Sample Attrition and Response Error: Do Two Wrongs Make a Right?

Bollinger, Christopher, and Martin H. David
Working paper no. 1995-15


Two important issues in the SIPP data are attrition and measurement error. Both of these issues have been previously studied (e.g. Marquis and Moore 1990); Bollinger and David (1993a,b); David (1989) and Zabel (1995)). However, the relationship between measurement error and sample attrition has not been previously studied. The work here examines the relationship between incidence of reporting error and incidence of sample attrition. We extend the work of Bollinger and David (1993b) by adding a variable representing the response pattern of the family unit in waves three through eight to a Probit model of reporting error for food stamp participation in the first wave. We find a strong and stable relationship between reporting error and non-response. The results support the “cooperator / non-cooperator” hypothesis and suggest that a latent variable is common to both reporting error and non-response. Further research is proposed.