Julie Y. L. Chow
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Julie Y. L. Chow has a total of 22 co-authors. Here are the top co-authors by number of shared publications:
- Evan James Livesey
School of Psychology, The University of Sydney, Sydney, Australia
4 shared publications Ben Colagiuri
School of Psychology, The University of Sydney, Sydney, Australia
3 shared publications- Jessica C. Lee
School of Psychology, University of Sydney
3 shared publications - Kelly Garner
School of Psychology, University of New South Wales
3 shared publications - Mike Le Pelley
School of Psychology, University of New South Wales
3 shared publications - Peter F. Lovibond
School of Psychology, University of New South Wales
3 shared publications
Latest Publications
- Open Access
- 26 June 2026
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 22 (6) , e1013220
Predicting continuous outcomes: Some new tests of associative approaches to contingency learning
Associative learning models have traditionally simplified contingency learning by relying on binary classification of cues and outcomes, such as administering a medical treatment (or not) and observing whether the patient recovered (or not). While successful in capturing fundamental learning phenomena across human and animal studies, these models are not capable of representing variability in human experiences that are common in many real-world contexts. Indeed, where variation in outcome magnitude exists (e.g., severity of illness in a medical scenario), this class of models, at best, approximate the outcome mean with no ability to represent the underlying distribution of values. In this paper, we introduce one approach to incorporating a distributed architecture into a prediction error learning model that tracks the contingency between cues and dimensional outcomes. Our Distributed Model allows associative links to form between the cue and outcome nodes that provide distributed representation depending on the magnitude of the outcome, thus enabling learning that extends beyond approximating the mean. Comparing the Distributed Model against a Simple Delta Model across four contingency learning experiments, we found that the Distributed Model provides significantly better fit to empirical data in virtually all participants. These findings suggest human learners rely on a means of encoding outcomes that preserves the continuous nature of experienced events, advancing our understanding of causal inference in complex environments.
- 26 June 2026
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 22 (6) , e1013220
Open AccessPredicting continuous outcomes: Some new tests of associative approaches to contingency learning
Associative learning models have traditionally simplified contingency learning by relying on binary classification of cues and outcomes, such as administering a medical treatment (or not) and observing whether the patient recovered (or not). While successful in capturing fundamental learning phenomena across human and animal studies, these models are not capable of representing variability in human experiences that are common in many real-world contexts. Indeed, where variation in outcome magnitude exists (e.g., severity of illness in a medical scenario), this class of models, at best, approximate the outcome mean with no ability to represent the underlying distribution of values. In this paper, we introduce one approach to incorporating a distributed architecture into a prediction error learning model that tracks the contingency between cues and dimensional outcomes. Our Distributed Model allows associative links to form between the cue and outcome nodes that provide distributed representation depending on the magnitude of the outcome, thus enabling learning that extends beyond approximating the mean. Comparing the Distributed Model against a Simple Delta Model across four contingency learning experiments, we found that the Distributed Model provides significantly better fit to empirical data in virtually all participants. These findings suggest human learners rely on a means of encoding outcomes that preserves the continuous nature of experienced events, advancing our understanding of causal inference in complex environments.
- Open Access
- 21 December 2025
- Published by Center for Open Science in OSF Preprints
Delaying reward feedback does not increase the influence of information on attentional priority in visual search
Demonstrations of information-seeking behaviour suggest that attention often acts in an exploitative way, prioritising stimuli that provide diagnostic information about upcoming events over stimuli associated with uncertainty. However, recent evidence from studies of attentional capture in visual search show an opposite pattern: automatic prioritisation of items associated with reward uncertainty over diagnostic stimuli. We hypothesise that this uncertainty-modulated attentional capture (UMAC) effect reflects ‘attention for learning’: that is, exploration of potential sources of new information. Here we investigated whether UMAC arises because immediate provision of reward feedback in prior studies rendered advance information redundant, attenuating exploitation of diagnostic items and promoting exploration. Accordingly, increasing the duration of anticipated uncertainty (and hence the value of advance information that allows us to escape uncertainty earlier) should promote prioritisation of diagnostic cues and lead to patterns of attentional exploitation. In two eye-tracking experiments, we compared attentional capture by a cue providing diagnostic reward information and a cue signalling uncertain reward, while manipulating the delay between response and feedback (i.e., the duration of anticipated uncertainty that advance information could forestall). We found a UMAC effect in all conditions: regardless of response–feedback delay, uncertain stimuli were more likely to capture attention than diagnostic stimuli. These results suggest that prioritisation of uncertainty is a robust pattern of behaviour in this task. Synthesising current and previous findings, we suggest that different modes of attentional information-seeking may reflect qualitative task differences, with exploration operating at an implicit, automatic level, and exploitation resulting from top-down, volitional processes.
- 21 December 2025
- Published by Center for Open Science in OSF Preprints
Open AccessDelaying reward feedback does not increase the influence of information on attentional priority in visual search
Demonstrations of information-seeking behaviour suggest that attention often acts in an exploitative way, prioritising stimuli that provide diagnostic information about upcoming events over stimuli associated with uncertainty. However, recent evidence from studies of attentional capture in visual search show an opposite pattern: automatic prioritisation of items associated with reward uncertainty over diagnostic stimuli. We hypothesise that this uncertainty-modulated attentional capture (UMAC) effect reflects ‘attention for learning’: that is, exploration of potential sources of new information. Here we investigated whether UMAC arises because immediate provision of reward feedback in prior studies rendered advance information redundant, attenuating exploitation of diagnostic items and promoting exploration. Accordingly, increasing the duration of anticipated uncertainty (and hence the value of advance information that allows us to escape uncertainty earlier) should promote prioritisation of diagnostic cues and lead to patterns of attentional exploitation. In two eye-tracking experiments, we compared attentional capture by a cue providing diagnostic reward information and a cue signalling uncertain reward, while manipulating the delay between response and feedback (i.e., the duration of anticipated uncertainty that advance information could forestall). We found a UMAC effect in all conditions: regardless of response–feedback delay, uncertain stimuli were more likely to capture attention than diagnostic stimuli. These results suggest that prioritisation of uncertainty is a robust pattern of behaviour in this task. Synthesising current and previous findings, we suggest that different modes of attentional information-seeking may reflect qualitative task differences, with exploration operating at an implicit, automatic level, and exploitation resulting from top-down, volitional processes.
- Open Access
Predicting continuous outcomes: Some new tests of associative approaches to contingency learning
Associative learning models have traditionally simplified contingency learning by relying on binary classification of cues and outcomes, such as administering a medical treatment (or not) and observing whether the patient recovered (or not). While successful in capturing fundamental learning phenomena across human and animal studies, these models are not capable of representing variability in human experiences that are common in many real-world contexts. Indeed, where variation in outcome magnitude exists (e.g., severity of illness in a medical scenario), this class of models, at best, approximate the outcome mean with no ability to represent the underlying distribution of values. In this paper, we introduce one approach to incorporating a distributed architecture into a prediction error learning model that tracks the contingency between cues and dimensional outcomes. Our Distributed Model allows associative links to form between the cue and outcome nodes that provide distributed representation depending on the magnitude of the outcome, thus enabling learning that extends beyond approximating the mean. Comparing the Distributed Model against a Simple Delta Model across four contingency learning experiments, we found that the Distributed Model provides significantly better fit to empirical data in virtually all participants. These findings suggest human learners rely on a means of encoding outcomes that preserves the continuous nature of experienced events, advancing our understanding of causal inference in complex environments. Author Summary: When we learn about cause and effect in everyday life—such as whether a medicine helps recovery from illness—we experience outcomes that vary in degree rather than simply happening or not happening. Traditional models of how humans and animals learn have largely focused on these all-or-nothing scenarios, essentially tracking the average value when outcomes are dimensional. We developed a model that extends on simple error-correction models to represent how people learn about relationships between cues and outcomes that can take on a range of values. Instead of just tracking the average, our Distributed Model captures the full spectrum of possible outcomes and their frequencies. We tested this model against a conventional single point-estimate approach across four experiments and found that our Distributed Model better matched how people make predictions in nearly every case. Our findings suggest that a relatively simple adjustment to conventional prediction-error learning algorithms that allows representation of outcome magnitudes provide a powerful way to capture the information that we preserve when we learn about variable outcomes. This has important implications for understanding how people make predictions and decisions in real-world situations where outcomes naturally vary, from medical treatments to environmental changes.
Open AccessPredicting continuous outcomes: Some new tests of associative approaches to contingency learning
Associative learning models have traditionally simplified contingency learning by relying on binary classification of cues and outcomes, such as administering a medical treatment (or not) and observing whether the patient recovered (or not). While successful in capturing fundamental learning phenomena across human and animal studies, these models are not capable of representing variability in human experiences that are common in many real-world contexts. Indeed, where variation in outcome magnitude exists (e.g., severity of illness in a medical scenario), this class of models, at best, approximate the outcome mean with no ability to represent the underlying distribution of values. In this paper, we introduce one approach to incorporating a distributed architecture into a prediction error learning model that tracks the contingency between cues and dimensional outcomes. Our Distributed Model allows associative links to form between the cue and outcome nodes that provide distributed representation depending on the magnitude of the outcome, thus enabling learning that extends beyond approximating the mean. Comparing the Distributed Model against a Simple Delta Model across four contingency learning experiments, we found that the Distributed Model provides significantly better fit to empirical data in virtually all participants. These findings suggest human learners rely on a means of encoding outcomes that preserves the continuous nature of experienced events, advancing our understanding of causal inference in complex environments. Author Summary: When we learn about cause and effect in everyday life—such as whether a medicine helps recovery from illness—we experience outcomes that vary in degree rather than simply happening or not happening. Traditional models of how humans and animals learn have largely focused on these all-or-nothing scenarios, essentially tracking the average value when outcomes are dimensional. We developed a model that extends on simple error-correction models to represent how people learn about relationships between cues and outcomes that can take on a range of values. Instead of just tracking the average, our Distributed Model captures the full spectrum of possible outcomes and their frequencies. We tested this model against a conventional single point-estimate approach across four experiments and found that our Distributed Model better matched how people make predictions in nearly every case. Our findings suggest that a relatively simple adjustment to conventional prediction-error learning algorithms that allows representation of outcome magnitudes provide a powerful way to capture the information that we preserve when we learn about variable outcomes. This has important implications for understanding how people make predictions and decisions in real-world situations where outcomes naturally vary, from medical treatments to environmental changes.
- Open Access
- 17 January 2025
- Published by Center for Open Science in PsyArXiv
Occasion setting in humans: Norm or exception?
In this commentary, we propose that occasion setting in humans may be even more widespread than Leising and colleagues assume. In contrast to animal work, our research using the feature negative procedure (A+ AX-) reveals substantial individual differences in what people learn about the feature (X). We discuss findings showing that the majority of participants in our experiments learn something akin to occasion setting, and present reasons for why this may be the case. We conclude that occasion-setting may be the norm in humans because it allows existing learning to be preserved, and for the possibility that the effect of a cue is unique to its accompanying target.
- 17 January 2025
- Published by Center for Open Science in PsyArXiv
Open AccessOccasion setting in humans: Norm or exception?
In this commentary, we propose that occasion setting in humans may be even more widespread than Leising and colleagues assume. In contrast to animal work, our research using the feature negative procedure (A+ AX-) reveals substantial individual differences in what people learn about the feature (X). We discuss findings showing that the majority of participants in our experiments learn something akin to occasion setting, and present reasons for why this may be the case. We conclude that occasion-setting may be the norm in humans because it allows existing learning to be preserved, and for the possibility that the effect of a cue is unique to its accompanying target.
- 1 July 2024
- Published by American Psychological Association (APA) in Journal of Experimental Psychology: Learning, Memory, and Cognition
- Vol. 50 (7) , 1167-1185
Using unobserved causes to explain unexpected outcomes: The effect of existing causal knowledge on protection from extinction by a hidden cause.
People often rely on the covariation between events to infer causality. However, covariation between cues and outcomes may change over time. In the associative learning literature, extinction provides a model to study updating of causal beliefs when a previously established relationship no longer holds. Prediction error theories can explain both extinction and protection from extinction when an inhibitory (preventive) cue is present during extinction. In three experiments using the allergist causal learning task, we found that protection could also be achieved by a hidden cause that was inferred but not physically present, so long as that cause was a plausible preventer of the outcome. We additionally showed complete protection by a physically presented cue that was neutral rather than inhibitory at the outset of extinction. Both findings are difficult to reconcile with dominant prediction error theories. However, they are compatible with the idea of theory protection, where the learner attributes the absence of the outcome to the added cue (when present) or to a hidden cause, and therefore does not need to revise causal beliefs about A. Our results suggest that prediction error encourages changes in causal beliefs, but the nature of the change is determined by reasoning processes that incorporate existing knowledge of causal mechanisms and may be biased toward preservation of existing beliefs. (PsycInfo Database Record (c) 2023 APA, all rights reserved).
- 1 July 2024
- Published by American Psychological Association (APA) in Journal of Experimental Psychology: Learning, Memory, and Cognition
- Vol. 50 (7) , 1167-1185
Using unobserved causes to explain unexpected outcomes: The effect of existing causal knowledge on protection from extinction by a hidden cause.
People often rely on the covariation between events to infer causality. However, covariation between cues and outcomes may change over time. In the associative learning literature, extinction provides a model to study updating of causal beliefs when a previously established relationship no longer holds. Prediction error theories can explain both extinction and protection from extinction when an inhibitory (preventive) cue is present during extinction. In three experiments using the allergist causal learning task, we found that protection could also be achieved by a hidden cause that was inferred but not physically present, so long as that cause was a plausible preventer of the outcome. We additionally showed complete protection by a physically presented cue that was neutral rather than inhibitory at the outset of extinction. Both findings are difficult to reconcile with dominant prediction error theories. However, they are compatible with the idea of theory protection, where the learner attributes the absence of the outcome to the added cue (when present) or to a hidden cause, and therefore does not need to revise causal beliefs about A. Our results suggest that prediction error encourages changes in causal beliefs, but the nature of the change is determined by reasoning processes that incorporate existing knowledge of causal mechanisms and may be biased toward preservation of existing beliefs. (PsycInfo Database Record (c) 2023 APA, all rights reserved).
- Open Access
Instruction on the Scientific Method Provides (Some) Protection Against Illusions of Causality
People tend to overestimate the efficacy of an ineffective treatment when they experience the treatment and its supposed outcome co-occurring frequently. This is referred to as the outcome density effect. Here, we attempted to improve the accuracy of participants’ assessments of an ineffective treatment by instructing them about the scientific practice of comparing treatment effects against a relevant base-rate, i.e., when no treatment is delivered. The effect of these instructions was assessed in both a trial-by-trial contingency learning task, where cue administration was either decided by the participant (Experiments 1 & 2) or pre-determined by the experimenter (Experiment 3), as well as in summary format where all information was presented on a single screen (Experiment 4). Overall, we found two means by which base-rate instructions influence efficacy ratings for the ineffective treatment: 1) When information was presented sequentially, the benefit of base-rate instructions on illusory belief was mediated by reduced sampling of cue-present trials, and 2) When information was presented in summary format, we found a direct effect of base-rate instruction on reducing causal illusion. Together, these findings suggest that simple instructions on the scientific method were able to decrease participants’ (over-)weighting of cue-outcome coincidences when making causal judgements, as well as decrease their tendency to over-sample cue-present events. However, the effect of base-rate instructions on correcting illusory beliefs was incomplete, and participants still showed illusory causal judgements when the probability of the outcome occurring was high. Thus, simple textual information about assessing causal relationships is partially effective in influencing people’s judgements of treatment efficacy, suggesting an important role of scientific instruction in debiasing cognitive errors.
Open AccessInstruction on the Scientific Method Provides (Some) Protection Against Illusions of Causality
People tend to overestimate the efficacy of an ineffective treatment when they experience the treatment and its supposed outcome co-occurring frequently. This is referred to as the outcome density effect. Here, we attempted to improve the accuracy of participants’ assessments of an ineffective treatment by instructing them about the scientific practice of comparing treatment effects against a relevant base-rate, i.e., when no treatment is delivered. The effect of these instructions was assessed in both a trial-by-trial contingency learning task, where cue administration was either decided by the participant (Experiments 1 & 2) or pre-determined by the experimenter (Experiment 3), as well as in summary format where all information was presented on a single screen (Experiment 4). Overall, we found two means by which base-rate instructions influence efficacy ratings for the ineffective treatment: 1) When information was presented sequentially, the benefit of base-rate instructions on illusory belief was mediated by reduced sampling of cue-present trials, and 2) When information was presented in summary format, we found a direct effect of base-rate instruction on reducing causal illusion. Together, these findings suggest that simple instructions on the scientific method were able to decrease participants’ (over-)weighting of cue-outcome coincidences when making causal judgements, as well as decrease their tendency to over-sample cue-present events. However, the effect of base-rate instructions on correcting illusory beliefs was incomplete, and participants still showed illusory causal judgements when the probability of the outcome occurring was high. Thus, simple textual information about assessing causal relationships is partially effective in influencing people’s judgements of treatment efficacy, suggesting an important role of scientific instruction in debiasing cognitive errors.
- Open Access
- 20 April 2024
- Published by Center for Open Science in PsyArXiv
Effects of instructed and experienced uncertainty on attentional priority
Previous research has demonstrated that attentional prioritisation is shaped by prior experience of reward uncertainty: attention is more likely to be captured by a stimulus associated with a variable (uncertain) reward than a stimulus that provides diagnostic information about available reward. This finding is noteworthy, because it runs counter to the principle that cognition is motivated to reduce uncertainty and hence surprise. Here we investigated whether this pattern of uncertainty-modulated attentional capture (UMAC) reflects a process of attention for learning, wherein uncertainty-related stimuli are prioritised in an attempt to learn about their true predictive status. To address this, we examined the distinct impact of two information sources that modulate potential for learning; explicit instruction versus ongoing experience of prediction error in reward feedback. Experiment 1 demonstrated that providing explicit instructions—and hence negating the need for further learning—did not reduce the magnitude of the UMAC effect, indicating that UMAC does not reflect attention for learning as a strategic approach for determining the task state. On the other hand, Experiment 2 showed that instructions alone were insufficient to generate a UMAC effect in the absence of reward feedback, suggesting that the impact of uncertainty on rapid attentional prioritisation is driven by direct experience of prediction error. Taken together, these findings point to two possibilities: UMAC may reflect attention for learning operating at an implicit level, or may evince an attentional system that is configured for rapid detection of sources of experienced uncertainty so that subsequent behaviour can be tailored appropriately.
- 20 April 2024
- Published by Center for Open Science in PsyArXiv
Open AccessEffects of instructed and experienced uncertainty on attentional priority
Previous research has demonstrated that attentional prioritisation is shaped by prior experience of reward uncertainty: attention is more likely to be captured by a stimulus associated with a variable (uncertain) reward than a stimulus that provides diagnostic information about available reward. This finding is noteworthy, because it runs counter to the principle that cognition is motivated to reduce uncertainty and hence surprise. Here we investigated whether this pattern of uncertainty-modulated attentional capture (UMAC) reflects a process of attention for learning, wherein uncertainty-related stimuli are prioritised in an attempt to learn about their true predictive status. To address this, we examined the distinct impact of two information sources that modulate potential for learning; explicit instruction versus ongoing experience of prediction error in reward feedback. Experiment 1 demonstrated that providing explicit instructions—and hence negating the need for further learning—did not reduce the magnitude of the UMAC effect, indicating that UMAC does not reflect attention for learning as a strategic approach for determining the task state. On the other hand, Experiment 2 showed that instructions alone were insufficient to generate a UMAC effect in the absence of reward feedback, suggesting that the impact of uncertainty on rapid attentional prioritisation is driven by direct experience of prediction error. Taken together, these findings point to two possibilities: UMAC may reflect attention for learning operating at an implicit level, or may evince an attentional system that is configured for rapid detection of sources of experienced uncertainty so that subsequent behaviour can be tailored appropriately.
- Open Access
Switching from Sugar- to Artificially-Sweetened Beverages: A 12-Week Trial
Background: Consumption of sugar-sweetened beverages (SSBs) forms the primary source of added sugar intake and can increase the risk of metabolic disease. Evidence from studies in humans and rodents also indicates that consumption of SSBs can impair performance on cognitive tests, but that removing SSB access can ameliorate these effects. Methods: The present study used an unblinded 3-group parallel design to assess the effects of a 12-week intervention in which young healthy adults (mean age = 22.85, SD = 3.89; mean BMI: 23.2, SD = 3.6) who regularly consumed SSBs were instructed to replace SSB intake with artificially-sweetened beverages (n = 28) or water (n = 25), or (c) to continue SSB intake (n = 27). Results: No significant group differences were observed in short-term verbal memory on the Logical Memory test or the ratio of waist circumference to height (primary outcomes), nor in secondary measures of effect, impulsivity, adiposity, or glucose tolerance. One notable change was a significant reduction in liking for strong sucrose solutions in participants who switched to water. Switching from SSBs to ‘diet’ drinks or water had no detectable impact on cognitive or metabolic health over the relatively short time frame studied here. This study was prospectively registered with the Australian New Zealand Clinical Trials Registry (ACTRN12615001004550; Universal Trial Number: U1111-1170-4543).
Open AccessSwitching from Sugar- to Artificially-Sweetened Beverages: A 12-Week Trial
Background: Consumption of sugar-sweetened beverages (SSBs) forms the primary source of added sugar intake and can increase the risk of metabolic disease. Evidence from studies in humans and rodents also indicates that consumption of SSBs can impair performance on cognitive tests, but that removing SSB access can ameliorate these effects. Methods: The present study used an unblinded 3-group parallel design to assess the effects of a 12-week intervention in which young healthy adults (mean age = 22.85, SD = 3.89; mean BMI: 23.2, SD = 3.6) who regularly consumed SSBs were instructed to replace SSB intake with artificially-sweetened beverages (n = 28) or water (n = 25), or (c) to continue SSB intake (n = 27). Results: No significant group differences were observed in short-term verbal memory on the Logical Memory test or the ratio of waist circumference to height (primary outcomes), nor in secondary measures of effect, impulsivity, adiposity, or glucose tolerance. One notable change was a significant reduction in liking for strong sucrose solutions in participants who switched to water. Switching from SSBs to ‘diet’ drinks or water had no detectable impact on cognitive or metabolic health over the relatively short time frame studied here. This study was prospectively registered with the Australian New Zealand Clinical Trials Registry (ACTRN12615001004550; Universal Trial Number: U1111-1170-4543).
- Open Access
- 6 January 2023
- Published by Ubiquity Press, Ltd. in Journal of Cognition
- Vol. 6 (1) , 19-32
Inhibitory Learning with Bidirectional Outcomes: Prevention Learning or Causal Learning in the Opposite Direction?
Influential models of causal learning assume that learning about generative and preventive relationships are symmetrical to each other. That is, a preventive cue directly prevents an outcome from occurring (i.e., "direct" prevention) in the same way a generative cue directly causes an outcome to occur. However, previous studies from our lab have shown that many participants do not infer a direct prevention causal structure after feature-negative discrimination (A+/AB-) with a unidirectional outcome (Lee & Lovibond, 2021). Melchers et al. (2006) suggested that the use of a bidirectional outcome that can either increase or decrease from baseline, encourages direct prevention learning. Here we test an alternative possibility that a bidirectional outcome encourages encoding of a generative relationship in the opposite direction, where B directly causes a decrease in the outcome. Thus, previous evidence of direct prevention learning using bidirectional outcomes may instead be explained by some participants inferring an "Opposite Causal" structure. In two experiments, participants did indeed report an opposite causal structure. In Experiment 1, these participants showed the lowest outcome predictions when B was combined with a novel cause in a summation test, and lowest outcome predictions when B was presented alone. In Experiment 2, B successfully blocked learning to a novel cue that was directly paired with a reduction in the outcome, and this effect was strongest among participants who endorsed an Opposite Causal structure. We conclude that previous evidence of direct prevention learning using bidirectional outcomes may be a product of excitatory rather than inhibitory learning.
- 6 January 2023
- Published by Ubiquity Press, Ltd. in Journal of Cognition
- Vol. 6 (1) , 19-32
Open AccessInhibitory Learning with Bidirectional Outcomes: Prevention Learning or Causal Learning in the Opposite Direction?
Influential models of causal learning assume that learning about generative and preventive relationships are symmetrical to each other. That is, a preventive cue directly prevents an outcome from occurring (i.e., "direct" prevention) in the same way a generative cue directly causes an outcome to occur. However, previous studies from our lab have shown that many participants do not infer a direct prevention causal structure after feature-negative discrimination (A+/AB-) with a unidirectional outcome (Lee & Lovibond, 2021). Melchers et al. (2006) suggested that the use of a bidirectional outcome that can either increase or decrease from baseline, encourages direct prevention learning. Here we test an alternative possibility that a bidirectional outcome encourages encoding of a generative relationship in the opposite direction, where B directly causes a decrease in the outcome. Thus, previous evidence of direct prevention learning using bidirectional outcomes may instead be explained by some participants inferring an "Opposite Causal" structure. In two experiments, participants did indeed report an opposite causal structure. In Experiment 1, these participants showed the lowest outcome predictions when B was combined with a novel cause in a summation test, and lowest outcome predictions when B was presented alone. In Experiment 2, B successfully blocked learning to a novel cue that was directly paired with a reduction in the outcome, and this effect was strongest among participants who endorsed an Opposite Causal structure. We conclude that previous evidence of direct prevention learning using bidirectional outcomes may be a product of excitatory rather than inhibitory learning.