Biological scientists have developed theories, tools, and methods to better understand why some species flourish and others do not. Published in peer-reviewed and popular journals, as well as books, a steady stream of biological publications thoughtfully consider population density (Andren, 1992; Hutchinson, 1991; Kingsland, 1995; Tarsi and Tuff, 2012). Yet the use of human population density as a variable in geographic research lacks similar development. The variable is not ignored but rather not fully explained, often leading to misunderstandings of the results.

For example, infectious disease studies, including those documenting the recent COVID-19 pandemic and related threats, include population density as a variable. But what it means in these studies has been problematic. For example, Keegan (2020) quoted former New York State Governor Andrew Cuomo's declaration that New York City's population density was too high, more specifically, the former governor said it was “destructive” and needed to be reduced. Keegan cautioned that areas with similar population densities had markedly different rates and overall density was not a predictor of incidence or death rates after initial exposure. Cuomo's remarks were heard by thousands, Keegan's were not.

At the onset of the pandemic, several of the first author's students were confused by what they were told about population density. New Jersey, the most densely settled state, and New York City, the most densely populated major city, had the highest infection and mortality rates in the nation. The students first heard that this situation was due to population density. Later, they were told that New York City and New Jersey were global entry points for the disease and that the region's mobility hubs served to spread it, not that the high rates were because of population density. What population density had to do with the spread of the epidemic was, to them, confusing at best.

The above examples are not unique. More often than not, we find ourselves disappointed by the lack of explanation of why population density is or is not included in health, economic, social, or political studies. Too often, it is used as a surrogate variable without adequate explanation. Given that much of the recent literature focuses on the COVID-19 pandemic, we illustrate the inconsistency in the use of population density as a surrogate variable in the literature below.

Starting with a good example, Hsu (2020) carefully explained that population density does not mean higher infection rates, pointing to public policies, personal behaviors, socioeconomic status, crowding, health care, transit, and other drivers of infection spread. In contrast, Biggs et al. (2021) analyzed the relationship between social vulnerability and COVID-19 incidence among Louisiana's census tracts, finding a positive statistical relationship. The authors indicate that population density (number of individuals per unit of space) may be a confounder and thus controlled for it in their analysis. However, population density is not a good surrogate variable for crowding, which is the product of not only density, but communications, contacts, and other activities such as mobility (Boots, 1979). For us, conflating population density with crowding weakens an analysis.

Another good example is Hazarie et al. (2021) examined the spread of SARS-COV-2 across four continents in 163 countries. The authors found that hotspots of disease were associated with far more than population density alone. In essence, they found that the key was to reduce the flow of people through hotspots by using telecommunicating. In contrast, Carnegie et al. (2022) examined the relationship between population density and non-communicable diseases in developed western nations. Using 54 studies published since 1990, the authors conclude that higher population density is associated with multiple kinds of cancer, chronic obstructive pulmonary disease, cardiovascular disease, asthma, and club foot. Conversely, they found low population density associated with diabetes. The authors concluded that population density is a risk marker for these diseases but note that upon “closer examination” it may be a surrogate for greenspace, the built environment, and man-made exposures. We believe that such post hoc theorization is problematic. Researchers should clearly lay out about why population density is included prior to their analyses, not after.

Three papers we found extremely thoughtful about the use of a population density variable are briefly noted here. Malpezzi (2013) wrote about changing population density and the significance of such changes for urban living, health, and other outcomes. Belikow et al. (2021) examined the relationship between population density and subjective rating of well-being using a survey of over 4000 residents of Montreal, Canada. Finding density to be a correlate, they also carefully described the potential impacts of noise, access to friends, and local amenities. Hanlon et al. (2012) examined maternal health services in 178 countries, finding that coverage increases with population density, which makes sense from a marketing perspective. The Hanlon et al. literature review is critical, reporting that some studies use population density as a dichotomous indicator whereas in others it is a continuous linear measure or other mathematical forms. Most importantly, the authors characterize some of the literature as treating population density as an “afterthought, rather than a determinant of interest.” It is this last observation that strongly resonates with us.

The afterthought observation about population density is an invitation for journalists, elected officials, and the general public to draw potentially misleading conclusions that can potentially lead to bad decisions.

After discussing whether population density is a direct or indirect driver of quality of life, we address the following questions.

How is population density associated with demographic, environmental, and institutional factors? Do these relationships change with geographic scale?

Does the association between population density and human quality of life change after accounting for other contributing factors? If yes, at which geographic scales?

We selected two variables that measure how well and how long humans live: access to broadband services (Tomer et al., 2020) and length of life (National Center for Health Statistics, 2022) for this study. We also offer three theoretical underpinnings for evaluating the relationship between human population density and quality of life.

marketing targets high density wealth areas,

socioeconomic status, race, ethnicity, age, and other factors affect access to services that contribute to human health outcomes,

policy decisions impact human quality of life.

One reason behind the expectation that population density is directly associated with human quality of life was originally developed by regional economists and economic geographers (Christaller, 1966; Losch, 1954; Berry, 1967). These authors definitively showed how high value goods and service sales centers would locate in urban places where they could reach many customers. For example, those interested in purchasing high value garments, jewelry, and other expensive products would find the widest variety in New York City, Chicago, Los Angeles, and other major urban centers (Berry, 1967).

The marketing literature has since evolved to focus on the density of wealth. In the United States, wealth creation began with the industrial revolution in the Northeast and along the Great Lakes, particularly in New York City, Philadelphia, Cleveland, Chicago, Detroit, and Buffalo (Pred, 1966). Here, grand avenues, communication and health services concentrated around clusters of wealthy residents, along with amenities such as America's first zoos, major museums, and urban parks (Cigliano and Landau, 1994). The telephone became the first critical device for marketing to these individuals. Whereas over 98% of Americans have at least one phone today, in 1920 only 35% had one (the cost was $10,000 in current dollars). Click Americana (2022) reproduced an article from 1927 reporting that the one-third of Americans who had a phone purchased two-thirds of the advertised goods (71% of the package cereals, 79% of the automobiles, 74% of the phonographs, 79% of the vacuum cleaners, 86 of the oil heaters). Reporting on surveys of people in cities, the article suggests that phone ownership was the best metric to track diffusion of markets.

After the Second World War, population shifts changed marketing tactics and investment in major parts of central cities declined while those targeting suburban markets increased (Berry, 1967). Movement of people, jobs and services continues today. Richard Florida (2002) shows that young and well-educated people are attracted to places with high wages, opportunities for social interactions, and good housing opportunities in rapidly growing cities like Austin, Overland Park, and San Diego. Where gentrification takes root and replaces poor populations in places like Philadelphia, San Francisco, Washington DC, New York City, there is also a rebound of wealth and services but with major negative impacts on displaced people and neighborhood culture. Overall, it follows that the providers of today's broadband access would first market their services in urban areas with large pools of potential well-paying customers before expanding offerings to rural areas or poor neighborhoods in urban centers. The Biden administration (NTIA, 2022) has made it a key part of its infrastructure quality-of-life program. State and local governments have also been trying to find grants to boost access to broadband in poor areas (Office of Regulatory Safety, South Carolina, 2021).

Similarly, length of life for overall populations tends to improve when overall access to health and other social services improves. But this relationship is strongly shaped by variations in socioeconomic status, race and ethnicity, age, and other factors that affect access to health services. Hence, we need to explore the extent to which population density is an indirect factor in access to the services that improve longevity. We limit our use of this variable to 2019 and earlier, however, as length of life data was notably impacted by the pandemic beginning in 2020, resulting in rapid shifts in both population mobility and the need for broadband access. Thus, including 2020 and 2021 provisional data might produce skewed results (Murray et al., 2013; Arias et al., 2022).

Government policy provides the third connection between population density, broadband access, longevity, and other services related to quality of life. Over the course of the twentieth century, strong political relationships were built in urban areas between elected officials, entrepreneurs, and labor that lead to investments in public education, health care, entertainment, and many other services. Powerful unions centered in populated cities exerted enormous pressures on mayors and city councils, governors and state legislatures, and the national political network to push through innovative social and environmental programs (Brinkley, 1996; Bondi, 2003). The powerful city-dominated politics of the mid-twentieth century has since diminished. Sanchez (2020) credits San Francisco, Boston, and New York City politicians for fighting for a progressive national platform, but notes that they are unable to manage the growing gap between the rich and poor in their own cities. Greenberg (2022) also shows that parts of these cities have become places for exclusive wealth and health clusters that have displaced the poor. Pew (2021) evaluated the current Red vs. Blue political divide, concluding it emanates from various types of group activism developed during the growth of cities, the New Deal, and the new Federalism under Richard Nixon. These groups left indelible markers that manifested in a relationship between high population density and progressive public policies. Given this reality, we expect population density to be directly associated with important services like broadband access and opportunities for improved length of life for populations.

Geographical scale complicates testing the explanations of these relationships between population density and two quality of life outcomes. The state is the most problematic scale because state data mix what are otherwise spatially concentrated phenomena. For example, at the state scale we expect a moderate to high correlation between population density and the proportion of Blacks and median household incomes, even though these factors may not concentrate near one another. County data may or may not reduce this problem as their populations are not usually equally distributed. While some counties may be totally rural, others may have one or more urban places within them. In some instances, the urban place takes up only a small part of the county area (e.g., Tacoma in Pierce County, Washington) while in others the county-city boundaries are congruent (e.g., Louisville with Jefferson County, Kentucky). Sometimes a county may share a large city with one or more other counties (e.g., Austin, Texas in Williamson, Travis and Hays Counties). This makes county data for population density somewhat challenging for the purpose of hypothesis testing.

The most accurate testing of the relationship between population density and other metrics is expected to be shown at the municipal and neighborhood scales where the drivers of outcomes are less clouded by spatial mixing and the diluting of phenomena. However, the closest scale to a neighborhood is the census tract, which is problematic because census tracts markedly vary in shape and population. In a large city we could compare tracts that are shaped like a rectangle with over 50,000 residents with others that look like snakes and have a population of 1050. In order to standardize the shapes and make the approach replicable, we estimated neighborhood population densities by drawing fixed radius circles around selected points and estimating the required information within the standardized circles from census tract data using a mapping tool called EJScreen as described below (U.S. EPA, 2022).