You write values questions people can agree or disagree to, and then you use some fancy statistical routines to be able to characterize who's in what group.