The underlying factors driving public transport expenditure requirements are complex and difficult to identify. Even academic literature is yet to reach consensus in some areas.

Public transport costs are determined by factors such as commuter volume, the properties of the transport network (such as the dominant modes of transport), and the geographical characteristics of the region itself, along with the interaction of all these factors. There is also the perception that policy-related factors, such as density goals and service quality, differ between States, thereby creating different cost structures. In Australia, regions differ significantly in terms of population distributions, economic activity patterns and topography, adding to the challenge of establishing a successful funding mechanism.

Finally, there is uncertainty in terms of the extent to which scale economies exist in the provision of public transport. That is, it is unclear whether, as network utilisation increases, public transport costs grow at a decreasing rate, or whether increased passenger volumes leads to costs growing at an increasing rate, implying higher congestion-related costs.

The CGC engaged us to identify a framework that can accommodate these various complexities whilst maintaining transparency. To ensure the robustness of this framework, any new supply and demand drivers that were identified needed to address the following concerns:

• Policy neutrality – it is vital to incorporate policy-related variables to remove the effect of policy factors on funding allocation
• Urban self-sufficiency – are labour markets of satellite cities integrated with their capital cities?
• Data availability – robust expenditure data was available for only 70 of Australia’s Significant Urban Areas (SUAs) and suitable proxies needed to be identified for incomplete explanatory variables
• Economies of scale – what is the best way to model this for SUAs of differing sizes?

The Solution

Firstly, Synergies developed a data assessment paper that outlined the strengths and weaknesses of candidate variables that could be included in the analysis.

This process ensured that:

• The characteristics of the SUAs included in the study were representative of those which were excluded on the basis of incomplete data
• We identified variables that were compatible with the demand, supply and cost factors that formed the analytical framework.

We also used spatial analysis to ascertain the extent to which satellite cities are self-sufficient from their nearest capital city. The degree of dependency was mapped using indices that considered the proportion of the population that works outside of their place of residence.

The outcomes from the data analysis informed the development of econometric models. At the centre of these models were the following considerations:

• Functional form – different specifications can be implemented to capture the diminishing or increasing expenditure arising from scale effects
• Correlation between variables – regressions can become unstable if too many similar factors are included.

We estimated a series of candidate equations and examined their merits on the basis of well-accepted criteria that included:

• Theoretical soundness and alignment with analytical framework
• Statistical selection criteria (such as R²)
• The predictive capabilities of the models (how well do they fit actual expense data?)
• Unbiasedness of results across States and Territories.

There was a strong emphasis on accessible visual analysis, which helped convey key issues to stakeholders. The incumbent model was used as a benchmark.

Our recommended model incorporated key demand, supply and cost variables that more accurately reflected actual expenditure patterns. These variables included:

• Density
• Journey to work distance
• Mean land slope (a measure of geographical complexity)
• Bus and train passengers (these were incorporated using a logarithmic form to capture scale effects).

The model was shown not to adversely favour any particular state over another, which was a central objective of the process. The final data used for the study represented over 96% of Australia’s population, which bolsters its future applicability.