In crowded cities, with growing populations moving around on busy streets, urban planners are faced with a need to both predict the future and shape current behaviours. Twenty years ago London attempted to change the incentive structures around driving with the introduction of a congestion charge, price “to encourage a sufficient change in travel behaviour”, with exemptions for specific groups, including residents and disabled people. The system was narrowly focused but it proved that such schemes provide useful new levers to address policy priorities.
Two decades later, policy makers and city leaders are still shaping people’s transport choices via the congestion charge and ULEZ schemes, but they have a wider range of issues to tackle: air quality, decarbonising travel, performance improvements, congestion, how to serve underrepresented groups, and measures to improve people’s health and wellbeing. This more complex set of goals makes crafting and implementing new policies more challenging, especially in areas where there is no historical data or precedent to guide us.
Designing the charge
The evidence base for the original congestion charge was made using TfL’s modelling, which was based on elasticities. At the time, this approach to planning a congestion charge was considered ground-breaking. Elasticities measure responses to a change, so if you increase prices by a certain amount, you can expect a corresponding change in traffic. These mechanisms are based on historical data, which is often hard to find, and doesn’t always capture the complexity of real decision-making.
An elasticity that links pricing to cars on the road makes it much harder to predict more fine-grained impacts, like which groups of people are most affected by an increase in costs, and how these impacts are distributed across a population. An easy critique of any flat congestion charge concerns equity or fairness – the greatest change of behaviour is made by those least able to pay.
Another of the dangers of modelling only at the aggregate level is that in the real world, there is no such thing as the average person. No family has 2.4 children, and real people are all unique in their plans, behaviours and restrictions. So how should you go about modelling their different responses to congestion charges, or anything else?
Agents of change – understanding the travelling public in the 2020s
Complexity is inherent in our transport system because it is inherent in our society. Patterns of living have become dynamic, flexible and unpredictable, especially as we continue to deal with the post-pandemic realities of hybrid working and new work/life patterns. The future no longer looks like the past. An elasticity-based approach is no longer useful; we’ve never seen these changes before, so we can’t use past performance to predict future behaviour.
That’s why in the City Modelling Lab, we’ve been developing a data-driven, simulation-based approach to transport modelling. We call these simulations “Agent-Based Models”, and they are helping us understand and respond to the complexity of modern transport policymaking.
A people-based approach to transport policy
Essentially, we create a computer simulation that represents the transport choices of millions of virtual people, or “agents”. Each agent has a set of activities – such as working, shopping, or going to the gym – that they want to complete over the course of the day. They can choose from all available modes of transport when making the journeys necessary to complete these activities. Through a process of trial and improvement, they each work out what works best for them to achieve their daily activity goals.
Each agent has the potential to indirectly affect other agents in the simulation. If, for example, every agent decides to drive, the roads become congested, so journeys take longer for everyone. Some agents will then try a different route, set out at a different time, or use a different mode of travel in an attempt to complete their plans more efficiently. Or, if too many agents start using transit, they might find it crowded or even be forced to wait longer for vehicles that are not too crowded to board. Some agents will find it harder than others to make changes that work for them.
Our models’ predictive power is grounded in data about individual behaviour. So, interactions between the agents and the transport network reproduce the complexity of the real world. ” Nick Bec London Data and Digital Leader
Armed with this individual-level detail within the simulation, we can then explore a huge range of different outcomes and test many different types of scenarios. We can look at who benefits most from proposed changes, analyse wider network impacts, and generate detailed carbon emission baselines.
Modelling and the future of congestion charging
So back to the Congestion Charge. How could one of these simulations help to assess a new charging scheme? We’ve been working with one of our clients in Ireland to study the effects of road charging and have been using one of our simulations to explore the impacts of different road charging scenarios. This is feeding into their long-term, country-wide strategy for the road network.
While these scenarios are exploring options at a very early stage, we’ve tested charges that vary based on vehicle type, time of day, type of road, and many combinations of these things. In doing so, we’ve explored a range of different questions:
How does a charge impact those on lower incomes?
Are people in rural areas unfairly penalised because driving is their only option?
How are journey times impacted?
This gives policymakers and planners better tools to test and plan new charging regimes before they go live in the real world. This is especially important as we aim to achieve multiple different competing outcomes and deliver policies that are fair for everyone.
Beyond road charging
Agent-based models are clearly very useful for understanding people’s responses to road charging in the here and now, but that is not all they can do for us. They also allow us to understand how behavioural responses might change in different hypothetical scenarios. For example, looking at travel patterns more holistically is really helpful when looking at short or long term changes – increased demand from a big event, balancing the needs of residents with those of visitors, and seeing how investing in infrastructure to support an event can deliver value for the area in the longer term. We can also look at impacts on air quality and carbon emissions.
Perhaps we want to model multiple scenarios that each represent a different demographic trend, e.g. the population becoming bigger/smaller, older/younger or wealthier/less wealthy. We can also look beyond people to things like the mix of different types of vehicles (EVs, combustion engines, hydrogen). Agent-based models are a great fit for these types of counterfactual analysis.
Beyond road charging, we are also using ABMs to:
simulate the impacts of increased walking and cycling
predict where new EV charging points will be required within the road network
assess the role of EVs in meeting decarbonisation targets
create bus strategies aimed at reducing the need for private vehicle use on short journeys
We are still in the early stages of using agent-based models to help shape policies that deliver better, more equitable, and more reliable transport outcomes. Still, these are exciting times in transport planning!