A Methodological Note on the Use of Response-Based Data, with an Application to Studying the Social Origins of Scientists

A crucial question facing today's sociologists is whether rare but sociologically interesting events should be excluded from statistical analysis just because of their rarity. This paper demonstrates, in a concrete way, why we should not give up sociological inquiries of rare events. Multivariate studies of rare events can be executed with existing statistical methods, and often by using existing data in innovative ways.

Two response-based samples are constructed from four independent but comparable existing national surveys in a sociological study of the social origins of scientists. The weighted maximum likelihood method is used for the estimation of logistic response models, with the dependent variable denoting whether or not the person is a scientist. It is shown that some ascribed characteristics, such as father's occupational status and place of childhood residence, affect the likelihood of becoming a scientist through education. Other ascribed characteristics, such as father's education, race, and father being a scientist, have direct effects net of education.