Using agent simulations to understand equity impacts of transport policies
Equity in transport systems
In the City Modelling Lab, we employ activity- and agent-based modelling (A2BM) techniques to simulate transport systems, draw policy insights and support decision-making. Often, the questions that we wish to answer on behalf of our clients revolve around the equity aspects of transport. In our context, an equity assessment typically looks at the distribution of benefits and disbenefits of a transport intervention among members of society. For example:
- How can we reduce emissions in a fair way?
- Do all groups benefit from a new infrastructure investment?
We recognise that people may have different transport needs and resources, different plans, and may experience very different outcomes in adapting to any changes. Current planning practice, while typically recognising the social dimension of transport policies, sometimes underplays its role in investment decisions. Our aim is to help bridge that gap through the development of an appropriate equity appraisal framework underpinned by analytical tools. Essentially helping our clients understand the equity impacts of policies in order to make better decisions.
We aspire towards a planning process that is socially responsible, supports more equitable access to opportunities (such as employment and basic services), considers vulnerable groups, and recognises people’s complex and interlinked transport requirements for achieving their activities within a day.
Why A2BMs?
The diversity of needs and outcomes is often hidden in averages. In our A2BM approach, we try to develop diverse representations of populations (agents) and their plans (see our blog series on PAM), with realistic distributions of personal and household characteristics, which are linked to their travel behaviour.
For example, our analysis of UK travel survey data shows that income is correlated with the distance and frequency of trips, as well as the choice of travel mode (as shown in the chart below); age is related to the likelihood of using public transport or active modes; and gender may be correlated to the number of activities undertaken within the day and the share of carer trips.
In our simulations, we seek to capture some of this heterogeneity of attributes and plans, and examine their interplay with environmental factors. Our models’ behavioural focus means that we can simulate the choices and constraints faced by different individuals throughout their day, and their priorities when making decisions.
For example, we may consider:
- trade-offs between time and cost: low-income individuals may favour a longer journey that is more cost-effective;
- household structures: parents may need to fit their travel and activities around their children's schedule;
- access to transport services: public transport-deprived areas may mean that people have no choice but to drive
Finally, the outputs of our A2BM offer high granularity; they tell us where every agent is and what they are doing at every second of the day. We can, therefore, flexibly slice and analyse model results in a multitude of ways to better examine the different aspects of equity impacts.
Setting up an analysis framework
Before jumping to modelling, we start by outlining an appropriate and consistent framework for evaluating and comparing the equity effects of transport policies. It is primarily a two-pronged approach:
a) Baseline assessment: first, we want to construct our equity questions, understand the spatial context, and identify groups of concern that may be impacted by the tested policy. We take into account the policy objectives (e.g. reduce congestion, lower carbon emissions, etc.), likely outcomes, the demographic characteristics of the wider area, the types of travel patterns, etc. The findings of this exercise are used to inform the scoping of our model: which demographic variables and behaviours should we represent in more detail, how should our scenarios be constructed, and what are our main hypotheses to test?
b) Equity evaluation: after we run our simulations, we consider appropriate ways to summarise the results, so that they can be transformed into useful and actionable insights. For example, we consider appropriate metrics to quantify equity effects, compare scenarios, and — where necessary — propose mitigation policies.
Measuring impacts
So, how do we “measure” transport equity? How can we select a “good” indicator to represent the priorities of vulnerable people? And how do we make sense of very complex and large model outputs? Of course, these are difficult questions with many possible answers, but we have made a start by testing some ideas on real case studies.
We start by simulating policy scenarios, such as a new public transport service or pricing scheme. Then we compare the results of the policy scenario against the “baseline” scenario — a counterfactual situation in the absence of the tested intervention. We dive into the different ways agents respond to the new conditions; for example, whether they spend more or less time travelling, wait less for transit, use more direct routes, spend less time at home or at leisure activity, etc.
As a measure for summarising such responses, we looked at agent utility, which encompasses and weights various aspects of travel experience (such as time, perception of cost, physical effort, mode preferences, ability to complete activities within certain time windows, etc). In our analysis, utility is effectively a measure of how well individuals were able to go about their day, and its composition can be informed by observed data on travel behaviour.
To analyse results, our first step is to group agents into “study” and “comparison” populations and assess any differences in their outcomes, similar to a quasi-experiment. The study group represents groups of concern from an equity perspective, linked to the characteristics of an area and the objectives of our scheme. The comparison group comprises every other agent outside the study group. That way, we can start commenting on distributional aspects: for example, whether a specific vulnerable group is experiencing less than ‘fair’ shares of benefits or disproportionately higher shares of disbenefits.
Going a step further, we start looking at distributions of impacts. We measure the policy’s impact on each individual agent, sort the results, and then summarise their distribution across the model’s population.
For example, the plot below shows the cumulative distribution function (CDF) of a scheme’s impacts on agents’ utility. That way, we can examine the distribution of impacts between “winners” and “losers” of each policy. We can understand whether there are small positive or negative impacts to a large number of agents, or vice versa, or what percentage of the population sees a certain share of benefits. And, more importantly, how those curves may differ between a study group and everyone else.
Finally, we are exploring clustering approaches to understand plan typologies. The aim is to understand whether people with similar plans (for example, night workers) are impacted in a distinct way. More on that to follow in a later post.
Conclusion
In this post, we described some of the transport equity-related questions that policymakers often struggle to answer, and how A2BMs enable us to explore these questions in detail.
By testing such an approach in real case studies, we believe that A2BMs offer a strong advantage in performing better equity analyses. They help us to see transport effects through the lenses of different groups and thus take their considerations on board when designing policies. A key component in this process is creating a feedback loop between analysis and design: identify whether expected negative effects can be mitigated (through policy strategies) or eliminated (by revisiting scheme design).
Theodore is a senior data scientist and R&D Lead at Arup’s City Modelling Lab.
Paola Bueno is the Social and Equity Lead of Arup’s Transport Economics and Growth Business area.
