People possess an abstract inferential rule system that is an intuitive version of the law of large numbers. Because the rule system is not tied to any particular content domain, it is possible to improve it by formal teaching techniques. We present four experiments that support this view. In Experiments 1 and 2, we taught subjects about the formal properties of the law of large numbers in brief training sessions in the laboratory and found that this increased both the frequency and the quality of statistical reasoning for a wide variety of problems of an everyday nature. In addition, we taught subjects about the rule by a "guided induction" technique, showing them how to use the rule to solve problems in particular domains. Learning from the examples was abstracted to such an extent that subjects showed just as much improvement on domains where the rule was not taught as on domains where it was. In Experiment 3, the ability to analyze an everyday problem with reference to the law of large numbers was shown to be much greater for those who had several years of training in statistics than for those who had less. Experiment 4 demonstrated that the beneficial effects of formal training in statistics may hold even when subjects are tested completely outside of the context of training. In general, these four experiments support a rather "formalist" theory of reasoning: People reason using very abstract rules, and their reasoning about a wide variety of content domains can be affects by direct manipulation of these abstract rules.
The CAUSE Research Group is supported in part by a member initiative grant from the American Statistical Association’s Section on Statistics and Data Science Education