Toward a comprehensive hazard-based duration framework to accomodate nonresponse in panel surveys
Many surveys suffer from low response rates and therefore carry a risk of nonresponse bias. The problem is more severe in panel surveys because sample units are subject to nonresponse repeatedly. This dissertation is concerned with nonresponse in longitudinal household travel surveys. It identifies the likely sources of nonresponse and investigates a model-based bias correction procedure for the subject of interest in the survey -- trip frequency. Low response rates often lead to a sample representativeness problem and threaten the validity of the survey. A better understanding of the survey participation behavior can provide guidance for survey design to increase the response rates and to build an effective nonresponse bias correction procedure. It is generally believed that nonresponse is a combined result of social environment, survey attributes, and characteristics of sample units. In addition, state dependence and the lagged impact of exogenous variables can not be ignored when considering repeated responses in panel surveys. The first stage of this work considers the repeated participation in panel surveys as a duration process and proposes a hazard-based duration model for the analysis. The model structure accommodates state dependence and the lagged effects in a straightforward manner. Various factors, especially the indicators of survey burden, are incorporated in the model for a comprehensive understanding of the survey participation decision. The empirical analysis based on the seven-wave Puget Sound Transportation Panel suggests that survey burden, in general, is negatively associated with the survey participation duration. The results also reflect an interactive impact of survey burden and time constraints on the survey participation. The second stage of this work further investigates the relationship between the survey participation and trip frequency. The model formulation incorporates observed and unobserved heterogeneity in the participation and travel decisions. It is found that trip frequency, especially for home-based non- work trips, is endogenously correlated with the survey participation decision and the ignorance of this endogenous correlation leads to a biased estimate for the trip frequency and survey participation duration.