International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
A Meta-Analysis of Transport Mode Choice Related Stated Preference Data in Several Metropolitan Areas in South Africa
Gary Patrick Hayes, Christo Venter

Last modified: 28 March 2017


As stipulated by the National Land Transport Act (Act No. 5 of 2009), all metropolitan municipalities in South Africa are obliged to prepare Comprehensive Integrated Transport Plans (CITP’s) every five years. To facilitate road and public transport planning and strategy formulation, metro-wide transportation models are developed for use as planning tools.

These models, mostly of the four-step type, require extensive data collection of socio-economic, demographic and trip related data. Since 2010, four metros (Tshwane (2010), Ekurhuleni (2013), Cape Town (2013) and Johannesburg (2014) have undertaken primary data collection in the form of stated preference surveys for the development of mode choice models. Both conjoint based models (Tshwane and Ekurhuleni) and discrete choice models (Cape Town and Johannesburg) were developed for mode choice simulation and the derivation of willingness to pay measures.

In addition, transportation models have recently been developed for the 25 Year Gauteng Integrated Transport Masterplan (GITMP25) (2013), the Tshwane Metropolitan Municipality (2014), and for the evaluation of the proposed extensions to the Gautrain Rapid Rail system (2015) based on a revision of the 2013 GITMP25 model. No primary data collection was undertaken for the development of the mode choice models in these studies, despite their outputs being used to motivate significant transport interventions. Rather, reliance was made on utility estimates and willingness to pay measures from older studies. This also in the context of previous efforts at forecasting Gautrain and BRT patronage and fare revenue having proved disappointing (Venter I. , 2014), (Venter I. , 2013).

That said, urban mode choice modelling in South Africa is complex. It is characterised by travellers with high-income differentials; several public transport modes (up to five in in Gauteng); significantly different public transport fare structures; uncertainty associated with mini-bus taxi routes and frequencies (despite this mode carrying about 67% of all public transport trips in South Africa (Africa, 2013)); and personal safety and security being of high concern to public transport users (Africa, 2013).

The paper first provides an insight into recent and current transport mode choice models in South Africa. The types of models used, their structure and utility attributes are presented. Willingness to pay measures such as the value of travel time savings (VTTS) are also presented and briefly discussed.

Then a meta-analysis of the four available SP data sets is presented for two types of discrete choice models, i.e. multinomial and mixed logit models. A total of approximately 3,000 questionnaires were completed for these surveys. Where possible, the data sets were segmented and analysed by income group and mode. The meta-analysis thus presents a useful insight into peak period urban traveller mode choice behaviour, and identifies gaps in the understanding of this behaviour for specific market segments. As an aside, a comparison of the statistical performance of the multinomial and mixed logit models is also shown. The mixed logit models provided a consistently better fit between the observed and modelled estimates of choice.

The results obtained from the new analyses varied significantly from the original findings in all cases. Meaningful insights were obtained into commuter behaviour and choice heterogeneity in all the metros, especially those in Gauteng, where it was possible to merge the Tshwane and Ekurhuleni data sets. The reasons for the variation were sometimes anticipated, for example the conjoint versus the DCM based estimates of utility. The advantages of discrete choice experiments over conjoint analysis are well documented (Louviere, Flynn, & Carson, 2010).

Deficiencies were found in the design and execution of the stated preference surveys. These deficiencies are discussed in some detail as some were not immediately obvious. In two cases they led to erroneous models being estimated and the distortion of the willingness to pay measures.

The new models revealed that there was consistency across all metros in regard the relative importance of the utility attribute coefficients, alternative specific constants and willingness to pay measures for the income groups and modes analysed, in particular, the value of travel time savings. This is an important finding, as it reveals consistent systemic behavioural factors for travellers across all the metros in the study. The findings also confirm the BRT user preferences highlighted by Venter in Johannesburg, i.e. lower than anticipated values of travel time savings,  (Venter C. , 2016), greater user preference for minimised BRT walk and wait times and fares, and less so for in-vehicle travel times.

95% confidence intervals were estimated for the VTTS (Armstrong, Garrido, & De Dios Ortuzar, 2001). The analyses revealed a wide variation in the ranges defined by the confidence intervals. The intervals with the narrower ranges were consistently derived from models with more significant estimators and larger sample sizes.

The paper ends with recommendations and guidelines in regard the design of stated preference surveys in South Africa urban areas and highlights potential pitfalls that were identified from the analyses. Suitable types of discrete choice models are also recommended, and the advantages and disadvantages of their application in the complex urban mode choice environment in South Africa are highlighted. Suggestions are also made for the application of more advanced DCM’s such as hybrid models.



Africa, S. S. (2013). National Household Travel Survey (Statistical Release P0320). Pretoria: Statistics South Africa.

Armstrong, P., Garrido, R., & De Dios Ortuzar, J. (2001). Confidence Intervals to Bound Values of Time. Transportation Research.

Louviere, J., Flynn, T., & Carson, R. (2010). Discrete Choice Analysis are not Conjoint Analysis. Journal of Choice Modelling, 57-72.

Venter, C. (2016). Are We Giving BRT Passengers what They Want? User Preference and Market Segmentation in Johannesburg. Southern Africa Transport Convention (SATC). Pretoria: Southern Africa Transport Convention.

Venter, I. (2013, February 8). Gautrain Passenger Number Tick Up but Below Initial Estimates. Engineering News.

Venter, I. (2014, July 11). Rea Vaya Battles Low Numbers. Engineering News.


Conference registration is required in order to view papers.