Young Children Form Overhypotheses About Uncertain Outcomes That Guide Predictions in a Causal Reasoning Task

• In Event: 1-088 - Children’s Early Ability to Reason in the Face of Uncertainty

Integrative Statement

A large literature demonstrates that young children learn and generalize causal relations from probabilistic evidence (see Gopnik & Wellman, 2012 for a review). However, the ability to track statistical contingencies between a cause and its effect is distinct from the ability to learn and apply higher-order ‘overhypotheses’ about a cause’s variability—the extent to which it produces outcomes that are predictable or certain. Can young children infer causal variability from their observations?

Preschoolers can learn abstract overhypotheses about the form of likely causal relationships based on the evidence they observe, including conjunctive vs. disjunctive causes (Lucas et al., 2013), object-based vs. relational causes (Walker et al., 2016), and trait-based vs. situational causes (Seiver et al., 2013). Additionally, 24-month-olds can discriminate between probabilistic and deterministic causes (Waismayer & Meltzoff, 2016). However, it remains unknown whether young children can use information about causal variability to learn overhypotheses that predict new outcomes.

In Experiment 1, N=87 children aged two (N=29, Mage=30.7 months), three (N=29, Mage= 41.9 months), and four (N=29, Mage= 53.6 months) were presented with a causal reasoning task (Figure 1). First, an experimenter introduced Pete the Parrot, who likes to eat blue and yellow pebbles from a “bird-feeder” composed of four vertical tubes and corresponding transparent jars underneath. The experimenter demonstrated the effects of two buttons (yellow and blue). The deterministic button (counterbalanced) caused a same-colored pebble to fall out of the first tube three times in a row, one with each press. The probabilistic button caused the other color to fall out of three of the four tubes in a random order. Pebbles remained visible in the transparent jars under each tube. At test, children saw that Pete wanted to eat from the tube leading to the empty jar—where neither button had produced a pebble. The experimenter asked, “Which button should we try if we want to make sure that Pete gets fed?” If children have not formed overhypotheses about the variability of each cause (button), they should choose at chance. If they have, then they should choose the probabilistic button, which is more likely to produce the unknown outcome.

All age groups chose the probabilistic button more often than chance. Sixty-nine percent of 2-year-olds (SD=.47, p=.03), and 76% of both 3- and 4-year-olds (SD=.43, p=.003) chose the probabilistic button, with no significant difference between the age groups, p=.56 (ns). Results suggest that children as young as two can learn overhypotheses about causal variability, which can be used to guide their inferences about unknown outcomes. Ongoing work is investigating whether children can differentiate between a structured sequence vs. random variation. This second experiment uses a “probabilistic” button that makes pebbles fall out in a random order from tubes 1– 3; by contrast, the “deterministic” button goes in sequence: 1, 2, 3. If Pete asks to be fed from the fourth tube, children should prefer the deterministic button; however, if Pete asks to be fed from a new, fifth tube, then the probabilistic button is the better bet.

Authors

• Mariel Goddu, University of California, Berkeley
  Presenting Author

• Caren Walker, University of California, San Diego
  Non-Presenting Author