Communities Most Vulnerable to COVID-19 are Slower to Adopt Social Distancing

Mobile phone data suggests that people in areas that will struggle most to cope with the spread of COVID-19 have reduced their movements the least.

By now, many of us are familiar with the practice of social distancing (SD). Reducing close contact with people outside our own households is one of the most important ways of stopping the spread of COVID-19.

While it is essential that everyone practice social distancing to the best of their abilities, regardless of where they live, it is absolutely critical in highly vulnerable communities.

In communities that are already plagued by under-resourced healthcare systems, pervasive poverty, high rates of chronic conditions, and other compounding risk factors, a COVID-19 outbreak can be all the more destructive.

We quantified and mapped this community vulnerability by developing the COVID-19 Community Vulnerability Index (CCVI), which we published last week, assigning a vulnerability score to each county in the United States.

We have already uncovered that the virus spreads faster in vulnerable communities once it reaches them than it does in less vulnerable ones. But what does social distancing look like in highly vulnerable communities compared with less vulnerable ones, and what does this mean for the future impact of COVID-19 in these areas?

To explore the relationship between vulnerability and social distancing, we used Unacast mobile phone location data to track the amount of social distancing across the country, and combined it with the CCVI map. We discovered that social distancing adherence is uneven across the nation: there’s wide variability at regional, state, and even county levels. Although many communities were quick to adopt social distancing, even before seeing a large number of confirmed COVID-19 cases, these early adopters tended to be communities least vulnerable to the impacts of the pandemic. In disheartening contrast, communities that are particularly vulnerable to COVID-19 have been among the slowest to social distance consistently.

Let’s examine the data in detail.

1. There are geographic differences in uptake of social distancing.

Before diving into the data, a quick explanation of how we measured social distancing is needed. Unacast used mobile phone location data to create a county-level scorecard, showing the percentage change in phone users’ movements compared with pre-pandemic days. This is a reasonable proxy for social distancing, because if a phone user is changing location throughout the day there’s a reasonable chance that they are coming into contact with other people (e.g., they’re working, shopping, or socializing) and are therefore not adequately social distancing. At the time of writing, we have social distancing data up to April 8.

The map below shows the amount of social distancing by county on March 18, five days after the White House declared the COVID-19 outbreak a national emergency. We consider counties that have reduced movement by more than 55% to be “highly social distancing”, those that have reduced by 25% “moderately social distancing”, and those reduced less than 25% “low social distancing.”

We can see that by March 18, the vast majority of counties had yet to significantly modify their behaviors, although many counties in the Northeast were already moderately social distancing. Certain states, such as Colorado and North Dakota, showed significant variability among their counties in the levels of social distancing.

Social Distancing on March 18. Counties in white are those where social distancing data is unavailable.

Fast forward three weeks to April 8, and the map now tells a different story. We see that much of the country is now practicing social distancing, albeit to different degrees. Nearly all of the Northeast has seen a reduction in movement at 40% to 70% (moderate to high social distancing) on April 8.

Still, only in a small number of counties do we see movement reduced by 55% or more (high social distancing), mostly in counties home to major urban areas such as New York City, Seattle, and San Francisco.

By comparison, in Southern states social distancing remains low, with reductions in movement of only around 20% to 40% (low to moderate social distancing).

Social distancing on April 8. Counties in white are those where social distancing data is unavailable

2. The most vulnerable states and counties are doing less social distancing.

While all states have, on average, seen increases in social distancing, these increases are unequal. When we overlaid social distancing data with our vulnerability index (CCVI), we found that those counties practicing less social distancing are also the most vulnerable.

For example, significant parts of the South and West rate as highly vulnerable on the CCVI, and these same vulnerable communities are also the places where there is less social distancing compared with the rest of the country.

(Note: On the maps below, we’ve reduced the CCVI’s five categories of vulnerability to three to make it easier to read. Each category — Low, Mid and High vulnerability — contains one-third of all US counties.)

CCVI + Social Distancing on March 18. Counties in white are those where social distancing data is unavailable

By April 8, we still see significant variation in social distancing rates within states (the graph below).

For example, on April 8 in Alabama, social distancing is at 40% reduction in movement in Jefferson, Shelby, and Madison counties. In contrast, Macon and Lowndes counties were not doing social distancing (no change in average amount of movement from before the pandemic), while Sumter County actually saw a jump in movement of +15%.

Despite its poor uptake of social distancing, Sumter County is much more vulnerable to the impact of COVID-19, with a CCVI of 0.975 (the highest possible score is 1), while better-performing counties like Madison are much less vulnerable: Madison County has a CCVI of just 0.252.

CCVI + Social Distancing on April 8. Counties in white are those where social distancing data is unavailable

3. Even as their case numbers grow, social distancing in vulnerable communities continues to lag behind less vulnerable ones.

The lag in uptake of social distancing among vulnerable counties becomes even more apparent when we take a look at the graphs below. Each county is plotted as a circle with a size proportionate to the number of confirmed COVID-19 cases per 100,000 residents in that county. Counties toward the top of the graph are more vulnerable (they have a high CCVI). Counties toward the right (negative %) are doing more social distancing. The vertical line represents the national median level of social distancing (i.e., half of all counties are to one side of the line, and half to the other).

The inverse relationship between vulnerability and social distancing rates emerges as early as March 18, before many counties in the US had seen any confirmed cases. We see from the cluster of circles to the lower right side of the line that less vulnerable counties were practicing above-average social distancing compared with the more vulnerable counties (the circles on the upper half of the graph, to the left of the line).

CCVI v.s. Social Distancing v.s. Confirmed cases per 100,000 residents on March 18

One week later (March 25), several counties, such as those that make up New York City, that had seen a larger number of confirmed cases per 100,000 residents (larger circles), were doing significantly more social distancing than the rest of the country.

CCVI v.s. Social Distancing v.s. Confirmed cases per 100,000 residents on March 25

By April 8 (the graph below), we see an explosion of confirmed cases across the country, including in areas that are vulnerable (note how much bigger the circles toward the top of the graph have grown). The average degree of social distancing also increased since March 18, from a 10% reduction in movement to 32%.

But worryingly, despite the fact that by April 8 most of the country was practicing some social distancing, the more vulnerable communities still lag behind.

The majority of counties toward the top of the graph are to the left of the median line, meaning they’re social distancing less than the national average.

For example, in Quitman County, GA, with a CCVI of 0.99 and where there were 4 confirmed cases by April 8, people actually are moving around the same amount as before the pandemic. In another example, in Terrell County, GA, with a high CCVI score of 0.93, confirmed cases have grown since March 18 from 0 to 93 (equivalent to 102 cases per 100,000 people), yet people are moving around 9% more on April 8 than on March 18.

CCVI v.s. Social Distancing v.s. Confirmed cases per 100,000 residents on April 8

Our finding that vulnerable communities are social distancing less is alarming. As the United States struggles to flatten the curve of new COVID-19 infections, effective social distancing may be our only hope. Less social distancing in the most vulnerable communities likely means more COVID-19 cases in the communities that will be hardest hit and least able to cope.

Stronger efforts by all levels of government and other leaders are needed to encourage and enable more people to practice social distancing to better protect these communities. In our next blog post, we’ll dive deeper and take a look at how different days of the week and other factors may contribute to the lack of social distancing in certain parts of the country.

This work was made possible by everyone at the Surgo Foundation, including but not limited to (in alphabetical order): Yael Caplan, Vincent Huang, Hannah Kemp, Tich Mangono, Sema Sgaier, Peter Smittenaar, Nick Stewart, and Staci Sutermaster.

Technical notes

  • Case data was retrieved on April 13th from JHU Github repository.
  • Cases are confirmed cases (including those that died, recovered, or are active; but excluding those never confirmed in a test).
  • JHU combines New York City’s five boroughs — Bronx, Queens, Kings (Brooklyn), Richmond (Staten Island), and New York (Manhattan) — into a single area when listing the number of confirmed cases. Manhattan has the lowest vulnerability among the city’s five boroughs, while some of the other boroughs have much higher vulnerability.
  • Want to analyze the numbers for yourself? The CCVI data is publicly available at census, county, and state level (link). The JHU data is available from Github.
  • Questions, concerns, comments? Get in touch at



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store