Modelling Propensity to Cycle As a Policy Tool



Modelling Propensity to Cycle As a Policy Tool

Authors

Neil Raha, SYSTRA Ltd, David Christie, Transport for London, Andrew Gordon, Mott MacDonald

Description

Cycling has been growing at a rapid rate in London. A new method has been developed to identify a segment of the population which has a high propensity to cycle. Future growth in cycle share can be modelled by increasing that population fraction.

Abstract

There has been rapid growth in cycle usage in London in recent years and the Mayor’s Transport Strategy calls for continued growth to achieve a significant mode share. This has been driven historically both by the availability of cycle facilities (such as the cycle hire scheme), infrastructure improvements, and changes in the perceived attractiveness of cycling as a mode. Most of the impacts of these drivers of cycle demand are not easily captured in generalised cost changes used as input to the Transport for London (TfL) strategic modelling suite, so TfL has been using an exogenously derived forecast for cycling to grow the cycling matrix for future years.

TfL recently appointed a consortium led by Jacobs supported by SYSTRA, Mott MacDonald and RAND Europe to undertake the development of a New Demand Model (NDM) within its major urban four-stage multimodal London Transportation Studies (LTS) model.

In the existing LTS model, exogenous cycle growth is matched through modifying constants in the distribution and mode split component. During the NDM development phase it was realised that using a similar approach through artificial changes in choice model parameters or cycle (“shadow”) costs would compromise the explanatory power of the new model and require a significant degree of forcing to match the anticipated high rates of growth in cycling.

An alternative method was identified which is to have a segment of the population for whom cycling is available (a propensity to cycle). The proportion of people for whom cycling is available can increase in future years and cause greater growth without significant cost changes. The method is similar to car ownership modelling.

The advantages of this method are that it can be linked to TfL policy removing barriers to cycling. It does not remove the requirement for an exogenous cycling forecast. Examples of policies which can be modelled include:
• What would happen if people the fraction of people considering cycling for leisure trips could be increased to generate a similar cycle mode share as currently observed in commuting?
• What would happen if people in one borough had a propensity to cycle equivalent to that in another?

However, it was specified that the method should be linked to evidence and it would be ideal if the segmentation and estimation phase of the NDM work could derive a group of people who have a propensity to cycle. Segmentation variables as well as components of generalised cost which can explain cycling mode choice were to be included where ever possible in the model estimation phase.

The cycling propensity analysis and the NDM segmentation and estimation were both based on the extensive rolling London Travel Demand Survey (LTDS) household survey dataset.

A methodology was developed to implement this approach. It comprised:

1. In the base year LTDS sample to be used for model estimation, flag all individuals with a “high cycling propensity” (HCP). All others are assumed to have low cycling propensity (LCP).
2. During model estimation, estimate separate model parameters (specifically the cycle mode constant, and also cost responsiveness where significant) for the LCP and HCP categories.
3. For base year model implementation, calculate the weighted average cycle mode constant, based on the ratio of HCP to LCP individuals in the base sample. This reduces run times by avoiding the need to explicitly segment the model between HCP and LCP.
4. For forecasting, calculate the weighted average cycle mode constant based on an exogenously defined ratio of HCP to LCP. This mechanism allows “what if” scenarios to be tested that reflect exogenous growth in the demand for cycling and the desire to move more socio-demographic groups into the HCP segment.

An initial approach to (1) was based on univariate analysis of an identified set of eight independent variables but this required the introduction of purpose-specific HCP thresholds, thus losing the rigour of association with individuals.

Bates (2016, private communication) proposed an alternative utility-based approach. We therefore performed logistic regression analysis against individual categories within the 8 variables to produce a cycle propensity utility measure for each individual within the LTDS record set, and defined a purpose-independent cut-off threshold to distinguish HCP from LCP. This approach proved very successful.

This paper discusses:
• the challenge from historic cycle usage trends and future policy;
• limitations of the current LTS approach to modelling walk cycle trips;
• the results of the cycle propensity analysis;
• the methodology for cycle modelling in NDM; and
• additional thoughts on scenario testing and variation in cycle usage due to weather.

Publisher

Association for European Transport