Causal Inference

Why Blame Bob: Probabilistic generative models, counterfactual reasoning, and blame attribution

by Tobias Gerstenberg

We consider an approach to blame attribution based on counterfactual reasoning in probabilistic generative models. In... more We consider an approach to blame attribution based on counterfactual reasoning in probabilistic generative models. In this view, people intervene on each variable within their model and assign blame in proportion to how much a change to a variable would have improved the outcome. This approach raises two questions: First, what structure do people use to represent a given situation? Second, how do they choose what alternatives to consider when intervening on an event? We use a series of coin-tossing scenarios to compare empirical data to different models within the proposed framework. The results suggest that people sample their intervention values from a prior rather than deterministically switching the value of a variable. The results further suggest that people represent scenarios differently when asked to reason about their own blame attributions, compared with the blame attributions they believe others will assign.

Ping Pong in Church: Productive use of concepts in human probabilistic inference

by Tobias Gerstenberg

How do people make inferences from complex patterns of evidence across diverse situations? What does a computational... more How do people make inferences from complex patterns of evidence across diverse situations? What does a computational model need in order to capture the abstract knowledge people use for everyday reasoning? In this paper, we explore a novel modeling framework based on the probabilistic language of thought (PLoT) hypothesis, which conceptualizes thinking in terms of probabilistic inference over compositionally structured representations. The core assumptions of the PLoT hypothesis are realized in the probabilistic programming language Church (Goodman, Mansinghka, Roy, Bonawitz, & Tenenbaum, 2008). Using “ping pong tournaments” as a case study, we show how a single Church program concisely represents the concepts required to specify inferences from diverse patterns of evidence. In two experiments, we demonstrate a very close fit between our model’s predictions and participants’ judgments. Our model accurately predicts how people reason with confounded and indirect evidence and how different sources of information are integrated.

Mechanistic evidence: Disambiguating the Russo-Williamson Thesis

by Phyllis Illari

International Studies in the Philosophy of Science. 25(2), (2011): 139–57.
DOI:10.1080/02698595.2011.574856

Russo and Williamson claim that establishing causal claims requires mechanistic and difference-making evidence. In this... more Russo and Williamson claim that establishing causal claims requires mechanistic and difference-making evidence. In this paper, I will argue that Russo and Williamson’s formulation of their thesis is multiply ambiguous. I will make three distinctions: mechanistic evidence as type vs object of evidence; what mechanism or mechanisms we want evidence of; and how much evidence of a mechanism we require. I will feed these more precise meanings back into the Russo-Williamson Thesis and argue that it is both true and false: two weaker versions of the thesis are worth supporting, while the stronger versions are not. Further, my distinctions are of wider concern because they allow us to make more precise claims about what kinds of evidence are required in particular cases.

Effects of Stress and Working Memory Capacity on Foreign Language Readers' Inferential Processing During Comprehension

by Lester Loschky

Although stress is frequently claimed to impede foreign language (FL) reading comprehension, it is usually not... more Although stress is frequently claimed to impede foreign language (FL) reading comprehension, it is usually not explained how. We investigated the effects of stress, working memory (WM) capacity, and inferential complexity on Spanish FL readers’ inferential processing during comprehension. Inferences, although necessary for reading comprehension, vary in inferential complexity and WM demands. We measured 55 intermediate-level Spanish FL learners’ reading comprehension, using questions with three levels of inferential complexity: non-inference (factual), bridging inference (pronoun referent), and pragmatic inference. We measured participants’ WM capacity and varied their stress level between blocks using a video camera. Results showed that higher WM learners were more accurate overall. Inference construction during comprehension was negatively related to inferential complexity. Stress increased processing time overall, with a trend toward greater effect on response times (RTs) for questions requiring greater inferential complexity. Higher WM learners showed a greater effect of inferential complexity on RTs than lower WM learners. More generally, and consistent with the Eysenck, Santos, Derekschan, and Calvo’s (2007) Attentional Control Theory, analyses showed that higher WM learners strategically traded reading speed (processing efficiency) for greater comprehension accuracy (processing effectiveness), whereas lower WM learners only did so under stress and did so less successfully. Thus, stress impedes FL reading comprehension through interactions between WM capacity and inferential complexity, and such effects are moderated by strategy use.

Understanding Counterfactuals and Causation

by Christoph Hoerl

Co-authored with Teresa McCormack and Sarah R. Beck. From C. Hoerl, T. McCormack & S.R. Beck (eds.). Understanding Counterfactuals, Understanding Causation: Issues in Philosophy and Psychology. Oxford University Press 2011

We provide an introduction to some of the key issues raised in this volume by considering how individual chapters bear... more We provide an introduction to some of the key issues raised in this volume by considering how individual chapters bear on the prospects of what may be called a 'counterfactual process view' of causal reasoning. According to such a view, counterfactual thought is an essential part of the processing involved in making causal judgements, at least in a central range of cases that are critical to a subject’s understanding of what it is for one thing to cause another. We argue that one fruitful way of approaching the different contributions to the volume is to think of them as providing materials, conceptual as well as empirical, for challenging counterfactual process views of causal thinking, or for responding to such challenges. Amongst the challenges we consider are ones that arise out of or parallel objections to counterfactual theories of causation in philosophy, or ones that appeal to apparent developmental dissociations between causal and counterfactual reasoning abilities. Possible responses turn on questions such as the following: What should count as engaging in counterfactual reasoning? How should we think of the cognitive prerequisites of such reasoning? Is it right to ask what the relationship is between causal and counterfactual reasoning, or are there in fact a number of different ways in which the two are connected?

Challenges to Inference in the Study of Crisis Bargaining

by Phil Arena

co-authored with Kyle Joyce, draft only

Directed acyclic graphs (DAGs) – The application of causal diagrams in epidemiology

by Juliane Hardt

Schipf S, Knüppel S, Hardt J, Stang A. [Directed acyclic graphs (DAGs) – The application of causal diagrams in epidemiology].[Article in German]. Gesundheitswesen 2011;73(12): 888-892.

Abstract in English available (see link to online copy of the paper) Abstract in English available (see link to online copy of the paper)

Directed acyclic graphs (DAGs) – Basic concepts and application of an approach for causal analyses in epidemiology

by Juliane Hardt

Hardt J, Brendler C, Greiser KH, Timmer A, Seidler A, Weikert C, Latza U. [Directed acyclic graphs (DAGs) – Basic concepts and application of an approach for causal analyses in epidemiology].[Article in German]. Gesundheitswesen 2011;73(12): 878-879.

DAGitty: A graphical tool for analyzing causal diagrams

by Juliane Hardt

Textor J, Hardt J, Knüppel S. DAGitty: A Graphical Tool for Analyzing Causal Diagrams.Epidemiology. 2011 Sep;22(5):745

Universal Laws and Prevention in Causal Structures

by Esteban Céspedes

"Somebody else's problem?" Staff perceptions of the sources and control of meticillin-resistant Staphylococcus aureus.

by Peter Griffiths

Morrow E, Griffiths P, Rao GG, Flaxman D. Am J Infect Control. 2011;39(4):284-91.

Background Meticillin-resistant Staphylococcus aureus (MRSA) is endemic within the United Kingdom health care sector.... more Background Meticillin-resistant Staphylococcus aureus (MRSA) is endemic within the United Kingdom health care sector. Recent campaigns to reduce health care-associated infection have rested on increasing staff accountability and ownership of the problem and its solutions. However, the existence of reservoirs of colonization in the community now creates ambiguity about sources, which may undermine preventative strategies.Methods The theoretical framework of causal attribution was applied to explore staff biases in perceptions and effects on infection control behaviors on both sides of the hospital/care home interface. Forty-four staff from 1 acute care hospital and 53 staff from 6 care homes estimated prevalence, risk, and sources of MRSA. Focus groups (6 care home and 8 hospital) were used to elicit group perceptions.Results Staff tended to attribute the causes of MRSA to external (not self) human factors including patient risk factors and poor infection control practices of others. Teams tend to attribute their "successes" in infection control to dispositional attributions (good team policy and performance) and attribute "lapses" to situational factors (client group, patient movement, work pressures).Conclusion Variations in information needs, ownership, and infection control practices could be addressed by better interorganizational working and support for staff teams to assess their own responses to the problem.

A review of causal inference for biomedical informatics

by Samantha Kleinberg

To appear in the Journal of Biomedical Informatics, 2011

Causality is an important concept throughout the health sciences and is particularly vital for informatics work such... more Causality is an important concept throughout the health sciences and is particularly vital for informatics work such as finding adverse drug events or risk factors for disease using electronic health records. While philosophers and scientists working for centuries on formalizing what makes something a cause have not reached a consensus, new methods for inference show that we can make progress in this area in many practical cases. This article reviews core concepts in understanding and identifying causality and then reviews current computational methods for inference and explanation, focusing on inference from large-scale observational data. While the problem is not fully solved, we show that graphical models and Granger causality provide useful frameworks for inference and that a more recent approach based on temporal logic addresses some of the limitations of these methods.

DAGs und andere Kausalitätstheorien.

by Felix Thoemmes

Das Gesundheitswesen.

Campbell's and Rubin's Perspectives on Causal Inference

by Felix Thoemmes

Psychological Methods

Solving the Flagpole Problem

by Alexander Gebharter

Gebharter, A. (submitted).

When learning order affects sensitivity to base rates: Challenges for theories of causal learning.

by Ulf-Dietrich Reips

Co-authored by Michael R. Waldmann, published in Experimental Psychology: Reips, U.-D., & Waldmann, M. (2008). When learning order affects sensitivity to base rates: Challenges for theories of causal learning. Experimental Psychology, 55, 9-22.

In three experiments we investigated whether two procedures of acquiring knowledge about the same causal structure,... more In three experiments we investigated whether two procedures of acquiring knowledge about the same causal structure, predictive learning (from causes to effects) versus diagnostic learning (from effects to causes), would lead to different base rate use in diagnostic judgments. Results showed that learners are capable of incorporating base rate information in their judgments regardless of the direction in which the causal structure is learned. However, this only holds true for relatively simple scenarios. When complexity was increased, base rates were only used after diagnostic learning, but were largely neglected after predictive learning. It could be shown that this asymmetry is not due to a failure of encoding base rates in predictive learning because participants in all conditions were fairly good at reporting them. The findings present challenges for all theories of causal learning.

The functional nature of induction: Flexibility in children's inferences about object properties

by Gedeon Deák

Deák, G.O. (2006). Representing object functions: The cognitive basis of tool-use by children. Proceedings of the Fifth International Conference on Development and Learning (ICDL’06), Indiana University-Bloomington.

The cognitive basis of tool-use in humans is the capacity for conceptual abstraction of object affordances. These... more The cognitive basis of tool-use in humans is the capacity for conceptual abstraction of object affordances. These abstractions support dynamic flexible representations, or simulations, of agents using instruments to cause effects. This cognitive capacity is largely mature by 4 years of age. It requires high-level networks that take many sources of input, including events that are embedded in social interactions. The content of children’s simulations is therefore dependent on social and exploratory experience, and on the capacities to acquire very abstract conceptualizations and to produce elaborate representations from minimal, or even distracting, perceptual information.

Investigating Causal Relationships In Stock Returns With Temporal Logic Based Methods

by Samantha Kleinberg

Working paper

We describe a new framework for causal inference and its application to return time series. In this system, causal... more We describe a new framework for causal inference and its application to return time series. In this system, causal relationships are represented as logical formulas, allowing us to test arbitrarily complex hypotheses in a computationally efficient way. We simulate return time series using a common factor model, and show that on this data the method described significantly outperforms Granger causality (a primary approach to this type of problem). Finally we apply the method to real return data, showing that the method can discover novel relationships between stocks. The approach described is a general one that will allow combination of price and volume data with qualitative information at varying time scales (from interest rate announcements, to earnings reports to news stories) shedding light on some of the previously invisible common causes of seemingly correlated price movements.

A Logic for Causal Inference in Time Series with Discrete and Continuous Variables

by Samantha Kleinberg

Will appear at IJCAI 2011 as both a poster and a talk

Many applications of causal inference, such as finding the relationship between stock prices and news reports, involve... more Many applications of causal inference, such as finding the relationship between stock prices and news reports, involve both discrete and continuous variables observed over time. Inference with these complex sets of temporal data, though, has remained difficult and required a number of simplifications. We show that recent approaches for inferring temporal relationships (represented as logical formulas) can be adapted for inference with continuous valued effects. Building on advances in logic, PCTLc (an extension of PCTL with numerical constraints) is introduced here to allow representation and inference of relationships with a mixture of discrete and continuous components. Then, finding significant relationships in the continuous case can be done using the conditional expectation of an effect, rather than its conditional probability. We evaluate this approach on both synthetically generated and actual financial market data, demonstrating that it can allow us to answer different questions than the discrete approach can.

What can God show us?

by Kevin Magill

Earlier version presented at the European Congress of Analytic Philosophy second conference, University of Leeds, September 1996

Criticises Peter van Inwagen's claim (in An Essay on Free Will) that it is possible to imagine a stone shattering a... more Criticises Peter van Inwagen's claim (in An Essay on Free Will) that it is possible to imagine a stone shattering a window and that God subsequently reveals that the window, although caused to shatter by the stone, did not have to shatter. The purpose of van Inwagen's thought experiment is to argue that `it is not part of the concept of causation that a cause - or even a cause plus the totality of its accompanying conditions - determines the effect'. I argue, first, that if a stone is thrown at a window and the window fails to shatter, our natural and reasonable assumption will be that the stone was not thrown hard enough or that the window is impact resistant: in short, we will assume that the window failed to shatter because of the absence of some necessary condition for its shattering. In that case, it is part of our concept of causation that a cause, plus the totality of its accompanying conditions, determines its effect. It is also part the concept of causation that effects are explained by citing their causes, and this is inconsistent with the supposition that, given the occurrence of a cause and the totality of its accompanying conditions, its effect might not have occurred. Therefore, if God were to persuade us that a window that is caused to shatter did not have to do so, this would constitute a revision of our concept of causation.
I go on to argue that, in any case, we cannot imagine anything that could count as revealing to us that when a window shatters, given everything as it was at the point of impact, the window did not have to shatter. I argue that even if a case could be made for saying that if we have reason to believe anything, we have reason to believe what God tells us or shows us, there is nothing we can imagine that would give us good reason to believe that it would really be God, and not a hellish impostor, who is doing the showing or telling. Therefore, what van Inwagen invites us to imagine is unimaginable. I conclude that we cannot conceive of any circumstance that can give us good reason for abandoning or modifying our natural and reasonable inclination to say that when causes fail to produce their expected effects, some necessary accompanying condition must have been missing.