Modelling correlation patterns in mode choice models estimated on multiday travel data

Understanding individual choices over time and measuring day-to-day variability in travel behaviour is important to capture the full range of travel behaviour. Although not very common, to date several multi-day travel surveys have been conducted and panel data is available to model different transport choices. However, determining the length of a panel that allows revealing variability in travel behaviour remains an open question. Also, no final agreement has been reached about modelling the various dimensions of correlation over the repeated observations. In this paper, we use the six-week panel data from the Mobidrive survey to estimate a mode choice model that accounts for correlation across individual observations over two time periods: all days of a single week and different days of the week (e.g. all Mondays) in the wave. We first analyse these effects separately, estimating different models for each type of correlation; then we try to disentangle the relative effects of each type of correlation, estimating both types jointly. We found that both types of correlation appeared highly significant when estimated alone, while only the correlation across a given day over the six-week period remained significant, when both types were estimated jointly. This implies that for the Mobidrive panel there is much less variability in mode choice across weeks than across the days of each week. It also suggests that one week could be an appropriate length for a panel to estimate modal choice and to correctly reveal day-to-day variability.