Reasoning About Alternative Outcomes Facilitates Anomaly Detection in Preschoolers

• In Event: 3-081 - The Emergence of Causal Reasoning and Causal Explanations in Childhood: Diverse Definitions and Methodologies

Integrative Statement

As learners, we frequently encounter evidence that is incompatible with our existing causal theories. While the presence of anomalous data plays a fundamental role in belief revision, learners often work to maintain their original hypotheses by ignoring, rejecting, or excluding anomalies (Chinn & Brewer, 1998). On the other hand, decades of empirical work indicate that even young children intuitively explore unexpected evidence, updating their existing hypotheses in light of new data (e.g., Gopnik, 2012). Here we explore whether drawing upon these intuitive causal reasoning abilities bolsters attention to the presence of conflicting evidence. In particular, we examine whether prompting preschool-aged children to explicitly generate alternative causal outcomes might facilitate anomaly detection, supporting their ability to differentiate candidate causes: one that accounts for all of the data they observe (no anomalies), and one that does not (anomalies observed).
Sixty-four 5-year-olds were randomly assigned to either control (n=32) or counterfactual (n=32) conditions, and introduced to a novel causal system in which some blocks made a machine light up while others did not. Each block had two painted sides, corresponding to two candidate causes. Over eight observations, one color (e.g., red) would correlate with the effect 100% of the time, while the other color (e.g., white) would correlate with the effect 75% of the time (Figure 1). Prior work has shown that children do not spontaneously detect the anomalous data in this context, and fail to privilege the 100% cause (Walker et al., 2016). However, given that counterfactual reasoning has been shown to prompt consideration of alternatives (e.g., Galinsky & Moskowitz, 2000), we assess whether generating counterfactuals (i.e., answers to “what if?” prompts) during learning might foster anomaly detection in this context. After each observation, children in the counterfactual condition were therefore asked to consider an alternative outcome (i.e., “What if the block had been blue/yellow? What would have happened to my machine?”), while those in the control condition simply reported what they actually observed. Afterwards, all participants were asked questions probing their understanding of each rule (no conflict questions). They were also asked conflict questions, which pit the 100% and 75% causes against one another. Critically, these conflict questions required that children notice the anomalous data in order to privilege the 100% cause.
As expected, all children successfully learned both candidate causes, as assessed by the no conflict questions, p<.01, with no difference between conditions, p=.76. However, for the conflict questions, only children in the counterfactual condition consistently privileged the 100% hypothesis (M=.81), p<.001, generalizing this newly learned cause to a novel case, p<.001. In line with previous work, children in the control condition selected between the two candidate causes at chance (M =.53), p=.13, with a significant difference between conditions, p<.03. These findings indicate that considering alternatives leads 5-year-olds to detect the presence of anomalous data, and privilege a hypothesis that accounts for more of their observations. Counterfactual prompts may therefore provide an important scaffold for supporting early scientific reasoning.

Authors

• Jae Engle, UCSD
  Presenting Author

• Hazel Baker-Harvey, UCSD
  Non-Presenting Author

• Caren Walker, University of California, San Diego
  Non-Presenting Author