements of compound
stimuli. Psychological Science,1, 298–302.
Wasserman, E. A., & Berglan, L. R. (1998). Backward
blocking and recovery from overshadowing in human
causal judgment: The role of within-compound
associations. Quarterly Journal of Experimental
Psychology,51B, 121–138.
Wasserman, E. A., & Castro, L. (2005). Surprise and
change: Variations in the strength of present and
absent cues in causal learning. Learning &
Behavior,33, 131–146.
Wasserman, E. A., Kao, S.-F., Van Hamme, L. J.,
Katagiri, M., & Young, M. E. (1996). Causation
and association. In D. R. Shanks, K. J. Holyoak, &
D. Medin L. (Eds.), The psychology of learning and
motivation (Vol. 34, pp. 207–264). San Diego,
CA: Academic Press.
Williams, D. A., & Docking, G. L. (1995). Associative
and normative accounts of negative transfer.
Quarterly Journal of Experimental Psychology,48A,
976–998.
Williams, D. A., Sagness, K. E., & McPhee, J. E. (1994).
Configural and elemental strategies in predictive
learning. Journal of Experimental Psychology:
Learning, Memory, and Cognition,20, 694–709.
THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2010, 63 (12) 2375
LEARNING ABOUT ABSENT CUES
Downloaded By: [Universidad de Deusto] At: 08:20 21 December 2010
... In other words, Pb exposure was associated with decreasing EMOCI score, while increasing DNA methylation at cg02901723 was associated with an increase in EMOCI score. When there is a larger magnitude of direct effect than total effect, or if the indirect effect is in the opposite direction as the direct effect, this suggests a suppressive effect by the mediator on the direct effect between exposure and outcome (55). Similarly, cg02901723 DNA methylation mediated the effect of T2 Pb exposure on 24-month ORIEN score [b Indirect ¼ 4.44 (À0.09, 10.68), P ¼ 0.06]. ...
... We provide preliminary evidence for both mediating and suppressive roles of DNA methylation in these relationships, depending on the locus. The latter of which can be interpreted as DNA methylation at a given gene suppressing the effect of Pb on a neurodevelopmental outcome (55). ...
Article
Full-text available
Early-life lead (Pb) exposure has been linked to adverse neurodevelopmental outcomes. Recent evidence has indicated a critical role of DNA methylation (DNAm) in cognition, and Pb exposure has also been shown to alter DNAm. However, it is unknown whether DNAm is part of the mechanism of Pb neurotoxicity. This longitudinal study investigated the associations between trimester-specific (T1, T2, and T3) maternal blood Pb concentrations, gene-specific DNAm in umbilical cord blood, and infant neurodevelopmental outcomes at 12 and 24 months of age (mental development index, psychomotor development index, and behavioral rating scale of orientation/engagement and emotional regulation) among 85 mother–infant pairs from the Early Life Exposure in Mexico to Environmental Toxicants (ELEMENT) study. In the mediation analysis for this pilot study, P < 0.1 was considered significant. DNAm at a locus in CCSER1 (probe ID cg02901723) mediated the association between T2 Pb on 24-month orientation/engagement [indirect effect estimate 4.44, 95% confidence interval (−0.09, 10.68), P = 0.06] and emotional regulation [3.62 (−0.05, 8.69), P = 0.05]. Cg18515027 (GCNT1) DNAm mediated the association of T1 Pb [−4.94 (−10.6, −0.77), P = 0.01] and T2 Pb [−3.52 (−8.09, −0.36), P = 0.02] with 24-month EMOCI, but there was a positive indirect effect estimate between T2 Pb and 24-month psychomotor development index [1.25 (−0.11, 3.32), P = 0.09]. The indirect effect was significant for cg19703494 (TRAPPC6A) DNAm in the association between T2 Pb and 24-month mental development index [1.54 (0, 3.87), P = 0.05]. There was also an indirect effect of cg23280166 (VPS11) DNAm on T3 Pb and 24-month EMOCI [2.43 (−0.16, 6.38), P = 0.08]. These associations provide preliminary evidence for gene-specific DNAm as mediators between prenatal Pb and adverse cognitive outcomes in offspring.
... Moreover, we found that motivation (intrinsic, autonomous, and controlled) successfully mediated the link between employees' perceptions of TL and work engagement. Interestingly, by negatively mediating the relationship between employees' perceptions of TL and work engagement, controlled motivation reduced the effect of employees' TL perceptions on work engagement (Castro & Matute, 2010). Finally, it is not surprising that amotivation was not a successful mediator. ...
Article
While research on the role of employees’ characteristics as core to transformational leadership theory is burgeoning, limited research has focused on the differing aspects of employees’ self-determined motivation as mediating mechanism through which transformational leadership may impact outcomes. Drawing on the self-determined theory of motivation, we build and test a theoretical model linking employees’ perceptions of transformational leadership with engagement through an intervening variable of differing aspects of employees’ self-determined motivation. Data from a sample of 155 participants revealed that employees’ perceptions of transformational leadership were positively related to employees’ self-determined motivation (intrinsic, autonomous, and controlled) and work engagement. Specifically, self-determined motivation (intrinsic, autonomous) was positively linked with work engagement while intrinsic, autonomous and controlled dimensions of self-determined motivation mediated the relationship between transformational leadership and work engagement. We discuss the theoretical and practical implications of the results.
Thesis
The health impacts of prenatal Pb exposure can be framed with the Developmental Origins of Health and Disease (DOHaD) theory, which postulates that exposure to environmental factors (i.e., diet, chemicals, stress, etc.) during sensitive periods of development (i.e., pre-conception, gestation, neonatal, etc.) alters an organism’s molecular biology, physiology, and metabolism, potentially leading to myriad effects on cognition, growth, and maturation. Around the world, exposure to lead (Pb) continues to be a persistent problem. A particularly vulnerable population includes pregnant women, where Pb from both current and past exposures can affect their developing fetuses. Disruptions in normal epigenetic processes during gestation impact gene regulation and subsequent outcomes. Accumulating evidence for DOHaD includes potential mechanisms such as epigenetic reprogramming via DNA methylation (5mC) and, less studied, DNA hydroxymethylation (5hmC), linking developmental exposures to later-life outcomes. This dissertation aims to examine whether Pb exposure influences DNA 5mC and 5hmC and susceptibility to early cognitive delays. A human longitudinal cohort, Early Life Exposure in Mexico to Environmental Toxicants (ELEMENT), was used to investigate whether developmental Pb exposure would alter epigenetics at birth and in adolescence, and if these alterations are associated with decreases in neurodevelopmental scores in early infancy. The first aim of this dissertation characterized the association between prenatal Pb exposure measured in maternal blood and bone and CpG site-specific DNA methylation at >750,000 loci at birth in infant umbilical cord blood cells. Linear regression models were used to identify differentially methylated CpG sites by prenatal Pb exposure, using the false discovery rate method to correct for multiple testing. In the second aim, we evaluated both gene-specific DNA 5mC and 5hmC in adolescent whole blood from the same cohort to determine whether the association of prenatal Pb exposure and epigenetics persisted into adolescence and determine if 5hmC is associated with the exposure. Finally, in the third aim we assessed whether DNA methylation at birth mediates the association between prenatal Pb exposure and adverse neurodevelopmental outcomes at 12- and 24-months of age. In Aim 1, we found that trimester-specific developmental Pb exposures were associated with gene-specific DNA methylation changes at birth, and most changes were identified in genes related to neurodevelopment. Next, we showed that some of these Pb-induced DNA methylation changes were stable into adolescence, particularly those associated with first trimester exposure. This potentially suggests that perturbations occurring during the first trimester are reprogramming the epigenome and these changes remain into adolescence. Within Aim 2, we observed widespread, variable 5hmC levels in adolescent whole blood. We also revealed that prenatal Pb exposure has independent associations with both 5mC and 5hmC. These associations were also sex-specific at some loci, indicating a sexually dimorphic effect on the epigenome. Finally, the Aim 3 mediation analysis showed DNA methylation mediated the association between second trimester Pb exposure and cognition scores at 24-months of age at multiple neurodevelopmental measures. Overall, this dissertation illustrates that prenatal exposures to Pb can modulate epigenetics, and this may be one mechanism linking exposure to persistent effects on development and offspring health. Additional studies in larger cohorts or animal studies are needed in order to confirm these findings. These results emphasize the importance of distinguishing between 5mC and 5hmC during epigenetic studies, thus presenting a new conceivable mechanism describing the underlying biology behind the association between exposure and outcome, 5hmC
Article
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.
Article
Ciiven the task of di the source of a patient's aUer^'ic reav-tion. college students jiuigcii the causal efficacy of common (A') and distinctive (A and Bj elements of compound stimuli: AX and BX. As the differential correlation of AX and BX with the occurrence and nonoccurrence ofthe allergic reaction rose from .00 to 1.00. ratings of ihe distinctive A and B elements diverged; most importantly, ratings ofthe common X element fell. These causal judgments of humans closely parallel the conditioned responses of animals in associa-tive learning studies, and clearly disclose that stimuli compete with one another for control over behavior.
Article
It has recently been suggested that patients with semantic breakdown may show the phenomenon of so-called "naming without semantics". If substantiated, this finding would clearly have a major impact on theories of face and object processing, all of which assume that access to semantic knowledge is a prerequisite for successful naming. In order to investigate this issue, we studied recognition, identification (the ability to provide accurate information), and naming of 50 famous faces by 24 patients with mild to moderate dementia of Alzheimer type (DAT) and 30 age-matched controls. The DAT group was impaired in all three conditions. An analysis of the concordance between identification and naming by each patient, for each stimulus item, established that naming a famous face was possible only with semantic knowledge sufficient to identify the person. Our data support the hypothesis that naming is not possible unless semantic information associated with the target is available. Naming without semantics, therefore, did not occur in patients with DAT. By contrast, there were 206 instances (17% of the total responses) in which the patients were able to provide detailed, accurate identifying information yet were unable to name the person represented. The implication of these findings far models of face identification and naming are discussed.
Article
Dickinson and Burke (1996) proposed a modified version of Wagner's (1981) SOP associative theory to explain retrospective revaluation of human causal judgments. In this modified SOP (MSOP), excitatory learning occurs when cue and outcome representations are either both directly activated or both associatively activated. By contrast, inhibitory learning occurs when one representation is directly activated while the other is associatively activated. Finite node simulations of MSOP yielded simple acquisition, overshadowing, blocking, and inhibitory learning under forward contingencies. Importantly, retrospective revaluation was predicted in the form of unovershadowing and backward inhibitory learning. However, MSOP did not yield backward blocking. These predictions are evaluated against the relevant empirical evidence anti contrasted with the predictions of other associative theories that have been applied to retrospective revaluation of human causal and predictive learning.
Article
It is hypothesized that causal explanations for an occurrence vary as a function of the causal background against which the occurrence is considered. Three experiments are presented that test propositions regarding the operation of the causal background in the selection of causal explanations. Findings indicate that factors previously shown to affect subjects' attributions-specifically, role (actor vs. observer), covariation information (consensus and distinctiveness), and quality of performance (positive vs. negative)-may do so by guiding subjects' selection of a causal background. Evidence indicates that these factors may not have the predicted effect on subjects' attributions when competing cues, such as context or wording of the causal question, suggest the relevance of conflicting causal backgrounds.
Article
Retrospective revaluation of causal judgments was investigated in a 2-stage procedure. In the 1st stage, compounds of 2 cues were associated with the outcome, whereas in the 2nd stage, a cue from each compound was trained by itself. Associating this cue with the outcome in the 2nd stage had no detectable effect on the causal rating of the other cue from the compound, whereas presenting it without the outcome enhanced the causal rating of the other cue. The retrospective revaluation of the causal rating of these productive cues and also of preventative cues depended on consistent pairing of the cues during compound training, suggesting a role for within-compound associations. These results favor associative accounts of retrospective revaluation that use separate excitatory and inhibitory learning processes rather than a general error-correcting learning algorithm.
Article
Three experiments tested a simple connectionist network approach to human categorization. The specific network considered consists of a layer of input nodes, each representing a feature of the exemplar to be categorized, connected in parallel to a layer of output nodes representing the categories. Learning to categorize exemplars consists of adjusting the weights in the network so as to increase the probability of making correct categorizations; weight changes are determined by the Rescorla-Wagner (1972) learning rule. The experiments used a simulated medical diagnosis procedure in which subjects have to decide which disease (the category) each of a series of patients is suffering from on the basis of the patients' symptoms (the features). After a series of trials, the subjects rated the extent to which particular symptoms were associated with particular diseases. In each of the experiments, it is shown that a process of selective learning occurs in this categorization task and that selection in turn depends on the relative predictiveness of the symptom for the disease. Such effects parallel results found in animal conditioning experiments and are readily reproduced by the connectionist network model. The results are also discussed in terms of a variety of traditional theories of categorization.