The Timing of Change for Automobile Transactions: A Competing Risk Multispell Specification

Abstract

With the increasing number of panel data sets available in transport, the opportunity exists for the study of the time frame a household uses in making transport decisions. Panels collect data at regular points in time, and record information related to occurrences over the period since the last wave. Some panels record the precise time of an event during a panel wave. The opportunity to record event histories complete with identification of the states associated with a phenomenon such as automobile ownership together with the duration spent in each state provides powerful data for modelling the ‘when’ component of change through time. The ability to trace the timing of change and to model it will give transportation planners one missing element of forecasting - the timing of change. This paper develops a number of competing risk multispell models to obtain insights into the time spent in each of three states of automobile transactions (no change, replace used with used vehicle, and replace used with new vehicle), the factors which affect the probability of leaving a state, the probability of staying with a state, the effect of past history on current behaviour, and whether the population segments into distinct groups with different change probabilities. A data set of 12 years of annual observations (1974-1985) of a sample of Sydney households is used to illustrate the application of event history models.