Probabilistic Causal Reasoning in Young Children

Probabilistic Causal Reasoning in Young Children

Paper presented at the International Conference on Infancy Studies, Minnesota, 2012

Authors: Anna Waisman [ & Andrew Meltzoff[ (University of Washington), Alison Gopnik [ (UC Berkeley)

Starting in infancy, children appear to learn about physical causes and effects from sparse evidence. While it is well known that older children are able to learn from observing both deterministic and probabilistic events, there is relatively little evidence that children younger than 4 years old are able to learn from probabilistic evidence. More recently, studies in infant probabilistic inference have shown that infants (12–14 months) will choose the more likely of two options based on knowledge of the sampling population (Denison, Xu) and that even younger infants (8-11 months) are sensitive to frequency information (Xu, Garcia 2008).Our work aims to combine the literature on causal and probabilistic inference, to investigate whether infants make causal inferences when observing probabilistic evidence.

In two experiments, we introduced an element of probability into a physical causal task in which 24-month-olds previously succeeded when presented with deterministic evidence.In experiment 1, we evaluated whether if given a choice between two probabilistic causes, toddlers would choose to act on the cause with a higher probability of producing an effect. In experiment 2, we controlled for frequency information to determine whether infants in experiment 1 were truly using the relative probabilities of the two possible causes (75% vs 25%), or were instead using the relative frequencies of the two possible causes (3 vs 1).

In experiment 1, 24-month-olds(M =24 months,SD =6 days) observed as two objects were placed on a machine. One of the objects when placed on the machine caused a marble to dispense from a marble dispenser in three out of four trials (75%) while the second object caused a marble to dispense in only one out of four trials (25%). At test, toddlers were given the chance to intervene on their own.Children chose to intervene on the higher probability object more than would have been expected by chance,p =.02. This experiment shows that 24-month-old children learn causal relations from observing probabilistically effective human interventions without the support of causal language describing the event or spatial contact between the cause and effect.

In experiment 2, 24-month-olds (M =24 months,SD =10 days) participated in the same procedure except that one of the objects caused the marble to dispense in three out of twelve trials (25%), while the second object caused the marble to dispense in three out of four trials (75%). Thus, the objects were equated for the frequency of trials on which marbles were dispensed, but one object continued to have a low probability of producing the effect. Preliminary evidence supports the conclusion that 24-month-olds do indeed use the probabilities to determine which object to intervene on rather than the relative frequency of the positive event.

These results along with prior work showing toddlers' learning from deterministic evidence demonstrate that toddlers are not only able to infer causal relationships by observing causal events but also that they use the evidence even in conditions of uncertainty to design their own interventions on the world.