International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Ofentse Hlulani Mokwena, Mark Zuidgeest

Last modified: 28 March 2017



South African universities are surrounded by formal and informal public transport services connecting students, staff and communities with the institutions. In order to connect students, university facilities and relevant activity centers, 6 of the 26 universities have official shuttle services exclusive for students. The lack of literature and practical knowledge about university student travel preferences in South Africa exposes a large gap in learner mobility—especially where learners pay for the service themselves. Furthermore, much of the stated preference research conducted on travel related themes in Africa have not investigated the use of behavioural theories in the choice models.


The gap in literature and practical knowledge on student travel preferences is even greater in the North West Province, where one university with three campuses attracts nearly 73 000 students from all over the country. In the capital city of the province, Mahikeng, pressure has emerged to improve the public transport services available: bus and minibus taxi. The city is a base for one of the three campuses of the North West University (NWU). This not only lays ground for investigating university student travel preferences in the Mahikeng area, but also exploring the application of behavioural theory in choice modelling.


This study reports on an unlabelled d-optimal stated preference survey of university students (N=121) conducted at the NWU Mafikeng Campus with a population of 11 842 students. Two hypotheses related to behavioural heterogeneity are tested. Firstly, students have unique compositions of behaviour influencing their intention to use bus and minibus taxi. Secondly, there are transport service preference differences between students who have positive and negative intent to use any public transport mode. Students were also asked to state which mode they would prefer between bus and minibus taxi.


It is assumed that preference manifests from psychological constructs that influence behaviour (observed through constructs). To test the hypotheses, this study connects the Theory of Planned Behaviour (TPB) with Hybrid Choice Modelling (HCM) framework. The TPB uses behavioural indicators to reveal a path dependency between behavioural constructs and their impact actual behaviour. HCM expands the behavioural framework of choice modelling to include: psychological indicators; indicators of history or dependence; behavioural heterogeneity; and latent constructs. In the HCM context, latent constructs have two interrelated functions and forms. Latent variables explain preference by contributing to the composition of choice. Latent classes segment preferences into class-specific preferences of choice. A latent class approach is applied in order to (a) segment students into behavioural classes for bus and taxi (class membership); and (b) estimate the mode preferences for each class.


The TPB postulates that action is closely observed through intention. The behavioural indicators are adopted from the TPB including attitude, perceived behavioural control and subjective norm are treated as determinants of intention. The likelihood that a student has a positive, negative, or neutral intention toward using a certain mode is the treated as their class membership probability. Class specific preferences and modal split(s) for minibus taxi and bus choice are also estimated in a Multinomial Logit form.

Two approaches were used to inform the sample size, based on discrete choice modelling literature, purporting that between 50 and 138 respondents are suitable (based on provincial modal splits, and survey design). The stated preference survey was constructed using the R statistical package. The choice models (membership and class specific) were estimated using BIOGEME.


In terms of class membership 55% of students have a negative intention to use bus that is influenced by: positive attitude (0.216), strong subjective normality (0.238) and their intention to use another mode (0.157). For minibus taxi, intention classes are nearly evenly spread between degrees of intent (33%) influenced by a strong negative attitude (-0.205), lower perception of normality of using minibus taxis (0.176) and intention to use another mode (0.18). This intimates that using minibus taxi is not as normal as using bus—whilst bus intention is exposed to greater intentions to use the alternative. Although, 93% of the students say they would prefer using taxi the non-class specific modal split is 79% and 21% for bus and taxi, respectively. Class specific modal split(s) indicate that 97% of students with positive intent to use public transport have a high value of time (R 4.70) and would choose taxi. In the negative intent class 81% of those with a negative intent would choose bus and a lower value of time (R 4.14). It is found that behavioural classes for university student travel preferences are unique between transport modes and across classes of positive, negative and neutral intention. The use of behavioural indicators in choice models seem to be a practical way to segment students and identify behavioural factors that may be used to attract or retain future and existing users, respectively.


This is especially true for policy interventions that nudge certain behaviours over others. For example, in Travel Demand Management this approach may enable the potential impact of voluntary interventions related to changing travel behaviour to be estimated. In the public transport marketing context—market segments that are more or less receptive of a type of advertorial may be identified and the likelihood of mode switching can be estimated. In terms of university student travel, behavioural segments are useful in targeting markets that have positive intentions to use public transport and retaining them through service and softer factors related to their behavioural segment. This study makes a unique contribution to stated preference research in Africa. Further research exploring and validating the application of approaches grounded on behavioural theory are necessary.


This study has a number of limitations. The sample size is suitable for the discrete choice models, but not for the TPB—if it were used independently: limiting the quality of tests related to consistency with TPB. Socio-demographic characteristics were excluded in the models because they were not statistically significant. Either implying that the differences between individuals were not significant in choice making, or that the survey did not capture relevant information.

Conference registration is required in order to view papers.