Intensification of the Covid-19 pandemic prompted politicians from all over the country to proclaim the obvious– the districts they represent are not New York. Usually, the denial was issued to justify weak social-distancing policies or moves to “reopen” their economies.
Kay Ivey, Governor of Alabama, defending her resistance to a statewide stay-at-home order, said “We’re not New York. We’re not even Louisiana.” South Dakota Governor Kristi Noem defended her policies with “South Dakota is not New York City.” California’s Gavin Newsome observed, “We’re not New York…there are very different conditions in the state of California.”
Even New Jersey Governor Phil Murphy felt compelled to note that his state is not New York–it’s worse! Murphy: “You know, we’re not New York. Somebody reminded me yesterday that if you drove from New York City to the Canadian border, it’s a 10-hour drive. It’s hard to get more than a three-hour drive in New Jersey.”
Sometimes the not-New-York disclaimer has a moralistic tinge or political edge, but more often it seems to be an allusion to New York’s assumed higher risk profile, and more specifically, to its greater population density. Vita G., a beachgoer in Florida, observed “I think we’re doing the right thing and we’re not high risk. We’re not New York.” Matt Tompter, a brewer and restauranteur in Anchorage, offered “We’re not New York City….the reason Alaska is able to open right now is we are naturally socially distanced.” David Morgan, Sheriff of Escambia County, Florida, was even more explicit. “We’re not New York City. We don’t have the density of population they have there.” Even New York’s Governor Cuomo bought into the density argument: “Why New York? Why are we seeing this level of infection? It’s very simple: It’s about density.”
From Governors on down, people in other parts of the country seem to think that because their home states or towns have a lower population density than New York City, their risk of a coronavirus outbreak is lower. Is that so? And does the new age of pandemic abruptly end the era of Superstar Cities?
The initial outbreak of the pandemic, with its epicenter in New York City, certainly triggered an early rush to blame density as a principal risk factor. In March, The New York Times ran an article entitled “Density is New York City’s Big ‘Enemy’ in the Coronavirus Fight.” Joel Kotkin, in the Los Angeles Times, wrote “…employment and housing patterns and transit modes appear to be very significant, if not decisive, factors” behind the differing coronavirus death rates in LA and NYC, contending that the pandemic vindicates LA’s sprawl. In USA Today, Glenn Harlan Reynolds simply concluded that “density kills.”
Scientific opinion has been more cautious, although some epidemiologists, public health experts, and economists have fingered density as one of the culprits. Steven Goodman, an epidemiologist at Stanford University, was quoted by the Times as saying: “Density is really an enemy in a situation like this. With large population centers, where people are interacting with more people all the time, that’s where it’s going to spread the fastest.” In the same article, Thomas Friedman, the former director of the CDC, said: “New York City is often the first to get hit because of how dense it is, and how many international travelers come through.” And in a working paper that attracted much attention, Jeffrey E. Harris of MIT linked the spread of the coronavirus in the City to its subway system, finding evidence of “subway-facilitated disease propagation.”
Other urban planners and public health experts are more skeptical about the role of density. Alon Levy, a mathematician and planner at NYU, attacked the methodology of the Harris paper. Economists at the World Bank argued that very dense cities like Singapore and Seoul contained the virus very well, and used detailed data on Chinese cities to conclude that “density is not an enemy in the fight against the coronavirus.” Two researchers from the Nature Conservancy and the San Francisco Bay Estuary Institute analyzed county-level data and concluded that “there is, at best, a weak relationship between density and the spread of Covid-19.”
The science journalist Adam Rogers and the urban planner Elek Pafka argued that broad metropolitan densities are not important, but density at the level of human interaction as in, say, offices or homes, is. Or as Pafka put it, it’s a matter of internal density, not external density. Researchers at NYU’s Furman Center provided independent support for that perspective by analyzing zip code level data for New York City. They found that higher rates of Covid-19 cases are not associated with neighborhood-level population density, but are associated with higher levels of housing unit crowding. The NYU researchers also found, as other experts have argued, that virus cases are associated with other factors such as the racial and ethnic makeup of neighborhoods and residents’ ability to work from home.
Intuitively, I find the assumption that density contributes to the spread of contagious diseases a natural one. Moreover, high external density necessarily implies greater internal density–smaller homes, more office employment, and especially, a greater reliance on mass transit. However, I don’t find the empirical data on a connection to density convincing. Also, I wonder whether a contagion spreading faster, as Drs. Goodman and Friedman suggest, is the same as spreading more extensively.
Most of the aforementioned opinion and research relies on the rapidly accumulating data on Covid-19 infections and mortality. Since the coronavirus is believed to spread through a mechanism very similar to influenza (by droplets emitted when an infected person coughs, sneezes or talks), however, I began to wonder if flu data could shed any additional light on the density-Covid question.
Since most bouts with the flu are not reported and don’t result in hospitalizations, it’s difficult, as it is with the coronavirus, to know exactly how many people are infected each year. The CDC’s current thinking is that about 3 percent to 11 percent of the U.S. population is infected with symptomatic influenza each year, depending on the severity of the flu season. The disease is estimated to generate 10- to 20 million medical visits annually and to result in about 150,000 to 800,000 hospitalizations, again depending on the severity of the season. U.S. deaths from influenza are estimated to have ranged from 12,000 in the 2011-2012 flu season to 61,000 in the 2017-2018 season. Serious cases requiring hospitalizations can involve a number of medical complications, but the most frequent one is bacterial pneumonia, and flu deaths are reported by the CDC as influenza/pneumonia deaths.
Unfortunately, the CDC does not report estimates of flu-related medical visits or hospitalizations by state, so I had to make do with state-level data on influenza/pneumonia deaths. The disadvantage is, obviously, that deaths don’t fully track the spread of the disease, being affected by a host of other medical and social factors. The statistical benefit is, as with many other medical and social indicators, terminal outcomes are typically those most reliably recorded. The CDC reports both the aggregate number of annual flu/pneumonia deaths by state as well as an age-adjusted death rates rate per 100,000 residents.
I looked most closely at the period encompassing the 2013-2014, 2014-2015, and 2015-2016 flu seasons. The death rates during that period ranged from a low of 7.0 per 100,000 residents in Vermont in 2015-2016, to a high of 27.4 in Hawaii in 2014-2015. Despite such wide variation in death rates, there is considerable–and puzzling–regularity in them. As the density argument predicts, New York is typically among the states with the highest flu/pneumonia death rate, averaging 19.3 (per 100,000) over the 3-year period. However, it is not the highest. Hawaii, Mississippi, Tennessee, Nevada and West Virginia all had higher 3-year death rates. Although there does seem to be a cluster of high-impact states in the border and Deep South (Georgia, Alabama, Kentucky, and Arkansas also tend to have high rates), it’s hard to think of what common characteristics might be shared by Hawaii, Mississippi, Nevada and New York. The states with the lowest death rates from influenza/pneumonia are also a motley crew: Vermont, Minnesota, Florida, Oregon, Washington, and Arizona. It’s difficult to identify in what ways those states are alike; arguably, Florida is the antithesis of Vermont.
I found those patterns intriguing enough to be motivated to collect more data and run a few regressions. The dependent variable was the 3-year average of state-level influenza/pneumonia death rates. To measure density, I used thousand residents per square mile in each state’s densest county. That measure of density was used since some states, such as Nevada and Arizona, are sparsely populated overall but a majority of their population lives in a single, relatively dense county. Since flu contagion tends to abate as Winter ends, I measured climate by the average daily high March temperature in each state’s densest city. Other explanatory variables evaluated were each state’s poverty rate, it’s adult flu vaccination rate, and a dummy variable indicating whether the state had opted into the ACA’s Medicaid expansion. To quantify the concept of “internal density,” I used the share of total civilian employment in the largest office-using industries (Information; Finance; and Professional, Scientific, and Business Services).
As cross-sectional regressions with 51 observations go, I’ve run worse. In the compact version the R-squared is .26 and the three explanatory variables are all significant at the 95 percent confidence level. The density variable is positive, as intuition might suggest, but the effect is modest. It suggests that a 10,000-person per square mile increase in the density of the state’s densest county is associated with a 1 person increase the flu/pneumonia death rate. That’s roughly the density difference between Dallas County, Texas (Dallas) and Suffolk County, Massachusetts (Boston).
Surprisingly, a 10-degree increase in average March temperature adds about the same– one person per 100,000– to the flu death rate. The March temperature difference between Dallas and Boston is more than twice that. That result contradicts the common assumption that influenza is less of a threat in warm climates as well as the indisputable seasonality of flu infections. I do not have a theory as to why this analysis finds a positive association between warm climate and flu-related deaths. Skeptics who might argue that warm climate proxies for poverty rate, the percentage of the population who are African-American, or the willingness of state governments to expand Medicaid access should note that those variables were tested with little effect on the climate coefficient.
An even bigger surprise was the result for the percentage of the workforce employed in office-using industries. If office employment increases “internal density” and hence the exposure of a state’s residents to flu infection, the coefficient should be positive. Instead, it is strongly negative and statistically significant. It suggests that a 5-percentage point increase in the share of office workers in the labor force–roughly the difference between New York and South Carolina–lowers the flu death rate by 2.1 persons per 100,000.
Skeptics might, once again, conclude that the percentage of office workers is a proxy for something else, perhaps the education level of the population. That was my reaction too. However, substituting the percentage of a state’s residents with a BA or higher degree for the office worker share produced much weaker results. So what is it that’s causing the office worker effect? I don’t know.
Even if the novel coronavirus does spread in a manner similar to the flu, this analysis is hardly definitive. In fact, at best it hints at a few patterns we might consider in contemplating the course of the present tragedy and what it means for big, dense cities like New York and for mostly urbanized or rural states. Yes, density may increase infection and death rates, but the effect might not be as large as is commonly assumed. Moreover, highly urbanized states may have other, as yet unidentified characteristics, that offset some or all of the impact of high population density.
In the end, many states may find that though they’re not New York, that might not protect them as much as they hope.