Longitudinal Determinants of Attrition in CATI Panel Surveys

Panel surveys are a precious tool to analyse phenomena that change over time in many fields of human knowledge: a panel survey, in fact, provides the data necessary for analyses that relate variables measured at different points of time, and these analyses provide a number of advantages for the estimation of behavioural models. On the other side, repeated interviews with a panel sample raise problems not found in single or repeated cross-sectional designs. Most significantly, erstwhile respondents may not be reached by the interviewer in later waves or, if they are, they may not be willing to collaborate again. This produces a sample loss, usually called attrition, which cumulates wave by wave.Analytic problems caused by attrition are much more closely linked to the nature than to the amount of such attrition, since if truly random, attrition rate harms only the efficiency of estimates made from the panel data, while non-random attrition may impart unacceptable biases.The present contribution aims at describing the nature of attrition in a panel survey conducted at the University of Padua, which involves a sample of 2818 students who graduated between autumn 2000 and spring 2002. The survey is longitudinal, and graduates are contacted for the first time at the completion of their study career with a self-administered questionnaire, and then for 6 waves at 6 months intervals by CATI interviews. The basic question is whether the loss in empirical collaboration rate is “physiologic”, and so becomes a random sample, or, instead, depends on structural causes which reduce progressively the sample representativeness. Hence, we analyse the characteristics of respondents who could not be reached from some wave on, and the characteristics of the ones who, after participating in some wave, refused the interview and dropped the sample, compared with the ones of graduates who completed their collaboration to the survey.The empirical analysis shows that the attrition is, to some extent, selective, and depends both on objective characteristics that make the respondent more difficult to reach (working condition, marital status, etc.), and on the graduate’s satisfaction about his own position, which makes the respondent more proud about himself and his career, and hence more willing to participate in the survey.