Probabilistic causal reasoning under time pressure

Abstract

Causal reasoning is a core facet of our cognitive abilities. However, the time-course of causal reasoning is not studied to its fullest. The duration of reasoning, including those that yield a reasoning error, might prove crucial in understanding the cognitive processes underlying causal reasoning. In two experiments we asked participants to make probabilistic causal inferences while manipulating external time pressure and measuring response times. We found that participants are less accurate under time pressure, which can be interpreted as a speed-accuracy-tradeoff. In addition, we found that conservative responding increased under time pressure. Surprisingly, we found that two other persistent reasoning errors - Markov violations and failures to explain away - appear insensitive to the time pressure we employed. These observations seemed related to the confidence that participants expressed in their causal judgment: Conservatism was associated with low confidence, whereas Markov violations and failures to explain were not. These findings shed doubt on existing explanations of causal reasoning that predict an association between time pressure and all reasoning errors. Our findings suggest that these errors should not be attributed to a single cognitive mechanism and emphasize the need for an understanding of causal judgements as the result of multiple processes.

Uncertainty can explain apparent mistakes in causal reasoning

“…Under a sampling account, giving participants more time to think should allow them to take more samples, which should decrease the size of their Markov violations (Davis & Rehder, 2020). In contrast to this prediction, manipulations of time pressure do not affect the magnitude of Markov violations in causal reasoning tasks, although they affect overall accuracy (Kolvoort et al, 2022;Rehder, 2014). Also, sampling-based accounts do not directly account for the effect of mechanistic information (Park & Sloman, 2013; 2014; see discussion above).…”

“…Figure 3 (panel C) shows that participants violated the Markov condition in the common cause structure in the Kolvoort et al (2022) data. Our model captures their qualitative pattern of judgments, which has also been found in the literature (Park & Sloman, 2013;Rehder, 2014;Mayrhofer & Waldmann, 2015;Rehder & Waldmann, 2017), although it does not perfectly capture the magnitude of the effect when the cause is present (i.e., Y = 1).…”

“…In what follows, we will use Eq. ( 4) to provide accounts of several purported normative violations: Markov in causal chains, Markov in common cause structures, explaining away in common effect structures, and conservativeness, by modeling data made publicly available by Kolvoort and colleagues (Kolvoort, Fisher, van Rooij, Schulz & van Maanen, 2022).…”

“…We test the predictions of our uncertainty-augmented model against data from a recent study. Kolvoort et al (2022) asked participants to make 27 types of inferences about each of 3 causal structures with binary variables: a common-cause, a common-effect, and a chain structure (see Figure 1). For each structure, participants first were given verbal and graphical descriptions of the causal model, and then learned the associated probabilities by viewing 32 representative samples generated from the model.…”

The Bayesian Mutation Sampler explains distributions of causal judgments

“…It proposes a sampling mechanism for how we generate causal judgments and has been successful in explaining average responses on a variety of tasks (Davis & Rehder, 2020). However, while accounts of average responses abound, the common observation of variability in causal judgments has received less attention and is often left unexplained (Davis & Rehder, 2020;Kolvoort et al, 2021Kolvoort et al, , 2022Rehder, 2014Rehder, , 2018Rottman & Hastie, 2016). This is an unfortunate gap in the literature, as variability in behavior can be informative of the cognitive mechanisms involved and so can help constrain the development of theories (e.g., as has been done in the domain of decision-making; Ratcliff, 1978).…”