Housing + Transportation Cost Methodology

A home that looks affordable can stop being affordable the moment you count the driving it requires. The Housing + Transportation (H+T) measure combines the two costs into one affordability lens: housing cost plus transportation cost as a share of a typical household's income. The approach was pioneered by the Center for Neighborhood Technology (CNT), whose published H+T Affordability Index established the widely used benchmark that a location is affordable when housing and transportation together take no more than 45% of income (housing alone no more than 30%, transportation no more than 15%).

CommunityScale maintains an independent, full replication of CNT's methodology, built from current public data and refreshed on our own schedule rather than frozen at CNT's release vintage. We compute it for every census block group and tract in the United States and publish the results on the CommunityScale atlas. This page documents how the model works, where it deliberately differs from CNT, and how closely it reproduces CNT's published index.

What we publish

For each block group we report, for a set of household profiles:

• H+T % of income: combined housing and transportation cost as a share of the profile's income.
• H % and T % of income: the two components separately.
• Annual transportation cost in dollars, and its inputs: predicted autos per household, predicted daily vehicle miles traveled (VMT) per household, and predicted transit commute share.

The household profiles follow CNT's definitions. The national typical household earns the national median income ($80,734, ACS 2020-2024 five-year), with the national average household size (2.53) and commuters per household (1.05). The regional typical household earns its own region's median income, resolved per block group to the metro area (CBSA) the county belongs to, or the county itself outside metro areas. The regional moderate household earns 80% of that regional median, the profile most relevant to workforce-housing analysis.

How the model works

The method follows CNT's three-model structure: transportation behavior is predicted from the built environment, then priced.

1. Measure the built environment. For every block group we compute a set of built-environment characteristics organized on the EPA Smart Location "D" framework: density (households, jobs, and population-weighted density), diversity (employment mix, jobs-housing balance, housing type shares), design (block size, intersection density, road-network density), transit service (distance to the nearest stop, service frequency from national GTFS schedules, CNT-style transit connectivity rings), and accessibility (gravity-weighted access to jobs and households, plus jobs reachable within 45 minutes by car and by transit). The gravity measures follow CNT's specification exactly: an inverse-square distance weight, not scaled below one mile, with the household intensities excluding the block group's own households.

2. Predict transportation behavior. Three regression models translate those characteristics, plus a small set of demographic controls, into behavior: automobiles owned per household, daily VMT per household, and transit commute share. The models are fit nationally at the block-group level on the same dependent variables CNT uses. For VMT we go one step beyond CNT's survey-modeled basis: the primary VMT model is fit on millions of actual odometer readings from state vehicle-inspection and DMV records (Colorado, Missouri, New York, and Oregon), which measure driving directly rather than inferring it from travel diaries. The odometer model serves roughly 95% of households; block groups outside its support fall back to a survey-anchored model built on the federal LATCH travel estimates, calibrated to observed county roadway VMT.

3. Price the behavior. Predicted behavior is converted to annual dollars using Bureau of Labor Statistics Consumer Expenditure Survey (2024) cost factors: $6,703 per vehicle per year in fixed ownership costs, $0.278 per mile in operating costs, and $1,560 per year per transit commuter. Housing cost is measured directly from the ACS: the tenure-weighted blend of median selected monthly owner costs for mortgaged owners (table B25088) and median gross rent (B25064), the same definitions CNT uses. Where the ACS suppresses a block group's housing cost, the containing tract's value fills the gap.

Dividing the annual costs by each profile's income yields the H%, T%, and H+T% shares.

Validation against CNT's published index

CNT publishes its index at the block-group level nationally, which lets us validate our replication block group by block group against a fully independent implementation. Across roughly 232,000 to 237,000 matched block groups (98.7% of the nation), agreement for the national typical household profile:

The regional profiles validate comparably (H+T correlation 0.89 for both the regional typical and regional moderate households). Medians agree within a few percent throughout; the small remaining level differences trace to documented vintage choices, chiefly our use of 2024 cost factors and 2020-2024 ACS income against CNT's older anchors.

Where we deliberately differ from CNT

The replication is faithful by construction, and every remaining deviation is deliberate, documented, and measured:

1. Fresher data. We build on ACS 2020-2024, LODES 2022/2023 jobs, TIGER 2024 geometry, current GTFS schedules, and BLS CES 2024 costs. CNT's published index is anchored on ACS 2018-2022. This is the main source of the small level differences above.
2. Observed VMT. CNT models VMT from the National Household Travel Survey via LATCH. Our primary VMT model is fit on actual odometer readings, which avoids the well-documented tendency of travel diaries to undercount driving.
3. Gravity destinations at the block-group level. CNT sums its accessibility measures over individual census blocks; we sum over block-group centroids, with block-group-resolution destinations (four times finer than tract). Measured against a block-level rebuild, the difference exceeds 50% in only about 2% of block groups, all in very sparse rural areas.
4. A 60-mile accessibility horizon. CNT extends its gravity sums nationwide using progressively coarser geography at distance; we truncate at 60 miles, beyond which the inverse-square weight makes contributions negligible.
5. Withholding predictions outside model support. In 265 block groups (0.11% of the nation, concentrated in frontier Alaska, Nevada, and Arizona) the average census block exceeds roughly 3,100 acres, beyond anything in the model's training range, and measured prediction error there is too large to publish. We report no transportation estimate for these rather than an unreliable one.

Notes and limitations

• Comparison over absolute level. As with all of our built-environment measures, the index is most reliable for comparing places. Rank correlations against CNT of 0.90 and above are the load-bearing validation statistic; absolute levels carry the cost-vintage choices documented above.
• Model predictions, not observations. Autos, VMT, and transit share are modeled from neighborhood characteristics, so they describe the expected behavior of a typical household in that location, not any particular household's behavior.
• Transit costs are a national benchmark. The headline index prices transit at a flat national pass benchmark, as CNT does. We additionally compute regional variants using actual realized fares by urbanized area (National Transit Database) and state auto-insurance costs (NAIC), which redistribute costs geographically without changing the national picture.
• Coverage gaps follow the source data. Puerto Rico has no LODES employment data and is excluded. Michigan and Alaska use the latest available LODES vintages (2021 and 2016).

Sources

U.S. Census Bureau ACS five-year estimates (2020-2024), LEHD LODES workplace jobs, TIGER/Line geometry, and the 2020 Decennial Census; aggregated national GTFS transit schedules; BTS Local Area Transportation Characteristics for Households (LATCH); state DMV and inspection odometer records (CO, MO, NY, OR); EPA NEI county roadway VMT; BLS Consumer Expenditure Survey (2024); Center for Neighborhood Technology, H+T Index Methods (November 2022). See Data sources and the Bibliography.