Sensitivity to Confounding in Causal Inference: From Childhood to Adulthood

A necessary condition for correctly assessing causality is the absence of confounding causes. This paper reports a pair of experiments that investigate whether people are sensitive to confounding when they infer causation. Two stories were constructed, one in which two candidate causes perfectly covaried with each other (confounded), and another in which the two candidate causes occurred independently of each other (unconfounded). In the confounded story, both causes covaried perfectly with an outcome; in the unconfounded story, only one of the two candidates covaried with the outcome. If people control for alternative causes while they evaluate a candidate cause, then subjects in the confounded condition should indicate that it is impossible to determine causality for either candidate alone, whereas those in the unconfounded condition should be able to judge that one of the candidates is causal and the other not. If people are not sensitive to confounding, however, subjects in the confounded condition should attribute causality to both candidates, and their judgments for these candidates should be the same as those for the target causal candidate in the unconfounded condition. Two experiments were conducted respectively with children and adults: Children received one or the other story, while adults received both. Both children and adults distinguished between confounded and unconfounded candidate causes when making attributions of causality. Our results show that children are able to state the indeterminacy of confounded candidate causes at an age much earlier than previously documented.