We make generalization based on limited data. How far we generalize is constrained by the goodness of the sample. A diverse sample is more informative than a narrow sample for drawing broader generalizations. For example, to find out the level of difficulty for all courses at Tsinghua University, it is more informative to generalize based on a sample of an intro level physics class and an advanced level literature class(diverse), than based on a sample of two intro level psychology classes (narrow).

Prior research showed that within the social domain, even 5-year-olds could evaluate that a diverse set of evidence is more generalizable than a narrow set of evidence. To find out about a toy preference of all children, 5-year-olds preferred to query the preferences of a diverse rather than a narrow group.

Here we are curious about children’s reasoning in various contexts of real-world situations.