Issues in Estimation and Application of Long Distance Travel Demand Models

Sammanfattning: Forecasts of the demand for long distance travel areessential for many decisions, such as infrastructureinvestments, operators?management policies etc. Manyresearch issues are related to this field, and the purpose ofthis thesis is to contribute to some of them. Paper I Modellingchoice of flight and booking class?a study using StatedPreference and Revealed Preference data goes into theinteresting issue of modelling departure time and ticket typechoices. The data used in the estimation process were acombination of revealed preference (RP) data and statedpreference (SP) data. A model was developed to estimaterecapture and buyup to improve the SAS (Scandinavian AirlinesSystem) yield management system. Paper II Endogenoussegmentation applied to long distance business trips in Swedenis related to the possibilities of identifying market segmentsthat differ in travel behaviour. The procedure of endogenoussegmentation has been explored to identify different tastesegments in a population. The results indicate that the methodis an efficient way of identifying market segments andoutperforms the traditional manual segmentation when modelefficiency is the objective.The restrictive assumption of independent distributed randomcomponents in the logit model utility function can be relaxedby finding a suitable nesting structure. The task ofestablishing such structures, another crucial issue in traveldemand model development, is further analysed in papers IIIExploring the HEV model to improve nesting structures of modelsfor Swedish long distance private trips and paper IV Enhancingmodel structure and treatment of incomplete geocodes in SAMPERSlong distance models. In paper III a technique to identifyscale parameters for separate alternatives is used. Based onthis, different nesting structures are tested and the resultsare compared with the old mode choice model for long distancetrips in Sweden. This paper leads to paper IV, where similarnesting structures were tested on the long distance mode anddestination choice model in the SAMPERS system. Paper IV alsodeals with data problems, as the destination coding wasincomplete. A final model formulation, related to these issuesis suggested. To turn the view a bit beside the pure estimationtechniques and structures used in the estimation of differentmodels, paper V Quantifying uncertainties in the SAMPERS longdistance forecasting model system considers uncertaintiesrelated to the fact that models are estimated on a sample of apopulation. The bootstrap method, a computer intensivestatistical method, can be used to compute statistical measuresfor very complex systems, without being bound to an analyticalapproach. The bootstrap method is applied on the SAMPERSsystem, and numerical results are presented on different modelsystem output levels.