However, there is precedent for mid-size data centers being beneficial neighbors. If a community accepts a data center, it should be under the condition that the data center’s waste heat be available for district heating and/or industrial processes, such as heating schools, swimming pools, and apartments.

Data centers are not going anywhere. They are the fastest-growing building category in the world. Cushman & Wakefield shows US data center construction surged from $8.5 billion in 2019 to over $31 billion in 2024 (while general office construction dropped from $71 billion to $50 billion). So, it is worth thinking about how they fit into our cities. This post explores the issues and how European cities are using data center heat.

Nearly all the electricity that enters a data center leaves as heat.

A data center is, thermodynamically, a space heater that does math. Every watt of electricity, through every chip, fan, and backup battery, ultimately becomes thermal energy. A 100 MW data center is a 100 MW thermal emitter, of which at least 50% could be captured as reusable heat. Air-cooled exhaust runs 25 to 35 degrees C (77 to 95 degrees F). Water-cooled systems deliver 50 to 60 degrees C (122 to 140 degrees F). Two-phase immersion cooling, where servers are submerged in dielectric fluid that boils and recondenses, reaches up to 90 degrees C. Typical exhaust lands around 38 degrees C. 

How do these temperatures work as an input to another process? Older district heating networks require very hot water. However, modern, energy-efficient networks are designed for 50°C to 70°C (122°F to 158°F). Data center waste heat at 30 to 60 degrees C is ideal for district heating, greenhouse agriculture, aquaculture, and swimming pools. 

Cities are already heating homes with server exhaust.

This is infrastructure operating at scale, heating real homes in real cities. At the large end, data centers connected to city-scale pipe networks. 

In Espoo and Kirkkonummi, Finland, 250,000 people get their heat from servers. Microsoft and Fortum built a 350 MW thermal system that will supply roughly 40 percent of all district heat for the area, reducing CO2 emissions by 400,000 tonnes per year. Seventy-five percent of the waste heat gets used annually.

Stockholm takes a multi-source approach. Stockholm Exergi’s Open District Heating network connects over 30 data centers to 3,000 km of pipes, warming 30,000 apartments. The utility pays operators roughly 2 million SEK per year per megawatt delivered (about $190,000). No single data center is critical. If one shuts down, others absorb the load. That redundancy makes the system investable in a way that a single-source network is not (Stockholm Exergi; EU Covenant of Mayors).

In Hamina, Finland, Google built a data center on the site of a former paper mill, cooled by Baltic seawater. It now provides 80 percent of local district heating, free of charge. For a small Finnish town, the data center is the heating system (Google Blog).

Here are some mid-size data centers being good neighbors in communities:

• Microsoft/Fortum — Espoo & Kirkkonummi, Finland — 350 MW thermal — 250,000 clients
• Stockholm Exergi — Stockholm, Sweden — 30+ DCs, individual sites 5–32 MW — 30,000 apartments
• Google Hamina — Hamina, Finland — 140 MW — Town-scale
• AWS Tallaght — Dublin, Ireland — 3 MW waste heat supplied — University-scale
• AECOM/London potential — London, UK — ~1,000–1,100 MW total London DC estate — 500,000 homes (potential)

Why is this not happening in the US? From the US data center operator’s perspective, interconnection takes longer and waste heat revenue may not offset the premium. The operator captures cheap power and externalizes the cost of wasted heat and community disruption, and transmission buildout. There is also not typically any existing district heating infrastructure. Stockholm has 3,000 km of pipes whereas US cities have nothing comparable. We heat buildings individually with gas furnaces, heat pumps, or electric resistance.

That being said, if a mid-size data center really wants to come to your city, you can stipulate conditions on how that land use arrives with zoning. Tangible benefits could include free heating for schools, as Google provides in Hamina and AWS provides in Tallaght. But you have to design the data center into the community from the start.

What a city should actually do.

Chicken and the egg problem. Data centers should not vent recoverable heat to the atmosphere while their neighbors burn gas for warmth. However, the absence of district heating is a cold-start problem. Requiring full waste heat connection before operations begin adds years and cost to project timelines. Instead, a condition could be connection points to future district heating networks, which would also give the host community time to set up receiving infrastructure (which might also be part of a community benefits agreement). The connection comes when the network reaches them, not as a precondition for switching on the servers, similar to requiring sewer connections where a treatment plant is planned but not yet built. 

Anchor the heat network with public buildings. The Tallaght model is the clearest demonstration of how to solve the cold-start problem. AWS provides heat free of charge to 47,000 square meters of public buildings, 135 apartments, and 3,000 square meters of commercial space in South Dublin. Savings: 1,500 tonnes CO2 per year, a 60 percent reduction. The trick was sequencing. Public buildings provided the guaranteed demand that made the district heating loop viable from day one. Libraries, community centers, social housing, schools: each one a baseload customer that does not negotiate seasonal discounts or threaten to switch suppliers. Identify a cluster of municipal buildings within 1 mile of a proposed or existing data center. Commit them as anchor customers (About Amazon; SEAI).

Establish the governance structure before the data center arrives. In Tallaght, South Dublin County Council created Heatworks, Ireland’s first not-for-profit energy utility. The community owns the pipes. The data center is a heat supplier, not a monopolist. If the supplier changes, the pipes remain. What happens if the data center closes? The same thing that happens when a power plant retires from a district heating network: backup sources maintain continuity through the contract. Gas boilers or large heat pumps carry the load until a new supplier connects. The logic mirrors municipal water utilities: public distribution, private supply through regulated contracts (CNBC). The Heatworks model deserves replication because it separates ownership of the network from dependence on any single heat source. That separation is what makes long-term investment rational.

Performance standards for other concerns. Approval conditions also need to address what data centers actually do to their immediate neighbors. Performance standards should address air quality limits on diesel generator testing and run hours, noise limits at the property line, full-cutoff lighting requirements to prevent light trespass, written utility capacity verification confirming the load can be absorbed without degrading service to existing customers, and water consumption documentation disclosing annual usage and source. The scale of the data center also matters. The negative externalities of a gigawatt data center are much harder to control within a community than mid-size data centers in the range of 2-50 MW.

The post How zoning for data centers could actually benefit your community appeared first on CommunityScale.

A bivariate approach to market analysis

The MVA measures each Census tract on two dimensions: current market condition (disinvested, somewhat competitive, or competitive) and rate of change over the past five years (not improving, keeping up, or changing quickly). This creates a 3×3 grid of nine neighborhood typologies, each calling for different strategies and interventions. The index draws on 17 input variables spanning market conditions, quality of life, and housing affordability, from building permits and home prices to air quality and walkability.

Data sources include ACS Census estimates, assessed property values, home sales and listing data, multifamily cap rates, school scores, blight designations, and mortgage denial rates. We prioritized nationally available datasets to ensure uniform coverage across the region’s mix of urban, suburban, and rural communities, and developed proxy measures where local data was unavailable.

From data to strategy

Beyond the base index, we developed three risk overlay analyses that add depth to the MVA map. A hedonic regression model identifies where homes sell above or below expected prices for their type, isolating the effect of location on value. An assessment-to-sale gap analysis highlights areas where appraisals may fall short of purchase prices, creating financing barriers for homebuyers. And a displacement risk index flags neighborhoods where long-term residents, particularly seniors and communities of color, may face pressure from changing market conditions.

We translated findings into a Strategy Toolkit with ten neighborhood vignettes covering all nine place types. Each vignette profiles a real community within the MAPA region, explains what is driving its MVA score, and identifies targeted policy responses. Strategies range from first-time homebuyer assistance and land bank activation in historically disinvested areas of North Omaha to zoning reform and proactive infrastructure planning in growing rural communities.

Built for ongoing use

The MVA is published as an interactive StoryMap with downloadable data, designed for four primary audiences: local planners anticipating trends for land use and transit planning, affordable housing developers identifying sites for investment, foundations targeting disinvested or at-risk neighborhoods, and policymakers comparing conditions across their districts. A Strategic Plan outlines how MAPA can maintain, update, and extend the tool over time, including recommendations for future data overlays and integration into regional planning processes like the upcoming Comprehensive Economic Development Strategy.

Project deliverables include the interactive map and data platform, a full report document, the Strategy Toolkit, an overlay analysis memo, and a comprehensive data library with all index inputs and scores by Census tract. The project included extensive stakeholder engagement through focus groups with planners, developers, foundations, and policymakers across both states.

The post Regional Market Value Analysis for Greater Omaha-Council Bluffs appeared first on CommunityScale.

Housing Forecast is a free dashboard covering every city, town, and county in the US. It draws on the American Community Survey, Census PUMS, Bureau of Labor Statistics, HUD, LODES, and private data sources to show how housing conditions are changing in a community, and what those changes mean for the people who live there.

What you’ll find

Start by searching for any place in the country. For McKinney, TX, for instance, you’ll see population projections showing how rapidly the city has grown and where that growth is headed. Scroll down and you’ll find household composition trends, a breakdown of the housing stock by type and tenure, and affordability metrics comparing the area median income to the income required to afford a typical home.

Each community’s dashboard concludes with a housing production target: an estimate of how many additional units would need to be built to accommodate projected growth and reduce the supply pressures that make housing less affordable. This is the number that turns a diagnostic into a conversation about action.

Charts include their data sources and methodology, and they update as new data is released. If you want to go deeper, the documentation is on GitHub.

Why we built it

At CommunityScale, we spend most of our time working directly with communities on housing needs assessments, zoning analyses, and planning studies. That work consistently reveals the same pattern: housing conversations stall when different stakeholders are working with different information. A planning board member cites one set of numbers, a developer another, and a neighborhood group a third. Before anyone can discuss solutions, they first have to agree on the problem.

Housing Forecast gives everyone the same starting point. It uses consistent definitions, data sources, and methodology across every geography, so comparisons between communities are apples-to-apples, and conversations can move from “is there a problem?” to “what should we do about it?”

We believe the most productive housing conversations focus on supply. When we look at communities that have made real progress on affordability, the common thread is that they’ve found ways to get more housing built, in more places, at more price points. Housing Forecast is designed around that framing: it highlights the gap between what a community has and what it needs, and points toward the production response that could close it.

How we think about the data

The American Community Survey is the gold standard for demographic and housing data, but its 5-year estimates inherently look backward. The 2024 release, for example, reflects conditions averaged across 2020 to 2024. That’s a rich foundation, but it can miss sharp turns.

We supplement ACS and PUMS with higher-frequency data: Zillow’s Home Value Index for monthly home value and rental trends, BLS data for inflation adjustment, and HUD income limits for the AMI calculations that anchor our affordability analysis. By blending the ACS’s comprehensive demographic detail with these more current market signals, Housing Forecast can produce analyses that are both demographically grounded and responsive to what’s happening now.

Population projections use a cohort-based approach, analyzing how age groups within a community have shifted over the past decade and projecting those trends forward. This captures the dynamics that matter most for housing: a community with a growing population of 30-somethings faces very different housing needs than one with a rapidly growing retiree population.

Who it’s for

We designed Housing Forecast for anyone with a stake in local housing conditions.

If you’re a planner or municipal staffer, you now have a shared baseline you can point to in conversations with elected officials, developers, and residents. If you’re a policymaker or advocate, you can pull charts directly into a presentation, a grant proposal, or public testimony to illustrate why a particular reform or investment matters. If you’re a developer or financial professional, you can quickly assess supply gaps and demand signals in communities you’re evaluating. And if you’re a curious resident, you can explore why housing feels harder to find in your community and what the data says about why.

What’s next

Housing Forecast is in beta. We’re continuing to add data sources, refine the methodology, and expand the set of metrics available for each community. We have custom dashboards in development for the Catawba Regional Council of Governments in South Carolina and the North Central Texas Council of Governments, and we’re actively working with other regional agencies and municipalities on tailored versions.

If your organization could benefit from a custom Housing Forecast, or if you need additional planning services around housing supply and zoning, get in touch. And if you want to stay informed as we add new features, sign up for updates.

We think every community deserves to understand its housing story. Housing Forecast is our way of making that possible.

The post Introducing the Housing Forecast app appeared first on CommunityScale.

The study encompasses six subregions across Will County, IL, combining a housing needs assessment, market analysis, suitability mapping, barriers analysis, and stakeholder engagement into an actionable strategic plan. The work responds to a set of interconnected challenges: an aging population, young professionals leaving the county, limited housing diversity, and rising costs that have pushed homeownership out of reach for many workers.

A County at a Crossroads

Will County has grown steadily since 2010, but that growth has been almost entirely in low-density, single-family subdivisions built on previously undeveloped land. Meanwhile, the county’s most walkable and amenity-rich areas have older housing stock and have actually seen a net decline in housing units. On average, 814 more people aged 20-34 leave Will County annually than move in, while the 65-and-over population is projected to increase significantly in every subregion.

These demographic shifts are changing what housing the county needs. Fewer families with children and more single-person households are driving demand toward smaller units, townhomes, condos, and apartments. But with only about 25% of the housing stock in multifamily buildings and limited rental options outside the Joliet area, the current inventory is not well positioned to serve these emerging needs. As a result, the county may struggle to attract the higher-paying employers and employees needed to help it compete economically with other counties in the region.

Understanding Market Gaps

The study’s market analysis uses household personas to illustrate how the county’s housing gaps play out for real people representing the types of workers and households Will County would like to attract and retain. A newly graduated registered nurse earning $72,000 and looking for a small rental finds only 8 to 20 listings per month in their price range countywide, with just 6 units built in the last 20 years. A retiring finance worker looking to downsize into a 1-2 bedroom home has somewhat more options, but a young couple with a child and a combined income of $117,000 faces a narrow and competitive market for starter homes.

Home prices remain affordable to median income earners in most subregions, but housing is considerably less affordable than it was a decade ago. Young, single-earning professionals just entering the workforce are effectively priced out of homeownership across most of the county.

Where Housing Can Go

A GIS-based housing suitability analysis scores locations across the county based on infrastructure capacity, access to jobs and services, land use diversity, walkability, and transit proximity. Areas scoring highest, concentrated around established downtowns and transit stations, are best positioned to support the kinds of denser and more diverse housing that would help attract the households and workforce the county needs to improve its competitive position in the region. The analysis also incorporates a spatial premium and discount model to identify where regulatory changes alone could unlock development versus where public investment or subsidies are needed to make projects feasible.

Downtown Joliet emerges as a particularly compelling opportunity, with the highest housing-readiness scores, substantial publicly owned land, and underutilized parking lots that could support catalytic mixed-use development. Transit-adjacent sites, including underused commuter parking lots with reduced post-pandemic ridership, present additional potential across the county.

Building a Strategic Action Plan

The study estimates that Will County needs approximately 25,000 new housing units over the next 10 years to keep up with projected household growth and address existing shortages. Nearly 5,000 rental units are needed immediately to fill gaps in the current housing stock, where low vacancy rates indicate that demand is outstripping supply.

Stakeholder engagement addressing a variety of topics, including market realities, regulatory barriers, housing products, site strategies, workforce retention, and political messaging, has shaped a set of strategic recommendations organized as policies, programs, and partnerships. Missing middle housing, including townhomes, duplexes, cottage clusters, and small-lot single-family, was the most frequently cited gap across all stakeholder sessions. The strategic action plan provides a best practices library and priority actions that municipalities can implement to promote the production of needed housing types.

The post Will County, IL Housing Study appeared first on CommunityScale.

Their key insight is the gap between theoretical AI capability and observed AI adoption. While 94% of tasks in Computer & Math occupations could theoretically be accelerated by AI, only 32% are actually seeing significant AI usage in professional settings. Office & Admin roles show the highest observed adoption at 42%, followed by Computer & Math (32%) and Business & Finance (28%). On the other end, physical and in-person occupations like agriculture (2%), food service (2%), and grounds maintenance (2%) remain largely untouched.

These occupation-level differences matter enormously at the community level, because every county has a different workforce mix. A county dominated by office and technology jobs will feel AI’s effects much sooner than one built around agriculture and manufacturing.

To see how this plays out geographically, we weighted each county's occupation mix (from the 2024 ACS 5-Year estimates, table C24010) by Anthropic's observed AI coverage rates. The result is a composite AI exposure index for every county in the country, covering 162 million workers across 3,222 counties.

The geographic pattern is striking. The highest-exposure counties cluster in the places you'd expect: the knowledge-economy corridors around Washington, D.C., San Francisco, and other major metros. Arlington County, VA leads among large counties at 20.4%, driven by its concentration of federal contractors, consulting firms, and technology companies. Neighboring Loudoun County, VA (20.2%) and Fairfax County, VA (19.2%) are close behind, forming a Northern Virginia corridor where roughly one in five workers' tasks overlap with current AI capabilities.

Other high-exposure counties include San Francisco (19.0%), Manhattan (18.8%), Santa Clara County (18.5%, home to Silicon Valley's nearly one million workers), and King County, WA (18.4%, the Seattle metro). The common thread is a workforce heavy on management, business, finance, and technology roles.

At the other end, rural counties with economies built around agriculture, manufacturing, and physical labor show exposure rates in the low teens or single digits. The national employment-weighted average sits at 15.6%, meaning the typical American worker is in a county where about one in six job tasks has meaningful AI overlap today.

What this means for communities

The Anthropic research found no systematic increase in unemployment for highly AI-exposed workers since late 2022, though there is suggestive evidence that hiring of younger workers (ages 22 to 25) has slowed in the most exposed occupations. This is not a crisis today. But it is a signal, and one that communities should take seriously for three reasons.

Workforce concentration creates vulnerability. Counties where a large share of workers are in AI-exposed occupations face correlated risk. If AI capabilities advance and adoption deepens (as they almost certainly will), the gap between Anthropic's "theoretical capability" and "observed exposure" will narrow. A county where 20% of tasks are already covered by AI could see that figure climb substantially as tools improve and deployment spreads. Communities with diversified workforces have more resilience built in.

AI exposure tracks with housing cost. The highest-exposure counties are, almost without exception, high-cost housing markets. San Francisco, Manhattan, Arlington, Santa Clara: these are places where housing affordability already depends on the high incomes that knowledge-economy jobs provide. Anthropic's data shows that workers in the most exposed occupations earn 47% more than those in unexposed roles. If AI disrupts these earnings, the housing markets that depend on them face a compounding challenge.

The young-worker signal matters for growing communities. The finding that entry-level hiring may be slowing in AI-exposed occupations is particularly relevant for communities that have been attracting young professionals. Places like Williamson County, TN (18.8% exposure, in the Nashville metro) and Douglas County, CO (19.2%, south of Denver) have seen rapid population growth fueled by young workers in exactly the kinds of jobs that show the highest AI adoption. If that pipeline of new workers slows, it affects everything from housing demand to municipal revenue projections.

Building on local data

This analysis is a starting point. The AI exposure index we've calculated reflects the current state of AI adoption, and as Anthropic notes, actual usage remains a fraction of what's theoretically possible. The gap between the blue bars (theoretical capability) and the red bars (observed exposure) in their research represents the future runway of disruption. As that gap closes, the geographic concentration of affected workers will only become more relevant for local planning.

Communities that understand their workforce composition have a head start. Explore your own county's workforce data on CommunityScale, and read the full Anthropic paper here.

Download a shareable social media card:

The post Mapping exposure to AI workforce disruption appeared first on CommunityScale.

For the municipalities, regional agencies, and states that CommunityScale works with, these bills represent both an opportunity and a call to action. Here’s what you need to know.

Federal zoning guidelines: a national playbook

Both bills direct HUD to publish federal guidelines on state and local zoning best practices. A task force of planners, architects, developers, community members, and local officials will draft recommendations covering:

• Reducing or eliminating parking minimums
• Expanding by-right multifamily development, including duplexes, triplexes, and quadplexes
• Legalizing accessory dwelling units
• Increasing allowable floor area ratios and building heights
• Transit-oriented development
• Streamlined, nondiscretionary development review
• Reducing minimum lot sizes and setbacks
• Eliminating barriers to manufactured and modular housing

The guidelines will be advisory, not mandatory. But they will set the standard against which communities are evaluated when competing for federal grants, and they will provide a framework that states can use to develop their own enabling legislation.

For communities already pursuing these reforms, this is validation. Massachusetts’ MBTA Communities Act, which required 177 municipalities to adopt by-right multifamily zoning near transit, anticipated many of these federal recommendations. CommunityScale worked with 13 communities on MBTA Communities compliance, and the lessons from that experience are directly applicable as federal guidelines emerge.

New money for housing plans and zoning reform

The House bill creates a new competitive grant program for housing planning and zoning reform. Regional planning agencies, states, cities, and counties can apply for funding to:

• Develop housing plans and housing strategies
• Update zoning codes
• Reduce barriers to housing supply
• Coordinate housing and transportation planning

This is a direct federal funding source for the kind of data-driven, community-specific planning work that produces results. Our work on housing needs assessments from downtown Syracuse to Stowe, Vermont to Northern Kentucky’s eight-county region shows what happens when communities invest in understanding their housing landscape: they develop actionable strategies that attract investment and build support for change.

Your CDBG allocation may depend on your housing growth rate

The Build Now Act adjusts Community Development Block Grant allocations based on whether a community is growing its housing supply faster or slower than before.

Every CDBG-receiving metropolitan city and urban county will be evaluated on a “housing growth improvement rate,” calculated as the ratio of recent housing unit growth to prior housing unit growth. Communities with above-median improvement receive bonus funding. Those below the median face a 10% reduction in their CDBG allocation.

This is significant. CDBG is a foundational federal program that funds everything from infrastructure improvements to public services. A 10% cut is a real financial consequence that creates real urgency for communities to identify and remove barriers to housing production.

The metric rewards improvement, not absolute volume. A small town that goes from building 10 units a year to 20 units a year shows the same improvement rate as a city that goes from 1,000 to 2,000. What matters is whether you are accelerating. Communities that have been losing ground can turn things around by adopting the kind of regulatory reforms and housing strategies that CommunityScale helps implement.

As the Build Now Act is formalized, CommunityScale will add the relevant metric to our Housing Forecast app.

$200 million per year for Innovation

The Senate bill also creates an Innovation Fund, authorized at $200 million per year, that awards competitive grants to communities that have demonstrated measurable improvement in housing supply growth. Grants range from $250,000 to $10 million, and the program explicitly prioritizes communities that have adopted zoning reforms like expanding by-right development, reducing parking requirements, legalizing ADUs, and streamlining permitting.

This creates a positive feedback loop. Communities that do the hard work of modernizing their zoning receive federal investment that makes their communities even better. The grants can fund community development activities, infrastructure improvements, and initiatives that expand attainable housing.

The data imperative

A thread running through both bills is the emphasis on data. The Build Now Act requires housing unit counts at the block level. The Innovation Fund requires applicants to document housing supply growth characteristics. CDBG recipients must report on 22 specific zoning reform categories.

Communities that invest in housing data infrastructure will be best positioned to respond. This means:

• Tracking building permits and certificates of occupancy in real time
• Monitoring housing unit growth against 5-year and 10-year baselines
• Benchmarking affordability metrics against HUD income limits and fair market rents
• Documenting zoning reforms and their impact on housing production
• Maintaining publicly accessible databases of publicly owned land

CommunityScale has built our practice around data-driven housing analysis, and we continue to develop new tools (including interactive dashboards and open-source mapping capabilities) to help communities understand and communicate their housing landscape.

What communities should do now

Regardless of the final legislative outcome, the direction of federal policy is clear. Communities that take the following steps now will be best positioned:

1. Assess your zoning. Evaluate your code against the 22-item checklist that CDBG recipients will be required to report on. Where do you stand on parking, ADUs, by-right multifamily, height limits, lot sizes, and permitting timelines?
2. Know your numbers. Calculate your housing unit growth rate over the past 5 and 10 years. Is it accelerating or decelerating? What is driving the trend?
3. Plan proactively. Develop or update your housing plan or housing needs assessment. This is both good planning practice and a prerequisite for many of the new federal programs.
4. Coordinate regionally. The federal programs emphasize regional approaches. Work with your regional planning agency, neighboring jurisdictions, and transit agencies to develop coordinated strategies.
5. Document your reforms. If you have already adopted progressive zoning or housing policies, make sure they are well documented. This evidence will be critical for competitive grant applications.

The federal government is recognizing what forward-thinking communities have known for years: housing supply is a national priority, and local regulatory reform is the key to unlocking it. At CommunityScale, we have helped communities across the country build the evidence base, update their zoning, and develop strategies that produce results. Contact us to discuss how these federal reforms affect your community.

The post Housing for the 21st Century Act & ROAD to Housing Act appeared first on CommunityScale.

This analysis works to understand potential predictors and patterns in US housing unit count growth data from 2020-2025. The most recent housing count record from 2025 considers more detailed, accurate, and timely data than the US Census can provide. Through these analyses, I hope to provide quantitative insight into what concrete changes will have influential factors on housing units, and subsequently, household counts. More generally, this project produces a preliminary understanding of the influence that various demographic factors have on housing unit counts at the county level. With a higher processing computer, the same workflow can be completed with census tracts or other more granular geometries. 

By implementing a Bayesian framework, this exploration considers uncertainty as an inherent contributor of the output distribution. For this specific exploration, the model considered a normal distribution of error centered around 0 for each input regressor. Consequently, the output of our model hypothesizes a distribution of values that include possible coefficients of our model and builds predicted value outcomes. When a coefficient distribution contains 0, we anticipate that the impact that this regressor has on our model may not be influential. This modeling technique also allows us to estimate changes to the posterior distribution of housing unit growth through posterior draws and subsequent manipulation.

The following findings detail regressors that, through our model, have confident influences on housing unit counts. Furthermore, by considering the model’s posterior estimates, I developed a function that considers how changes in the regression input could output a different expected housing unit change distribution. This interactive element is helpful in determining what changes will influence housing unit counts more than others. An important caution with this model is that these relationships might not be explicitly causal and should not be treated as such. 

Preliminary modeling

In addition to exploring vacancy percentages and household growth data through Bayesian techniques, the bulk of my exploration considered housing units as the dependent variable. After initially considering tract level geometries, I focused my exploration to counties. This exploration considered variables from the following Census and LEHD Origin-Destination Employment Statistics (LODES) Dataset categories: median income, renting percentage, standardized 2020 county population, median home value, county racial demographics, county citizenship demographics, wage changes from 2015 to 2020, vacancy percentage (excluding second homes), percentage experiencing poverty, and initially around 50 job categories from the LODES dataset. This last section was ultimately trimmed down to 28 variables using an exploratory data analysis process. After collinearity checks, this model ultimately considered 42 predictive variables of housing unit growth from 2020 to 2025. 

In this dataset containing over 3000 US counties, only 18 of them had negative or zero housing unit growth. For the sake of modeling accuracy, I removed these counties from the analysis. The codes for those counties are listed as follows: 02013, 02016, 02050, 02066, 02070, 02100, 02105, 02158, 02164, 02180, 02185, 02188, 02240, 02275, 02282, 15005, 48261, 48301. After this cleaning, I Winsorized my outliers: a standard outlier transformation technique used for model fitting ease that redefines all datapoints whose outcome is greater than the 95th percentile value to that upper-value limit. After this transformation, I applied a gamma distribution to fit the positive, rightly-skewed housing unit counts histogram. In an exploratory modeling phase, I considered a linear regression to predict changes in the model considering all of the above regressors. The linear regression showed predictive power and potential, but the fit increased with the gamma model. The output of that model can be found here. Consequently, for the Bayesian exploration, I continued to use a gamma fit.  

Bayesian exploration

After running 42 variables in the Bayesian regression, the model determined distributions of all of the regressor coefficients and determined an estimated posterior distribution. A visualization of the coefficient distributions are found in the density plots below. The variables that did not contain 0, and thus confidently influence the housing unit count outcome, are shown in green in the chart. The second graph shows the absolute values of all regressor coefficient distributions in ascending order and stratified by if they contain 0. The last visualization shows the coefficient distributions that do not contain zero, and their numeric summaries are in the chart that follows. 

In addition to these coefficient outputs, I modeled posterior predicted values to simulate the housing unit growth distribution based on our model. This distribution comparison is seen in the histogram below.

Using this synthetic data, I developed a function that considers alterations in the seven significant regressors and outputs how the posterior distribution may be expected to change. For example, the density distribution below is what is expected of the housing counts across the nation when the standardized number of jobs with earnings $1251/month to $3333/month decreases to 25% of its current value. This depicts the relationship between some lower paying jobs being offered at a high rate and fewer housing units growing. This tool can be manipulated in various ways to show numerous variable changes or specific regressor changes independently. 

Additional findings and next steps

In addition to building this model, I considered the potential influence of spatial autocorrelation on housing unit growth for various counties. Before implementing a spatial component to the Bayesian analysis through a spatial prior, I used a Global Moran’s I test to test the hypothesis that there is a random distribution of housing unit growth increase across adjacent and nonadjacent counties. After building a county adjacency matrix from Census data, I found a p-value of < 2.2e-16 which rejects the possibility that there is random spatial structure. This finding motivates the implementation of a Spatial Autoregressive Model to better fit housing unit count growth. The results of this model would consider county proximity to counties with increasing housing unit counts as a potential contributor to each county’s own housing unit growth. The gamma regression from this past exploration would be considered alongside a matrix defining county proximities, and the output coefficients of this model may vary from the nonspatial due to the new predictive considerations. 

Ultimately, this exploration provided a baseline understanding of how housing unit counts can be modeled and predicted across US counties. This research could be helpful in determining present day household counts, a statistic that is not readily available but could be determined through “now-casting”-like analyses. More generally speaking, Bayesian explorations of housing unit count data provide meaningful avenues for understanding the interaction of potentially influential demographic qualities on housing realities and discrepancies across the nation.

Thank you to Avery Cutler for the research and guest post.

The post Bayesian exploration of housing unit counts appeared first on CommunityScale.

Housing Forecast is a housing planning tool for communities in the US: an out-of-the-box dashboard highlighting local housing trends and needs with interactive maps, charts, and guiding text. Housing Forecast is meant for curious residents, researchers, planners, developers, financial professionals, and policymakers alike.

For each place, Housing Forecast considers demographic and employment trends, evaluates existing and planned housing supply, and identifies misalignments between available units and resident needs. Each Housing Forecast concludes with a local housing production target: an estimate of how much housing a place would need to build to accommodate projected growth and reduce supply strains that make housing less affordable. Charts include sources and methodology, and are updated on at least an annual basis as new data is released.

The post Housing Forecast App appeared first on CommunityScale.

Situated within the Charlotte metropolitan area’s sphere of influence, the Catawba region faces a set of interconnected housing challenges. Rapid job growth in York and Lancaster counties has not been matched by sufficient housing production, while Chester and Union counties are working to attract new investment and revitalize their communities. With the study focused on households earning between 80 and 150 percent of the area median income, the work targets the segment of the workforce most directly affected by gaps in housing availability, affordability, and diversity.

Growth Without Enough Housing to Match

The region has added jobs at a strong pace but housing production has not kept up, particularly for the types of units the workforce needs most. In York County, single-family homes are being permitted at roughly six times the rate of multifamily units, and 73% of the housing stock has three or more bedrooms.

At the same time, demographic trends are shifting what housing the region needs. The population aged 65 and over is the fastest-growing age cohort, while the share of residents aged 20 to 34, those entering the workforce, has remained flat since 2010. Households are getting smaller, with single-person households and adults living with roommates or parents growing faster than families with children. These shifts point toward a need for more smaller units, more rentals, and more variety in housing types than the current stock provides.

A Workforce Squeezed by the Market

The study’s workforce housing assessment uses real worker profiles to illustrate how market conditions affect people at different income levels and life stages. A registered nurse earning $91,000 per year and managing childcare costs can realistically afford only about 7% of homes recently listed in York County. An accountant looking to downsize after their children leave home finds that only 4% of listings meet their criteria for a smaller, newer unit on a manageable lot. Meanwhile, employers report that managers relocating to the region struggle to find rental apartments near job sites, in part because many rental units are advertised informally through word of mouth rather than conventional listings.

Vacancy rates reinforce the picture. Rental vacancy in York County has hovered around 5% since the early 2010s, below the 7.5% threshold typically associated with a healthy market. Ownership vacancy rates have also dropped below healthy levels. With the typical home in York County priced at $381,000, most individual job salaries in the region do not cover the cost of homeownership without combining incomes from multiple earners.

What Employers Are Saying

An employer survey conducted as part of the study found that 43% of respondents consider housing availability and affordability for their workforce a critical or important factor. One in three reported job candidates asking about housing during recruiting, and 20% said they have lost employees or candidates because they could not find suitable housing within a reasonable commute. Some employers are even factoring housing availability into site selection decisions, and others are considering automation as a response to persistent workforce challenges.

Stakeholder focus groups with municipal staff, economic development agencies, real estate professionals, and service providers surfaced a range of barriers on both the supply and demand sides. Zoning restrictions, minimum lot sizes, parking requirements, and material standards limit what can be built. High land costs, long permitting timelines, and the expense of extending infrastructure make it difficult for builders to deliver units at workforce price points. On the demand side, cash buyers outcompete mortgage-dependent households, childcare costs consume a significant share of family budgets, and limited transit options add commuting costs that further reduce what workers can spend on housing.

Strategies for a Regional Response

The study’s recommendations are organized into a strategies toolkit spanning seven topic areas: zoning and development standards, development economics, transportation and infrastructure, housing options, household finance, workforce development, and public sentiment and communication. Strategies are tiered by level of commitment, from policy changes such as updating zoning codes, to programs involving ongoing investment, to partnerships that bring catalytic public resources to bear on development opportunities.

Complementing the toolkit, a GIS-based housing readiness analysis identifies areas across the region best positioned to support denser and more diverse housing based on infrastructure capacity, access to jobs and amenities, and walkability. The study’s findings and interactive data are presented through an online housing data dashboard designed to support an informed community conversation about housing needs and opportunities across the four-county region.

The post Catawba, SC Regional Workforce Housing Study appeared first on CommunityScale.

The migration chain effect

When someone moves into a new apartment, they leave their previous home behind. A second household moves into that vacancy, leaving behind their home, and so on. Economist Evan Mast tracked these “migration chains” across 12 major U.S. cities, following 52,000 residents of new market-rate buildings through six rounds of moves. The findings are striking: by the sixth link in the chain, roughly 40% of movers were coming from neighborhoods with below-median incomes.

Mast’s simulation shows that 100 new market-rate units create between 45 and 70 equivalent units in below-median income neighborhoods. Even under conservative assumptions, the ripple effect extends deep into lower-income parts of the metro area. As CityLab reported, even pricey new units free up a significant amount of existing housing, and the process doesn’t take decades to unfold. Meaningful effects emerge in two to five years.

Real-world results

Minneapolis has been building more housing than comparable Midwestern cities for years, and the results are clear. Adjusted for local earnings, average rents in Minneapolis are down more than 20% since 2017, while rents in five other similarly sized and growing cities have risen over the same period. As one Minneapolis Star Tribune column put it, the chain reaction from new construction reaches into lower-rent districts every time a unit is added at the top of the market.

International evidence tells the same story. After Auckland, New Zealand upzoned large areas in 2016, its 25% rent premium over Wellington was erased within six years. The pattern is consistent: when cities allow more housing to be built, affordability improves across the market.

What this means for communities

In our housing shortage analysis, we’ve documented how supply deficits ripple through local markets, pushing costs higher at every income level. The filtering research confirms what we see in the data: housing markets are deeply interconnected. Restricting supply in one segment affects every other segment.

This is how new construction improves affordability. The chain reaction creates openings throughout the market. And importantly, the effect is strongest in tight markets with low vacancy rates, precisely where shortages are most acute and the need is greatest.

We’ve explored this interconnection from several angles, including the relationship between housing costs and household income, the fiscal case for housing growth, and the role of policy choices in shaping local housing outcomes. In each case, the conclusion points in the same direction: more supply leads to better outcomes for residents at every income level.

Supply is necessary, not sufficient

Filtering does not reach the very lowest-income households who cannot afford any market rent. Direct subsidies and dedicated affordable housing remain essential for those populations. As Timothy Taylor noted in his review of the research, market-rate construction and direct housing assistance are complementary strategies, not competing ones. But restricting supply makes every affordability challenge worse.

The bottom line

New housing benefits more than the people who can afford to live in it. Every new unit sets off a chain of moves that opens doors in neighborhoods throughout the region at a variety of income levels.

The post How filtering helps with affordability appeared first on CommunityScale.