Mapping Manufactured Housing

Manufactured homes, often referred to as "mobile homes," are an essential supply of affordable housing for millions of Americans. Built in a factory and then transported to the site, manufactured housing units cost considerably less to build per square foot than site-built dwellings and are one of the  most affordable housing options today . This makes manufactured housing a lifeline for families seeking stability in an increasingly unaffordable housing market.

The American Community Survey collects data documenting the percentage of manufactured homes across the country. These data show that manufactured homes are primarily concentrated in Southern and Western states (see Map 1).

Most manufactured homes are located in either a residential subdivision or a manufactured home park. In parks, residents usually own their homes but rent the land beneath them. As a result, homeowners are vulnerable to increases in land rent or  forced eviction due to park closures . This makes owning a home in a manufactured home park considerably less secure than owning a manufactured home in a residential subdivision, where residents often own both the home and the land.

Despite the importance of land tenure and its impact on housing stability for manufactured home owners, we know surprisingly little about how parks and subdivisions are distributed across the country.

The  Department of Homeland Security  maintains a registry of "mobile home parks" across the country, while the  Department of Housing and Urban Development  and the  Texas Office of the Attorney General  maintain registries of "colonias" — informal subdivisions with poor-quality infrastructure and housing conditions and high shares of manufactured housing — found along the U.S.-Mexico border.

Maintaining complete and accurate registries of manufactured housing, whether located in parks or subdivisions, is important for efforts to monitor housing and infrastructure conditions and to intervene to mitigate damage in the event of storms and natural disasters. However, the existing registries have a number of limitations.

The  HUD  and  Texas OAG  databases  only document the location of colonias developed before 1990 and within 150 miles of the US-Mexico border , criteria laid out in the National Affordable Housing Act of 1990. As I show elsewhere, however,  informal subdivisions with limited infrastructure and poor-quality housing continued to develop throughout the border region after 1990 . Although pre-1990 subdivisions, like the one in Hidalgo County, Texas, shown in Figure 2 (see blue shading), are documented in state and federal colonias registries, newer subdivisions, like the large subdivision on the left-hand side of Figure 2, are not systematically documented in any state or federal registries.

Millions of manufactured homes are also located in subdivisions outside the US-Mexico border region. For example, Aristocrat Ranchettes (see Map 2) is a subdivision and census designated place northeast of Denver, Colorado, where approximately one-third of the more than 500 housing units are manufactured homes (American Community Survey, 2018-2022 5-Year Estimates). No state or federal registries track the location of or monitor conditions in non-border subdivisions like Aristocrat Ranchettes.

In some of these subdivisions, manufactured homes make up the vast majority of housing units. Figure 3 shows an example of one of these manufactured home subdivisions — see the cluster of yellow parcels, which denote manufactured housing on residential parcels, in the center of the map. Manufactured home subdivisions like this are often difficult to distinguish from manufactured home parks based on their appearance in satellite imagery alone. Indeed, in the DHS registry, this community is flagged as a park (see the black pushpin in Figure 3) although it is in fact a subdivision — the error is likely due to the visual similarity of manufactured home subdivisions and manufactured home parks.

In other subdivisions, site-built buildings make up the majority of housing units, but manufactured housing is also common. This is true in many state- and federally designated colonias, as well as in other subdivisions outside the border region like Aristocrat Ranchettes.  Elsewhere, I refer to these as informal subdivisions  because they are typically developed on inexpensive land with minimal infrastructure, and homeowners often  self-build their homes or live in affordably priced (and often second-hand) manufactured homes . The left-hand side of Figure 3 shows an informal subdivision in Webb County, Texas, where between 10% and 30% of parcels have a manufactured home (see yellow shading). This is contrasted with the formal subdivision at the bottom right where there are no manufactured homes.

Unlike the colonias registries, which only track the location of subdivisions within 150 miles of the US-Mexico border, the DHS "mobile home park" registry is national in scope (see Map 3). However, it also has important limitations.

For example, the DHS dataset documents the location of more than 56,000 "parks," although many of these locations are in fact subdivisions. Figure 5 provides an example in Webb County, Texas. Two communities on the map are included in the DHS dataset, as indicated by the black pushpin. The community on the left is a true manufactured home park, as indicated by the large parcel (outlined in blue) on which dozens of manufactured homes sit. The community on the right, however, is a manufactured home subdivision, with each building located on a separate parcel (those parcels recorded as manufactured homes on residential land are shaded yellow). In total, we estimate that as many as 15% of the DHS locations in Texas are subdivisions rather than parks.

Moreover, the DHS database is also likely missing thousands of parks nationwide. For example, the two parks shown in blue outlines in Map 4 are not tracked by DHS but were identified via our analysis of building footprints and parcel boundaries. In total, we estimate that approximately 8% of parks in Texas are not included in the DHS registry. Nationwide, this means that there may be as many as 4,000 parks that are not currently tracked by the DHS

Our databases document the predicted location of buildings in parks and subdivisions nationwide. We provide access to these predictions at different spatial scales: buildings, census block groups, and counties. These data can be used to supplement existing registries of these communities and can inform emergency management response efforts and the allocation of funding for housing and infrastructure conditions in these communities. However, our databases have a number of limitations that users should keep in mind.

Most importantly, our national databases are based on predictions of the location of location of parks and subdivisions, rather than verified locations. Thus, although our models achieve accuracies of 95-99%, they are accompanied by prediction errors. In particular, our estimates are highly sensitive to the location of singlewide manufactured homes, because these buildings have a distinct, elongated building footprint. Parks and subdivisions with doublewide manufactured homes are less likely to be included in our database.

Moreover, our models also have difficulty distinguishing between manufactured home subdivisions and manufactured home parks — this is not surprising given the similar appearance of these communities. For example, in prior work I found that I was able to correctly distinguish between parks and subdivisions  only 80% of the time  by satellite imagery alone.

If your goal is to systematically verify the location of parks and subdivisions in a particular area, our building footprint-level database is a good place to start. But we recommend that you combine it with additional data, such as parcel boundaries and land use classifications in property records, to accurately distinguish between parks and subdivisions. We have found that pairing satellite imagery, building footprints, and parcel records is the most reliable way to do so.

You can also download our block group and county estimates, or you can use the maps above to examine the distribution of parks and subdivisions at these scales. Pairing our block group and county-level databases with estimates of manufactured housing from the American Community Survey (ACS) is also helpful. Estimates from the ACS are useful for accurately mapping the distribution of manufactured housing in general, but provide no insight into land tenure arrangements; our databases can be used to supplement ACS estimates to more reliably distinguish between parks and subdivisions at these scales.

To build our database, we first compiled a national dataset of building footprints by combining building footprint polygons published by  Microsoft  and  OpenStreetMap .

We then used a sample of 5 million buildings across 19 Texas counties to create a validated set of buildings in parks and subdivisions. We identified approximately 1,500 verified manufactured home parks by systematically reviewing the DHS locations and triangulating them with satellite imagery and parcel records. To identify informal or manufactured home subdivisions, we used county property records to find buildings located in residential subdivisions where 10% or more of the buildings were manufactured homes.

We used the  High-Performance Computing Center at Michigan State University  to predict whether buildings were located in a manufactured home park and informal or manufactured home subdivision. To do so, we measured the size, shape, and arrangement of each building and its 5, 10, and 20 nearest neighbors using the Python programming language. We then used a machine learning technique called Random Forests to predict if each building in the nation was located in a park or subdivision.

Our models achieve a relatively high accuracy of 95-99%. However, our models have lower precision (70% for parks, 46% for subdivisions) and recall (87% for parks, 63% for subdivisions). This is attributable, in part, to the fact that manufactured homes make up a small fraction of buildings. However, our models are also particularly responsive to the location of singlewide manufactured homes, which have a distinct, elongated shape that is relatively easy to identify using building footprints. Many of the doublewide manufactured homes located in parks and subdivisions will not be included in our database.