An exploratory alternative approach for student non response weight adjustment

Large scale surveys in education have to face non-response issues that might bias the results. Non-response can occur at three levels: (i) a school refuses to participate, (ii) a sample student fails to participate and (iii) a participating student refuses to answer a particular question. Until now schools and student non-response have been counterbalanced by a non-response weight adjustment. This assumes that both the school and student non-respondents have similar characteristics to the school and student respondents, respectively, within classes. In this article results of analyses conducted on the Student Tracking Form data of the OECD/PISA 2000 survey are presented. The non-randomness of student absenteeism or refusal is demonstrated. Then, a simulation that compares the relative efficiency of the student weight adjustment with a multiple imputation method is presented where the superiority of the multiple imputation method, in particular for educational systems with a small school variance is shown. Finally, the multiple imputation method applied to some PISA 2000 countries identifies biases that become substantial in a longitudinal perspective.