Survey non-response in relation to ability and family background: structure and effects on estimated earnings functions

Non-response in a 1983 re-interview of a 1952 random sample of Dutch sixth-graders is analysed with a probit model, relating non-response to family background, ability and school-related variables. Family background appears irrelevant, while ability and class failures have the expected effect. Differences between procedures for males and females (a follow-up interview on the mail survey for male non-respondents only) have no effect on structural relations. The selectivity bias term inserted in a simple wage equation has no significant effect.