A mode choice model for elasticities of panel data with inertia effects
In the present paper, a mode choice model is estimated for: car driver, car passenger, BTM (bus-tram-metro), train, cycling and walking. Two types of effects are modelled: inertia and latent effects. The estimated models show a significant inertia effect for all transport modes.
The largest inertia effect was found for the mode bicycle. The models also show that both travel time and travel cost are statistically significant for mode choice, with random taste variations of travel time across respondents. And, substantially different effects can be found across waves of the MPN data. Personal characteristics also play an important role, despite of the effects of the LOS variables. Perceptions and preferences are manifested as groups of respondents (PT lover, car dependent, etc.) in the MPN sample. Furthermore, the results show that panel effects are significantly relevant for modelling mode choice.
Both car users and cyclists are the noticeably inert travelers. While PT users show a lower tendency to maintain the usual choice (habit). PT users and pedestrians show similar inertia levels. Life events were significant in the presence of inertia effects, as longitudinal effects are considered. The inertia model shows smaller probabilities to travel by car. Therefore, ignoring inertia effects might lead to overestimations of car travelers. Furthermore, we calculate the sensitivity of the market share to changes in travel related variables. We calculated both baseline situation (without changes) and scenarios situations; with changes of 10% in travel costs by BTM, car and train, respectively. The results show that comparing the 3 scenarios of travel cost changes, car market share is the most elastic demand, while BTM is the least elastic. Regarding the elasticities per km, the analysis shown a clear tendency to change (increase or decrease) the distance travelled for both work and non-working trips according to car-trip cost. Non-working trips show substantially larger reductions in distance, as reasonable. Differences substantially change per year, and a more disperse pattern can be observed in 2015/2014 and 2014/2013; while the observations in 2016/2015 are more concentrated.