Visiting Fellows at the Bennett Institute, Kaveh Jahanshahi and Ying Jin have developed a new methodology, connecting two data science techniques, for capturing the commuting geographies of people across Britain in more granular and precise ways. In this blog, they reflect on the significant similarities between the travel patterns they identify in different kinds of place, and the incidence and spread of COVID-19.
As people in England who cannot work from home are encouraged by Boris Johnson to find a way to go back to their jobs, the prosaic question of how people commute has suddenly assumed considerable importance.
In terms of available options for commuting, much of what we have taken for granted in previous decades is now highly uncertain. The Prime Minister asked commuters to avoid public transport if they can, and traffic jams are already building up in some busy areas despite the relatively small number of people driving to work (traffic on London’s strategic road network is 29% down from mid May 2019). How to negotiate the new challenges of commuting in the era of social distancing will become a major issue of concern for many citizens.
The UK is actually well endowed in terms of its expertise in understanding travel behaviour. There is a rich eco-system of transport research across the UK, and it runs the world’s best travel survey at the national level [1]. The basic idea of the UK National Travel Survey (NTS) is quite similar to the COVID-19 surveillance survey that is currently being rolled out. The NTS has been asking a carefully designed sample of households and individuals across the UK about their travel habits, and the full range of personal and wider circumstances that shape their choices, since 1988.
Could the data generated by this source tell us anything about the new challenges we now face?
We think it can. After all, for the coming months at least, we are talking about people going back to their previous workplaces. The layout of settlements and transport networks may change, in the coming months, but only gradually. And the choices we make about how we get to work are underpinned by a fairly unchanging set of psychological impulses and influences.
How we understand the personal and wider factors that influence these choices is the subject of a paper we have just published (open access here). In this we report on a new methodology we have developed for interpreting the results of the NTS, combining the data science tools of structural equation and latent class modelling (where structure equation modelling disentangles the interactions among a myriad of different influences, and the latent classes account for the diverse needs and circumstances of travel). Our analysis illuminates the complex interaction of personal and wider circumstances which together comprise a web of influences that can be mapped onto a ‘latent geography’ which is distinctive from conventional geographies used in current research and modelling. The methodology behind this latent geography, we suggest, might well provide important insights into some of the issues associated with the response to the coronavirus epidemic.
Latent geography of commuting choices in the UK
This latent geography consists of five main commuter ‘types’ who are residents living in eight kinds of areas (including three transition areas between the main types; see the left panel of Figure 1). Compared with the conventional rural-urban classification of Britain’s geography, which all current analyses adopt in one way or another – as indeed does the NTS itself (shown on the right panel of Figure 1) – the latent geography that we identify provides a more precise and clear delineation of commuter distances, commuting durations, shares of car travel and uses of public transport (See Figure 2).
Mapping the latent commuting geography that we identify onto the residential population of the eight kinds of areas we look at, it is clear that only around 8% of the population in the metropolitan/urban core areas can readily respond to the government’s calls to do more walking and cycling, for those whose commuting distances are the shortest. Parallel research by Steve Denman and Ying Jin, in collaboration with Professor Robert Cervero at University of California, Berkeley, indicates that women are far more reluctant to cycle and walk than men, and more sensitive to stress on the cycle network [3]. In the suburban areas around the main cities – where people have started using public transport more regularly, and doing more walking and cycling – car-usage levels are not that different from the remote towns and rural areas (Figure 2). These areas include the outer boroughs of London and other metropolitan cities.
This suggests that people in local areas are likely to respond very differently to the new proposals coming from government, and this will generate a new kind of commuting geography. The government’s current advice has left considerable room for very different responses from citizens. And it will need to clarify its guidance rapidly to avoid chaos on the road network – the Blackwall tunnel has already seen long tailbacks, even with only a limited number of people back at work.
Figure 1 – A comparison of geographies: our latent classes based on travel behaviours (left panel) vs conventional area types based on administrative boundaries (right panel)
Figure 2 – A comparison of geographies: our latent classes based on travel behaviours vs conventional area types based on administrative boundaries
In view of the likely move away from public transport in the context of COVID-19, the dense urban areas we look at may see more e-bikes and similar vehicles appearing for the longer commutes, besides a greater take-up of cycling and walking for shorter journeys. A particular challenge may emerge for those living on the peripheries of metropolitan conurbations – such as the so-called ‘Golden Arc’ running through Hampshire, Berkshire, Oxfordshire, Buckinghamshire, Hertfordshire, and Cambridgeshire to the coast, where the local roads have weak and easily congested intersections, especially around the suburban science and innovation labs that may have a critical role to play in the battles against the virus. It is in these areas where the government’s calls for people to drive to work are most likely to be heeded, and where traffic growth will be most apparent.
Latent commuting geography and COVID-19 incidence
We have started to compare the commuting geography described above with the emerging patterns of identified COVID-19 cases and deaths (Figure 3 and Figure 4). And we find considerable similarity between them. Our initial thoughts for this parallel are that levels of population density and interrelated commuting patterns may have an important role to play in both the risk of transmission and the severity of the disease. This will need to be carefully analysed of course, alongside the consideration of other factors such as socioeconomic background and poverty. We are working on this topic, and hope that the improved COVID-19 surveillance data will enable this analysis to progress quickly.
Scope for new investigations
Since most countries have done travel surveys with methods like the UK NTS at the national or regional level, we are currently expanding this analysis to other countries, and hope that colleagues in this field will collaborate with us. For post COVID-19 planning, these travel surveys can do more than support the day-to-day planning of transport operations. The new methodology we have been using can help us surpass barriers to understanding the plentiful information they provide about different influences on travel decisions.
Our sense is that there may be considerable potential for using the methodology behind the techniques we have employed to interpret the incidence of COVID-19 across diverse places and communities. As the COVID-19 surveillance programmes start to produce dependable data, the same methodological questions will arise, regarding variable interaction and endogeneity, and the discovery of latent geographies that emerge under multiple and interactive influences. For instance, in investigating the high incidence of COVID-19-related deaths in the BME communities [3], a key challenge is to make a distinction between the impacts of spatial proximity, social connections, commuting patterns, genetics, etc – precisely the kind of analytical question that our new methodology is designed to address.
Figure 3 – A comparison with the geography of COVID-19 cases: England
Figure 4 – A comparison with the geography of COVID-19 deaths: England and Wales
References:
[1] The UK National Travel Survey. See https://www.gov.uk/government/collections/national-travel-survey-statistics (Accessed 8 May 2020).
[2] Kaveh Jahanshahi and Ying Jin (2020). Identification and mapping of spatial variations in travel choices through combining structural equation modelling and latent class analysis: findings for Great Britain. Published as ‘First Online’ at Transportation (https://doi.org/10.1007/s11116-020-10098-9; permanent open access at https://www.repository.cam.ac.uk/handle/1810/303371). The authors work for Cambridge University’s Martin Centre for Architectural and Urban Studies, and they are Visiting Fellows at the Bennett Institute for Public Policy (respectively 2019-2020 and 2018-2019); Kaveh Jahanshahi is also a Senior Lecturer at the ONS Data Science Campus. The research has been supported by UK Department for Transport’s Transportation Research Innovation Grant (T-Trig), although the views expressed in the paper are those of the authors alone and do not necessarily reflect those of the authors’ workplaces or UK DfT.
[3] Cervero, R, S Denman and Y Jin (2018). Network Design, Built and Natural Environments, and Bicycle Commuting: Evidence from British Cities and Towns. Transport Policy. Vol 74: 153-164. See https://www.sciencedirect.com/science/article/pii/S0967070X1830101X; permanent open access at https://www.repository.cam.ac.uk/handle/1810/285496).
[4] The Guardian (2020) ‘Ethnic minorities dying of Covid-19 at higher rate, analysis shows (https://www.theguardian.com/world/2020/apr/22/racial-inequality-in-britain-found-a-risk-factor-for-covid-19; Accessed 8 May 2020).
The views and opinions expressed in this post are those of the author(s) and not necessarily those of the Bennett Institute for Public Policy.
