The Gospel Desert Index is a structural read of one question: whether the observable conditions that make sustained evangelical witness tractable are present in a place. This page is the complete reference — the domains and their weights, the data behind them, what the index refuses to score, and what it cannot tell you.
The Gospel Desert Index adapts the food desert concept from urban planning to the domain of evangelical witness. A food desert is an area where residents lack reasonable access to nutritious food because of structural factors — not personal failure. The GDI applies the same logic: a Gospel Desert is a Census tract where the structural conditions that typically enable sustained evangelical witness are absent or severely degraded.
This is a deliberately structural claim, not a spiritual one. The index does not purport to measure where the Spirit is at work. It measures whether the external, observable infrastructure of Gospel presence — accessible churches, institutional anchors, neighborhood vitality — exists at a level that makes incarnational ministry tractable.
The GDI identifies where incarnational presence is most structurally needed, not where God is absent. A tract in the Desert tier may hold faithful believers, house churches, and Spirit-driven fruit invisible to any dataset. The index is a tool for resource allocation, not a map of divine sovereignty.
The GDI rests on four explicit theological commitments that both motivate and constrain what it claims to measure.
Reformed ecclesiology holds that God ordinarily works through means: the Word preached, the sacraments administered, the community gathered. The structural absence of a faithful, accessible church is therefore missiologically significant — not because God cannot work without one, but because He has ordained that He ordinarily does work through one.
Harvie Conn’s incarnational urban missiology holds that Gospel access requires not merely the existence of a church within a geographic radius, but the presence of believers embedded in the social fabric of a community — who know its people, share its material conditions, and bear witness from within. Geographic proximity is necessary but insufficient.
Jonathan Edwards argued that the chief evidence of genuine conversion is consistent, habitual practice shaped by holy love. What begins as inward regeneration eventually presses outward into how a person serves their neighbor, and those practices, repeated across many lives, aggregate into community-level patterns. The GDI measures structural preconditions for access. What follows from the Gospel is watched by two further instruments, both built on the Shalom composite: need pressure, the weight a place is carrying right now, and the Gospel Impact Metric, the change in that weight between two eras. Three distinct readings, to be used together and never blended into one.
Following Nassim Taleb, Gospel Desert conditions are not simply barriers to ministry — they are the conditions under which resilient faith communities are forged. A church planted in a Desert-tier tract that survives and grows over a decade has demonstrated a robustness that a church planted in favorable conditions cannot.
The GDI is constructed from three independent data layers joined at the Census tract FIPS key — an 11-digit identifier combining state (2), county (3), and tract (6) codes. The United States contains roughly 84,000 Census tracts, each designed to capture a socially coherent neighborhood unit of 1,200–8,000 residents.
OpenStreetMap is the primary source for point locations of places of worship. At national scale the pipeline pulls Geofabrik state-level PBF extracts rather than querying the live API. Records are tagged amenity=place_of_worship and carry name, denomination, and religion fields. Coverage is strong in urban areas; denomination metadata is inconsistent nationally.
The IRS 990 Business Master File supplies the complete register of 501(c)(3) organizations with NTEE classification codes. Codes X00–X99 designate religion-related organizations. Per the file’s own code definitions, X20 is Christianity — a broad umbrella — X21 is Protestant, and X22 is Roman Catholic. The layer is built from X20 and X21 together, restricted to foundation code 10 (“Church, IRC 170(b)(1)(A)(i)”) so that clergy associations, parachurch ministries and think tanks registered under the same umbrella do not enter a church count. X22 is out of scope: most Catholic parishes file under a diocesan group ruling rather than individually.
Two further sources join them. ChurchesUSA is a purchased congregation directory. Overture Places is a web-derived places corpus, added in September 2026, which contributes congregations the other three miss — small, storefront and immigrant churches are systematically under-registered in all of them — and backfills websites, addresses and phone numbers onto records that had none. It is filtered to Protestant traditions and, like the whole denominator, excludes Latter-day Saints and Jehovah’s Witness records. All four sources are deduplicated against each other before being joined to tract geography.
A record is matched to an existing one on distance and name similarity, with the radius set per source and the newest source given the lowest priority, so it may add evidence to a church already known but never overwrite a name or a coordinate that a surveyed or registered source supplied.
Four fields per tract come out of this layer: church_count (organizations inside the tract), nearest_church_m (distance from the tract centroid to the nearest church, which may sit outside the tract), church_density_per_sqmi (normalized to land area), and church_denominations (count of distinct denominations present).
church_count counts congregations inside the tract. church_density_per_sqmi is churches per square mile of its land. Only density and distance feed the score. Substituting the count for the density means a threshold of one church per square mile is cleared by any tract holding a single church, however large.
The Census Bureau’s LEHD Origin-Destination Employment Statistics link employer payroll records with Census demographics to produce block-level counts of jobs by industry, wage, and worker characteristics. Two derived fields matter here. wac_pedestrian_index sums retail, food service, and healthcare jobs — the three sectors most reliably associated with daytime pedestrian traffic. od_activation_ratio is (workers arriving + 1) / (workers leaving + 1): above 1.0 marks a tract that receives workers, below 1.0 a residential exporter whose sidewalks are empty during the day.
LODES covers formal employment only. Informal economy activity, volunteer presence, and the relational foot traffic Jane Jacobs identified as the animating fabric of neighborhood life are not captured.
ACS 5-year estimates provide tract-level demographic and economic context. The 5-year averaging window substantially reduces sampling error at tract level, making estimates reliable without the margin-of-error concerns that affect block-group data. The pipeline reads B01003 (total population), B17001 (poverty), B25002 (housing vacancy), B23025 (labor force and unemployment), and B03002 (race and ethnicity).
Every domain score is computed by percentile rank within the full national tract set, so the GDI measures relative severity: a tract scoring 75 is more structurally desert-like than 75% of U.S. tracts, not past some absolute threshold. Each domain runs 0–25 and the four are summed. They carry equal weight — no domain outranks another in the formula.
Tier boundaries are fixed cut points on the composite, not quantiles, so the tiers are not evenly populated. Over four fifths of all scored tracts land in Sparse or Dry, and fewer than one in a hundred reaches Desert. Counts below are from the index this site currently serves.
Of 84,414 tracts, 83,344 carry a score and 1,070 do not. A tract with no score carries no score: writing one as 0.0 is what once put open water and industrial land in the Desert tier. Unscored is rendered as its own color on the map so data absence is never mistaken for a reading.
Separately, the pipeline flags tracts whose score rests on thin ground. Four conditions raise a flag: fewer than 200 residents, fewer than 50 housing units, more than 5,000 jobs against fewer than 100 residents, or more than four fifths water. 1,373 tracts carry at least one — 1,065 of them unscored, and 308 of them scored and displayed with a tier, including 34 Arid and 4 Desert. Those qualifiers travel with the tract and are stated wherever its score appears.
Empirically, distance to the nearest church is the strongest single input to a tract’s GDI (r = +0.44 against log miles). The count of congregations inside the tract is very nearly uncorrelated with it (r = +0.02) and is not a scoring input at all. A GDI score therefore cannot be read as a statement that a tract has, or lacks, a congregation of its own — a tract can sit 800 feet from a church just over its own boundary and hold none itself. The church count is its own field, and it is the only thing that says that.
Every index embeds assumptions. Identifying them explicitly is not a concession to methodological weakness — it is a precondition for responsible use.
The GDI assumes geographic distance between a resident and the nearest evangelical church is a meaningful indicator of access. Cultural, linguistic, and economic distance can render a physically proximate church functionally inaccessible. The index treats physical proximity as necessary but not sufficient.
LODES measures formal, payroll-reported employment. It does not capture the informal economy, gig work, or relational foot traffic. A Desert-tier tract in a dense urban environment may have vibrant informal social networks invisible to this layer.
Five-year estimates smooth year-to-year volatility, improving reliability but reducing sensitivity to rapid change. A rapidly gentrifying tract may carry outdated figures. The GDI is a baseline condition measure, not a real-time status indicator.
Any dataset of existing churches captures only churches that survived. Failed plants and dissolved congregations are absent from OSM and IRS data by definition, so the church access domain may undercount a tract’s structural difficulty: prior failure is invisible. A failed-church dataset is maintained as a planned correction to Domain A and is not yet in the score.
Following Taleb’s Lindy effect, structural conditions that have persisted for decades are likely to persist for decades more. A Gospel Desert that has been one since 1970 is unlikely to self-correct without deliberate, sustained investment. That gives the GDI long-horizon validity as a diagnostic, while cautioning against treating any single build as definitive.
The GDI is designed to serve a three-stage church planting site selection workflow. Each stage corresponds to a different geographic scale and a different output.
Network leadership decides which cities to enter. Tracts aggregate to CBSA, producing a ranked list of metros by Desert-tier tract count and population exposed — the presentation layer for a General Assembly or planting summit.
Regional directors narrow from metro to 8–12 candidate tracts, filtering the map to Desert and Arid tiers with domain breakdowns visible. This is where the GDI replaces gut instinct with a defensible field of candidates.
Planting teams assess candidates in depth using the field diagnostic: the tract’s Shalom read, the five domains ranked by pressure, and the GIM movement between 2020 and 2024. The GDI narrows the field; the Shalom read assesses the ground.
A denomination should not make a planting commitment on GDI scores alone. The index identifies where a site visit and a full field diagnostic are warranted — not where a church should be planted.
The most analytically useful read is the 2×2 of structural access against standing need: GDI versus need pressure, split at a GDI of 60 and a pressure of 50. Comparing them — not blending them — is where the pastoral insight lives. This matrix reads the level, not the movement; GIM is the change between two eras and is a separate layer, named Gospel Shift on the map.
The GDI produces real and useful information. It also has real and important limits. Both deserve plain statement.
Numbers can tell you a tract has a 38% poverty rate and no nearby evangelical churches. They cannot tell you about the widow in Building C who leads a Bible study for twelve of her neighbors. Both are real. The data serves the pastor; it does not replace him.
“We have this treasure in jars of clay, to show that the surpassing power belongs to God and not to us.” — 2 Corinthians 4:7
The Gospel Desert Index is a jar of clay. Use it to see more clearly. Let it push you toward the street, not away from it. And never let the score replace the face of the neighbor you are called to love.