Content uploaded by Julie Y.L. Chow
Author content
All content in this area was uploaded by Julie Y.L. Chow on Jun 06, 2022
Content may be subject to copyright.
Running head: Inhibitory causal structures 1
Reversal of inhibition by no-modulation training but not by extinction in human causal learning
Peter F. Lovibond
Julie Y.L. Chow
Cheryl Tobler
Jessica C. Lee
The University of New South Wales
This study was funded by an Australian Research Council Discovery Project Grant
[DP190103738] to Peter Lovibond. Jessica Lee was supported by a Discovery Early Career
Researcher Award from the Australian Research Council [DE210100292].
Please address correspondence to p.lovibond@unsw.edu.au
Inhibitory causal structures 2
Abstract
One of the many strengths of the Rescorla-Wagner (1972) model is that it accounts for both
excitatory and inhibitory learning using a single error-correction mechanism. However, it
makes the counterintuitive prediction that non-reinforced presentations of an inhibitory
stimulus will lead to extinction of its inhibitory properties. Zimmer-Hart and Rescorla (1974)
provided the first of several animal conditioning studies that contradicted this prediction.
However, the human data are more mixed. Accordingly, we set out to test whether extinction
of an inhibitor occurs in human causal learning after simultaneous feature negative training
with a conventional unidirectional outcome. In two experiments with substantial sample sizes
we found no evidence of extinction after presentations of the inhibitory stimulus alone in either
a summation test or causal ratings. By contrast, two “no-modulation” procedures that
contradicted the original training contingencies successfully reversed inhibition. These results
did not differ substantially as a function of participants’ self-reported causal structures
(configural/modulation/prevention). We hypothesise that inhibitory learning may be
intrinsically modulatory, analogous to negative occasion-setting, even with simultaneous
training. This hypothesis would explain why inhibition is reversed by manipulations that
contradict modulation but not by simple extinction, as well as other properties of inhibitory
learning such as imperfect transfer to another excitor.
Keywords: inhibition, extinction, prediction error, feature negative discrimination, modulation,
causal structure
Inhibitory causal structures 3
No extinction of inhibition in human causal learning with a unidirectional outcome
In the 1960s Robert Rescorla played a leading role in re-invigorating the study of
inhibition in associative learning (Rescorla & LoLordo, 1965; Rescorla, 1967, 1969). In
particular, he investigated Pavlov’s (1927) conditioned inhibition procedure, in which a
training excitor A is paired with the unconditioned stimulus (US) when presented alone but not
when combined with an inhibitory stimulus B, and extended it to aversive conditioning and
negative correlation designs. He showed that such a procedure (now often referred to as feature
negative training) slowed subsequent excitatory learning to B (retardation test), and led to
transfer of B’s inhibitory properties to a separately trained excitor (summation test).
When Rescorla and Wagner formulated their famous error-correction theory (Rescorla
& Wagner, 1972; Wagner & Rescorla, 1972), they maintained Pavlov’s view of inhibition as
the direct opposite of excitation. This opposition was implemented in their use of a single bi-
directional scale for associative strength, with opposite signs for excitation and inhibition. The
model accounted for the known properties of inhibitory learning, including the conditions for
its establishment (negative prediction error) and the outcomes of summation and retardation
tests. Furthermore, it did so with exactly the same formula as was used to account for
excitatory learning and recently discovered stimulus competition effects such as blocking
(Kamin, 1969), correlation learning (Rescorla, 1967) and relative validity (Wagner, 1969). The
Rescorla-Wagner model has become the definitive model of associative learning and has also
been highly influential in the human causal and predictive learning literature (e.g., Gershman,
2015; Shanks & Dickinson, 1987; Van Hamme & Wasserman, 1994).
However, the Rescorla-Wagner model makes the problematic prediction that non-
reinforced presentations of an inhibitory stimulus alone will lead to extinction of its inhibitory
properties. This prediction is counterintuitive because the outcome that occurs on such trials
(no US) appears to be the same as that predicted by the inhibitor (no US), which should lead to
Inhibitory causal structures 4
no change in associative strength. But because inhibition is represented with a negative sign,
the model indicates that there will in fact be positive prediction error, arising from the
difference between the actual outcome (zero) and the associative strength of the inhibitor
(which is negative). Characteristically, Rescorla himself was the first to identify this potential
shortcoming and to publish empirical findings that contradicted the model’s prediction
(Zimmer-Hart and Rescorla, 1974). Subsequent studies in animal conditioning have largely
confirmed this result (e.g., Lysle & Fowler, 1985; Williams & Overmier, 1988).
In the present research we were concerned with extinction of inhibition in humans. This
topic is of particular relevance to the human causal learning literature because inhibitory
learning has been seen as a potential model for how people learn about preventive relationships
(e.g., Shanks & Dickinson, 1987). However, the evidence regarding extinction of inhibition in
human causal learning is mixed. Yarlas, Cheng and Holyoak (1995) provided a brief report of
an experiment in which they used a hypothetical medical scenario to pair different
combinations of cues (biochemical substances) with the presence or absence of an outcome
(disease). After feature negative training, they presented the inhibitory feature I without the
disease in one group (direct extinction condition) but not in a control group. There was no
change in causal rating or outcome predictions for I in the Extinction group, nor any difference
between the extinction and control groups. Thus, they found no evidence for extinction of
inhibition after exposure to the inhibitor alone, consistent with the animal findings.
Melchers, Wolff and Lachnit (2006) revived the issue when they suggested that
extinction of inhibition might occur if a bidirectional outcome were employed during training,
since this would align better with the assumptions of the Rescorla-Wagner model regarding
inhibition being opposite to excitation. The idea was that with such a procedure the inhibitor
would elicit an expectancy that the outcome level would decrease, which would differ from the
actual consequence given during the extinction phase (no change in outcome level), thus
Inhibitory causal structures 5
generating prediction error. Melchers et al. (2006) tested this proposal by assessing extinction
of inhibition as a function of the type of outcome employed, in a medical prediction task. They
found that non-reinforced presentations of the inhibitor did in fact lead to a loss of its
inhibitory strength in a group trained with a bidirectional outcome (hormone level with a non-
zero baseline) but not in a group trained with a unidirectional outcome. This result was
replicated, with additional controls, by Lotz and Lachnit (2009).
Although the results from the bidirectional conditions in Melchers et al. (2006) and
Lotz and Lachnit (2009) make sense from an expectancy perspective, they do not really resolve
the original concern with the Rescorla-Wagner model. In particular, the results from the
unidirectional conditions are still problematic for the model, which does not postulate a
bidirectional outcome dimension. Rather, it was designed from the outset to explain inhibitory
learning with unidirectional outcomes such as food and shock which can vary in magnitude in
a positive direction but cannot take on negative values. Furthermore, Baetu and Baker (2010)
replicated the design used by Melchers et al. (2006) and Lotz and Lachnit (2009), modifying
certain aspects such as the rating scale, and found evidence for extinction of inhibition in both
the bidirectional and unidirectional groups. Their results thus favored the predictions of the
Rescorla-Wagner (1972) model. Finally, Kutlu and Schmajuk (2012) observed some extinction
of inhibition in a summation test after training with an outcome that was ostensibly
unidirectional (high/low height of a bar, where the low outcome was near zero).
In the present research, we were specifically interested in the question of extinction of
inhibition when a conventional unidirectional outcome is employed, because this is the
condition relevant to evaluation of the Rescorla-Wagner (1972) model, and it is also the
condition where the published results are in conflict. We investigated this question within a
conventional causal judgment procedure developed by Van Hamme and Wasserman (1994),
the allergist task. In this task, the predictive stimuli are foods eaten by a hypothetical patient in
Inhibitory causal structures 6
a given meal, and the outcome is an allergic reaction which is either present or absent. Such an
outcome is intrinsically unidirectional since it cannot take on negative values. We have
conducted several studies on conditioned inhibition using this task, and have established a
protocol that yields reliable inhibition in a summation test relative to a conservative control
stimulus trained in a non-reinforced compound (Lee & Lovibond, 2021; Lovibond & Lee,
2021).
Using this procedure, we have identified a factor that may be critical in determining the
impact of non-reinforced presentations of an inhibitor, namely the causal structure that
participants infer when exposed to an inhibitory contingency. Specifically, we found that
participants report having learned a variety of different causal structures after simultaneous
feature negative training (A+/AB-). Some describe the inhibitory feature B as directly
preventing the allergic outcome. We take this “prevention” structure to be analogous to the
type of inhibitory learning inherent in the Rescorla-Wagner (1972) model, where inhibition
directly opposes excitation. Others report that B prevents the training excitor from causing the
outcome. We label this structure “modulation” and consider it to be analogous to Holland’s
view of negative occasion-setting, where the feature is thought to gate the association between
the training excitor and the outcome (see Fraser & Holland, 2019, for a review). Finally, some
participants report simply having learned what follows each combination of stimuli, with little
consideration of the properties of individual stimuli. We have labelled this type of response as
“configural” as it is similar to strategies that have previously been identified in causal learning
(e.g., Williams et al., 1994) as well as in configural theories of learning (e.g., Pearce, 1987). In
the present project, we tested whether these individual differences might be associated with
different patterns of inhibitory extinction. In particular, we hypothesised that modulation
participants would be unaffected by extinction of the inhibitory feature, because this type of
learning is not directly contradicted by presentation of the inhibitor without its target excitor.
Inhibitory causal structures 7
Conversely, prevention participants might show a loss of inhibition as predicted by the
Rescorla-Wagner (1972) model.
Experiment 1
The aim of Experiment 1 was to test the effect of non-reinforced presentations of an
inhibitor after training with a simultaneous conditioned inhibition (feature negative) design
using a unidirectional outcome. We also included an assessment of self-reported causal
structure in order to test whether it predicts the degree of extinction observed. The overall
procedure followed that of our previous work (e.g., Lovibond & Lee, 2021). The experimental
design is shown in Table 1. Letters represent foods eaten by a fictitious patient, and +
represents the occurrence of an allergic reaction. In Phase 1, A+ and AB- trials were used to
establish B as a putative inhibitor. C+ trials established C as an excitor for the summation test,
and DE- trials provided a control cue D that had been trained in a similar way to B but not in
the presence of an excitor. Causal structure was assessed by both open-ended and forced choice
questions administered immediately after the inhibitory training. Phase 2 was the only phase
that differed between groups. The Extinction group received B- trials to test the Rescorla-
Wagner (1972) prediction of extinction of inhibition. The No-Mod group received A+ and
AB+ trials intended to directly contradict B’s ability to modulate A’s relationship to the
outcome. This “no-modulation” procedure was also employed by Zimmer-Hart and Rescorla
(1974), who found that it effectively counteracted B’s inhibitory properties. The Control group
did not receive any trials involving B.
In the Test phase, CB trials served as a summation test for B with the separately trained
excitor C, and CD trials provided a control. After the Test phase, participants rated the causal
strength of individual stimuli on a cause-prevent scale and completed a second assessment of
causal structure. Any reduction in the inhibitory properties of B would be seen in less transfer
in the summation test and less inhibitory causal ratings. We expected the impact of the
Inhibitory causal structures 8
extinction procedure to vary as a function of causal structure, as noted above. Conversely, we
expected the no-modulation procedure to effectively reverse the inhibitory properties of B in
all participants, because it contradicts all inhibitory structures. For example, the Rescorla-
Wagner predicts that A+/AB+ training will increase B’s associative strength to an asymptote of
zero. Thus the No-Mod group served as a reference condition for comparison with the
extinction manipulation.
Method
Participants
In this experiment 206 undergraduate students (129 female, M age = 19.3, SD age =
3.3) participated in exchange for course credit. This sample size was based on the rate of
exclusions and the number of participants classified in the causal structure subgroups in our
previous research, with the goal of achieving >80% power for detecting medium (d=0.5) effect
sizes in between-group comparisons.
Apparatus and Stimuli
As in Lee and Lovibond (2021), the experimental stimuli (A-L) were selected randomly
from a pool of 16 food pictures that included a verbal label (e.g., “chicken”). The allergic
reaction outcome consisted of the text “Allergic Reaction!” accompanied by a sad face
emoticon. No outcome consisted of the text “No allergic reaction”. The experiment was
programmed using the jspsych library (de Leeuw, 2015), hosted using JATOS (Lange, Kühn,
& Filevich, 2015) and run on participants’ web browsers.
Procedure
The project was approved by the University of New South Wales Human Research
Ethics Advisory Panel C (approval number 3136), and participants provided online consent.
Table 1 shows the sequence of phases. The design and procedure followed that used in our
Inhibitory causal structures 9
previous research (Lee & Lovibond, 2021; Lovibond & Lee, 2021). In brief, participants were
asked to play the role of an allergist trying to work out which foods cause allergic reactions in a
patient “Mr X”. Each trial represented a meal eaten by the patient. Participants were presented
with the text “Mr X eats” followed by pictures of either one or two foods with their verbal
labels. After 500ms, participants were asked to rate the likelihood that Mr X would show an
allergic reaction on a visual analogue scale from “Definitely NO ALLERGIC REACTION” to
“Definitely ALLERGIC REACTION”. The predictive ratings were recorded on a numerical
scale from 0 to 100. When participants had made their rating, the prediction scale and prompt
were replaced by either the allergic outcome or no outcome. After 2s, the stimuli and feedback
disappeared and the 1-s blank inter-trial-interval (ITI) period commenced.
Phase 1. As shown in Table 1, Phase 1 followed our previous training protocol (Lee &
Lovibond, 2021; Lovibond & Lee, 2021) in which A+ and AB- trials were used to establish B
as a conditioned inhibitor. C+ trials trained C as an excitatory (causal) stimulus for the later
summation test. Stimulus D provided a control stimulus for B, having been similarly trained in
a non-predictive compound but with no source of excitation. The remaining stimuli were
fillers, included to help prevent participants from inferring general rules such as all single
foods predict the allergic outcome. The order of trials was randomised, with the restriction that
the same trial type could not be presented on successive trials. Each trial type was presented 6
times (twice within each of 3 blocks of 16 trials). The left-to-right presentation of the
compound stimuli was counterbalanced within each block.
Phase 2. A similar procedure was followed in Phase 2, where each group received 5
trial types designed to implement the between-group manipulation. The Extinction group
received B- trials to test for extinction of inhibition, while the No-Mod group received A+ and
AB+ trials, intended to more strongly contradict the inhibitory training to B. Each group was
also given novel trials (J+/JK+ or L-) to match the contingencies experienced by the other
Inhibitory causal structures 10
group. The Control group received both sets of novel trials, and no trials involving B. All
groups received C+ and F- trials to maintain some continuity with Phase 1.
Between Phases 1 and 2, participants completed an initial assessment of their inferred
causal structure for stimulus B. This consisted of two parts. The first was an open-ended
question asking them to describe what they had learned about stimulus B. The second was a 3-
alternative forced choice (3AFC) question asking them to choose one of three descriptions
regarding the causal status of food B. The three options (prevention, modulation, configural)
were the same as in our previous studies (Lee & Lovibond, 2021; Lovibond & Lee, 2021).
Specifically, the prevention option was “It prevented allergic reactions in general”, the
modulation option was “It prevented allergic reactions caused by specific foods”, and the
configural option was “It is hard to know the exact role of individual foods such as this one. I
concentrated on remembering which combinations of foods caused an allergic reaction and
worked from there”. The order of the options was randomized for each participant.
Test phase. After training was completed, participants were told that they would now be
presented with more meals and that they should continue to make predictive ratings, but they
would no longer receive any feedback regarding the presence or absence of the allergic
outcome. The trial types presented in the Test phase consisted of a) all the trial types from the
Training phase except for the filler compound GH+; b) individual stimuli that had previously
only been presented in a compound (D, E); c) a novel stimulus I; and d) two compounds for the
summation test (CB and CD). Each trial type was tested twice, with the left-to-right
presentation of compound stimuli counterbalanced.
After the Test phase, participants were presented with all of the individual stimuli
except for the filler stimuli G and H. They were asked to rate “to what extent each food tended
to prevent or cause an allergic reaction” on a visual analogue scale. The scale ranged from
Inhibitory causal structures 11
“Strongly PREVENTED ALLERGIC REACTION” to “Strongly CAUSED ALLERGIC
REACTION”, with “No effect” labelled in the middle of the scale. The causal ratings were
recorded on a numerical scale from -100 to +100. Finally, they completed the causal structure
assessment for a second time, in order to test whether either of the reversal manipulations had
altered participants’ perception of the way in which the inhibitory stimulus B operated.
Demographic questions were included at the start of the experiment, and an additional question
at the end asked participants if they had written anything down during the experiment.
Table 1. Design of Experiment 1.
Group
Phase 1
Phase 2
Test
Causal ratings
Extinction
A+, AB-
C+, DE-
,
F-, GH+
B-
J+, JK+
C+, F-
A, AB, CB, CD,
C, DE, D, E,
F, I
A, B,
C, D, E,
F, I
No-Mod
A+, AB+
L-
C+, F-
Control
J+, JK+
L-
C+, F-
Note: Letters represent stimuli (foods); + represents the outcome (allergic reaction); –
represents the absence of the outcome. The procedure for Phase 1, Test and Causal ratings
were identical for the three groups. The critical Phase 2 manipulations are shown in bold type.
Arrows indicate administration of the causal structure assessment.
Inhibitory causal structures 12
Data Analysis
We analysed the data with planned contrasts using the PSY package (Bird, Hadzi-
Pavlovic & Isaac, 2000). The critical analyses concerned the test data. Within-participant
contrasts compared CB to CD in the outcome prediction ratings, and B to D in the causal
ratings. Group contrasts tested all three pairwise comparisons between the Extinction, No-Mod
and Control groups. Causal structure subgroup contrasts compared a) the modulation subgroup
with the prevention subgroup, and b) the average of these two groups with the configural
subgroup. We tested all contrasts as main effects (i.e., averaged over all other factors), as well
as all possible interactions between contrasts. For each contrast of interest, we report the p
value as well as the corresponding standardized 95% confidence interval (CI). The mean of the
CI represents standardised effect size (Bird, 2004).
Results
Exclusion Criteria
The primary exclusion criteria were those used in Lee and Lovibond (2021) and
Lovibond & Lee (2021). Specifically, participants were excluded if they reported having
written anything down during the experiment (n=26) or if they failed the acquisition criterion
which was an average rating > 75 for the stimuli that predicted the outcome (A+, C+, GH+)
and an average rating < 25 for the stimuli that predicted no outcome (AB-, DE-, F-) in the last
block of Phase 1 (n=26). In addition, we excluded participants who failed the instruction check
more than twice (n=5). After applying all three criteria, 152 participants remained (47 in the
Control group, 47 in the Extinction group, and 58 in the No-Mod group).
Causal Structure Assessment 1
Examination of the open-ended responses regarding causal structure at the first
assessment revealed no novel causal structures not already covered by the forced choice
Inhibitory causal structures 13
options. Therefore, in line with our previous research (e.g., Lee & Lovibond, 2021), we divided
participants into subgroups based on their response to the first 3AFC question. These
subgroups were used to analyse any effects of the initial causal structures that participants
reported having formed during acquisition of the A+/AB- discrimination. There were 34
participants in the configural subgroup, 104 in the modulation subgroup, and 14 in the
prevention subgroup. As shown in Table 2, there was no significant difference between the
groups in their causal structure choice at the first assessment (X2(df=2, N=152) = 3.50, p=.48).
Table 2. Causal structure classification by group at the first 3AFC assessment in Experiment 1
Group
Configural
Modulation
Prevention
Total
Control
11
34
2
47
Extinction
11
29
7
47
No-Mod
12
41
5
58
Total
34
104
14
152
Phase 1
Figure 1 shows the mean outcome prediction ratings across trials in Phase 1. Initial
analyses confirmed very similar acquisition patterns across groups and causal structure
subgroups, as would be expected given that all participants received the same training. There
was a small but significant difference in mean ratings between the Control and Extinction
groups (49.8 vs. 47.8 respectively; F(1,149) = 5.72, p=.02, 95% CI = 0.01, 0.14). However,
none of the group or causal structure subgroup contrasts interacted significantly with the
within-participant contrasts that assessed acquisition (largest F(1,149) = 2.61, p=.11, 95% CI =
-0.04, 0.36); hence Figure 1 is collapsed across both group and subgroup.
Inhibitory causal structures 14
Participants rapidly acquired differential responding to stimuli that predicted the
outcome versus those that predicted no outcome. This resulted in a significant main effect for
the contrast that compared predictive and non-predictive stimuli (F(1,151) = 4082.5, p<.001,
95% CI = 2.57, 2.73), as well as an interaction between this contrast and linear trend over trials
(F(1,151) = 2102.9, p<.001, 95% CI = 1.32, 1.43). Ratings on the first AB- trial were higher
than to the other stimuli, due to approximately half of the participants having previously
received an A+ trial. This pattern led to a significant main effect for the contrast comparing
AB- to the other two non-reinforced trial types (DE- and F-) averaged over trials (F(1,151) =
64.3, p<.001, 95% CI = -0.47, -0.29), as well as an interaction between this contrast and linear
trend over trials (F(1,151) = 72.3, p<.001, 95% CI = 0.28, 0.45). Nonetheless, by the end of
Phase 1 ratings were similarly low for the three non-reinforced trial types.
Figure 1. Mean outcome prediction ratings (+/– 1 SE) during Phase 1 in Experiment 1.
Phase 2
Figure 2 shows the mean outcome prediction ratings across trials in Phase 2, separated
by group. There were only small differences between the causal structure subgroups in this
phase so Figure 2 is collapsed across this factor. All groups maintained low ratings to F and
Inhibitory causal structures 15
high ratings to GH throughout the phase. They also learned about the novel filler cues as
expected. Statistical analysis focused on the critical trial types within the two experimental
groups. In the Extinction group, prediction ratings to the inhibitory stimulus B were low from
the first trial, despite the fact that these participants had never seen B outside of the AB
compound previously. Nonetheless, the analysis revealed a small but significant decreasing
linear trend across subsequent B- trials (F(1,44) = 5.14, p=0.03, 95% CI = -0.43, -0.03). This
contrast also interacted with the comparison between the Configural and Prevention subgroups,
reflecting a slightly higher mean rating for B on the first two trials in the Configural group
(F(1,44) = 5.16, p=0.03, 95% CI = -1.19, -0.07). In the No-Mod group, initial ratings to the A
and AB trial types were consistent with the end of Phase 1 training. However ratings to AB
rapidly increased as participants learned that this compound now signalled the outcome. This
pattern led to an interaction between the A vs AB comparison and linear trend over trials
(F(1,55) = 143.4, p<.001, 95% CI = -1.48, -1.06). This pattern did not further interact with
either of the causal structure subgroup contrasts, Fs < 1.
Figure 2. Mean outcome prediction ratings (+/– 1 SE) for each group during Phase 2 in
Experiment 1.
Inhibitory causal structures 16
Test
For analysis of the outcome prediction test data, each participant’s prediction ratings were
averaged over the two trials for each trial type. Ratings for the trained stimuli were very similar
to the end of acquisition. Figure 2 shows mean ratings for the critical summation compounds for
each group (see Supplemental Materials for full data). Looking first at the top panel, it can be
seen that ratings to CB were clearly lower than to the control compound CD in the Control group,
indicating successful transfer of the inhibitory properties of B to the novel excitor C. The data
pattern for the Extinction group was very similar. By contrast, in the No-Mod group, ratings to
CB were much higher and were close to CD. These differences were reflected in the statistical
analysis. There was an overall main effect for the CB-CD difference, averaged over groups
(F(1,143) = 25.30, p<.001, 95% CI = -1.40, -0.61). This contrast interacted with both the Control
vs No-Mod comparison (F(1,143) = 16.53, p<.001, 95% CI = -3.19, -1.10) and the Extinction vs
No-Mod comparison (F(1,143) = 36.80, p<.001, 95% CI = -3.34, -1.70). Importantly, however,
the contrast comparing CB with CD did not interact with the Control vs Extinction comparison
(F<1). Thus there was no evidence that the B- trials in Phase 2 in the Extinction group had any
impact on B’s ability to suppress responding to C in the summation test.
The bottom panel of Figure 3 shows the prediction test data further separated by self-
reported causal structure as recorded after Phase 1. The three causal structure subgroups each
followed a similar pattern to the sample as a whole. The ordering of ratings for the test compound
CB was consistent with our previous work (strongest transfer in the Prevention subgroup and
weakest transfer in the Configural subgroup; Lee & Lovibond, 2021; Lovibond & Lee, 2021).
However, none of the interactions between the CB-CD comparison and the subgroup
comparisons reached significance (largest F(1,143) = 2.68, p=0.10, 95% CI = -0.13, 1.40).
Inhibitory causal structures 17
Figure 3. Top panel shows mean outcome prediction ratings (+/– 1 SE) for the summation
(CB) and control (CD) compounds for each group in the Test phase of Experiment 1. Bottom
panel is separated by causal structure subgroup.
Causal rating test
Causal ratings for the common training stimuli were as expected on the basis of their
reinforcement history. The top panel of Figure 4 shows mean causal ratings for the critical test
stimuli B and D (see Supplemental Materials for full data). This test differed from the outcome
prediction test in that participants rated the causal strength of the individual stimuli on a cause-
prevent scale, rather than making outcome predictions. Overall, participants gave significantly
lower (more preventive) ratings for the inhibitory stimulus B than for the control stimulus D
(F(1,143=36.70, p<.001, 95% CI = -1.55, -0.79). However, as in the prediction test data, this
difference was not equivalent across groups. Interaction contrasts showed that the B-D
Inhibitory causal structures 18
difference was significantly smaller in the No-Mod group than in either the Control group
(F(1,143=51.17, p<.001, 95% CI = -2.96, -1.68) or the Extinction group (F(1,143=79.12,
p<.001, 95% CI = -2.77, -1.77). However, the B-D difference did not differ between the
Control and Extinction groups (F<1). We repeated these group comparisons for stimulus B
alone and obtained the same pattern of results. Finally, we examined the causal structure
subgroup factor (bottom panel of Figure 4). The pattern of group differences was consistent
across subgroups. Although the Prevention participants in the No-Mod group appeared to show
more negative ratings to B and D than the other participants, there were only 5 participants in
this subgroup. There were no significant interactions involving the causal structure factor in
either the B-D analysis (all Fs<1) or the analysis of B alone (largest F(1,143)=1.40, p=0.24,
95% CI = -0.26, 1.03).
Inhibitory causal structures 19
Figure 4. Top panel shows mean causal ratings (+/– 1 SE) for the critical test stimuli for each
group in Experiment 1. Bottom panel is separated by causal structure subgroup. Positive
ratings indicate causation, negative ratings indicate prevention, and zero indicates no effect.
Causal Structure Assessment 2
In the second causal structure assessment, 42 participants selected the configural
option, 82 selected the modulation option, and 26 selected the prevention option. However, this
pattern was not the same across groups (X2(df=2, N=152) = 38.5, p<.001). As shown in Table
3, a greater proportion of participants in the No-Mod group chose the Configural option at the
second assessment compared to the other two groups. This difference is presumably due to the
contradictory experiences that the No-Mod group received in Phases 1 and 2.
Table 3. Causal structure classification by group at the second 3AFC assessment in Experiment 1
Group
Configural
Modulation
Prevention
Total
Control
5
31
11
47
Extinction
7
26
14
47
No-Mod
32
25
1
58
Total
44
82
26
152
Discussion
The results from Experiment 1 were very clear. Non-reinforced presentations of the
inhibitor B in the Extinction group had no discernible influence on its inhibitory properties in
either the summation test or causal ratings. The test data for this group were very similar to
those from the Control group, which did not experience B- trials. By contrast to the Extinction
group, the No-Mod group showed a complete reversal of B’s inhibitory properties, on both test
Inhibitory causal structures 20
measures. This condition is important because it demonstrates that it is possible to reverse
inhibition, and that the tests we employed were sensitive enough to detect such a reversal.
In this experiment, the majority of participants reported inferring a configural or
modulatory causal structure for B, as opposed to the prevention structure that would
correspond to an inhibitory association that acted directly on the outcome representation
(Wagner & Rescorla, 1972). The small number of participants who endorsed a prevention
structure did not respond differently to the extinction procedure compared to the other
subgroups. In fact, there were essentially no differences between participants in their response
to the group manipulation as a function of their self-reported causal structure for B. This
pattern will be considered further in the General Discussion.
Even though the results of Experiment 1 were clear cut, there are nonetheless two
issues that impact on its interpretation. First, it is possible that interposing the causal structure
assessment between Phase 1 and Phase 2 influenced the way in which participants represented
the knowledge they had acquired in Phase 1, or the way in which they responded to the
extinction manipulation in Phase 2. Second, the No-Mod procedure might have exerted its
strong reversal effect by virtue of the fact that it directly contradicted the training
contingencies. Specifically, the AB compound was non-reinforced in Phase 1 and reinforced in
Phase 2. We addressed both of these issues in Experiment 2.
Experiment 2
In this experiment we included two design changes to extend and clarify the findings of
Experiment 1. First, to eliminate any potential impact of the initial causal structure assessment
on subsequent responding, we omitted this assessment in Experiment 2. Thus, there was no
longer any break between Phases 1 and 2. Second, we added a fourth group to test a subtler
version of the no-modulation procedure, as suggested by Baetu and Baker (2012). In this
Inhibitory causal structures 21
group, labelled No-Mod Novel, participants were given intermixed M+ and MB+ trials, where
M was a novel stimulus introduced in Phase 2. This design counteracts the modulatory role of
B but does not directly contradict the training trials with stimulus A.
Method
Participants
Two hundred and forty four undergraduate students (163 female, M age = 20.0, SD age
= 3.6) participated in Experiment 2 in exchange for course credit. The exclusion criteria were
the same as for Experiment 1 with the addition that participants could not have participated in
Experiment 1.
Apparatus and Stimuli
The apparatus and stimuli were the same as in Experiment 1, except that 13 rather than
12 food stimuli were randomly selected to serve as experimental stimuli A-M.
Procedure
The procedure for Experiment 2 followed that of Experiment 1, with two main
exceptions. First, an additional group was included, No-Mod Novel. This group was treated in
the same way as the No-Mod group, except that in Phase 2 the no-modulation training involved
a novel stimulus M. Thus, this group received M+ and MB+ trials in Phase 2. The second
change was that the first 3AFC causal structure assessment was omitted. Training proceeded
from Phase 1 to Phase 2 with no demarcation. A final minor change was to include stimulus B
alone in the Test phase. The full design table is shown in Table S2 in Supplemental Materials.
Data Analysis
We followed the same analysis strategy as in Experiment 1, except that we did not
analyse causal structure subgroups since we only had a post-experimental assessment and
Experiment 1 had demonstrated that the Phase 2 manipulations significantly altered
Inhibitory causal structures 22
participants’ responses from the first to the second assessment. In this experiment the planned
contrasts for the group factor compared a) the Extinction with the Control group, 2) the No-
Mod with the No-Mod Novel group, and c) the average of the Extinction and Control groups
with the average of the No-Mod and No-Mod Novel groups.
Results
Exclusion Criteria
Eighteen participants failed the write check, 24 failed the instruction check, and 37
failed the acquisition criterion. After all exclusions had been applied, 175 participants remained
(41 in the Extinction group, 48 in the No-Mod group, 45 in the No-Mod Novel group, and 41
in the Control group).
Phase 1
The training data were very similar to Experiment 1. There were no substantial
differences between groups, so Figure 5 is collapsed across this factor. Participants again
learned to discriminate cue combinations that predicted the outcome from those that did not.
Acquisition was demonstrated by a significant difference between the reinforced and non-
reinforced cues, averaged over trials (F(1,163)=2846.0, p<.001, 95% CI = 2.77, 2.98) as well
as an interaction between this contrast and linear trend over trials (F(1,163)=1663.1, p<.001,
95% CI = 1.40, 1.55).
Inhibitory causal structures 23
Figure 5. Mean outcome prediction ratings (+/– 1 SE) during Phase 1 in Experiment 2.
Phase 2
The Phase 2 predictive ratings are shown for each group in Figure 6. The data patterns
for the Extinction, No-Mod and Control groups were very similar to those seen in Experiment
1. The Extinction group again gave low ratings to the inhibitory stimulus B on the first trial,
and in this experiment showed no significant reduction over subsequent trials (linear trend:
F<1). In the No-Mod Novel group, initial ratings to the novel cue M were near the middle of
the scale but rapidly increased to asymptote. Ratings to the MB compound were somewhat
lower, but also increased rapidly to asymptote. By contrast, in the No-Mod group, ratings to the
AB compound increased more slowly, presumably due to the conflict with the Phase 1
contingency. These two groups were analysed together, to test the above pattern. There was a
significant interaction between the No-Mod vs No-Mod Novel group comparison and the
element (A or M) vs compound (AB or MB) comparison, (F(1,87) = 37.00, p<.001, 95% CI =
1.20, 2.37), confirming that averaged over trials, the A vs AB difference in group No-Mod was
smaller than the M vs MB difference in group No-Mod Novel. This contrast further interacted
with linear trend, indicating that the group difference in responding to the compound relative to
Inhibitory causal structures 24
the element diminished across trials as participants learned the new contingencies, (F(1,87) =
18.01, p<.001, 95% CI = -1.60, -0.61).
Figure 6. Mean outcome prediction ratings (+/– 1 SE) for each group during Phase 2 in
Experiment 2.
Test
Figure 7 shows the mean predictive ratings to the critical test stimuli for each group
(see Supplemental Materials for full data). In addition to the CB and CD summation test
compounds, data for A and AB trials are included to allow comparison between the two no-
modulation procedures on the original training discrimination. B trials are also included as a
further test of the impact of the group manipulations. The results for the three groups in
common with Experiment 1 were very similar to that experiment, except that the difference
between the summation test compound CB and the control compound CD was somewhat
smaller in this experiment. However, the pattern across groups replicated that seen in
Experiment 1 and extended it to the No-Mod Novel condition which showed a very similar
pattern to the No-Mod group. Averaged over groups, ratings to CB were lower than to CD
(F(1, 163)= 21.67, p<.001, 95% CI = -0.64, -0.26), confirming the overall inhibitory properties
of B. This contrast interacted with the comparison between the two no-modulation groups and
Inhibitory causal structures 25
the other two groups (F(1, 163)= 22.81, p<.001, 95% CI = -1.29, -0.54), demonstrating that the
no-modulation manipulations had reduced inhibitory transfer. The interaction between the CB
vs CD comparison and the No-Mod vs No-Mod Novel comparison was non-significant (F<1),
reflecting the very similar pattern seen in these two groups. Interestingly, the interaction
between the CB vs CD comparison and the Extinction vs Control comparison just reached
significance, in the direction of slightly greater inhibitory transfer in the Extinction group (F(1,
163)= 5.04, p=.026, 95% CI = 0.07, 1.11).
Figure 7. Mean outcome prediction ratings (+/– 1 SE) for critical test stimuli in each group in
the Test phase of Experiment 2.
Although the two no-modulation groups showed a similar loss of transfer of B’s
inhibitory properties to the test excitor C, and both groups experienced B paired with the
outcome in Phase 2, the No-Mod Novel group maintained low ratings to the AB compound
whereas the No-Mod group showed much higher ratings. This pattern led to a significant
interaction between the No-Mod vs No-Mod Novel comparison and the A vs AB comparison
(F(1, 163)= 31.01, p<.001, 95% CI = -3.35, -1.60). We also compared the groups on their
predictive ratings to stimulus B alone. Here, the principal finding was higher ratings to B in the
two no-modulation groups compared to the Extinction and Control groups, consistent with the
Inhibitory causal structures 26
summation test results (F(1, 163)= 8.92, p=.003, 95% CI = -1.07, -0.22). There was no
difference between the two no-modulation groups (F<1) or between the Extinction and Control
groups (F(1, 163)= 2.62, p=.11, 95% CI = -0.11, 1.05) in their ratings to B.
Causal Rating Test
Figure 8 shows mean causal ratings for the critical test stimuli B and D (see Supplemental
Materials for full data). Averaged over groups, participants gave significantly lower (more
preventive) ratings for the inhibitory stimulus B than for the control stimulus D
(F(1,163)=33.64, p<.001, 95% CI = -1.21, -0.60). However, as in Experiment 1, the magnitude
of this difference differed across groups. Specifically, the B-D difference was significantly
smaller in the average of the two no-modulation groups than in the average of the Control and
Extinction groups (F(1,163)=10.59, p=.001, 95% CI = -1.63, -0.40). However, the B-D
difference did not differ between the No-Mod and No-Mod Novel groups, or between the
Control and Extinction groups (Fs<1). We repeated these group comparisons for stimulus B
alone and obtained the same pattern of results.
Figure 8. Mean causal ratings (+/– 1 SE of the mean) for test stimuli in each group in
Experiment 2.
Inhibitory causal structures 27
Discussion
This experiment replicated the main finding from Experiment 1, namely that there was
no evidence of extinction of inhibition, in either the summation test or causal ratings. If
anything, we saw a small effect in the opposite direction for the CB-CD comparison in the
summation test. This outcome indicates that the absence of extinction in Experiment 1 was not
due to the inclusion of a causal structure assessment prior to extinction. Nonetheless, the
overall strength of inhibitory learning observed in both the summation test and causal ratings
appeared to be slightly lower than in Experiment 1. This difference suggests that asking
participants in Experiment 1 to reflect on causal structure, or exposing them to the options in
the 3AFC question, enhanced their inhibitory performance and transfer.
Experiment 2 also showed that a no-modulation training procedure with a novel excitor
(No-Mod Novel condition) was as effective in reversing inhibitory learning as the same
procedure with the training excitor (No-Mod condition). This finding suggests that the critical
feature needed for reversal of inhibitory learning is training that the inhibitor no longer has
modulatory properties. Although no-modulation training with the training excitor might more
successfully interfere with the original inhibitory learning, no-modulation training with a novel
excitor might encourage greater transfer in a summation test with yet another excitor. Perhaps
these two factors cancelled out in the present design. In any case, it is clear that both
procedures were effective in reversing inhibition.
General Discussion
Both experiments showed that non-reinforced presentations of a conditioned inhibitor,
established by feature negative training with a conventional unidirectional outcome, do not
lead to any measurable reduction in its inhibitory properties. This outcome is consistent with
the results reported by Yarlas et al. (1995) and the unidirectional outcome conditions in
Inhibitory causal structures 28
Melchers et al. (2006) and Lotz and Lachnit (2009). In the present experiments, the lack of
extinction of inhibition was observed in both a summation test with a novel excitor and in
causal ratings, relative to a control group that did not receive presentations of the inhibitor
alone. Both experiments had a substantial sample size, and the success of the no-modulation
conditions in reversing inhibition demonstrated that the procedure was able to detect reversal if
it occurred. To further assess the strength of the null findings for the Extinction-Control
comparisons, we pooled the data from these two conditions from each experiment and carried
out a Bayesian analysis. Bayes Factors showed that the evidence for the null hypothesis – no
effect of the extinction manipulation relative to the control – was between 3 and 4 times higher
than for the alternative hypothesis (BF01 = 3.81 for the CB-CD difference in predictive ratings,
and BF01 = 3.77 for the B-D difference in causal ratings).
The results of the present experiments conflict with those of the Unidirectional group in
Baetu and Baker (2010, Experiment 1), which showed evidence for extinction of inhibition.
Comparison of the procedures suggests a critical difference that could account for the
diverging outcomes. Specifically, the outcome used by Baetu and Baker (2010), change in
hormone level from an unspecified baseline level, could have been interpreted by participants
as bidirectional even in the Unidirectional group. The only procedural difference between the
Unidirectional and Bidirectional groups was that the Unidirectional group did not experience
any trials in which the hormone level decreased. Thus, the Unidirectional participants may still
have inferred that the inhibitory cue would have produced such a decrease if presented alone.
Direct support for this interpretation comes from their Test phase data, which were collected
using a bidirectional magnitude scale that ranged from “decrease” through “not change” to
“increase” in both groups. The Unidirectional group gave strong negative (decrease) ratings for
both of the unextinguished inhibitory cues – in fact, stronger than for the Bidirectional group.
Therefore it seems likely that the extinction of inhibition seen in the Unidirectional group in
Inhibitory causal structures 29
Baetu and Baker (2010) occurred for the same reason as for the Bidirectional group, and in
accordance with the account proposed by Melchers et al. (2006) – that is, presentations of the
inhibitor with no change in outcome level contradicted the expectation that it would produce a
decrease. By contrast, in the present research, we used a unidirectional outcome that could only
vary between zero (no allergic reaction) and a positive value (allergic reaction).
A similar argument may apply to the results of Kutlu and Schmajuk (2012), who
reported a loss of inhibitory strength after extinction of an inhibitory feature. These researchers
used an outcome (line height) that was putatively unidirectional. However, they presented no
causal scenario or instructions to participants to clarify whether cues led to a particular absolute
line height or a change in line height from some starting level. Note that the outcome used by
Melchers et al. (2009), Lotz & Lachnit (2009) and Baetu & Baker (2010), hormone level, is
also intrinsically unidirectional – it is not possible to have negative values of a hormone. The
outcome only became bidirectional in the Bidirectional conditions of those experiments by
virtue of instructions stating that cues could cause a relative increase or a decrease from a non-
zero starting value. In the Kutlu and Schmajuk (2012) design, the starting point for the line
height was ambiguous, potentially allowing some participants to treat this outcome as
bidirectional.
It therefore appears that when a conventional unidirectional outcome is employed, as in
the present experiments and the majority of animal research, presentations of an inhibitor do
not decrease its inhibitory properties. What does this mean for the way in which inhibitory
learning is encoded? As noted earlier, it is problematic for the Rescorla-Wagner (1972) model,
which incorrectly predicts a loss of inhibition. Rescorla (1979) pointed out that one way to
accommodate this finding is in terms of the theory put forward by Konorski (1948). According
to this theory, an inhibitor acts to increase the threshold required for an excitor to activate the
US (outcome) representation. As such, non-reinforced presentations of an inhibitor would not
Inhibitory causal structures 30
lead to extinction. However, this theory shares a limitation with the Rescorla-Wagner (1972)
model, namely that it predicts strong transfer in a summation test, an outcome that is rare in
animal research and even rarer in human research (e.g., Karazinov & Boakes, 2004; Lee &
Livesey, 2012; Williams, 1995). Finally, unlike the Rescorla-Wagner (1972) model,
Konorski’s (1948) theory fails to capture the idea of expectancy violation that appears to be
critical for inhibitory learning. It is also difficult to apply directly to causal learning in humans
in terms of aligning with a distinct causal structure.
An alternative approach is suggested by the literature on negative occasion-setting (e.g.,
Rescorla, 1987; Fraser & Holland, 2019). This type of learning, which in animals is promoted
by the use of a serial arrangement during feature negative training (A+ / B→A–), appears to
endow the inhibitory stimulus with the ability to modulate responding to the target excitor.
Importantly, this type of learning is typically somewhat specific to the target excitor and hence
supports only partial transfer in a summation test. Furthermore, the occasion-setting property
of a stimulus appears to be relatively unaffected by manipulations involving its direct
association with the outcome. Both logically and empirically, presentation of an occasion-setter
with or without the outcome does not affect its modulatory properties with respect to another
stimulus (Fraser & Holland, 2019). Several aspects of the current results are consistent with the
idea that inhibitory learning followed a modulatory associative structure. First, transfer to a
novel excitor in the summation test was only modest. Second, non-reinforced presentations of
the inhibitory stimulus did not affect its ability to modulate either the training excitor or a
novel excitor. Third, the two manipulations that did contradict the modulatory properties of the
inhibitor, No-Mod and No-Mod Novel, were highly effective in reversing inhibition. Finally,
many of the participants reported having learned a modulatory causal structure in the self-
report assessment in Experiment 1.
Inhibitory causal structures 31
There are also aspects of the present results that do not align so well with the idea that
participants were learning inhibition in a modulatory way. First, we used a simultaneous
feature negative design, whereas in animals, modulatory learning (occasion-setting) is most
commonly seen with serial designs. However, in our previous research using a similar
procedure to the current experiments, we have found that a serial arrangement has little impact
on self-reported causal structure (Lovibond & Lee, 2021). Perhaps humans are more
predisposed to interpret inhibitory associations in a modulatory way, even when the stimuli are
presented simultaneously. A second issue is that we only tested inhibitory learning with a
summation test, not with a retardation test (Pavlov, 1927; Rescorla, 1969). An important topic
for future research would be to test the prediction that follows from the occasion-setting
literature, namely that an inhibitor with modulatory properties may not in fact show retarded
excitatory acquisition, due to the apparent independence of modulatory and direct associations
(Fraser & Holland, 2019).
A third complication with a modulatory account of our results is that not all participants
reported a modulatory causal structure when we assessed this in Experiment 1 – a substantial
number reported a configural or prevention structure. When we first investigated individual
differences in inhibitory learning, we hypothesised that the different self-reported causal
structures represented qualitatively different structures (Lee & Lovibond, 2021). Our
subsequent work has led us to question this assumption, for several reasons. First, some
participants might choose the configural option because it is the most conservative, rather than
because it represents a discrete causal structure. In the present experiments, these participants
gave substantial negative ratings to B in the causal rating test, suggesting they were in fact able
to infer an inhibitory role for this element when it was presented outside the compound.
Second, several manipulations we have tested, including serial vs. simultaneous training, and
prior modulatory training of the test excitor, turned out to have a similar impact across all
Inhibitory causal structures 32
causal structure subgroups (Lovibond & Lee, 2021). These results suggest an alternative
possibility, that the differences in transfer we have consistently observed in a summation test
(Lee & Lovibond, 2021; Lovibond & Lee, 2021) instead reflect participants’ willingness to
generalise the properties of the inhibitory stimulus to a novel excitor. Furthermore, this
variation may be quantitative rather than qualitative, with preventers showing the greatest
transfer in a summation test, configural participants the least, and modulators in between.
Consistent with this perspective, we have recently argued that a more productive way to
conceptualise the summation test is in terms of principles of generalisation, rather than the
arithmetic summation of associative strengths (Chow, Lee & Lovibond, in press; see also
Bonardi, Robinson & Jennings, 2017). Thus, participants who chose the prevention option may
have had a greater tendency to generalize the inhibitory properties of the feature to a new
excitor, presumably due to differences in their pre-experimental their reinforcement history. If
participants vary primarily in their willingness to generalise, then it is less surprising that they
respond equally to manipulations that are not related to this dimension, including the present
manipulations of extinction and no-modulation.
Accordingly, it remains at least possible that there is a single causal structure that
underlies inhibitory learning in humans, and that this structure is modulatory in nature. Such a
view can explain the three key findings of the present experiments: no extinction of inhibition,
successful reversal of inhibition by procedures that contradict modulation, and no substantial
differences in these effects as a function of self-reported causal structure. It is possible that
inhibitory learning in animals is also fundamentally modulatory in nature, even with
simultaneous feature negative training. The apparent reversal of such inhibition by direct
reinforcement (e.g., Holland, 1989) could be explained in terms of the superimposition of
excitatory learning on the original modulatory learning. By this account, inhibition would be
the symmetrical opposite not of excitation but of facilitation (positive occasion-setting), a view
Inhibitory causal structures 33
put forward by Rescorla (1987) and extended to causal reasoning by Baetu and Baker (2012).
Furthermore, it would not require abandonment of the central prediction error component of
the Rescorla-Wagner (1972) model, only a modification of the way in which learning is
represented: direct association in the case of excitatory learning, and modulation of an
excitatory association in the case of inhibitory learning and facilitation. Whatever the merits of
this particular position, it is clear that the insights provided by Robert Rescorla over his 50-
year career will continue to play a central role in advancing our understanding of associative
learning.
Inhibitory causal structures 34
References
Baetu, I. & Baker, A.G. (2010). Extinction and blocking of conditioned inhibition in human
causal learning. Learning & Behavior, 38, 394-407.
Baetu, I. & Baker, A.G. (2012). Are preventive and generative causal reasoning symmetrical?
Extinction and competition. Quarterly Journal of Experimental Psychology, 65, 1675-
1698.
Bird, K. D., Hadzi-Pavlovic, D. & Isaac, A. P. (2000). PSY [Computer software]. Available
from http://www.psy.unsw.edu.au/research/psy.htm
Bird, K.D. (2004). Analysis of variance via confidence intervals. London: Sage Publications.
Bonardi. C., Robinson, J., & Jennings. D. (2017). Can existing associative principles explain
occasion setting? Some old ideas and some new data. Behavioural Processes, 137, 5-18.
Chow, J.Y-L., Lee, J.C. & Lovibond, P.F. (in press). Inhibitory summation as a form of
generalisation. Journal of Experimental Psychology: Animal Learning and Cognition.
de Leeuw, J. R. (2015). jsPsych: A JavaScript library for creating behavioural experiments in a
web browser. Behaviour Research Methods, 47, 1-12.
Fraser, K. M., & Holland, P. C. (2019). Occasion Setting. Behavioral Neuroscience, 133, 145-
175.
Gershman S.J. (2015). A unifying probabilistic view of associative learning. PLoS
Computional Biology, 11, e1004567.
Holland, P. C. (1989). Transfer of negative occasion setting and conditioned inhibition across
conditioned and unconditioned stimuli. Journal of Experimental Psychology: Animal
Behavior Processes, 15, 311–328.
Inhibitory causal structures 35
Kamin, L. J. (1969). Predictability, surprise, attention and conditioning. In B.A. Campbell &
R.M. Church (Eds.), Punishment and aversive behavior (pp. 279-296). New York:
Appleton-Century-Crofts.
Karazinov, D.M. & Boakes, R.A. (2004). Learning about cues that prevent an outcome:
Conditioned inhibition and differential inhibition in human predictive learning. Quarterly
Journal of Experimental Psychology, 57, 153-178.
Konorski, J. (1948). Conditioned reflexes and neuron organization. Cambridge: Cambridge
University Press.
Kutlu, M.G. & Schmajuk, N.A. (2012). Deactivation and reactivation of the inhibitory power
of a conditioned inhibitor: Testing the predictions of an attentional-associative model.
Learning & Behavior, 40, 83-97.
Lange, K., Kühn, S., & Filevich, E. (2015). “Just Another Tool for Online Studies” (JATOS):
An easy solution for setup and management of web servers supporting online studies.
PLoS ONE, 10(6): e0130834.
Lee, J.C. & Livesey, E. J. (2012). Second-order conditioning and conditioned inhibition:
Influences of speed versus accuracy on human causal learning. PLoS ONE, 7(11),
e49899.
Lee, J.C. & Lovibond, P.F. (2021). Individual differences in causal structures inferred during
feature negative learning. Quarterly Journal of Experimental Psychology, 74(1), 150-
165.
Lotz, A & Lachnit, H. (2009). Extinction of conditioned inhibition: Effects of different
outcome continua. Learning & Behavior, 37, 85-94.
Rescorla, R.A. & LoLordo, V.M. (1965). Inhibition of avoidance behavior. Journal of
Comparative & Physiological Psychology, 59, 406–412.
Inhibitory causal structures 36
Lovibond, P.F. & Lee, J.C. (2021). Inhibitory causal structures in serial and simultaneous
feature negative learning. Quarterly Journal of Experimental Psychology, 74, 2165-2181.
Lysle, D.T. & Fowler, H. (1985). Inhibition as a "slave" process: Deactivation of conditioned
inhibition through extinction of conditioned excitation. Journal of Experimental
Psychology: Animal Behavior Processes, 11, 71-94.
Melchers, K.G., Wolff, S. & Lachnit, H. (2006). Extinction of conditioned inhibition through
nonreinforced presentation of the inhibitor. Psychonomic Bulletin & Review, 13, 662-
667.
Pavlov, I. P. (1927). Conditioned Reflexes. London, UK: Oxford University Press.
Pearce, J. M. (1987). A model for stimulus generalization in Pavlovian conditioning.
Psychological Review, 94, 61-73.
Rescorla, R.A. (1967). Pavlovian conditioning and its proper control procedures. Psychological
Review, 74, 71–80.
Rescorla, R.A. (1969). Pavlovian conditioned inhibition. Psychological Bulletin, 72, 77–81.
Rescorla, R.A. (1979). Conditioned inhibition and extinction. In A. Dickinson and R.A.
Boakes (Eds.), Mechanisms of Learning and Motivation: A memorial volume to Jerzy
Konorski (pp. 83-110). Hillsdale, N.J.: Erlbaum.
Rescorla, R.A. (1987). Facilitation and inhibition. Journal of Experimental Psychology:
Animal Behavior Processes, 13, 250-259.
Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the
effectiveness of reinforcement and nonreinforcement. Classical conditioning II: Current
research and theory, 2, 64-99.
Shanks, D.R. & Dickinson, A.R. (1987). Associative accounts of human causality judgment.
The Psychology of Learning and Motivation, 21, 229-261.
Inhibitory causal structures 37
Van Hamme, L.J & Wasserman, E.A. (1994). Cue competition in causality judgments: The
role of nonpresentation of compound stimulus elements. Learning and Motivation, 25,
127-151.
Wagner, A.R. (1969). Stimulus validity and stimulus selection in associative learning. In N.J.
Mackintosh & W.K. Honig (Eds.), Fundamental issues in associative learning (pp. 90-
122). Halifax: Dalhousie University Press.
Wagner, A.R. & Rescorla, R.A. (1972). Inhibition in Pavlovian conditioning: Application of a
theory. In R.A. Boakes and H.S. Halliday (Eds.), Inhibition and Learning. London:
Academic Press.
Williams, D.A. (1995). Forms of inhibition in animal and human learning. Journal of
Experimental Psychology: Animal Behavior Processes, 21, 129-142.
Williams, D. A. & Overmier, J. B. (1988). Some types of conditioned inhibitors carry collateral
excitatory associations. Learning & Motivation, 19, 345-368.
Williams, D.A., Sagness, K.E. & McPhee, J.E. (1994). Configural and elemental strategies in
predictive learning. Journal of Experimental Psychology: Learning, Memory and
Cognition, 20, 694-709.
Yarlas, A. S., Cheng, P. W. & Holyoak, K. J. (1995). Alternative approaches to causal
induction: The probabilistic contrast versus the Rescorla–Wagner model. In J. F. Lehman
& J. D. Moore (Eds.), Proceedings of the Seventeenth Annual Conference of the
Cognitive Science Society (pp. 431-436). Hillsdale, NJ: Erlbaum.
Zimmer-Hart, C. L. & Rescorla, R. A. (1974). Extinction of Pavlovian conditioned inhibition.
Journal of Comparative & Physiological Psychology, 86, 837-845.
Inhibitory causal structures 38
Supplemental Materials
Experiment 1
Table S1. Causal structure classification at the two time points in Experiment 1
3AFC2
3AFC1
Configural
Modulation
Prevention
Total
Configural
14
16
4
34
Modulates
28
63
13
104
Prevents
2
3
9
14
Total
44
82
26
152
Note: 3AFC1 refers to the first forced choice assessment of causal structure (after Phase 1) and
3AFC2 refers to the second assessment (at the end of the experiment). Bold font indicates
participants whose choice was stable; that is, they selected the same option at each time point.
Inhibitory causal structures 39
Figure S1. Mean outcome prediction ratings (+/– 1 SE) for all test stimuli in the Test phase of
Experiment 1, separated by group and causal structure subgroup.
Inhibitory causal structures 40
Figure S2. Mean causal ratings (+/– 1 SE) for all test stimuli in Experiment 1, separated by
group and causal structure subgroup.
Inhibitory causal structures 41
Table S2. Design of Experiment 2
Group
Phase 1
Phase 2
Test
Causal ratings
Extinction
A+, AB-
C+, DE-,
F-, GH+
B-
J+, JK+
C+, F-
A, B, AB,
CB, CD, DE,
C, D, E, F, I
A, B,
C, D, E,
F, I
No-Mod
A+, AB+
L-
C+, F-
No-Mod
Novel
M+, MB+
L-
C+, F-
Control
J+, JK+
L-
C+, F-
Notation as per Table 1 in the main manuscript.
Inhibitory causal structures 42
Figure S3. Mean outcome prediction ratings (+/– 1 SE) for all test stimuli in the Test phase of
Experiment 2, separated by group.
Inhibitory causal structures 43
Figure S4. Mean causal ratings (+/– 1 SE) for all test stimuli in Experiment 2, separated by
group.
A preview of this full-text is provided by American Psychological Association.
Content available from Journal of Experimental Psychology: Animal Learning and Cognition
This content is subject to copyright. Terms and conditions apply.
