Development of weights for a choice-based panel survey sample with attrition

Rising costs associated with conducting surveys, coupled with the need to study infrequent choices, have contributed to a growing interest in and employment of choice-based sampling schemes. More recently, several panel surveys are being conducted in the United States and Europe to measure changes in travel behavior, with a sample much smaller than repeated cross-sectional surveys would require. It is natural to take advantage of these two survey methods and conduct a panel survey with choice-based sampling. In choice-based sampling schemes, where the sample entities are selected based on the endogeneous variable under study, the resulting sample will not be representative of the general population. The issue is more complicated with a panel survey because of several possibilities by which choice-based sampling can be conducted over time. This paper shows how appropriate weights can be developed and applied to observations from a choice-based sample so as to draw generalizable inferences. The study is the first attempt to develop weights for a choice-based transportation panel sample. Using a data set from the Puget Sound Region, transitions in discrete travel choice and longitudinal changes in demographic characteristics are inferred.