Disparities in COVID-19, Mobility, and Housing Crowding
Using mobility measures to understand disparities in exposure to COVID-19. Also: why improvements in ER care overstate the reduction in crime
A big focus of this newsletter is the positive agglomeration economies from cities. It’s really valuable to have people close to other people, and important to think about ways we could improve these spillovers even more. But cities typically teeter on the edge between both benefits as well as the costs—the negative externalities from agglomerations. Today I want to talk about two of these congestion effects—disease, and crime.
Racial Disparities in COVID-19 Exposure
There is a pretty striking graphic from the Marshall Project breaking out excess mortality across time and race. Excess mortality numbers abstract from difficulties in case measurement, and show a pretty striking pattern here: non-whites are 2-4x more likely to die from Covid-19 than whites.
Now, what’s interesting here is that the best medical studies we have so far actually find little evidence of disparities in outcomes among hospitalized patients. So these disparities in deaths are attributable to disparities in case exposure. But where do those come from?
I look at this question in a new paper with Milena Almargo, Joshua Coven, and Angelo Orane. We focus on New York City, where the disparities in Covid outcomes across neighborhoods are incredibly stark:
We use mobile phone geolocation data to investigate the risk factors behind Covid exposure. This is all the rage among kids these days, but we are able to push our data a little further. We are able to connect individual devices to hospitals, allowing us to measure hospitalization at an individual level (mobility-measured hospitalizations correlate well with official measures—in this period, most people in the hospital were there for Covid-related reasons).
Our analysis highlights two key risk factors:
Time spent outside of home tract, which we find is associated with commuting to essential work.
We find pretty sizable effects—someone who spends above the median amount of time outside of home has a two percentage point higher chance of winding up in the hospital, relative to a less mobile individual. And we find that both of these risk measures—mobility out of home, and housing crowding—show racial disparities. Basically, structural inequalities lead disadvantaged groups to disproportionately live in crowded housing and specialize in jobs that require physical presence. These underlying inequalities in job presence and housing create temporary pockets of density through which SARS-CoV-2 virus propagates.
What Happened in New York City this Spring
So our narrative of the crisis in NYC runs something like this. Starting in early March, the crisis became pretty apparent in New York City. A pretty large chunk of the city—10-15% of Manhattan for instance—left right at that point (another paper coming out soon documents this urban flight in more detail). Of the remaining population: those who could work from home hunkered down. Racial minorities and lower-income workers, however, continued to commute and head to essential work, and transmitted the disease through that route. This was, of course, the period when the CDC and other health authorities were still downplaying the role of masks.
At some point in mid-April or so, many of these workers were laid off. The relative role of the two channels switched at this point, and disease transmission still continued through the housing crowding channel. We talk about the complementarity between these two channels—locking down may help to limit some forms of disease transmission, but if people instead spend more time at home that may change the nature of disease spread.
I find some similar results in other work with Boyeong Hong, Bartosz Bonczak, Lorna Thorpe, and Constantine Kontokosta. We divided up neighborhoods through a clustering approach. The wealthier parts of Manhattan and Brooklyn, where individuals are able to flee/shelter show really drastic reductions in activity relative to the outer boroughs with more income and racial diversity, where mobility patterns continued at more normal rates.
I think these results also complement findings in other countries. Anup Malani and co-authors, for instance, find maybe 58% of individuals in Mumbai slums were positive, but 17% of people in non-slums were positive. Other studies have found 40% positivity rates in some Cape Town areas. These areas wind up hitting up against the raw herd immunity limits, due to the crowded housing environments and essential work patterns of residents.
I really enjoyed working on this project, the collaboration for which was purely virtual. Here are some of the other takeaways I have:
It’s essential to introduce isolation homes for individuals who test positive, to limit spread to others. We’ve talked about this before in the context of isolation strategies elsewhere in the world, and relevant for Universities as well. But basically this household channel seems important and we still don’t have great policy to address this.
I think people have internalized more of the potential costs from spreading the disease over time—but we still want to make sure (through sick leave, for instance) that individuals who have symptoms or have tested positive are able to remain at home.
Crowded home situations are a product of tight zoning, and we should incorporate this in our planning decisions. Unnecessary restrictions on new construction in urban areas don’t just raise prices. It also lowers available quantity, resulting in worse housing stock for people who are trying to access urban labor markets.
Improved ER Survival Rates and Measured Violence
A key narrative of urban renaissance is the drastic drop in crime rates over the last few decades. American cities became plagued with violent crime, triggering out-migration and urban decay portrayed in the Kurt Russell classic Escape from New York.
Why exactly this situation improved remains a topic of active debate. Steve Levitt and John Donohue, controversially, connected the drop in abortion with the subsequent reduction in crime (a result that was contested then defended etc.). Another common narrative centers on the role of lead, which was phased out of gasoline and paint and to associate with crime and other bad outcomes in some well-done studies. Now, whatever the reason—the reduction in crime rates was associated with an urban renaissance (or just gentrification if you are the curmudgeonly type).
But another prosaic reason is—what if also got better at treating patients in the ER, cloaking some of this decline in murder rates?
An interesting paper on the topic by Anthony Harris, Stephen Thomas, Gene Fisher, and David Hirsch documents a large decrease in the lethality of homicide rates until 1999, accompanied with an re-classification of non-lethal crimes as “aggravated assaults.” This continued until about 1992, after which point assaults declined as well (though not reaching the previous lows).
One plausible explanation is that the victimization lethality rate declined due to improvements in medical care, especially in the ER, alongside other public health advances. It turns out this improvement was particularly strong in rural areas, where the authors argue the improvement in medical advances has been strongest. If we had a lethality rate closer to the pre-Vietnam War rates, for instance, the total murder rate today would look a closer to those historic highs.
Now, I don’t think this is the full story—it does look like other violent crimes (such as other property crimes) declined in this period. And we could have done a better job at measuring low-violent offenses, artificially biasing down the lethality rate. At the same time, it appears have been upgrading their weaponry—substituting fewer knives and going for higher caliber guns. So one interpretation is that we took some of the “gain” in mortality resulting from better medical treatment, and criminals responded by increasing the lethality of their own weapons.
That data go up through 1999, but another paper here follows up documents further mortality improvements:
Lower case-fatality rates over time for firearm assaults were evident in our data as well. After a case fatality rate high of 27% in 2002, the case fatality rate for firearm homicide/assaults reached their lowest point of 18% by 2012.
I find this an interesting narrative, because it sort of pushes back against some common views behind the changing decades. It suggests that—though America is indeed safer than it once was—our improvements in medical treatment have masked or hidden the high amount of crime that still happens. Homicide and murder are just more salient to our society than non-lethal assault, so this changing composition in the type of crime removes it from the headlines.
Mortality Improvements have also Masked the Costs of War
Tanisha Fazal has found similar trends in war. A conventional narrative, emphasized by Steven Pinker and others, highlights declining rates of war. Part of this may be real—but part of it is that improvements in public health and battlefield mortality have lowered the body-counts resulting from wars (while still leaving a large imprint in non-fatal war injuries). Here is a telling plot from her work:
So we had something like a 3:1 wounded:killed ratio up until the 1950s, but a ~7:1 ratio in our most recent wars. So if the Vietnam war had the mortality rate of more recent wars, the US would have lost maybe closer to 30 thousand dead; rather than close to 60 thousand. Combat deaths in Iraq, by contrast, would be close to 9 thousand, rather than about 4,400, with the medical technology prevailing in Vietnam.
This turns out to have consequences in how we view the peacefulness of the post-WWII period. For instance, many counts of armed conflicts employ battle death thresholds that do not account for improved battle casualty rates, biasing down our conflict estimates. And public support for wars may be different with lower headline casualty numbers.
My colleagues Theresa Kuchler and Johannes Stroebel (along with Georij Alekseev, Safaa Amer, Manasa Gopal, JW Schneider, and Nils Wernerfelt) have a really interesting paper on using Facebook to do a large survey of business owners. This confirms some things you may have thought already—that businesses are facing huge cash flow shocks, female employment is disproportionately affected, and firms are doing more work online. One interesting finding is that worse financial conditions lower product prices. More broadly I think this illustrates the strength of the Facebook platform to ask really interesting surveys in close to real-time.
There is a nice Op-Ed here by Stijn Van Nieuwerburgh and Ralph Koijen about the complementarities between health and life insurance. Coronavirus shows a huge problem with tying health insurance to employment—in a pandemic recession, health coverage evaporates exactly when it is most needed. Stijn and Ralph point out in another excellent paper that we already have another insurance entity that is in a much better place to internalize investments in longevity—life insurance companies. So allowing individuals to tap into their life insurance policies a bit would help them bridge temporary health insurance payments in a manner that helps everyone. And more broadly—transforming life insurance itself away from something that pays off only at death, towards something that is more like a generic savings vehicle for people could really add a lot of value.
We discussed last week some of the issues with NEPA, a major piece of national environmental legislation. The California equivalent is CEQA, and it appears that this legislation limits controlled burns and tree thinning. Our failure to do this forest management is the proximate cause of all of these West Coast fires, as tinder and brush builds up explosively. Also, CEQA by design ignores Zoning, because the assumption is that all zoning is environmentally beneficial. This is of course catastrophically wrong—because California makes it so hard to build on the coast, people have to move further inland, with longer commutes and greater impact on the local wildlife. So really a pretty horribly bad piece of legislation all around.
There has been some discussion about infrastructure as stimulus. A nice paper by Andrew Garin shows the limits of this approach—infrastructure projects associated with the ARRA seem to have had little spillover stimulus benefit. As infrastructure projects get more specialized and complicated, they stop being generic “shovel-ready” stimulus.
On the College Covid case issue—I think we are seeing basically all of the mass testing regimes work well across the US, and many of the no or little testing regimes really fail. Of course, students appear to have lower hospitalization and mortality risk from Covid (though the long-term effects remain uncertain). Still, the concern is that widespread prevalence among College students may ultimately spread back to the population. For now, it does look like case are rising nation-wide, and many of the places seeing the fastest increases are in College towns.