Traditional measures of segregation rely on where people live, but daily life extends far beyond the home. This map visualizes the National Experienced Racial-Ethnic Diversity (NERD) dataset, which uses anonymized mobile phone location data to measure the diversity people actually encounter throughout their day: at work, during errands, and in leisure activities.
Take the quick tour, or jump right in.
Use the City menu at the top left to jump to any U.S. metropolitan area. The map flies there and loads data for every census tract.
The Metric menu sets what the colors show: Total Experienced Diversity, Total Residential Diversity, or the Difference between them. The Legend just below decodes the color scale.
Toggle Weekday / Weekend and the time of day (Morning through Late night) to see how the diversity people experience shifts through the day.
The Metric Distribution histogram summarizes the current view. Click any tract to compare its experienced vs. residential diversity and see its census characteristics, or use Search Places to fly to a specific address.
This dataset is made available under a custom license. Copyright 2023 Wenfei Xu. Please cite the associated publication when using this data.
Xu, W., Wang, Z., Attia, N., Attia, Y., Zhang, Y., & Zong, H. (2024). An experienced racial-ethnic diversity dataset in the United States using human mobility data. Scientific Data, 11, 638. https://doi.org/10.1038/s41597-024-03490-y
| Column | Description |
| STATEFP10 | 2010 Census state FIPS code |
| COUNTYFP10 | 2010 Census county FIPS code |
| TRACTCE10 | 2010 Census census tract code |
| GEOID10 | Census tract identifier (state FIPS + county FIPS + tract code) |
| STATE | State Name |
| COUNTY | County Name |
| CBSA Title | Core Based Statistical Area name |
| total_pop | Total Population (2020 Census) |
| white_perc | % Non-Hispanic White Alone (2020) |
| black_perc | % Non-Hispanic Black Alone (2020) |
| indigenous_perc | % Non-Hispanic American Indian and Alaska Native Alone (2020) |
| asian_perc | % Non-Hispanic Asian Alone (2020) |
| pac_isl_perc | % Non-Hispanic Native Hawaiian and Pacific Islander Alone (2020) |
| other_perc | % Non-Hispanic Some Other Race Alone (2020) |
| two_more_perc | % Non-Hispanic Two or More Races (2020) |
| hispanic_perc | % Hispanic (2020) |
| ba_higher_perc | % Bachelor's Degree or Higher (2020) |
| median_inc | Median household income (2021 inflation-adjusted dollars) |
| white_exp_exposure | White Experienced Diversity (2022) |
| black_exp_diversity | Black Experienced Diversity (2022) |
| asian_exp_diversity | Asian Experienced Diversity (2022) |
| hispanic_exp_diversity | Hispanic Experienced Diversity (2022) |
| other_exp_diversity | Other Race Experienced Diversity (2022) |
| total_exp_diversity | Total Experienced Diversity (2022) |
| residential_diversity | Residential Diversity (2020 Census) |
| diff | Difference between Total Experienced and Residential Diversity |
Click a state to download its GeoJSON file from the OSF repository.
The National Experienced Racial-Ethnic Diversity (NERD) dataset provides estimates of experienced diversity for the entire United States at the census tract level. Using anonymized mobile phone location data from over 66 million opted-in devices, the dataset measures the diversity people actually encounter throughout their day, not just where they live.
Diversity in potential social interactions is estimated at 38.2 m × 19.1 m spatial resolution and 15-minute temporal resolution for a representative year, then aggregated to the census tract level for data privacy. The dataset includes experienced diversity broken down by race/ethnicity (White, Black, Asian, Hispanic, Other), time of day (morning 6:00 AM–11:59 AM, afternoon 12:00 PM–4:59 PM, evening 5:00 PM–9:59 PM, late evening 10:00 PM–11:59 PM, late night 12:00 AM–5:59 AM), and day of week (weekday vs. weekend), along with census demographic characteristics.
The dataset is built from anonymized, opted-in mobile-device location data provided through Cuebiq's Spectus Data for Good program, covering roughly 66 million devices across the United States. To capture seasonal variation, four representative two-week periods spanning 2022 (March, June, September, and December) are used. GPS pings originate from everyday mobile applications (weather, navigation, games, and similar), with an average location accuracy of about 21 meters.
Each device is assigned a home location by scoring where it dwells over the preceding month (favoring frequent, overnight, and long-duration stays), updated daily to reflect moves. Home locations are up-leveled to 2010 Census block groups for privacy, and only devices with an identifiable home are retained. Because individual race and ethnicity are not observed, each device is given a probabilistic demographic profile drawn from the American Community Survey composition of its home block group, across five groups: Non-Hispanic White, Non-Hispanic Black, Non-Hispanic Asian, Hispanic, and Non-Hispanic Other.
Co-locations (devices whose stays overlap in both space and time) are detected on a geohash-8 grid (about 38.2 m × 19.1 m) in 15-minute windows. For each cell and time window, experienced diversity is computed as a Gini-Simpson index:
D = 1 − Σj ( nj / N )2
where nj is the number of people of group j present and N is the total. D is the probability that two people drawn at random from that place and time belong to different racial-ethnic groups. Cell-level values are aggregated into five times of day and averaged separately for weekdays and weekends.
Geohash-level diversity is aggregated to 2010 Census tracts using an activity-weighted sum, so busier places contribute proportionally. To protect privacy, tracts with fewer than 20 unique devices are excluded (613 tracts nationwide), and all outputs are aggregated so that no individual or sensitive location, such as a school or place of worship, can be singled out.
For comparison, residential diversity applies the same index to each tract's residential racial-ethnic composition from the 2017–2021 ACS. The map's Difference metric is experienced minus residential diversity: positive values mark places where people encounter more diversity through daily activity than the neighborhood's residents alone would suggest, and negative values the reverse.
Device coverage tracks Census population closely (Spearman ρ ≈ 0.93 nationally), and inferred home and workplace locations align with the Census LEHD/LODES employment data (r > 0.8), supporting the representativeness of the sample.
Xu, W., Wang, Z., Attia, N., Attia, Y., Zhang, Y., & Zong, H. (2024). An experienced racial-ethnic diversity dataset in the United States using human mobility data. Scientific Data, 11, 638. https://doi.org/10.1038/s41597-024-03490-y