Statistical Generalization

A statistical generalization infers a conclusion about the proportion of members of a subject kind have a certain predicate, from knowledge about the proportion of subject-members in a sample group that have the relevant predicate. For example:

Map 41:

To assess a statistical generalization, you must consider whether the sample is large and proportionally representative. All things being equal, a larger sample makes for a stronger generalization, but whether the sample is representative matters more than how large it is. To be representative, the sample has to range over all the differences there are among members of the subject kind which there is reason to think might make a difference to whether the predicate applies to them.

In a statistical generalization, the sample must not just be representative in the sense of including some subject-members which each of the characteristics that may be relevant to the predicate, it must include the same proportion of members with each of these characteristics as exists in the whole population of subject-members. For example, if there is reason to think that whether a dentist practices in an urban or rural area is relevant to whether he would recommend flossing, then it is not enough that one has both urban and rural dentists in one’s sample, the proportion of urban to rural dentists in one’s sample has to match the proportion in the general population of dentists.