The city and the virus
Cities around the world are the epicentres of the coronavirus pandemic. Why is this, why have some places been hit so much harder than others, and – as countries start to move out of lockdown – what will come next?
There is no shortage of possible explanations. The catastrophe of COVID-19 has triggered a parallel wave of research and analysis, as we struggle to understand the virus and its impacts. By the end of April, scientists had already published over 7,500 articles on COVID-19. Economists had put out over 110 working papers (one has even managed a book). Urbanists have also been busy: hundreds of rapid-response pieces have been written about COVID-19 and cities.
In this piece I look over what’s been said about cities and the virus: by academics, journalists and the policy community. I test some of the competing theories about cities’ role in the spread of COVID-19; discuss its rolling impacts on urban economic and social life; and review predictions of what ‘post-virus’ cities might look like.
It’s a snapshot; I’m not an epidemiologist; so it’s inevitably incomplete, and any errors are mine.
1/ The Upside Down
Physical proximity and face to face contact are central to urban living, and to cities’ success as social organisms. So what happens when proximity becomes a friction, and face to face communication becomes dangerous?
Lockdown has drastically thinned out urban life. For those not working on the frontline, it’s now experienced at middle distance, or through the windows of our homes. Perhaps not surprisingly, many early articles on COVID-19 and cities had an elegaic tone; laments to lost urbanity, sometimes accompanied by photographs or video of deserted locations usually crowded with people. I’ve been collecting these here; the best is probably this one. The underlying sense of loss from lockdown is real, as are its mental impacts. Although in some cases, the amateur psychogeography is easy to overdramatise— especially given the grim experiences of those working in hospitals, care homes or public transport.
2/ Why have cities borne the brunt?
From the mass of hot takes, we can pick out six competing theories linking cities and COVID-19. Richard Florida also has a good overview here. Broadly speaking, these focus on one of places, people or policymaking.
The first comes from urban economics. Urban density and interaction are good for social and economic life, but are also vectors for the spread of disease. Cities provide more opportunities for bad interactions as well as good ones. The history of cities and pandemics provides much support for this view (see Ed Glaeser and The Economist). Just as agglomeration economies scale with city size, so may bigger cities’ vulnerability to disease. In more developed countries, we’ve got used to thinking of urban downsides as congestion, pollution and high living costs: having conquered urban public health problems, we’ve largely forgotten about them. (For an elegant reformulation of these ideas into network science, see Mike Batty’s piece here.) In its simplest form (here, here, and here), this line of argument makes density the enemy: cities’ economic strengths have become their public health weaknesses.
There are two variants of this story. One is that global flows of people and goods are additional channels for viruses to spread; so dense global cities — London, Paris, New York — will be particularly affected by COVID-19 compared to their wider nations. (Critical voices like Mike Davis, whose 2005 book on Avian Flu now feels prescient, would see the virus flowing through the networks of globalised urban capitalism.)
Another focuses on specific features of urban form, such as public transport networks. Commuting is one of the main ways in which citydwellers come into close contact with others, so this is a plausible transmission mechanism. In this view, people in cities like LA and Milton Keynes — built around cars and suburban neighbourhoods — should fare better than those in older, denser places. A striking early analysis suggested that the NYC subway system ‘seeded’ the epidemic in the city. In London, over 20 bus drivers have so far died from the virus.
The fourth theory focuses on urban living conditions, especially overcrowded housing and multigenerational households in expensive cities like London. Crowding brings people into sustained close contact, and there is a clear association between crowding and COVID-19 cases (here and here). This view implies that cases — and deaths — will be higher in poorer vs. richer cities, in more crowded neighbourhoods within cities, and among groups most affected by urban housing crises — including poorer people and some black and minority ethnic (BAME) groups. Some social structures — such as prevalance of multigenerational households — may turn out to be a particular problem.
A fifth view is that urban economic structures are the driving force: specifically the large numbers of high-contact, low-wage jobs that cannot be done remotely. Urban labour markets have increasingly polarised into high-wage knowledge-intensive work and low-wage service roles. Both sets of jobs depend on face to face interaction, but while the first group of workers have the means — and often the choice — to work from home, the second group often cannot. These workers are then exposed both physically and economically — to the virus if they keep working, and to loss of income if they do not, or if their workplaces close. Again, this suggests that COVID-19 will have unequal impacts both across and within cities. In the UK, younger, poorer (and some BAME) people outside the Greater South East are least likely to be working from home. In the US, it’s poorer lower-skilled people, often without easy access to healthcare. New ONS analysis picks out jobs most likely to be in frequent contact with people and exposed to disease. This workforce — including health and care sector workers, rank and file police, hairdressers, bar staff, primary and nursery teachers — is more female than male, around 1 in 5 minority ethnic (twice the population share) and has a large minority earning below median wages. Many of these workers — as well as security guards and bus/taxi drivers — also have very high COVID-19 death rates.
A sixth theory focuses on the role of institutions and political leadership, or the lack of it. In the US, where public heath is largely devolved, incompetent national government contrasts with some more activist state and city leaders. Lockdown timing, for example, seems to have a strong link to cases and death tolls. In the UK, we have mounting evidence of our own national government’s errors and failings. Worse, city leaders have far less policy autonomy to counterbalance this. Austerity has stripped out local government’s capacity, especially in urban areas. In turn, that implies that places with weaker local government — and tax bases — will be harder hit and less able to manage recovery.
3/ Who’s right?
Here is a paradox (courtesy of Matt Singh). We know from individual health records that older people are more vulnerable to Coronavirus. So we might expect places with more older people — say, over-70s — to have more COVID-19 cases, and deaths, per population. In fact, we see a slightly negative link.
Expecting higher death rates in ‘older places’ is a classic case of an ecological fallacy: here, reading off individual outcomes from area characteristics. It turns out that places with a smaller share of over-70s have slightly more cases per 100k people. So is that because of something about those places? Or the kinds of people in them?
For example, there is a long history of blaming urban built form for problems that have their roots in wider social or economic conditions. Is it correct this time? Globally, there are more COVID-19 cases, and deaths, in urban areas. US and UK analysis also suggests that case and death rates are also higher in urban areas, especially when London and New York are included — but that socio-economic and health factors probably explain at least some of this, as well as density and global city status (see this by Jed Kolko and this by Valentine Quinio, for example).
The growing evidence for links between COVID-19 outcomes, area deprivation and demography are especially disturbing. This huge study of NHS patient data suggests that COVID-19 deaths in hospital are linked to old age in particular, as well as pre-existing health conditions and being male; but also to minority ethnic status and to higher deprivation. This ONS analysis suggests that all minority groups have higher death rates than white people, Black and Pakistani/Bangladeshi men and women in particular. As the ONS suggests, that probably partly reflects socio-economic disadvantage, but may also reflect (largely urban) location, type of work, household structures or all of these.
There are four big challenges in figuring out what is happening, at least with area-level data. First, as Jed points out, it is hard to establish causal links: the patterns may be clear but not the reasons. Second, patterns may not be that clear: picking out the role of specific features — physical or otherwise — is challenging because they are all closely related to each other. The biggest, densest cities are most likely to have large public transport systems; higher density housing; higher costs of living, leading to overcrowding; and higher levels of income inequality. It turns out, for example, that it as easy to show that cars ‘caused’ COVID-19 in NYC as the subway.
Third, the epidemic and responses to it are dynamic: as outbreaks move through space, people change their behaviour if they can. Mobile phone data from Google and other platforms shows dramatic drops in urban mobility as social distancing has kicked in. This may help explain why in New York, for example, density seems to be less important in explaining cases over time. But we’d expect bigger behaviour responses in a) places with more cases, and b) places where more jobs can be done from home, as UK data for city centres seems to suggest. And more broadly, the extent to which people can socially distance is shaped by their economic and household circumstances — those in ‘essential’ jobs and crowded homes can’t do this easily. It’s hard to separate out the effect of lockdown policies from these wider features.
Finally, we have to take wider factors into account. Some very large, dense East Asian cities — notably Seoul, Singapore and Hong Kong — have contained their outbreaks more successfully than their European or US counterparts. That implies policy choices and systems — not to mention recent experience of SARS and other pandemics — are at least as important as any urban features.
With that in mind, what does the data suggest for England? Here are some first impressions — at best these are associations, not causal linkages, and the patterns aren’t always clear. Overall, we are still some way off understanding the role of urban places in the pandemic.
At first sight, the case data from Public Health England supports the density and global city hypotheses. Figure 2 shows confirmed hospital cases in English city-regions over time (I use combined authority boundaries). London has more than almost all other conurbations put together, and has a faster growth rate (although we know that this data is a substantial underestimate of the true number of cases).
Figure 3 controls for population and changes the picture drastically. We can now see London’s case rate growth gradually caught by a number of other city-regions, especially those in the North West and North East.
Figure 4 breaks down case growth within city-regions, specifically the Unitary/Upper Tier Local Authorities used in the PHE data. We can see some huge variation at the local level within London, and in the conurbations — like Birmingham, Liverpool and Newcastle-Gateshead — where overall cases have grown the most.
The highest cases rates are now outside London. The Newcastle city-region has become a national hotspot, with Sunderland (435 per 100k as of 1 May) and Gateshead (430) the two highest case rates in the country. South Tyneside (397) is also heavily affected, as is Middlesborough in the nearby Tees Valley (410). Knowsley (387) is the most affected in the Liverpool city-region, followed by St Helens (376). Oldham (353) has the highest case rates in Greater Manchester, followed by Salford (320). In Birmingham, Walsall is the most affected (363) followed by Wolverhampton (336). London’s highest case rates are now a mixture of Outer boroughs (Brent, 417; Harrow, 377) and inner boroughs (Southwark, 383; Lambeth, 354). Remember, all these figures are undercounts of the true case rates, since they miss care homes and the wider community.
What might be driving this? Figure 5 is a binscatter plot showing how the simple relationship between COVID-19 cases and population density changes over time. The left hand graph plots the link as of lockdown, 23 March. The right hand graph shows a rather weaker link on 1 May, as COVID-19 has spread across the country. That weakening density link is in line with this study for New York.
Figure 6 repeats the analysis, adding in controls for urban form (public transport use, household size, household crowding), demography (male, under 30s, over 70s) and economic conditions (workers by SOC1 groups, median wage, self-employment, IMD). Now the case-density relationship is much flatter at lockdown, and still close to zero by 1 May. The problem here is that many of these features are highly correlated with each other as well as with density. This makes it harder to draw firm conclusions about what we’re seeing.
Figure 7 gives a sense of the underlying complexity. We can see that population density is positively linked to COVID-19 cases (top left). But public transport use (bottom left), household size (top right) and household overcrowding (bottom right) all have an equally strong connection. It’s challenging to pick these features of cities apart.
Figure 7 also shows that there are fewer over-70s living in denser cities and in crowded conditions, except when they are part of larger — probably multigenerational — households. If some older people in cities are more vulnerable than others because of their household structures, that may help explain the puzzle in Figure 1 above.
Finally, Figure 8 looks at the link between COVID-19 cases and deprivation, as measured by districts’ rank on the Index of Multiple Deprivation (IMD). This relationship also holds — albeit weaker — controlling for demographic, urban and economic features. Here we see that a weakly negative relationship at lockdown has, alarmingly, reversed and strengthened by 1 May. There are now more cases for people in more deprived places, a relationship that also holds for deaths from the virus. (Tom Forth has done a very detailed analysis of this relationship both between and within cities, which is worth reading in full.)
4/ The wider (economic) impacts
As the pandemic has spread, governments and commentators have turned their attention to the likely impacts. It’s already clear that the UK and most other countries are heading into severe recessions. In the first week of UK lockdown, over 50% of workers in one survey had already done less work, 8% had already lost their jobs, and a third expected to do so by August. The UK has seen over 2m new benefit claims. We also know that younger, poorer people are mostly likely to bear the immediate costs of this. That’s because these groups are already being most affected by lockdowns and social distancing. There is also a geography to these shifts: as this US paper shows, places with more exposed workers are already seeing falls in employment and wages.
However, these patterns are less clear in the UK — so far. Simply predicting vulnerability is of limited use here. One analysis, for example, suggests that small and medium sized towns will be most at risk given their industrial and occupational mix. Another suggests that every city in the UK has at least 20% of vulnerable or very vulnerable jobs, and there is no clear geography of impact.
A challenge facing all predictive studies is the all-encompassing nature of the pandemic, and the very large set of effects it generates (before we even think about how these might interact with each other). As Henry Overman argues here, past experience may not help given the nature of this shock.
A second body of work gets around this by using high-frequency data to describe impacts across groups and space in close-to-real time (see this and this, for example). In the UK much of the credit belongs to the ONS, which has worked overtime on new datasets and analysis. In the US, a number of more … specialist tech firms than Google and Apple have turned over a huge amount of mobile phone location data to public use (here and here, for example). My UCL colleagues James Cheshire and Terje Trasberg use similar UK data to show a massive decline in retail activity in big British cities.
Two other UK studies – here and here – show alarming drops in consumer spending using banking transactions data (See also these from Denmark, Spain and the US.) Both British studies also emphasise the unusual geography of the coming slowdown. The St Andrews team, working at regional level, find the biggest proportional shifts in London, and among higher-income people. The Tortoise analysis highlights that consumer spending is falling most in both rich cities like Oxford (down 56%) and Canterbury (53%) – but also coastal towns like Whitby (66%) and Penzance (69%), or tourist hotspots like Kendal (62%). The only three areas of the country where spending has risen are Bradford, Yeovil and Milton Keynes. This is not going to be a normal recession.
5/ The 90% world
Many countries are starting to lift lockdown restrictions this week. As people venture back out into the world, what will they find, and how might they act?
Claims that the world ‘will never be the same’ seem off the mark. They miss the historical reality that the world and its cities have coexisted with pandemics for a long time. They also forget the very recent experience of urbanised pandemics like SARS (2002), avian flu (2005) and H1N1 (2009). In the mainly East Asian countries and cities where they concentrated, everyday life has shifted, but is still very much recognisable. A perceptive piece in The Economist, which is worth reading in full, describes a ‘90% world’ in which life will be close to ‘normal’ but subtly different: we will be in the uncanny valley until a vaccine is developed.
What might shape our behaviour in such a world? First, our own habits and biases. A mass of evidence suggests that we are generally risk-averse, but also habit-bound. Those impulses will be also be governed by two other dynamics: what governments encourage or mandate us to do, and what is left of the economy. As countries start to work out how to to live with social distancing, there are tensions between the three.
We are starting to see signs of what 90% cities might look like. This perceptive piece by Gideon Lichfield, written in mid-March, still provides a useful map.
A first zone of change is in urban infrastructure. These have already started with informal, ‘emergency urbanism’ — like pavement extensions or these painted-in bike lanes. But given that underlying road systems are often not good enough — in London,for example, 2/3 of pavements are not wide enough for social distancing — cities now need to shift to more formalised planning. Given the possible links between COVID-19 and air pollution, this will mean both making public transport safer, and shifting people to walking, cycling or electric scooters (here is a good roundup of city leaders’ current thinking).
In the UK, cities like London, Manchester and Birmingham have already been building physical cycling infrastructure. But a major shift in commuting mode— even if jobs are safe to go back — requires huge behaviour change. As Paul Swinney points out, on distance alone over 2/3 of commuters in England and Wales could walk or cycle to work; but even in cities, over 2/3 of people commute by car. Moving to — say — Dutch levels of cycling can take decades. For most cities, I suspect that means the bigger short term goal is making public transport a safe option — through some mix of masks, health checks or lower density ridership. In turn, that will mean finding ways to fniancially support public transport systems that rely on fare income.
A second feature of post-virus cities will be be living with greater surveillance, as the examples of Singapore, Israel and Korea suggest. Contact-tracing apps involve some basic technological tradeoffs: John Naughton has an excellent series of posts on centralised vs decentralised approaches to contact tracing. They are also not a ’solution’: at minimum they also require effective public health systems and high take-up.
A third feature of ‘post-virus’ cities will be shifts in ways of working. Large firms are already planning how lower-density office spaces might work: options include fewer people in buildings; staggered working hours; ‘distancing standards’; ending hotdesking and long meetings; temperature sensing and deep cleaning; more remote working. But not all organisations will be able to make such changes: how can employees in smaller, less cash-rich companies be safe?
The remote-working world is part of the death of distance, a bad future for cities that has never actually arrived. In most countries, far more jobs could — in theory — be done at home than currently are. Dingel and Neiman estimate ‘teleworkable’ jobs for a number of countries on the basis of tasks involved. For the UK, 43.5% of jobs could be done from home. But in practice, only 27% of British workers have ever done so. Even leaving aside the majority of workers who cannot do their jobs from home — including most of those in lower-paid occupations — closing that gap raises a number of major questions.
The social and econmic impacts of pervasive remote working are still poorly understood. There is suggestive evidence that — at least for some routine tasks — providing the option of working from home raises productivity. If remote working is the norm, however, it is much harder to have the rich interactions that much work requires. A number of studies with scientists and researchers suggest that microgeographies of office and lab space make a differences to the quantity and quality of collaborations. Remote-working technologies seem to bring their own downsides. All of this suggests that a remote-working future would be less innovative, as well as less socially rich.
While it’s possible to work through single issues, it’s far harder to imagine how urban systems overall might change further ahead. Only a few have been bold enough to try. Mike Batty (here), Paul Cheshire and Christian Hilber (here), Richard Florida (here) and Mark Kleinman (here) all suggest a near/far future in which high-value economic activity concentrates even more in a few urban cores; the amenity value of big cities falls; home-working creates a push towards suburban living and lower-frequency commuting; some supply chains deglobalise, and there is far less inter-urban travel (especially air travel). This might be a greener urban world, but it could also be a more divided one.
Cities will come through this crisis, as they have in the past. Whether they are better or worse places to be is still, in large part, up to us.
Many thanks to Mike Batty, Adam Dennett, Mark Kleinman, Liz Moor and John Tomaney for comments, suggestions and encouragement.
You can download code and data here.
