Bob Rehder - Academia.edu (original) (raw)

Papers by Bob Rehder

Research paper thumbnail of Causal status and coherence in causal-based categorization

PsycEXTRA Dataset, 2009

Research has documented two effects of interfeature causal knowledge on classification. A causal ... more Research has documented two effects of interfeature causal knowledge on classification. A causal status effect occurs when features that are causes are more important to category membership than their effects. A coherence effect occurs when combinations of features that are consistent with causal laws provide additional evidence of category membership. In this study, we found that stronger causal relations led to a weaker causal status effect and a stronger coherence effect (Experiment 1), that weaker alternative causes led to stronger causal status and coherence effects (Experiment 2), and that "essentialized" categories led to a stronger causal status effect (Experiment 3), albeit only for probabilistic causal links (Experiment 4). In addition, the causal status effect was mediated by features' subjective category validity, the probability they occur in category members. These findings were consistent with a generative model of categorization but inconsistent with an alternative model.

Research paper thumbnail of Modeling Category Learning with Exemplars and Prior Knowledge

Proceedings of the Annual Meeting of the Cognitive Science Society, 2006

An open question in category learning research is how prior knowledge affects the process of lear... more An open question in category learning research is how prior knowledge affects the process of learning new concepts. Rehder and Murphy's (2003) Knowledge Resonance (KRES) model of concept learning uses an interactive neural network to account for many observed effects related to prior knowledge, but cannot account for the learning of nonlinearly separable concepts. In this work, we extend the KRES model by adding exemplar nodes. The new model accounts for the fact that linearly separable concepts are not necessarily easier than nonlinearly separable concepts (Medin & Schwanenflugel, 1981), and more importantly, accounts for a notable interaction between the presence of useful prior knowledge and linear separability (Wattenmaker, Dewey, Murphy, & Medin, 1986). Two architectural variants of the model were tested, and the dependence of good results on a particular architecture, indicates how formal modeling can uncover facts about how the prior knowledge which influences concept learning is used and represented.

Research paper thumbnail of Causal status, coherence, and essentialized categories

Proceedings of the Annual Meeting of the Cognitive Science Society, 2007

Research paper thumbnail of Causal-Knowledge and Information Search During Categorization

Research paper thumbnail of Eye movements and knowledge in category learning

Research paper thumbnail of A causal model approach to dynamic control

Cognitive Science, Dec 31, 2018

Acting effectively in the world requires learning and controlling dynamic systems, that is, syste... more Acting effectively in the world requires learning and controlling dynamic systems, that is, systems involving feedback relations among continuous variables that vary in real time. We introduce a novel class of dynamic control environments using Ornstein-Uhlenbeck processes connected in causal Markov graphs that allow us to systematically test people's ability to learn and control various dynamic systems. We find that performance varied across a range of test environments, roughly matching with complexity defined by a set of models trained on the task (an optimal model, a deep Reinforcement Learning agent, and a PID controller). The testbed of dynamic environments and class of models introduced in this paper lay the groundwork for the systematic study of people's ability to control complex dynamic systems.

Research paper thumbnail of Attention allocation in inference learning

Proceedings of the Annual Meeting of the Cognitive Science Society, 2007

The majority of experimental research on human concept learning has relied on the supervised clas... more The majority of experimental research on human concept learning has relied on the supervised classification task. Recently it has been argued that a complete understanding of concept learning requires expanding laboratory techniques to account for other uses of categories. Yamauchi and Markman (1998) examined the relationship between the use of categories and their representation by contrasting traditional supervised classification learning and inference learning. Participants learned about a 4-dimensional family resemblance category structure by either classifying items into 2 categories or by making inferences about missing features with the items’ category label present. The learning phase continued until participants reached an accuracy criterion. In a subsequent test phase all participants made inferences about every dimension of every item. Inference learners were significantly more likely then classification learners to make inferences in accordance with the category prototype. This was the case even when the to-be-predicted dimension had displayed an exception feature (i.e., a feature from the opposite category) when that item had appeared during training. This result has been interpreted as evidence that inference learning induces learners to represent the internal structure of categories, including typical features and within-category correlations (Chin-Parker & Ross 2002; Yamauchi & Markman, 1998). In contrasts, classification learning seems to encourage learning only the dimensions that are diagnostic of category membership. However, an alternate explanation for inference learners’ preference for prototype consistent features is that participants may be learning simple bidirectional rules relating the category label to each feature (Johansen & Kruschke, 2005). This explanation is reasonable because the inference learners only predicted features that were consistent with the category prototype during learning. A replication of Yamauchi & Markman (1998) Experiment 1 was conducted with an eye tracker because the two hypotheses make different predictions about attention allocation. If inference learners are motivated to learn the structure of a category they should attend multiple dimensions; in contrast, a category label-based rule would only require attending to the label. The main behavioral results replicated, including the fact that the inference condition (M = 7.95) required significantly fewer blocks than the classification condition (M = 18.61). The theoretically interesting effect from the test phase was also observed in that inference learners responded to exception-feature queries in accordance with the category prototype significantly more than classification learners (M = .94, M = .68).

Research paper thumbnail of Feature inference and eyetracking

In addition to traditional supervised classification learning, people can also learn categories b... more In addition to traditional supervised classification learning, people can also learn categories by predicting the features of category members. It has been proposed that feature inference learning promotes the learning of more within-category information and a prototype representation of the category, as compared to classification learning that promotes learning of diagnostic information. We tracked learners' eye movements during inference learning and found (Expt. 1) that they indeed fixated other features (even though those features were not necessary to predict the missing feature), providing the opportunity to extract within-category information. But those fixations were limited to only those features that needed to be predicted on future trials (Expt. 2). In other words, inference learning promotes the acquisition of within-category information not because participants are motivated to learn that information, but rather because of the anticipatory learning it induces. Whenever a person classifies an object, describes a concept verbally, engages in problem solving, or infers missing information, they must access their conceptual knowledge. As a result, the study of concept acquisition has been a critical part of understanding how people experience the world and how they interact with it in appropriate ways. Concept researchers have developed sophisticated formal theories that explain certain aspects of concept acquisition. These theories are largely based on the study of what has come to be known as standard supervised classification—a task that occupies the majority of experimental research in this area (Solomon, Medin, & Lynch, 1999). However, an emerging literature is focused on expanding the range of tasks that can be used to inform our models of concept acquisition. By studying different learning tasks we can understand other aspects of concept acquisition, including the interplay between category use and the type of concept learned (Brooks, 1978; Yamauchi & Markman, 1998, 2000, 2002; Chin-Parker & Ross, 2002). Within this research, the distinction between inference and classification tasks has received the most attention, perhaps because those two tasks can be more easily equated. In fact, Anderson (1991) has argued that inference and classification can be treated identically if category labels and category features are interchangeable (however see Yamauchi & Markman, 2000). Research on classification versus inference learning has revealed apparent differences in the types of category representations formed. Whereas classification promotes learning the most diagnostic features for determining category membership, inference may foster learning additional category information (Chin-Parker & Ross, 2004; Medin et al., 1987; Shepard, Hovland, & Jenkins, 1961; Rehder & Hoffman, 2005a). Classification versus inference learning also affects the ease with which different category structures are acquired. Linearly separable (family-resemblance) category structures are more easily acquired through inference relative to classification (Yamauchi & Markman, 1998). However, when a comparable non-linearly separable category structure is used, classification yields a significant learning advantage (Yamauchi & Markman, 2002). Differences in how category information is acquired across classification and inference tasks have been explained in terms of exemplars and prototypes. Yamauchi and Markman have argued that inference learners form representations consistent with prototype models because they seem to extract family-resemblance information such as typical features and typical feature relations. In contrast, by focusing on diagnostic information, classification encourages representations consistent with learning rules and exceptions (perhaps via exemplar memorization). Nevertheless, this interpretation has been challenged by arguments noting the many differences between the classification and inference tasks. This debate is worth discussing in detail. Yamauchi and Markman (1998, Exp. 1) contrasted classification and inference learning by training groups of participants on a family resemblance category structure, consisting of four exemplars per category (see Table 1). Each item consisted of a label and four binary feature dimensions. The members of both categories were derived from category prototypes, A = 0000 and B = 1111. All items had one dimension that contained a feature value taken from the opposite category prototype, i.e., an exception feature. Participants either classified the eight exemplars into two categories or they predicted a feature missing from every exemplar. One critical aspect of their design was that participants were never required to predict a missing exception feature. For example, they were never presented with the item 000x labelled as a member of category A and asked to predict (on

Research paper thumbnail of Reasoning With Causal Cycles

Cognitive Science, Nov 17, 2016

This article assesses how people reason with categories whose features are related in causal cycl... more This article assesses how people reason with categories whose features are related in causal cycles. Whereas models based on causal graphical models (CGMs) have enjoyed success modeling category‐based judgments as well as a number of other cognitive phenomena, CGMs are only able to represent causal structures that are acyclic. A number of new formalisms that allow cycles are introduced and evaluated. Dynamic Bayesian networks (DBNs) represent cycles by unfolding them over time. Chain graphs augment CGMs by allowing the presence of undirected links that model feedback relations between variables. Unfolded chain graphs are chain graphs that unfold over time. An existing model of causal cycles (alpha centrality) is also evaluated. Four experiments in which subjects reason about categories with cyclically related features provided evidence against DBNs and alpha centrality and for the two types of chain graphs. Chain graphs—a mechanism for representing the equilibrium distribution of a dynamic system—may thus be good candidates for modeling how people reason causally with complex systems. Applications of chain graphs to areas of cognition other than category‐based judgments are discussed.

Research paper thumbnail of Emerging Insights from Eye-Movement Research on Category Learning

Proceedings of the Annual Meeting of the Cognitive Science Society, 2010

Research paper thumbnail of Attention Allocation in Inference Learning - eScholarship

Proceedings of the Annual Meeting of the Cognitive Science Society, 2007

Research paper thumbnail of A Generative Model of Causal Cycles

Cognitive Science, 2011

Causal graphical models (CGMs) have become popular in numerous domains of psychological research ... more Causal graphical models (CGMs) have become popular in numerous domains of psychological research for representing people's causal knowledge. Unfortunately, however, the CGMs typically used in cognitive models prohibit representations of causal cycles. Building on work in machine learning, we propose an extension of CGMs that allows cycles and apply that representation to one real-world reasoning task, namely, classification. Our model's predictions were assessed in experiments that tested both probabilistic and deterministic causal relations. The results were qualitatively consistent with the predictions of our model and inconsistent with those of an alternative model.

Research paper thumbnail of How does this thing work? Evaluating computational models of intervention-based causal learning

Research paper thumbnail of Variability in causal judgments

People's causal judgments exhibit substantial variability, but the processes that lead to thi... more People's causal judgments exhibit substantial variability, but the processes that lead to this variability are not currently understood. In this paper, we use a repeated-measures design to study the within-participant variability of conditional probability judgments in common-cause networks. We establish that these judgments indeed exhibit substantial within-participant variability. This variability differs by inference type and is related to the extent to which participants commit Markov violations. The consistency and systematicity of this variability suggest that it may be an important source of evidence for the cognitive processes that lead to causal judgments. The systematic study of both within- and between-person variability broadens the scope of behavior that can be studied in causal cognition and promotes the evaluation of formal models of the underlying process. The data and methods provided in this paper provide tools to enable the further study of within-participant ...

Research paper thumbnail of Knowledge Effect and Selective Attention in Category Learning: An Eyetracking

Two experiments tested the effect of prior knowledge on attention allocation in category learning... more Two experiments tested the effect of prior knowledge on attention allocation in category learning. Using eyetracking, we found that (a) knowledge affects dimensional attention allocation, with knowledge-relevant features being fixated more often than irrelevant ones, (b) this effect was not due to initial attention bias to the relevant dimensions but rather gradually emerged in response to observing category members, and (c) the effect grew even after the last error trial, that is, in the absence of error. These results pose challenge to current models of knowledge-based category learning. Because of the importance of categories for human cognition, the manner in which people learn categories has received intensive study. Among many procedures, supervised classification learning has been popular, and a

Research paper thumbnail of Causal status, coherence, and essentialized categories

Causal Status, Coherence, and Essentialized Categories ShinWoo Kim (shinwoo.kim@nyu.edu) Bob Rehd... more Causal Status, Coherence, and Essentialized Categories ShinWoo Kim (shinwoo.kim@nyu.edu) Bob Rehder (bob.rehder@nyu.edu) Department of Psychology, New York University 6 Washington Place, New York, NY 10003 USA Essential Chain Keywords: Categorization; causal status; coherence; feature weight; psychological essentialism E Y E Z X Control Y Z E X Y Z Figure 1 most Lake Victoria Shrimp had a high body weight and that some had a low body weight). The ratings were subjected to regression with predictors that coded the presence or absence of each feature (which provided a measure of f eature import ance) and predictors that coded the 2-way interactions between features (which provided a measure of coherence). T he results are presented in Fig. 2. First, a stronger causal status effect (feature weight X > Y > Z) obtained in the Essential-Chain condition than in the Chain condition. This result confirms Rehder's (2003) claim that the causal status effect depends on the features be...

Research paper thumbnail of Modular versus Integrated Causal Learning

Cognitive Science, 2016

Many pieces of information are potentially important to causal inference. Determining whether vit... more Many pieces of information are potentially important to causal inference. Determining whether vitamin C prevents colds may entail knowing the frequency with which colds occur without vitamin C, other cold inhibitors, and the frequency of vitamin C use. Do reasoners integrate all this information to create coherent beliefs? In contrast to models emphasizing modular causal learning (e.g., Cheng, 1997), McDonnell, Tsividis, & Rehder (2013) proposed an integrated model, positing that individuals simultaneously update their beliefs about all components of a causal network. We tested modular versus integrated learning in two experiments using a retrospective inhibition design. In both, participants learned about two causes of headaches sequentially across two phases. We manipulated the base rate of headaches in phase II to be either consistent or inconsistent with phase I learning. Across experiments, participants failed to use base rate information as predicted by the integrated model, s...

Research paper thumbnail of Do Causes or Effects Dominate Categorization Decisions? A Test of a Causal Model Theory of Categorization

Research paper thumbnail of Causal Structure Learning with Continuous Variables in Continuous Time

Cognitive Science, 2018

Interventions, time, and continuous-valued variables are all potentially powerful cues to causati... more Interventions, time, and continuous-valued variables are all potentially powerful cues to causation. Furthermore, when observed over time, causal processes can contain feedback and oscillatory dynamics that make inference hard. We present a generative model and framework for causal inference over continuous variables in continuous time based on Ornstein-Uhlenbeck processes. Our generative model produces a stochastic sequence of evolving variable values that manifest many dynamical properties depending on the nature of the causal relationships, and a learner’s interventions (manual changes to the values of variables during a trial). Our model is also invertible, allowing us to benchmark participant judgments against an optimal model. We find that when interacting with systems acting according to this formalism people directly compare relationships between individual variable pairs rather than considering the full space of possible models, in accordance with a local computations model...

Research paper thumbnail of The Paradox of Time in Dynamic Causal Systems

Entropy

Recent work has shown that people use temporal information including order, delay, and variabilit... more Recent work has shown that people use temporal information including order, delay, and variability to infer causality between events. In this study, we build on this work by investigating the role of time in dynamic systems, where causes take continuous values and also continually influence their effects. Recent studies of learning in these systems explored short interactions in a setting with rapidly evolving dynamics and modeled people as relying on simpler, resource-limited strategies to grapple with the stream of information. A natural question that arises from such an account is whether interacting with systems that unfold more slowly might reduce the systematic errors that result from these strategies. Paradoxically, we find that slowing the task indeed reduced the frequency of one type of error, albeit at the cost of increasing the overall error rate. To explain these results we posit that human learners analyze continuous dynamics into discrete events and use the observed re...

Research paper thumbnail of Causal status and coherence in causal-based categorization

PsycEXTRA Dataset, 2009

Research has documented two effects of interfeature causal knowledge on classification. A causal ... more Research has documented two effects of interfeature causal knowledge on classification. A causal status effect occurs when features that are causes are more important to category membership than their effects. A coherence effect occurs when combinations of features that are consistent with causal laws provide additional evidence of category membership. In this study, we found that stronger causal relations led to a weaker causal status effect and a stronger coherence effect (Experiment 1), that weaker alternative causes led to stronger causal status and coherence effects (Experiment 2), and that "essentialized" categories led to a stronger causal status effect (Experiment 3), albeit only for probabilistic causal links (Experiment 4). In addition, the causal status effect was mediated by features' subjective category validity, the probability they occur in category members. These findings were consistent with a generative model of categorization but inconsistent with an alternative model.

Research paper thumbnail of Modeling Category Learning with Exemplars and Prior Knowledge

Proceedings of the Annual Meeting of the Cognitive Science Society, 2006

An open question in category learning research is how prior knowledge affects the process of lear... more An open question in category learning research is how prior knowledge affects the process of learning new concepts. Rehder and Murphy's (2003) Knowledge Resonance (KRES) model of concept learning uses an interactive neural network to account for many observed effects related to prior knowledge, but cannot account for the learning of nonlinearly separable concepts. In this work, we extend the KRES model by adding exemplar nodes. The new model accounts for the fact that linearly separable concepts are not necessarily easier than nonlinearly separable concepts (Medin & Schwanenflugel, 1981), and more importantly, accounts for a notable interaction between the presence of useful prior knowledge and linear separability (Wattenmaker, Dewey, Murphy, & Medin, 1986). Two architectural variants of the model were tested, and the dependence of good results on a particular architecture, indicates how formal modeling can uncover facts about how the prior knowledge which influences concept learning is used and represented.

Research paper thumbnail of Causal status, coherence, and essentialized categories

Proceedings of the Annual Meeting of the Cognitive Science Society, 2007

Research paper thumbnail of Causal-Knowledge and Information Search During Categorization

Research paper thumbnail of Eye movements and knowledge in category learning

Research paper thumbnail of A causal model approach to dynamic control

Cognitive Science, Dec 31, 2018

Acting effectively in the world requires learning and controlling dynamic systems, that is, syste... more Acting effectively in the world requires learning and controlling dynamic systems, that is, systems involving feedback relations among continuous variables that vary in real time. We introduce a novel class of dynamic control environments using Ornstein-Uhlenbeck processes connected in causal Markov graphs that allow us to systematically test people's ability to learn and control various dynamic systems. We find that performance varied across a range of test environments, roughly matching with complexity defined by a set of models trained on the task (an optimal model, a deep Reinforcement Learning agent, and a PID controller). The testbed of dynamic environments and class of models introduced in this paper lay the groundwork for the systematic study of people's ability to control complex dynamic systems.

Research paper thumbnail of Attention allocation in inference learning

Proceedings of the Annual Meeting of the Cognitive Science Society, 2007

The majority of experimental research on human concept learning has relied on the supervised clas... more The majority of experimental research on human concept learning has relied on the supervised classification task. Recently it has been argued that a complete understanding of concept learning requires expanding laboratory techniques to account for other uses of categories. Yamauchi and Markman (1998) examined the relationship between the use of categories and their representation by contrasting traditional supervised classification learning and inference learning. Participants learned about a 4-dimensional family resemblance category structure by either classifying items into 2 categories or by making inferences about missing features with the items’ category label present. The learning phase continued until participants reached an accuracy criterion. In a subsequent test phase all participants made inferences about every dimension of every item. Inference learners were significantly more likely then classification learners to make inferences in accordance with the category prototype. This was the case even when the to-be-predicted dimension had displayed an exception feature (i.e., a feature from the opposite category) when that item had appeared during training. This result has been interpreted as evidence that inference learning induces learners to represent the internal structure of categories, including typical features and within-category correlations (Chin-Parker & Ross 2002; Yamauchi & Markman, 1998). In contrasts, classification learning seems to encourage learning only the dimensions that are diagnostic of category membership. However, an alternate explanation for inference learners’ preference for prototype consistent features is that participants may be learning simple bidirectional rules relating the category label to each feature (Johansen & Kruschke, 2005). This explanation is reasonable because the inference learners only predicted features that were consistent with the category prototype during learning. A replication of Yamauchi & Markman (1998) Experiment 1 was conducted with an eye tracker because the two hypotheses make different predictions about attention allocation. If inference learners are motivated to learn the structure of a category they should attend multiple dimensions; in contrast, a category label-based rule would only require attending to the label. The main behavioral results replicated, including the fact that the inference condition (M = 7.95) required significantly fewer blocks than the classification condition (M = 18.61). The theoretically interesting effect from the test phase was also observed in that inference learners responded to exception-feature queries in accordance with the category prototype significantly more than classification learners (M = .94, M = .68).

Research paper thumbnail of Feature inference and eyetracking

In addition to traditional supervised classification learning, people can also learn categories b... more In addition to traditional supervised classification learning, people can also learn categories by predicting the features of category members. It has been proposed that feature inference learning promotes the learning of more within-category information and a prototype representation of the category, as compared to classification learning that promotes learning of diagnostic information. We tracked learners' eye movements during inference learning and found (Expt. 1) that they indeed fixated other features (even though those features were not necessary to predict the missing feature), providing the opportunity to extract within-category information. But those fixations were limited to only those features that needed to be predicted on future trials (Expt. 2). In other words, inference learning promotes the acquisition of within-category information not because participants are motivated to learn that information, but rather because of the anticipatory learning it induces. Whenever a person classifies an object, describes a concept verbally, engages in problem solving, or infers missing information, they must access their conceptual knowledge. As a result, the study of concept acquisition has been a critical part of understanding how people experience the world and how they interact with it in appropriate ways. Concept researchers have developed sophisticated formal theories that explain certain aspects of concept acquisition. These theories are largely based on the study of what has come to be known as standard supervised classification—a task that occupies the majority of experimental research in this area (Solomon, Medin, & Lynch, 1999). However, an emerging literature is focused on expanding the range of tasks that can be used to inform our models of concept acquisition. By studying different learning tasks we can understand other aspects of concept acquisition, including the interplay between category use and the type of concept learned (Brooks, 1978; Yamauchi & Markman, 1998, 2000, 2002; Chin-Parker & Ross, 2002). Within this research, the distinction between inference and classification tasks has received the most attention, perhaps because those two tasks can be more easily equated. In fact, Anderson (1991) has argued that inference and classification can be treated identically if category labels and category features are interchangeable (however see Yamauchi & Markman, 2000). Research on classification versus inference learning has revealed apparent differences in the types of category representations formed. Whereas classification promotes learning the most diagnostic features for determining category membership, inference may foster learning additional category information (Chin-Parker & Ross, 2004; Medin et al., 1987; Shepard, Hovland, & Jenkins, 1961; Rehder & Hoffman, 2005a). Classification versus inference learning also affects the ease with which different category structures are acquired. Linearly separable (family-resemblance) category structures are more easily acquired through inference relative to classification (Yamauchi & Markman, 1998). However, when a comparable non-linearly separable category structure is used, classification yields a significant learning advantage (Yamauchi & Markman, 2002). Differences in how category information is acquired across classification and inference tasks have been explained in terms of exemplars and prototypes. Yamauchi and Markman have argued that inference learners form representations consistent with prototype models because they seem to extract family-resemblance information such as typical features and typical feature relations. In contrast, by focusing on diagnostic information, classification encourages representations consistent with learning rules and exceptions (perhaps via exemplar memorization). Nevertheless, this interpretation has been challenged by arguments noting the many differences between the classification and inference tasks. This debate is worth discussing in detail. Yamauchi and Markman (1998, Exp. 1) contrasted classification and inference learning by training groups of participants on a family resemblance category structure, consisting of four exemplars per category (see Table 1). Each item consisted of a label and four binary feature dimensions. The members of both categories were derived from category prototypes, A = 0000 and B = 1111. All items had one dimension that contained a feature value taken from the opposite category prototype, i.e., an exception feature. Participants either classified the eight exemplars into two categories or they predicted a feature missing from every exemplar. One critical aspect of their design was that participants were never required to predict a missing exception feature. For example, they were never presented with the item 000x labelled as a member of category A and asked to predict (on

Research paper thumbnail of Reasoning With Causal Cycles

Cognitive Science, Nov 17, 2016

This article assesses how people reason with categories whose features are related in causal cycl... more This article assesses how people reason with categories whose features are related in causal cycles. Whereas models based on causal graphical models (CGMs) have enjoyed success modeling category‐based judgments as well as a number of other cognitive phenomena, CGMs are only able to represent causal structures that are acyclic. A number of new formalisms that allow cycles are introduced and evaluated. Dynamic Bayesian networks (DBNs) represent cycles by unfolding them over time. Chain graphs augment CGMs by allowing the presence of undirected links that model feedback relations between variables. Unfolded chain graphs are chain graphs that unfold over time. An existing model of causal cycles (alpha centrality) is also evaluated. Four experiments in which subjects reason about categories with cyclically related features provided evidence against DBNs and alpha centrality and for the two types of chain graphs. Chain graphs—a mechanism for representing the equilibrium distribution of a dynamic system—may thus be good candidates for modeling how people reason causally with complex systems. Applications of chain graphs to areas of cognition other than category‐based judgments are discussed.

Research paper thumbnail of Emerging Insights from Eye-Movement Research on Category Learning

Proceedings of the Annual Meeting of the Cognitive Science Society, 2010

Research paper thumbnail of Attention Allocation in Inference Learning - eScholarship

Proceedings of the Annual Meeting of the Cognitive Science Society, 2007

Research paper thumbnail of A Generative Model of Causal Cycles

Cognitive Science, 2011

Causal graphical models (CGMs) have become popular in numerous domains of psychological research ... more Causal graphical models (CGMs) have become popular in numerous domains of psychological research for representing people's causal knowledge. Unfortunately, however, the CGMs typically used in cognitive models prohibit representations of causal cycles. Building on work in machine learning, we propose an extension of CGMs that allows cycles and apply that representation to one real-world reasoning task, namely, classification. Our model's predictions were assessed in experiments that tested both probabilistic and deterministic causal relations. The results were qualitatively consistent with the predictions of our model and inconsistent with those of an alternative model.

Research paper thumbnail of How does this thing work? Evaluating computational models of intervention-based causal learning

Research paper thumbnail of Variability in causal judgments

People's causal judgments exhibit substantial variability, but the processes that lead to thi... more People's causal judgments exhibit substantial variability, but the processes that lead to this variability are not currently understood. In this paper, we use a repeated-measures design to study the within-participant variability of conditional probability judgments in common-cause networks. We establish that these judgments indeed exhibit substantial within-participant variability. This variability differs by inference type and is related to the extent to which participants commit Markov violations. The consistency and systematicity of this variability suggest that it may be an important source of evidence for the cognitive processes that lead to causal judgments. The systematic study of both within- and between-person variability broadens the scope of behavior that can be studied in causal cognition and promotes the evaluation of formal models of the underlying process. The data and methods provided in this paper provide tools to enable the further study of within-participant ...

Research paper thumbnail of Knowledge Effect and Selective Attention in Category Learning: An Eyetracking

Two experiments tested the effect of prior knowledge on attention allocation in category learning... more Two experiments tested the effect of prior knowledge on attention allocation in category learning. Using eyetracking, we found that (a) knowledge affects dimensional attention allocation, with knowledge-relevant features being fixated more often than irrelevant ones, (b) this effect was not due to initial attention bias to the relevant dimensions but rather gradually emerged in response to observing category members, and (c) the effect grew even after the last error trial, that is, in the absence of error. These results pose challenge to current models of knowledge-based category learning. Because of the importance of categories for human cognition, the manner in which people learn categories has received intensive study. Among many procedures, supervised classification learning has been popular, and a

Research paper thumbnail of Causal status, coherence, and essentialized categories

Causal Status, Coherence, and Essentialized Categories ShinWoo Kim (shinwoo.kim@nyu.edu) Bob Rehd... more Causal Status, Coherence, and Essentialized Categories ShinWoo Kim (shinwoo.kim@nyu.edu) Bob Rehder (bob.rehder@nyu.edu) Department of Psychology, New York University 6 Washington Place, New York, NY 10003 USA Essential Chain Keywords: Categorization; causal status; coherence; feature weight; psychological essentialism E Y E Z X Control Y Z E X Y Z Figure 1 most Lake Victoria Shrimp had a high body weight and that some had a low body weight). The ratings were subjected to regression with predictors that coded the presence or absence of each feature (which provided a measure of f eature import ance) and predictors that coded the 2-way interactions between features (which provided a measure of coherence). T he results are presented in Fig. 2. First, a stronger causal status effect (feature weight X > Y > Z) obtained in the Essential-Chain condition than in the Chain condition. This result confirms Rehder's (2003) claim that the causal status effect depends on the features be...

Research paper thumbnail of Modular versus Integrated Causal Learning

Cognitive Science, 2016

Many pieces of information are potentially important to causal inference. Determining whether vit... more Many pieces of information are potentially important to causal inference. Determining whether vitamin C prevents colds may entail knowing the frequency with which colds occur without vitamin C, other cold inhibitors, and the frequency of vitamin C use. Do reasoners integrate all this information to create coherent beliefs? In contrast to models emphasizing modular causal learning (e.g., Cheng, 1997), McDonnell, Tsividis, & Rehder (2013) proposed an integrated model, positing that individuals simultaneously update their beliefs about all components of a causal network. We tested modular versus integrated learning in two experiments using a retrospective inhibition design. In both, participants learned about two causes of headaches sequentially across two phases. We manipulated the base rate of headaches in phase II to be either consistent or inconsistent with phase I learning. Across experiments, participants failed to use base rate information as predicted by the integrated model, s...

Research paper thumbnail of Do Causes or Effects Dominate Categorization Decisions? A Test of a Causal Model Theory of Categorization

Research paper thumbnail of Causal Structure Learning with Continuous Variables in Continuous Time

Cognitive Science, 2018

Interventions, time, and continuous-valued variables are all potentially powerful cues to causati... more Interventions, time, and continuous-valued variables are all potentially powerful cues to causation. Furthermore, when observed over time, causal processes can contain feedback and oscillatory dynamics that make inference hard. We present a generative model and framework for causal inference over continuous variables in continuous time based on Ornstein-Uhlenbeck processes. Our generative model produces a stochastic sequence of evolving variable values that manifest many dynamical properties depending on the nature of the causal relationships, and a learner’s interventions (manual changes to the values of variables during a trial). Our model is also invertible, allowing us to benchmark participant judgments against an optimal model. We find that when interacting with systems acting according to this formalism people directly compare relationships between individual variable pairs rather than considering the full space of possible models, in accordance with a local computations model...

Research paper thumbnail of The Paradox of Time in Dynamic Causal Systems

Entropy

Recent work has shown that people use temporal information including order, delay, and variabilit... more Recent work has shown that people use temporal information including order, delay, and variability to infer causality between events. In this study, we build on this work by investigating the role of time in dynamic systems, where causes take continuous values and also continually influence their effects. Recent studies of learning in these systems explored short interactions in a setting with rapidly evolving dynamics and modeled people as relying on simpler, resource-limited strategies to grapple with the stream of information. A natural question that arises from such an account is whether interacting with systems that unfold more slowly might reduce the systematic errors that result from these strategies. Paradoxically, we find that slowing the task indeed reduced the frequency of one type of error, albeit at the cost of increasing the overall error rate. To explain these results we posit that human learners analyze continuous dynamics into discrete events and use the observed re...