The influence of geographic mobility and gender on Indiana wages

Abstract

The study examines wage differentials of individuals experiencing unemployment episodes using a multivariate analysis of wage and unemployment records. The focus is the wage effect of small distance geographic mobility (micro-mobility) during job seeking. The results identify limitations on geographic micro-mobility as a source of wage disparity in the re-employment market. The study isolates persistent gender differences in geographic mobility rates and hypothesizes this as a potential source of gender-wage disparity in both the re-employment and greater labor market.

The data and methods are unique. The dataset is Indiana administrative wage records over a ten-year period for individuals that experience unemployment episodes. The study assesses unemployment as an exogenous shock on wages to determine underlying influences in the labor market. The novel approach is unconstrained by limitations associated with aggregated or proxy data.

Keywords

Unemployment geographic mobility wages gender-wage bias

1. Introduction

This paper examines the influence of geographic mobility on wages in the re-employment market. Unemployment negatively affects incomes or wages, which the paper uses interchangeably. In most instances, the self-interest of the individual is to minimize the length of the unemployment episode and re-establish with the labor market. Unfortunately, the skills possessed by an individual may not match the need of the local labor market. To maximize advantage and successfully navigate the re-employment market, individuals often seek employment in a locale that is short of their skills. If such geographic mobility is not possible, the individual may wait longer to re-establish with the labor market, which can negatively influences the resulting wage. Alternatively, the individual might take a local position that is less suited to their skills and accept diminished wages.

The body of regional unemployment and wage research is extensive. The reduction of wages associated with unemployment negatively affects local economies. The long-term economic distress of rural America and increased poverty is partially due to concentrated unemployment [9, 12, 28, 40]. The rural poor are limited in income opportunities due to isolation and limited opportunities [12, 48].

The potentially offset of rural poverty is wage convergence. If regional wage disparities exist, companies should arbitrage the wage opportunity. If wages are low in rural areas, firms should increasingly locate to rural areas to take advantage of lower wages. The increased rural labor demand should lift wages and diminish the wage disparity between rural and urban communities.

In seminal work on income distribution, Kuznets [27] uses country data to examine the historical relationship between the country’s level of development (GDP per capita) and inequalities in income distribution. In early stages of country development, the inequality grows. In later stages of development, inequality decreases.

Williamson [49] examines regional income disparities in the United States. The analysis compares statewide GDP per capita data with regional income disparities internal to each state. The results are similar to those Kuznets; advancing levels of GDP per capita reduce regional income disparities within the state, while disparity increases at low levels of GDP per capita. More recently, papers supporting these results examine European regional disparities [3, 13].

Research also counters the wage convergence theory. Myrdal [34] suggests that growth is a spatially cumulative process driven by factors such as economies of scale, increasing returns to scale, and externalities associated with human capital. For this reason, regional income disparities increase over time. Economic growth will increasingly concentrate in the few areas that contain these elements. Income growth will be spatially limited to specific regions leading to income divergence. Several lines of research including development studies [14, 20], geographical studies (Krugman, 1991: Thisse, 2000), and urban studies [38, 19] support the theory.

The potential sources of wage disparity are numerous. One the most cited factors altering wages is education. The link between educational achievement and wages in academic literature is well established and defined by imperfect substitution between work and the availability of labor skills. The labor market maintains a positive wage bias favoring skills and human capital investment. Card [7] surveys a large and consistent body of literature establishing the connection.

Tilak [42] suggests that wage inequality decreases with increasing levels of education. Initially, enhanced levels of skill afford higher wages. As more people pursue higher wages through educational attainment, however, the supply of higher skilled workers increases and exerts downward price pressure on high-skilled jobs. At the same time, fewer people pursuing low-skilled jobs pushes wages higher due to the lack of supply. From this perspective, education provides wage convergence by altering the supply of job skill. Goldin and Margo [16] support this view using empirically results from pre-1970 data on wages and job skill attainment. Teulings [43, 44] also concludes that initially highly educated people in complex jobs demand higher salaries. In the long-term, people seeking higher salaries obtain more education. The supply of highly educated people for the fewer complex jobs puts downward pressure on the wages of the complex jobs. High-educated workers fall into less complex jobs and lower wages.

Others disagree with the notion of long-term wage convergence due to educational attainment. Acemoglu [1] argues that diminishing returns to education are unlikely. The increase in human capital from education induces greater levels of investment in technology, which promotes innovation. Innovation is a positive externality derived from education. The accumulation of innovation offsets and surpasses any potential diminishing returns. This argument is consistent with numerous studies using post-1970 wages. The studies indicate increasing wage inequality in the labor market due to skill requirement differentiation [5, 6, 22, 23, 26, 31].

Thurow [45] and Revenga [37] suggest that the explanation for the observed increase in income inequality after 1970 is a function of job mix. Low-wage jobs replaced high-wage jobs in traditional declining industries such as manufacturing. The change in job mix suppresses wages and maintains disparity. This view of job creation and wage growth is not universally accepted [10, 11]. While unsettled, the common view is that the restructuring of the employment mix has had extended effects on rural areas and caused wage stagnation. Once a source of civic pride and steady high-wage employment, industries such as manufacturing have witnessed decades of employment declines [39, 41].

The literature on gender-wage disparity is robust. While much of the existing literature focuses on symptom assessment, less focuses on source identification. One area of source research on gender-wage disparity examines gender and job transfer [29, 24, 15].

Geographic mobility and spatial wage studies examine both physical re-location and allotments for commuting time. Mills [32] and Muth [33]address the issue of firm migration from city centers and resulting wage impacts. Rees and Schultz [36] investigate the wage differences of city center employment versus their suburban counterpart.

Women often carry increased family responsibilities that affect their ability to relocate or commute extensively for employment [8, 17, 18, 35]. The concern is that the limitation of geographic mobility becomes a gender trap for lower wages. Prior research supporting this thesis focuses on city center versus suburban employment [18, 30, 35, 21].

The study adds to existing research by applying a novel approach to wage analysis. It uses administrative wage data and unemployment shocks to assess the link between geographic mobility and wages. The study supports prior results when assessing short distance mobility, described as micro-mobility. Micro-mobility is the movement between Indiana counties measured in kilometers. The study also broadens the thesis by expanding the argument beyond the city center versus suburban framework that currently dominates the literature. The gender divergence of geographic mobility partially explains the gender disparity in wages. Numerous factors, including family considerations, limit mobility and detrimentally influence wages. As females often carry a disproportionate weight of family considerations, they also carry its wage influence.

2. Methodology

The study uses a longitudinal dataset of de-identified individual Indiana unemployment claimant data linked to quarterly wage records.1 The data ascertain the individual’s wages before and after each episode of unemployment, and include observations from the 1 ${}^{\text{st}}$ quarter of 2006 through the 4 ${}^{\text{th}}$ quarter of 2015. Portions of the dataset came from individuals when applying for unemployment benefits. Each incidence of unemployment is a unique claims episode, and individuals may appear multiple times during the collection period. In an effort to protect record unanimity, information about repeat episodes of unemployment is not available.2

Each claims episode matches to the individual’s wages pre- and post-episode. The study examines the wage impact of workforce re-integration, so complete wage records are integral. To ensure complete records, the study excludes the following:

•
Records of individuals without wage matches on either side of UI claims.
•
Records of individuals coming into Indiana for work or leaving the state for employment.
•
The self-employed because wage records only exist for those employed by companies covered by unemployment insurance.
•
Records of individuals living in Indiana, but employed outside the state.
•
Records of individuals below the age of 18 or older than 80 years of age.3

The final dataset consists of 1,515,722 observations. Each observation represents a unique claim episode between 2006 and 2015 with matching wage data.4 The distribution of yearly observations is in Fig. 1.

Figure 1.
Observations.

A wage variable measures the difference between an individual’s wages entering the claims system against the wages once reintegrated into the labor market. The wage difference is Eq. (1):

$\displaystyle\textit{Wage Difference}=(\textit{Avg. Wages Post-UI Claim}){}-(% \textit{Avg. Wages Pre-UI Claim})$ (1)

A positive wage difference reflects higher wages after the episode of unemployment, while a negative wage difference reflects decreased wages after unemployment. The wage difference is the dependent variable in the regression analysis.

Independent variables assess influence on the wage discount. The study focusses on county-to-county mobility within Indiana, or micro-mobility.5 To assess the impact, a variable measures the movement of individuals during episodes of unemployment. The county of residence of the unemployed individual prior to the episode is self-reported when applicants apply for unemployment benefits. Indiana wage records contain the location of the employer both pre- and post-unemployment episode. A constructed 92 $\times$ 92 diagonal matrix compares geographic midpoint distances between all Indiana counties. For each unemployment episode, the distance (in kilometers) from individual’s residence to the employer prior to unemployment episode is determined from the matrix. If it is the same county, the distance is zero. The distance from residence to the employer post-unemployment episode is also determined the same methodology. The kilometers between the pre- and post-unemployment episode is a proxy for the increased willingness of an individual to commute or move for employment. The difference is Eq. (2):

$\displaystyle\textit{Kilometers to New Job}=(\textit{Kilometers Post UI % Episode}){}-(\textit{Kilometers Pre UI Episode})$ (2)

A positive difference reflects an individual willing to travel further for a new position (increased mobility). A negative difference reflects a new position closer to prior residence. Zero indicates no difference in distances.6

The earning power of an individual prior to an episode of unemployment is a likely influence on the wage difference. A wage before unemployment variable (Wage Before Unemployment)7 is added. The total time collecting UI benefits is the total number of weeks of UI benefits received. The variable (Total Weeks Unemployment Claimed) includes all benefits, including state and federal.8

Explanatory variables include claimant characteristics such as gender, age, and race/ethnicity. The variables are self-reported when applicants apply for unemployment benefits. A binary gender variable (Gender Male) is included and is affirmative for male. A binary variable (White) results from individuals self-reporting as white or Caucasian. The observations segregate into the following binary age categories.

•
Ages 18 to 30
•
Ages 31 to 50
•
Ages 51 to 65
•
Ages 66 to 80

When individuals experience unemployment, they often attempt to re-attach in positions similar to previous employment to maximize knowledge and experience. In some instances, such as temporary layoffs, an unemployed person may re-attach with their previous employer. A binary variable (Same Employer) is affirmative if the individual re-attaches with the same employer prior to unemployment. If the person cannot find employment at the same firm, they may attempt to leverage their knowledge in the same field. A binary variable (Same Industry (NAICS)) is created if the individual re-attaches with a new employer in the same industry as prior employment, as denoted by a common two-digit NAICS code. The Indiana yearly Gross Domestic Product (GDP) variable controls for the economic environment during unemployment. Yearly binary variables account for yearly variation.

The highest achieved level of educational attainment by the unemployed is included as binary variables. Education is self-reported when applicants apply for unemployment benefits. Figure 2 shows the education level of applicants. The data skew towards lower education levels as the probability of unemployment increases with decreased education.

Figure 2.
Education.

Occupational variables are included. The Standardized Occupational Code (SOC) is self-reported when applicants apply for unemployment benefits. It represents the position of the applicant at separation. Unfortunately, the SOC for new positions are not available. The model uses occupational binary variables (SOC).

Regional differences are likely to affect the unemployment experience. The Indiana Department of Workforce Development (IDWD) divides the state into 12 Economic Growth Regions (EGRs) by counties clustered by economic commonality. Binary variables are created (Home EGR) for the individual’s home location.

Table 1 provides a list of summary statistics. The average unemployment wage discount is $-$ $431.06 per quarter. During the period of 2006 to 2015, an individual going through a period of unemployment in Indiana received on average $431.06 less per quarter (or $1,724.24 annually) in the new position compared to their prior employment.

Table 1
Summary statistics

Table 2
Results

The analysis uses an ordinary least squares (OLS) model. The dependent variable is the difference in wages (Wage Difference). The econometric models in this study are implemented using Stata version 16.0 (64 bit) on an Intel Core I5 CPU laptop running at 1.60 GHz with 16 GB of memory, running Windows 10 64-bit Enterprise operating system.

$\displaystyle\textit{Model}\quad Y=\beta_{0}+\sum_{1}^{i}\beta_{i}X_{i}+\varepsilon$ (3)

An examination of prior research did not yield a similar approach. The approach appears novel in three aspects. First, this approach focuses on micro-mobility over short distances. Distances are measured over Indiana counties and provide results at the kilometer level. Secondly, previous wage studies are often limited because they assess diverse individuals over time while controlling for inherent differences. They attempt to compare dissimilar entities over long periods. This study compares the labor market reaction to a specific person when subjected to an episode of unemployment. The labor market is responding to an exogenous shock applied to an individual over a short duration. The shock is common, but the reaction of the labor market and wage discount is not. Biases that alter the wage discount reveal underlying labor market preferences. This approach is free from the complexities of previous studies when comparing dissimilar individuals. Finally, the study examines the relationship between micro-mobility and wage disparity by gender.
3. Results

The results indicate a positive relationship between post-unemployment wages and mobility. Micro-mobility is positive and highly significant (Kilometers to New Job). Micro-level mobility, or movement over shorter distances within Indiana, positively affects wages and is consistent with prior studies. The increased mobility is presumably a function of either increasing commuting times or relocation for employment. A 1-kilometer increase in mobility is associated with a gain of $0.34 in quarterly wages.

High wages prior to unemployment and a longer duration of unemployed increases the potential of a worsening wage discount. The coefficients of the Wage Before Unemployment and the Total Weeks Unemployment Claimed variables are negative and significant. The unemployment discount is more severe the higher the wages were prior to unemployment. Likewise, the length of time unemployed is detrimental to finding comparable wages post-unemployment.

The gender variable is a positive and significant influence for post-unemployment wages. The results indicate that males re-employ with less of a wage discount. The magnitude is $793.10 in quarterly wages. The results for racial binaries (White) are also significant, but the impact of this wage differential is comparatively modest.

The working age brackets are significant and negative, with the 31 to 50 bracket removed for collinearity. Compared to prime working years, those in the younger bracket have a slightly higher unemployment wage discount. The wage discount is progressively more severe for those in the older age brackets. Figure 3 illustrates the age bias on unemployment discounts.

Figure 3.

Age bracket influence on unemployment discount.

Finding employment with the same employer or in the same industry is beneficial in maintaining a comparable wage post-unemployment. The coefficients for the Same Employer and the Same Industry (NAICS) binary variables are both positive and significant. By returning to the same place of employment and maintaining the same line of work, an individual can leverage their skills and knowledge to achieve a better wage when re-attaching to the workforce. The coefficient for Indiana’s GDP is significant and positive. The economic environment influences the resulting wage in the re-employment market.

Education is a large influence on the ability of an individual to successfully navigate unemployment and achieve higher wages after workforce re-attachment. The variable for less than a High School education is omitted and remaining education variable coefficients are significant and positive. Increasing levels of education produce higher wages in the re-employment market, as represented by Fig. 4.

Figure 4.

Education influence of unemployment discount.

The occupation held before unemployment is an important determinant on subsequent post-unemployment wages. Training and experience are influential factors on the unemployment experience, as indicated by most of the SOC coefficients being significant. The variables differ in magnitude, as the level of training vary by field. Farming and Military occupations were omitted as they are less likely to be influenced by market conditions and to prevent collinearity. The EGR control variables are also largely significant, with EGR 12 omitted for collinearity. Analysis of two and five-year cohorts ensures robust results. These results are not included but available.

4. Discussion

The results indicate that geographic micro-mobility over short distances (kilometers) is a source of wage disparity in the re-employment market. Mobility rates vary by demographic category, which can be a source of consistent wage inequality. The study examines the differing rates of mobility by gender. The rates of micro-mobility between genders in the dataset differ by an average of approximately 22% (Fig. 5). While the magnitude of the mobility is incrementally small, the aggregated effect compounded over a lifetime can partially explain the consistent wage disparity found in prior research.

Figure 5.

Micro-mobility: Kilometers to new job.

Figure 6.

Micro-mobility: Residence to old job (in kilometers).

The study further assessed 2 years increments of data. This is done both as a robustness check and to determine if the results vary over time. Within the 2-year segments, males typically express greater rates of geographic micro-mobility before unemployment. Figure 6 indicates the distance (in kilometers) to the place of employment prior to the episode of unemployment. The economic crisis of 2008/2009 affected workers increasingly close to their residence. The gap between micro-mobility rates by gender remains relatively constant.

Figure 7.

Micro-mobility: Residence to new job (in kilometers).

Figure 7 indicates geographic micro-mobility rates of the unemployed after re-attachment to the labor market. The figure indicates the distance (in kilometers) to the place of employment post-unemployment. Again, a persistent gender gap in mobility rates exists. On close examination, a slight narrowing of the mobility gender difference during the middle three periods compared to the first and last periods. Behavior becomes increasingly similar during economic distress. Before and after economic distress, the gender gap widens.

5. Conclusion

Geographic micro-mobility is an important factor in determining post-unemploy- ment wages when measured at the kilometer level. Limiting the job search a specific location or approximate vicinity is detrimental to post-unemployment wages.

Mobility may partially explain lingering wage gaps for segments of society such as gender. Household commitments and other considerations may limit female geographic mobility, which hinders wage attainment and allows for persistent wage gaps. The effect is evidenced in the re-employment market but may extend into the overall labor market. An extension of the analysis can assess the effects of geographic micro-mobility on other demographic factors.

Footnotes

A random identifier is applied and the researcher never has access to identified data, insuring record anonymity. The results are provided in aggregate.

The data was requested, but was prohibited due to privacy concerns. The data would be helpful as many of the unemployment claimants may appear multiple times in the dataset.

This represents 742 observations and 416 observations respectively. As the study determines movement through the unemployment process, those outside these bounds are considered atypical and removed to prevent bias.

Indiana wage data are imperfect and several accommodations are required. Wage data are presented in quarterly aggregates. The data does not delineate part-time labor, full-time labor, weeks worked, or the number of hours worked within the quarter. Given it unlikely that people separate or re-integrate into the labor market precisely on the first or last days of the respective quarters, a potential for measurement error exists. To adjust for the potential error, the study uses an average of the second and third quarters directly preceding separation and the second and third quarters directly following workforce re-integration. The quarterly wage data directly preceding or post-claims are discarded, due to the high likelihood that the quarterly wage data represents only a partial quarter and underrepresents actual earnings.

Location by zip or other method of granularity is not permitted due to privacy concerns.

Data by zip code is preferred, but not allowed due to anonymity concerns. It is acknowledged that some people may move resulting from job displacement; therefore it is not assumed that this variable only shows an increase in commuting time. It is also acknowledged that there are discrepancies in recorded counties for residence of both the individuals and companies (particularly if the company operates physical locations in multiple counties).

This represents the average of the second and third quarters preceding unemployment. As the quarter directly preceding unemployment is likely a partial quarter, it is discarded.

Some federal benefit programs, particularly those covered by Trade Adjustment Programs can extent for a very long period during re-training. However, these types of extended programs are infrequent.

References

Acemoglu

. Technical Change, Inequality, and the Labor Market. Journal of Economic Literature. Mar 2002; 7-72.

Baltagi

. Econometric Analysis of Panel Data. Chichester: Wiley; 1995.

Barrios

Strobl

. The Dynamics of Regional Inequalities. Economic Papers of the European Commission, Brussels. 2005; 229.

Bureau of Economic Analysis. Table Data: Real GDP by state (millions of chained 2009 dollars). www.bea.gov. Indiana GDP data extracted March 16, 2017.

Blackburn

. What can explain the increase in earnings inequality among males? Industrial Relations. 1990; (29): 441-456.

Bound

Johnson

. Changes in the structure of wages in the 1980’s: An evaluation of alternative explanations. American Economic Review. Jun 1992; 82: 371-392.

Card

. The Casual Effect of Education on Earnings. Chapter 30 in O. Ashenfelter and D. Card (Eds.) Handbook or Labor Economics, Vol 3. Amsterdam: Elsevier; 1999.

Carlson

. The availability of women workers: The effects on company location in the suburbs. New York: Garland; 1997.

Deavers

. Rural america: Lagging growth and high poverty…do we care? Choices. 1989; (2): 4-7.

10.

Dickens

Lang

. A test of dual labor market theory. American Economic Review. Sep 1985; 75: 792-805.

11.

Dickens

Lang

. Where Have all the Good Jobs Gone? Deindustrialization and the Labor Market Segmentation. In Unemployment and the Structure of Labor Markets. Edited by Kevin Lang and Jonathon Leonard. New York. Blackwell; 1987.

12.

Duncan

Tickamyer

. Poverty research and the policy for rural america. Am Sociol. 1988; 19(3): 243-259.

13.

Ezcurra

Pascual

. Regional polarisation and national development in the european union. Urban Studies. 2007; 44(1): 99-122.

14.

Fleming

. External economics and the doctrine of balanced growth. Economic Journal. 1955; 65: 241-256.

15.

Fuller

. Job mobility and wage trajectories for men and women in the United States. American Sociological Review. Feb 2008; 73(1): 158-183.

16.

Goldin

Margo

. The great compression: The wage structure in the United States at mid-century. Quarterly Journal of Economics. Feb 1992; 107: 1-34.

17.

Hanson

Pratt

. Spatial dimensions of the gender division of labor in a local labor market. Urban Geography. 1988; 9: 367-78.

18.

Hanson

Pratt

. Job search and the occupational segregation of women. Annals of the Association of American Geographers. 1991; 81: 229-53.

19.

Henderson

. Urban Development. Oxford University Press. 1958.

20.

Hirschman

. The Strategy of Economic Development. New Haven, Conn.: Yale University Press. 1958.

21.

Ihlanfeldt

. Intra urban wage gradients: Evidence by race, gender, occupational class, and sector. Journal of Urban Economics. 1992; 32: 70-91.

22.

Karoly

. Changes in the Distribution of Individual Earnings in the United States: 1967–1986. Review of Economics and Statistics. Feb 1992; 74: 107-115.

23.

Katz

Murphy

. Changes in relative wages, 1963–1987: Supply and demand factors. Quarterly Journal of Economics. 1992; 107: 35-78.

24.

Keith

McWilliams

. The returns to mobility and job search by gender. ILR Review. Apr 1999; 52(3): 460-477.

25.

Killian

Bloomquist

Pendleton

McGranahan

. Symposium on Rural Labor Markets Research Issues, Staff Report No. AGES860721. Washington D.C.: U.S. Dept. Agric. Econ. Res. Serv. 1986.

26.

Kosters

. Workers and Their Wages: Changing Pattern in the United States. Washington D.C.: American Enterprise Institute. 1991.

27.

Kuznets

. Economic growth and income inequality. American Economic Review. 1955; 45: 1-28.

28.

Levy

. Dollars and Dreams: The Changing American Income Distribution. New York: Russell Sage. 1987.

29.

Loprest

. Gender differences in wage growth and job mobility. The American Economic Review. May 1992; 82(2): 526-532.

30.

Madden

. Why women work closer to home. Urban Studies. 1981; 18: 181-94.

31.

Mehta

. Quality of education, productivity changes, and income distribution. Journal of Labor Economics. Apr 2000; 18(2): 252-281.

32.

Mills

. Studies in the structure of the urban economy. Baltimore: John Hopkins University Press. 1972.

33.

Muth

. Cities and housing: The spatial patterns of urban residential land use. Chicago: University of Chicago Press. 1969.

34.

Myrdal

. Economic Theory and Underdeveloped Regions. Hutchinson. 1957.

35.

Nelson

. Labor demand, labor supply and the suburbanization of low-wage office work. In Publication, work territory: The geographical anatomy of industrial capitalism. ed. A.J. Scott and M. Storper, 149-69. Boston: Allen Unwin. 1986.

36.

Rees

Schultz

. Workers and wages in an urban labor market. Chicago: University of Chicago Press. 1970.

37.

Revenga

. Exporting jobs? Quarterly Journal of Economics. Feb 1992; 107: 255-284.

38.

Segal

. Are there returns of scale in city size? Review of Economics and Statistics. 1976; 89: 393-413.

39.

Singelmann

. From Agriculture to Services: The Transformation of Industrial Employment. Beverly Hills, CA.: Sage. 1987.

40.

Tickamyer

Duncan

. Work and Poverty in Rural America. In Rural Policy for the 1990’s; ed. J. Christenson, D. Flora. Boulder/London: Westview. In Press. 1990.

41.

Tienda

. Industrial Restructuring in Metropolitan and Nonmetropolitan Labor Markets: Implications for Equity and Efficiency. See Killian et. al. 1986; 33-70.

42.

Tilak

JBG

. Education and Its Relation to Economic Growth, Poverty, and Income Distribution: Past Evidence and Future Analysis. Washington D.C.: World Bank. 1989.

43.

Teulings

. The wage distribution model of the assignment of skills to jobs. Journal of Political Economy. 1995; 103(2): 280-315.

44.

Teulings

. Comparative advantage, relative wages, and the accumulation of human capital. Journal of Political Economy. 2005; 113(2): 425-461.

45.

Thurow

. A surge in inequality. Scientific American. May 1987; 256(5): 26-33.

46.

U.S. Census, county population data, extracted 2017, wwwcensus.gov.

47.

van de Klundert

. Wage differentials and employment in a two-sector model with a dual labor market. Metroeconomica. Oct 1989; 40: 235-256.

48.

Williams

Kornblum

. Growing Up Poor. Lexington, Mass: Heath. 1985.

49.

Williamson

. Regional inequality and the process of national development: A description of patterns. Economic Development and Cultural Change. 1965; 13: 3-45.