Do Local Economic Conditions Affect Homelessness? Impact of Area Housing Market Factors, Unemployment, and Poverty on Community Homeless Rates

Abstract

This article estimates the impact of local housing and labor market conditions on area homelessness using the U.S. Department of Housing and Urban Development's (HUD’s) annual point-in-time counts of homelessness from 2007 to 2014. In cross-sectional models, the median rent, the share of households in rental housing, and the poverty rate have strong positive impacts on homelessness. Once area-fixed effects are included, only the median rent remains positive and significant. However, fixed-effect models find a positive relationship between poverty and homelessness in communities that maintain right-to-shelter policies, suggesting constraints in shelter bed supply may limit responses of homelessness to changes in economic conditions.

Keywords:

• Homeless
• labor market
• income
• low-income housing

Notes

1. Although it might be desirable to use denominators for the family and individual homeless rates that are based on the total number of persons living in families with children and the total number of individuals living alone, this analysis does not do this because the data source used to construct the denominator (the U.S. Census Bureau annual estimates of population by county) is not disaggregated by family status.

2. Individuals not living with family members include people living in a household of size one and people living with nonrelatives.

3. The analysis did experiment with some specifications that included very rough estimates of federal prevention dollars spent in each community over the analysis period, but did not find that including this variable changed the results.

4. The six CoC with larger areas excluded include the Oklahoma City CoC (42% excluded), the Connecticut Balance of State CoC (32% excluded), the Kansas City, Independence, Lee's Summit/Jackson County CoC (21% excluded), the Brookline/Newton CoC (18% excluded), the Tulsa City and County CoC (13% excluded), and the Massachusetts Balance of State CoC (11% excluded).

5. Most of the consolidated CoC were linked to one county; however, two consolidated CoC in Massachusetts were linked to an aggregated county group. The Springfield/Hamden CoC and the Pittsfield/Berkshire, Franklin, Hampshire Counties CoC were combined to form an aggregated CoC that was matched to a county group that included Hampden, Hampshire, Franklin and Berkshire Counties. In addition, the Lowell, Cambridge, Quincy/Weymouth, Malden/Medford, Somerville, Brookline/Newton, and the MA Balance of State CoC were matched to a county group that includes Norfolk and Middlesex counties.

6. For example, the geographic boundaries of the Miami/Dade County CoC include 100% of the area of PUMA numbers 234 and 235, and 64% of the area of PUMA number 236. Thus, the estimated poverty rate for the Miami/Dade CoC would be: where i,234 is an index for each individual in the Census living in PUMA 234; i,235 is an index for each individual living in PUMA 235; and i,236 is an index for each individual living in PUMA 236. The variable Poor_i is a dummy variable that equals 1 if the ith individual has an income below the poverty line, and is 0 otherwise. The variable w_i is the sample weight from the American Community Survey.

7. These estimates assume that any CoC which contains overlapping PUMA should be consolidated into a larger CoC. It collapses 76 of the original 381 CoC into 30 aggregated CoC, leaving a total of 335 CoC.

8. To calculate proportional responses, this analysis uses the estimated coefficients and the means reported in Table 2. For example, using the coefficient in the first column of Table 3 (top), a 10% increase in the rental share would be associated with a proportional increase in homelessness of 0.10*33.21*0.823/20.52 = 13.3%.

9. Shelter-all policies based on email exchange with Sharon McDonald of the National Alliance to End Homelessness on February 6, 2016, and confirmed by conversation with Lisa Thornquist of the Hennepin County to End Homelessness on June 24, 2016, and emails from Dee Norton, Director of IT with the Road Home in Salt Lake City, Utah on June 21, 2016, and on June 17, 2016, with the DC Interagency Council on Homelessness. Washington DC has had, since 1984, a policy that guarantees access to shelter to families and individuals when the temperature including wind chill index falls below 32°F. They state that, in practice, individuals can access shelter year round, and they enacted a policy in 2016 that guarantees year-round access to shelter for families as well. The Road Home adopted a policy of no waiting lists for shelter for families and individuals in October 2010.

10. In shelter-all communities, the mean homeless rate is 48.76 per 10,000, the mean value of the median rent is $1,142, and the mean poverty rate is 15.69%. Thus, a 10% increase in poverty rates in shelter-all communities would be associated with an increase in homelessness of 0.10*15.69*(−0.360 + 1.583)/48.76 = 0.05, or 5%.

11. All models include controls for the percentage of the population that is Black, Hispanic, a veteran, a baby boomer, or an immigrant, or that has more than a high school education. Each of these controls was tabulated from the American Community Survey.

12. To calculate the proportional response implied by each coefficient, the coefficient is multiplied by 0.1 times the mean of the independent variable and then divided by the mean of the dependent variable. Thus, the proportional impact of an increase in state poverty rates is 0.1*14.896*0.4249/2.030 = 0.312 or 31%, and the impact of an increase in area rents is 0.1*934.3*0.0064/2.030 = 0.295 or 30%.