Geospatial · Areas and surfaces

Calibrate Gravity Model

Fits the Gravity Model's attractiveness and distance-decay settings to what customers actually did, such as loyalty-card visits per area and store, and reports how well the fit reproduces each store.

Also known as: Calibrate Huff, distance decay

How it works

Wire three tables: demand (origins), locations, and observed data with one row per origin and location, or raw transactions, holding visits, customers or spend. Pick the Observed origin id column, Observed location id column and Observed value column, then the same geometry, id, Demand and Attractiveness columns, Distance measure, Distance decay and Search limit you use on the Gravity Model being calibrated. Fit chooses Both exponents (α and β), the default, or Distance decay only (β) with a Fixed attractiveness exponent (α), 1 by default.

The exponents are fitted by maximum likelihood. With Output set to Parameters the result is one row: alpha, beta, decay, distance_measure, search_limit, r_squared, mape, pairs_used, origins_used and locations_used. Fit by location instead adds observed, predicted, error, error_pct and captured_demand to each location so you can see which stores the fit reproduces well.

Inputs and outputs

Settings

{
  originGeometryColumn: string
  originIdColumn: string
  demandColumn?: string
  locationGeometryColumn: string
  locationIdColumn: string
  attractivenessColumn?: string
  observedOriginColumn: string
  observedLocationColumn: string
  observedValueColumn: string
  distanceMeasure: "straight"|"drive-time"|"drive-distance"
  decay: "power"|"exponential"
  searchLimit: number
  fit: "both"|"decay-only"
  fixedAlpha: number
  output: "parameters"|"locations"
}

Example

You have loyalty-card visits per postcode and store. Wire the postcode table, the store table and the visits table, pick the id and value columns, keep Fit at Both exponents and Output at Parameters, then wire the result into the Gravity Model's params input so its forecast uses the alpha and beta fitted from your own data.

More in Areas and surfaces

• Trade Areas — Builds catchment areas around each location: distance rings, drive-time bands, areas traced from demand, or rings cut into sectors.
• Gravity Model — Splits each area's demand between the locations in range by their attractiveness and distance (the Huff model). Gives the demand each location captures, such as a new store's sales and what it takes from the others, the top location per area, or a probability for every area and location.
• Voronoi — Draws one area per point covering everywhere closer to it than to any other point, such as the nearest-store territory for every shop.
• H3 Aggregate — Puts features into hexagon cells of a chosen size and summarizes each cell, for an even grid on density maps.
• Spatial Interpolate — Estimates a smooth surface from measured sample points, on a hexagon or square grid. For example, a pollution surface from sensors.