Cognitive biases in human causal learning

The main aim of this work was to look for cognitive biases in human inference of causal relationships in order to emphasize the psychological processes that modulate causal learning. From the effect of the judgment frequency, this work presents subsequent research on cue competition (overshadowing, blocking, and super-conditioning effects) showing that the strength of prior beliefs and new evidence based upon covariation computation contributes additively to predict causal judgments, whereas the balance between the reliability of both, beliefs and covariation knowledge, modulates their relative weight. New findings also showed "inattentional blindness" for negative or preventative causal relationships but not for positive or generative ones, due to failure in codifying and retrieving the necessary information for its computation. Overall results unveil the need of three hierarchical levels of a whole architecture for human causal learning: the lower one, responsible for codifying the events during the task; the second one, computing the retrieved information; finally, the higher level, integrating this evidence with previous causal knowledge. In summary, whereas current theoretical frameworks on causal inference and decision-making usually focused either on causal beliefs or covariation information, the present work shows how both are required to be able to explain the complexity and flexibility involved in human causal learning.     

A probabilistic contrast model of causal induction

Deviations from the predictions of covariational models of causal attribution have often been reported in the literature. These include a bias against using consensus information, a bias toward attributing effects to a person, and a tendency to make a variety of unpredicted conjunctive attributions. It is contended that these deviations, rather than representing irrational biases, could be due to (a) unspecified information over which causal inferences are computed and (b) the questionable normativeness of the models against which these deviations have been measured. A probabilistic extension of Kelley's analysis-of-variance analogy is proposed. An experiment was performed to assess the above biases and evaluate the proposed model against competing ones. The results indicate that the inference process is unbiased.     

Causes versus enabling conditions.

People distinguish between a cause (e.g., a malfunctioning component in an airplane causing it to crash) and a condition (e.g., gravity) that merely enables the cause to yield its effect. This distinction cannot be explained by accounts of reasoning formulated purely in terms of necessity and sufficiency, because causes and enabling conditions hold the same logical relationship to the effect in those terms. Proposals to account for this apparent deviation from accounts based on necessity and sufficiency may be classified into three types. One approach explains the distinction in terms of an inferential rule based on the normality of the potential causal factors. Another approach explains the distinction in terms of the conversational principle of being informative to the inquirer given assumptions about his or her state of knowledge. The present paper evaluates variants of these two approaches, and presents our probabilistic contrast model, which takes a third approach. This approach explains the distinction between causes and enabling conditions by the covariation between potential causes and the effect in question over a focal set--a set of events implied by the context. Covariation is defined probabilistically, with necessity and sufficiency as extreme cases of the components defining contrasts. We report two experiments testing our model against variants of the normality and conversational views.     

经验信息与共变信息对判断两原因共同作用的影响

本研究考察了被试在使用经验信息和共变信息进行两原因共同作用因果判断时所具有的特点。研究结果表明:(1)两原因的性质对于被试的选择有着显著影响。当两原因的可信度相等时,更多被试认为两原因都可以引起结果。(2)两原因都不出现时结果出现的概率P(e/~i~j)对于被试的选择没有显著影响。(3)两原因单独出现时结果出现的概率对于被试的选择有着显著的影响,但是这一变量是与两原因的可信程度共同产生影响的。     

Expectations and interpretations during causal learning

In existing models of causal induction, 4 types of covariation information (i.e., presence/absence of an event followed by presence/absence of another event) always exert identical influences on causal strength judgments (e.g., joint presence of events always suggests a generative causal relationship). In contrast, we suggest that, due to expectations developed during causal learning, learners give varied interpretations to covariation information as it is encountered and that these interpretations influence the resulting causal beliefs. In Experiments 1A-1C, participants' interpretations of observations during a causal learning task were dynamic, expectation based, and, furthermore, strongly tied to subsequent causal judgments. Experiment 2 demonstrated that adding trials of joint absence or joint presence of events, whose roles have been traditionally interpreted as increasing causal strengths, could result in decreased overall causal judgments and that adding trials where one event occurs in the absence of another, whose roles have been traditionally interpreted as decreasing causal strengths, could result in increased overall causal judgments. We discuss implications for traditional models of causal learning and how a more top-down approach (e.g., Bayesian) would be more compatible with the current findings.     

Distinguishing genuine from spurious causes: a coherence hypothesis

Two opposing views have been proposed to explain how people distinguish genuine causes from spurious ones: the power view and the covariational view. This paper notes two phenomena that challenge both views. First, even when 1) there is no innate specific causal knowledge about a regularity (so that the power view does not apply) and 2) covariation cannot be computed while controlling for alternative causes (so that the covariation view should not apply), people are still able to systematically judge whether a regularity is causal. Second, when an alternative cause explains the effect, a spurious cause is judged to be spurious with greater confidence than otherwise (in both cases, no causal mechanism underlies the spurious cause). To fill the gap left by the traditional views, this paper proposes a new integration of these views. According to the coherence hypothesis, although a genuine cause and a spurious one may both covary with an effect in a way that does not imply causality at some level of abstraction, the categories to which these candidate causes belong covary with the effect differently at a more abstract level: one covariation implies causality; the other does not. Given this superordinate knowledge, the causal judgments of a reasoner who seeks to explain as much as possible with as few causal rules as possible will exhibit the properties that challenge the traditional views. Two experiments tested and supported the coherence hypothesis. Both experiments involved candidate causes that covary with an effect without implying causality at some level, manipulating whether covariation that implies causality has been acquired at a more abstract level. The experiments differed on whether an alternative cause explains the effect.     

Driven by power? Probe question and presentation format effects on causal judgment

It has been proposed that causal power (defined as the probability with which a candidate cause would produce an effect in the absence of any other background causes) can be intuitively computed from cause-effect covariation information. Estimation of power is assumed to require a special type of counterfactual probe question, worded to remove potential sources of ambiguity. The present study analyzes the adequacy of such questions to evoke normative causal power estimation. The authors report that judgments to counterfactual probes do not conform to causal power and that they strongly depend on both the probe question wording and the way that covariation information is presented. The data are parsimoniously accounted for by an alternative model of causal judgment, the Evidence Integration rule.     

Cause and covariate: Development of two related concepts

This study used a rule-analytic technique to investigate the role of event covariation in causal judgment. Junior high school and college subjects were shown information about the co-occurrences of two potentially related events and were asked to make either causal or covariation judgments about the two events. Subjects often failed to identify covariates as causes or identified as causes events which were either unrelated or related in the opposite direction to the event to be explained. Rule analyses indicated that use of mathematically flawed strategies resulted in erroneous covariation and causal judgments. Comparisons between the junior high and college samples showed parallel improvement with increasing age for the two judgments. Strategy analyses of the covariation and causal judgments showed that males defined causes and covariates by similar rules, but that females used different rules to make the two judgments.     

On the origin of personal causal theories

Detecting the causal relations among environmental events is an important facet of learning. Certain variables have been identified which influence both human causal attribution and animal learning: temporal priority, temporal and spatial contiguity, covariation and contingency, and prior experience. Recent research has continued to find distinct commonalities between the influence these variables have in the two domains, supporting a neo-Humean analysis of the origins of personal causal theories. The cues to causality determine which event relationships will be judged as causal; personal causal theories emerge as a result of these judgments and in turn affect future attributions. An examination of animal learning research motivates further extensions of the analogy. Researchers are encouraged to study real-time causal attributions, to study additional methodological analogies to conditioning paradigms, and to develop rich learning accounts of the acquisition of causal theories.     

单结果多原因情况下的因果判断研究

本研究以直接呈现每种因果类型的频次的方式考察单结果多原因情况下影响因果判断的因素,同时检验概率对照模型,效力PC理论和因果模型理论。结果发现：（1）影响因果判断的因素有：事件原因与促进条件的性质差异,原因的熟悉度,原因和结果的共变程度;（2）同题抽象性对因果判断没有影响;（3）概率对照模型和因果模型理论在一定的情况下适用,但是都不能解释所有的情况。     

Is inferential reasoning just probabilistic reasoning in disguise?

Oaksford, Chater, and Larkin (2000) have suggested that people actually use everyday probabilistic reasoning when making deductive inferences. In two studies, we explicitly compared probabilistic and deductive reasoning with identical if-then conditional premises with concrete content. In the first, adults were given causal premises with one strongly associated antecedent and were asked to make standard deductive inferences or to judge the probabilities of conclusions. In the second, reasoners were given scenarios presenting a causal relation with zero to three potential alternative antecedents. The participants responded to each set of problems under both deductive and probabilistic instructions. The results show that deductive and probabilistic inferences are not isomorphic. Probabilistic inferences can model deductive responses only using a limited, very high threshold model, which is equivalent to a simple retrieval model. These results provide a clearer understanding of the relations between probabilistic and deductive inferences and the limitations of trying to consider these two forms of inference as having a single underlying process.     

The knowledge-structure and inductivist strategies in causal attribution: A direct comparison

Two different strategies for making causal attributions are distinguished. The first is the classic inductivist approach, which uses covariation information to arrive at causal attributions. The second is the knowledge-structure approach, which uses information relevant to knowledge about plans and goals to explain behaviour. Two experiments are reported in which information activating both types of strategy is given. The results indicate that goal-relevant information activates expectancies that resist the presence of explicit covariation information. The results are interpreted as indicating that expectancies generated by knowledge-structures are therefore different to those activated by verbs, which do not resist the effect of explicit covariation information. It is concluded that knowledge-structures constitute an alternative strategy of causal attribution to the inductivist strategy, and the nature of the relationship between the two strategies is considered.     

The Role of Probability and Intentionality in Preschoolers' Causal Generalizations

Three experiments examined whether preschoolers recognize that the causal properties of objects generalize to new members of the same set given either deterministic or probabilistic data. Experiment 1 found that 3- and 4-year-olds were able to make such a generalization given deterministic data but were at chance when they observed probabilistic information. Five-year-olds reliably generalized in both situations. Experiment 2 found that 4-year-olds could make some probabilistic inferences, particularly when comparing sets that had no efficacy with sets in which some members had efficacy. Children had some difficulty discriminating between completely effective sets and stochastic ones. Experiment 3 examined whether 3- and 4-year-olds could reason about probabilistic data when provided with information about the experimenter's beliefs about causal outcomes. Children who were more successful on standard false-belief measures were more likely to respond as if the data were deterministic. These data suggest that children's probabilistic inferences develop into early elementary school, but preschoolers might have some understanding of probability when reasoning about causal generalization.     

Time as a guide to cause

How do people learn causal structure? In 2 studies, the authors investigated the interplay between temporal-order, intervention, and covariational cues. In Study 1, temporal order overrode covariation information, leading to spurious causal inferences when the temporal cues were misleading. In Study 2, both temporal order and intervention contributed to accurate causal inference well beyond that achievable through covariational data alone. Together, the studies show that people use both temporal-order and interventional cues to infer causal structure and that these cues dominate the available statistical information. A hypothesis-driven account of learning is endorsed, whereby people use cues such as temporal order to generate initial models and then test these models against the incoming covariational data.     

因果力比较范式下对效力PC理论的检验

在用图形方式集中呈现信息的条件下,用因果力大小比较的实验范式检验效力PC理论。233名大学生被试对不同化学药物影响动物基因变异的能力做大小比较判断。结果发现,对单一因果关系因果力大小的比较判断具有以下3个特点：（1）不对称性：在预防原因条件下的因果力判断一般符合效力PC理论,而在产生原因条件下的因果力判断一般不符合效力PC理论;（2）在同时变化协变值DP和结果基率P(E|~C)的产生原因条件下,多数被试使用DP规则。这与概率对比模型的预测相一致,而不支持效力PC理论;（3）在固定协变值DP而只变化结果基率P(E|~C)的产生原因条件下,多数被试使用变异比RP规则。这是目前所有的因果推理理论都不能解释的现象。     

Exemplar models as a mechanism for performing Bayesian inference

Probabilistic models have recently received much attention as accounts of human cognition. However, most research in which probabilistic models have been used has been focused on formulating the abstract problems behind cognitive tasks and their optimal solutions, rather than on mechanisms that could implement these solutions. Exemplar models are a successful class of psychological process models in which an inventory of stored examples is used to solve problems such as identification, categorization, and function learning. We show that exemplar models can be used to perform a sophisticated form of Monte Carlo approximation known as importance sampling and thus provide a way to perform approximate Bayesian inference. Simulations of Bayesian inference in speech perception, generalization along a single dimension, making predictions about everyday events, concept learning, and reconstruction from memory show that exemplar models can often account for human performance with only a few exemplars, for both simple and relatively complex prior distributions. These results suggest that exemplar models provide a possible mechanism for implementing at least some forms of Bayesian inference.     

Betting on transitivity in probabilistic causal chains

Causal reasoning is crucial to people’s decision making in probabilistic environments. It may rely directly on data about covariation between variables (correspondence) or on inferences based on reasonable constraints if larger causal models are constructed based on local relations (coherence). For causal chains an often assumed constraint is transitivity. For probabilistic causal relations, mismatches between such transitive inferences and direct empirical evidence may lead to distortions of empirical evidence. Previous work has shown that people may use the generative local causal relations A → B and B → C to infer a positive indirect relation between events A and C, despite data showing that these events are actually independent (von Sydow et al. in Proceedings of the thirty-first annual conference of the cognitive science society. Cognitive Science Society, Austin, 2009, Proceedings of the 32nd annual conference of the cognitive science society. Cognitive Science Society, Austin, 2010, Mem Cogn 44(3):469–487, 2016). Here we used a sequential learning scenario to investigate how transitive reasoning in intransitive situations with negatively related distal events may relate to betting behavior. In three experiments participants bet as if they were influenced by a transitivity assumption, even when the data strongly contradicted transitivity.     

BUCKLE: a model of unobserved cause learning

Dealing with alternative causes is necessary to avoid making inaccurate causal inferences from covariation data. However, information about alternative causes is frequently unavailable, rendering them unobserved. The current article reviews the way in which current learning models deal, or could deal, with unobserved causes. A new model of causal learning, BUCKLE (bidirectional unobserved cause learning) extends existing models of causal learning by dynamically inferring information about unobserved, alternative causes. During the course of causal learning, BUCKLE continually computes the probability that an unobserved cause is present during a given observation and then uses the results of these inferences to learn the causal strengths of the unobserved as well as observed causes. The current results demonstrate that BUCKLE provides a better explanation of people's causal learning than the existing models.