In 2004, a groundbreaking study 1 emerged, detailing a novel model to understand the global dissemination of Severe Acute Respiratory Syndrome (SARS), a newly identified infectious disease originating in China in 2003. As SARS spread globally, sparking public health concerns, Lars Hufnagel, then a postdoc at the Max-Planck-Institute for Dynamics and Self-Organization in Göttingen, Germany, recognized a critical pattern. He realized that the intricate network of global air transportation, with few exceptions, dictated long-distance human movement. Consequently, this network held immense influence over the worldwide spread of contagious diseases. Lars hypothesized that deciphering this extensive mobility network of national and international flights would enable the modeling of these dynamic phenomena, paving the way for predictive mathematical models and real-time projections, akin to weather forecasting. At that time, I, Dirk Brockmann, had recently completed my dissertation on anomalous diffusion processes at the same institute. Lars approached me, seeking my expertise in diffusion processes and random walks to collaborate on his compelling idea. Intrigued by the challenge, we embarked on developing a model for the global spread of SARS. Initially, I held reservations, believing the complexities of human infectious diseases and mobility patterns would hinder accurate predictions. However, to my surprise, a pivotal outcome of our SARS research (our SARS study) revealed that global spreading patterns are indeed significantly shaped by the global mobility network.
Shortly after publishing our SARS paper, a crucial realization dawned upon me. While our global approach was effective on a macro scale, it fell short in accurately depicting epidemic spread at intermediate and shorter distances. Within a single nation, neglecting shorter-range traffic—daily commutes by car, local public transportation, and medium-distance travel by trains—could lead to inaccurate results. I began contemplating methods to gather human mobility data across all spatial scales worldwide, aiming to create a comprehensive dataset capturing the nuances of every mode of human movement. My ambitious goal was to construct an international multi-scale mobility network, a task I knew was, at best, exceedingly difficult, and potentially impossible.
Engrossed in these thoughts, I attended a physics conference in Montreal. Post-conference, I decided to visit Dennis Derryberry, a college friend residing within driving distance in the picturesque Green Mountains of Vermont, where he worked as a cabinet maker. After a pleasant highway drive, Dennis and his family warmly welcomed me to their beautiful woodland home. During my visit, Dennis, known for his sharp wit, posed a question one evening as we relaxed on his porch with beers: “So Dirk, what are you working on?” “I’m fascinated by the underlying patterns of human travel,” I replied, elaborating on my endeavors to comprehend human mobility and our objective of developing more precise models for epidemic spread. “The challenge lies in compiling all this necessary data,” I explained, highlighting the significant hurdle in my research. Dennis paused thoughtfully, then inquired, “Do you know this website, www.wheresgeorge.com?”
I was unfamiliar with it. But in that instant, everything began to fall into place. I asked Dennis about the website, and he described it as a sort of online bill tracking system. That evening on the porch, I only grasped a vague notion of its purpose. The next morning, he showed me the WheresGeorge.com website, and in a flash, I understood its potential to resolve several of our most pressing data acquisition problems. WheresGeorge.com meticulously tracks the geographic circulation of individual dollar bills within the United States. A vast community of enthusiasts, known as “Georgers,” mark individual bills. When someone comes across a marked bill, they visit the WheresGeorge.com website and input their current zip code along with the bill’s serial number. Once the bill re-enters circulation, it can be reported again at a different time and location by another person, effectively charting the bill’s journey across the country. For each registered bill, these movements are logged, allowing for the study of individual bill trajectories and the finder-posted logs. Visualizing millions of these dollar bill journeys in my mind, I became convinced that analyzing this data would unveil essential properties of human mobility, the very engine driving the dispersal of banknotes. Dennis’s insightful question sparked a series of human mobility studies rooted in this ingenious concept, all thanks to WheresGeorge.com.
Upon returning to Göttingen after my Vermont visit, I immediately shared the revelation of WheresGeorge.com with Lars. We began discussing the feasibility of utilizing this data for our scientific research and decided to reach out to Hank Eskin, the creator of WheresGeorge.com, to inquire about accessing his data. We sent an email to Hank and eagerly awaited his response. In the interim, we thoroughly examined the publicly available information on the WheresGeorge.com website. We discovered that all the data we required was already accessible; the only missing element was an automated method for data collection. While awaiting Hank’s reply, Lars developed a small program designed to systematically scrape bill reports from the website. Each morning in our office, we monitored the increasing volume of reports downloaded from WheresGeorge.com.
Daily reach traffic on wheresgeorge.com demonstrating a significant peak on January 24th, 2006, coinciding with the publication of “The scaling laws of human travel.” Data sourced from Alexa.com.
The probability p(r) of a dollar bill traveling a distance r within a short timeframe follows a simple mathematical relationship known as a power law. We calculated this power law from over a million dollar bill trajectories across the United States, data sourced from WheresGeorge.com.
Meanwhile, Hank noticed unusually high-frequency visits to his website originating from our program and detected that someone from Germany was extracting data. Unaware of this at the time, we were surprised when one day we could no longer access WheresGeorge.com from our office computers. Initially, we suspected a local network issue, but that wasn’t the case. As a precautionary measure, Hank had blocked access to the entity he identified as reading out the data. In fact, his caution extended to blocking access from the entire city of Göttingen. We were understandably disappointed but realized we were likely responsible for this access denial. However, we had already successfully downloaded over a million individual dollar bill trajectories from WheresGeorge.com, which proved sufficient for our initial analysis. For our preliminary investigation, I decided the most straightforward and informative metric to compute was the probability of a bill traveling a specific distance within a day. Initially, I was quite skeptical and didn’t anticipate any discernible pattern. Imagine my excitement and surprise when I discovered that this probability adhered to a very simple mathematical law (this probability follows a very simple mathematical law)! Intuitively, longer journeys, such as 1000 miles, are less frequent than shorter trips of a few miles. Yet, the specific manner in which this probability diminishes with distance followed a remarkably simple relation – a power law. Drawing from my work on anomalous diffusion, I recognized the profound implications: the dispersal of dollar bills is scale-free, self-similar, and fractal. I was thrilled to uncover these fundamental mathematical laws governing dollar bill movement, and it became apparent that further simple patterns related to mobility were embedded within the WheresGeorge.com data. We synthesized these discoveries into a manuscript that was published in early 2006 2.
This publication garnered immediate attention from the mainstream press (immediate response from the mainstream press), and shortly after its release, Hank Eskin observed an extraordinary surge in traffic to his WheresGeorge.com website. He faced an overload of requests and was also contacted by journalists inquiring about the German scientists utilizing his data in their human mobility and disease spread study. He quickly realized that these were the same German researchers who had contacted him over a year prior. Hank, along with many “Georgers”, were thrilled that WheresGeorge.com had become central to a study that, for the first time, mathematically analyzed human mobility across distances ranging from a few to thousands of miles. This website, WheresGeorge.com, had facilitated the discovery of the scaling laws of human mobility and showed promise in enhancing pandemic disease forecast models. Leveraging WheresGeorge.com data, we estimated multi-scale human mobility networks. These networks served as the foundation for our computational model predicting the most likely progression of the H1N1 pandemic (swine flu) in the United States in early 2009.
Following the publication, Hank contacted Lars, and we have maintained close communication ever since. Hank generously provided his complete dataset for our research, which now forms the core of more advanced projects (more sophisticated projects). Notably, in April 2009, we utilized WheresGeorge.com data to model the spread of swine flu across the United States (model the spread of swine flu through the United States) and generated projections of its temporal spread. These large-scale computer simulations would have been impossible without WheresGeorge.com. We continue to investigate the structure of human mobility using bill tracking data and are optimistic that WheresGeorge.com data will reveal further insights. We extend our deepest gratitude to Hank Eskin for his generous data provision, to the vast community of Georgers who have contributed to this data over the years, and lastly to Dennis Derryberry, the cabinet maker and friend whose insightful question on a Vermont porch at the perfect moment sparked this entire line of inquiry regarding WheresGeorge.com. (Ed.: Daniel Grady) Download PDF of this page
[1] L. Hufnagel, D. Brockmann and T. Geisel: Forecast and control of epidemics in a globalized world**. Proc Natl Acad Sci USA 101**, 15124 (2004). pdf
[2] D. Brockmann, L. Hufnagel and T. Geisel: The scaling laws of human travel**. Nature 439**, 462 (2006). pdf
