Local housing costs and basic household needs

California’s population is about 48% larger than Texas’ population, so if interstate migration rates were uniform and individuals randomly chose a new state of residence, one would still expect differences even without cost-of-living considerations.

The data can be found at: http://​www.​huduser.​org/​portal/​datasets/​fmr.​html.

Blomquist et al. (1988) estimate pecuniary values for quality of life across 253 urban counties and note that income can be adjusted for the implicit quality of life due to variations in amenities. The fact that both wages and costs vary by locality means that housing affordability—usually measured as the ratio of housing cost to household income—is not interchangeable with housing cost. Affordability is often denoted as spending more than 30% of household income on housing (Stone 2006; Schwartz and Wilson 2007).

Rising housing costs may improve well-being for homeowners due to housing wealth effects (Bostic et al. 2009). In addition, the annual cost of home ownership (known as the “user cost”) is low when house prices are appreciating (Himmelberg et al. 2005).

Hoynes (2000) finds that worsening local labor market conditions (such as the employment-to-population ratio) are associated with longer welfare spells and higher recidivism rates.

Yelowitz (2007) finds that housing prices lead to modest increases in living with parents for young adults. Rogers and Winkler (2014) find that a $100 increase in monthly rent reduces the probability of living independently by economically small 0.01 percentage points.

Glaeser et al. (2016) find a positive association between rents and subjective well-being in the Behavioral Risk Factor Surveillance System (BRFSS).

Bartfeld and Dunifon (2006) also include the state unemployment rate and find that it impacts food insecurity. Neither Fletcher et al. (2009) or Harkness et al. (2009) include regional employment characteristics.

Between 9–14% of SIPP respondents do not answer the well-being questions depending on the SIPP panel. In early panels, responses are coded as missing, and in later panels responses are imputed using hot-deck methods. Non-response to the well-being questions was more likely among males, non-whites, and younger reference persons, while marital status and educational attainment were uncorrelated with non-response. The results on the key policy variables are insensitive to the inclusion or exclusion of these imputed values.

Although many households, especially those with children, will demand housing units with more than two bedrooms, the FMR in this study is meant to proxy for general housing costs in an area. HUD scales other bedroom sizes relative to the two-bedroom FMR using information from the decennial Census; thus, the changes within a metropolitan area over time will be highly correlated regardless of bedroom size.

Starting in 2001, HUD began calculating FMRs at the 50th percentile in a small number of metropolitan areas. However, no such bridge file was found to convert FMRs to the 45th percentile for the 2003 data. The objective in rescaling the FMR was to give housing authorities a tool to deconcentrate voucher program use patterns.

Nord (2000) and Kurre (2003) show that living costs are substantially lower in non-metro areas.

The 35 low-cost areas include: Adams County, IN; Albany, NY; Albuquerque, NM; Bakersfield, CA; Baton Rouge, LA; Beaumont, TX; Birmingham, AL; Buffalo, NY; Corpus Christi, TX; El Paso, TX; Fayetteville, NC; Fort Myers, FL; Fresno, CA; Greensboro, NC; Greenville, SC; Harrisburg, PA; Indianapolis, IN; Knoxville, TN; Lakeland, FL; McAllen, TX; Memphis, TN; Mobile, AL; Oklahoma City, OK; Palm Bay, FL; Pensacola, FL; Pittsburgh, PA; Rochester, NY; Rockford, IL; San Antonio, TX; Scranton, PA; Springfield, OH; Syracuse, NY; Toledo, OH; Tulsa, OK; Utica, NY. The 41 high-cost areas include: Atlanta, GA; Austin, TX; Boston, MA; Charlotte, NC; Chicago, IL; Cincinnati, OH; Cleveland, OH; Colorado Springs, CO; Columbus, OH; Dallas, TX; Denver, CO; Detroit, MI; Eugene, OR; Honolulu, HI; Houston, TX; Jacksonville, FL; Lansing, MI; Los Angeles, CA; Madison, WI; Miami, FL; Milwaukee, WI; Minneapolis, MN; Nashville, TN; New Orleans, LA; New York, NY; Norfolk, VA; Orlando, FL; Philadelphia, PA; Phoenix, AZ; Portland, OR; Raleigh, NC; Sacramento, CA; Salt Lake City, UT; San Diego, CA; San Francisco, CA; Seattle, WA; Springfield, MA; St. Louis, MO; Stockton, CA; Tampa, FL; W. Palm Beach, FL.

Angrist and Pischke (2009) discuss advantages of the linear probability model. For the preferred specification (“Specification 3”), the vast majority of the predicted probabilities lie within the 0/1 interval for each of the eleven binary variables (varying between 73 and 93%, depending on the outcome). Therefore, the potential bias in the linear probability model is reduced (Horrace and Oaxaca 2006). Each specification was also estimated as a probit model, and the substantive conclusions are unchanged.

For the 82 MSAs examined, the correlation between FMR and employment-to-population is weak; for example, the long-run pairwise correlation for the within-MSA changes in these two variables between 1992 and 2003 is −.12 and insignificant. The short-run pairwise correlations are −.36, .20, and −.07 between 1992/1995, 1995/1998, and 1998/2003, with the first two correlations being significant at the 10% level.

Approximately 15% of renters live in subsidized housing. It is possible that housing costs could affect unsubsidized renters differently than subsidized renters. However, the conclusions from Specification 5 on both the FMR and labor market are unchanged when only focusing on unsubsidized renters.

See, for example, http://​thedataweb.​rm.​census.​gov/​pub/​sipp/​1991/​s91tm2dd.​asc (variable TM8706).

See http://​www.​fhfa.​gov/​DataTools/​Downloads/​Pages/​House-Price-Index-Datasets.​aspx . For households in localities without a corresponding FHFA index, the value was set to zero, and an additional dummy variable for “FHFA index missing” was included.