International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Multimodal travel groups and preference mode for different travel purpose: a latent class cluster analysis
xiaochen ma, Oded Cats, Serge paul Hoogendoorn

Last modified: 28 March 2017


Multimodal travel groups and preference mode for different travel purpose: a latent class cluster analysis

Mode choice as a part of the travel demand models is essential for assessing policies designed to influence modal split. Most commonly it involves attracting more people to reduce car use. It is important to understand the travellers’ behavior to develop those policies. So it may be helpful to identify multimodal travellers, that is, travellers who make use of more than one mode of transport but not exclusively uses a single mode irrespective of context (e.g. Aarts et al., 1998). Kroesen (2014) found that multimodal users are more likely than single-mode users switch from one travel behaviour pattern to another. It is indeed necessary to identify multimodal traveller groups and understand the nature of the group in order to facilitate more such behaviour. Nobis (2007) found that multimodality is disproportionately high among adolescents, older people, and residents of population centres. Blumenberg and Pierce (2014) found that lower-income Americans are less multimodal than those with higher incomes. Molin and Mokhtaria (2016) identified (multi) modal groups by applying latent class analysis in which the indicators are mode use frequency. Mode perception and attitudes were included to predict who belongs to multimodal travel groups. Various researchers have examined that travellers who habitually use the same mode may have different attitude characteristic, perception on other modes. (See, Van Exel and Rietveld (2009). In specific, it is suggested that frequent car users who solely use car perceive a longer travel time towards PT transport than frequent car users who also travel by public transport (Diana and Mokhtarian, 2009a, 2009b). Vij et al. (2013) extended the study from main descriptive perspective to modal. They estimated a latent class choice model, the results show different multimodal style are related to long-term travel decisions and travel time sensitivity.

To the best of the authors’ knowledge, however, mode preferences of multimodal travellers have received little attention, partly because it is hard to obtain the data required for performing such an analysis. Diana and Mokhtarian (2009a) examined the nature of various multimodal clusters, but their analysis was limited to a few socioeconomic characteristics. They classed the users by their actual, perceived and desired mobility levels by different transport modes. It examined the relation between the actual use and desire use levels.  However there is not direct relates of the self-reported preference for different modes to the multimodal travel group analysis. And the approach in the their analysis is cluster analysis. The aim of this paper is to address this gap in the literature, in particular by exploring the relations between mode-related preference for different purpose trips and belonging to a particular travel behaviour group in addition to socio-demographic. An interesting question is whether travellers’ travel pattern is in line with their preference. Which group has the greatest misalignment with their preference and what are the reasons resulting in such an inconsistency.

In this paper, travellers will be classified based on their reported mode use frequency by using latent class cluster analysis (LCCA). The approach for distinguishing the different travel groups is similar to the approach taken by Molin (2016). An advantage of the LCCA over conventional cluster analysis is a model-based approach that assigns travellers to clusters probabilistically. This analysis includes two steps: First class the travellers into different groups according to the frequency of use of various transport modes. Second is a membership model, which predicts the probability of individuals belonging to each of the identified clusters based on personal characteristics such as socio-demographic, and preference mode for work and non-work trip. The dataset in the paper is panel data, collected by KiM in the Nederland, consisting of three days travel diary surveys of 2500 households (older than 12 years old) in 2013-2015. The respondents reported their frequency of using various modes, which varies from more than 4 days in a week to less than once in a year. Respondents self-reported frequency of one year though has less-detail than diary data, e.g., Nobis (2007), Buehler and Hamre (2015) and Kuhnimhof et al (2012) used one-week travel diary data, and Vij et al (2013) relied on six-week travel diary data, it still has advantages that our data has much longer time periods can be taken into account. Especially multimodality is defined as use of various transport modes in a certain time period (Nobis, 2007). So it is better to classify a traveler as a (non) multimodality based on a longer time period. And in our datasets the travel diary are also included for both work and non-work trips, it can be used as reference to check whether self-reported mode frequency are in line with their actual travel. In addition, the survey data contains information about socio-demographic characteristics of the respondents and the preferred mode for work and non-work trips of each respondent. The results of this study first can show the probability that an individual belongs to a particular class, as a function of the preference and characteristics of the travellers. Our work would be of interest to understand whether travellers’ preference for various transports modes influencing their mode choice and corresponding use frequency, or different multimodal segmentations with different mode use frequency has significant various modes preference.  Second, it is also expected that our result can improve the predictive accuracy and better describe travelers’ mode choice behavior. In specific, by using an approach where individuals are grouped using latent cluster analysis, individuals’ travel behavior estimation can be expected visibly improved. E.g. Ding (2016) proved that the individual grouping benefits the mode choice estimation.  The advantage of the model with grouping compared to that without grouping is that, it takes full account of the travelers’ pattern. This is important in analyzing travelers’ behavior using discrete choice model. The results will also help in practice policy makers to examine approaches to attract different user segments to use sustainable modes for different travel circumstances.

Conference registration is required in order to view papers.