demographics_segregation_county_year
County × year. Residential segregation computed transparently from ACS 2019–2023 5-year tract race counts (table B03002), aggregated to county: the index of dissimilarity (0–100, higher = more segregated) for three pairings — Black/White, Hispanic/White, Asian/White — plus Theil's multigroup entropy index H (0–1). These are derived single-construct statistics (evenness of group distribution across tracts) — NOT a multi-construct composite index like SVI/EJI. Tier B (ACS estimates). Validated against the prior CHR-sourced Black/White index (r=0.84); replaces it with a multi-pairing, transparently-computed measure. A pairing is NULL where its reference group's county total is 0 (~3% for Black/White).
overview
Computed from 84,400 tracts across 3,144 counties. Black/White dissimilarity present for ~97% of counties; Hispanic/White and Asian/White nearly complete; Theil's H present except for fully homogeneous counties. Connecticut covered via 2022 planning regions (unlike the prior CHR version).
current vintage — ACS 2019–2023 5-year (end year 2023)
history — Single vintage loaded (2019–2023). ACS publishes B03002 for prior 5-year windows; not yet ingested.
source & licensing
fields
| name | type | definition |
|---|---|---|
| country_idkey | string | ISO alpha-2 country code (always 'US' for domestic tables).Part of primary key. |
| county_idkey | string | 5-character FIPS code identifying the county.Part of primary key. Joins dim.counties on county_id. |
| segregation_asian_white_index | float64 · index (0–100) | Index of dissimilarity between non-Hispanic Asian and non-Hispanic White residents across the county's census tracts, measuring the share of either group that would have to relocate to achieve an even residential distribution. Higher values indicate greater residential segregation; NULL where either group's county total is 0. |
| segregation_black_white_index | float64 · index (0–100) | Index of dissimilarity between non-Hispanic Black and non-Hispanic White residents across the county's census tracts, measuring the share of either group that would have to relocate to achieve an even residential distribution. Higher values indicate greater residential segregation; NULL where either group's county total is 0. |
| segregation_hispanic_white_index | float64 · index (0–100) | Index of dissimilarity between Hispanic and non-Hispanic White residents across the county's census tracts, measuring the share of either group that would have to relocate to achieve an even residential distribution. Higher values indicate greater residential segregation; NULL where either group's county total is 0. |
| state_idkey | string | 2-character FIPS code identifying the state.Part of primary key. Joins dim.states on state_id. |
| theil_h_multigroup | float64 · index (0–1) | Theil's multigroup entropy index (H) measuring how much county-wide racial/ethnic diversity is lost at the tract level, computed across 5 groups. Ranges from 0 (tracts perfectly mirror the county composition, fully even) to 1 (complete segregation); higher values indicate greater segregation. NULL for fully homogeneous counties. |
| tract_count | int64 · count | Number of census tracts in the county used to compute the segregation indices. Filter out low values, as dissimilarity and entropy estimates become unstable when computed over few tracts. |
| yearkey | int64 | Reference year of the observation.Part of primary key. |
joins
how to use this table
Index of dissimilarity D = ½ Σ_i |a_i/A − b_i/B| over tracts i within each county (×100), where a/b are the two groups' tract counts and A/B their county totals, for Black/White, Hispanic/White, and Asian/White. Theil's multigroup entropy index H = Σ_i t_i(E − E_i)/(T·E) over 5 groups (NH White, NH Black, NH Asian, Hispanic, NH Other), where E is county entropy, E_i tract entropy, t_i tract population, T county population. Computed in dbt from raw.acs_tract_race_2023.
Individual-level inference (this is an ecological, area-level measure). Counties with very few census tracts (the index is unstable; single-tract counties are 0 by construction — filter on tract_count). Cross-year trend analysis without accounting for ACS 5-year window overlap between vintages. Note on grain: county dissimilarity UNDERSTATES segregation in fragmented metros — the sociological literature measures segregation primarily at the metro/MSA level (Brown's Diversity & Disparities project, Governing). County is defensible (it is our join grain) but is not the literature default.
Indices are unstable for counties with few tracts or small group populations; single-tract counties are 0 by construction (filter on tract_count). A pairing is NULL where the reference group's county total is 0. ACS 5-year overlap means consecutive vintages are not statistically independent. Decennial DHC would give a 100%-count anchor but injects differential-privacy noise into small tract race cells.