# The 21 features & 6 themes Reference table of all 21 model features and the 3 context fields shown in the app. LookAlike scores neighborhoods using 21 observable features grouped into 6 thematic clusters. These are the same features used to build each brand’s archetype and to rank resemblance across the continental US. Each feature is pulled from open US Census data (ACS 2023 5-year estimates) and Overture Maps Places. They are organized into groups so that correlated features—like income, per-capita income, home value, and rent, which all correlate above r = 0.7—count as one theme rather than four independent pieces of evidence. This prevents any single correlated cluster from dominating the similarity distance. ## The 21 scored features Feature Unit Group Meaning Median household income $ Income, wealth & housing cost Typical household earnings in the neighborhood. Per-capita income $ Income, wealth & housing cost Average income per person. Median home value $ Income, wealth & housing cost Typical owner-occupied home value. Median gross rent $/mo Income, wealth & housing cost Typical monthly rent including utilities. Poverty rate % Income, wealth & housing cost Share of people below the federal poverty line. Bachelor’s degree or higher % Education Share of adults 25+ with at least a 4-year degree. Population density /km² Density & transportation People per square kilometer of land. Households without a vehicle % Density & transportation Proxy for walkable, transit-served areas. Walk / bike / transit commute % Density & transportation Share commuting without a private car. Renter-occupied housing % Density & transportation Share of homes that are rented, not owned. Median age yrs Age & household type Half the residents are older, half younger. Average household size people Age & household type Average number of people per household. Households with children % Age & household type Share of households with someone under 18. White (non-Hispanic) % Race & ethnicity Share of residents who are white, non-Hispanic. Hispanic / Latino % Race & ethnicity Share of residents of Hispanic or Latino origin. Asian (non-Hispanic) % Race & ethnicity Share of residents who are Asian, non-Hispanic. Places of interest /km² Retail & dining context All Overture points of interest per km²—commercial intensity and walkability. Grocery stores /km² Retail & dining context Grocery stores and supermarkets per km²—grocer competition and co-location. Restaurants /km² Retail & dining context Restaurants and fast-food per km². Cafés & coffee /km² Retail & dining context Coffee shops and cafés per km². Shopping & apparel /km² Retail & dining context Clothing, department, and shopping-center POIs per km². These 21 features are the complete set used by the similarity model. They are chosen to be interpretable, fully documented, and traceable to their data source. See [data sources](https://docs.lookalikeiq.com/method/data-sources/) for provenance. ## The 3 context fields Three additional fields appear in neighborhood detail panels and reports but are **not scored**—they are shown for reference only: Field Unit Source Meaning Total population people US Census ACS 2023 Total residents in the tract. Black (non-Hispanic) % US Census ACS 2023 Share of residents who are Black, non-Hispanic. Unemployment rate % US Census ACS 2023 Share of the labor force that is unemployed. ## How features are scaled and weighted Before a feature enters the similarity distance, all features are: 1. **Clipped to the national 1st–99th percentile** to prevent a handful of extreme tracts (top-coded incomes, ultra-dense urban cores) from distorting the national scale. 2. **Standardized as deviations from the brand’s archetype**, using each feature’s own spread so the distance accounts for how tightly the brand clusters on that feature. 3. **Given a per-feature variance floor** (λ = 0.5): `sigma_eff = sqrt(sigma_brand² + (λ · nat_std)²)`. The floor raises the effective spread of a feature the brand is artificially tight on, so that feature cannot dominate the score; it never reduces the spread of a feature the brand is already wider on than the nation. See [how it works](https://docs.lookalikeiq.com/method/how-it-works/#computing-distance) for the full explanation. 4. **Averaged within each theme**, then **averaged across all six themes**, so each thematic group counts once regardless of how many features it contains. These features measure observable neighborhood similarity only. They do **not** predict whether a store will succeed, what revenue it will generate, or whether a location is a good business opportunity. High resemblance to a brand’s archetype means a neighborhood looks like where that brand already operates—nothing more. See [what it is](https://docs.lookalikeiq.com/guides/what-it-is/) for the full honesty statement. ## Feature groups explained **Income, wealth & housing cost** (5 features) Captures the affluence level and housing cost burden of the neighborhood. These five features are all strongly correlated (r > 0.7) and are grouped to prevent the affluence cluster from dominating the score. **Education** (1 feature) Bachelor’s degree or higher is the only education feature; it distinguishes knowledge-economy, professional, and affluent-suburban neighborhoods from others. **Density & transportation** (4 features) Distinguishes urban, walkable, transit-served neighborhoods from car-dependent suburban and rural areas. Population density, vehicle ownership, commute mode, and tenure (renter vs. owner) together describe place type and car dependency. **Age & household type** (3 features) Captures whether a neighborhood skews older or younger, and whether it has families or is dominated by individuals and couples. These distinguish family-oriented suburbs from young-professional urban neighborhoods and retiree communities. **Race & ethnicity** (3 features) Documents neighborhood racial and ethnic composition. These three (White non-Hispanic, Hispanic/Latino, Asian non-Hispanic) represent the largest demographic shares; Black non-Hispanic is tracked as context only. **Retail & dining context** (5 features) Commercial density and competitive retail environment, sourced from Overture Maps Places (June 2026 release). “Places of interest” is a catch-all commercial density; the other four measure specific retail clusters that may attract similar customer types or indicate co-location opportunities. You can adjust how much weight each theme has in the similarity score using the **factor weight sliders** in the app. This lets you explore scenarios (e.g., “What if I only cared about affluence?” vs. “What about urbanism?”). See [tuning the factors](https://docs.lookalikeiq.com/guides/factor-weighting/) for details. ## Data quality notes - **ACS data** (demographics): Carries a multi-year lag (2023 estimates are current as of mid-2024) and sampling error, which is worst in small, sparse tracts. - **Overture Maps** (retail context): Coverage is comprehensive for large metropolitan areas but thinner in rural regions; some categories may lag reality. - **Store locations** (from OpenStreetMap): Good coverage for large national chains, less complete for regional or franchised brands. See [data sources](https://docs.lookalikeiq.com/method/data-sources/) for full lineage and download endpoints. Canonical URL: https://docs.lookalikeiq.com/method/features/