Inferring causal networks from observations and interventions

Information about the structure of a causal system can come in the form of observational data—random samples of the system's autonomous behavior—or interventional data—samples conditioned on the particular values of one or more variables that have been experimentally manipulated. Here we study people's ability to infer causal structure from both observation and intervention, and to choose informative interventions on the basis of observational data. In three causal inference tasks, participants were to some degree capable of distinguishing between competing causal hypotheses on the basis of purely observational data. Performance improved substantially when participants were allowed to observe the effects of interventions that they performed on the systems. We develop computational models of how people infer causal structure from data and how they plan intervention experiments, based on the representational framework of causal graphical models and the inferential principles of optimal Bayesian decision‐making and maximizing expected information gain. These analyses suggest that people can make rational causal inferences, subject to psychologically reasonable representational assumptions and computationally reasonable processing constraints.     

Biased belief in the Bayesian brain: A deeper look at the evidence

A recent critique of hierarchical Bayesian models of delusion argues that, contrary to a key assumption of these models, belief formation in the healthy (i.e., neurotypical) mind is manifestly non-Bayesian. Here we provide a deeper examination of the empirical evidence underlying this critique. We argue that this evidence does not convincingly refute the assumption that belief formation in the neurotypical mind approximates Bayesian inference. Our argument rests on two key points. First, evidence that purports to reveal the most damning violation of Bayesian updating in human belief formation is counterweighted by substantial evidence that indicates such violations are the rare exception—not a common occurrence. Second, the remaining evidence does not demonstrate convincing violations of Bayesian inference in human belief updating; primarily because this evidence derives from study designs that produce results that are not obviously inconsistent with Bayesian principles.     

Focus induced tunnel vision in managerial judgment and decision making: The peril and the antidote

Managers often must assess singular strategic options. Four studies of such assessments demonstrate a tunnel vision effect: Focal managerial options often are favored in an evidentially unjustifiable manner. Study 1 concerns new product development, and demonstrates that a prototype that has become focal tends to be judged overly favorably, and is chosen for launch with unwarranted enthusiasm. Study 2 shows that this tunnel vision effect generalizes to judgments and decisions about general strategy. Study 3 focuses on the information search patterns underlying the effect, and Study 4 replicates the tunnel vision effect among experienced executives, and demonstrates the utility of a debiasing procedure. Data in all of the studies implicate selective processing as the driver of the tunnel vision effect, and further understanding of how selective processing affects choice. Several alternative operationalizations of the empirically tested debiasing procedure are discussed.     

Reasoning,biases and dual processes: The lasting impact of Wason (1960)

Wason (1960) published a relatively short experimental paper, in which he introduced the 2-4-6 problem as a test of inductive reasoning. This paper became one of the most highly cited to be published in the Quarterly Journal of Experimental Psychology and is significant for a number of reasons. First, the 2-4-6 task itself was ingenious and yielded evidence of error and bias in the intelligent participants who attempted it. Research on the 2-4-6 problem continues to the present day. More importantly, it was Wason's first paper on reasoning and one which made strong claims for bias and irrationality in a period dominated by rationalist writers like Piaget. It set in motion the study of cognitive biases in thinking and reasoning, well before the start of Tversky and Kahneman's famous heuristics and biases research programme. I also show here something for which Wason has received insufficient credit. It was Wason's work on this task and his later studies of his four card selection task that led to the first development of the dual process theory of reasoning which is so dominant in the current literature on the topic more than half a century later.     

Further attempts to clarify the importance of category variability for categorisation

The issue of how category variability affects classification of novel instances is an important one for assessing theories of categorisation, yet previous research cannot provide a compelling conclusion. In five experiments we reexamine some of the factors thought to affect participant performance. In Experiments 1 and 2, participants almost always classified the test item as belonging to the high variability category. By contrast, in Experiment 3 we employed an alternative experimental paradigm, where the difference in variability of the two categories was less salient. In that case, participants tended to classify a test item as belonging to the low variability category. Two additional experiments (4 and 5) explored in detail the differences between Experiments 1, 2 on the one hand, and 3 on the other. Some insight into the underlying psychological processes can be provided by computational models of categorisation, and we focus on the continuous version of Anderson's (1991) Rational Model, which has not been explored before in this context. The model predicts that test instances exactly halfway between the prototypes of two categories should be classified into the more variable category, consistent with the bulk of empirical findings. We also provided a comparison with a slightly reduced version of the Generalised Context Model (GCM) to show that its predictions are consistent with those from the Rational Model, for our stimulus sets.     

Inferring action structure and causal relationships in continuous sequences of human action

In the real world, causal variables do not come pre-identified or occur in isolation, but instead are embedded within a continuous temporal stream of events. A challenge faced by both human learners and machine learning algorithms is identifying subsequences that correspond to the appropriate variables for causal inference. A specific instance of this problem is action segmentation: dividing a sequence of observed behavior into meaningful actions, and determining which of those actions lead to effects in the world. Here we present a Bayesian analysis of how statistical and causal cues to segmentation should optimally be combined, as well as four experiments investigating human action segmentation and causal inference. We find that both people and our model are sensitive to statistical regularities and causal structure in continuous action, and are able to combine these sources of information in order to correctly infer both causal relationships and segmentation boundaries.     

Identifying Hypothesis Confirmation Behaviors in a Simulated Murder Investigation: Implications for Practice

Investigators have been shown to be prone to accessing information that confirms their preferred hypothesis. This tendency has been termed hypothesis confirmation. Hypothesis confirmation behaviour was explored in two studies using a computer administered simulation of a murder investigation. In Study 1, hypothesis confirmation behaviour did not differentiate successful and unsuccessful participants. However, unsuccessful participants stored more confirmatory information than non‐confirmatory information. Successful participants did not show this pattern. In Study 2, unsuccessful participants tended to settle early on a suspect and then acquired information that supported their decision. Successful participants tended to consider a number of suspects for longer and accessed a greater diversity of information. Taken together, these findings are highly suggestive of the role of information acquisition as an important component of hypothesis confirmatory behaviour, which is associated with unsuccessful information processing in the context of a simulated murder investigation. Implications for practice are presented. Copyright © 2012 John Wiley & Sons, Ltd.     

The “analysis of competing hypotheses” in intelligence analysis

The intelligence community uses “structured analytic techniques” to help analysts think critically and avoid cognitive bias. However, little evidence exists of how techniques are applied and whether they are effective. We examined the use of the analysis of competing hypotheses (ACH)—a technique designed to reduce “confirmation bias.” Fifty intelligence analysts were randomly assigned to use ACH or not when completing a hypothesis testing task that had probabilistic ground truth. Data on analysts' judgement processes and conclusions were collected using written protocols that were then coded for statistical analyses. We found that ACH‐trained analysts did not follow all of the steps of ACH. There was mixed evidence for ACH's ability to reduce confirmation bias, and we observed that ACH may increase judgement inconsistency and error. It may be prudent for the intelligence community to consider the conditions under which ACH would prove useful and to explore alternatives.     

Reducing tunnel vision with a pen‐and‐paper tool for the weighting of criminal evidence

In order to prevent tunnel vision, and ultimately miscarriages of justice, police, prosecutors, and judges must remain open to alternative scenarios in which the suspect is in fact innocent. However, it is not evident from the literature that people are sufficiently aware of how alternative scenarios should be employed in the decision‐making process. In the present research, participants read a case vignette and formed an impression of the suspect's guilt. Some participants were made familiar with an alternative scenario. Others were not only presented with an alternative scenario but were also instructed to score (with pen and paper) the extent to which every piece of evidence fitted in the primary and the alternative scenario. Findings suggest that this pen‐and‐paper task helped to reduce tunnel vision.     

Bayesian inference for finite mixtures of generalized linear models with random effects

We present an hierarchical Bayes approach to modeling parameter heterogeneity in generalized linear models. The model assumes that there are relevant subpopulations and that within each subpopulation the individual-level regression coefficients have a multivariate normal distribution. However, class membership is not known a priori, so the heterogeneity in the regression coefficients becomes a finite mixture of normal distributions. This approach combines the flexibility of semiparametric, latent class models that assume common parameters for each sub-population and the parsimony of random effects models that assume normal distributions for the regression parameters. The number of subpopulations is selected to maximize the posterior probability of the model being true. Simulations are presented which document the performance of the methodology for synthetic data with known heterogeneity and number of sub-populations. An application is presented concerning preferences for various aspects of personal computers.     

Stereotype threat may not impact women's inhibitory control or mathematical performance: Providing support for the null hypothesis

Underpinned by the findings of Jamieson and Harkins (2007; Experiment 3), the current study pits the mere effort motivational account of stereotype threat against a working memory interference account. In Experiment 1, females were primed with a negative self- or group stereotype pertaining to their visuospatial ability and completed an anti-saccade eye-tracking task. In Experiment 2 they were primed with a negative or positive group stereotype and completed an anti-saccade and mental arithmetic task. Findings indicate that stereotype threat did not significantly impair women's inhibitory control (Experiments 1 and 2) or mathematical performance (Experiment 2), with Bayesian analyses providing support for the null hypothesis. These findings are discussed in relation to potential moderating factors of stereotype threat, such as task difficulty and stereotype endorsement, as well as the possibility that effect sizes reported in the stereotype threat literature are inflated due to publication bias.