Neighbourhood income and physical distancing during the COVID-19 pandemic in the United States

Abstract

Physical distancing has been the primary strategy to control COVID-19 in the United States. We used mobility data from a large, anonymized sample of smartphone users to assess the relationship between neighbourhood income and physical distancing during the pandemic. We found a strong gradient between neighbourhood income and physical distancing. Individuals in high-income neighbourhoods increased their days at home substantially more than individuals in low-income neighbourhoods did. Residents of low-income neighbourhoods were more likely to work outside the home, compared to residents in higher-income neighbourhoods, but were not more likely to visit locations such as supermarkets, parks and hospitals. Finally, we found that state orders were only associated with small increases in staying home in low-income neighbourhoods. Our findings indicate that people in lower-income neighbourhoods have faced barriers to physical distancing, particularly needing to work outside the home, and that state physical distancing policies have not mitigated these disparities.

Extended Data Fig. 1 |. Proportion of smartphone users staying home all day by level of urbanicity.

Notes: Income quintile 1 represents the lowest-income group. Outcomes are presented as weekly averages. Period covered is January 6, 2020, through May 3, 2020. Levels of urbanicity are National Center for Health Statistics classifications. Sample comprises 210,288 census block groups with mean 89 active devices per block group per day.

Extended Data Fig. 2 |. Proportion of smartphone users staying home all day by region.

Notes: Income quintile 1 represents the lowest-income group. Outcomes are presented as weekly averages. Period covered is January 6, 2020, through May 3, 2020. Regions are U.S. Census Bureau classifications. Sample comprises 210,288 census block groups with mean 89 active devices per block group per day.

Extended Data Fig. 3 |. Visits to places of worship by weekday/weekend.

Notes: Visitor income is calculated for each place of worship (n = 97,379) based on visitor home census block group (BG) from January and February 2020. Median visitor income quintile is based on the median of household income values from visitors, weighted by the number of visits per BG. Unlike Fig. 4, this plot does not omit visits of > 4 hours, since data on visit duration were only available by week, not day/date. Counts were aggregated by week, weekday/weekend, and income quintile for this visualization.

Extended Data Fig. 4 |. Proportion of devices working outside the home, by income quintile and age composition.

Notes. Age composition is the proportion of residents within each age category, based on 2018 American Community Survey estimates, for (a) ages 15–17 and (b) ages 18–21. Metrics are aggregated by week, income quintile, and age composition. Cut points do not represent quintiles. Sample comprises 210,288 census block groups with mean 89 active devices per block group per day.

Extended Data Fig. 5 |. SafeGraph physical distancing metrics, 2019 vs. 2020.

Notes. Figures compare 2019 and 2020 SafeGraph physical distancing metrics. Metrics are aggregated by week and income quintile. a, Proportion of devices at home all day, January 7, 2019 through May 3, 2020; (b) Proportion of devices working outside the home, January 7, 2019 through May 3, 2020, including dashed lines representing Memorial Day, 4th of July, Labor Day, Thanksgiving, and Christmas holidays; (c) Ratio of devices at home all day, comparing January 6-May 3, 2020 to the same weeks of 2019; (d) Ratio of devices working outside the home, comparing January 6-May 3, 2020 to the same weeks of 2019. Sample comprises 210,288 census block groups.

Extended Data Fig. 6 |. SafeGraph work behaviors vs. Google workplace mobility data.

Notes. Comparison of SafeGraph work behaviors indicator (as used/explained in main manuscript) vs. Google COVID-19 Community Mobility Reports workplace visit metrics for the period of February 15, 2020, through May 1, 2020. SafeGraph data have been normalized using the same schema as Google data, as explained here: https://www.google.com/covid19/mobility/data_documentation.html?hl=en#about-this-data. Counties (n = 30) were randomly selected from counties with populations exceeding the median U.S. county population, because Google data were suppressed in some smaller counties and because estimates were expected to be more stable in larger counties. Google data were obtained from public sources using the tidycovid19 R package.

Extended Data Fig. 7 |. Proportion of devices working outside the home, by income quintile and unemployment level.

Notes. Unemployment is the proportion of working-age adults who are unemployed, based on 2018 American Community Survey estimates. Metrics are aggregated by week, income quintile, and unemployment level. Cut points do not represent quintiles. Sample comprises 210,288 census block groups with mean 89 active devices per block group per day.

Fig. 1 |. Proportion of smartphone users staying home all day.

Income Q1 represents the lowest income group. Outcomes are presented as weekly averages. Period covered is 6 January to 3 May 2020. Sample comprises 210,288 census BGs with a mean of 89 active devices per BG per day.

Fig. 2 |. Proportion of smartphone users working outside the home.

Income Q1 represents the lowest income group. Outcomes are presented as weekly averages. Period covered is 6 January to 3 May 2020. Sample comprises 210,288 census BGs with a mean of 89 active devices per BG per day.

Fig. 3 |. Visits to non-work locations.

Note that visitor income is calculated for each point of interest on the basis of visitor home census BG from January and February 2020. Median visitor income quintile is based on the median of household income values from visitors, weighted by the number of visits per BG. Non-work visit counts were calculated at the point of interest level by subtracting visits assumed to be from workers (duration >4 h) from total visit counts and adjusting the remaining visits to account for variation in device-to-population ratio among visitors’ home BGs. The resulting counts were aggregated by week, location type and income quintile for this visualization. Period covered is 6 January to 3 May 2020. Sample comprises 414,946 POI.

Fig. 4 |. Event study linear regression analysis: effects of state physical distancing orders on staying home all day.

Each panel reports the result of an event study regression within a single income stratum. Models are similar to DiD models reported above, except that we replaced the binary policy indicator with binary indicators for living in intervention states in a series of 1-d periods up to 14 d before and after policy changes. The reference group was being in a comparison state or being in an intervention state on the day before policy enactment (day –1). Time period is limited to dates before 20 April 2020, when the first state physical distancing order was lifted (South Carolina). Sample comprises 210,288 census BGs with a mean of 89 active devices per block.