Thursday, 23 June 2011

Categorization and Selection Bias in the Study of Ethnic Groups

Presentation by Oliver Strijbis (University of St. Gallen) at the CO:STA Colloquium

Oliver Strijbis will present his work on the problem of sample selection bias: "Categorization and Selection Bias in the Study of Ethnic Groups". In his presentation, Oliver will propose a solution to the problem that his data on ethnic minority protest events might contain a biased sample of ethnic groups.

What is sample selection bias? When we think of selection bias, we tend think of a biased i.e. non-random selection into a treatment X. Educational attainment is correlated with mental ability and so the formers effect on earnings is cannot be inferred from a simple comparison. This is the well-known problem of endogeneity, omitted variable bias, unobserved heterogeneity or however it is called elsewhere in the universe of statistics language.

However, there is a another selection bias, namely that of a biased selection into the sample. In general, sample selection bias refers to problems where the dependent variable is observed only for a restricted, nonrandom sample. Reconsidering the example given above, we will only observe the earnings of those respondents who are part of the labor force and for these the relation between earnings and educational attainment might be different than for the population at large. The best known version of this bias is sampling on the dependent variable.

Winship, D. and Mare R. D. (1992): Models for Sample Selection Bias. Annual Review of Sociology 18, 327-350