Sample Selection Bias in Models of Commuting Time

Abstract

This research conceptualises, measures and evaluates the effects of sample selection bias on models of commuting time. Data are drawn from the Public Use Microdata Sample of the 1990 US Census for the Boston metropolitan area. The major finding of the analysis is that the process that determines entry into employment introduces sample selection bias into the estimates of commuting-time models. The degree of sample selection bias observed differs by race/ethnicity and gender on such key variables as marital and parental status and reliance on public transport, because the influence of these variables on employment differs by race/ethnicity and gender. These variables are important for evaluating both the spatial mismatch and the spatial entrapment hypotheses and therefore the contribution of previous analyses should be reconsidered.