Using Bayesian statistics to predict housing growth in the U.S.

Posted 01/15/14

Information about housing densities at fine spatial scales is critical to understanding how human development impacts species of interest. However, available data is at the county level, which is too large a scale for most ecological studies. Nick Keuler, statistician for the SILVIS lab, and Roger Hammer at Oregon State University are using Bayesian statistics to project housing densities into the past and future at smaller units of scale, called partial block groups (PBGs).

Information about housing densities at fine spatial scales is critical to understanding how human development impacts species of interest. However, available data is at the county level, which is too large a scale for most ecological studies. Nick Keuler, statistician for the SILVIS lab, and Roger Hammer at Oregon State University are using Bayesian statistics to project housing densities into the past and future at smaller units of scale, called partial block groups (PBGs).Beginning in 2000, the US census began to ask heads of households when their housing unit was built at the partial block groups (PBGs) level. Prior censuses only collected this data at the county level. This means that while there is good housing estimates at the county level for censuses prior to 2000 the data doesn’t exist at the PBG level. Nick and Roger are using the 2000 census data to backcast how many houses there were in each PBG at different points in time, going back to the 1940 census. For example, the 2000 census might report PBG-level information about 900 housing units in a particular county in 1990, whereas it is known that 1000 housing units existed in that county in that year. The discrepancy of 100 housing units could be because owners didn’t report or didn’t know then their units were built. Nick is using Bayesian statistics to predict to which PBGs within the county those missing 100 units should be allocated. Bayesian statistics allows him to use information on neighboring PBGs, as well as past patterns of development, to make those predictions.

Information about historical housing units is interesting and useful, but so is information about the future patterns of housing density. Using roughly the same method but applying it to existing predictions of future county housing unit numbers through 2100, Nick and Roger are forecasting where future housing units will be built within PBGs.These data will be important to a wide variety of researchers at SILVIS and beyond, and some are already seeing use. For example, SILVIS postdoc Eric Wood is using the new housing projections to predict the impact of development around protected areas on bird richness.”