Stanford Computer Model Predicts How COVID-19 Spreads in Cities

As COVID-19 cases increase across the country, city officials have been given the difficult balancing act of preventing the spread of infections and supporting businesses. A computer model from Stanford University demonstrates mobility and contact patterns in a way that its creators hope will help to guide community leaders' decision-making.

The Stanford team says that their model’s specificity could serve as a valuable tool for officials, as the simulation reveals the tradeoffs between new infections and lost sales if establishments open at limited capacities.

A major conclusion: According to the model (and the above video from Stanford University), capping occupancy at 50 percent of the maximum will lead to the economy losing only 5 to 10 percent of visits, while reducing the overall number of infections by over 50 percent.

Using anonymized large-scale data from cell phones, the Stanford team analyzed movement patterns in 10 of largest metropolitan areas in the United States, including Atlanta, Dallas, and New York City — a group totaling over 98 million people.

The computer model accurately predicted the spread of COVID-19 in the ten major cities this spring by analyzing three factors that drive infection risk: where people go in the course of a day, how long they linger; and how crowded the places get at one time.

A small percentage of infections at points of interest," it turns out, account for a large percentage of infections.

The study, published this month in the journal Nature , used a combination of demographic data, epidemiological estimates, and anonymous cellphone location information, to predict that most COVID-19 transmissions outside the home occur at “superspreader” sites," where people remain in close quarters for extended periods.

“We built a computer model to analyze how people of different demographic backgrounds, and from different neighborhoods, visit different types of places that are more or less crowded. Based on all of this, we could predict the likelihood of new infections occurring at any given place or time,” said Jure Leskovec , a Stanford computer scientist and lead researcher.

Leskovec and his team concluded that density caps, or restricting the occupancy of establishments, reduces infections overall, as well as disparities between communities impacted by COVID-19. The model suggests that mobility patterns drive disproportionate risks.

"It turns out that low-income groups are more likely to frequent places in which densities are high," said study co-author David Grusky, a professor of sociology at Stanford’s School of Humanities and Sciences (in the above video). "For example, grocery stores in low-income neighborhoods tend to be higher in density, and tend to be more crowded."

A new computer model predicts the COVID-19 infection vs. activity trade-off for Chicago. According to the figure, COVID-19 infections will rise as the number of visits to businesses and public places approach pre-pandemic levels. However, restricting maximum occupancy can strike an effective balance; for example, a 20 percent occupancy cap would still permit 60 percent of pre-pandemic visits while risking only 18 percent of the infections that would occur if public places were to fully reopen. (Image Credit: Serina Yongchen Chang)

Grusky, who also directs the Stanford Center on Poverty and Inequality, said the model demonstrates how reopening businesses with lower occupancy caps tend to benefit disadvantaged groups the most.

“Because the places that employ minority and low-income people are often smaller and more crowded, occupancy caps on reopened stores can lower the risks they face,” Grusky said. “We have a responsibility to build reopening plans that eliminate – or at least reduce – the disparities that current practices are creating.”

How Stanford Gathered the Data

SafeGraph, a company that aggregates anonymized location data from mobile applications, showed the Stanford modelers where people went; for how long; and, most importantly, what the square footage of each establishment was so that researchers could determine the hourly occupancy density.

The cities in the Stanford study included New York, Los Angeles, Chicago, Dallas, Washington, D.C., Houston, Atlanta, Miami, Philadelphia and San Francisco.

In Phase one of the study, from March 8 to May of this year, mobility data was used to predict transmission rate of the coronavirus. In their model, after incorporating the number of COVID-19 infections reported to health officials each day, the researchers developed and refined a series of equations to compute the probability of infectious events at different places and times.

The predictions tracked closely with the actual reports from health officials, giving the researchers confidence in the model’s reliability.

The team, which included PhD student Emma Pierson, has made its tools and data publicly available so other researchers can replicate and build on the findings.

In a short Q&A below, Pierson tells Tech Briefs why the model suggests that a reopening strategy does not have to be "all-or-nothing."

Tech Briefs: With the model itself, what kind of data is being collected that allows a kind valuable "specificity," especially compared to existing modeling methods?

Emma Pierson: We use anonymized, aggregated data from SafeGraph, a company that tracks human movement patterns using cell phone data. Our data records how many people go to points of interest (POIs) like restaurants and grocery stores at every hour, and also records the neighborhoods they come from.

Our analysis is based on data from ten large U.S. metro areas from March to May 2020 (the first wave of infections). This fine-grained mobility data allows us to model who is infected, where they are infected, and when they are infected.

Tech Briefs: What was the most important conclusion, do you think, that was drawn from your model?

Emma Pierson: There are a number of conclusions that flow from our analysis, but two of the most important are:

  • Reopening does not have to be “all-or-nothing”: strategies like reducing maximum occupancy can enable us to reopen more efficiently by providing a large reduction in infections for a relatively small reduction in visits.
  • Our model also suggests that racial and socioeconomic disparities are driven in part by mobility: they’re not inevitable, but can be influenced by short-term policy decisions. Therefore, in evaluating reopening strategies, it’s important not just to consider the impact on the population as a whole, but also the impact on disadvantaged groups. This supports steps being taken by California and the Biden-Harris transition team to specifically consider the impact of reopening policies on disadvantaged populations.

Tech Briefs: How can officials use your model most effectively?

Emma Pierson: The two findings above are directly policy-relevant, and help us develop more effective and equitable reopening strategies. We are also building an online tool that can allow policy-makers and members of the public to interact with and learn from our model. Finally, we are working on extending the analysis on more updated data, since the original analysis is based on data from the spring, and many things have changed since then.

What do you think? Share your questions and comments below.