Spatial Mismatch of the Poor: An Explanation of Recent Declines in Job Isolation

ABSTRACT:

Despite years of spatial mismatch hypothesis research, the poor’s access to jobs is not well understood. This article provides a detailed analysis of the distributional imbalance between the residential locations of poor and nonpoor families and jobs over the 1990s in U.S. metropolitan areas. Descriptive and multivariate evidence are presented addressing whether job access for the poor has improved over the 1990s. If so, what factors drove the improvement, and do they help explain the disparity in job access between the poor and nonpoor? Using data from the U.S. Census, the Economic Census and the Zip Code Business Pattern files, the analysis reveals how the change in metropolitan factors over the 1990s explains the gap in job access between the poor and nonpoor. Specifically, the factors which increase parity or exacerbate it are identified. Results indicate that job access for the poor has improved significantly, particularly for black poor over the decade. A decomposition of job isolation indicates that change in dynamic metropolitan-wide factors such as job sprawl and the variation of housing affordability overall impedes movement toward parity in access to jobs between the poor and nonpoor.

Notes

1 Segregation among the poor is a prominent feature of American society in that issues surrounding class determine access to particular neighborhoods. Although we expect less heterogeneity concerning job access for the poor, we expect that white poor will depict more heterogeneity in access as a result of their ability to penetrate neighborhoods that other ethnic minorities are unable to penetrate. The white poor are less segregated in poverty concentrated neighborhoods (Jargowsky, 2003).

2 As a result of the economic boom black unemployment by 2000 was at an all time low. In 1999, the black unemployment rate was 8%. While this was nearly double the national unemployment rate, the annual rate of 8% is the lowest recorded value for black unemployment rates since the Bureau of Labor Statistics began to collect separate data for African Americans in 1972. See Table B-42 in the Economic Report of the President, U.S. Government Printing Office, 2001.

3 See Raphael and Stoll (2002) for an empirical analysis of the narrowing spatial mismatch between blacks and jobs in the 1990s.

4 See Jargowsky (2003) for a descriptive analysis of the dramatic declines of concentrated poverty over the 1990s.

5 It is important to acknowledge that over this period the economic conditions of black men may be overstated since unemployment does not capture the disparaged worker or men capable of working but not counted in the unemployment measure because they are within the institutionalized population (Jensen & Slack, 2003).

6 See Martin (2001) for one of the first empirical analyses of spatial mismatch to use the dissimilarity index to operationalize the imbalance in jobs-to-people.

7 The U.S. Bureau of Labor Statistics for 2001 estimates that retail trade accounts for 18% of all jobs.

8 The estimates were explored for poor families and all poor individuals including poor children and there were no measurable differences in the estimates. However, selection of the family unit allows one to average poverty over all the members of the family making it a more conservative approach.

9 To be sure, a mismatch index based on the dissimilarity measure does not actually measure the physical distance between the average member of a given populations and jobs. The index measures the imbalance across geographic subunits of the metropolitan area (for example, zip-codes or census tracts) between members of the population and jobs. To take an extreme example, suppose that all poor residents resided in one zip code of a city while all jobs were located in a different zip code. Whether these two zip codes are one mile apart from one another or 20 miles apart will not influence the dissimilarity measure. In both instances, the dissimilarity index will be equal to 100. Nonetheless, as a summary measure, the dissimilarity measure does allow comparison of geographic areas over time as well as comparisons across geographic areas. For mismatch measures that take into account distance between populations and jobs, see Raphael (1998).

10 Unfortunately, 1994 is the earliest year of the Zip Code Business Pattern data files, and data on total employment by zip code is not provided in the 1992 Economic Census. Hence, one is forced to use the 1994 total employment data for the 1990 jobs/total employment mismatch indices.

11 The metropolitan areas used in the analysis are Metropolitan Statistical Areas (MSAs) and Primary Metropolitan Statistical Areas (PMSAs) as defined by the Office of Management and Budget (OMB) in 1999 for Census 2000. Consolidated Metropolitan Statistical Areas (CMSAs), which are usually much larger than MSAs or PMSAs, were not included among these metropolitan areas.

12 An alternative decomposition would add and subtract w¹⁹⁹⁰₁I²⁰⁰⁰_i to the original expression for the change in the index value. After factoring, this would yield the decomposition:

where again, the first term is the component driven by within-area improvements in the index and the second term is the component driven by between-area migration. These two decompositions may differ slightly depending on the average changes in the index values and the distribution of the changes in weights. To account for these differences, the decomposition in the analysis is based on the average of these two equations (as is the convention). Specifically, the estimate of the within-area improvement component is calculated by computing both decompositions (given by Equations (4) and the one above) taking the average of the first terms from the two equations. Estimates of the between-area contribution to the improvement are calculated by taking the average of the second terms from the two equations. Since both decompositions yield very similar results, conclusions do not appear to be sensitive to the averaging or the choice of decomposition.

13 Paired means tests were used to determine statistical significance at 95% probability. These results are not included in the article.

14 Glaeser and Vigdor (2001) show that overall black/non-black segregation levels are currently at their lowest point since the 1920s. The 1990s continued a three-decade trend toward decreasing segregation throughout the United States.

15 In addition to the 5 mile radius, I experimented with boundaries that are located outside of a 3 and 10 mile radius centered on the metropolitan area’s Central Business District to examine the sensitivity of this 5 mile boundary to alternative distances. These alternative measures of job sprawl were highly correlated with that shown here (though they differed in levels as to be expected) and did not produce qualitatively different results than those presented in this analysis. Moreover, this measure of sprawl has been used elsewhere, and is correlated with other measures of sprawl, such as the concentration/centralization of people (since the spatial distribution of all people and jobs is highly correlated), and with measures typically used by economists to measure employment density, such as spatially based employment density gradients (Glaeser & Kahn, 2001a; Kahn, 2001).

16 As constructed, the variable captures whether or not an MSA received Hope VI funding over the decade. This was the author’s best attempt to construct this variable compared to an aggregation of total monetary funds invested within each MSA standardized per capita. The total per capita expenditure, especially in large MSAs, is inconsequential. For example, in Los Angeles, the second largest MSA in the United States, an estimated $73 million was invested across two public housing authorities over the decade; that translates into $4,478 per 1,000 people. There appear to be limitations with the use of this variable; overall, the program and the funding generated from the program may be too small to show an effect.

17 Multiple measures of affordable housing were tested, Hope VI funding targeted at individual MSAs, dispersion of households with a housing cost burden different from 30% and 25%, and a rental payment differing from fair market rent. Only the inclusion of the coefficient of variation representing the difference in housing cost burden from 25% seemed to consistently improve models as indicated by the Adjusted R Squares for the model and hence this variable was preferred and included in multivariate models.

18 For example, there is no statistical difference in the magnitude of the coefficients in equations predicting job isolation for the restricted versus unrestricted sample (for those independent variables that were available). This occurred despite the fact that the omitted metropolitan areas were generally of smaller size (with respect to population).

19 The author determined a significant positive relationship through correlation and bivariate tests; the results are not provided in this article.

20 FDIC, Housing Bubble Concerns and the Outlook for Mortgage Credit Quality, FDIC Outlook (2004), accessed December 21, 2005: http://www.fdic.gov/bank/analytical/regional/ro20041q/na/infocus.html document. “Strong demand for housing, facilitated by low interest rates, has pushed home prices to their highest rates of appreciation in more than a decade. But this sturdy price appreciation has not been accompanied by equally strong personal income growth. Since 2000, annual home price appreciation has averaged roughly 7%, while disposable per capita personal income gained 4% per year, on average sales of existing homes rose to record levels in 2003, surpassing previous records set in 2002.”

21 Essentially, the federal government enacted the LIHTC to provide ten years of tax credits to investors who back developments in which a portion of units are made affordable for lower-income renters for at lease 15 years. The Internal Revenue Service (IRS) administers the program and developers apply to the agency to receive tax credit allocations in exchange for building units that are affordable to low-income households (Freeman, 2004).

22 See Freeman and Botein (2002). Overall, LIHTC units do not inevitably “drag down” their surrounding neighborhoods.

23 Arthur C. Nelson and others, “The Link Between Growth Management and Housing Affordability: The Academic Evidence” (Washington, DC: Brookings Institution, 2002).