Trip Generation Modelling: Theoretical Considerations and Practical Applications
Reza Tolouei, AECOM
The study investigates different methods of modelling trip generation and apply these to estimate partial effects of public transport accessibility and travel time on trip generation.
Trip generation model is a key part of the demand model, estimating total number of trips produced in each model zone as a function of individual, household, and zonal characteristics. Category analysis, the classical method of modelling trip generation, suffers from certain disadvantages including limitation on the number of categories to avoid low sample problems and lack of flexibility to investigate interaction effects of different factors. Use of statistical models provides the opportunity to estimate the isolated effects of a large number of explanatory variables and their interacting effects. Given the discrete nature and distribution of trips, Poisson and negative binomial regression models are the most commonly-used model forms.
Daly and Miller (2006) argue that trip generation model should be based on the utility theory, particularly when it is part of a wider transport model which includes other travel choices based on the utility theory. Multinomial and nested Logit models are most commonly-used examples of such models. Alternatively, Larsen (2003) proposes a Logit-Poisson model: a Logit model for the choice of whether to make a trip, and a Poisson model to estimate number of trips conditional on making a trip. The other potential issue associated with trip generation models is the inter-dependency between different journey purposes, which are usually modelled separately and treated as independent observations. According to Larsen (2003), people tend to combine trips with different purposes within a tour; hence, the correlation between journey purposes should be reflected in the trip generation models. He proposes a simultaneous model that includes a Logit-Poisson model for the total number of trips and a multinomial model for the distribution of them between purposes.
There is limited evidence on comparative performance of these models, especially between Poisson / Negative binomial models, Logit-Poisson models, and nested Logit models, in modelling trip generation. These modelling techniques are investigated in this study in terms of their suitability in developing a robust statistical procedure to estimate total number of trips, by purpose, as a function of various explanatory variables. The paper includes a comparison between the performance of the applied models and a discussion on practical considerations of using them as part of the wider transport model.
The other objective of this study is to estimate the isolated effects of different person, household, and area-specific factors on trip generation rates using a detailed household interview survey, undertaken as part of the development of a multi-modal transport model by AECOM. Explanatory analysis of data shows that trip rates tend to vary by factors such as age, gender, nationality, household size, income, working status, and car ownership. The transport model being developed is intended to be used to test a range of policies including provision of widespread public transport and various measures to reduce the severe congestion in peak periods. Therefore, apart from the isolated effects of the above variables, the trip generation model is required to be capable of providing answers to the following questions:
- How do trip generation rates vary by individuals’ accessibility to public transport?
- How is trip frequency influenced by changes in travel time and costs?
Therefore, it was hypothesised whether improved accessibility to public transport and reductions in travel times increases trip frequency, when all other factors are controlled for. These hypotheses were tested through explicit representation of these factors as explanatory variables in the model.
Building on previous developments in the area of trip generation modelling (e.g. Daly, 1997; Larsen, 2003; Daly and Miller, 2006), this study attempts to bring together the previous research and provide insights into comparative performance of proposed theoretical approaches, with a particular consideration given on practical aspects of applying these methods.
Daly, A.J. (1997). "Improved Methods for Trip Generation". In: PTRC European Transport Forum, PTRC.
Larsen, O. I. (2003). “Estimating Independent and Simultaneous Trip Frequency Models for all Travel Purposes with Combined Logit/Poisson”, proceedings of the European Transport Conference, Strasbourg.
Daly, A.J. and S.P. Miller (2006). “Advances in Modelling Traffic Generation”, proceedings of the European Transport Conference, Strasbourg.
Association for European Transport