different results, but researchers have also estimated differing index values for the same areas even when using similar methods and data (e.g., compare Blackley, Follain, and Lee, 1986, and Thibodeau, 1989). The work at BLS to extend and improve the hedonic methodology so that the results are more stable with respect to such factors as the choice of reference area or independent variables is very promising, but this effort is still developmental. Moreover, data problems remain: the data source with the largest sample size and coverage (the decennial census) has limited information on housing characteristics, while other data sources that are richer in content (the CPI database and the American Housing Survey) are smaller in size and restricted in the areas they cover.11
Yet despite all the methodological problems and uncertainties, it is clear that the cost of housing differs across geographic location. For example, HUD fair market rents differ significantly across areas even when they are adjusted for the median income of the area. Overall, we believe the findings support the importance of an adjustment of the poverty thresholds for geographic variations in housing costs.
Furthermore, despite the problems and uncertainties, the literature helps indicate the size of geographic area for which an adjustment would be feasible and appropriate. Data are not available with which to develop housing cost indexes for every city and town in the United States, but an adjustment for areas classified by population size within region would accord with findings that intraregional differences are highly correlated with population: larger cities or metropolitan areas within a region are more expensive than smaller areas. This pattern is evident in the results from Kokoski, Cardiff, and Moulton (1992, 1994), and in other studies as well (e.g., Thibodeau, 1989); Ruggles (1990) recommends an adjustment of this type.
At the current state of knowledge, we conclude that a feasible way to move toward a comprehensive interarea price index with which to adjust the poverty thresholds is first to develop an interarea price index for shelter. Not only are housing costs a large component of a poverty budget, but housing cost
The national component of the American Housing Survey is conducted every two years and currently includes about 57,000 housing units; the sample is designed to produce national estimates, and the geographic identification made available to users is limited to four regions and central city-suburb and urban-rural classifications. The metropolitan component currently includes samples of about 5,000 housing units in each of 44 metropolitan areas; 11 areas are surveyed each year on a rotating cycle. The CPI database (described above) obtains price data for about 85 areas, most of which are combined for publication into size classes within each of four regions.