Modeling user perception of transit quality using partial proportional odds models

It is necessary for the public transport operators to assess the perception of the quality of service of their users in order to propose improvements. However, many attributes can impact user satisfaction, hence a method to limit the number of variables that correlate with overall satisfaction is needed. Partial proportional odds model is proposed as an improvement over the ordered logit model used currently as the former does not depend on the proportional odds assumption. The analysis was performed in a survey of user satisfaction for the public rail transportation system in Costa Rica. The study demonstrated that the partial proportional odds model is significantly better than the ordered logit model for the data analyzed. Only 6 variables of the original 35 were significantly correlated with overall satisfaction: punctuality, ventilation inside the train, trip duration, smell inside the train, waiting time, and overall train condition. The models showed similar coefficients for the variables that met the parallel lines assumption. On the other hand, the coefficients for waiting time showed important differences between levels which demonstrated that a single coefficient was not appropriate for this variable. The results of the stated importance survey for the system analyzed are unworkable since almost all the variables resulted “very important” while only six variables in the model were correlated with overall satisfaction.