ArticlePDF Available

Revisiting the role of within-compound associations in cue-interaction phenomena

Authors:

Abstract and Figures

Although it is thought that within-compound associations are necessary for the occurrence of both backward blocking and unovershadowing, it is not known whether this variable plays a similar role in mediating the two phenomena. Similarly, the roles of within-compound associations in forward blocking and in reduced overshadowing have not been tested independently. The present experiments evaluated how the strength of within-compound associations affects backward blocking, unovershadowing, forward blocking, and reduced overshadowing. Using an allergy task, the strength of within-compound associations was varied by taking advantage of the participants' prior knowledge of common and uncommon food pairings. Backward blocking and unovershadowing effects were present only when highly memorable compound cues were used. Moreover, the magnitudes of both retrospective revaluation effects were affected by the strength of within-compound associations. Forward blocking and reduced overshadowing effects were independent of within-compound associations. These results have important theoretical implications for causal learning research.
Content may be subject to copyright.
Revisiting the role of within-compound associations
in cue-interaction phenomena
David Luque & Amanda Flores &
Miguel A. Vadillo
Published online: 1 July 2012
#
Psychonomic Society, Inc. 2012
Abstract Although it is th ought that within-compound
associations are necessary for the occurrence of both back-
ward blocking and unovershadowing, it is not known
whether this variable plays a similar role in mediating the
two phenomena. Similarly, the roles of within-compound
associations in forward blocking and in reduced oversha-
dowing have not been tested independently. The present
experiments evaluated how the strength of within-
compound associations affects backward blocking, unover-
shadowing, forward blocking, and reduced overshadowing.
Using an allergy task, the strength of within-compound
associations was varied by taking advantage of the partic-
ipants prior knowledge of common and uncommon food
pairings. Backward blocking and unovershadowing effects
were presen t only when highly memorable compound cues
were used. Moreover, the magnitudes of both retrospective
revaluation effects were affected by the strength of within-
compound associations. Forward blocki ng and reduce d
overshadowing effects were independent of within-
compound associations. These results have important theo-
retical implications for causal learning research.
Keywords Backward blocking
.
Blocking
.
Causal
learning
.
Reduced overshadowing
.
Unovershadowing
Learning causal relations from the environment allows us to
both predict and control events that are relevant for survival.
It is not surprising, therefore, that the specific mechanisms
that underlie causal learning have been a subject of active
research during the last few decades. A family of empirical
phenomena, usually described as cue-interaction effects, has
played a crucial role in the experimental study of causal
learning (De Houwer, 2009; Gopnik et al., 2004; Holyoak &
Cheng, 2011; López & Shanks, 2008; Mitchell, De Houwer,
& Lovibond, 2009; Shanks, 2007, 2010). According to the
main results of the relevant literature, the successful learning
of cueoutcome relationships depends on both the contin-
gency (or statistical relationship) between a particular cue
and an associated outcome and the existence of alternative
predictors of the outcome in question.
Imagine, for instance, that after an experience with eating
yogurt, the taste of yogurt becomes a good predictor of
developing a stomachache. If this highly predictive cue is
then presented along with a new food (e.g., honey) on some
other occasion, and if the combination of foods also predicts
the onset of a stomachache, it is unlikely that the added food
will be considered a true cause of stomachache. This is an
illustration of the so-called blocking effect, which was ini-
tially demonstrated in the area of animal conditioning
(Kamin, 1968) and which is an effect that has played a
major role in the development of human contingency learn-
ing theories (Dickinson, Shanks, & Evenden, 1984). In a
typical blocking experiment, participants are first exposed to
a series of pairings of a cue, A, and an outcome, O (i.e.,
AO). In a subseq uent experimental phase, the participants
D. Luque
:
A. Flores
University of Málaga,
Málaga, Spain
M. A. Vadillo
University of Deusto,
Bilbao, Spain
M. A. Vadillo
University College London,
London, UK
D. Luque (*)
Departamento de Psicología Básica, Facultad de Psicología,
Universidad de Málaga,
Campus de Teatinos, s/n,
29072 Málaga, Spain
e-mail: david.luque@gmail.com
Learn Behav (2013) 41:6176
DOI 10.3758/s13420-012-0085-3
are exposed to trials on which a novel cue, B, is presented in
conjunction with cue A, and the compound containing cues
A and B is then a predictor of the outcome (i.e., ABO).
The usual result is that participants fail either to learn the
BO associati on or to express knowledge of this associa-
tion if it has been learned. In other words, learn ing the
AO association impairs, or blocks, the learning of the
BO association.
Standard associative learning theories can easily account
for this blocking effect. For instance, in the previous exam-
ple, the well-known learning algorithm that was proposed
by Rescorla and Wagner (1972) predicts that the association
between honey and the outcome (stomachache) will not be
strengthened during the compound trials, because yogurt
alone already perfectly predicts the presence of the outcome.
Thus, honey would be a redundant cue; the presence or
absence of honey offers no additional information about
the subsequent occurrence of the illness (Rescorla &
Wagner, 1972; see also, e.g., Sutton & Barto, 1981).
Some of the classical theories of learning, including the
RescorlaWagner model cited above, fail to account for
another group of cue-interaction phenomena, which are
known as retrospective revaluation effect s. These effects
are characterized by the modification of a previously learned
cueoutcome link that results from the subsequent develop-
ment of an association between another cue and the same
outcome (or between another cue and the absence of the
outcome). For instance, in a backward blocking design, a
compound that contains two separate c ues, A and B, is
paired with an outcome. Later, one of the elements of the
compound, A, is paired repeatedly with the same outcome
(i.e., a series of ABO trials is followed by a series of
AO trials; see Table 1). If the association between the
other element of the compound, B, and the outcome, O, is
subsequently tested, participants who have been exposed to
AO pairings generally have a lower subjective estimation
of the causal strength between B and O than do control
participants who were not exposed to AO pairings.
The experimental design in studies of the unovershadow-
ing effect is very similar to the design used in studies
investigating backward blocking; the only difference be-
tween the two experimental setups is that the cue presented
in isolati on is associated with the absence of the outcome O
(i.e., a series of EFO trials is followed by a ser ies of
EnoO trials; see Table 1). In a posttraining test of the
causal strength of cue F, the typical result is a higher sub-
jective estim ation of the stre ngth of the causal r elation
between F and O among participants who were exposed to
EnoO trials than among participants in the control condi-
tion (e.g., Wasserman & Berglan, 1998).
According to classical associative models, modifications
of a cueoutcome association can occur only during trials on
which this cue is present. Thus, classical models of associa-
tive learning predict that no learning about cue B or cue F
will occur during the second phases of backward blocking
or unovershadowing experiments, respectively (Shanks,
1985). In other words, these models cannot account for the
retrospective actualizations of the cueoutcome links that
characterize retrospective revaluation phenomena.
However, it is relatively easy to adapt or extend classical
models so that they can successfully account for retrospec-
tive learning about missing cues (Dickinson & Burke, 1996;
Markman, 1989 ; Tassoni, 1995; Van Hamme & Wasserman,
1994).
Some of the revised models that have been proposed
assume that a within-compound association between the
cues in a given compo und develops because the cues are
paired with each other and with the outcome during the first
training stage of a retrospective revaluation experiment.
This within-compound associatio n will then mediate any
subsequent cue-interaction processes that may take place.
For instance, the models proposed by Dickinson and Burke
Table 1 Summary of experimental designs
Group Condition Phase 1 Phase 2 Test
Experiment 1 UsC [common pairs of foods:
e.g., macaroni and cheese]
Backward blocking ABOAOB?
Control (overshadowing) CDO C? D?
Unovershadowing EFOEnoO F?
UnC [uncommon pairs of
foods: e.g., grapes and
noodles]
Backward blocking ABOAOB?
Control (overshadowing) CDO C? D?
Unovershadowing EFOEnoO F?
Experiment 2 UsC [common pairs of foods:
e.g., macaroni and cheese]
Forward blocking AOABOB?
Control (overshadowing) CDOC?D?
Reduced overshadowing EnoO EFOF?
UnC [uncommon pairs of
foods: e.g., grapes and
noodles]
Forward blocking AOABOB?
Control (overshadowing) CDOC?D?
Reduced overshadowing EnoO EFOF?
62 Learn Behav (2013) 41:6176
(1996) and Van Hamme and Wasserman (1994) assume that
the associative strengths of both cues that are present and
cues that are absent (but that are expected) can and do
change. The mechanism by which absent cues can be
expected is the presence of a within-compound association
between the absent cue and other cues that happen to be
present. Therefore, the representation of cue B (or F) is
somehow active during the second phase of the experiment,
which allows participants to learn about the cue.
In a different vein, the comparator hypothesis (Miller &
Matzel, 1988; Stout & Miller, 2007) proposes that the acti-
vation of the representation of O in a test will depend on a
comparison of the relative associative strengths of the cues
that appear within a compound stimulus in the first phase of
an experiment. The comparator hypothesis also notes that
this comparison is triggered by a within-compound associ-
ation. Backward blocking would then take place because the
second learning phase results in the development of an
association between cue A and outcome O that is stronger
in the experimental condition than in the control condition.
Therefore, the association between B and the outcome is
poorly expressed, relative to the control condition.
Similarly, unovershadowing occurs because the EO asso-
ciation is weakened during the second phase of the experi-
ment; thus, when the EOandFO associations are
compared in a posttest, the FO association is expressed
without opposition. Thu s, the models of causal learning that
are most relevant for our study, including the comparator
hypothesis, make a fundamental assumption that retrospec-
tive revaluation depends on the formation of a within-
compound association between two cues that are presented
in the same compound during the first learning phase (see
Ghirlanda, 2005, and Kruschke & Blair, 2000, for alternate
accounts of causal learning that are independent of within-
compound associations).
The aim of Experiment 1 was to examine whether within-
compound associations are independently related to back-
ward blocking and unovershadowing effects. According to
the aforementioned models, both of these effects should be
affected by the degree to which cue A is associated with cue
B during the first learning phase. Although many studies
have shown that within-compound associations are involved
in retrospective revaluation effects in general (Aitken,
Larkin, & Dickinson, 2001; Dickinson & Burke, 19 96;
Larkin, Aitken, & Dickinson, 1998; Melchers, Lachnit, &
Shanks, 2004, 2006; Mitchell, Killedar, & Lovibond, 2005;
Vandorpe, De Houwer, & Bec kers, 2007 ), it is difficult to
determine the independent effects of these associations on
backward blocking and unovershadowing in many of these
studies, because the unovershadowing and backward block-
ing conditions have often been used as controls for each
other. A condition that could be used as a control for both
effects, such as an overshadowing condition (as proposed by
Wasserman & Berglan, 1998, and Shanks, 1985; the same
control was also proposed by Dickinson & Burke, 1996,
footnote 3), is excluded from most of these experiments, so
it is impossible to separately draw firm conclusions about
backward blocking and unovershadowing.
For instance, Dickinson and Burke (1996) manipulated
the strengths of the within-compound associations in an
experiment that used both backward blocking and unover-
shadowing. In their consistent condition, cue A appeared in
conjunction with cue B durin g the first learning phase,
which allowed for the normal formation of an association
between the two cues. However, in their varied condition,
cue B was associated with cue A on some trials and with
other distracting cues on others. It was then expected that
the association between cues A and B would be weaker in
the varied condition than in the consistent condition, but the
strengths of the within-compound associations in the two
conditions were not measured. The experimental design of
this experiment included both backward blocking and uno-
vershadowing conditions, but it did not include any other
control cond ition(s). When the participants were subse-
quently asked to assess the causal strength of cue B in a posttest,
their ratings were lower in the backward blocking condition than
in the unovershadowing condition, but this observation was true
only for participants who had been exposed to the condition in
which the cues were consistently paired.
Although Dickinson and Burkes(1996) result showed
that it is possible to learn about cues that are absent during
the second learning phase and that within-compound asso-
ciations are likely to be involved in the process, it is not
possible to determine whethe r t he obse rved effect influ-
enced backwa rd blocking, uno vershadowing, or b oth
effects. Unfortunately, other studies that have been con-
ducted to assess the role of within-compound associations
in retrospective revaluation have similar methodological
problems (Aitken et al., 2001; Dickinson & Burke, 1996;
Melchers et al., 2004, 2006; Mitchell, Killedar, & Lovibond,
2005; Vandorpe et al., 2007).
Interestingly, it appears that backw ard blocking and uno-
vershadowing could result from different processes to some
extent. This hypothesis is supported by previous research
that shows that unovershadowing effects are easier to obtain
and/or stronger than backward blocking effects. For in-
stance, a significant unovershadowing effect was observed
in an experiment by Wasserman and Berglan (1998), where-
as the backward blocking effect in their experiment did not
become significant until they removed participants with
lower levels of within-compound associations from the
analysis (see also Larkin et al., 1998). Unfortunately, it is
difficult to appropriately assess the hypothesis that unover-
shadowing and backward blocking are mediated by different
processes, because most retrospective revaluation experi-
ments lack an overshadowing control condition.
Learn Behav (2013) 41:6176 63
We find a similar confound in the published studies of
forward cue interaction. Forward cue-interaction effects are
parallel to retrospective revaluation effects, but they occur
when the learning phases are reversed (Table 1). Many
previous studies of forward blocking and reduced oversha-
dowing also suffer from the lack of an independent over-
shadowing control condition (e.g., Melchers et al., 2004,
2006). In Experiment 2, we tested the extent to which
within-compound associations affect forward blocking and
reduce ove rshadowing. Although the comparator theory
predicts that both forward blocking and reduced oversha-
dowing effects will necessarily be similarly affected by the
strength of within-compound associations, other models,
such as one that was proposed by Van Hamme and
Wasserman (1994), predict that these forward cue-
interaction effects will remain relatively unaffected by
within-compound associations.
Experiment 1
The main goal of Experiment 1 was to determine whether
the backward blocking and unovershadowing effects in
causal learning are similarly affected by changes in the
strengths of within-compound associations. To achieve this
goal, we manipulated
1
the strengths of the within-compound
associations in stimuli that were used in a standard allergy
task (Wasserman, 1990). The design of our experime nt
included a within-subjects retrospective revaluation manip-
ulation that was similar to the one used in a study conducted
by Wasser man and Berglan (1998). Thus, our design includ-
ed a backward blocking condition, an unovershadowing
condition, and importantly, an overshadowing control con-
dition that could be used to assess the magnitudes of the
backward blocking and unovershadowing effects indepen-
dently. To change the strengths of the within-compound
associations, we used a two-level, between-subjects manip-
ulation. In the unusual-compounds (UnC) group, the stimuli
that were used as cues were the same as those us ed by
Wasserman and Berglan. In general, these foods are rarely
eaten together. Alternatively, in the usual-compounds (UsC)
group, the stimuli utilized as cues were pairs of food that are
frequently eaten t ogethe r (e.g., strawberries and cream)
(Mitchell, Lovibond, & Gan, 2005).
In addition, two complementary tests were used to mea-
sure the strengths of the within-compound associations at
the end of the experiment. The first of these tests was the
recognition test used by Wasserman and Berglan (1998; see
also Aitken et al., 2001; Melchers et al., 200 4, 2006;
Wasserman & Castro, 2005). In this test, the participants
were asked to identify pairs of cues that they h ad seen
during the first learning phase among other distractor cue
compounds. Because there were only four compounds that
could be correctly recalled, the recognition test scores
ranged from 0 to 4.
After the recognition test, the participants were asked to
participate in an additional test that was aimed at increasing
the variability of the measure of the strength of the within-
compound associations (see Larkin et al., 1998, for a related
memory test). In the second test, participants were asked to rate
the extent to which they were confident in their previous
selections in the recognition test. Thus, each participant pro-
vided two related measures of his or her ability to remember
the compounds that had been used in the experimentnamely,
the number of correctly identified compounds and the level of
confidence that each participant had in his or her selec-
tions. An additional goal of this experiment was to
determine whether this double test of the strength of
within-compound associations offers any advantage over
the classic recognition test.
Method
Participants and apparatus
In total, 92 students at the University of Málaga School of
Psychology received course credit for voluntary participa -
tion in this study. The students were randomly assigned to
the two experimental condition groups; 52 students were
included in the UnC condition group, and 40 students were
included in the UsC condition group.
Each participant performed the task on 1 of 10 PCs in a
semi-isolated cubicle. Each PC had been equipped with
custom-built software that was programmed in Visual
Basic 2005 (Microsoft, USA). Participants indicated their
responses using the keyboard and the mouse of the PC.
Stimuli and procedure
Stimuli Eight different food names were taken from
Wasserman and Berglans(1998) study and were used as
cues in the UnC condition. These food names were walnut s,
mushrooms, grapes, yoghurt, noodles, orang es, chicken,
and carrots. Cue pairings in the UnC condition were coun-
terbalanced. A different set of eight food names (making up
four pairs) that had been adapted from Mitchell, Lovibond,
& Gan, (2005) was used in the UsC condition and included
1
Although we have used the term manipulate in this context, we did
not manipulate this variable in a narrow sense, because the larger
within-compound associations between the food pairs that were used
in the UsC condition were not developed via laboratory training.
Despite this, however, we think that the results of the within-
compound association tests help to mitigate concerns about possible
confounding extraexperimental differences in the participants judg-
ments of the stimuli.
64 Learn Behav (2013) 41:6176
the following pairs: macaroni and cheese, strawberries and
cream, tea and coffee, and bread and butter. The assignment
of the various food names to individual cues in the UsC
condition was only partially counterbalanced so that the
food pairings that commonly co-occur in extraexperimental
contexts also appeared together as compound cues during
the experiment. The cues were written in black type and
appeared at the center of the screen. For trials on which two
cues were presented, the position of each cue on the screen
(right or left) was randomized across trials. The phrases
Allergy or No Allergy were used as outcomes (O and
noO, respectively; see Table 1) and appeared as text that was
positioned below th e cues in the center of the screen.
Participants were given feedback about their choices on each
trial via messages stating either Incorrect! (written in red) or
Correct! (written in green).
Procedure With the except ion of the additional memory test
that was used after the traditional recognition test, all of the
details of our procedure, instruc tions, and ques tionnaires
were kept similar to those used in Wasserman and
Berglans(1998) study. The participants began the experi-
ment by reading the instructions on the computer screen. In
the instructions, they were asked to imagine that they were
allergists who were attempting to determine which foods
were causing a fictitious patient, Mr. X, to experience an
allergic reaction. To do so, the participants were required to
study the results of a series of daily allergy tests in which
Mr. X did or did not suffer from an allergic reaction after
eating certain foods.
After they had finished reading all of the instructions and
their questions had been answered, the participants were
asked to rate the likelihood that each of the foods that would
eventually be used as a cue could cause an allergic reaction
in an ordinary person. Participants were instructed to answer
on a scale ranging from 0 to 9, in which a score of 0 indicated
that it [the food] does not cause any allergic reaction at all
and a score of 9 indicated that it [the food] causes a strong
allergic reaction.
After obtaining these preliminary data, the participants
began performing the actual causal learning task. Each
learning trial represented an allergy test performed on Mr.
X. These trials began with the presentation of the foods that
had been eaten by Mr. X (i.e., the cues) in that particular
allergy test at the center of the top of the screen. After 3 s,
the participants were asked to decide whether Mr. X would
suffer an allergic reaction and to indicate their decisions by
pressing one of two keys: the A key was used to
indicate that they expected Mr. X to have an allergic
reaction, and the N key was used to indicate that they
did not expect him to have an allergic reaction. After a
participant had pressed the key that indicated his or her
choice, the outcome and the feedback appeared on the
screen, thereby allowing the participants to learn the
correct cueoutcome associations.
During the first learning phase of the experim ent, the
participants had to learn to identify four food pairings that
could potentially cause an allergic reaction in Mr. X. In this
phase, compounds AB, CD, and EF were always paired with
an allergic reaction, and the filler compound GH was always
paired with the absence of a reaction. Each compound was
presented 30 times in a pseudoran domized presentation
order, with the constraint that each compound was presented
6 times during each 24-trial series. After the first learning
phase, participants were instructed to respond to the same
questionnaire that had been presented at the beginning of the
experiment; however, they were instructed to answer the
questions on the questionnaire using the information that
they had learned about Mr. X during the first learning phase.
The participants were not given access to the ratings that
they had provided in previous questionnaires.
After completing the second questionnaire, each partici-
pant began the trials that made up the second learning phase.
In this phase, cue A was always paired with the allergic
reaction, and cues E and G were always paired with the
absence of an allergic reaction. Each cue was presented 30
times according to a pseudorandom order of presentation; the
constraint for ordering during this phase was that each cue had
to be presented 6 times in every block of 18 trials. After the
second learning phase, participants were again instructed
to respond to the questionnaire regarding the causal
efficacy of each food in producing an allergic reaction.
In this case, the participants were asked to consider all
of the information that they had learned throughout both
phases of the experiment. The ratings that the partici-
pants provided while responding to this questionnaire
(specifically, the ratings for cues B, C, D, and F) were
used as depe n de nt va r ia ble s i n the s ubs e que n t me as u re-
ment of cue-interaction phenomena.
After completing the third questionnaire, each participant
performed two mem ory tests that were aimed at measuring
his or her ability to remember the compounds that had been
established during the first training phase. The first of these
tests was the recognition test used by Wasserman and
Berglan (1998). In this test, the participants had to select
the four compounds that had actually been presented during
the task among 16 pairs of food. The 12 incorrect food
pairs were new compounds that were formed by pairing
the eight foods that had been tested in a different way.
The second memory test requested that the participants
assess their levels of confidence in each of the selec-
tions that they had made in the recognition test.
Participants were asked to rate their confidence levels
on a scale of 0 to 9, in which a score of 0 indicated
being completely unconfident and a score of 9 indicated
being absolutely confident.
Learn Behav (2013) 41:6176 65
Data analysis
In contrast to previous studies of cue interaction, we used a
nonparametric approach to the analysis of our data. This
decision deserves some comment. The nonpara metric ap-
proach is highly re commended for analyzing the current
data because the dependent variables are ordinal in nature
(single Likert-type items). The main dependent variables
that were used in our experiment were ratings associated
with the probability that each food would cause the fictitious
patient, Mr. X, to have an allergic reaction. Participants were
asked to make these ratings by choosing among a set of
ordered categories (from 0 to 9). We can determine whether
a participant rates one causal relation as being stronger than
a different causal relation, but we can not determine the
magnitude of the distance between these ratings, because
we do not have a measure of the relation between our
measurement tool and the causal belief of the participants
(the underlying factor in generating ratings). For instance, it
is possible that the difference between 0 and 1 reflects a
greater difference in the underly ing factor than does th e
difference between 1 and 2. However, according to the
Likert-type item score, the distance between each pair of
valuesnamely, one pointis the same.
Although the recognition test (which measured the num-
ber of compounds that were correctly remembered) was not
a Likert-type item, this variable was also an ordinal variable,
because the correct recognition of a higher number of com-
pounds indicates the presence of stronger within-compound
associations. Again, we cannot determine the magnitudes of
the distances between the different levels of the underlying
factor (the memory for compounds) that produce variations
in the measured variable.
The critical reader might argue that because parametric
analyses facilitate the analysis of complex interactions, it is
preferable to assume that there are some minor inaccuracies
in the statistical analyses and to co nduct the parametric tests,
even when an experi ment uses ordinal dependent variables.
It is important to note that several articles that discourage the
use of parametric analyses for ordinal dependent variables
have been published in other fields. For examp le, treatment
efficacy in psychiatric research is frequently measured via
rating scales that are very similar to the scales that are used
to measure t he dependent variables in causal le arning
experiments. Munzel and Bandelow (1998) strongly criti-
cized the generalized use of parametric analyses in the
studies that used ratings to measure treatment efficacy. In
the field of plant pathology, the severity of plant diseases is
often assessed according to an ordinal rating scale, rather
than a continuous scale. Shah and Madden (2004) criticized
the use of parametric analyses in these studies, and they
proposed the use of nonparametric alternatives. In the field
of epidemiology, Kahler, Rogausch, Brunner, and Himmel
(2008) observed that even though many measures of health-
related quality of life aspects are ordinal, the data are typi-
cally analyzed using parametric tests. These authors exam-
ined whether applying parametric tests (instead of the
appropriate nonparametric tests) to these quality of life data
could affect the results and found that using parametric tests
to analyze ordinal data could lead to erroneous results (see
also Norris, Ghali, Saunders, Brant, & Galbraith, 2004;
Singer, Poleto, & Rosa, 2004).
The ordinal nature of our dependent variables was not the
only reason that parametric analyses of our data were not
feasible; most of the dependent variables were also non-
normally distributed. For example, the judgments about
both cue B and cue F in the final questionnaire were not
normally distributed. We attempted to improve the nor-
mality of our data by performing some common data
transformations, including log(X + 1), sqrt(X), and 1/(X + 1).
However, KolmogorovSmirnov tests showed that the distri-
butions that resulted from all of these transformations were
still significantly different from the normal distribution.
In summary, the reasons to avoid applying parametric
analyses to our data have been described many times (for
a classical reference, see Siegel & Castellan, 1988, Chapter
3). The statistical model s that underlie parametric analyses
assume that the dependent variables are continuous and
normally distributed. Applying this type of model to a
dependent variable that is ordinal and nonnormal may pro-
duce deviations in the obtained p value. Even if the devia-
tion in the p value is not dramatic, the impact of it is difficult
to quantify. Thus, a nonparametric analysis is strongly rec-
ommended, because it is more conservative than the para-
metric alternative and because a nonparametric analysis
maximizes statistical power.
Even so, because we are aware that very different statis-
tical tests have been used to analyze these types of variables
in the past, and to avoid any suspicion regarding an
attempt to avoid ambiguous results from other types of
statistical tests, we have also repeated all of our analy-
ses using equivalent parametric tests. These complemen-
tary parametric analyses were performed in the same
manner as in previous experiments; that is, p arametric
analyses of variance (ANOVAs) were used in conjunc-
tion with additional t-tests when necessary (e.g., Melchers et
al.,
2004). Any differences between the results of the para-
metric and nonparametric analyses that we observed are
reported below.
An important limitation of the use of ordinal variables is
that they can neither be summed nor subtracted. This limi-
tation implies, for instance, that we cannot average the mean
responses to cues C and D to compute an oversh adowing
control variable, although this is commonly done in the
general research area of the present study (e.g., Melchers
et al., 2006). Therefore, we conducted all of the analyses
66 Learn Behav (2013) 41:6176
that involved the overshadowing control condition twice;
the first analysis used cue C to represent the control condi-
tion, and the second used cue D. For the sake of simplicity,
we have reported only the results of analyses in which cue C
represented the control condition for cases in which both
analyses yielded the same results.
In addition, because nonparametric analyses cannot be
used to examine complex interactions, we needed to com-
pute new variables that represented the strengths of the
backward blocking and/or the unovershadowing effects for
some analys es (Melchers et al., 2006). The magnitude of the
backward blocking magnitude variable (mBB) was devel-
oped to sort the participants according to the degree to
which they experienced backward blocking (the resultant
variable was also ordinal). Thus, participants with lower
ratings for cue B than for cues C and D were categorized
as being participants who experienced higher l evels of
backward blocking (labeled as +2). For instance, a partici-
pant who provided a rating of 3 for cue B and ratings of 4
and 9 for cues C and D, respectively, would have a mBB
value of +2. The next category, labeled as +1, included
participants who had lower ratings for B than for C or D.
Participants were labeled as + 1 i f the ratings that they
provided for one of the control cues (cue C or cue D) were
higher than the ratings that they provided for cue B and if
the ratings for the other control cue were equal to their
ratings for cue B. For example, a participant who provided
a rating of 3 for cues B and C and a rating of 9 for cue D
would have a mBB value of +1. The next mBB category
was 0. This category included all of the participants who had
provided equal ratings for cues B, C, and D. It also included
participants who gave a rating to one of the control cue s that
was higher than the rating that they had given to cue B but
who gav e a lower rating to the other control cue. For
instance, a participant who had given ratings of 3, 2, and 9
to cues B, C, and D, respectively, would have been assigned
an mBB value of 0. The mBB category below 0 was
1, which included participants who had given one
control cue a rating that was lower than the rating that
they had given to cue B and who had given the other
control cue a rating that was equal to the rating that
they had attributed to cue B. For instance, a participant
whohadratedcuesB,C,andDwithscoresof3,3,
and 2, respectively, would have an mBB value of 1.
The lower extreme of this ordinal variable was the 2
category. The 2 mBB category included participants
who had given lower ratings to both of the control cues
(C and D) than they had given to cue B.
The ordinal variable that was used to measure the mag-
nitude of unovershadowing (mUN) was computed accord-
ing to the same rules, but it was used as an index of the
unovershadowing effect instead of the backward blocking
effects (i.e., instead of ratings for cue B, ratings for cue F
relative to ratings for the control cues were used in the
computation of the mUN value for each participant).
To differentiate between participants who had either
strong or weak within-compound associations, we used both
the number of correctly remembered compounds from the
recognition test (an ordinal varia ble that was ordered from 0
to 4) and the subsequent ratings that the participants p rovid-
ed during the confidence tests. The first recognition test
allowed us to divide the sample into two groups. One group
comprised the participants who had correctly identified all
four compounds (the good recognition, [Good-Rec] group) ,
and the other group comprised the remaining participants
(the poor recognition [Poor-Rec] group; see Wasserman &
Berglan, 1998). A similar criterion was used to group the
participants on the basis of their performance on the second
(confidence) test. Participants who had correctly identified
all four compounds and who had confidence ratings of 9 for
their selections were included in the good recognition
and confidence group (Good-Rec&Con), whereas the
remaining participants were included in the poor recog-
nition a nd confidence group (Poor-Rec&Con). The
effects of both memory tests on backward bloc king
and uno vershadowing w ere eva luated inde pendently.
Pairwise comparisons were used to control for type I
errors at the .05 significance level.
Results and discussion
The medians of the causal ratings for each of the cues with
their respective outcomes in each of the three rating periods
are provided in Table 2. These results indicate that the
participants generally had low causal ratings at the begin-
ning of the experiment. After the first phase of learning, the
participants adjusted thei r ratings to be consistent with the
programmed contingenc ies. Both the backward blocking
and unovershadowing effects were evaluated using the rat-
ings that were provided in the third questionnaire as depen-
dent variables; these data were supplied after the completion
of the second learning phase.
A Friedman test was conducted with the whole sample of
participants, regardless of group, to evaluate the difference
in the median ratings for cues B (median 0 4), C (median 0
5), and F (median 0 6), and the observed difference was
significant, χ
2
(2, N 0 92) 0 15.07, p < .001. Follow-up
comparisons were conducted using a Wilcoxon test, which
revealed that the median of the ratings for cue F was signif-
icantly higher than the medians of the ratings for both cue C
(p < .01) and cue B (p < .001). Moreover, the median of the
ratings for cue C was significantly higher than that of the
ratings for cue B (p < .01). Thus, we can confirm that we
obtained reliable backward blocking and unovershadowing
effects. In addition, we also compared the strengths of the
backward blocking and unovershadowing effects by making
Learn Behav (2013) 41:6176 67
a direct comparison of the mBB (median 0 0) and mUN
(median 0 0) variables. The results of a Wilcoxon test were
not significant, z 0 0.58, p > .5.
AMannWhitne y U-test was co nducted to determine
whether our within-compound manipulation (UnC vs.
UsC) affected the participants memory for compounds in
the expe cted manner. We used the number of correctly
remembered compounds as the dependent variable in this
test. The result of the test was statistically significant and
was in the expected direction, z 0 2.32, p <.05;this
result indicates that, on average, the members of the
UsC group had better memories for compounds (average
rank 0 52.51) than the members of the UnC group
(average rank 0 41.88).
We then evaluated the backward blocking and unover-
shadowing effects in the UsC and UnC conditions. The
medians and semi-in terquartile ranges of these effects in
the two groups suggest that both backward blocking and
unovershadowing were obtained only in the UsC condition
(see Fig. 1). This impression was confirmed using two
Friedman tests; one test was used to assess each condition.
The effect of cue (B vs. C vs. F) was not significant in the
UnC condition, χ
2
(2, N 0 52) 0 3.06, p > .1, but it was
highly significant in the UsC condition , χ
2
(2, N 0 40) 0
15.24, p < .001. Wilcox on tests were used to conduct
follow-up comparisons of the data from Gro up UsC, and
the results of these tests showed that the median rating for
cue F was higher than the median rating for cue C (p 0
.061), although the test result did not meet our significance
criterion. However, the results of a parallel comparison
between the median rating for cue F and the median rating
for the alternative control cue, D, were statistically significant
(p < .05). In addition, the median rating for cue F was signif-
icantly higher than the median rating for cue B (p <.001),and
the median rating for cue C was also higher than that for cue B
(p <.005).
Although the results of the Friedman test were not
significant in the UnC condition, we also conducted
pairwise comparisons of the data obtained in this con-
dition, for theoretical reasons. None of the comparisons
that were made using these tests were significant; all p
values were > .09. Thus, it appears that the strengths of
the within-compound associations were a key factor in
obtaining both backward blocking and unovershadowing
effects.
To look for convergent evidence regarding the role of
within-compound associations in cue-interaction effects, ad-
ditional MannWhitney U-tests wer e conducted to deter-
mine whether our independent variable (UnC vs. UsC) had
also affected the values of the mBB and mUN variables.
None of these tests yielded significant results, although
there was a trend in the mBB variable, with higher values
of mBB in the UsC t han in the UnC conditions [mBB,
z 0 1.63, p 0 .103; the same analysis (UnC vs. UsC) with
mUN variable, z 0 1.03, p > .3]. A parametric approach to
examining these effects offers a similar pattern of
results, with the exception of the mBB effect, which
reaches statistical significance, indicating that the mBB
Table 2 Median ratings of all cues in Experiments 1 and 2 according to experimental conditions UnC and UsC
Experiment 1 (Medians) Experiment 2 (Medians)
Condition Cue Rating
period 1
Rating
period 2
Rating
period 3
Rating
period 1
Rating
period 2
Rating
period 3
UnCA179199
B265101
C175105
D065005
E 1 6.5 0 0 0 0
F165009
G100100
H100100
UsC A 1 6.5 9 1 9 9
B281.5405
C 1 8 5.5 0.5 0 5
D1652.505
E060100
F188.5208
G000100
H1.500200
68 Learn Behav (2013) 41:6176
measure differed between the UnC and UsC groups
2
[mBB, two-sample separate variance, t(68.56) 0 2.12,
p <.05;mUN,t(90) 0 1.01, p >.3].
Thus, it appears that the comparison between UnC and
UsC was not sufficiently sensitive to show a clear effect on
the values of mBB and mUN. There is, however, a more
direct way to evaluate the role of within-compound associ-
ations in unovershadowing and backward blocking. Instead
of dividing participants on the basis of an experimental
manipulation, it is possible to sort them as a function of
their memory test performance scores (i.e., their abilities to
remember the compounds that were shown during training).
Indeed, we expected that this grouping strategy might be
more sensitive than a strategy of grouping on the basis of the
types of stimuli that were used in the experimental task,
because it is reasonable to expect that some participants in
the UnC group were capable of learning t he within-
compound associations perfectly and that some participants
in the UsC group were not capable of correctly learning the
within-compound associations.
Therefore, we sorted participants into groups on the basis
of their abilities to remember the within-compound associ-
ations, and we subsequently investigated whether there were
any differences in the mBB and mUN values of these two
groups (Table 3). We initially grouped participants accord-
ing to the results of the standard memory test (Good-Rec vs.
Poor-Rec) alone, but the results of the MannWhitney U-
tests that we conducted for both variables were not signifi-
cant (mBB, z 0 1.47, p > .1; mUN, z 0 0.35, p > .5).
Importantly, we also assessed this hypothesis after using the
second memory test to assign participants to the Good-
Rec&Con and Poor-Rec&Con groups. After grouping the
participants in this manner,theresultsofbothMann
Whitney U-tests in the mBB and mUN variables met
our significance criterion (mBB, z 0 3.03, p < .005;
mUN, z 0 2.06, p < .05), which confirms that partic-
ipants with stronger within-compound associations also
demonstrated more substantial backward blocking and
unovershadowing effect s.
In summary, the statistical analysis of our data revealed
reliable effects of both backward blocking and unoversha-
dowing. These results are interesting in and of themselves
because previous studies have failed to obtain both effects in
a single experiment. For example, Larkin et al. (1998)
obtained unovershadowing but not backward blocking,
which led some authors to the conclusion that unoversha-
dowing is a stronger effect than backward blocking and that
it is easier to obtain (e.g., De Houwer, Beckers, & Vandorpe,
2005). Indeed, s imulations of a model proposed by
Dickinson and Burke (1996) have shown that their model
predicts that the unovershadowing effect in a given situation
will be stronger than the backward blocking effect (Aitken
& Dickinson, 2005). In contrast to this prediction, both
backward blocking and unovershadowing effects were
obtained in the present experiment.
The most significant result was that both the backward
blocking and unovershadowing effects were mediated by
the strengths of the within-compound associations.
Therefore, we can say that the results from Experiment
1
were consistent with the predictions of current associative
learning theories, such as the associative learning models
that have been proposed by Van Hamme and Wasserman
(1994) and by Miller and Matzel (1988), in which the role of
2
Parametric analyses were performed by applying the standard proce-
dure that has typically been used in other articles about cue interaction
in which causal judgments were treated as the dependent variable. For
instance, descriptive statistics for the mBB and mUN variables were
computed by calculating a single measure of overshadowing from the
means of the ratings for cues C and D. The magnitudes of the unover-
shadowing and bac kward blocking effects were then calculated by
subtracting the new mean CD from the ratings for F (mUN) and then
subtracting ratings for B from this mean (mBB).
Fig. 1 Medians of foodallergy causality ratings at the end of phase 2
of each exper iment (a Experiment 1; b Experiment 2). Error bars
indicate semi-interquartile ranges. We represent overshadowing scores
as the median of the values of the C
Learn Behav (2013) 41:6176 69
within-compound associations is crucial in the mediation of
both backward blocking and unovershadowing effects.
Experiment 2
For theoretical reasons, de bate regarding the roles that
within-compound associations play in cue-interaction
effects has focused on retrospective revaluation phenomena.
The reason for this focus is that most acco unts of retrospec-
tive revaluation phenomena propose that within-compound
associations affect cue-interaction effects only in their back-
ward versions (Dickinson & Burke, 1996; Van Hamme &
Wasserman, 1994).
Nevertheless, some experiments have also included for-
ward cu e-interaction conditions that are similar to retrospec-
tive revaluation conditions but that reverse the order of the
learning phases (see Table 1). Experiments that have
assessed the role of within-compound associations in for-
ward cue-interaction phenomena have been aimed at inves-
tigating whether, as the comparator hypothesis predicts,
within-compound associations play the same roles in for-
ward cue interaction and in retrospective revaluation (Stout
& Miller, 2007). The results of several causal learning
experiments that have been conducted using human partic-
ipants appear to refute this prediction (Aitken et al., 2001;
Dickinson & Burke, 1996; Melchers et al., 2004, 2006), but
a study by Amundson, Witnauer, Pineño, and Miller (2008)
showed that within-compound associations could mediate
overshadowing in three different conditioning experiments
that utilized rats. Although there are some cases in which an
overshadowing condition can be used as a control condition
for other types of cue-interaction effects, such as Experiment 1
of this study, overshadowing itself can be considered a cue-
interaction phenomenon, because cues that are presented to-
gether com pet e wit h e ach o the r. O n th e bas is o f th eir
ownresults,Amundsonetal. concluded that within-
compound associations not only are necessary for retro-
spective revaluation, but also are necessary for forward
cue-interaction effects.
An important limitation of previous work regarding the
role of within-compound associations in cue-interaction
phenomena is that none of the human causal learning
experiments that have assessed the role of within-
compound associations in forward cue-interaction phenom-
ena have included an appropriate control for overshadow-
ing. This makes it imp ossible to draw any conclusions about
the specific role(s) of within-compound associations in for-
ward blockin g and/or reduced overshadowing. Althoug h
Vandorpe and De Houwer (2005) studied both effects in
human causal learning and used the correct control condi-
tion, these authors did not assess the roles that within-
compound associations played in mediating the two cue-
interaction effects. They did find that the reduced oversha-
dowing effect was generally larger than the forward block-
ing effect in their study, which they have interpreted as
being inconsistent with associative learning models and
consistent only with an inferential account of cue interaction
in causal learning.
The main goa l of the follow ing experiment was to deter-
mine whether within-compound associations have an impact
on forward blocking and reduced overshadowing. Moreover,
it seems relevant to test the generality of the result that was
obtained by Vandorpe and De Houwer (2005)bytryingto
replicate it, particularly because they concluded that their
result was inconsistent with the predictions of associative
learning models. Thus, replicating their result was a secondary
goal of the present experiment.
Method
Participants, apparatus, stimuli, and procedure
In total, 59 students at the University of Malaga School of
Psychology received course credit for their voluntary par-
ticipation in our study. The students were randomly assigned
to the two main experimental condition groups; specifically,
31 students were included in the UnC condition group, and
28 students were included in the UsC condition group. All
of the experimental app aratuses, stimuli, and procedures
Table 3 Percentages of each category for the mBB and mUN variables in Experiment 1
Groups mBB (%) mUN (%)
2 10122 10 1 2
UnC 9.6 15.4 51.9 7.7 15.4 5.8 13.5 51.9 9.6 19.2
UsC 5 7.5 52.5 12.5 22.5 10 0 50 22.5 17.5
Poor-Rec 14.3 10.7 57.1 3.6 14.3 3.6 3.6 64.3 7.1 21.4
Good-Rec 4.7 12.5 50 12.5 20.3 9.4 9.4 45.3 18.8 17.2
Poor-Rec&Con 9.3 14.8 61.1 5.6 9.3 7.4 9.3 61.1 7.4 14.8
Good-Rec&Con 5.3 7.9 39.5 15.8 31.6 7.9 5.3 36.8 26.3 23.7
70 Learn Behav (2013) 41:6176
were the same as those used in Experiment 1,withthe
exception that the order in which learning phases 1 and 2
took place was reversed (see Ta ble 1).
Results and discussion
We followed the same strategy for analyzing the data that
we used in Experiment 1. We conducted a Friedman test of
ratings from the third quest ionnaire in which we contrasted
the median ratings of cues B (median 0 4), C (median 0 5),
and F (median 0 9) to determine whether we had obtained
effects of forward blocking a nd reduced overshadowing.
The results of this analysis were significant, χ
2
(2, N 0 59) 0
24.06, p < .001. Pairwise comparisons that were conducted
using Wilcoxon tests showed that the median rating for cue F
was higher than those for both cues B (p < .001) and C
( p < .001). In addition, the median rating that was
associated with cue C was higher than that for cue B
(p < .05). Thus, we obtained reliable forward blocking and
reduced overshadowing effects. As a means of evaluating the
generality of the results that were obtained by Vandorpe and
De Houwer (2005), we also compared the strengths of the two
forward cue-interaction effects by directly comparing the var-
iable that we used as a measure of the magnitude of forward
blocking, mFB (median 0 1), with the variable that we used as
a measure of the magnitude of reduced overshadowing, mRO
(median 0 1). A nonsignificant result was obtained from a
Wi lcoxon test (mFB vs. mRO: z 0 0.24, p > .8). Our
result was s till far from reaching statistical significance,
t(58) 0 0.82, p > .4, when using dependent variables
and a parametric approach to analyzing the data that were
identical to those used by Vandorpe and De Houwer (2005).
AMannWhitney U-test was co nduc ted to dete rmine
whether the UnC versus UsC experimental manipulation af-
fected the participants memory for compound stimuli as
expected. As in Experiment 1, we used the number of correctly
remembered compounds as the dependent varia ble. The result
of this test was statistically significant and was in the expected
direction, z 0 3.85, p < .001, which indicated that, on average,
participants in the UsC group were better at remembering the
identities of the trained compounds (average rank 0 37.07) than
were participants in the UnC group (average rank 0 23.61).
We then evaluated forward blocking and reduced oversha-
dowing effects in both the UsC and UnC groups. The medians
and semi-interquartile ranges of these effects in the two groups
suggest that there was a reliable cue-interaction effect in both
UsC and UnC groups (see Fig. 1). The effect of cue (B vs. C vs.
F) was significant in the UnC group, χ
2
(2, N 0 31) 0 16.90,
p < .001. Follow-up comparisons were conducted using
W ilcoxon tests and showed that the median rating for cue F
was higher than those for both cue C (p < .005) and cue B
(p < .001). In addition, the median rating for cue C was higher
than that for cue B (p < .05 ). The effect of cue (B vs. C vs. F)
was also significant for participants in the UsC group, χ
2
(2, N 0
28) 0 7.70, p < .05. Follow-up comparisons showed that the
median rating for cue F was higher than those for cues C
(p < .05) and D (although the comparison between the median
ratings for cue F and the control cue D was only marginally
significant, p 0 .054). The median rating for cue F was also
higher than that for cue B (p < .005). In contrast, the median
rating for cue C did not differ from the median rating for cue B
(p > .3). Thus, a simple forward blocking effect was not
observed among members of the UsC group.
In accordance with the logic behind the analyses that were
conducted in Experiment 1,additionalMannWhitney U-tests
were conducted to investigate whether the manipulation of the
stimuli (UnC vs. UsC) had also affected the values of the mFB
and mRO variables. However, none of these tests yielded
significant results (mFB, z 0 1.56, p 0 .119; mRO,
z 0 1.03, p > .8), indicating no difference in the magnitude
of forward blocking and reduced overshadowing between
both groups (UnC vs. UsC).
To further assess the role(s) played by within-compound
associations in forward blocking and reduced overshadow-
ing, we sorted participants into different groups according to
their memories for the within-compound associations, as in
Experiment 1. We then tested for between-group (UnC vs.
UsC) differences in the values of the mFB and mRO vari-
ables (Table 4). We first grouped the participants on the
basis of their standard recognition test scores alone (which
Table 4 Percentages of each category for the mFB and mRO variables in Experiment 2
Groups mFB (%) mRO (%)
2 10 1 2 2 10 1 2
UnC 3.2 9.7 29 29 29 3.2 9.7 35.5 29 22.6
UsC 3.6 17.9 35.7 32.1 10.7 10.7 3.6 32.1 35.7 17.9
Poor-Rec 6.3 6.3 37.5 37.5 12.5 0 6.3 43.8 43.8 6.3
Good-Rec 2.3 16.3 30.2 27.9 23.3 9.3 7 30.2 27.9 25.6
Poor-Rec&Con 3.8 7.7 34.6 34.6 19.2 3.8 3.8 34.6 42.3 15.4
Good-Rec&Con 3 18.2 30.3 27.3 21.2 9.1 9.1 33.3 24.2 24.2
Learn Behav (2013) 41:6176 71
resulted in the generation of the Good-Rec and Poor-Rec
groups). Subsequent Mann Whitney U-tests showed no
significant difference between Good-Rec and Poor-Rec
groups for either mFB or mRO (mFB, z 0 0.25, p > .8;
mRO, z 0 0.44, p > .6). We also assessed this hypothesis
using the results of the second memory test to assign par-
ticipants to the Goo d-Rec&Con and Poor-Rec&Con
groups. A gain, the results of both MannWhitney U-
tests were far from reaching statistical significance
(mFB, z 0 0.45, p >.6;mRO,z 0 0.55, p >.6).
In summary, the results of Experiment 2 demonstrated the
presence of a reliable forward cue-interaction effect. We also
obtained evidence of both forward blocking and reduced
overshadowing. Importantly, our results supported the idea
that within-compound associations are not involved in any
forward cue-interaction phenomena; thus, our results are
consistent with the Van HammeWasserman and Dickinson
Burke models of associative learning. Moreover, the absence
of evidence supporting the notion that within-compound asso-
ciations mediate forward blocking and reduced overshadow-
ing effects cannot be explained by the comparator hypothesis
(Stout & Miller, 2007).
We also assessed the generality of the asymmetrical cue-
interaction effects that were obtained by Van dorpe and De
Houwer (2005). We were not able to replicate their results in
the present experiment, even when we used identical depen-
dent variables and exactly the same method of analysis.
Surprisingly, we failed to find a significant forward
blocking effect among members of the UsC group. It is
important to note, however, that the strength of the forward
blocking effect did not differ betwe en the UnC and UsC
groups and that when the entire sample was considered, the
forward blocking effect was significant using both control
conditions, C and D. Thus, the nonsignificant forward
blocking effect in the UsC group can be attributed to a loss
of statistical power. In any case, the main findings of
Experiment 2 were that (1) we found significant forward
blocking and reduced overshadowing effects and (2) these
effects did not appear to be impacted by various aspects of
within-compound associations.
In addition, we conducted some cross-experimental anal-
yses. The results of these analyses should be interpreted
with caution because between-experiment comparisons are
never as convincing as within-experiment comparisons.
However, given the similarity between the procedures that
were used in the two experiments, we explored some theo-
retically interesting comparisons. Specifically, comparing
the results of Experiments 1 and 2 could allow us to exam-
ine the effect that the order of the training phases used in
each paradigm (backward vs. forward, respectively) has on
the magni tudes of the cue-interaction effects in groups UnC
and UsC. On the one hand, we did not expect to find any
differences in cue-interaction effects in the UsC g roup,
regardless of the order of the phases. On the other hand,
however, we expected to find that the forward condition
would be associated with larger cue-interaction effects
in the UnC group when compared with the cue-
interaction effects that w ere associated with the back-
ward condition.
The dependent variables that we used in these analyses
were the magnitudes of the cue-interaction effects. We had
two dependent variablesnamely, the magnitude of back-
ward blocking or forward blocking (BB/FB) and the mag-
nitude of unovershadowing or reduced overshadowing (UN/
RO). We conducted two separate Mann Whitney U-tests for
each within-com pound condition (UnC and UsC) and for
each of the two dependent variables (BB/FB and UN/RO).
The independent variable in each of the analyses was the
order of the phases (forward vs. backward). The analysis of
the effect on participants in the UnC group showed that the
order of the phases significantly affected the BB/FB cue-
interaction effect, z 0 2.69, p < .01 (forward average rank 0
50.77; backward average rank 0 36.77), and that there was a
similar trend for the UN/RO effect, z 0 1.589, p 0 .11
(forward average rank 0 47.10; backward average rank 0
38.96). In contrast to this finding, no significant effects were
observed in the analyses of the data from participants in the
UsC group (p values > .5). It appears that the cue-interaction
effects in the UsC group were generally large regardless of
the order in which the phases were presented, so there were
no significant effects of the order of the learning phases on
the cue-interaction effects that were measured among par-
ticipants in this condition.
General discussion
In light of the results of the present study and the results of
previous studies, it appears to be quite clear that within-
compound associations play an important role in mediating
retrospective revaluation effects (e.g., Dickinson & Burke,
1996). To date, however, and despite evidence in support of
the role of within-compound associations in retrospective
revaluation effects, it has not been determined whether these
associations have identical impacts on backward blocking
and unovershadowing. In the present study, we have
shown that within-compound associations play roles in
the modulation of both backward blocking and unover-
shadowi ng (Ex per im ent 1). However, we failed to es-
tablish the presence of a within-compound association
effect on forward cue-interaction phenomena, such as
forward blocking and reduced overshadowing (Experiment
2). The general pattern of results that were obtained in the
present study is consistent with the predictions made by
modern learning theories (e.g., Van Hamme & Wasserman,
1994).
72 Learn Behav (2013) 41:6176
Despite the numerous experiments that have already
assessed the role of within-compound associations in cue
interaction, it is remarkable that nearly all of the previous
studies in which the strengths of the within-compound asso-
ciations were manipulated failed to include control condi-
tions that were appropriate for the independent assessment
of backward blocking and unovershadowing (e.g.,
Dickinson & Burke, 1996). A notable exception is a study
conducted by Larkin et al. ( 1998). In the first experiment of
their study, they manipulated within-compound associations
using an experimental design that included both backward
blocking and unovershadowing conditions; their design was
very similar to the design that we used for Experiment 1.
Larkin et al. (1998) manipulated the strength of the
within-compound associations in their study by varying
the number of compound trials (three vs. six). Although this
manipulation affected the participants abilities to remember
the identities of vario us compound stimuli, it had no effect
on either backward blocking or unovershadowing. Indeed,
in both Experiments 1 and 2 of their study, they only
obtained unovershadowing, not back ward blocking. In
Experiment 3 of their study, they manipulated the within-
compound associations in accordance with a strategy that
was proposed by Dickinson and Bur ke (1996 ); namely, they
included a varied condition in which they paired the target
cues with several filler cues in an attempt to weaken the
within-compound associations. Using that paradigm, Larkin
et al. (1998) found that the unovershadowing effect was
weaker among participants in the varied condition group
than among participants in the consistent condition group.
Larkin et al. (1998) interpreted these results as being
contrary to the model proposed by Van Hamme and
Wasserman (1994). Van Hamme and Wasserman adapted
the well-known RescorlaWagner model to include the
complementary assumption that the associative strengths
of cues that are absent but expected can also change.
However, the associative change that occurs due to an
absent cue is the opposite of the change that occurs due to
a cue that is present; associations are weakened if the out-
come that is associated with an absent cue is presented, and
they are streng thened if the outcome is also absent. For
instance, in a backward blocking design, the model pro-
posedbyVanHammeandWassermanpredictsthatthe
associative link BO will become weaker during the sec-
ond learning phase because the cue B is absent, but expected
(by virtue of its within-compound association with cue A),
and the outcome associated with it is pr esent. Thus, the
model proposed by Van Hamme and Wasserman predicts
both backward blocking and unovershadowing effects, and
it suggests that both effects are mediated by within-
compound associations (or other factors that cause an absent
cue to be expected during the second learning phase of a
causal learning experiment). Thus, it is difficult to reconcile
the results that were obtained by Larkin et al. (1998) with
the predictions of Van Hamme and Wassermans model.
Larkin et al. (1998) argued that their results could be
better accounted for by Dickinson and Burke s(1996) ad-
aptation of Wagners(1981) SOP model. The original SOP
model proposed that the representation of each stimulus is
composed of simpler elements. Each one of these elements
can be in one of three different activation states: an inactive
(I) state, a high activity (A1) state, or a low activity (A2)
state. According to Wagners model, the elements of which
a stimulus is composed pass from state I to state A1 when
the stimu lus is presented. This activation decreases ov er
time, and the stimulus elements p ass progressively from
state A1 to state A2 and then from stat e A2 to state I. In
addition, elements that have been activated by an associative
link (instead of being activated by the direct presentation of
the stimulus) pass directly from state I to state A2.
According to the original SOP model, learning occurs only
when elements of a cue are in state A1; if the elements that
make up the outcome are in state A1 at the same time that
the elements of the cue are in state A1, the excitatory link
between the cue and the outcome is strengthened. In con-
trast, if elem ents of the outcome are in state A2, an inhibi-
tory link between the cue and a given outcome e lement
develops.
Two more assumptions were added to the revised SOP
model proposed by Dickinson and Burke (1996). First, if
both cue and outcome elements are in state A2 at the same
time, the excitatory link between these elements is strength-
ened. Second, if the cue elements are in state A2 and the
associated outcome elements are in state A1, the inhibitory
link between these elements is strengthened.
It is easy to explain the unovershadowing effect that was
observed by Larkin et al. (1998) within this framework.
During the second learning phase of the design of their
unovershadowing experiment, some elements of cue F are
in activation state A2 at the time that cue E is presented
because of a within-compound association that has formed
between cues E and F. In these trials, the presen tation of cue
E also results in the activation of some of the outcome
elements to state A2 as a result of the EO association that
was learned during the first learning phase. Thus, elements
of both F and O are simultaneous ly in state A2 during these
trials, and the excitatory association between cue F and
outcome O is therefore strengthened.
Interestingly, Larkin et al. (1998; see also Aitken &
Dickinson, 2005) showed that the predictions of the modified
SOP model are more complicated in the case of backward
blocking. During the second learning phase of a backward
blocking experiment, the presentation of cue A results in the
activation of elements of cue B to state A2. The outcome
elements in these trials are also in state A2 because of activa-
tion that has spread from cue A. However, because the
Learn Behav (2013) 41:6176 73
outcome is actually present, some of its elements will also be
in state A1. Thus, the outcome has some elements in state A1
and some elements in state A2, which then allows for the
simultaneous development of both excitatory and inhibitory
associations between the absent cue, B, and the outcome. The
mixture of oppositely signed associations between cue B and
the outcome would then cause the absence of backward block-
ing (for further details, see Aitken & Dickinson, 2005). Note
that this analysis predicts the absence of backward blocking,
regardless of the strength of the within-compound associa-
tions, because within-compound associations in this context
would promote the formation of both excitatory and inhibitory
associations between cue B and the outcome.
Our results do not concur with the results that were
obtained by Larkin et al. (1998), nor do they support the
predictions of the revised SOP model. We obtained reliable
effects of both backward blocking and unovershadowing.
Furthermore, both of these effects depended strongly on the
strength of the within-compound associations. Thus, it
appears that the model that was advanced by Van Hamme
and Wasserman (1994) offers a better account of the results
of the present s tudy th an does the modified SOP model
proposed by Dickinson and Burke (1996).
In accordance with a very different line of re asoning,
some researchers have attempted to explain cue-interaction
phenomena in terms of inferential processes. From their
point of view, associative learning processes do not play
any role in human causal learning. Instead, they argue that
learning occurs as a result of inferential and proposition-
based reasoning (De Houwer, 2009). For instance, partici-
pants in an unovershadowing experiment would create the
proposition that when patient X eats foods E and F, he or
she develops an allergic reaction during the first learning
phase. During the second phase, these participants would
subsequently add another premise to their proposition
namely, when patient X eats food E alone (without F), he
or she does not develop an allergic reaction. On the basis of
these two premises, the participants would then conclude
that food E does not contribute to the patients allergy and
that F must, therefore, be the genuine cause of the allergy.
The inferential account also predicts that within-
compound associations should affect retrospective revalua-
tion phenomena and should not affect forward cue interac-
tion phenomena (Mitchell, Killedar, & Lovibond, 2005). To
continue with the example of unovershadowing, concluding
that cue F is the genuine cause of the allergic reaction
requires that during the second learning phase, the partici-
pant must remember that the cue that accompanied cue E
during the first phase was cue F. This memory is not rele-
vant in the case of forward cue-interaction effects, because
the inference could be made from trials in the second learn-
ing phase in which both cues are presented together. Thus,
an inferential view could reasonably account for the main
findings of our experiments.
It is reasonable to wonder why some of the results of the
present experiments differ from the results of very similar
previous experiments. In Experiment 1 , we found similar
magnitudes for both the backward blocking and unoversha-
dowing effects; this finding is contrary to the results of other
experiments, such as the results of Experiments 1 and 2 in
the aforementioned study by Larkin et al. (1998). A reason-
able explanation for this is that the participants in
Experiment 1 of our study remembered the within-
compound associations particularly well. Indeed, our
manipulation of the within-compound associations attemp-
ted to improve participants abilities to remember the asso-
ciations, so it is the refore not surprising that the vast
majority of the participants remembered at least three of
the four compounds. Thus, if we admit that within-
compound associations are crucial for obtaining a backward
blocking effect, backward b locking effects would be
obtained only in experiments in which the participants were
good at remembering the identities of the compounds.
Because a reliable effect of unovershadowing was
obtained in all three of the experiments conducted by
Larkin et al. (1998), one way of accounting for their results
could be that the magnitudes of backward blocking and
unovershadowing effects differ in terms of their relation to
a within-compound association variable. More research is
needed to further elucidate the details of the relationship
between within-compound associations and the aforemen-
tioned cue-interaction effects.
As was m entioned above, Vandorpe and De Houwer
(2005) obtained reduced overshadowing effects that were
stronger than the forward blocking effects t hat t hey ob-
served. We were not able to replicate their result, even when
using the same method of data analysis, rather than the
nonparametric analysis method that has been used extensively
in this report. Similarly, Chapman (1991, Experiment 1)also
found both effects in the same experiment. In light of the
evidence presented in both Chapmans study and ours, and
considering that the experiment conducted by Vandorpe and
De Houwer included only 18 participants, it seems reasonable
to question their conclusion that a reduced overshadowing
effect can be obtained more easily than a forward blocking
effect.
Finally, we think that some methodological issues in this
study warrant a brief comment. First, the results of Experiment
1 showed that using a combina tion of the resu lts of the
recognition and confidence tests predicted retrospective reval-
uation effects more accurately than did using data from the
recognition test alone. Thus, we encourage researchers to use
this type of Likert-like item when measuring the strength of
within-compound associations (see also Larkin et al., 1998).
74 Learn Behav (2013) 41:6176
Second, we have shown that cue-interaction phenomena can
also be studied by treating Likert-type items as ordinal varia-
bles. Treating the Likert-typ e items as ordinal variables in this
experiment neither limited the number of hypotheses that we
were able to test nor prevented us from performing all of the
usual analyses, including analyses in which the magnitudes of
the effects were used as dependent variables. In addition, we
controlled for type I and II errors without making any assump-
tions about the particular distribution(s) of our dependent var-
iables. Importantly, our analysis methods also allowed us to
minimize the extent to which we made assumptions about the
underlying metric used by the participants to score the percep-
tions of causal strength that they reported in the questionnaires
(for a s imilar argument, see Chapman, 1991).
Authors Note This research was supported by Grant P11-SEJ-7898
from the Junta de Andalucía to David Luque and Grant PI2012-56
from the Basque Government to Miguel A. Vadillo. Amanda Flores
was supported by an FPI scholarship that was awarded by Junta de
Andalucía. We thank Joaquín Morís for his help with statistical anal-
yses. Correspondence concerning this article should be addressed to
the Departamento de Psicología Básica, Facultad de Psicología, Uni-
versidad de Málaga, Campus de Teatinos, s/n, 29072, Málaga,
Spain e-mail: david.l uque@gmai l.com
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, 147159. doi:10.3758/BF03196059
Aitken,M.R.F.,Larkin,M.J.W.,&Dickinson,A.(2001).
Reexamination of the role of within-compound associations in
the ret rospective revaluation of causal judgments. Quarterly
Journal of Experimental Psychology, 54B, 2751.
Amundson, J. C., Witnauer, J. E., Pineño, O., & Miller, R. R. (2008).
An inhibitory within-compound association attenuates oversha-
dowing. Journal of Experimental Psychology. Animal Behavior
Processes, 34, 133143. doi:10.1037/0097-7403.34.1.133
Chapman, G. B. (1991). Trial order affects cue interaction in contin-
gency judgment. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 17, 837854.
De Houwer, J. (2009). T he propositional approach to associative
learning as an alternative for association formation models.
Learning & Behavior, 37, 120. doi:10.3758/LB.37.1.1
De Houwer, J., Beckers, T., & Vandorpe, S. (2005). Evidence for the
role of higher order reasoning processes in cue competition and
other learning phenomena. Learning & Behavior, 33, 239249.
doi:10.3758/BF03196066
Dick inson, A., & Burke, J. (1996). Within-compound associations
media te the retrospective revaluation of causality judgements.
Quarterly Journal of Experimental Psychology, 49B, 6080.
Dickinson, A., Shanks, D. R., & Evenden, J. L. (1984).
Judgement of actoutcome contingency: The role of selective
attribution. Quarterly Journal of Experimental Psychology,
36A, 2950.
Ghirlanda, S. (2005). Retrospective revaluation as simple associative
learning. Journal of Experimental Psychology. Animal Behavior
Processes, 31, 107111. doi:10.1037/0097-7403.31.1.107
Gopnik, A., Glymour, C., Sobel, D. M., Schulz, L. E., Kushnir, T., &
Danks, D. (2004). A theory of causal learning in children: Causal
maps and Bayes nets. Psychological Review, 1 11, 332. doi:10.1037/
0033-295X.111.1.3
Holyoak, K. H., & Cheng, P. W. (2011). Causal learning and inference as a
relational process: The new synthesis. Annual Review of Psychology,
62, 135163. doi:10.1146/annurev.psych.121208.131634
Kahler, E., Rogausch, A., Brunner, E., & Himmel, W. (2008). A
parametric analysis of ordinal quality-of-life data can lead to
erroneous results. Journal of Clinical Epidemiology, 61, 475480.
Kamin, L. J. (1968). Attention-like processes in classical condition-
ing. In M. R. Jones (Ed.), Miami Symposium on the Prediction of
Behavior, 1967: Aversive Stimulation (pp. 931). Coral Gables,
FL: University of Miami.
Kruschke, J. K., & Blair, N. J. (2000). Blocking and backward block-
ing involve learned inattention. Psychonomic Bulletin & Review,
7, 636645. doi:10.3758/BF03213001
Larkin, M. J., Aitken, M. R. F., & Dickinson, A. (1998). Retrospective
revaluation of causal judgments under positive and negative con-
tingencies. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 24, 13311352.
López, F. J., & Shanks, D. R. (2008). Models of animal learning and
their relations to human learning. In R. Sun (Ed.), Handbook of
computational cognitive modelling (pp. 589611). Cambridge,
MA: Cambridge University Press.
Markman, A. B. (1989). LMS rules and the inverse base-rate effect: Comment
on Gluck and Bower (1988). Journal of Experimental Psychology.
General, 118, 417421. doi:10.1037//0096-3445.118.4.417
Melchers, K. G., Lachnit, H., & Shanks, D. R. (2004). Within-compound
associations in retrospective revaluation and in causal learning: A
challenge for comparator theory. Quarterly Journal of Experimental
Psychology, 57B, 2553. doi:10.1080/02724990344000042
Melchers, K. G., Lachnit, H., & Shanks, D. R. (2006). The comparator
theory fails to account for the selective role of within-compound
associations in cue-selection effects. Experimental Psychology,
53, 316320. doi:10.1027/1618-3169.53.4.316
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.
5192). San Diego, CA: Academic Press.
Mitchell, C. J., De Houwer, J., & Lovibond, P. F. (2009). The propo-
sitional nature of human associative learning. The Behavioral and
Brain Sciences, 32, 183198. doi:10.1017/S0140525X09000855
Mitchell, C. J., Killedar, A., & Lovibond, P. F. (2005). Inference-based
retrospective revaluation in human causal judgments r equires
knowledge of within-compound relationships. Journal of
Experimental Psychology. Animal Behavior Processes, 31, 418
424. doi:10.1037/0097-7403.31.4.418
Mitchell, C. J., Lovibond, P. F., & Gan, C. Y. (2005). A dissociation
between causal judgment and outcome recall. Psychonomic
Bulletin & Review, 12, 950954. doi:10.3758/BF03196791
Munzel, U., & Bandelow, B. (1998). The use of parametric vs. non
parametric tests in the statistical evaluation of rating scales.
Pharmacopsychiatry, 31, 222224.
Norris, C. M., Ghali, W. A., Saunders, L. D., Brant, R., & Galbraith, P.
D. (2004). Systematic review of statistical methods used to ana-
lyze Seattle Angina Questionnaire scores. The Canadian Journal
of Cardiology, 20, 187193.
Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian
conditioning: Variations in the effectiveness of reinforcement
and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.),
Classical conditioning II: Current research and theory
(pp. 64
99). New York: Appleton-Century-Crofts.
Shah, D. A., & Madden, L. V. (2004). Nonparametric analysis of ordinal
data in designed factorial experiments. Phytopathology, 94,3334.
Learn Behav (2013) 41:6176 75
Shanks, D. R. (1985). Forward and backward blocking in human
contingency judgment. Quarterly J ournal of Experimental
Psychology, 37B, 121.
Shanks, D. R. (2007). Associationism and cognition: Human contin-
gency learning at 25. Quarterly Journal of Experimental
Psychology, 60, 291309. doi:10.1080/17470210601000581
Shanks, D. R. (2010). Learning: From association to cog nition .
Annual Review of Psychology, 61, 27130 1. doi:10.1146/
annurev.psych.093008.100519
Siegel, S., & Castellan, N. J. (1988). Nonparametric statistics for the
behavioral sciences (2nd Ed.). New York: McGraw-Hill.
Singer, J. M., Poleto, F. Z., & Rosa, P. (2004). Parametric and non-
parametric analyses of repeated ordinal categorical data.
Biometrical Journal, 46, 460473.
Stout, S. C., & Miller, R. R. (2007). Sometimes competing retrieval (SOCR):
A formalization of the comparator hypothesis. Psychological Review,
114, 759783. doi:10.1037/0033-295X.114.3.759
Sutton, R. S., & Barto, A. G. (1981). Toward a modern theory of
adaptive networks: Expectation and prediction. Psychological
Review, 88, 135170.
Tassoni, C. J. (1995). The least mean squares network with information
coding: A model of cue lea rning. Journal of Experiment al
Psychology: Learning, Memory, and Cogniti on, 21, 193204.
doi:10.1037//0278-7393.21.1.193
Vandorp e, S., & De Houw er, J. (2005). A comparison of forwa rd
blocking and reduced overshadowing in human causal learning.
Psychonomic Bulletin & Review, 12, 945 949. doi:10.3758/
BF03196790
Vandorpe, S., De H ouwer, J., & Beckers, T. (2007). The role of
memory for compounds in cue competiti on. Learning and
Motivation, 38, 195207. doi:10.1016/j.lmot.2007.03.001
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, 127151.
Wagner, A. R. (1981). SOP: A model of automatic memory processing
in animal behavior. In N. E. Spear & R . R. Miller (Eds.),
Information processing in animals: Memory mechanisms (pp. 5
47). Hillsdale, NJ: Erlbaum.
Wasserman, E. A. (1990). Attribution of causality to common and
distinctive elements of compound stimuli. Psychological
Science, 1, 298302. doi:10.1111/j.1467-9280.1990.tb00221.x
Wasserman, E. A., & Berglan, L. R. (1998). Backward blocking and
recovery from overshadowing in human causal judgement: The
role of within-compound associations. Quarterly Journal of
Experimental Psychology, 51B, 121138.
Wasserman, E. A., & Castro, L. (2005). Surprise and change: Variations
in th e strength of present and absent cues in causal learning.
Learning & Behavior, 33, 131146. doi:10.3758/BF03196058
76 Learn Behav (2013) 41:6176
... For example, the Rescorla and Wagner (1972) model can account for forward blocking but not backward blocking without recourse to a within-compound association, and the model can often (not always, e.g., Westbrook et al., 2002) account for context specificity of conditioned behavior as seen here in forward blocking based on direct associations to the context. Supporting the view that backward blocking but not forward blocking depends on there being a within-compound association, Luque et al. (2013) manipulated the strength of the within-BLOCKING AND CUE INTERFERENCE 11 compound association in blocking by assigning cues based on their participants' preexperimental associations. They found that when there were weak or no within-compound associations, backward blocking, but not forward blocking, was greatly reduced. ...
... The conclusion that backward blocking is seemingly not a "pure" cue competition effect, as is overshadowing, is not new. For example, based on evidence of a role for the within-compound association in backward blocking but not forward blocking, Luque et al. (2013) concluded that backward and forward blocking arise at least in part from different mechanisms. Luque et al. manipulated the strength of the within-compound association by assigning cues based on their participants' preexperimental associations. ...
Article
Full-text available
Blocking (i.e., reduced responding to cue X following YX-outcome pairings in Phase 2 as a consequence of cue Y having been paired with the outcome in Phase 1) is one of the signature phenomena in Pavlovian conditioning. Its discovery promoted the development of multiple associative models, most of which viewed blocking as an instance of pure cue competition (i.e., a decrease in responding attributable to training two conditioned stimuli in compound). Two experiments are reported in which rats were examined in a fear conditioning paradigm (i.e., lick suppression), and context dependency of retrieval at test was used as an index of associative cue interference (i.e., a decrease in responding to a target cue as a result of training a second cue with the same outcome but without concurrent presentation of the two cues). Specifically, we observed renewal of forward-blocking which parallels renewal of proactive interference, and renewal of backward-blocking which parallels renewal of retroactive interference. Thus, both backward-blocking (Experiment 1, embedded in a sensory preconditioning design) and forward-blocking (Experiment 2, conducted in first-order conditioning) appear to be influenced by retroactive and proactive interference, respectively, as well as cue competition. Consequently, blocking, long regarded as a benchmark example of pure cue competition, is sometimes a hybrid of cue competition and associative interference. Finally, the Discussion considers whether stimulus competition and associative interference are two independent phenomena or products of a single underlying process. (PsycInfo Database Record (c) 2022 APA, all rights reserved).
... In the second case, when the previous belief is that A prevents the outcome from occurring, the detection of a causal relationship between the second cause B and the outcome will be facilitated (i.e., this particular case of cue interaction is generally known as superconditioning; Rescorla, 1971). Many cue interaction experiments, both with animals and humans, show that learning about the relationship between a potential cause and an outcome can result altered when the potential cause is presented in compound with another potential cause that has been previously associated either with the outcome or its absence (Aitken, Larkin, & Dickinson, 2000;Arcediano, Matute, Escobar, & Miller, 2005;Dickinson, Shanks, & Evenden, 1984;Kamin, 1968;Luque, Flores, & Vadillo, 2013;Luque & Vadillo, 2011;Mor ıs, Cobos, Luque, & L opez, 2014;Rescorla, 1971;Shanks, 1985). ...
... Nevertheless, what is clear given the present results is that the group who developed a high illusion about the ineffective Medicine A tended to assess the effective Medicine B as less effective than the group who developed a weaker illusion. Similar cue interaction effects have been clearly established in other causal learning research (Aitken et al., 2000;Arcediano et al., 2005;Dickinson et al., 1984;Luque & Vadillo, 2011;Luque et al., 2013;Mor ıs et al., 2014;Shanks, 1985). Moreover, cue interaction effects are beautifully predicted by current theories of learning (Rescorla & Wagner, 1972). ...
Article
Full-text available
Cognitive illusions are often associated with mental health and well-being. However, they are not without risk. This research shows they can interfere with the acquisition of evidence-based knowledge. During the first phase of the experiment, one group of participants was induced to develop a strong illusion that a placebo medicine was effective to treat a fictitious disease, whereas another group was induced to develop a weak illusion. Then, in Phase 2, both groups observed fictitious patients who always took the bogus treatment simultaneously with a second treatment which was effective. Our results showed that the group who developed the strong illusion about the effectiveness of the bogus treatment during Phase 1 had more difficulties in learning during Phase 2 that the added treatment was effective.
... Finally, participants are asked to judge which foods are causing Mr. X to suffer an allergic reaction, and this judgment may also take the form of a probability or cued recall judgment. The food allergist task has been used to study cue competition effects such as blocking and overshadowing (e.g., Shanks and Lopez, 1996;Aitken et al., 2000;Lovibond et al., 2003;Beckers et al., 2005a;Mitchell et al., 2005Mitchell et al., , 2006Vandorpe et al., 2007;Livesey et al., 2013Livesey et al., , 2019bLuque et al., 2013;Uengoer et al., 2013), learning of preventative relationships such as in the case of conditioned inhibition Boakes, 2004, 2007), complex rule learning tasks such as the patterning task (Shanks and Darby, 1998;Wills et al., 2011;Don et al., 2020), as well as a host of phenomena related to learned attentional changes including the learned predictiveness effect (Le Pelley and McLaren, 2003;Don and Livesey, 2015;Shone et al., 2015), the inverse baserate effect (Don et al., 2019), outcome predictability effects (Griffiths et al., 2015;Thorwart et al., 2017), and other related transfer effects (Livesey et al., 2019a). Food allergies are relatively commonplace such that, by the time they enter the laboratory, participants have a lifetime of experience with food and its ability to cause allergic reactions in oneself or others. ...
Article
Full-text available
Causal and predictive learning research often employs intuitive and familiar hypothetical scenarios to facilitate learning novel relationships. The allergist task, in which participants are asked to diagnose the allergies of a fictitious patient, is one example of this. In such studies, it is common practice to ask participants to ignore their existing knowledge of the scenario and make judgments based only on the relationships presented within the experiment. Causal judgments appear to be sensitive to instructions that modify assumptions about the scenario. However, the extent to which prior knowledge continues to affect competition for associative learning, even after participants are instructed to disregard it, is unknown. To answer this, we created a cue competition design that capitalized on prevailing beliefs about the allergenic properties of various foods. High and low allergenic foods were paired with foods moderately associated with allergy to create two compounds; high + moderate and low + moderate. We expected high allergenic foods to produce greater competition for associative memory than low allergenic foods. High allergenic foods may affect learning either because they generate a strong memory of allergy or because they are more salient in the context of the task. We therefore also manipulated the consistency of the high allergenic cue-outcome relationship with prior beliefs about the nature of the allergies. A high allergenic food that is paired with an inconsistent allergenic outcome should generate more prediction error and thus more competition for learning, than one that is consistent with prior beliefs. Participants were instructed to either use or ignore their knowledge of food allergies to complete the task. We found that while participants were able to set aside their prior knowledge when making causal judgments about the foods in question, associative memory was weaker for the cues paired with highly allergenic foods than cues paired with low allergenic foods regardless of instructions. The consistency manipulation had little effect on this result, suggesting that the effects in associative memory are most likely driven by selective attention to highly allergenic cues. This has implications for theories of causal learning as well as the way causal learning tasks are designed.
... In contrast allergist tasks are usually run with multiple filler cues that do not predict the outcome and most participants in these experiments show very clear discrimination between trained and novel cues. carrots or dates (i.e., B < C or D, see for example, Luque et al., 2013). Figure 1 depicts an example of a simple standard associative network after training and its interaction with the events in the world. ...
Article
Full-text available
Associative learning theories offer one account of the way animals and humans assess the relationship between events and adapt their behavior according to resulting expectations. They assume knowledge about event relations is represented in associative networks, which consist of mental representations of cues and outcomes and the associative links that connect them. However, in human causal and contingency learning, many researchers have found that variance in standard learning effects is controlled by “non-associative” factors that are not easily captured by associative models. This has given rise to accounts of learning based on higher-order cognitive processes, some of which reject altogether the notion that humans learn in the manner described by associative networks. Despite the renewed focus on this debate in recent years, few efforts have been made to consider how the operations of associative networks and other cognitive operations could potentially interact in the course of learning. This paper thus explores possible ways in which non-associative knowledge may affect associative learning processes: (1) via changes to stimulus representations, (2) via changes to the translation of the associative expectation into behavior (3) via a shared source of expectation of the outcome that is sensitive to both the strength of associative retrieval and evaluation from non-associative influences.
... Therefore , following these authors, at least two different systems appear to serve human associative learning, one propositional—responsible for previous results obtained using more standard, slow-paced tests—and another associative, which would be responsible for data obtained in fast-paced tests. For example, in [14]'s Experiment 4, a dissociation between verbal ratings and speeded performance in a priming test was reported regarding the effects of explicit instructions in the blocking effect (i.e., an effect whereby no predictive value is attributed to a cue—the blocked cue—when this blocked cue is presented together with another cue—the blocking cue—with a higher predictive value [15, 16] ). Associative models explain blocking as a consequence of the operation of an associative learning algorithm. ...
Article
Full-text available
Propositional and associative processes have been proposed to explain human associative learning. Our main objective in this study was to evaluate whether propositional knowledge may gain control over behavior even under high time-pressure conditions, as suggested by propositional single-process models. In the experiment reported, different groups of participants had to learn a series of cue-outcome relationships on a trial-by-trial basis under different time pressure conditions. Later, a simple verbal instruction indicated that one of the cues had reversed its contingency (informed condition). The other cue had also changed its contingency, though in an unanticipated way (uninformed condition) whilst other contingencies did not change (no-change condition). The results showed that, in the absence of instructions, interference (i.e., uninformed vs. no-change effect) was greater in the high time than in the low time-pressure group. This result indicates that those responses which were previously relevant are more difficult to inhibit when there is little time to respond. However, time pressure had no detectable effect on the use of the verbal instruction, since an equivalent instruction advantage (i.e., uninformed vs. informed effect) was obtained in both time pressure groups. These results reveal that propositional knowledge can override those cue-outcome relationships that were learnt trial-by-trial even under conditions of high cognitive demand. This pattern of results is consistent with a propositional single-process model of associative learning.
... These include questions of how manipulation of one cue may affect learning about another cue with which it is simultaneously reinforced (e.g., overshadowing, potentiation), with which it has been previously paired 37 (e.g., sensory preconditioning, retrospective revaluation), or with which it is subsequently paired (e.g., forward blocking). This general set of questions involves theoretical issues of whether compound stimuli are processed elementally or configurally (e.g., Allman & Honey, 2006;Gonzalez, Quinn, & Fanselow, 2003;Harris, Gharaei, & Moore, 2009;Lachnit, Schultheis, Konig, Ungor, & Melchers, 2008), under what conditions stimulus elements compete with each other for associative strength (Pearce, Graham, Good, Jones, & McGregor, 2006;, and in what ways reinforcement of one stimulus can affect responding to an absent but associated stimulus (Blaser, Couvillon, & Bitterman, 2004;Bradfield & McNally, 2008;Dopson, Pearce, & Haselgrove, 2009;Luque, Flores, & Vadillo, 2013). The results of these experiments suggest that a number of parametric variables such as the CS-US interval, pairing frequency, US intensity, and stimulus modality, may affect cue interactions, although in most cases additional exploration and replication are required to establish these effects. ...
Article
The study of learning has long held a central position in the field of comparative psychology. Here we present a survey of the past 20 years of comparative learning research, covering publications from 1994 to 2013. We selected seven journals with a strong focus on comparative learning, and identified five major topics of study represented by the publications in those journals: non-associative learning, associative learning, perceptual/object learning, social learning, and neural correlates of learning. Of these topics, associative learning was by far the most popular, comprising about 85% of the research in comparative learning. We therefore subdivided this topic into seven subcategories of research questions, which included causal reasoning, compound cue interactions, extinction, stimulus control, outcome learning and motivation, spatial learning, and temporal integration or timing. The number of publications addressing each topic or research question, as well as the number of citations received by these publications, was examined for the combined seven journals across the 20 year period of review. The subject of spatial learning has grown rapidly over the past 20 years, and has attracted robust interest by researchers both in and outside of the field of comparative psychology. Although much less popular in terms of publication number, recent growth was also identified for studies of causal reasoning, social learning, and perceptual or object learning.
Article
The correlation between blocking and within-compound memory is stronger when compound training occurs before elemental training (i.e., backward blocking) than when the phases are reversed (i.e., forward blocking; Melchers et al., 2004; 2006). This trial order effect is often interpreted as problematic for performance-focused models that assume a critical role for within-compound associations in both retrospective revaluation and traditional cue competition. The present manuscript revisits this issue using a computational modeling approach. The fit of sometimes competing retrieval (SOCR; Stout & Miller, 2007) was compared to the fit of an acquisition-focused model of retrospective revaluation and cue competition. These simulations reveal that SOCR explains this trial order effect in some situations based on its use of local error reduction.
Article
Full-text available
Blocking refers to the finding that less is learned about the relationship between a stimulus and an outcome if pairings are conducted in the presence of a second stimulus that has previously been established (via pretraining) as a reliable predictor of that outcome. Attentional models of associative learning suggest that blocking reflects a reduction in the attention paid to the blocked cue. We tested this idea in three experiments in which participants were trained in an associative learning task using a blocking procedure. Attention to stimuli was measured 250 ms after onset using an adapted version of the dot probe task. This task was presented at the beginning of each associative learning trial (Experiments 1 & 2) or in independent trials (Experiment 3). Each experiment found evidence of reduced attention to blocked stimuli as compared to pretrained stimuli, but no evidence of a corresponding difference in a control condition. In addition, this attentional bias correlated positively with the magnitude of blocking in associative learning, as measured by predictive-value judgments. Moreover, Experiments 2 and 3 found evidence of an influence of learning about predictiveness on memory for episodes involving stimuli. These findings are consistent with a central role of learned attentional biases in producing the blocking effect, and in the encoding of new memories. Although these attentional biases were likely created in an effortful way during learning, they exerted an influence on learning and memory very rapidly and independently of participants’ ongoing task goals.
Article
Full-text available
Five experiments involving human causal learning were conducted to compare the cue competition effects known as blocking and unovershadowing, in proactive and retroactive instantiations. Experiment 1 demonstrated reliable proactive blocking and unovershadowing but only retroactive unovershadowing. Experiment 2 replicated the same pattern and showed that the retroactive unovershadowing that was observed was interfered with by a secondary memory task that had no demonstrable effect on either proactive unovershadowing or blocking. Experiments 3a, 3b, and 3c demonstrated that retroactive unovershadowing was accompanied by an inflated memory effect not accompanying proactive unovershadowing. The differential pattern of proactive versus retroactive cue competition effects is discussed in relationship to amenable associative and inferential processing possibilities.
Article
Additivity-related assumptions have been proven to modulate blocking in human causal learning. Typically, these assumptions are manipulated by means of pretraining phases (including exposure to different outcome magnitudes), or through explicit instructions. In two experiments, we used a different approach that involved neither pretraining nor instructional manipulations. Instead, we manipulated the causal structure in which the cues were embedded, thereby appealing directly to the participants' prior knowledge about causal relations and how causes would add up to yield stronger outcomes. Specifically, in our "different-system" condition, the participants should assume that the outcomes would add up, whereas in our "same-system" condition, a ceiling effect would prevent such an assumption. Consistent with our predictions, Experiment 1 showed that, when two cues from separate causal systems were combined, the participants did expect a stronger outcome on compound trials, and blocking was found, whereas when the cues belonged to the same causal system, the participants did not expect a stronger outcome on compound trials, and blocking was not observed. The results were partially replicated in Experiment 2, in which this pattern was found when the cues were tested for the second time. This evidence supports the claim that prior knowledge about the nature of causal relations can affect human causal learning. In addition, the fact that we did not manipulate causal assumptions through pretraining renders the results hard to account for with associative theories of learning.