The overarching goal of our analysis was to create a long-term dataset on housing density change that is accurate, spatially detailed, and consistent across the United States. The maps and data were designed for strategic decision making and for the visualization of housing growth patterns over large areas. The data are too coarse for most county- or township-level land use planning, and more detailed and current data are often available for local applications.
To calculate our housing growth estimates, we relied on existing data sources and made assumptions about past growth rates and settlement patterns. Our maps represent a conservative estimate of change. It is important for users to understand how the dataset was generated, what errors the source data may contain, and upon which assumptions it is built. The objective of this technical documentation is to provide this information.
The decennial census is the only national, comprehensive source of longitudinal information on housing patterns across the U.S. However, digitized census boundaries at fine scales only became available in 1990, and boundaries of census tracts, block groups, and blocks change so frequently that earlier data at these scales cannot be easily mapped. We thus developed: Digital data on the smallest census units for which long-term housing data can be analyzed (partial block groups), and Algorithms to estimate housing density for each partial block group, by decade, back to 1940 and forward to 2030.
Partial Block Group Formation
The smallest units for which the Census Bureau provides both detailed social data and digital geographic boundaries are block groups. However, block groups are often subdivided into smaller spatial units by other boundaries such as those forming incorporated places, legal and census-designated county subdivisions, and rural/urban areas. Block groups thus frequently consist of multiple "partial" block groups for which data tabulations are also provided. And housing unit density is often different among the partial block groups that make up a single block group. Analyzing housing density at the block group level obfuscates important housing density differences that are revealed by using partial block groups.
The formation of partial block groups began with 2000 Census blocks (U.S. Census Bureau, 2004). Each Census block contains a sequence of identifiers denoting both the block group to which it belongs, and any other geographic boundaries intersecting that block group (see above). Together the identifiers indicate the partial block group to which each block belongs. We used these identifiers to remove all census block boundaries that fell within a given partial block group, dissolving census block boundaries to form partial block groups. It is important to note that partial block groups sometimes consist of multiple polygons that are not necessarily spatially contiguous. In our designation of partial block groups, we assumed that Census blocks with zero housing units in 2000 also had no houses in any prior decade and thus we excluded them from the partial block groups.
While the Census Bureau does not provide partial block group boundaries readily, it does tabulate housing data at this level. The Summary File 3A (U.S. Bureau of the Census, 2002) includes tabulations for these partial block groups.
Historic Estimates and Adjustments
Census takers ask "in what year was this housing unit built?" In 2000, the "year housing unit built" question was coded with the following response options: 1999 or 2000, 1995 to 1998, 1990 to 1994, 1980 to 1989, 1970 to 1979, 1960 to 1969, 1950 to 1959, 1940 to 1949, and 1939 or earlier. We aggregate the three initial categories between 1990 and 2000 to produce consistent decadal estimates.
Initial estimates of historic housing density (i.e., backcasts) were derived by summing all housing units built prior to a given decade for each partial block group (e.g., for a 1970 estimate we sum housing units built from 1939-earlier through 1960). However, it is important to note that these initial estimates exclude housing units that were present in the past, but were no longer present in 2000. We adjusted for these 'missing' housing units by comparing county-level totals of our initial estimates with historic county-level totals reported by the Census Bureau. If historic Census counts were greater than the county-level totals initially estimated, we applied a correction and distributed it across the partial block groups. Essentially, we raised the housing density by a uniform factor for all partial block groups in the county to ensure that the historic estimates summed to the county level matched the recorded historic county totals.
Census blocks can capture areas with no housing such as military and some other public lands. However, no accurate, fine-scale, and consistent public land ownership data is available for the United States, and thus, no actual land ownership data was used in generating our maps and data. This means that some partial block groups contain portions of public lands where housing is excluded, and may give the false impression that there are houses in areas where houses do not exist. Users should be aware of this feature of our data.
For further details on the estimation of historic housing densities, please consult Hammer et al. 2004. Characterizing spatial and temporal residential density patterns across the U.S. Midwest, 1940-1990. Landscape and Urban Planning 69: 183-199
Future Estimates and Adjustments
The 2000 partial block group geography created for the backcasts was maintained in producing forecasts for 2010, 2020, and 2030. Initial estimates were produced by extrapolating 1990s housing growth rates over the coming three decades. To control for potential over- or underestimation arising from unique, localized growth patterns occurring during the 1990s, initial future estimates were adjusted using commercially produced population estimates from Woods and Poole Economics (W&P) (http://www.woodsandpoole.com/). W&P population projections (Woods & Poole 2004) were converted to housing unit projections based on county-specific household size in 2000, as reported in the same dataset. This results in a conservative estimate of future housing unit projections in relation to population because household size may continue its decline in the future, thus raising the number of housing units associated with a given population. We used the resulting county housing projections to control our partial block group estimates for future decades, just as we use the historic county Census totals to modify backcasts (i.e., summing to the county level and comparing our county estimate to the county W&P control total, adjusting our county estimate as needed, and distributing the adjustment to the constituent partial block groups).
U.S. Census Bureau, 2004. "2004 TIGER/Line Files".
U.S. Census Bureau, 2002. "2000 Census of Population and Housing, Summary File 3".
Woods & Poole Economics Inc. 2004. "2004 Regional Projections and Database". Washington D.C.