Data sources
Where every number in MapDove comes from: OpenStreetMap and Overture Maps places, US Census and WorldPop demographics, Census market boundaries and modeled visit estimates. Sources, licences, methods and limits.
MapDove covers every US metropolitan and micropolitan area, every rural county outside them, and Dubai, and all of it runs on open data. Places come from OpenStreetMap merged with Overture Maps Foundation Places; demographics come from the US Census Bureau and, for Dubai, WorldPop; US market and neighborhood boundaries come from the Census Bureau’s boundary files and the 2023 metropolitan and micropolitan delineations. Visit counts are modeled estimates, not measurements: they are produced by a gravity model driven by the real places and the real demographics. MapDove does not collect or use data from mobile devices, location panels, payment cards, cameras or Wi-Fi sensors, and none of the data describes individual people.
This page lists every source, how each field is derived and what the data can and cannot support.
At a glance
| Item | Value |
|---|---|
| Markets | 2,225 in the US (USD, imperial): 387 metropolitan and 538 micropolitan areas from the 2023 OMB delineations, plus 1,300 rural counties outside them; Puerto Rico is not covered. Plus Dubai (AED, metric) |
| Neighborhoods | US: Census places, ZIP code areas inside the largest cities and a “Rest of … County” piece per county. Dubai: named districts |
| Places | About 2.7 million points of interest in 14 categories, from OpenStreetMap merged with Overture Maps Foundation Places (US) and from OpenStreetMap (Dubai) |
| Brands | Chains recognised from the brand data in Overture and OpenStreetMap (a chain gets one brand per category, e.g. a grocer and its fuel stations); other places are independents |
| Hex cells | ~0.8 km² each in built-up areas; larger cells in the rural parts of a market |
| US demographics | US Census Bureau, ACS 2019–2023 5-year estimates (block groups) and LEHD LODES 8 jobs (2023) |
| Dubai demographics | WorldPop 2025 population (100 m grid); other Dubai attributes are modeled estimates |
| Daily visits | Modeled estimates for 2025-11-01 to 2026-10-31 (365 days) for every open place |
| Monthly history | Modeled monthly totals for the 24 months before (2023-11 to 2025-10), from the same model’s weekday, seasonal, holiday and growth terms |
| Live description | GET /api/v1/dataset returns which dataset an installation serves, its version and sources |
Sources and licences
| Source | Used for | Licence and attribution |
|---|---|---|
| OpenStreetMap | Places: names, categories, brands, addresses where mapped, building outlines | Open Database License (ODbL) 1.0. Places © OpenStreetMap contributors. The place data derived from OpenStreetMap remains available under the ODbL. |
| Overture Maps Foundation, Places (2026-10-21 release) | US places, merged with OpenStreetMap: names, categories, brands, addresses | Community Data License Agreement – Permissive 2.0 (CDLA-Permissive 2.0). © Overture Maps Foundation. |
| US Census Bureau, American Community Survey 2019–2023 5-year estimates | US population, households, income, age, education, family households, rent, housing built since 2020, employed residents | Public domain (US Government work) |
| US Census Bureau, 2023 cartographic boundary files | US block groups, counties, metropolitan and micropolitan areas and places; 2020 ZIP Code Tabulation Areas (market and neighborhood boundaries) | Public domain |
| US Census Bureau, 2023 metropolitan and micropolitan statistical area delineations (US Office of Management and Budget) | Which counties make up each US market | Public domain |
| US Census Bureau LEHD, LODES 8 workplace area characteristics (2023) | US jobs by census block (daytime population) | Public domain |
| US Census Bureau LEHD, LODES 8 origin-destination, workplace area characteristics and crosswalk (2024) | Workforce & commuters where it has been loaded (not yet in most markets, see below): jobs by home and work census block. Each workforce result names the source it was computed from (see below) | Public domain |
| US Census Bureau, American Community Survey 2016–2020 5-year county-to-county migration flows | County migration in the metro areas where it has been loaded: people who moved between counties, or in from abroad, with their margins of error. Each migration result names the source it was computed from (see below) | Public domain |
| US Census Bureau, 2020 cartographic boundary files (1:20,000,000) | County outlines and points on the migration map | Public domain |
| WorldPop, 2025 constrained population, 100 m (R2025A) | Dubai population | CC BY 4.0 — WorldPop (www.worldpop.org), University of Southampton |
Places
Every place is a point of interest inside a market, assigned to one of MapDove’s 14 categories. In the US, places come from Overture Maps Foundation Places: operating places whose Overture category maps to one of MapDove’s categories. OpenStreetMap adds the places Overture lacks, and a place found in both (within 75 m and in the same category, with the same brand or a similar name; within 30 m, a looser name match) is kept once. In Dubai, places come from OpenStreetMap. OpenStreetMap places are categorised from their tags (for example amenity=cafe → Coffee & Cafés, shop=supermarket → Grocery, amenity=fast_food → Quick Service, amenity=fuel and shop=convenience → Fuel & Convenience).
- Names and addresses are the ones in the source. Where no street address is recorded, MapDove shows the neighborhood.
- Chains come from the brand recorded in Overture or OpenStreetMap (OpenStreetMap’s
brandtag). In Dubai, a name that appears at least five times in one category also counts as a chain. Chains with 20 or more locations get a brand strength of 1.15, those with 5–19 get 1.05, smaller chains 1.0 and independents 0.9. - Floor area comes from the mapped building outline when it is plausible for the category (¼ to 4× the category’s typical size), otherwise from the typical size (80% of it for independents).
- Duplicates (a shop mapped both as a point and as a building) are merged when they share a name within 60 m.
- Opening dates are used when OpenStreetMap records one inside the data year; new places then ramp up over 90 days.
Neither source is complete, and coverage varies by neighborhood and category. A store missing from both is missing in MapDove until it appears in one of them and the data is refreshed.
Demographics
United States
ACS block groups are spread onto the hex grid by area overlap: a hex receives each block group’s counts in proportion to the share of the block group’s area it covers, and medians are population-weighted.
| Field | How it is derived |
|---|---|
| Population, households | ACS total population; total households |
| Median household income | ACS median household income (population-weighted across overlapping block groups) |
| Median age | ACS median age |
| College share | ACS population 25+ with a bachelor’s degree or higher |
| Family share | ACS family households ÷ all households |
| Daytime population | Residents who don’t commute out (population − employed residents) + jobs located in the hex (LODES) |
| Growth | Share of housing units built 2020 or later (ACS), spread over the ~3.5 years it covers and capped at 15% a year. A proxy for recent growth, not a measured population change |
| Rent index (0–100) | Percentile composite of median gross rent (ACS, 65%) and daytime density (35%), within each city |
| Traffic index (0–100) | Percentile composite of OpenStreetMap points of interest within 1 km (60%) and daytime density (40%), within each city |
| Lifestyle segment | The segment rules applied to the hex’s real age, income, density and family share |
Hexes without residents (parks, water, industrial land) carry the city-wide figures for medians so that nothing is left blank; their population is zero.
Dubai
Population comes from WorldPop’s 2025 100 m population grid, summed per hex. No open small-area data exists for income, age, education, household size, rent or daytime population in Dubai, so those fields remain modeled estimates from neighborhood-level parameters, scaled to the real population. The traffic index uses real OpenStreetMap activity, as in the US cities.
Workforce & commuters (US)
Availability. Workforce data is being rebuilt for national coverage, so most US markets don’t have it yet; the app says so where it is missing.
The Workforce panel of a place and the Workforce tool show how many jobs are located in an area, where the people holding them live and how far they commute. In the US, MapDove’s source for these numbers is the US Census Bureau’s LEHD Origin-Destination Employment Statistics (LODES 8, 2024): administrative counts of jobs by work and home census block, summed onto the hex grid. They are measured, not modeled.
Every workforce result is labeled with the data it was computed from: “Measured · US Census LEHD LODES 2024” for LODES, “Demo data” for the synthetic version, or a notice that workforce data isn’t available (Dubai) or hasn’t been loaded on that installation yet. GET /api/v1/dataset lists the LODES origin-destination source once an installation has loaded it.
| Figure | How it is derived |
|---|---|
| Jobs in the area | LODES jobs whose work block lies in the hexes covered by the circle (hexes cut by the circle count by their share inside it) |
| Where workers live | Home blocks inside the study area are mapped by hex and summed by neighborhood; homes beyond it are named by census place (else county); homes in another state are counted as such |
| Commute distance | Straight line between the home and work blocks, in bands; the median is interpolated within its band. Not a road distance or travel time |
| Live in the area | Share of the area’s jobs held by people whose home is also inside the circle |
| Industry, earnings, age | LODES workplace area characteristics (NAICS sectors in 8 groups) and the origin-destination earnings (≤ $1,250, $1,251–3,333, more per month) and age (≤ 29, 30–54, ≥ 55) bands |
LODES counts jobs, not people (someone with two jobs counts twice) and the Census Bureau adds noise to small block-level counts to protect confidentiality, so larger areas are more reliable than very small ones. LODES covers the United States only: Dubai has no workforce data, and the app says so. Development and self-hosted installations that run on the synthetic baseline show a synthetic version labeled as demo data.
County migration (US metros)
The Migration section of the Market page shows where people moving into a metro come from, where people moving out go, and each metro county’s net migration. It is available for the metro areas where it has been loaded, not for every market. MapDove’s source for these numbers is the US Census Bureau’s American Community Survey (ACS) 5-year county-to-county migration flows, 2016–2020: survey answers to “Where did you live 1 year ago?”, tabulated by county of residence now and one year earlier. They are measured survey estimates, not modeled, and every figure comes with its margin of error.
Every migration result is labeled with the data it was computed from: “Measured · US Census ACS 2016–2020 county-to-county migration” for the ACS flows, “Demo data” for the synthetic version, or a notice that migration data isn’t available (Dubai) or hasn’t been loaded on that installation yet. GET /api/v1/dataset lists the ACS migration source once an installation has loaded it.
Why 2016–2020. It is the latest county-to-county release. From the 2017–2021 estimates on, the Census Bureau publishes migration flows from each state to each county only (a decision of its Disclosure Review Board), which name no origin county and include no moves out of a county. The figures therefore describe moves in a typical year of 2016–2020, before more recent shifts.
Which counties make up a metro. A metro is a metropolitan statistical area as delineated by the US Office of Management and Budget (July 2023). For example:
| City | Metropolitan statistical area | Counties |
|---|---|---|
| Austin | Austin-Round Rock-San Marcos, TX | Bastrop, Caldwell, Hays, Travis, Williamson |
| Houston | Houston-Pasadena-The Woodlands, TX | Austin, Brazoria, Chambers, Fort Bend, Galveston, Harris, Liberty, Montgomery, San Jacinto, Waller |
| Phoenix | Phoenix-Mesa-Chandler, AZ | Maricopa, Pinal |
| Miami | Miami-Fort Lauderdale-West Palm Beach, FL | Broward, Miami-Dade, Palm Beach |
| Figure | How it is derived |
|---|---|
| Moved in, moved out (per county) | People 1 year and over who moved into the county from another US county (50 states and DC), or out of it to one, including moves between the metro’s own counties |
| Net migration (per county) | Moved in − moved out; also per 1,000 residents aged 1 year and over |
| From abroad | Moved in from abroad, Puerto Rico or the US Island Areas, by world region. Kept apart from net migration (see below) |
| Metro totals | The same for the metro as a whole: moves between its own counties don’t count |
| Top origins and destinations | Counties outside the metro by the number of people who moved from them into the metro (or from the metro to them), and the same by state. Smaller counties are summed per state, so state totals stay complete |
| Flow map | A line between each top county and the city’s centre; the metro’s counties shaded by net migration. County outlines and points come from the Census Bureau’s 2020 cartographic boundary files |
Margins of error. The ACS is a sample survey, so each estimate carries a margin of error at the 90 % confidence level: the true value lies within estimate ± margin with 90 % confidence. Flows between two counties, their net and each county’s population and moves from abroad show the margins the Census Bureau publishes; figures that add several of them up (a county’s moves in or out, metro totals, states) use the Bureau’s approximation, the square root of the sum of the squared margins. Small flows are uncertain: a flow whose margin is larger than the estimate is not reliably different from zero, so compare the larger flows.
The ACS surveys people living in the United States and Puerto Rico, so moves out of the country aren’t measured: net migration covers moves between US counties only, and moves from abroad are shown separately. The ACS covers the United States only: Dubai has no migration data, and the app says so. Development and self-hosted installations that run on the synthetic baseline show a synthetic version labeled as demo data.
Visits are modeled estimates
MapDove does not measure visits. Each place’s daily visits, visitors and dwell time are estimated:
base = category_base_visits · A · (local_demand / city_avg_demand)^0.5
visits = base
· weekday_weight(category, day of week)
· (1 + amplitude · cos(2π · (day_of_year − peak_day) / 365)) -- city seasonality
· (1 + growth · (day − 2026-05-01) / 365) -- growth proxy of the place's hex
· holiday factors -- US Thanksgiving, Black Friday, Christmas; December season
· opening ramp-up · small day-to-day noise
Ais the place’s attractiveness:(floor area / category norm)^0.7 × brand strength.local_demand = Σ_h (population_h + 0.4 · daytime_pop_h) · exp(−d / 3 km)over hexes within 8 km, using the real demographics above.- Weekday weights per category (e.g. coffee strong Mon–Fri, apparel and malls strong on weekends).
- Seasonality: amplitude and peak day per city (Austin 6% peaking late April, Houston 5% in April, Phoenix 12% mid-February, Miami 12% early February, Dubai 15% mid-January). These are modeled parameters, not measured seasonality.
- Ramp-up: places opened during the data year start at 55% of potential and reach 100% after 90 days.
- Visitors = visits × category visitor ratio; dwell = category dwell time, each with small day-to-day variation.
Holidays and events
- Holiday season: a bell-shaped bump centered on 20 December, scaled by each category’s holiday sensitivity (strong for apparel, electronics, department stores and malls; negative for fitness).
- Black Friday (28 Nov 2025) in US cities: a one-day spike for holiday-sensitive categories.
- Thanksgiving and Christmas Day in US cities: most categories at 30% of normal (hotels, fuel and cinemas stay open).
- No local events. The model does not add festivals, storms or other one-off events; a spike or dip at a single place is not in the data.
Hourly profiles
Hourly heatmaps combine the modeled average visits per weekday with a fixed hourly curve per category. Hour-of-day patterns are therefore category-typical, not place-specific.
Everything built on visits inherits this: trends, rankings, market overviews, trade areas, cross-shopping and the calibration factor of the Huff forecast. Treat them as consistent, model-based estimates for comparing places and sites, not as counts of real people.
Limitations
- Visits are not observed. They reflect location, size, brand, competition and demographics, not how well a particular store is run.
- Trade areas and cross-shopping are modeled from the gravity model, not from visitor origins.
- OpenStreetMap and Overture are incomplete in places and may include closed businesses or miss new ones.
- ACS figures are 5-year estimates (2019–2023) with sampling error, most visible in small block groups.
- County migration is from 2016–2020 (the latest county-to-county release) and has survey margins of error; moves out of the country aren’t measured.
- Dubai demographics other than population are modeled.
- Neighborhoods are statistical areas. In the US they are Census places, ZIP Code Tabulation Areas and the rest of each county, which don’t always match local neighborhood names; in Dubai they are geometric (Voronoi cells around real district centres), not official boundaries.
- Drive-time rings are a speed-based approximation, not road-network isochrones (see Catchments).
- One modeled year. “Change vs previous period” compares modeled periods.
Use MapDove to compare candidates consistently and to frame decisions; confirm important decisions with site visits and, where it matters, your own or licensed measured data.
Your own data (Enterprise)
On the Enterprise plan, MapDove can be deployed with customer-supplied or licensed data in place of, or alongside, the open data:
- Your places: store lists, sizes, opening dates and formats, so calibration uses your actual network.
- Your traffic: store-level counts (door counters, transactions, aggregated visit estimates from a provider you license) replace modeled visits. MapDove only ingests data that is already aggregated to place and day (or hour) level.
- Demographics: other official statistics or licensed estimates, re-aggregated to the hex grid.
- Custom markets and segments beyond the hosted markets.
- Network drive times as precomputed isochrones.
The models, API and app stay the same; only the inputs change. Real feeds are governed by the Data Processing Addendum, and the Acceptable Use Policy prohibits loading or deriving data that identifies individuals.
Development and self-hosted installations
Local development, tests and self-hosted installations start with a synthetic baseline generated by the database migrations (fictional brands, modeled demographics), so they work without any download. GET /api/v1/dataset reports "kind": "synthetic" there and "kind": "open" once the open-data bundle has been imported.