South East Wales Transport Model: Frequency model results
13 Sep 2018
This report documents the methodology and estimation results of the frequency model components of the South East Wales Transport Model (SEWTM).
Developing a New Transport Model for South East Wales
Researchers developed models capable of predicting levels of transport demand in South East Wales in response to a range of different policy options and in light of demographic changes.
They based the model on an existing one they had developed for the West Midlands, modifying it as needed based on local data. The resulting model closely predicts share of transport modes, though at times under-predicted car driver share.
Background
In 2018 the Welsh Government commissioned RAND Europe, together with Mott MacDonald and David Simmonds Consultancy, to develop a new transport model for South East Wales. The new model was required for the assessment of a range of transport policy options, including schemes to improve the public transport services across the region.
Goals
The aim of RAND Europe’s contribution to the project was to develop new models capable of predicting the levels of transport demand in South East Wales, in response to a range of different policy options and in the light of demographic changes. Separate models were designed in relation to different purposes for travel, as well as the various modes of transport.
Methods
The approach that RAND Europe followed in this study was to take the detailed transport models they had developed together with Mott MacDonald for the West Midlands region, and transfer them to a South East Wales context. This transfer approach allowed a detailed, policy responsive model to be developed for South East Wales, both at a lower cost and over a shorter timescale compared to developing a new model from scratch.
Findings
What travel frequency variations were identified?
- Variation in travel frequency largely depends on employment status, age, household size and car availability. There are also significant differences in average travel frequency rates which depend on travel purposes.
- Employers’ business travel was underreported in household interview data, one of the methods used to inform models.
How well did the models transfer to the South East Wales context?
- For four “home-based” travel purposes, the transfer process worked well, with no significant differences in scale identified between the West Midlands and South East Wales contexts. Only with the “commute” and “home–tertiary” travel purposes were the scales in the South East Wales context significantly lower.
- For non-home-based purposes, two of the scale parameters were 0.5 and 0.4, and the rest of the parameters were fixed to one, meaning that for those two purposes the models were less able to predict travel behaviour in the South East Wales context.
How well do the models replicate observed travel behaviour in South East Wales?
- Analysis demonstrated a close match between predicted model shares and those observed in local household interview data.
How do the model sensitivities compare to WebTAG realism test guidance?
- Compared to WebTAG realism test guidance, the sensitivities of the models to fuel cost changes were relatively low, while the car time elasticities and all but two of the public transport fare elasticity values were consistent.
How well does the implemented model make predictions?
- The share of transport modes predicted by the implemented model was on the whole close to those observed in local household interview data. For some purposes however, the car driver share was found to be under-predicted, owing to how the model predicts the distribution of persons across different segments.
- The implemented model also predicted reasonably well against the tour length information observed in the National Travel Survey for South East Wales, though there was some tendency to over-predict car passenger tour lengths and under-predict train tour lengths.
Recommendations
Household interview data is important for providing a representative evidence base for model estimation. We recommend a minimum of 4,000 household interviews be collected which include full destination information, to allow the ability to take account of variations in cost sensitivity in South East Wales. Interviewers should also be carefully briefed on the difference between commuter and employer business travel when household interview data is collected.
The approach to estimating frequency models for education purposes needs revisiting if there are any changes to the school leaving age.
We recommend that realism tests are undertaken to verify that the model elasticities are WebTAG compliant.
