How We Score Duty Stations
The data sources, scoring rules and limitations behind your duty station comparisons.
Last reviewed: September 21, 2026
How scoring works
Market factors compare eligible observations across up to 299 U.S. military housing markets. Market comparisons use available observations in their respective datasets; the supplemental factor table currently covers 298 markets. Survey and installation-directory factors use their own eligible comparison pools. Army tenant surveys compare stateside posts meeting the response floor. On-base amenities compare distinct MWR directories, counting combined directories once.
Factor grades use fixed score bands: A at 80 or above, B at 60, C at 40, D at 20, and F below 20. Most factors are percentile ranks. Safety and things-to-do scores blend comparisons and are not direct percentile ranks; the food, arts and recreation score blends county comparisons and is not itself a percentile rank. Area livability grades disclose the contributing pillars and any unavailable pillar beside the letter. Hover a grade for its comparison and source limitations, or a factor for its comparison details. These are not national quality ratings.
Selected priorities receive descending weights from the number selected (N) to one. Eligible readings retain their full original weights; a missing or ineligible priority contributes to neither the weighted sum nor its denominator. Partial scores remain ordered in lists and suggestions, with scored counts and unscored priorities disclosed; different stations can therefore score different subsets of your priorities. No eligible priorities means no score.
Market housing cost and your BAH
This planning benchmark is not a quote for a particular home. Housing type, size and neighborhood affect your actual cost. The blend's dollar error has not been established.
The Duty Station Decider ranks estimated monthly market housing cost, with lower costs scoring better. The estimate averages two dollar inputs with fixed 50/50 weights: the market's HUD 3-bedroom gross-rent benchmark (H) and a BAH-informed estimate (D*). D* equals the fixed mean HUD benchmark across the 299-market reference set (Hbar, approximately $2,168) multiplied by a grade-invariant BAH market index (I_D), normalized to mean 1 over that same reference set. Thus market housing cost is 0.5 x H + 0.5 x Hbar x I_D. Hbar and the BAH index are fixed in the published model, never recalculated for your selected markets or household. The estimate, housing-cost score and rank stay the same for every pay grade and dependency status. Your own BAH is displayed separately at its published dollar amount. The benchmark is displayed rounded to the nearest $25. Scoring uses the unrounded estimate. Equal costs share a percentile and market rank.
Both dollar inputs sit on HUD's national dollar scale. D* is a BAH-informed estimate on that scale, not DoD's published local three-bedroom cost. Anchoring D* to Hbar is a modeling assumption, not independent confirmation of the dollar level. The fixed BAH index takes the geometric mean of all 24 grades and both dependency states in each market, then divides by the arithmetic mean of those market values across the fixed 299-market reference set. It combines allowances tied to differing housing profiles under DoD's data-collection methodology; it measures relative market cost and is not a three-bedroom price. HUD supplies the model's three-bedroom reference, which is a product choice rather than a description of your home. The blend is a three-bedroom, 40th-percentile-anchored benchmark, not an observed local percentile or a quote for a particular home.
The ordering-optimal weight on HUD measured on the labels is about 0.60, the level-optimal weight is below 0.45 by both the de-contaminated triangulation direction and the independent-label MAPE. The two identification attempts point in opposite directions and straddle 0.50, and 0.50 is robust across the sensitivity sweep. That is a decision under unidentified uncertainty, not an estimated optimum. The triangulation direction depends on the measurement-model assumptions; it does not identify a bound on the true optimal weight. The labels were selected because published figures looked high, so their comparisons are diagnostic and do not establish calibration for all markets. The weights are not estimated from an ordering tie, and no downward correction factor is applied.
HUD FMR and SAFMR target the 40th percentile of gross rent for standard-quality units occupied by recent movers, including utilities except telephone. HUD's gross-rent definition also excludes television service and internet. DoD's BAH methodology uses median market rent plus average utilities for six housing profiles, with allowances varying by pay grade and dependency status. Its grade structure does not represent a common three-bedroom home. Even a bedroom-matched comparison would still mix a median-rent target with a 40th-percentile rent benchmark.
This tool displays your BAH and the market benchmark as separate facts. It does not subtract one from the other or compute a BAH coverage verdict: the benchmark is anchored to three bedrooms, while DoD uses different housing profiles for different grades and dependency states. The benchmark also includes a BAH-derived market index, so comparing it with your allowance would be partly self-referential. Neither the benchmark nor the ranking establishes whether your allowance covers the home you choose.
Across the 299 markets, median instrument disagreement is 9.5%, calculated as the median of each market's absolute difference between H and D* divided by its blended estimate. This is a descriptive diagnostic, not an uncertainty bound. Agreement between the two inputs does not remove uncertainty, and their minimum and maximum are not a bracket for actual cost.
The HUD input uses FY2026 revised SAFMR where published and county FMR otherwise, weighted by ACS 2024 five-year ZCTA population across the existing commuter ZIP ring. ZIP inputs are aggregated without intermediate rounding, then rounded once to the nearest dollar. ZIPs without positive published population weights are recorded as unweighted and omitted. Every positively weighted ZIP requires its applicable HUD rent input: an invalid published SAFMR is not replaced by county FMR, and conflicting county values remain unresolved. A missing required HUD benchmark or BAH index leaves housing cost unscored with a partial-score disclosure. No other rent source or single-input estimate is substituted.
HUD's FY2026 methodology uses private rent data including the Zillow Observed Rent Index (ZORI). Local private inflation factors can therefore introduce shared, market-varying error. DoD's currently published BAH collection description also includes web listings such as Zillow and Trulia alongside MLS, commercial datasets, property managers, landlords and local real-estate professionals. HUD and BAH are substantially different instruments in what they measure, geography and collection, but share some inputs and are not independent. Treating Zillow as an independent third indicator cannot identify the blend weights without additional assumptions.
Scoring version 2 identifies this housing-cost meaning separately from the data vintage. Saved preferences and share links that selected the retired BAH value factor require explicit priority review: the full priority list is cleared because removing one factor changes the remaining weights, while station and household choices are retained. Older selections without that factor keep their priority order. Unsupported scoring versions also require review. The 2026 home-buying report and its search metadata retain their frozen 3-bedroom rent inputs for rent-versus-buy and rental-income calculations, so their figures intentionally differ from the decider's market housing-cost model.
Schools
Schools use Stanford SEDA 6.0 achievement of the districts serving each market's commuter ZIPs (county-level estimate where district coverage is thin, labeled). At the 13 stateside installations with DoD-run schools on post (Bragg, Campbell, Benning, Lejeune, and others, per the DoDEA Americas directory), the Schools row discloses that on-post children attend those schools, which the district grade does not cover.
SEDA 6.0 pools state test results from the 2009 through 2019 school years, placed on a common scale linked to NAEP, and was published in 2025. A school score here describes those years, not the current school year, and the tool says so everywhere a school score appears. We publish only derived statistics from SEDA, a percentile of each market's achievement estimate against the other military markets we cover. We do not republish the SEDA files.Cite as: Reardon, S. F., Ho, A. D., Shear, B. R., Fahle, E. M., saliba, j., and Kalogrides, D. (2025). Stanford Education Data Archive (Version 6.0). purl.stanford.edu/xh833nn4025.
Safety
Safety compares reported violent and property offenses per 100,000 residents among the population covered by police agencies in each market's dominant county that reported a complete 2024 to the FBI and passed the reporting-stability screen. Both offense rates and the covered-population-to-ACS-county-population ratio appear together; each reading names its population. The scored comparison covers 187 markets. Higher scores mean lower reported rates in this comparison, not a prediction of personal safety. Scored markets require an estimated county share of at least 60% of the housing market, together with complete reporting and sufficient county coverage. Each reading leads with that estimated market share; when that share is below 100%, the remaining market is not measured by these rates. Rates always measure only the reporting agencies' covered population. County shares never weight or split offenses.
Local county context: 44 more markets show both rates for the named county's reporting agencies. Each reading first names the county's estimated share of the housing market and states that the rest of the market is not measured by these rates, then shows agency population and its ratio to the county's 2023 Census population. They are shown as reported, not ranked, and the Safety row reads Not scored. These rates do not change the off-base grade or your match score.
Limited reporting: 60 markets show how many agencies reported a complete year, their covered population and its ratio to the county population. The reporting sample is too incomplete for an area comparison, so no crime rate or safety grade is shown. Check locally: 8 markets have no usable full-year agency evidence. For either group, check recent incidents near each home with local police.
Every market stays in the tool. If another selected priority can be scored, the partial match score uses those remaining priorities. If safety is your only priority, unscored markets have no match score or ranked recommendation. Reporting ratios are not clamped and can exceed 100% because agency and Census population estimates differ.
Schools, safety and amenities come from the livability engine (SEDA 6.0 district grain, 2009-2019 assessments; FBI CDE 2024; Census CBP 2023).
Housing, childcare and the spouse job market
Market housing cost averages the HUD 3-bedroom gross-rent benchmark and a fixed BAH market index expressed on HUD's national dollar scale. This is a three-bedroom, 40th-percentile-anchored benchmark; the blend itself is not a measured rent percentile. HUD gross rent includes specified tenant-paid utilities. BAH uses median rent plus average utilities for housing profiles that vary by pay grade and dependency status. Its market index is not a published DoD 3-bedroom price. The instruments share some inputs, and both blend components use HUD's national dollar scale by assumption; their agreement does not establish the dollar error. Your published BAH is shown separately because these measures do not support a member-specific coverage verdict.
Lower estimated cost scores better; equal estimates share a percentile and market rank. The estimate stays the same across pay grades and dependency states. Missing required inputs remain unscored with no substitute source. The VA loan report retains its frozen rent inputs. Priorities use scoring version 2; saved BAH value selections require you to choose and order priorities again. Housing methodology and modeling assumptions are detailed above.
Childcare: DoL National Database of Childcare Prices. The measure is each county's median weekly price for full-time center-based infant care, multiplied by 52 and blended across the counties in the market's commuter ring by population. That is a civilian market price. It is not the fee schedule at an on-base Child Development Center, which DoD sets from total family income, and it does not account for waitlists or the cost of care for older children. A market is priced only when every county in its ring reports in the pinned comparison year, so contributors are never averaged across vintages. Source: DoL Women's Bureau National Database of Childcare Prices (through 2022; modeled).
Some states report to the federal survey in regional clusters rather than county by county, so separate housing markets can inherit one cluster price. 50 of the 286 priced markets share an identical childcare price with another market in the same state, across 21 groups. Markets inside a group are tied on childcare rather than ranked against each other, and any gap you see between them comes from other factors.
12 of the 298 markets carry no childcare price and no childcare score. Every one is a gap in the federal source rather than a judgment about the market, and we checked each ring county against the database in every published year instead of assuming it. New Mexico replaced its childcare market rate survey with a cost estimation model, so the database prices no New Mexico county in any year. Indiana publishes subsidy reimbursement rates, which the federal report states cannot be used to calculate market rates, so no Indiana county is priced either. In Missouri the database prices 21 of the state's 115 counties and independent cities, and none of the counties in either affected market, which are Fort Leonard Wood and Whiteman AFB. The last gap is Kauai, where the federal report lists the county among those it says are represented while the published workbook carries no price for it, so the source contradicts itself and we do not guess which side is right. Each unpriced market names its own cause on the result card.
Where a state publishes a figure of its own that we could verify against the primary document, we show it as evidence and still refuse to score it. 2 markets carry one today. Missouri publishes a 75th-percentile daily rate for licensed centers, while the federal measure is a median weekly price. Ranking one against the other would rank the disagreement between two instruments instead of the difference between two places, which is the same trap the housing work already walked into once. So the state figure appears on the card in the percentile, setting, age band and unit the state published, with no conversion applied, and the row still reads “Not scored.”
Spouse job market: BLS county unemployment (modeled estimate, a whole-market proxy, not a military-spouse figure). Source: BLS LAUS 2025 annual county unemployment (modeled estimate).
Amenities and family activities: Census County Business Patterns 2023. Source: Census CBP 2023 (museums, amusement, recreation per 10k residents).
Walkability, transit, airports and climate
Walkability: EPA National Walkability Index (2021). Source: EPA National Walkability Index (Smart Location Database v3.0, 2021).
Transit: Census ACS 2019-2023. Source: Census ACS 2019-2023 5-yr, public-transit commute share. Airports: FAA CY2024 hub classes, straight-line distance. Source: FAA CY2024 enplanements (large/medium hubs) + OurAirports coordinates, straight-line distance.
Climate: NOAA 1991-2020 normals, nearest station. Source: NOAA U.S. Climate Normals 1991-2020, nearest station.
Hometown reference points: Census 2024 ZCTA Gazetteer (documentation checked September 17, 2026). ZIPs without a point use a labeled approximate ZIP-prefix center when available. Distances are straight-line geographic comparisons, not travel distances.
Barracks and on-post family housing (Army tenant surveys)
Barracks quality (Army single members only): the Army's first FY25 Unaccompanied Housing Tenant Satisfaction Survey (a 1-100 CEL satisfaction index), rankable only where an installation logged at least 30 responses; posts with fewer show their score as labeled context, and this factor compares Army posts to each other, not to the 299 markets.
On-post family housing (Army families only): the Army's FY25 Privatized Family Housing Tenant Satisfaction Survey (same 1-100 CEL index; 48 installations, 28.2% response), rankable at posts with at least 30 responses, always shown with the private company that operates that post's housing.
No other branch publishes per-installation barracks data yet. Barracks reporting beyond the survey: curated facts from GAO reports, the DoD Barracks Task Force, and named projects, each cited inline and never scored or ranked.
On-base amenities
On-base amenities (single members; Army, Navy, and Marine Corps): facility counts from each installation's official MWR directory (armymwr.com garrisons, Navy MWR installation sites, MCCS installation sites), retrieved September 2026, with DeCA commissary and BOSS/Liberty/Single Marine Program presence flagged; ranked within installations that publish a directory.
Counts are what the directory lists; absence from a directory does not prove absence on post, larger installations naturally list more, and combined directories (Stewart-Hunter, Lejeune-New River) are labeled. The Air Force and Space Force publish no central facility directories, so their installations carry no amenity score rather than a fabricated one.
Explore mode and better-fit suggestions
Explore mode and the better-fit suggestions run the identical scoring over every stateside installation we track for your branch, excluding admin headquarters, closed posts, and overseas bases, and showing one installation per metro so a single market cannot flood a shortlist; suggestions appear only when nothing on your own list scores at least 60/100 on your priorities, and only for markets that beat your best base.
Overseas and territory stations
Overseas and territory stations (the few we cover with a base guide) are shown as an evidence panel, never a ranked score: the factors above are U.S. datasets that do not exist abroad, so an overseas card shows only sourced intel that travels, standard tour length (JTR OCONUS supplement), command sponsorship (Korea), spouse employment under the SOFA, the OHA housing ceiling and overseas COLA index for your pay grade (from our Overseas Station Planner datasets, DTMO-sourced and representative, with a link to run your full pay there), the on-base amenities the MWR directory lists, DoDEA schools (overseas, DoDEA is the school system), barracks where surveyed, and distance from your hometown.
Limitations and what this tool does not claim
These are not national quality ratings. Different results can score different subsets of your priorities; partial scores disclose the scored counts and unscored priorities. The housing benchmark is not a member-specific housing budget, and the tool does not compute a BAH coverage verdict.
Ranked scoring covers stateside markets; a few overseas guides are shown for reference with sourced country intel, not a score. This tool informs your preferences; assignments follow the needs of the service.