Attrition Bias in Panel Data: A Sheep in Wolf’s Clothing? A Case Study Based on the MABEL Survey

Melbourne Institute Working Paper No. 14/14

Date: 2014


Terence C. Cheng
Pravin K. Trivedi


This paper investigates the nature and consequences of sample attrition in a unique longitudinal survey of medical doctors. We describe the patterns of non-response and examine if attrition affects the econometric analysis of medical labour market outcomes using the estimation of physician earnings equations as a case study. We compare the econometric estimates obtained from a number of different modeling strategies: balanced versus unbalanced samples; an attrition model for panel data based on the classic sample selection model; and a recently developed copula-based selection model. Descriptive evidence shows that doctors who work longer hours, have lower years of experience, are overseas trained, and have changed their work location are more likely to drop out. Our analysis suggests that the impact of attrition on inference about earnings of General Practitioners is small. For specialists, the impact of attrition is statistically and economically significant, but is on the whole not very large. Finally we discuss how the top-up samples in the MABEL survey can be used to address the problem of panel attrition.