Effects of Monotone and Nonmonotone Attrition on Parameter Estimates in Regression Models with Educational Data: Demographic Effects on Achievement, Aspirations, and Attitudes
Using the High School and Beyond longitudinal study, we investigate the participation patterns across four waves of data. Because nonrespondents from one wave are recontacted at subsequent waves, both monotone and nonmonotone attrition patterns arise. We discuss correlates of these two types of attrition in an attempt to describe individuals who may be at-risk of attrition. Gender and incomplete participation in the base-year (respondents who exhibit item nonresponse on key variables) are important predictors of later attrition. Estimated effects of monotone and nonmonotone attrition on parameter estimates in regression models suggest that certain demographic effects will be biased due to sample attrition. The evidence for bias is neither pervasive nor consistent, but suggests a systematic inflation of the Black-White achievement disparity.