Children’s causal interventions combine discrimination and confirmation

Like scientists, children have a sharp sense of when and how to seek evidence, but when it comes to generating causal interventions, their performance often falls short of normative information-theoretic metrics such as the expected information gain (EIG). We looked at whether this deviation resulted from mixing discriminatory strategies such as maximizing EIG with confirmatory strategies such as the positive test strategy (PTS). Thirty-nine 5- to -7-year-olds solved 6 puzzles where they had one opportunity to intervene on a three-node causal system to identify the correct structure from two possibilities. Children's intervention choices were better fit by a Bayesian model that incorporated EIG and PTS compared to alternative models that only considered a single strategy or selected interventions at random. Our findings suggest that children's intervention strategy may be a combination of discrimination and confirmation.