JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000, pp. 473–497
THE CAUSES OF REGIONAL VARIATIONS IN U.S. POVERTY:
A CROSS-COUNTY ANALYSIS*
William Levernier
Department of Finance and Economics, Georgia Southern University, Statesboro, GA,
30458 U.S.A. E-mail:
Mark D. Partridge
Department of Economics, St. Cloud State University, St. Cloud, MN 56301, U.S.A.
E-mail [email protected]
Dan S. Rickman
Department of Economics and Legal Studies, Oklahoma State University, Stillwater,
OK 74078, USA. E-mail: [email protected]
ABSTRACT. The persistence of poverty in the modern American economy, with rates of
poverty in some areas approaching those of less advanced economies, remains a central
concern among policy makers. Therefore, in this study we use U.S. county-level data to
explore potential explanations for the observed regional variation in the rates of poverty.
The use of counties allows examination of both nonmetropolitan area and metropolitan
area poverty. Factors considered include those that relate to both area economic performance and area demographic composition. Specific county economic factors examined
include economic growth, industry restructuring, and labor market skills mismatches.
1.
INTRODUCTION
Despite long periods of U.S. economic growth in the 1980s and 1990s, the
relative economic position of low-income families has deteriorated. For example,
the historical link between economic growth and reduced poverty appeared to
weaken in the 1980s (Blank and Card, 1993) as family poverty rates rose above
their levels of the late 1970s. Moreover, there are still significant differences in
poverty across areas within the United States (Triest, 1997). These facts, along
with federal attempts to reform welfare have heightened interest in the underlying causes of poverty.
*Earlier versions of this paper have been presented at: University of Oklahoma and Oklahoma
State University economics workshops; the 34th Missouri Valley Economic Association Meetings,
Kansas City, MO; the 38th European Congress of the Regional Science Association International,
Vienna, Austria; and the 45th North American Meetings of the Regional Science Association
International, Sante Fe, NM. We thank participants at these presentations and the anonymous
referees for their many helpful comments.
Received August 1998; revised June 1999 and December 1999; accepted January 2000.
© Blackwell Publishers 2000.
Blackwell Publishers, 350 Main Street, Malden, MA 02148, USA and 108 Cowley Road, Oxford, OX4 1JF, UK.
473
474
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
Numerous explanations are suggested for recent poverty trends. For example, many studies focus on the causes of the decline in the low-skilled wage rate.
Demand-side explanations include the loss of manufacturing jobs (Bluestone,
1990), a shift in labor demand towards high-skilled occupations (Cutler and
Katz, 1991), and the decline of unions (Freeman, 1993). Similarly, increased
supply of low-skilled labor through immigration and increased labor market
competition associated with increased female labor force participation have been
found to reduce low-skilled male wages (Topel, 1994).
Other studies emphasize demographic components of poverty. For example,
an increased number of families headed by females is associated with increased
poverty (Blank and Hanratty, 1992). Poverty among blacks in central cities has
also worsened. Reasons suggested for poverty among inner-city blacks include
discrimination (Kirschenman and Neckerman, 1991), spatial mismatches between residence and job location (Holzer, 1991), and negative neighborhood
effects associated with inner cities (Corcoran et al., 1992; Cutler and Glaeser,
1995). Yet, other studies suggest that the lack of generosity of U.S. transfer
payments such as welfare underlie higher U.S. poverty (Blank and Hanratty,
1992).
To shed further light on national poverty trends and regional patterns in
poverty, in this study we examine differences in 1990 family poverty rates across
all counties and independent cities in the lower 48 states, resulting in over 3,000
observations. The use of county data allows us to examine the causes of both
nonmetropolitan and metropolitan poverty. Although the nonmetropolitan poverty rate is higher than that of metropolitan areas, nonmetropolitan poverty has
received considerably less attention in the literature.
In what follows, we examine the extent to which differences in county
poverty rates can be explained by various economic and demographic factors. In
particular we explore whether counties that experienced recent employment
growth have lower poverty. County-level data provide an alternative basis for
assessing the link between area economic performance and poverty than multistate regional data (e.g., Blank and Card, 1993; Triest, 1997) because aggregation can mask important labor market effects. In addition to assessing the role
of industry composition in influencing area poverty, we explore whether counties
that underwent recent structural change have higher poverty. We also attempt
to discover whether higher poverty in central-city counties and nonmetropolitan
counties is related to a mismatch between area labor force job skills and area
skills composition of employment. Finally, we address whether these economic
factors interact with county type, education, and race.
2.
MODEL OF REGIONAL POVERTY
Conceptual Model
Poverty rates can vary across geographic areas because of differences in
both person-specific and place-specific characteristics. For example, an area may
have a higher rate of poverty simply because it contains disproportionately
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
475
higher population shares of demographic groups associated with greater poverty.
Alternatively, area poverty may be more related to place-specific factors such as
its economic performance. A strong area economy may sufficiently reduce the
poverty rate among all groups such that the overall poverty rate is lower.
Moreover, relative poverty rates of particular demographic groups may be
interrelated with area economic conditions.
Regarding person-specific characteristics, poverty rates are relatively
higher nationally for most minority groups. One factor suggested to underlie
higher minority poverty is discrimination or racial preferences in hiring (Kirschenman and Neckerman, 1991; Ihlanfeldt and Young, 1996). Alternatively, it
is argued that the key to reducing poverty among minorities is to improve their
quality of education and to increase their education completion rates (e.g., Smith
and Welch, 1986). Also focusing on the supply side, Mead (1992) argues that
reservation wages of blacks lead them to not accept available jobs. Viscusi (1986)
suggests that relatively higher rates of return to crime may be one reason for
this.
Poverty rates also are higher for female-headed families across all racial
groups (Blank and Hanratty, 1992). Besides being the sole potential wage earner
for the family, female family heads are disproportionately young, lesser educated, and less skilled. Moreover, child-care constraints can adversely affect their
job performance further. Thus, female heads receive lower wage rates and are
less likely to participate in the labor force. Indeed, Blank and Hanratty (1992)
suggest that some of the relative increase in U.S. poverty in the 1980s compared
to Canada was due to the relatively greater increase in U.S. female-headed
households.
Low-skilled workers in general are more likely to experience poverty. One
suggested primary cause for the relative decline of low-skilled wages is a
hypothesized relative demand shift that has favored high-skilled occupations
(Juhn, Murphy, and Pierce, 1993). Along with technological change, a prominent
explanation for the skill shift is the decline in manufacturing and “good” paying
jobs for those with lesser job skills (e.g., Cutler and Katz, 1991). Also, labor force
participation fell for those whose wage rates dropped (Topel, 1993). As an
example of the interrelationship between demographic patterns of poverty and
economic performance, Wilson (1987) argues that structural changes and demand shifts particularly hurt blacks who are relatively lesser skilled and lesser
educated.
Declining low-skilled wages also have been argued to be caused by supply
shifts. For example, the relative growth in the supply of college graduates
relative to high school dropouts dramatically declined in the 1980s compared to
previous decades (Juhn, 1999). This supply-side change has been attributed as
being a primary cause of the relative decline of high school dropout wages
compared to college graduate wages since the late 1970s. Increased immigration
of disproportionately low-skilled workers also has been linked to increased male
wage inequality (Topel, 1994). However, at the regional level less-skilled natives
may out-migrate in response to the arrival of immigrants (Frey, 1995), which
© Blackwell Publishers 2000.
476
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
greatly complicates the regional wage effects of immigration (e.g., Borjas,
Freeman, and Katz, 1996). Increased labor supply resulting from increased
female labor force participation in recent decades also has been linked to
increased male wage inequality (Topel, 1994), indirectly increasing poverty
among families of low-skilled males. Nevertheless, increased labor force participation by wives in low-income families somewhat offset the earnings losses of
husbands (Cancian, Danziger, and Gottschalk, 1993). Therefore, the net effects
of immigration and female labor force participation on family poverty are
conceptually ambiguous.
Much of the work on the link between area economic performance and
poverty has been done at the metropolitan level. On one hand, total metropolitan
area employment growth and tight labor markets have been reported to benefit
low-income individuals more than high-income individuals, particularly young
black workers (Freeman, 1991; Bartik, 1996). For example, strong labor demand
may provide employment opportunities for low-skilled individuals that otherwise would not exist. On the other hand, in-migration of low-skilled workers
that have more experience, or are more educated, can mitigate the potential
benefits of employment growth for low-skilled natives (Larson, 1989; Sawicki
and Moody, 1997). Moreover, accompanying shifts in skill-level demand can
offset the beneficial effects of increased aggregate job availability (Cutler and
Katz, 1991). Consequently, the historical positive link between growth and
reduced poverty may have been weakened in the 1980s (e.g., Blank and Card,
1993).
A relatively unexplored aspect of the relationship between local economic
conditions and poverty is the degree to which changes in industry structure
affect area poverty. That is, if there are adjustment costs associated with
changing sectors, longer-term unemployment may result (Partridge and Rickman, 1998). In addition, post-displacement earnings are typically lower than
pre-displacement earnings (Carrington and Zaman, 1994), where a likely causal
factor is job-specific training. Therefore, areas that experience significant industrial restructuring (aside from any losses of manufacturing jobs) are expected
to have increased poverty. That is, the process of switching sectors (e.g., from
manufacturing to services) can reduce income and increase poverty.
Related to area economic performance, a substantial literature exists on the
contribution to poverty rates of “spatial mismatch” factors in central cities. For
example, besides the general decline in jobs, another trend is the relocation of
manufacturing jobs from central cities to their suburbs. This relocation may
increase the locational imbalance between the demand for low-skilled workers
in suburbs and the supply of low-skilled workers in inner cities. Regarding racial
aspects of spatial mismatch, blacks, who are disproportionately concentrated in
inner cities, have been observed to be less likely to increase their commutes to
offset the relocation of inner-city jobs to suburban areas (Holzer, Ihlanfeldt, and
Sjoquist, 1994). Also, housing discrimination (Turner, 1992) and suburban
zoning practices (O’Regan and Quigley, 1991) may prevent inner-city residents
from moving closer to the jobs, whereas in-migrants to a metropolitan area may
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
477
be more likely to locate near the newly created jobs than inner-city residents
(Sawicki and Moody, 1997). As a weaker form of spatial mismatch, the importance of neighborhood effects such as peer pressure, poor role models, and scarce
information about jobs may explain inner-city poverty (Corcoran et al., 1992;
O’Regan and Quigley, 1991). Cutler and Glaeser (1995) argue that broader social
problems that affect youths growing up in poverty-stricken areas determine
neighborhood effects, not proximity to jobs. However, the issue of spatial mismatch in its various forms remains unsettled (Holzer, 1991).
Rural areas possess higher poverty rates than their urban counterparts for
many reasons (Rural Sociological Society Task Force on Persistent Rural Poverty,
1993). Some of the higher rural poverty may be related to lower cost of living,
reliance on agricultural and other extractive industries, demographic characteristics, and less human capital in the labor force (Brown and Warner, 1991).
However, an unexplored question is whether employment growth and human
capital have differential effects on rural poverty versus urban poverty. Nevertheless, other aspects of rural areas may contribute to higher poverty after
controlling for these effects. For example, geographic isolation of rural residents
and their unwillingness to migrate to nearby growth centers may contribute to
a mismatch between jobs and worker skills that may be more severe than those
in urban areas (Brown and Warner, 1991; Rural Sociological Society Task Force
on Persistent Rural Poverty, 1993). Yet, with poverty more diffused in rural areas
than central cities (Rural Sociological Society Task Force on Persistent Rural
Poverty, 1993), fewer negative neighborhood effects may exist and “middle-class”
values among low-income households may be more prevalent.
Empirical Model
In our empirical model, we examine the poverty rate level. Conversely,
Madden (1996) regressed the 1980 to 1990 change in MSA-level poverty rates
on 1980–1990 changes in the independent variables. There are several other
differences between the current study and Madden’s, including: our orientation
on counties versus metropolitan areas, our consideration of nonmetropolitan
areas, and different choices of independent variables and tested hypotheses. The
two approaches complement each other despite the many differences. For
example, our approach appears better at exploring how a county with a high
poverty rate differs from a county with a low poverty rate. Using changes in
poverty rates is not as well suited for this comparison. For example, consider
two metropolitan areas: one whose poverty rate increased from 25 percent in
1980 to 28 percent in 1990, and another one whose poverty rate increased
from 5 percent to 8 percent. A first-difference approach would generally treat
the two areas as equivalent because they both had a three-percentage-point
increase in their poverty rate even though the former had a significantly greater
poverty problem than the latter. Our levels approach directly addresses the
difference between low and high pockets of poverty. Conversely, the changes
approach may be better in examining why some metropolitan areas had declining poverty rates while others had increasing poverty rates, which is a more
© Blackwell Publishers 2000.
478
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
challenging question for studies that only employ one cross section. Yet, both
approaches should generally yield similar responses.1
Therefore, we assess the importance of the above factors in explaining area
differences in poverty using the following levels-based empirical model. The
poverty rate POV in county i, state s, is regressed on several independent
variables that are intended to capture the effect of person-specific and placespecific county characteristics discussed above and their interrelationships as
follows2
(1)
POVis = α1 + β1 CTY_TYPEis + γ1 DEMOGis + φ1 ECONis + σs + εis
where CTY_TYPE is a vector that represents the type of county; DEMOG
denotes demographic characteristics of the population; ECON contains variables related to area economic performance; σs denotes state fixed effect; α1, β1,
γ1, and φ1 represent corresponding coefficients; and ε is the error term with the
usual assumptions.3
State fixed effects account for the poverty effects of omitted state-level
variables, which if correlated with the included independent variables would
otherwise bias the coefficients of the remaining variables by their omission.
Omitted factors may include cost of living, amenity effects, cultural influences
(Partridge, Partridge and Rickman, 1998) and state government policies including
welfare programs. The inclusion of state fixed effects means that the slope
coefficients reflect variation across counties within states because their inclusion
1
Suppose that the 1980 and 1990 poverty rates for MSA or county i in state s can be depicted
by the following
POVRATEis1990 = βXis1990 + eis1990
POVRATEis1980 = βXis1980 + eis1980
with X representing the matrix of independent variables, β the coefficient vector, and e the residual.
Our model estimates the first equation. Conversely, Madden (1996) estimates the difference of the
two equations
∆POVRATEis1990–1980 = β∆Xis1990–1980 + ∆eis1990–1980
The two approaches should yield equivalent coefficient estimates (that is, the same β). The changes
approach has the advantage if there is a separate county fixed effect (besides the state fixed effect)
because it would be differenced away. However, our levels approach has an advantage because
10-year changes may not reflect shorter-term poverty-rate changes. As Bartik (1993) notes, economic
conditions early in a decade may completely differ from those at the end of the decade. Thus, our
levels approach allows poverty rates to be determined by more contemporaneous effects than
ten-year changes.
2
The official poverty rate is based on (before-tax) money income including public assistance.
Nonetheless, official poverty rates are not adjusted for several factors such as in-kind public welfare
programs like medicaid and regional cost-of-living differentials. See Slesnick (1993) for a detailed
discussion of problems with official poverty thresholds.
3
All of the variables in Equation (1) are from the 1990 Census of Population with the exception
of the 1988–1990 employment growth, 1990 employment, and 1988–1990 structural change (ISC)
variables in the ECON vector. These three variables are from U.S. Department of Labor, USA
Counties CD-ROM, and U.S. Department of Commerce, Regional Economic Information System.
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
479
is the equivalent of using first differences around the state mean. Nevertheless,
the sensitivity of the results to their inclusion is examined.
CTY_TYPE includes dummy variables for: (1) whether a county contains
the central city of metropolitan area; (2) whether the county is a suburb in a
large metropolitan area; (3) whether the county was a suburb in a small
metropolitan area; and (4) whether the county is a single-county metropolitan
area. To avoid perfect collinearity, the omitted category is nonmetropolitan
counties. We chose a population of 350,000 as the division between large and
small metropolitan areas. All else equal, poverty should be higher in central-city
counties relative to suburban counties if mismatches in job skills and neighborhood effects exist. For similar reasons, poverty is expected to be higher in
nonmetropolitan counties than suburban counties. However, whether the poverty rate differs between nonmetropolitan counties and central-city counties
may depend on the relative strengths of county-type labor market mismatches
and neighborhood effects. Moreover, the reader should be cautioned that many
“central city” counties also include large suburban populations which act to
make such counties more homogeneous. However, descriptive statistics described below show that suburban and central-city counties (combined with
single-county metropolitan areas) significantly differ in terms of economic and
demographic characteristics. Population size of metropolitan areas and nonmetropolitan counties is also included. Population may be related to factors such
as agglomeration and skills matching in the labor market, though population’s
role may depend on how increased size is correlated with distances between
residence and employment and density of economic activity.
Economic factors ECON include county-level measures of the 1988–1990
employment growth rate, one-digit industry shares (minus one), and a measure
of recent structural adjustment.4 A negative sign for employment growth would
support the hypothesis that tight labor markets reduce poverty. In particular,
Raphael (1998) strongly argues that employment growth is a much superior
measure to employment levels in detecting economic outcomes such as spatial
mismatch. One-digit industry shares capture the influence of manufacturing (or
other sectors) on area poverty with a lower than average coefficient expected for
manufacturing. Also included in ECON is recent industrial structural change
(ISC), which is simply a dissimilarity index measured as the sum of absolute
changes in the share of one-digit industry employment between two periods,
divided by two (see Allen and Freeman, 1995). The 1988–1990 ISC measures
what share of the labor force would have to shift one-digit sectors such that 1988
4
We chose county-level labor-market data to capture the most disaggregated labor market
information available. The reason for this is that counties more closely reflect the appropriate labor
market for many low-skilled individuals in large metropolitan areas and for workers in rural areas.
Hence, our use of the smallest possible geographic area for the labor market does not suffer from
the aggregation bias that likely occurs when using larger areas (e.g., regions, states, or large
metropolitan areas.) However, to the extent that counties do not accurately reflect the appropriate
labor market (particularly in metropolitan areas), our results should be interpreted with some
caution.
© Blackwell Publishers 2000.
480
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
and 1990 would have the same one-digit sectoral composition. A positive coefficient suggests adjustment costs in the reallocation of labor across sectors that
worsens the economic outcomes at the lower end through some combination of
increased unemployment and lower wage rates. Labor force participation rates
by gender are accounted for in ECON. The coefficients reflect effects on labor
force participation beyond labor demand, although employment growth in
neighboring counties may also influence labor force participation in the
county.
Also included in ECON is the employment/labor-force ratio, which is both
a proxy of labor market tightness and nearby employment accessibility. It is
expected that poverty rates will be negatively correlated with the employment/
labor-force ratio. Job accessibility concerns may be more severe for some central-city county residents and (isolated) nonmetropolitan county residents.
Hence, we also include interactions of the employment/labor-force ratio with the
central-city and nonmetropolitan county dummies. Similar employment/laborforce measures have been utilized in Spatial Mismatch Hypothesis studies (e.g.,
see Kain, 1992; Madden, 1996).
Demographic variables DEMOG include age and racial categories, the
percent of families headed by single females, and education attainment levels.
Poverty is expected to be lower for counties with higher education attainment
levels whereas the percent of families headed by single females is expected to
be positively related to poverty. Inclusion of race variables allows for examination of whether poverty differences across racial groups remain after controlling
for the potential effects on poverty of other variables correlated with race.
3.
EMPIRICAL RESULTS
Descriptive Statistics
Table 1, column 1, presents the unweighted descriptive statistics for the
entire sample of 3,023 counties in the 48 contiguous states. These statistics are
presented separately for nonmetropolitan counties and metropolitan area counties (MSAs) in columns 2 and 3, where only about one-fourth of U.S. counties are
in MSAs. In columns 4 and 5 we present within-metropolitan-area statistics for
central-city counties (and single-county MSAs) and suburban counties.
The unweighted average family poverty rate is 13 percent for the entire
sample, but there is significant dispersion across county type.5 Metropolitan
counties had poverty rates that were less than two-thirds of nonmetropolitan
poverty rates. Suburban family poverty rates were just one-half those of nonmetropolitan counties and even central-city counties and single-county MSAs
had poverty rates that were about 4 percentage points below nonmetropolitan
areas (where central-city county and single-county MSAs had poverty rates of
5
For comparison, family poverty rates were 10.3 percent nationally in 1989. Since 1960
national family poverty rates have ranged from 18.1 percent in 1960 to 8.8 percent in 1973–1974;
the rate was 10.3 percent in 1997.
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
481
TABLE 1: Descriptive Statisticsa
Variable
(1)
(2)
(3)
Full Sample Nonmetropolitan Metropolitan
Counties
Counties
Dependent Variable:
Family Poverty Rate
Metropolitan/
Nonmetropolitan:
Single County MSA
Small MSA Suburban Countyb
Large MSA Suburban Countyc
Central City Countyd
MSA Population
Nonmetropolitan County
Population
Economic Development:
1988–1990 ISC
1988–1990 Employment Growthe
Employment/Labor Force
%Civilian Female Labor
Force Participation
%Civilian Male Labor
Force Participation
Industry Composition:
%Agriculture, Forestry, Fisheries
13.04
(7.00)
14.35
(7.12)
0.053
(0.22)
0.03
(0.17)
0.11
(0.31)
0.05
(0.22)
na
na
na
na
na
na
23,586
(22,972)
10.42
(4.59)
(5)
Suburban
Counties
7.89
(4.40)
0.22
0.51
na
(0.41)
(0.50)
0.12
na
0.21
(0.33)
(0.41)
0.45
na
0.79
(0.50)
(0.41)
0.21
0.49
na
(0.41)
(0.50)
1,067,578
692234
1,350,654
(1,369,440) (1,332,789) (1,329,534)
na
na
na
0.034
(0.018)
0.07
(0.12)
1.05
(0.31)
0.036
(0.019)
0.07
(0.13)
1.06
(0.26)
0.028
(0.013)
0.09
(0.09)
1.04
(0.44)
0.022
(0.009)
0.07
(0.06)
1.20
(0.23)
0.032
(0.015)
0.10
(0.11)
0.92
(0.51)
51.94
(7.14)
50.22
(6.56)
57.29
(6.17)
56.72
(5.30)
57.7
(6.7)
70.35
(7.19)
68.97
(7.05)
74.64
(5.82)
73.35
(4.74)
75.6
(6.4)
2.6
(2.2)
26.0
(8.0)
7.1
(2.1)
21.5
(2.5)
6.1
(2.2)
31.7
(5.9)
5.1
(3.3)
2.2
(2.2)
23.4
(7.0)
6.7
(1.7)
22.3
(2.3)
6.4
(2.2)
34.3
(5.1)
4.9
(2.9)
2.9
(2.3)
28.0
(8.1)
7.4
(2.3)
21.0
(2.5)
5.9
(2.1)
29.7
(5.6)
5.2
(3.6)
8.6
(8.8)
%Goods Producing
27.2
(10.3)
%Transportation, Public Utilities 6.5
(2.0)
%Trade
19.7
(3.5)
%FIRE
4.4
(1.8)
%Services
29.1
(5.9)
%Public Administration
4.8
(3.0)
© Blackwell Publishers 2000.
na
8.97
(4.65)
(4)
Central City
Counties
10.6
(9.2)
27.5
(10.9)
6.3
(2.0)
19.1
(3.5)
3.8
(1.3)
28.3
(5.7)
4.7
(2.8)
482
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
Variable
Human K/Demographic:
%High School Graduate/
Some College
(1)
(2)
(3)
Full Sample Nonmetropolitan Metropolitan
Counties
Counties
56.0
(7.6)
%4 Year College Graduate
13.5
(6.6)
%Female-Headed Family
12.8
(5.4)
%18–24 years old
9.2
(3.7)
%60–64 years old
4.7
(1.0)
%65 and over
15.0
(4.4)
Average Children per Family
0.91
(0.14)
%African American
8.58
(14.25)
%Non African American Minority
3.90
(7.66)
%Same County in 1985
79.78
(8.28)
N
3023
55.8
(8.0)
11.8
(4.9)
12.3
(5.4)
8.7
(3.6)
4.9
(1.0)
15.9
(4.2)
0.91
(0.15)
8.16
(14.95)
3.85
(8.25)
80.90
(7.64)
2288
56.7
(6.1)
18.8
(8.3)
14.3
(4.8)
10.7
(3.5)
4.2
(0.8)
11.9
(3.4)
0.90
(0.13)
9.89
(11.76)
4.03
(5.43)
76.30
(9.19)
735
(4)
Central City
Counties
(5)
Suburban
Counties
56.0
(5.8)
20.3
(6.7)
16.7
(5.0)
11.5
(3.8)
4.2
(0.8)
12.5
(3.4)
0.90
(0.12)
12.00
(12.23)
5.73
(6.75)
78.46
(7.95)
316
57.2
(6.4)
17.7
(9.2)
12.6
(3.8)
10.0
(3.2)
4.1
(0.9)
11.5
(3.4)
0.89
(0.13)
8.31
(11.15)
2.75
(3.69)
74.67
(9.72)
419
a
The total sample originally had 3,109 counties but 86 counties were omitted due to data
availability.
b
Suburban MSA counties are defined as all counties in a multiple-county MSA that do not
contain the largest city in the metropolitan area. A small MSA is defined as a total MSA population
of less than 350,000.
c
Suburban MSA counties are defined as all counties in a multiple-county MSA that do not
contain the largest city in the metropolitan area. A large MSA is defined as a total MSA population
of greater than 350,000.
d
A central city MSA county is defined as the county containing the largest city in a multiplecounty MSA.
e
Change in employment during the time span divided by the level of employment at the
beginning of the period, or the employment growth ratio.
10.8 percent and 10.0 percent respectively). That is, despite the concerns of policy
makers regarding urban and central city poverty (e.g., Bradbury, Kodrzycki, and
Mayer, 1996), poverty rates are highest in nonmetropolitan counties.
A comparison of columns 2 and 3 shows that relative to nonmetropolitan
counties, MSA counties have higher average income, faster employment growth
in the late 1980s, and higher male and female labor force participation. In
addition, metropolitan counties experienced smaller sectoral reallocations (1988–
1990 ISC). For example, it would require 2.8 percent of the typical MSA’s labor
force to change one-digit sectors in 1988 and 1990 to equate industry composition
across the two periods, compared to 3.6 percent for nonmetropolitan counties.
Metropolitan and nonmetropolitan counties have similar employment/labor
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
483
force ratios. Nonmetropolitan counties have relatively higher employment shares
in agriculture and lower shares in FIRE and services. Metropolitan counties
have higher shares of college graduates and female-headed families whereas
nonmetropolitan counties have disproportionately more senior citizens (persons
65 and over) and fewer minorities.
Columns 4 and 5 show that within metropolitan areas, suburban counties
experienced more employment growth, more structural change, and lower
employment/labor force ratios than central-city counties. Suburban counties
have a higher employment share in the goods-producing sector but lower shares
in trade and services. Male and female labor-force participation rates are higher
in suburban areas whereas the share of adults with college degrees is higher in
central-city counties. Central-city counties also have higher minority population
shares. Also compared to nonmetropolitan counties, central-city counties experienced less industrial structural change and the same level of employment
growth.
Base Regression Results
Regarding how these characteristics explain differences in poverty rates
across counties, we turn to the regression analysis of Equation (1). We use the
full sample of counties for the contiguous 48 states, less 86 counties because of
sectoral employment nondisclosure problems in constructing the ISC variables.6
Table 2 shows the results for various formulations of Equation (1) beginning
with the base model in column 1 followed by more parsimonious specifications.
The purpose of presenting the alternative specifications is that they help
disentangle how various interrelated effects are influencing the final results.
For example, many demographic factors affect the poverty rate both directly and
also indirectly by influencing labor market outcomes. By considering a model
that includes demographic factors but not labor market variables, both the direct
and indirect demographic effects are reflected in the estimated coefficients for
that model. In addition, the results from more parsimonious models give some
indication of the degree to which multicollinearity influences the results in the
base model.
Column 1 of Table 2 shows the base regression model. The negative countytype coefficients show that all four MSA county types have significantly lower
poverty rates than nonmetropolitan counties (the omitted group) after a wide
range of characteristics are accounted for. Central-city counties have slightly
lower poverty rates than the other MSA county types, ceteris paribus, which
suggests that the higher average poverty rates in central-city counties
versus suburban counties are mostly produced by differing average levels
of their measured characteristics.7 Greater MSA and nonmetropolitan county
6
There appears to be nothing systematic about these 86 counties. The descriptive statistics
are almost the same in the entire sample of 3,109 counties and the subsample of 3,023 counties.
7
To analyze a premise related to the Spatial Mismatch Hypothesis, the following ratio was
derived: the central-city county share of the MSA minority population divided by the central-city
© Blackwell Publishers 2000.
484
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
TABLE 2: Poverty Regression Resultsa
Variable
Metropolitan/
Nonmetropolitan:
Single County MSA
Small MSA Suburban Countyb
Large MSA Suburban Countyc
Central City Countyd
MSA Population
Nonmetropolitan County
Population
Economic Development:
1988–1990 ISC
(1)
(3)
(4)
(5)
(6)
–2.35
(4.87)
–2.21
(4.99)
–2.13
(5.22)
–2.77
(3.59)
–2.3E-7
(3.57)
–3.02
(6.39)
–2.51
(5.56)
–2.54
(6.36)
–4.18
(5.48)
–4.5E-7
(7.25)
–2.03
(7.09)
–1.77
(6.29)
–1.75
(7.17)
–2.76
(8.69)
–2.4E-7
(3.16)
–2.78
(10.38)
–2.10
(7.38)
–2.27
(9.42)
–3.22
(10.65)
–3.6E-7
(5.05)
–4.63
(11.42)
–5.74
(11.05)
–7.82
(19.17)
–4.44
(12.18)
7.3E-8
(0.44)
–2.81
(1.96)
–2.87
(2.39)
–2.23
(2.11)
–5.32
(5.34)
–6.7E-7
(2.50)
–1.5E-5
(5.32)
–2.7E-5
(8.72)
–2.1E-5
(6.25)
–3.2E-5
(9.23)
–5.2E-5
(11.11)
–3.1E-5
(4.12)
9.18
(3.15)
1988–1990 Employment Growthe –0.37
(0.69)
Employment/Labor Force
–0.37
(1.35)
Employment/Labor Force
–0.81
Nonmetropolitan
(2.01)
Employ/Labor Force
–0.09
Central City
(0.14)
%Civilian Female Labor Force
–0.30
Participation
(16.05)
%Civilian Male Labor Force
–0.13
Participation
(7.17)
Industry Composition:
%Agriculture, Forestry, Fisheries 0.15
(5.79)
%Goods Producing
–0.04
(2.17)
%Transportation, Public
–0.007
Utilities
(0.21)
%Trade
0.08
(2.97)
%FIRE
–0.09
(2.01)
%Services
0.05
(2.48)
%Public Administration
na
Human K/Demographic:
%High School Graduate/Some
–0.30
College
(17.04)
%4 Year College Graduate
–0.14
(8.11)
© Blackwell Publishers 2000.
(2)
6.91
(2.18)
0.50
(0.92)
–0.38
(1.50)
–0.54
(1.41)
0.75
(1.20)
–0.31
(18.80)
–0.15
(8.21)
14.51
(4.57)
–0.77
(1.29)
10.93
(2.80)
–0.99
(1.29)
–0.76
(0.73)
–0.99
(0.84)
1.94
(1.49)
-0.29
(9.31)
–0.17
(5.74)
0.21
(9.03)
0.006
(0.42)
0.09
(2.93)
0.16
(6.36)
–0.06
(1.28)
0.09
(5.28)
na
0.04
(1.19)
–0.14
(6.15)
0.01
(0.27)
–0.04
(1.07)
–0.25
(4.56)
0.06
(1.94)
na
0.19
(3.32)
–0.04
(0.85)
0.01
(0.15)
0.10
(1.88)
–0.24
(2.20)
0.06
(1.03)
na
–0.30
(21.83)
–0.14
(8.09)
–0.41
(20.57)
–0.37
(20.71)
–0.42
(19.98)
–0.30
(20.57)
–0.29
(11.04)
–0.12
(2.55)
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
Variable
%Female-Headed Family
(1)
0.56
(20.39)
%18–24 years old
–0.12
(4.23)
%60–64 years old
–0.34
(2.84)
%65 and over
–0.27
(8.04)
Average Children per Family
3.70
(3.86)
%African American
–0.03
(3.61)
%Non African American Minority 0.09
(7.22)
%Same County in 1985
0.02
(1.76)
State Fixed Effects
Y
.893
R2
N
3023
(2)
(3)
(4)
0.54
(19.38)
–0.12
(3.72)
–0.60
(4.94)
–0.25
(7.79)
5.39
(7.10)
–0.03
(3.62)
0.08
(6.48)
0.05
(4.81)
N
.866
3023
0.58
(18.11)
0.02
(0.80)
0.34
(2.43)
0.02
(0.49)
7.33
(7.09)
–0.04
(3.43)
0.08
(5.70)
–0.003
(0.19)
Y
.853
3023
0.56
(18.65)
0.10
(3.62)
0.63
(4.36)
0.16
(4.78)
11.37
(10.53)
–0.02
(1.44)
0.11
(7.50)
–0.04
(2.33)
Y
.830
3023
485
(5)
(6)
Y
.468
3023
0.54
(13.09)
–0.05
(0.84)
–0.22
(0.99)
–0.20
(2.87)
8.20
(4.06)
–0.04
(3.01)
0.07
(4.06)
0.03
(1.29)
Y
.842
1204
a
The t-statistics use the White heteroskedasticity correction.
Suburban MSA counties are defined as all counties in a multiple-county MSA that do not
contain the largest city in the metropolitan area. A small MSA is defined as a total MSA population
of less than 350,000.
c
Suburban MSA counties are defined as all counties in a multiple-county MSA that do not
contain the largest city in the metropolitan area. A large MSA is defined as a total MSA population
of greater than 350,000.
d
A central city MSA county is defined as the county containing the largest city in a multiplecounty MSA.
e
Change in employment during the time span divided by the level of employment at the
beginning of the period, or the employment growth ratio.
b
population are negatively associated with the poverty rate.8 The greater magnitude of the nonmetropolitan population coefficient suggests that population
reduces poverty more in nonmetropolitan counties, but nonmetropolitan county
population never becomes large enough such that they have lower poverty rates
than MSA counties, ceteris paribus. One implication of the Metro/Nonmetro
variables is that the economic and demographic characteristics of nonmetropolitan counties do not entirely explain their higher average poverty rates.
county share of MSA employment. This variable was then interacted with the central-city county
dummy and included in the model shown in column 1. A value greater than 1.0 for this variable
implies that the minority population is concentrated in the central-city county without a corresponding concentration of jobs suggesting that job accessibility may be problematic for central-city
minorities and increasing poverty rates. However, the minority share/employment share coefficient
was negative and statistically significant (not shown) whereas the other results were unaffected.
8
Including MSA population for MSA counties and nonmetropolitan county population for
nonmetropolitan counties is the equivalent of interacting an MSA indicator variable with MSA
population and a nonmetropolitan county dummy with nonmetropolitan population (where the
nonmetropolitan county dummy reflects the omitted county type).
© Blackwell Publishers 2000.
486
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
Most of the Economic Development variables are statistically significant. As expected, the industrial structural change variable is positive and
significantly related to poverty rates, suggesting that structural change at the
regional level creates obstacles for the less-skilled that shifts many families
below the poverty threshold aside from its net impact on overall employment
growth. In sensitivity analysis (not shown), we also experimented with long-term
measures of structural change over the 1985–1990 and 1980–1990 periods, but
these variables were insignificant. The insignificance of the longer-term structural change measures suggests that the poverty impact of structural change is
not persistent, despite short-term adjustment problems.
Employment growth is not significantly related to poverty, which is consistent with the macroeconomic findings of Cutler and Katz (1991) and Blank and
Card (1993).9,10 Particularly at the county level, in-migration can mitigate
potentially poverty reducing effects of employment growth.11 In column 1, the
main employment/labor-force coefficient (for suburban counties and singlecounty MSAs) is negative, but only statistically significant at the 10 percent
level in a one-tail test. The nonmetropolitan interaction coefficient is negative
and statistically significant, supporting the hypothesis that job accessibility
plays a key role outside of MSAs in affecting poverty. The central-city county
interaction coefficient is negative but not statistically significant. Yet, it should
be noted that a large number of central-city jobs are filled by commuters which
may affect the interpretation of this coefficient.12
9
Employment growth between 1985–1990 and 1980–1990 was also added to the model to see
if long-term measures of employment growth reduced poverty rates. These measures tend to be
positively related to poverty rates, suggesting that long-run employment growth does not reduce
poverty rates within a locality, perhaps because in-migrants or commuters fill many of the new jobs.
However, this does not mean that national job growth does not reduce poverty rates. Just that
above-average employment growth in a region may have disappointedly small poverty-reducing
effects due to migration from other regions. To test the notion that employment-growth in areas with
higher concentrations of low-skilled workers should be particularly effective at reducing poverty, the
two-year employment-growth rate was interacted with the percent of the adult population without
a high school degree. Nonetheless, this variable is statistically insignificant.
10
These results are inconsistent with Bartik’s (1996) conclusions for overall MSA poverty
rates. However, Bartik used a different methodology, a different measure of poverty (125 percent of
the poverty line), and different geographic units of observation (MSAs versus metropolitan and
nonmetropolitan counties). More consistent with our findings is Madden (1996) who found an
insignificant relationship between ten-year changes in overall MSA employment growth and MSA
poverty rates.
11
In a separate regression county-level employment growth from 1980–1990 led to significant
in-migration to the county over the same period (t-statistic = 30.3).
12
In another model, we interacted the employment/labor force measure with all county-type
dummies. These results confirm that there is no statistically significant difference in effects between
the employment/labor force ratio in central-city counties and suburban counties. Likewise, in a
different model (not shown), another job-accessibility measure was created. Specifically, it is the ratio
of the number of blue-collar jobs to the number of low-skilled workers. Following Raphael (1998),
blue-collar workers was defined as those employed in manufacturing, retail trade, and services and
low-skilled workers was defined as the number of high school dropouts in the labor force (which we
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
487
Both male and female labor force participation variables are negative and
statistically significant. Employment growth and structural change are included
in the regression so the significant labor force participation coefficients suggest
additional factors underlie labor force participation and its effect on poverty.
Also, the female coefficient is over twice the magnitude of the male coefficient,
suggesting that reducing barriers for women to enter the labor market would
be particularly effective in reducing poverty. A one-standard deviation increase
in female labor force participation reduces poverty by about 2.1 percentage
points, which is approximately the amount that national family poverty rates
increased between 1989–1993, a period of sluggish economic growth.13
Many of the Industry Composition coefficients are statistically significant. In particular, counties with above-average employment shares in agriculture, trade, and services have greater poverty rates. Above-average employment
shares in goods-producing industries and FIRE are associated with lower
poverty rates (where the industry composition coefficients are measured relative
to public administration, the omitted sector). The negative goods-producing
sector coefficient suggests that the decline of the relatively high-paying manufacturing and mining sectors have hurt families on the margin. The FIRE share
coefficient is even more negatively associated to poverty than the goods share,
but Table 1 shows that FIRE is only about one-sixth the size of the goodsproducing sector, indicating that its absolute poverty-reducing impact is much
smaller. Note that the relative decline of agriculture over time may have reduced
poverty rates in nonmetropolitan areas (after a suitable adjustment period).
As expected, greater educational attainment reduces poverty. For example,
reducing the high school dropout share of the population by one percentage point,
while increasing the share of high school graduates by one percentage point
implies a 0.3 percentage point reduction in the poverty rate. A greater share of
college graduates has only about one-half of the poverty-reducing effect, illustrating that high school attainment is more effective in lifting poor families out
of poverty because college attainment more likely lifts families into the middle
and upper classes.14
approximate by taking the high-school dropout share of the adult population multiplied by the
number in the labor force). However, the blue-collar ratio coefficient was positive (and significant),
which runs counter to the spatial mismatch hypothesis where greater low-skilled job availability
relative to the low-skilled labor supply should result in better labor market outcomes (Raphael also
found weak results).
13
Caution should be exercised in interpreting this result because employment growth in
neighboring counties is a possible factor also. Nevertheless, strong employment growth in a
neighboring county may spill over into the county, producing a positive correlation. Thus, the
coefficient on county employment growth likely captures some of this effect (though it is statistically
insignificant).
14
The percent of the population that immigrated between 1985–1990 was also included but
its coefficient was negative and significant. Possible reasons for the result include the possibility
that (low-skilled) natives out-migrate in response to immigration. In a separate regression (not
shown), we found that net out-migration of previous county residents occurred approximately in
© Blackwell Publishers 2000.
488
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
Not surprisingly, more female-headed families is positively associated with
family poverty rates; a one-percentage point rise in the share of female-headed
families increases the poverty rate by almost 0.6 percent. More children per
family, a greater share of the population that are young adults, and an older
population (over 60) are all positively related to family poverty rates. After
controlling for state fixed effects (which account for current and residual effects
of discrimination at the state level), there is a negative association between
African-American population shares and poverty rates (consistent with Madden,
1996). This suggests that greater African-American poverty rates in subnational
areas are especially related to factors such as labor force participation, educational attainment, and female headship. Yet there is a positive association
between non–African-American minority share and the poverty rate.
The percent of the 1990 population that lived in the county in 1985 (or 100
minus the percent of residents that migrated to the county in the previous five
years) is a measure of geographic mobility. The positive percent in the samecounty coefficient indicates that counties with greater gross migration rates
have less poverty, although the coefficient is only significant at the 10 percent
level. Thus, there is some evidence that families moving their residence to where
there is greater job availability and better labor market matches can reduce
poverty, implying that policies that reduce household relocation costs and
provide better geographic labor market information merit more attention.15
The state fixed effects control for common effects within the state such as
state right-to-work laws, minimum-wage rates, welfare programs, amenity
levels, or cultural influences such as racial attitudes.16 Including the state fixed
proportion to the rate of immigration (confirming the state-level findings of Partridge and Rickman,
1999) which would offset immigration’s positive low-skilled supply-side effect. In addition, immigrants may be attracted to areas with lower poverty, further confounding interpretation of the
coefficient. For these reasons we omit immigration from our final model.
15
The official definition of poverty is based on a fixed income threshold that is unadjusted for
regional cost of living. Thus, we experimented with a couple of models that attempted to control for
cost-of-living effects. First, we included the log of the county’s average family income. However, one
concern is that greater average income mechanically implies a lower poverty rate due to the fixed
nominal poverty threshold. One characteristic of these results is that the magnitude of many of the
variable coefficients were reduced because they are ultimately a factor behind average family
income. For example, because labor force participation is a causal factor behind greater family
income, including income washes out some of participation effect. Nonetheless, most of the major
coefficient patterns were unchanged in this model. Alternatively, to control for cost of living we
included in another sensitivity model the county’s median home value and median monthly rent. As
expected the median rent was negatively related to the poverty rate (t = –5.04), and the median housing
value was positively related (t = 2.00) whereas the other coefficients were virtually unchanged.
16
One obvious public-policy issue is the degree to which basic government policies underlie
the estimated state fixed effects. To assess this issue we estimate some simple regressions that use
the state fixed effect as the dependent variable (not shown). This approach is fruitful in assessing
the effectiveness of state-wide public policies because demographic and economic factors are already
accounted for in the panel model. These results showed that once the demographic and labor market
effects are controlled for, none of the government policy variables (unemployment insurance, public
welfare expenditures, state minimum wage rates, percent unionized) significantly reduced family
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
489
effects in essence creates a first-difference model around the state means. Yet,
including the state fixed effects may in essence “over control” for some of the
factors. For example, high-growth sun belt states have experienced rapid growth
whereas their state dummy coefficient could capture the poverty-reducing
effects of economic growth. Thus, by omitting the state fixed effects, the model
in column 2 explores the possibility that the state fixed effects are capturing
some of the other variables’ effects.
The MSA county-type dummies are a little more negatively related to
poverty when the state fixed effects are omitted. This suggests that lower MSA
poverty rates relative to nonmetropolitan areas are somewhat related to unmeasured effects at the state level. Regarding the independent variables of
interest (e.g., ISC, labor force participation, educational attainment, female
headship), it is remarkable how robust their coefficients are to the omission of
the state fixed effects.17 Most of the other control-variable coefficients are also
approximately equivalent to the results in column 1. Overall, the results in
column 2 indicate the base-model results in column 1 are not an artifact of the
state fixed effects.
The model in column 3 is an alternative model that omits the employment/labor force and the labor force participation variables from the specification in column 1. One purpose of this is to gauge the “upper-bound” effect for the
other labor market and demographic variables because they may be causal
factors behind labor force participation (these results would be unaffected if
employment/labor force was included). Consistent with structural change increasing poverty indirectly through reducing labor force participation the structural change (ISC) effect is much greater in column 3. Yet, the employmentgrowth coefficient remains insignificant, supporting the hypothesis that
employment-growth has a relative small effect in influencing cross-region
variations in poverty (probably due to migrants or commuters).18 Similarly, the
poverty rates (even at the 20 percent level using a two-tail best). The insignificant results for
unemployment insurance benefits and welfare payments are consistent with Gottschalk’s (1993)
conclusion that U.S. income-transfer programs are ineffective at reducing income inequality. In a
sense this supports welfare reform proponents who claim that traditional welfare programs have
been ineffective at reducing poverty. The low R2 in this regression (about 0.20) suggests that there
is still a relatively large unexplained state-specific poverty effect, where cultural and social factors
(e.g., Partridge, Partridge, and Rickman, 1998), amenities and cost of living are possible explanations.
17
The nonmetropolitan employment/labor force ratio interaction coefficients are no longer
individually statistically significant. However, note that summing the main employment/labor force
coefficient and the nonmetropolitan interaction coefficient still produces a negative and statistically
significant total nonmetropolitan result (F = 17.2, p = .0001).
18
In other regressions we regressed the total labor force participation rate and the female
labor force participation rate on 1988–1990 employment growth, the ISC structural change variable,
and the state fixed effects. As expected, employment growth was positively and statistically
significant and structural change was negative and statistically significant, showing the importance
of labor demand conditions in determining participation. Yet, the results in column 3 indicated that
in terms of poverty rates, the interrelationship between structural change and labor force participation is more important than the employment growth–labor force participation interrelationship.
© Blackwell Publishers 2000.
490
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
magnitude of the education coefficients increased with the omission of the labor
force participation variables. This implies that one way in which education
reduces poverty is by inducing greater labor force participation. Also, we infer
that female-headed families experience poverty for reasons beyond just labor
force participation because the female-headed family coefficient was unaffected
by excluding female labor force participation. Finally, some of the Industry
Composition dummy coefficients were affected by the omission of the labor
force participation variable. A probable cause is a linkage between labor force
participation and industry structure.
The model shown in column 4 omits all of the Economic Development
and the Industry Composition variables from the model. Implicitly, this
specification assumes that poverty rates are only a function of the demographic
variables and geographic location. In a sense, this model may be viewed as a
reduced-form model in which demographic factors ultimately determine labor
market outcomes, perhaps due to human capital or tastes and preferences
toward work. Although some of the control variables are affected by this change
(e.g., the age coefficients are larger in magnitude), it is striking how little most
coefficients change between columns 3 and 4. In particular, note the stability of
the female-head coefficient across all of the specifications. This suggests that
female headship influences poverty rates through avenues unrelated to the
economic conditions of the locality. Thus, simple jobs-related economic development may be less effective at reducing poverty in counties with disproportionate
numbers of female-headed families and other interventions such as child care
or work transportation may be needed.
Column 5 of Table 2 shows the regression results of the most parsimonious
specification that only includes county-type dummies, MSA population, nonmetropolitan county population, and state fixed effects. This model considers
sensitivity of the county-type dummy coefficients to omission of all of the other
independent variables. MSA population is now insignificant while nonmetropolitan county population remains negative and statistically significant. The
MSA county-type dummies are still negative and statistically significant, suggesting that MSA counties have about 4 percent to 8 percent lower poverty rates
than the least populated nonmetropolitan counties when only adjusting for state
and population. There is a smaller poverty rate differential for more populated
nonmetropolitan counties, where the very largest nonmetropolitan counties
have approximately the same poverty rates as central-city counties. Overall, the
general pattern of the sensitivity analysis in columns 2 to 5 is that multicollinearity of the variables is not strongly affecting the results even with the close
theoretical interrelationships between many of the independent variables.
However, one concern is that high-poverty-rate counties are influenced
differently by the independent variables than low-poverty counties. If the
coefficients are completely unstable then pooling counties with high and low
poverty rates into one model is a misspecification. For example, the pool of
marriageable men may be smaller in high-poverty locales or high poverty rates
may be associated with higher divorce rates whereas both may influence the
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
491
female-head and participation effects.19 In column 6 we report the results of a
model that only includes the 1,204 counties with an above-average poverty rate
using the same specification as column 1.20 Although there are some modest
changes most of the coefficients are quite consistent with the full sample’s
results in column 1. For example, all of the MSA county-type coefficients are
negative and statistically significant, while the labor force participation, education, and female-headed family coefficients are virtually unchanged. The invariance of the coefficient associated with single female-headed households across
subsamples provides some evidence against the existence of simultaneity between poverty and the share of female-headed households. The total nonmetropolitan employment/labor force effect is significantly less than zero (F = 16.74,
p = 0.0001), even though the employment/labor force ratio main coefficient and
its interaction with the nonmetropolitan county dummy are individually
insignificant.
Results of Interactions
The results in Table 2 suggest that on average county employment growth
does not reduce the poverty rate. However, this may be misleading if these effects
do not apply equally across county types or demographic groups. Similar
statements can be made about the impact of structural change and education.
To explore these possibilities, Table 3 shows the results from several different
regressions that were performed by adding various interaction variables to the
specification shown in column 1. Unless otherwise stated, the findings for the
other control variables were not changed in this analysis.
The right-hand side of Panel a in Table 3 shows the influence of adding two
race–employment growth interactions to the model. The F-statistic indicates
that these two interaction variables are jointly statistically significant, suggesting separate employment effects for minorities. The African-American
employment-growth interaction is negative (the total African-American effect is –0.59 = –0.05 – 0.54), which (weakly) implies that employment growth
reduces poverty rates more in counties with greater population shares of
African-Americans. Hence, it may be possible that targeted economic development efforts focused on counties with significant population shares of AfricanAmericans may have more success. Conversely, the non–African-Americanminority employment-growth interaction variable is positive, suggesting that
counties with higher percentages of such population groups benefit less from
employment growth than other counties. (Note that adding this interaction
coefficient to the main employment-growth coefficient still suggests that the
19
Simultaneity between poverty and single female-headed households would positively bias
the female-head coefficient suggesting that caution should be exercised in interpreting those results.
20
Dividing the sample into high-poverty and low-poverty counties is akin to using a county
fixed effect for high poverty counties (that is interacted with slope coefficients). Thus, unaccounted
for characteristics that make them high-poverty counties are controlled for in this model.
© Blackwell Publishers 2000.
492
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
TABLE 3: Alternative Employment, Education, Race, and Structural
Change Interactionsa
Panel a
Employment Interactions
Employment Growth × Race
1988–1990 Employment Growth
× %African American
1988–1990 Employment Growth
× %Non African American Minority
1988–1990 Employment Growth
F-Interactions
(p-value)
Employment Growth × County Type
–0.05
(1.40)
0.09
(2.49)
–0.54
(0.86)
7.33
(p = .0007)
1988–1990 Employment Growth
× Single-City MSA
1988–1990 Employment Growth
× Small MSA Suburb
1988–1990 Employment Growth
× Large MSA Suburb
1988–1990 Employment Growth
× Central-City County
1988–1990 Employment Growth
F-Interactions
(p-value)
3.83
(2.49)
0.19
(0.11)
0.67
(0.70)
5.82
(1.68)
–0.52
(0.86)
1.05
(p = .378)
Panel b
Education Interactions
Education × Race
%High School/Some College
× %African American
%High School/Some College
× %Non African American Minority
%College Graduate
× %African American
%College Graduate
× %Non African American Minority
%High School/Some College
%College Graduate
F-Interactions
(p-value)
Education × County Type
–0.002
(2.43)
–0.006
(3.08)
–0.002
(4.15)
–0.003
(2.03)
–0.26
(15.00)
–0.12
(6.57)
17.38
(p = .0001)
%High School/Some College
× Single-City MSA
%High School/Some College
× Small MSA Suburb
%High School/Some College
× Large MSA Suburb
%High School/Some College
× Central-City County
%College Graduate
× Single-City MSA
%College Graduate
× Small MSA Suburb
%College Graduate
× Large MSA Suburb
%College Graduate
× Central-City County
%High School/Some College
%College Graduate
F-Interactions
(p-value)
© Blackwell Publishers 2000.
0.06
(1.16)
0.10
(3.42)
0.08
(4.16)
0.22
(7.19)
0.01
(0.41)
–0.04
(1.35)
0.08
(2.70)
0.08
(2.70)
–0.31
(16.99)
–0.15
(6.73)
5.98
(p = .0001)
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
493
Panel c
Structural Change Interactions
ISC × Education
1988–1990 ISC
× High School/Some College
1988–1990 ISC
× College Graduate
1988–1990 ISC
F-Interactions
(p-value)
ISC × County Type
–1.45
(3.40)
0.95
(1.45)
76.35
(3.31)
13.22
(p = .0001)
1988–1990 ISC
× Single-City MSA
1988–1990 ISC
× Small MSA Suburb
1988–1990 ISC
× Large MSA Suburb
1988–1990 ISC
× Central-City County
1988–1990 ISC
F-Interactions
(p-value)
–19.16
(1.99)
–4.02
(0.44)
–1.82
(0.24)
20.03
(1.09)
9.66
(3.02)
0.41
(p = .805)
ISC × Race
1988–1990 ISC
× %African American
1988–1990 ISC
× %Non African American Minority
1988–1990 ISC
F-Interactions
(p-value)
0.19
(0.91)
–0.69
(1.12)
9.17
(2.41)
2.33
(p = .097)
a
The coefficients reflect the estimates when the interaction variables are added to the model
shown in column 1 of Table 2. In parentheses are t-statistics and F-statistic p-values.
employment point estimate for the Non–African-American minority groups is
negative: –0.45 = –0.54 + 0.09).
The left-hand side of Panel a shows the results of adding employment
growth interacted with the county-type variables. The F-statistic indicates
that these interaction variables are jointly insignificant (where there may be
a separate single-county MSA effect). Thus, the effect of employment growth
does not appear to vary across metropolitan and nonmetropolitan areas or
across suburbs and central cities within MSAs, which suggests that economic
development policies should not be targeted to particular types of counties. For
example, an enterprise-zone policy aimed at increasing employment in central
cities or in nonmetropolitan areas without regard to other county characteristics
may not be more effective than other types of economic development policies
elsewhere.
Panel b shows how the impact of education varies across race and county
type. The left-hand side of Panel B shows that the race–education interaction
coefficients are negative whereas the t-statistics and F-statistics indicate that
the interactions are jointly significant. This suggests that increasing educational
attainment is especially effective in reducing poverty rates in counties with large
minority populations. The left-hand side of Panel b shows the education–
county–type interactions, where the F-statistic indicates that these interactions
© Blackwell Publishers 2000.
494
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
are jointly significant. The positive coefficients (with one exception) suggest that
greater education ameliorates nonmetropolitan poverty (the omitted group)
more than MSA poverty, particularly relative to central-city counties. Overall,
Panel b implies that poverty rates can be reduced more by increasing educational
attainment in nonmetropolitan areas and in counties with large populations of
minorities—but this policy would be less effective for MSA counties that are
predominantly white.
Panel c shows how education, racial composition, and county type interact
with industrial structural change (ISC). The upper right-hand side shows that
the high school–ISC interaction coefficient is negative and significant. That is,
counties with modestly higher educational attainment (fewer high school dropouts and more high school graduates) suffer smaller increases in poverty rates
as a result of structural change. The college graduate interaction is positive and
insignificant. This does not imply that college graduates are not negatively
affected by structural change, just that such families are rarely pushed below
the poverty threshold. The lower right-hand side shows the interaction between
racial composition and structural change. The joint statistical significance of
these interactions is modest (p = 0.097). As is the case for employment growth,
poverty in counties with larger population shares of African-Americans is more
affected by labor market structural change (although the results are relatively
weak). Finally, the F-statistic in the left-hand side of Panel c indicates that the
impact of structural change does not jointly vary either across MSAs or within
MSAs as a group (with the possible exception of single-county MSAs). This
finding is consistent with the findings for employment growth by county-type.
4.
SUMMARY AND CONCLUSION
Using data for all counties in the contiguous 48 states, we ascertain the
reasons for differences in area poverty across the United States. Specifically, the
roles of both person-specific and place-specific characteristics in influencing area
poverty were assessed.
Higher area poverty was found to be associated with single-female family
headship and lower educational attainment levels. After controlling for these
and other factors, poverty was found to be higher for counties with larger
population proportions of non–African-American minorities, but not for counties
with larger population proportions of African-Americans. Regarding area economic performance, recent employment growth on average did not reduce the
poverty rate; however, employment growth did relatively (and absolutely) reduce
poverty in counties with greater population shares of African-Americans. However, the generalizability of the county employment-growth results to the
national level is complicated by the fact that in-migration responses to
employment growth are more likely at the county than the national level,
making it less likely that employment growth can reduce poverty at the county
level. Structural change increased poverty in the short run with its effects
disappearing within five years. Nevertheless, structural change hurt counties
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
495
with greater population shares of African-Americans and adults without high
school degrees. Migration is an adjustment that can occur to mitigate the effects
of structural change so it is likely that the effect of structural change on poverty
is even larger at the national level.
Greater goods-producing employment is also associated with lower poverty.
Higher labor force participation is associated with lower poverty rates, particularly among females. The results suggest that skills mismatches were most
severe in nonmetropolitan areas, likely associated with geographic isolation of
their residents. On the other hand, educational attainment reduced poverty
more in nonmetropolitan counties than in metropolitan area counties.
Regarding policy conclusions, the results point to increasing education as
key to reducing poverty particularly for counties with minorities and nonmetropolitan counties. Along with educational attainment, central-city poverty appears more related to female-family headship and the number of children in the
family than to job-skills mismatches. Nevertheless, targeted economic development and assistance for displaced workers are suggested by the results to reduce
poverty in counties with African-American populations. Similarly, policies that
increase labor force participation among females may be warranted. More
research is needed into whether potential job-skills mismatch effects in nonmetropolitan areas are related to a lack of labor market information, less
transferable job skills, or rational residence choices by nonmetropolitan residents involving amenity considerations.
REFERENCES
Allen, Steven G. and Richard B. Freeman. 1995. “Quantitative Flexibility in the U.S. Labor Market,”
(unpublished manuscript), Department of Economics, North Carolina State University.
Bartik, Timothy J. 1993. “Who Benefits from Local Job Growth: Migrants or the Original Residents,”
Regional Studies, 27, 297–311.
———. 1996. “The Effects of Metropolitan Job Growth on the Size Distribution of Family Income,”
Journal of Urban Economics, 40, 150–178.
Blank, Rebecca M. and David Card. 1993. “Poverty, Income Distribution, and Growth: Are they Still
Connected?” Brookings Papers on Economic Activity, 2, 285–325.
Blank, Rebecca M. and Maria Hanratty. 1992. “Down and Out in North America: Recent Trends in
Poverty Rates in the United States and Canada,” Quarterly Journal of Economics, 107, 232–254.
Bluestone, Barry. 1990. “Comment,” in G. Burtless (ed.), A Future of Lousy Jobs? Washington, DC:
Brookings Institution, pp. 68–76.
Borjas, George J., Richard Freeman, and Lawrence Katz. 1996. “Searching for the Effect of
Immigration on the Labor Market,” American Economic Review, 86, 246–251.
Bradbury, Katherine L., Yolanda K. Kodrzycki, and Christopher J. Mayer. 1996. “Spatial and Labor
Market Contributions to Earnings Inequality: An Overview,” New England Economic Review,
May/June 1–10.
Brown, David and Mildred Warner. 1991. “Persistent Low-Income Nonmetropolitan Areas in the
United States: Some Conceptual Challenges for Development Policy,” Policy Studies Journal, 19,
22–41.
Cancian, Maria, Sheldon Danziger, and Peter Gottschalk. 1993. “Working Wives and Family Income
Inequality among Married Couples,” in S. Danziger and P. Gottschalk (eds.), Uneven Tides: Rising
Inequality in America. New York: Russell Sage Foundation, pp. 195–222.
Carrington, William J. and Asad Zaman. 1994. “Interindustry Variation in the Costs of Job Displacement,” Journal of Labor Economics, 12, 243–275.
© Blackwell Publishers 2000.
496
JOURNAL OF REGIONAL SCIENCE, VOL. 40, NO. 3, 2000
Corcoran, Mary, Roger Gordon, Deborah Laren, and Gary Solon. 1992. “The Association between
Men’s Economic Status and Their Family and Community Origins,” Journal of Human Resources,
27, 575–601.
Cutler, David M. and Edward L. Glaeser. 1997. “Are Ghettos Good or Bad?” Quarterly Journal of
Economics, 112, 827–872.
Cutler, David M. and Lawrence F. Katz. 1991. “Macroeconomic Performance and the Disadvantaged,”
Brookings Papers on Economic Activity, 2, 1–74.
Freeman, Richard. 1991. “Employment and Earnings of Disadvantaged Young Men in a Labor
Shortage Economy,” in C. Jencks and P. Peterson (eds.), The Urban Underclass. Washington, DC:
The Brookings Institution, pp. 103–121.
———. 1993. “How Much Has De-unionization Contributed to the Rise in Male Earnings Inequality?”
in S. Danziger and P. Gottschalk (eds.), Uneven Tides: Rising Inequality in America. New York:
Russell Sage Foundation.
Frey, William H. 1995. “Immigration and Internal Migration ‘Flight’ from US Metropolitan Areas:
Toward a New Demographic Balkanisation,” Urban Studies, 32, 733–757.
Gottschalk, Peter. 1993. “Changes in Inequality of Family Incomes in Seven Industrial Countries,”
American Economic Review, 83, 136–142.
Holzer, Harry J. 1991. “The Spatial Mismatch Hypothesis: What Has the Evidence Shown?” Urban
Studies, 28, 105–122.
Holzer, Harry J., Keith Ihlanfeldt, and David Sjoquist. 1994. “Work, Search and Travel among White
and Black Youth,” Journal of Urban Economics, 35, 320–345.
Ihlanfeldt, Keith R. and Madely Young. 1996. “The Spatial Distribution of Black Employment
between the Central-City and the Suburbs,” Economic Inquiry, 34, 693–707.
Juhn, Chinhui. 1999. “Wage Inequality and Demand for Skill: Evidence from Five Decades,”
Industrial and Labor Relations Review, 52, 424–443.
Juhn, Chinhui, Kevin Murphy, and Brooks Pierce. 1993. “Wage Inequality and the Rise in the Returns
to Skill,” Journal of Political Economy, 101, 410–442.
Kain, John F. 1992. “The Spatial Mismatch Hypothesis: Three Decades Later,” Housing Policy Debate,
3, 371–460.
Kirschenman, Joleen and Katherine Neckerman. 1991. “We’d Love to Hire Them But . . .,” in C.
Jencks and P. Peterson (eds.), The Urban Underclass, Washington, DC: The Brookings Institution,
pp. 203–232.
Larson, Donald K. 1989. “Transitions of Poverty Amidst Employment Growth: Two Nonmetropolitan
Case Studies,” Growth and Change, 20, 19–34.
Madden, Janice Fanning. 1996. “Changes in the Distribution of Poverty Across and Within the US
Metropolitan Areas, 1979–89,” Urban Studies, 33, 1581–1600.
Mead, Lawrence. 1992. The New Politics of Poverty. New York: Basic Books.
O’Regan, Katherine M. and John M. Quigley. 1991. “Labor Market Access and Labor Market
Outcomes for Urban Youth,” Regional Science and Urban Economics, 21, 277–293.
Partridge, Jamie S., Mark D. Partridge, and Dan S. Rickman. 1998. “State Patterns in Family Income
Inequality,” Contemporary Economic Policy, 15, 277–294.
Partridge, Mark D. and Dan S. Rickman. 1998. “Regional Differences in Chronic Long-Term
Unemployment,” Quarterly Review of Economics and Finance, 38, 193–215.
———. 1999. “A Note on the Benefits to Current Residents of State Employment Growth: Is There
an Industry Mix Effect on Migration?” Journal of Regional Science, 39, 167–181.
Raphael, Steven. 1998. “The Spatial Mismatch Hypothesis and Black Youth Joblessness: Evidence
from the San Francisco Bay Area,” Journal of Urban Economics, 43, 79–111.
Rural Sociological Society Task Force on Persistent Rural Poverty. 1993. Persistent Poverty in Rural
America. Boulder, CO: Westview Press.
Sawicki, David and Mitch Moody. 1997. “The Effects of Intermetropolitan Migration on the Labor
Force Participation of Disadvantaged Black Men in Atlanta,” Economic Development Quarterly,
11, 45–66.
Slesnick, Daniel. 1993. “Gaining Ground: Poverty in the Postwar United States,” Journal of Political
Economy, 107, 1–38.
© Blackwell Publishers 2000.
LEVERNIER, PARTRIDGE, & RICKMAN: VARIATIONS IN U.S. POVERTY
497
Smith, James P. and Finis R. Welch. 1986. Closing the Gap: Forty Years of Economic Progress for
Blacks. Santa Monica: Rand Corporation.
Topel, Robert H. 1993. “What Have We Learned from Empirical Studies of Unemployment and
Turnover?” American Economic Review, 83, 110–115.
———. 1994. “Regional Labor Markets and the Determination of Wage Inequality,” American
Economic Review, 84, 17–22.
Triest, Robert K. 1997. “Regional Differences in Family Poverty,” New England Economic Review,
Jan/Feb, 3–17.
Turner, Margery. 1992. “Discrimination in Urban Housing Markets: Lessons from Fair Housing
Audits,” in Housing Policy Debate, Washington DC: Fannie Mae.
Viscusi, W. Kip. 1986. “Market Incentives for Criminal Behavior,” in R. Freeman and H. Holzer (eds.),
The Black Youth Employment Crisis, Chicago: University of Chicago Press, pp. 301–351.
Wilson, William. 1987. The Truly Disadvantaged. Chicago: University of Chicago Press.
© Blackwell Publishers 2000.