Encouraging Bayesian inference in a task that assesses 5- and 6-year-olds’ Backwards Blocking

Abstract

Causal reasoning is an important ability for learning how the world works. Yet, there is considerably less consensus among researchers and theorists about how children reason causally. Another issue concerns how children reason about multiple candidate causes; most studies on causal reasoning in children rely on two objects. Understanding how children reason about multiple candidate causes is important theoretically because it can provide greater insight into the processes that might support causal reasoning in the real world. One study that attempted to address both questions was Benton et al. (2023). While it is understood that there was a shift from Bayesian inference to associative learning as the information processing demands of the task increased (Benton et al., 2023), there’s still uncertainty regarding whether we can predispose them to rely on Bayesian inference under greater information-processing demands, given that researchers agree that Bayesian inference is a more rational strategy than associative learning? The aim of the current study is to investigate whether manipulating the base rate, which signifies the likelihood of an object being a cause and is a key parameter in standard Bayesian models, can promote Bayesian reasoning tendencies among 5-6-year-old children. In our experiment and computational models, we introduced the base rate prior to demonstrating the complex events described in Benton et al. (2023), aiming to assess whether this intervention can increase the probability of children deviating from their default associative learning strategies in a Backwards-Blocking task, a retrospectively reevaluating task. This study shows that while base rate information can shape children’s causal judgments, their retrospective reasoning under complex conditions is best explained by associative learning rather than Bayesian inference.