Children integrate prior knowledge with new statistical evidence in a Bayesian way, overturning initially likely hypotheses¶

Claim · All claims
Evidence · 1 study (2 entries) · theoretical r3 · q3
Currency · newest study 2012 · not yet checked against its sources · not reviewed by a person · overdue for a recheck since 2017: baselines and conditions change

Subclaims¶

q3 i? Children take baseline frequency (priors) into account in backward blocking, making different inferences when blickets are rare versus common. → Gopnik 2012 q3 i? Children override a strong contact prior when statistics indicate remote causation, and gradually shift cross-domain inferences with accumulating evidence as a Bayesian model predicts. → Gopnik 2012 (2)

Evidence¶

Gopnik 2012¶

Gopnik, A., & Wellman, H. M. (2012). Reconstructing constructivism: Causal models, Bayesian learning mechanisms and the theory theory. Psychological Bulletin, 138(6), 1085-1108. https://doi.org/10.1037/a0028044

q3 · i? · theoretical · r3

The review reports Sobel, Tenenbaum, and Gopnik (2004) and Griffiths et al. (2011): children's backward-blocking inferences differed with told baseline frequency of blickets, showing prior probabilities are combined with likelihoods in the Bayesian manner.

"If blickets were rare, for example, children were less likely to conclude that a block was a blicket than if blickets were common."

Gopnik 2012 (2)¶

Gopnik, A., & Wellman, H. M. (2012). Reconstructing constructivism: Causal models, Bayesian learning mechanisms and the theory theory. Psychological Bulletin, 138(6), 1085-1108. https://doi.org/10.1037/a0028044

q3 · i? · theoretical · r3

Kushnir and Gopnik (2007) showed children initially assumed blocks must touch the detector, but revised this given statistical evidence for remote activation. Relatedly, Schulz et al. (2007) found accumulating statistical evidence systematically shifted children's acceptance of cross-domain causes.

"When they saw that the machine was most likely to activate when an object was waved above it, rather than touching it, they concluded that contact was unnecessary."

Discussion¶

The same article argues that hierarchical causal models can represent this learning at two levels at once: specific causal theories and the more abstract framework theories they sit in.

Related Claims¶

• Preschoolers make probabilistic backward-blocking causal inferences, with capacities appearing by 18 months — related
• Children aged 2-4 infer causal structure from covariation and use it to design effective interventions — related
• Children extend statistical causal learning to psychological causes, inferring others' desires by 20 months — a narrower finding that bears on this claim
• Four-year-olds infer unobserved hidden causes when observed variables cannot explain the evidence