Two years ago, I started my term as Director of the Center for Urban Science and Progress (CUSP) at New York University’s Tandon School of Engineering. Filling the very big shoes of past Directors comes with a lot of stress, but also with rewarding opportunities. One of those opportunities was to expand the academic offerings of the center,  and to promote interdisciplinary training of engineering students. I launched a pilot Ph.D. Track on Urban Science, with methodological foundations in complex systems, data science, and sensing. These areas, in my view, encompass a lot of what an urban scientist may need to advance the science of cities and contribute technical advancements that could improve the life of urban communities.

While we were already strong on the latter two areas, we did not have specific courses on complex systems, which led us to create a  Complex Urban Systems class at CUSP. The preparation of the class was an amazing experience, and I cannot stress enough how lucky I was to have the timing of the class  sync with the publication of the book on the same topic by Luis Bettencourt. The book is just amazing — a refreshing look at an exciting area of research through the lens of network science, mathematical modeling, and statistical mechanics, just to name a few of the many disciplines on which Luis draws upon to advance the science of cities. I spent months immersing myself into the topic, and when the time came to teach, my students and I had a blast.

The topic that mostly drew my interest was urban scaling. I was stunned to learn that allometries would be so common in urban science. Even more surprising was to learn that socioeconomic variables, like GDP, would scale superlinearly with population size, in contrast with variables associated with infrastructure, like road surfaces, that would scale sublinearly with population size. All was aligning in favor of growing urbanization: the bigger the city, the higher the wealth and the lower the infrastructure needs per capita (at least on average). Is that really true? Is it always the case that the bigger the better? In my personal experience, things are generally more nuanced. A smaller strawberry might be tastier than a bigger one, or a smaller car may be more efficient than a big one. In fact, a deeper dive into the topic would reveal that criminality scales superlinearly with population size: crime per-capita is higher in big cities than rural areas.

The study of criminality is, indeed, what attracted me to CUSP five years ago, and is a key focus of my research.  I am particularly interested in firearm-related harm, a leading cause of death in the US. You can clearly see where this is going, but I promise there are some twists to the story. Right after the class ended, I approached Rayan Succar — an amazingly talented Ph.D. student whose training reminded me how much I enjoy my work — to work together on urban scaling of firearm violence, accessibility, and ownership in the US. We were pumped up after the class, and worked tirelessly over the summer on the project.

In agreement with our predictions, we discovered superlinear scaling of both armed robberies and firearm homicides. Surprisingly, the prevalence of firearm ownership and licensed firearm dealers scale sublinearly with population size — there are more guns and gun dealers per-capita in rural areas than in cities. Guns behave like some sort of infrastructure, akin to road surface area. The parallel may be more than circumstantial and rather telling of how the country views guns. For example, consider the debates and state-level actions at the beginning of the COVID-19 pandemic, when firearm-related businesses were deemed part of the critical infrastructure that would remain open.

These findings are part of our study published in Nature Cities, but they are not all of it. During the course of the class, Rayan tinkered with ideas to overcome the correlational nature of urban scaling towards insights into causal associations between urban variables. Throughout the preparation of the study, we refined these ideas into a working methodology to tackle causal associations in urban datasets. Our approach takes various urban features as inputs and outputs a directed acyclic graph encapsulating causal associations between the variables. The main ingredient for the analysis is scale-adjusted metropolitan indicators, measuring how much a particular city would deviate from the scaling law.

By applying the methodology to our dataset, we offer backing to the theory of self-protection as a driver of firearm ownership, whereby the prevalence of firearm ownership is associated with the incidence of homicides and the prevalence of licensed dealers. In other words, people will tend to buy guns under the fear of being the victim of violence, provided they have easy access to a dealer. Statistical backing to this theory was not found in previous studies on nation- or state-levels, supporting the value of granular analyses on city-level data.

Ahead of us, we have a lot of questions to address, mostly related to effective policy making in cities and scalability of the approach to other urban domains. We are thankful to the National Science Foundation for funding our work and to the students of the Complex Urban Systems Class for their enthusiasm and precious insights.