Attrition in longitudinal studies: who do you lose?

Abstract Objective: To describe the risk factors for various types of attrition in three age cohorts of women in a longitudinal study and to discuss strategies to minimise attrition. Methods: Analysis of survey data from the Australian Longitudinal Study on Women's Health, collected by mailed questionnaire. In 1996, the study recruited and surveyed a national random sample of ‘younger’ (18–23 years, n=14,247), ‘mid‐age’ (45–50 years, n=13,716), and ‘older’ women (70–75 years, n=12,432), and began a staggered cycle of mailed follow‐up questionnaires: 1998 (mid‐age), 1999 (older), 2000 (younger) and so on. Demographic, health and social risk factors for attrition were examined using multivariate analysis. Results: Attrition at survey 2 was highest among younger women (32%), mainly because of participants not being contactable (21%), and lower among the older (16%) and mid‐age women (10%). At survey 1, the survey 2 non‐respondents were more likely to report having less education, being born in a non‐English‐speaking country and being a current smoker, in all cohorts, and had poorer health (mid‐age and older cohort) and more diffculty managing on their income (younger and mid‐age). Conclusion: Although the magnitude of different types of attrition was found to differ by age, there were several risk factors for attrition that remained consistent. These fndings are important to inform future studies on ways to lessen or prevent systematic loss of participants. Implications: Recruitment and follow‐up methods in longitudinal studies should be tailored to maximise retention of participants at higher risk of dropout.