Perth tour-based mode choice model: Better planning and policy

The iMOVE project Tour-based mode choice model development has been completed. This work responds to the need for improved mode choice modelling tools that accurately represent how people plan and complete travel across entire tours, rather than isolated trips. The final report is available for download below.

Partners with iMOVE on the project were the Department of Transport (WA), Main Roads Western Australia, and the University of Western Australia.

Significance and objectives

Tour-based modelling is now widely recognised as best practice because it reflects how people actually organise their daily travel, not as isolated trips, but as linked movements that form complete tours. This provides a more realistic picture of travel behaviour, including intermediate stops, trip chaining, and return-home patterns.

This project was undertaken to support the Western Australian Department of Transport in updating its Strategic Transport Evaluation Model (STEM). The current STEM mode choice component relied on older data (2002-2006), and the availability of new Revealed Preference (RP) and Stated Preference (SP) data from the Perth Area Travel and Household Survey (PATHS, 2018-2021) created an opportunity to build more behaviourally sound and locally relevant models.

The study reviewed the PATHS RP and SP datasets to assess their usefulness for tour-based modelling and to understand how SP data might support updates to STEM. The literature also confirmed the value of tour-based approaches, while noting challenges such as Perth’s low multimodality and the added complexity of modelling full tours.

Based on these considerations, the project’s objectives were to:

• Assess the PATHS RP and SP datasets for their suitability in tour-based mode choice modelling;
• Review and validate the data processing steps and underlying assumptions;
• Develop clearer travel profiles, trip expositions, and tour structures for Perth;
• Examine modelling options for simple and complex tours, including conditional and dynamic discrete choice models; and
• Provide recommendations on how RP, SP, or a combination of both could be used to update STEM’s mode choice component.

Travel profiling

Travel profiling examined how Perth residents organise their daily travel, using detailed information from the PATHS (2018-2021). The analysis looked at how people structure their activities across an entire day; including the number of trips made, the order in which activities occur, where tours begin and end, and how different modes are used within the same day.

The profiling confirmed several key patterns in Perth’s travel behaviour:

• Motorised modes dominate, with private vehicles used in 81% of tours;
• Active travel forms a small but meaningful share, mainly walking (11%);
• Public transport use is limited, with bus (3%), train (2%), and taxi/rideshare (1%);
• Tour distances vary widely across modes, from 1 km walking tours to 51 km train tours; and
• Most tours are simple, with 67.6% involving a single purpose and very few (5.3%) being multimodal.

Exposition of trips

The analysis examined trips within the context of the full tours they were part of, rather than treating them as isolated movements. The PATHS data shows that most daily travel follows simple, purpose-driven patterns, with only a small share involving multiple modes or complex chaining.

Different modes also produce very distinct tour lengths and characteristics, highlighting how the structure and purpose of a tour influence overall travel behaviour.

Tours

The study examined how full tours; not just individual trips are structured and completed. Using PATHS data, tours were classified by their anchor points (such as home-based tours), the number and purpose of intermediate stops, and whether travellers used a single mode or switched modes during the day.

The findings confirmed that most tours in Perth are simple and motor-vehicle-based, with only a small share involving multiple modes or complex activity chains.

Network skims to generate revealed preference data

To support RP modelling, the project reviewed the network skim matrices previously developed for Perth’s STEM Model. These skims provide essential attributes, such as travel times, costs, distances, and public transport access and waiting components that describe the travel options available to individuals in the PATHS dataset.

The review examined how accurately and consistently these skim inputs reflected real-world travel conditions and identified areas where skim development could be improved so that the inputs better represent full-tour conditions rather than individual trip segments.

Revealed Preference (RP) mode choice

The analysis of RP data focused on how observed travel behaviour could inform tour-based modelling. RP evidence from the PATHS dataset provided behaviourally credible results, showing strong sensitivity to factors such as travel time, parking costs, transfers, and access/egress conditions.

However, the project also noted that RP modelling required imputing attributes for non-chosen alternatives; particularly travel times and costs which introduced potential uncertainty at the tour level. Despite this, RP data remained the most reliable foundation for understanding how Perth travellers make mode choices in real-world conditions

Stated Preference (SP) data

The project reviewed the SP dataset collected through PATHS to understand how travellers respond to hypothetical transport scenarios. SP data were useful for identifying the directional effects of key variables such as travel time, cost, reliability, and transfers.

However, the SP design had several limitations; including pivoting inconsistencies, unrealistic attribute ranges, and the removal of more than 20% of observations as outliers, which reduced its suitability for calibration and positioned it mainly as a supplementary source for sensitivity testing.

Choice modelling results for SP

SP-only models captured expected behavioural trade-offs; such as time versus cost and provided insight into how travellers might respond under improved or hypothetical transport conditions.

Yet the explanatory power of the SP models was modest, and several design issues weakened their reliability. The modelling showed that SP results were directionally reasonable, but could not match the behavioural realism produced by the RP models. This reinforced the need to prioritise RP data for model calibration, while using SP results only as supplementary indicators.

Joint RP-SP choice models

The project explored joint RP/SP estimation to combine the strengths of both datasets. Joint models improved behavioural clarity by incorporating observed choices alongside sensitivity tests from hypothetical scenarios, and they also highlighted state dependence effects; showing how past mode choices influence current decisions.

Joint models offered additional behavioural insight but required careful handling of scale factors and data alignment. Overall, RP models remained the most reliable basis for Perth’s operational modelling.

Operationalising mode choice models

The study outlined how the refined modelling insights could be applied in practice when updating STEM. Key considerations included:

• Prioritising RP-based models as the foundation for STEM calibration, given their behavioural realism in the Perth context;
• Avoiding donor-style model structures, which rely on parameters imported from other cities and were shown to misrepresent Perth’s travel behaviour;
• Representing tours and first-trip choices appropriately within model systems;
• Ensuring skim matrices accurately reflect full-tour conditions rather than single trips;
• Distinguishing between bus, rail, park-and-ride, and active modes based on observed behaviour; and
• Incorporating socio-demographic variables to capture traveller heterogeneity.

These steps outline how a practical, Perth-specific tour-based mode choice model can be developed for future STEM updates.

Conclusion and limitations

The project strengthened the tour-based modelling framework by clarifying how tours, trips, and traveller decisions should be represented using Perth’s RP and SP data. RP models provide the strongest basis for future STEM calibration, while SP data offer useful directional insights but cannot be used on their own.

Remaining limitations include small samples for complex tours, issues in SP design, and the need for better skim inputs. Donor models were also shown to misrepresent Perth’s behaviour and should not be used. Future work will focus on improving skim development, refining SP design, and building an operational Perth-specific tour-based model.

Download the final report

Download your copy of the final report, Estimating Simplified Main Tour Mode Using Revealed Preference and Stated Preference Data, by clicking the button below.

DOWNLOAD THE REPORT