Studies of correlated willingness-to-pay indicators for public transport

摘要

Much research on the estimation of willingness-to-pay measures is based on mixed logit models. In recent studies, the importance of testing the distributional assumptions in these models has been emphasized. It has also been shown that models with several mixing dimensions should allow for correlation. This paper investigates how to specify models allowing for correlation between different willingness-to-pay measures. The purpose is to capture correlation by a parsimonious specification for several public transport modes. For each mode, the distributional assumptions are tested in the final models. We find significant correlation between willingnessto- pay measures for all modes. For each model we develop a model that cannot be rejected by the most general model allowing for correlation. A main contributor to the correlation is a random scale which can be partly explained by background variables. The correlation is seen to change the evaluation of willingness to pay for some modes. This highlights the importance of finding a suitable representation for correlation. The tests of the distributional assumptions give valuable information on how to improve the models. The principal conclusion is that correlation structures should be included in modelling but that the most reasonable structure might be neither the simplest nor the most complex.