To provide a mid-decade update of rental housing conditions and needs, the original study Affordability and Availability of Rental Housing in Pennsylvania approximated official comprehensive housing affordability strategy (CHAS) tabulations for 2000 by preparing equivalent estimates for 2005-06 from micro-data from the 2005 and 2006 ACS. The ACS micro-data do not identify each county, but instead identify public-use micro-data areas (PUMAs). PUMAs must have a population of at least 100,000 and therefore are aggregates of smaller counties when necessary. Because of limitations due to sample size and other procedural differences, the authors recognized that their 2005-06 estimates would differ from what official CHAS tabulations would show for the same counties if official estimates had been provided for this two-year span (2005-06).
The possible sources of procedural differences between the two estimates, as described in greater detail below, include:
In addition to these procedural differences, estimates for the 2005-07 period may differ from those of 2005-06 because of real changes in local conditions that are reflected in the additional year of data.
HUD’s official income limits. The income limits and affordability thresholds used in the 2005-07 CHAS tabulations of the ACS are based on HUD’s official income limits for each county that incorporate all necessary statutory adjustments.1 The original study also used HUD’s official income limits for each county identified. But when a PUMA contained more than one county, the authors developed weighted income limits, with each component county’s income limits weighted by the number of households in that county. This process is described in Appendix E of the original study.
The procedure used for 2005-06 income limits should, on average, approximate the official HUD limits used in developing CHAS data. Therefore, any differences in the income distributions of renters between the two periods should reflect only sampling variability or shifts in income distributions between 2006 and 2007.
Procedural differences in the treatment of cost burden. The original study calculated cost burden in 2005-06 for households reporting no cash rent but some utility payments because the authors judged it desirable to use all available data, and utility payments alone may well be unaffordable for the lowest income renters. The Census Bureau’s CHAS data, following bureau procedures, define cost burden as not available for such households. This procedural difference, as discussed in Appendix E of the study, would be expected to increase the incidence of rent burdens and severe burdens among extremely low-income (ELI) renters in 2005-06.
As anticipated, more ELI renters were recorded as having severe rent burden in 2005-06 than in 2005-07 at the state level, and in all but three of the counties that can be directly compared (Blair, Lebanon, and York). Nevertheless, the main results of the earlier study stand. In both periods, the incidence of severe rent burden was much higher among ELI renters than for very low-income (VLI) and low-income (LI) renters. Moreover, in general, counties with a low incidence of severe rent burden among ELI renters in 2005-06 (e.g., Fayette, Westmoreland, and Blair) also had a low incidence in 2005-07. Similarly, counties with a high incidence of severe rent burden among ELI renters in 2005-06 (e.g., Centre, Monroe, Delaware, and Bucks) also had a high incidence in 2005-07.
Inclusion of rental units. The 2005-06 estimates include all rental units, whereas those from 2005-07 count only standard rental units by excluding those with incomplete kitchen or plumbing facilities.
Differences due to sampling variability, particularly for smaller counties. Because the ACS is a sample survey, rather than a complete census of households, its results are subject to random sampling error. For any result, the margin of error provided by the Census Bureau estimates the amount of random sampling error. The larger the margin of error, the less likely it is that the reported results are close to the “true” figures for the entire population. In general, the larger the population of the county, the smaller the margin of error, and vice-versa. The tables showing the incidence of any housing problems illustrate this by reporting the margin of error as a percentage of the base number of households for ELI renters and owners. For Philadelphia, Pennsylvania’s largest county, the margin of error for the incidence of housing problems among ELI renters is only 3 percent. By contrast, for Juniata County, the county with the fewest reported renter households, this margin of error is 32 percent.
Conclusion: Comparisons between the 2005-06 and 2005-07 results reinforce the conclusions of the original study. The observed disparities in severe burden incidence among ELI renters are broadly consistent with the different treatment of cost burden, and the other differences observed between results from the two periods are likely to reflect small actual changes, minor procedural differences, and some sampling variability.
The confirmation of the original study’s results supports the utility of analyzing ACS micro-data for measuring trends in housing conditions among households classified as LI using the HUD program eligibility rules that define income in relation to HUD-adjusted AMI. The results moreover suggest that in the future similar use of the ACS micro-data to explore changes observed in CHAS tabulations (or not covered by CHAS tabulations) is warranted.