SEARCH

SEARCH BY CITATION

ABSTRACT

  1. Top of page
  2. ABSTRACT
  3. 1. INTRODUCTION
  4. 2. DATA AND METHODOLOGY
  5. 3. ESTIMATION AND RESULTS
  6. 4. CONCLUSION
  7. REFERENCES

This note examines the correlation between income inequality and economic growth using a panel of income distribution data for 3,109 counties of the U.S. Using the System Generalized Method of Moments (GMM) approach, we find that for the entire sample of 3,109 counties, an increase in a county's level of inequality has a significant negative relationship with future economic growth. In reality, however, the magnitude, sign, and significance of this relationship is not necessarily uniform across all regions of the U.S., motivating the need to examine regional differentials in the relationship between inequality and growth. Consequently, we split our dataset into metropolitan and nonmetropolitan counties, into the eight Bureau of Economic Analysis regions, and into regions of different political affiliation and economic-dependence typology. Our results show considerable heterogeneity in the relationship between inequality and growth across these regions.

1. INTRODUCTION

  1. Top of page
  2. ABSTRACT
  3. 1. INTRODUCTION
  4. 2. DATA AND METHODOLOGY
  5. 3. ESTIMATION AND RESULTS
  6. 4. CONCLUSION
  7. REFERENCES

The connection between income inequality and growth has captured the attention of economists for over 50 years, since the seminal work of Kuznets (1955). Different models and methods of analysis, and different data sets have yielded different results, sometimes sharply different, sometimes modestly. Panizza (2002) notes that a possible reason for these conflicting results is that many studies on the inequality-growth nexus typically use cross-country data. A well-known problem with cross-country studies is the quality and comparability of the cross-country inequality data. To sidestep this problem, Partridge (1997), and Hasanov and Izraeli (2011), among others, examine the inequality-growth relationship using data on the 48 contiguous U.S. states. Fallah and Partridge (2007) and Atems (2013) investigate the county-level determinants of inequality on growth; Glaeser, Resseger, and Tobio (2009) study inequality in cities, while Wheeler and La Jeunesse (2008) examine trends in neighborhood income inequality in the U.S.

The aforementioned studies, however, do not consider a detailed examination of heterogeneity in the relationship between growth and inequality. Fallah and Partridge (2007) note that spatial differences in agglomeration economies and economic incentives can produce different channels through which inequality affects growth. One such mechanism, suggested by Galor and Zeira (1993) is that high inequality makes it difficult for individuals at the lower end of the income distribution to invest in their human capital, thereby decreasing growth. To the extent that credit barriers to the poor exist, capital markets may be unable to uplift low income individuals from their cycle of low human capital, further weakening long run growth. Due to the immense spatial variation in rates of educational attainment, and hence human capital across U.S. counties, it is necessary to examine heterogeneity in the inequality-growth relationship. Furthermore, Hasanov and Izraeli (2011), using U.S. state-level data document nonlinearity in the relationship between inequality and growth, while using a spatial econometric approach, Atems (2013) examines the spatial dynamics of growth and inequality using U.S. county-level data. To the extent that the inequality-growth relationship may be nonlinear, and that there exist a spatial dimension to it suggests that there is need to consider geographic heterogeneity or regional differentials in the relationship between inequality and growth.

Another channel through which inequality affects growth, emphasized by Alesina and Rodrik (1994) and Alesina and Perotti (1996) is the role of political (in)stability on investment and growth. Alesina and Perotti (1996) hypothesize that high inequality in income distribution may lead to widespread discontent among the impoverished population, which may translate into civil unrest. Economic agents respond to such civil unrest by decreasing the scope of their economic activities to minimize risks, thereby stifling the rate of growth. Persson and Tabellini (1992), and Bertola (1993) discuss several other political economy channels. While such political economy channels are much less prevalent for advanced economies like the U.S., the rise of the Occupy Wall Street movement and similar movements have made it clear that the role of political economy channels, and the inextricable link between politics, inequality and growth need to be considered. Volscho and Kelly (2012) document that congressional shifts to the Republican Party in the U.S. that diminished union membership and lowered top tax rates played a strong role in the rise of the “super-rich.” Consequently, we examine the inequality-growth nexus across counties of different political affiliation. That is, we split our sample of 3,109 counties into three subsamples: Counties in Republican-leaning (Red) states, Democratic-leaning (Blue) states, and Independent (Purple) states, and examine the inequality-growth link across counties in these political regions.

Furthermore, because regions' economic and social characteristics are important determinants of their income distribution and growth dynamics, it is necessary to examine the inequality–growth relationship across regions of varying economic-dependence typology. At the cross-national level, the manufacturing sector and industrial development have had important roles in the growth of countries such as China, South Korea, Taiwan, and Indonesia. Some of these countries have managed to achieve growth with equity, while other countries, particularly in agriculture-dependent Sub-Saharan Africa, are witnessing slow growth and high inequality (Kniivila, 2007). To the extent that the inequality-growth linkage varies in countries with varying economic dependence typology, it may also vary across U.S. counties with similar differences in economic-dependence typology. Consequently, we examine the link between growth and inequality in counties in six nonoverlapping economic-dependence classes: farming-dependent, manufacturing-dependent, mining-dependent, federal and state government-dependent, services-dependent, and nonspecialized counties.

To further motivate the need to consider regional differentials in the inequality-growth nexus, consider Figure 1 which shows the spatial variation in average annual per capita real income growth rates of U.S. counties in the 1970s and 2000s. A glimpse of Figure 1 shows that in the 1970s, real per capita income grew approximately 3 percent annually (between 0.02 and 0.06 in most regions). A particularly striking feature in panel A of Figure 1 is that counties in the Great Plains experienced relatively low growth, with many experiencing negative growth. Several counties on the East and West coasts recorded positive growth. Three decades later (panel B), while some counties in the Great Plains still had low growth rates, others had progressed in terms of growth. In the Great Plains, except for counties in the central Midwestern states, many northern and southern counties experienced positive growth (between 0.02 and 0.06). The East and West Coasts, which had recorded positive growth in the 1970s witnessed reduced growth rates (0.0–0.02) in the 2000s. Equally noteworthy are counties in the Gulf Coast which saw declining per capita growth in the 2000s.

image

Figure 1. Spatial Variation in U.S. County-Level Per Capita Income Growth Rate.

Note: (a) Real Income Growth Rate: 1970–1980; (b) Real Income Growth Rate: 2000–2007.

Download figure to PowerPoint

The growth patterns described above have been accompanied by a reduction in the share of income earned by the bottom 90 percent of households. Figure 2 shows the spatial variation in inequality in U.S. counties in 1970 and 2000 measured by the Gini coefficient. The figure shows that in the 1970s (panel A), inequality was generally low (less than 0.38) in many U.S. counties. High inequality (over 0.38), however, is present in counties in southern and Gulf states such as Arizona, New Mexico, and Texas. Patches of high inequality are also apparent in the Dakotas, particularly in Dewey (0.54) and Shannon (0.52) counties of South Dakota and Emmons (0.45) county in North Dakota. Panel B shows the distribution of family income 30 years later. By 2000, inequality had risen across many counties, even in the hitherto low-inequality counties. The West and Northeastern coasts, which in the 1970s recorded low inequality, were, by 2000, experiencing severe income inequality (over 0.44). Counties in the southern states, as well as those in the Gulf coast which were already experiencing high inequality in 1970, saw higher inequality in 2000.

image

Figure 2. Spatial Variation in U.S. County-Level Income Inequality.

Note: (a) County-Level Inequality: 1970; (b) County-Level Inequality: 2000.

Download figure to PowerPoint

Figures 1 and 2 show significant regional variations in growth and inequality. Levernier, Partridge, and Rickman (1995, 1998, 2000) document similar variations in U.S. state-level and county-level income inequality, while Fallah and Partridge (2007) point out that the relationship between inequality and growth may differ between more and less populated counties. Splitting their sample into metropolitan and nonmetropolitan counties, Fallah and Partridge (2007) find a significant positive relationship for urban counties, and a significant negative relationship in rural counties. These regional variations in growth and inequality shown in Figures 1 and 2 and further documented by Levernier et al. (1995, 1998, 2000), and that the relationship between inequality and growth may vary between urban and rural counties suggests that it may also vary across major geographical regions of the U.S. Scatter plots of growth against inequality seem to support this idea. Panel A of Figure 3 shows a simple scatter plot of the average annual real per capita growth rate against inequality for metropolitan and nonmetropolitan counties, and the eight Bureau of Economic Analysis (BEA) regions of the U.S., namely the Far West, Great Lakes, Mideast, New England, Plains, Rocky Mountains, Southeast, and Southwest. Panel B shows a scatter plot of the relationship between changes in the two variables. Both panels of Figure 3 seem to suggest that regional differentials do exist in the relationship between (changes in) growth and (changes in) inequality. Consequently, there is need to examine such regional heterogeneity in the relationship between inequality and growth in greater detail.

image

Figure 3. Real Per Capita Income Growth vs. Inequality: Regional Differences.

Note: (a) Income Growth vs. Gini; (b) Change in Growth vs. Change in Gini.

Download figure to PowerPoint

The purpose of this note is to explicitly examine regional heterogeneity in the relationship between growth and inequality using a panel of income distribution data for 3,109 counties or county-equivalent administrative units of the U.S. One advantage of using county-level data is that much of the measurement errors that sometimes plague cross-country studies, such as intertemporal and international comparability of data are reduced, since these data are collected using the same techniques for the same time periods within the U.S. In addition, as pointed out by Ravallion (1998), using aggregate data might in fact conceal the effects of inequality on growth. Therefore, the use of county-level data provides a more comprehensive analysis of the link between inequality and growth, and pushes this inquiry one step closer toward more tenable results. Unlike previous county-level inequality-growth studies (e.g., Fallah and Partridge, 2007) that use cross-sectional (between) data, our analysis employs panel data to estimate the dynamic relationship between inequality and growth using the System GMM (within) approach. Therefore, it should be noted that the results presented hereinafter are not directly comparable to previous county-level studies. As noted by Partridge (2005), cross-sectional (or between) approaches focus on the long-run relationship between inequality and growth, while the within approach employed in this paper focuses on the short-run relationship between changes in inequality and changes in growth.

Using various model specifications, our results indicate that for the full sample of 3,109 counties, increases in inequality are significantly related to decreases in future growth. Following Barro (2000) and Hasanov and Izraeli (2011), we test for nonlinearity in the inequality-growth relationship, but find no significant evidence in favor the nonlinear specifications. We however, find significant evidence of regional differences. When we split the data into metropolitan and nonmetropolitan counties, and into the eight BEA regions, our results show negative and significant relationships for both metro and nonmetro counties, and counties in the Far West, Plains, Southeast, and the Southwest. Insignificant positive relationships are found for the Great Lakes and New England, while for the Rocky Mountains, and the Mideast, the relationship is positive and significant. In addition, we find significant positive relationships for counties in blue and purple states, but significant negative relationships for counties in red states. Furthermore, when we split the sample into the six economic-dependence classes mentioned earlier, the coefficient on inequality is positive and significant for federal/state government-dependent, services-dependent and nonspecialized counties; positive but insignificant for farming-dependent counties; and negative and significant for mining- and manufacturing-dependent counties. Finally, when we aggregate the data to the U.S. state level, the coefficient on inequality remains negative, but turns insignificant.

2. DATA AND METHODOLOGY

  1. Top of page
  2. ABSTRACT
  3. 1. INTRODUCTION
  4. 2. DATA AND METHODOLOGY
  5. 3. ESTIMATION AND RESULTS
  6. 4. CONCLUSION
  7. REFERENCES

The primary sources of data for the analysis are the U.S. Census Bureau and the Bureau of Economic Analysis Regional Economic Information System (BEA-REIS). The data set contains observations for 3,109 U.S. counties from 1970 to 2007. The growth model estimated considers the determinants of the average annual real per capita income growth over 10-year horizons (except for the 2000–2007 period where the average annual growth rate is over a seven-year period) at the county level.

Data on per capita personal incomes and the ensuing growth rates come from the BEA-REIS. Following the convention in most economic growth and inequality studies, the average annual real per capita income growth over 10-year periods is used as the dependent variable (see Partridge, 1997; Barro, 2000). On the one hand, while this long-term perspective is constrained by data limitations, on the other hand, such a long-term perspective follows the norm in the growth literature that strives to explain long- rather than short-run variations and reduces annual serial correlation from business cycle fluctuations.

The choice of independent variables that affect growth are well grounded in the literature, and their inclusion in the model shown hereinafter rests on the availability, and reliability of data for the counties of interest. These independent variables are dated at the start of each decade. By so doing, there should not be any direct endogeneity issues. By dating the variables at the start of each decade, the model also follows the convention in the literature that posits that economic growth converges to an equilibrium path based on initial conditions (Durlauf and Quah, 1999). To this end, the level of per capita income at the start of each decade is included as a regressor to account for convergence across counties. Data on other control variables come from a variety of sources. County-level income inequality data come from Nielsen (2002). Four education variables—the percentage of individuals with (i) less than high school education, (ii) high school, (iii) some college, and (iv) college education—are included to adequately proxy for the effects of human capital on growth. Other control variables used include the average wage per county per job, the county population density, the employment-to-population ratio, total nonfarm earnings per county, and the unionization rate per state.1 We begin by specifying a base growth model typical in this literature:

  • display math(1)

where inline image is the natural log of real per capita income of county i in period inline image. The left hand side is the average annual growth rate of real per capita income of county i in period t. inline image is the Gini coefficient at time inline image. inline image contains the other period inline image variables that affect growth mentioned above. inline image denotes the unobservable time effects, inline image, the time-invariant county effects and, inline image is the disturbance term. Table 1 provides the descriptive statistics, definitions and sources.

Table 1. Summary Statistics, Variable Transformations, and Data Sources
 MeanStandard DeviationMinimumMaximum
  1. Data sources:

  2. a

    Economic Information System (BEA);

  3. b

    Nielsen (2002);

  4. c

    Economic Research Service (USDA);

  5. d

    National Association of Counties (NACo);

  6. e

    Hirsch, McPherson, and Vroman (2001).

y: Real per capita personal income (log)a9.7960.3858.42111.457
ineq: Gini coefficient of inequalityb0.3890.0480.0130.605
lhsc: Less than high school educationc0.3730.1660.0300.884
hsc: High school completion ratec0.3300.0750.0580.543
assoc: Some college educationc0.1750.0850.0590.449
coll: College completion ratec0.1220.0700.0110.637
density: Population density per county (log)d3.6521.658−3.31712.031
empratio: Employment-to-population ratioa0.4690.2770.04915.957
wage: Average wage per job (log)a10.0450.2678.99911.315
union: State union membership ratese0.1340.0730.0290.349
earnings: Total nonfarm eranings (log)a11.8201.8185.82919.311

3. ESTIMATION AND RESULTS

  1. Top of page
  2. ABSTRACT
  3. 1. INTRODUCTION
  4. 2. DATA AND METHODOLOGY
  5. 3. ESTIMATION AND RESULTS
  6. 4. CONCLUSION
  7. REFERENCES

Estimation

Equation (1) can be estimated using OLS, or the fixed and random effects methods specified in the panel data literature. In fact, a Hausman specification test (inline image) rejects the random effects model in favor of the fixed effects (FE) model. However, several problems arise if Equation (1) is estimated using OLS, fixed or random effects methods.2 To avoid these problems, many dynamic panel studies of inequality and growth use the Arellano and Bond GMM estimator. Arellano and Bond (1991) show that for short dynamic panels (inline image, and T is fixed), Equation (1) is first-differenced to eliminate the individual effects inline image. First differencing is essentially equivalent to controlling for county-specific time-invariant characteristics. This gives:

  • display math(2)

where inline image.3 To include FE, the variables in the above differenced equation should be taken in deviation from their period means.

Blundell and Bond (1998) argue that using lagged levels are poor instruments for first differences. They show that as inline image in (1), the Arellano and Bond GMM estimator suffers a downward finite sample bias. They propose the use of the System GMM estimator. Blundell, Bond, and Windmeijer (2000) show that the System GMM estimator for equations such as Equation (1) outperforms the first difference GMM estimator. Roodman (2009) further notes that for short panels such as the one in this paper (3, 109 × 4) the system GMM seems appropriate because if T is large, the number of instruments in system GMM tends to explode, while the Arellano–Bond test for autocorrelation becomes unreliable when N is small. To estimate the dynamic panel data model in this paper, we therefore use the System GMM estimator. Table 2 presents the estimates of the System GMM. For the purpose of comparability, the OLS, FE, the continuously updating GMM (CUE), and the Arellano and Bond GMM (ABGMM) estimates are also presented.

Table 2. Estimation Results for the Dynamic Panel Data Model
Variableinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
Notes
  1. Boldface numbers indicate significance at the 5 percent level. For the fixed effects model, R2 is the within-R2. HAC-corrected standard errors are reported in parentheses. The GMM estimates reported are all two step, except inline image which reports the one-step System GMM results. inline image estimates the two-step procedure without period dummies. inline image includes a inline image term to the baseline specification in column 5, while inline image, includes a Gini-income interaction term.

inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline image       inline image 
inline image        inline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
R2inline imageinline imageinline image      
inline image   inline imageinline imageinline imageinline imageinline imageinline image
inline image   inline imageinline imageinline imageinline imageinline imageinline image
inline image  262.340      
inline image  inline imageinline imageinline imageinline imageinline imageinline imageinline image

Results

Full sample

To show the importance of considering heterogeneity, we begin by analyzing the effect of inequality on growth for our entire sample of 3,109 counties before proceeding to estimate the relationship for various geographical, political, and economic-dependence typology regions. Table 2 shows the result of estimating Equation (1) across different specifications. Several observations can be made from Table 2. First, regardless of the estimation method, the coefficient on inequality is negative and significant (except for the nonlinear specifications in columns 8 and 9). Since our primary focus is on the estimates of the system GMM estimator (inline image), the estimate of −0.318 (standard error, 0.152) indicates a significantly negative relationship between changes in inequality at the start of each decade, and changes in growth through out the decade. With P-values of 0.220 and 0.606, the Hansen J-test of overidentifying restrictions and the Arellano and Bond test for second order serial correlation respectively indicate valid identifying assumptions and no second order serial correlation.

Second, the coefficient on initial income (inline image) is negative and significant across all specifications, providing evidence of convergence across U.S. counties. Blundell and Bond (2000) note that for models such as (1), the OLS estimate of inline image will be upward biased, while the FE and inline image estimates will be downward biased and inconsistent. A consistent estimate of inline image is therefore expected to lie between the upward-biased OLS and the downward-biased FE and inline image estimates. Clearly, the coefficient estimate of inline image from the system estimator (inline image) of −0.071 lies between the upward-biased OLS estimate of −0.009 and the downward-biased FE and inline image estimates of −0.091 and −0.080, respectively. Furthermore, because no formal tests have yet been developed for evaluating the strength of the instrument set for system GMM, Bun and Windmeijer (2010) point out that “the Wald test statistic using continuously updating GMM (CUE) estimation results and their proposed variance estimator performs well in a static panel data model estimated in first differences. As the number of instruments in this panel data setting grow quite rapidly with the time dimension of the panel, this may be a sensible approach also for the system moment conditions.” Accordingly, we estimate Equation (1) in differences using the CUE that is robust to weak instruments (column 3). The coefficient on inline image using the CUE of −0.072 is identical to the inline image estimate, validating the suggestion of Bun and Windmeijer (2010). Also, rather than rely on first-stage GMM regressions in isolation, we compute the Cragg–Donald Wald statistic which may be more informative than the F-statistics from the first-stage regressions in models with more than one endogenous variable. When compared to the critical values developed by Stock and Yogo (2005), the computed Cragg–Donald statistic of 262.34 rejects the hypothesis that the instruments are weak. Overall, these findings provide some degree of confidence in our results despite the fact that no formal tests for weak instruments in system GMM estimation currently exist.

Third, the coefficients on the education variables are generally positive and significant, supporting the view that counties with higher rates of education are expected to experience faster growth rates. The coefficients on the other variables are expected and consistent with previous studies. Looking at inline image, the significant negative coefficient on inline image suggests that densely populated counties are expected to grow much slower than their less dense counterparts. The model also includes the average wage per job inline image and the employment-to-population ratio (inline image) to capture local labor market trends. The inline image coefficient is negative and significant, while the positive (although insignificant) relationship between the employment-to-population ratio and growth is consistent with previous studies that document positive employment elasticities or intensities. A positive coefficient on aggregate nonfarm earnings is found, and a negative relationship between unionization and growth is documented, consistent with, among others (Card, 1992).

The results presented in the next section estimate Equation (1) for various regions using the two-step system GMM specification in column 5 (inline image), so it is important to conduct further sensitivity analyses to validate the chosen model. A known problem with two-step estimation is that even though the standard covariance matrix is robust to panel-specific heteroskedasticity and autocorrelation, the standard errors tend to be biased downward in finite samples. Consequently, we re-estimate the one-step variant of the inline image. inline image in column 6 shows the estimates of this specification. The results, including the tests for second-order serial correlation and overidentifying restrictions are identical to the two-step results shown in column 5 (inline image). As an alternative robustness check, we estimate the model without the period dummies (column 7). The results show that the signs and significance for many of the estimates remain as previously found, however, the Hansen test of overidentifying restrictions shows that the instruments in this specification are invalid (p = 0). This confirms that time dummies should be included in the model.

Several authors, including Barro (2000), and Hasanov and Izraeli (2011) contend that the relationship between inequality and growth is nonlinear. Barro (2000) shows that for rich countries—countries with per capita GDP over $2000 (1985 U.S. dollars)—the inequality-growth relationship is positive, whereas for poor countries, the relationship is negative. Hasanov and Izraeli (2011), using U.S. state-level data, confirms Barro's findings. One way to test for nonlinearity, following Barro (2000), and Hasanov and Izraeli (2011) is to introduce a Gini-income interaction term, a Gini squared term, or both. Column 8 includes a Gini squared term to the baseline specification in column 5, while column 9 includes a Gini-income interaction term. As both columns show, the coefficient on the nonlinear terms as well as the linear inequality terms are insignificant, suggesting that the relationship between growth and inequality is, in fact, the linear model in column 5. Partridge (2005), using U.S. state-level data, and Forbes (2000), using country-level data experiment with nonlinearities, and also reject the nonlinear specifications in favor of the linear specification.

Regional differences in the relationship between inequality and growth

Table 2 shows a robust negative relationship between changes in inequality and changes in subsequent economic growth. In reality, the magnitude, sign, and significance of the relationship is not necessarily uniform across all regions of the U.S. To investigate whether regional differences in the relationship between growth and inequality exist, and the magnitude of these regional differences, if any, we estimate the model for metro and nonmetro counties, and the eight BEA regions. The resulting inequality coefficients are reported in Table 3. As the table shows, the coefficients vary significantly across counties in the eight BEA regions. The negative and statistically coefficients for metro and nonmetro counties, as well as for counties in the Far West, Plains, Southeast, and Southwest regions indicate that inequality is inversely related to future economic growth in these regions. On the contrary, for counties in the Mideast and the Rocky Mountain, the effect of inequality is positive and significant, while for counties in the Great Lakes and New England regions, inequality produces positive effects, although statistically insignificant.4

Table 3. Regional and Income Differentials in the Relationship between Inequality and Growth
VariableCoefficient on IneqStandard ErrorCountiesinline imageJ-TestLong-Period Cross-Section
Notes
  1. Boldface numbers indicate significance at the 5 percent level. Numbers in parentheses are robust standard errors. For each sample, we use the two-step system GMM estimator including time dummies in the baseline specification as in column (5) of Table 2. Complete results are available from the author upon request.

Standard analysis      
Whole sampleinline image0.1523,1090.6060.220inline image
Regional groups      
Metroinline image0.1081,0860.0440.330inline image
Nonmetroinline image0.2082,0230.0410.168−0.012
Far West (inline image)inline image0.2521550.1270.947inline image
Great Lakes (inline image)inline image0.0584350.0220.313−0.007
Mideast (inline image)inline image0.0491770.0160.085inline image
New England (inline image)inline image0.138670.7400.708inline image
Plains (inline image)inline image0.1066180.1450.407inline image
Rocky Mountain (inline image)inline image0.1832150.0760.225−0.003
Southeast (inline image)inline image0.0561,0640.1080.127inline image
Southwest (inline image)inline image0.2193770.4610.328inline image
Political Groups      
Blue Statesinline image0.0958560.0620.154inline image
Red Statesinline image0.0851,7730.7200.152inline image
Purple Statesinline image0.0704800.6510.056inline image
Economic-dependence groups      
Farminginline image0.0764400.8980.114inline image
Mininginline image0.2391260.1390.208inline image
Manufacturinginline image0.5809000.8020.598inline image
Federal/State Governmentinline image0.0963700.5510.523inline image
Servicesinline image0.0633390.3510.108inline image
Nonspecializedinline image0.0459340.6760.199−0.001
Aggregation      
States−0.4780.607490.7050.611inline image

Before proceeding to the results for other regions, it is important to note an important observation regarding the results presented so far. Fallah and Partridge (2007), find a significant positive coefficient on inequality for metro counties while our results show a significant negative coefficient. It should be noted, however, that our findings are not directly comparable or necessarily contradictory to those reported in previous county-level studies. Previous work (e.g., Fallah and Partridge, 2007) use OLS or instrumental variables (between approaches) to estimate some variant of Equation (1). Our analysis estimates the dynamic relationship between inequality and growth using the System GMM (within) approach. Partridge (2005) notes that between or cross-sectional approaches focus on the long-run relationship between the level of inequality and the level of growth, while the within approach employed in this paper focuses on a short-run relationship between changes in inequality and changes in growth. As Figure 3 shows, the relationship in levels (panel A) can vary noticeably from the relationship in differences (panel B). To further assess whether the system GMM estimates mostly reflect short-run responses, the last column of Table 3 reports results of a long-period cross-sectional model for each region. That is, using OLS, we regress the average annual growth rate between 1970 and 2007 on variables measured in 1970. Several important observations can be made from these results. First, consistent with Fallah and Partridge (2007), the coefficient on inequality is now positive and significant for metro counties, and negative (although insignificant) for nonmetro counties. Second, while inequality may be detrimental to growth in the short run, the effect of inequality on growth may be positive in the long run (for example the coefficient for the entire sample is −0.318 using inline image, while the long-period model coefficient is 0.015). This finding supports the incentives and Schumpeterian arguments on the relationship between growth and inequality which suggest that in the short run, high inequality may be detrimental to growth as low income individuals view society as unfair, and opportunities for upward mobility are limited. However, as inequality persists, it increases entrepreneurial and work effort, and brings about a higher level of capital accumulation that increases efficiency and growth in the long run.

As shown above, the geographic location of counties can have significant impacts on their development process, which in turn affects their income distribution and growth dynamics. Similarly, counties' economic characteristics can have significant effects on their development process, as well, and this can affect their income distribution and growth dynamics too. The Economic Research Service (ERS) of the U.S. Department of Agriculture (USDA) has developed county-level typology codes that capture differences in economic and social characteristics across counties. These codes separate U.S. counties into six nonoverlapping economic dependence classes, including farming, mining, manufacturing, services, Federal/State government, and unspecialized counties. We examine the relationship between inequality and growth across these six economic classes. Such an examination is important because it provides policy-relevant information on what kind of policies are needed to alleviate inequality while enhancing growth in different economic areas. Table 3 shows that the relationship between growth and inequality varies significantly across counties of different economic dependence typology. The table shows that for mining- and manufacturing-dependent counties, changes in inequality are inversely related to future growth. On the other hand, federal and state government-dependent, services-dependent, and nonspecialized counties have significant positive inequality coefficients, suggesting that inequality is growth-enhancing in these counties. These findings are important because they suggest that policies aimed at decreasing income inequality must be region-specific, and based on the economic dependence typology of the county.

Another aspect of inequality and growth which has been left largely unexamined, is whether the relationship varies across regions based on their political affiliation. It is well known that the political affiliation of regions determines the policies of these regions, which in turn affect income and inequality. Pogge (2008) notes that in regions with high inequality, the rich have greater opportunities to influence political and economic policy to their advantage, and these policies affect the growth of the region. Volscho and Kelly (2012) find that over the 1980–2008 period, congressional shifts to the Republican Party in the U.S. that diminished union membership and lowered top tax rates played a strong role in the rise of the “super-rich.” It is surprising that the link between inequality and growth has not be examined across different political regions. To examine this linkage, we split our sample into counties in red states, blue states, and purple states, and re-estimate the specification in column 5. The negative coefficient on inequality for red states shown in Table 3 indicates that for these states, high inequality is associated with decreases in future growth. On the contrary, the table shows that for purple and blue states, inequality is growth-augmenting.

Several studies on the relationship between growth and inequality typically use aggregate data. One limitation of aggregate data is that aggregation can potentially conceal the underlying relationship between inequality and growth. Ravallion (2001) notes that “spurious inequality effects in an aggregate growth regression can arise from the assumptions made in aggregating across microrelationships, given credit market failures.” Using a microgrowth model estimated on farm-household data for China, Ravallion (1998) finds that aggregation severely biases conventional tests of whether inequality impedes growth. To investigate whether the relative significance of inequality on growth varies with different levels of spatial aggregation of observational units, we aggregate the county-level data to the U.S. state level, and re-estimate inline image. Table 3 shows that the coefficient on inequality is now −0.478 and is statistically insignificant, consistent with Ravallion (1998). This finding is equally significant because it suggests that previous studies that use more aggregate data may in fact be concealing the underlying relationship between inequality and growth.

One drawback of estimating a separate regression for each region (as shown in Table 3) is that each region specification does not tell whether there are any statistical differences in inequality across regions (for example between metro and nonmetro counties). As a robustness check, we estimate the base model (inline image) in Table 2 including dummy variables for various regions, and then interact these dummy variables with the inequality variable. Specifically, we generate inline image, an interaction between ineq and inline image (where inline image for nonspecialized counties, and 0 for specialized counties) to capture statistical differences in inequality between nonspecialized counties and counties of other economic-dependence typology. Similarly, inline image is included to capture statistical differences in inequality between Republican-leaning (Red) counties and other counties; inline image to identify differences between the Southeast and other BEA regions; while inline image attempts to examine statistical differences between metro and nonmetro counties.5 Table 4 shows the result of this specification. The coefficient on inequality is positive but insignificant, while inline image is negative and insignificant, suggesting (and consistent with Table 3) that there are no statistically significant differences in the relationship between inequality and growth between metro and nonmetro counties. In addition, the coefficients on inline image and inline image are significantly negative and that for inline image is positive and significant. Specifically, the significantly negative coefficient on inline image indicates that on average, the effect of inequality on growth is lower for nonspecialized counties than for counties with specialized economic dependence typology, and the significantly negative coefficient on inline image indicates that the effect of inequality on growth is lower for counties in Red states than for counties in states of different political affiliation. These findings suggest that statistically significant differences in the inequality-growth linkage exist across regions, and is consistent with the view that different transmission mechanisms of economic incentives, opportunities and obstacles, and agglomeration economies exist across different geographic, political, and economic regions.

Table 4. Sensitivity Analysis
VariableEstimateVariableEstimateVariableEstimateVariableEstimate
inline imageinline imageinline imageinline imagePLinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageServicesinline image
inline imageinline imageinline imageinline imageSEinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageinline imageinline imageinline imageinline imageinline imageinline image
inline imageinline imageNEinline imageManufacturinginline imageinline imageinline image

4. CONCLUSION

  1. Top of page
  2. ABSTRACT
  3. 1. INTRODUCTION
  4. 2. DATA AND METHODOLOGY
  5. 3. ESTIMATION AND RESULTS
  6. 4. CONCLUSION
  7. REFERENCES

This note reconsiders the correlation between growth and inequality using U.S. county-level data covering the decades of 1960–2000. Using the System GMM approach, we first examine the relationship between growth and income inequality for the entire sample of 3,109 counties, and find a significant negative relationship between changes inequality and changes in subsequent periods.

The robust negative relationship between changes in inequality and changes in growth does not necessarily imply that the magnitude and/or sign of this relationship is the same across all regions of the U.S. Because studies by Levernier et al. (1995, 1998, 2000), and Fallah and Partridge (2007) document significant variations in U.S. state-level and county-level income inequality, we investigate whether such regional variations affect the relationship between growth and inequality. We split our sample into the eight major regions of the U.S.—the Far West, the Great Lakes, the Mideast, New England, Plains, Rocky Mountain, the Southeast, and the Southwest—and also into metropolitan and nonmetropolitan counties. We find negative and significant coefficients on inequality for the Far West, Plains, the Southeast, the Southwest, metropolitan, and nonmetropolitan counties; positive and significant coefficients for counties in the Mideast and the Rocky Mountains; and positive but insignificant relationships between growth and inequality for counties in the Great Lakes and New England. We further examine the relationship between growth and inequality in counties in six nonoverlapping economic-dependence classes, as well as counties in states of different political affiliation. We find significant growth-enhancing effects of inequality in Federal and state government-dependent, services-dependent, and nonspecialized counties, as well as in counties in blue and purple states; and inverse relationships between growth and inequality for counties in red states, and manufacturing- and mining-dependent counties. These findings suggest that different transmission mechanisms of economic incentives, opportunities and obstacles, and agglomeration economies exist across regions of various geographic, political, and economic-dependence typology.

A key policy implication of the results presented in this paper is that because of geographic, political, and economic differentials in the relationship between growth and inequality, a one-size-fits-all income-equalizing policy may be detrimental to some regions, while beneficial to others. For the Rocky Mountain, Mideast, services-dependent, federal and state government-dependent, and nonspecialized counties, our findings imply that redistribution policies aimed at decreasing income inequality might also lead to a reduction in growth, whereas these same policies might be growth-augmenting for the Far West, Plains, the Southeast, the Southwest, mining- and manufacturing-dependent counties. Therefore, perhaps policies aimed at decreasing income inequality ought to be more decentralized and/or localized.

  1. 1

    Data on union membership rates are not available at the county-level, so the state-level rates are used.

  2. 2

    See Forbes (2000, pp. 875–876) for a detailed discussion of these issues.

  3. 3

    To see this more clearly, note that inline image. Therefore Equation (1) can be written as inline image. Adding inline image to both sides yields: inline image. First differencing this equation yields Equation (2), where now inline image.

  4. 4

    One reviewer pointed out that the finding of a different inequality-growth relationship in Table 3 might indicate that the results reported in Table 2 are just driven by region fixed effects, and suggested re-estimating the models in Table 2, controlling for region specifications (dummies) as listed in Table 3. We, however, find no significant changes in the results presented in Table 2 after controlling for region dummies.

  5. 5

    We thank one of the anonymous reviewers for suggesting this.

REFERENCES

  1. Top of page
  2. ABSTRACT
  3. 1. INTRODUCTION
  4. 2. DATA AND METHODOLOGY
  5. 3. ESTIMATION AND RESULTS
  6. 4. CONCLUSION
  7. REFERENCES
  • Alesina, Alberto and Roberto Perotti. 1996. “Income Distribution, Political Instability and Investment,” European Economic Review, 40(6), 12031228.
  • Alesina, Alberto and Dani Rodrik. 1994. “Distributive Politics and Economic Growth,” Quarterly Journal of Economics, 109(2), 465490.
  • Arellano, Manuel and Stephen Bond. 1991. “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations,” Review of Economic Studies, 58(2), 277297.
  • Atems, Bebonchu. 2013. “The Spatial Dynamics of Growth and Inequality: Evidence Using U.S. County-Level Data,” Economics Letters, 118(2), 1922.
  • Barro, Robert. 2000. “Inequality and Growth in a Panel of Countries,” Journal of Economic Growth, (5), 532.
  • Bertola, Giuseppe. 1993. “Factor Shares and Savings in Endogenous Growth,” American Economic Review, 83(5), 11841198.
  • Blundell, Richard and Stephen Bond. 1998. “Initial Conditions and Moment Restrictions in Dynamic Panel Data Models,” Journal of Econometrics, 87(1), 115143.
  • Blundell, Richard, Stephen Bond, and Frank Windmeijer. 2000. “Estimation in Dynamic Panel Data Models: Improving on the Performance of the Standard GMM Estimator,” in Badi H. Baltagi, Thomas B. Fomby, and R. Carter Hill (eds.) Nonstationary Panels, Panel Cointegration, and Dynamic Panels (Advances in Econometrics, Volume 15). Emerald Group Publishing Limited, pp. 5391.
  • Bun, Maurice and Frank Windmeijer. 2010. “The Weak Instrument Problem of the System GMM Estimator in Dynamic Panel Data Models,” Econometrics Journal, 13(1), 95126.
  • Card, David. 1992. “The Effect of Unions on the Distribution of Wages: Redistribution or Relabelling?NBER Working Paper No. 4195, 161.
  • Durlauf, Steven and Danny Quah. 1999. “The New Empirics of Economic Growth,” in John B. Taylor and Michael Woodford (eds.), Handbook of Macroeconomics. Amsterdam: Elsevier, 235308.
  • Fallah, Belal and Mark Partridge. 2007. “The Elusive Inequality-Economic Growth Relationship: Are there Differences Between Cities and the Countryside?Annals of Regional Science, 41(2), 375400.
  • Forbes, Kristin. 2000. “A Reassessment of the Relationship between Inequality and Growth,” American Economic Review, 40(4), 869887.
  • Galor, Oded and Joseph Zeira. 1993. “Income Distribution and Macroeconomics,” The Review of Economic Studies, 60(1), 3552.
  • Glaeser, Edward, Matt Resseger, and Kristina Tobio. 2009. “Inequality In Cities,” Journal of Regional Science, 49(4), 617846.
  • Hasanov, Fuad and Oded Izraeli. 2011. “Income Inequality, Economic Growth and the Distribution of Income Gains: Evidence From the U.S. States,” Journal of Regional Science, 51(3), 518539.
  • Hirsch, Barry, David McPherson, and Wayne Vroman. 2001. “Estimates of Union Density by State,” Monthly Labor Review, 124(7), 5155.
  • Kniivila, Matleena. 2007. “Industrial Development and Economic Growth: Implications for Poverty Reduction and Income Inequality,” in Industrial Development for the 21st Century: Sustainable Development Perspectives. New York: United Nations, Department of Social and Economic Affairs, 295332.
  • Kuznets, Simon. 1955. “Economic Growth and Income Inequality,” American Economic Review, 45(1), 128.
  • Levernier, William, Mark Partridge, and Dan Rickman. 2000. “The Causes of Regional Variations in U.S. Poverty: A Cross-County Analysis,” Journal of Regional Science, 40(3), 473497.
  • Levernier, William, Mark Partridge, and Dan Rickman. 1998. “Differences in Metropolitan and Nonmetropolitan U.S. Family Income Inequality: A Cross-County Comparison,” Journal of Urban Economics, 44(2), 272290.
  • Levernier, William, Mark Partridge, and Dan Rickman. 1995. “Variation in U.S. State Income Inequality:1960-1990,” International Regional Science Review, 18, 355378.
  • Nielsen, Francois. 2002. “Income Inequality in U.S. Counties (Gini Coefficients),” Retrieved June 10, 2010, from http://www.unc.edu/~nielsen/data/data.htm
  • Panizza, Ugo. 2002. “Income Inequality and Economic Growth: Evidence from American Data,” Journal of Economic Growth, 7(1), 2541.
  • Partridge, Mark. 1997. “Is Inequality Harmful for Growth? Comment,” American Economic Review, 87(5), 10191032.
  • Partridge, Mark. 2005. “Does Income Distribution Affect U.S. State Economic Growth?Journal of Regional Science, 45(2), 363394.
  • Persson, Torsten and Guido Tabellini. 1992. “Growth, Distribution and Politics,” European Economic Review, 36(2-3), 593602.
  • Pogge, Thomas. 2008. “Growth and Inequality: Understanding Recent Trends and Political Choices,” Dissent, 55(1) 6675.
  • Ravallion, Martin. 1998. “Does Aggregation Hide the Harmful Effects of Inequality on Growth?Economics Letters, 61(1), 7377.
  • Ravallion, Martin. 2001. “Growth, Inequality and Poverty: Looking Beyond Averages,” World Development, 29(11), 18031815.
  • Roodman, David. 2009. “How to do xtabond2: An Introduction to Difference and System GMM in Stata,” Stata Journal, 9(1), 86136.
  • Stock, James and Motohiro Yogo. 2002. “Testing for Weak Instruments in Linear IV Regression,” NBER Technical Working Papers 0284.
  • Volscho, Thomas and Nathan Kelly. 2012. “The Rise of the Super-Rich: Power Resources, Taxes, Financial Markets and the Dynamics of the Top 1 Percent, 1948 to 2008,” American Sociological Review, 77(5), 679699.
  • Wheeler, Christopher and Elizabeth La Jeunesse. 2008. “Trends in Neighborhood Income Inequality in the U.S.: 1980–2000. Journal of Regional Science, 48(5), 979891.