Examining the Relationship Between the Poverty Rate and Economic Conditions in the Southern United States

Transcription

1 Examining the Relationship Between the Poverty Rate and Economic Condions in the Southern Uned States Gary A. Hoover Department of Economics, Finance, and Legal Studies Universy of Alabama Geoffrey Wallace Department of Public Affairs and Economics Universy of Wisconsin January 2005 Abstract: This paper documents the historical trend of higher poverty rates in the South compared to the rest of the country. It examines any differences that might exist in how the poverty population in the country as a whole and the poverty population in the South respond to different economic and demographic factors. Economic factors such as unemployment and wages impact poverty populations similarly, in terms of sign and significance but have smaller impacts on poverty populations. After testing, is shown that poverty populations in the South do not respond in similar fashions to the rest of the country regarding demographic factors such as households headed by single females or non whes. JEL: I30, I31 This project was supported wh a grant from the UK Center for Poverty Research through the U.S. Department of Health and Human Services, Office of the Assistant Secretary for Planning and Evaluation. The opinions and conclusions expressed herein are solely those of the authors and should not be construed as representing the opinions or policy of the UKCPR or any agency of the Federal government. The authors wish to thank James Peoples and two anonymous referees for helpful comments. The usual caveats apply regarding errors. Correspondence: Hoover, Universy of Alabama, Department of Economics, Box , Tuscaloosa, AL , PH: , ghoover@bama.ua.edu

2 INTRODUCTION Over the last forty years, many resources and much effort have been devoted by policy makers and academics to understanding the causes of poverty and how certain economic and demographic factors impact poverty populations. Researchers such as Blank (1991), Freeman (2001), and LeBlanc (2000) have explored the causes and responses of poverty using aggregate measures. Interesting results have come from these endeavors which have been helpful in highlighting effective anti-poverty policy. One limation of this type of research has been the limed number of observations due to the lack of long and sustained time-series data. Blank and Card (1993), Beeghley (1988), Sawhill (1988), and Moen (1989) were among the first researchers to use regional data to examine the determinants of poverty. By creating these panels, the impacts and determinants of poverty could be examined more thoroughly. Addionally, because there are large differences in demographics and economic condions across regions in the Uned States, researchers have been able to learn more about the determinants of national and regional poverty through the use of regional poverty data. Their results revealed that certain economic and demographic characteristics, such as female headed households and increases in the unemployment rate were significantly linked to poverty both nationally and regionally although not necessarily uniformly. The South has been one region of the country that has historically higher poverty rates than the rest of the country. Figure 1 gives a graphical illustration of historical poverty levels for the Uned States. Panels A and B show those Census divisions that have historically been below the national level and those that have been above. Table 2 provides simple summary statistics for the Uned States as a whole and each of the nine Census divisions. It is interesting to note that, as was illustrated in Figure 1, the three divisions (5,6, and 7) wh the highest average poverty 2

3 rates over the sample period of 1980 through 2001 comprise the South region for the Uned States. 1 In addion, these divisions consistently had lower wages and have suffered through higher rates of unemployment. It is well established that education is directly correlated to income earning capacy and these regions have historically had lower levels of college graduates. In most categories highlighted in Table 2 those divisions comprising the South have lagged. Researchers have shown that investments in human capal and infrastructure have been significant determinants of this historical trend. Recent work by Levernier (2003) reveals that understanding differences in the demographic characteristics of metropolan and nonmetropolan Southern populations are crical in developing effective anti-poverty policy. This analysis did not compare the South wh non-southern populations but examined intra-regional differences in poverty populations. Bishop et al. (1992) show how income distributions of Southern and non-southern populations have been converging over time. Their analysis was not specific to poverty populations, however. The analysis that follows will have results that will be two-fold. We will examine whether poverty populations in the South respond to certain economic and demographic factors in a manner that is consistent wh the rest of the country. This will be an important aspect of our analysis since the majory of public policy that is aimed at the poor is national. In addion, we will examine whether any differences that may appear are systematic. As stated previously, Figure 1 shows graphically that over the last 20 years those Census divisions comprising the South have had sustained poverty rates above the national average. By understanding that poverty populations in the South respond differently to certain factors and are impacted to a lesser degree by others, a more effective anti-poverty policy can be formulated. 1 See Table 1 for break downs by regions and divisions. 3

4 THE DATA In this study we use division level panel data constructed from the March Current Population Survey (CPS) to investigate differences in the relationship between poverty and economic and demographic condions in the South and the rest of the country. The primary advantage of using regional level panel data is one of sample size. Instead of using one relatively short time-series to estimate the relationship between economic condions and poverty, we can make use of 9 divisional time series. As noted by Blank and Card (1993), another important advantage of divisional level panel data is that allows for the use of unrestricted year effects. The inclusion of year effects are potentially important as they can be used to control for any omted variables that have similar effects across divisions or regions. Nearly all of the data used in this analysis come from the Current Population Survey: Annual Demographic Files (March CPS). The March CPS is a large annual survey of 50 to 100 thousand US households administered by the Census Department. Each March (since 1968), survey participants are asked about their background, living arrangements, and income over the prior calendar year. To form the sample used in the analysis that follows, we computed poverty rates in nine census divisions for each year from 1980 to (See Table 1 for identifiers of each division and region.) Using the data from the March CPS, we estimate a family measure of poverty that is similar to that of the Census Bureau, wh the exception that we designate individuals as families wh only one member. When the federal government adopted an official poverty measurement methodology, assigned responsibily for implementing annual and decennial measures of poverty to the U.S. Census Bureau. The Census Bureau uses a set of money income thresholds that vary by family size and composion to determine who is poor. If a family's total income is 4

5 less than the applicable threshold for that type of family, then the family, and every individual in, is counted as poor. Over time the thresholds are adjusted for inflation using the Consumer Price Index (CPI-U). Crics of the official poverty measure note that does not vary geographically and s reliance on money income before taxes, which does not include capal gains nor noncash benefs, is an incomplete measure of a family s well-being. Further, the official poverty measures exclude persons in milary barracks, instutional group quarters (e.g., prisons) and unrelated individuals under age 15 (such as foster children). Researchers, such as Formby et al. (2001), have used alternative poverty measures to investigate the impact of certain economic and demographic variables on poverty, concluding that the differences have not generally been that dramatic. 2 The previously collected poverty data was merged wh data on divisional and regional unemployment rates obtained from the Bureau of Labor Statistics. These regional unemployment rates were created by aggregating state-level data on the number of persons employed and in the labor force and the number of employed workers. Our selection of principle determinants or controls for poverty come from those used by previous researchers such as Freeman (2003), Levernier (2003) and Blank and Card (1993). We calculated the fraction of individuals in each of three education categories: less than high school, high school graduate wh no college, and some post-secondary school training. In our data we include the percentage of families headed by single females. Because single-earner families have less income than dual earner families, are subject to more frequent and severe income shocks, and women earn less than men, female-headed families are more likely to be in poverty than 2 The U.S. Census Bureau has experimented wh alternative poverty measures suggested by a Panel of the National Academy of Sciences (see Cro and Michael, 1995). For further information concerning poverty measurement methods in the U.S. the reader is directed to the U.S. Census Bureau s webse at See Zheng (1997) for a comprehensive review of alternative poverty measures. 5

6 male-headed families or married couple families. Our data also includes the fraction of married couple families by division and region. To these economic and family structure variables we add the percentage of families headed by an individual who are 65 years of age or older (elderly) and the percentage of families headed by non-whe persons. 3 Our final control variable is the percentage of families living in the central cy. Blank and Card (1993) include the percentage of persons living in rural areas. After 1991, the Census Bureau stopped collecting this information; thus, is not included in the analysis. In addion, Levernier (2003) examines differences in poverty rates in metropolan versus non metropolan counties in the South region exclusively. Our data will allow us to examine whether a metropolan effect exists across regions. THE MODEL To investigate the extent to which labor market condions and demographic factors affect the poverty rate we will make use of the following Ordinary Least Squares (OLS) specification: (1) Povertyrate + β College 5 = α + β Urate 0 + β Nwhe β Wages + β Sfemale β Married 3 + β Ccy 8 + β Diploma + β Elderly 9 4 +Θ i + δ + ε t where Povertyrate, Urate, and Wages are the poverty rate, unemployment rate, and log of wages in division i ( i= 1,...,9) during year t (t = 1980,,2001). Married, Diploma, and College are the percentage of the division that have married heads of household, percentage of heads of household wh high school and college degrees, respectively. To these we add Nwhe, Sfemale, Ccy, and Elderly which are the percentage of non whe heads, single female heads, the percentage of households in the central cy, and elderly heads of households. The parameters Θ i and δ t are division and year specific effects and ε is an iid error term. Through 3 Previous researchers have principally used the percentage of the population that is black. Due to increases in other non whe ethnic groups a measure that is more inclusive of all non whe ethnic groups may hold more explanatory power. 6

7 pooling of division level data we are able to create the region data that is used in most of the regression results presented. This model is similar to those of Blank and Card (1993), Levernier (2003), and Formby et al. (2001). A slight difference in our model is the un of observation for regional poverty rates. 4 In previous work the regional poverty rate was defined as the weighted sum of the poor families divided by the estimated number of families whin each region. Because the CPS family weights are equivalent to the person weights for the family head, a family wh 10 people need not get greater weight in this calculation then a family consisting of a single individual. Our contention is that larger families should receive more weight. For this reason we use a headcount measure of the regional poverty rate whereby the poverty rate is computed as the estimated number of poor persons divided by the estimated population whin each region. As stated before, Figure 1 Panel B shows that divisions 5 (South Atlantic), 6 (East South Central), and 7 (West South Central) all had poverty levels above the national level. All other divisions had poverty levels below the national level. RESULTS Table 3 begins the econometric analysis by examining data for the entire Uned States. In each column of Table 3 the model specification varies by the inclusion of time and region fixed effects. The fullest and most stable of these specifications is presented in Column 4 which has both time and region effects included. The estimated coefficients on the variables match our a priori predictions concerning signs. As expected, unemployment, being elderly, households 4 See Blank and Card (1993) for an excellent discussion of the CPS and how defines family uns. 7

8 headed by single females, and households headed by non whe persons all had posive and significant estimated coefficients. 5 These results are consistent wh the lerature. In addion, wages, living in the central cy, and education had estimated coefficients that were negative and significantly related to poverty. The estimated coefficient on married heads had a negative sign but this variable was not significant at conventional levels. As stated earlier, we set out to investigate the impact of these demographic and economic factors on poverty in the South and determine whether they vary from the nation as a whole. We begin this part of the investigation wh Table 4 which separates the country into two subsets, namely the South and the rest of the country. The specifications presented in Table 4 are primarily those of Columns 1 and 2 of Table 3. We will focus our discussions of the results to Columns 2 and 4 in Table 4 which include time fixed effects. The most direct comparison of the South to the entire country would be to compare the specifications that do not include region fixed effects. As noted earlier, the results in this specification for Column 2 of Table 3 are not as robust as those that use the full specification (Column 4 of Table 3) and should be noted because of the obviously spurious results on married heads of households and residency in the central cy. In addion, the smaller sample size for the South specifications greatly reduce the degrees of freedom. Despe the smaller number of observations, the impact of the demographic variables on poverty in the South are consistent wh those of Levernier (2003). Focusing on differences that exist between variables that had statistical significance in both models reveals that the posive impact of unemployment on poverty is slightly higher for 5 Since official poverty measures use income to determine poverty status those elderly persons that might have sizable amounts of wealth but ltle income would be considered to be in poverty and influence the posive estimated coefficient seen in the latter two columns of Table 3. 8

9 the South. In addion, wages do not have as great an impact in the South and living in the central cy has a larger posive impact on poverty in the South than in the country as a whole. These comparisons between the country as a whole and the South region are revealing but might be skewed by the fact that the South is included in the estimates of the entire country. The last two columns of Table 4 show estimates when the rest of the country, excluding the South, are examined. In this way, we are able to separate out any results that might have been biased by the inclusion of the South region. In both Columns 2 and 4 of Table 4 there are several estimated coefficients that have the same sign and are statistically significant although the size of the estimated coefficients vary. These include the impact of unemployment, wages, and having a high school diploma. This information, in self is important given that public policy administered at the national level may impact the country as a whole in the same fashion but not at the same level of effectiveness. In other words, although unemployment is posively correlated wh poverty through out the country s impact is not uniform and appears to have a much larger impact on poverty in the South region. 6 There were differences eher in the sign of a statistically significant estimated coefficient or in the statistical significance in a few variables. These were the impact of college, being a non whe head of household, single female household heads, and being elderly. Once again, given the limations of small sample sizes we tested whether these differences were spurious or whether they constute significant differences in how these areas poor populations responded to changes in these variables. To that end, we employed an F-test, sometimes referred to as a Chow test. Results from the Chow tests are reported in Table 5. We 6 We ran the reported regressions at the division level but had a great deal of spurious results due to the limed number of observations. Still, even in these models, unemployment had a significantly stronger posive impact on poverty in the divisions of the South. 9

10 reconfigured the data so that a the break would occur at the separation of Southern and non- Southern regions. The Chow test then can be used to test whether a structural break is occurring and display s significance. The results of this test provide further evidence that some of the variables in question do not have the same impact on poverty rates in the South as in the rest of the county. Those variables that had statistical significance were the poverty rate self, households headed by single females, and households headed by non whes. It should be noted however, that this test does not say in which direction these differences are occurring nor does provide any insight into the magnudes of differences. POLICY IMPLICATIONS The policy implications of this paper are that national anti-poverty plans that are not necessarily designed for regional variations in poverty populations may loose effectiveness. What this paper shows is that even though some factors effect poverty populations in the same manner i.e. unemployment, is not clear that different regions are effected to the same degree. We were able to investigate whether these impacts are uniform by examining the differing impacts of certain variables on poverty rates at the regional level. Another important aspect of this work is that presents evidence that these differences in poverty populations appear to be somewhat systematic. This would imply that effective policy will need to have a more farsighted time horizon regarding expected results. Through analysis and testing we show that certain economic and demographic factors impacted poverty significantly in both the South as well as the rest of the country but not to the same degree. In addion, we provide insight that reveals that households headed by women and non whes are impacting poverty in the South in ways that are not the same as the rest of the country. Policy makers looking to reduce poverty in this region should be aware of this fact in 10

11 developing effective policy. Our speculation would be that greater investments in human capal for these effected subgroups would help in reducing poverty for the region. CONCLUSIONS This paper explores several questions relating to national poverty and adds to the growing lerature on regional poverty. First, we explore the impact of key economic and demographic factors on poverty over the last 20 years. We investigate the impact of unemployment and movements in unemployment over the period and confirm that the labor market is a significant condu by which poverty is affected. In addion, we were able to show the impact that other economic factors, such as wages, have on poverty wh results that are consistent wh the previous lerature. We finally explore differences in responses of regional poverty populations to changes in certain economic and demographic factors. 11

12 REFERENCES Becker, G. (1964). Human Capal and the Goal Distribution of Income: an Analytical Approach, New York: Columbia Universy Press Beeghley, L. (1988). Individual and Structural Explanations of Poverty Population Research and Policy Review, 7(3), Blank, R.M. (1991). Why were poverty rates so high in the 1980s? Working Paper No. 3878, National Bureau of Economic Research. Blank R. and David Card (1993). "Poverty, Income, and Growth: Are They Still Connected?" Brookings Papers on Economic Activy, 2 (Fall). Bishop, J.A., J.P. Formby and P. D. Thistle (1992) Convergence of the South and Non-South Income Distributions, , American Economic Review, 82(1), Cro, C.F. and R.T. Michael, (1995). Measuring poverty: A new approach (National Academy Press, Washington D.C.). Freeman, D., (2003). Poverty and the Macroeconomy: Estimates from U.S. Regional Data, Contemporary Economic Policy, 21(3), Freeman, R., (2001). The Rising Tide Lifts? National Bureau of Economic Research, Working Paper Formby, J., Hoover, G., and Kim, H., (2001). Economic Growth and Poverty in the Uned States: Official Statistics versus Sen's Index of Poverty, Journal of Income Distribution, 10 (3-4), Heckman, Lochner, Smh, and Taber (1997), The effects of government policy on human capal investment and wage inequaly, Chicago Policy Review, Vol. 1, pp LeBlanc, M., (2000). Poverty, Policy, and the Macroeconomy, Uned States Department of Agriculture, Technical Bulletin Levernier, W. (2003). An Analysis of Poverty in the American South: How are Metropolan areas different from Nonmetropolan Areas? Contemporary Economic Policy, 21(3), Moen, J. R. (1989). Poverty in the South, Federal Reserve Bank of Atlanta Economic Review, 8(3), Sawhill, I.V. (1988). Poverty in the U.S.: Why is so Persistent? Journal of Economic Lerature, 26(3),

13 Zheng, B. (1997). Aggregate poverty measures, Journal of Economic Survey, Vol. 11, pp

14 TABLE 1 Geographical Regions and Divisions Northeast Division 1 (New England ) Division 2 (Middle Atlantic ) Midwest Division 3 (East North Central) Division 4 ( West North Central) South Division 5 (South Atlantic) Division 6 (East South Central) Division 7 (West South Central) West Division 8(Mountain) Division 9 (Pacific) ME, NH, VH, MA, RI, CT NY, NJ, PA OH, IN, IL, MI, WI MN, IA, MO, ND, SD, NE, KS DE, MD, VA, WV, NC, SC, GA, FL KY, TN, AL, MS AR, LA, OK, TX MT, ID, WY, CO, NM, AZ, UT, NV WA, OR, CA, AK, HI 14

15 TABLE 2 Simple Means By Division US Div 1 Div 2 Div 3 Div 4 Div 5 Div 6 Div 7 Div 8 Div 9 Poverty Status High School College Unemployment Log Wages Income Female Non Whe Married Central Cy Elderly

16 TABLE 3 Estimates of Economic and Demographic Factors on National Poverty Status Independent Variables (1) (2) (3) (4) Intercept *** 0.222** 0.266*** (1.29) (2.88) (2.31) (2.88) Unemployment 0.394*** 0.414*** 0.496*** 0.335*** (7.83) (6.28) (11.81) (5.85) Log Wages *** *** *** *** (-13.40) (-9.29) (-3.46) (-4.58) Married Head 0.403*** 0.236** (3.79) (2.45) (0.30) (-0.57) High School *** *** ** (-4.91) (-0.94) (-4.56) (-2.21) College *** *** (-5.01) (-1.04) (-5.42) (-1.13) Non Whe 0.085*** 0.071*** *** (4.33) (4.12) (0.87) (2.83) Single Female ** 0.176* (0.10) (1.55) (2.54) (1.78) Central Cy *** * * (-0.01) (2.99) (-1.81) (-1.76) Elderly * 0.188** (-0.47) (-0.29) (1.67) (2.19) Year Effects No Yes No Yes Region Effects No No Yes Yes Adjusted R # of Obs ***,**,* denote significance at the 1%, 5%, and 10% levels, respectively. 16

17 TABLE 4 Estimates of Economic and Demographic Factors on Regional Poverty Status (1) (2) (3) (4) Independent Variables South South Rest of US Rest of US Intercept *** (-0.80) (0.35) (0.02) (2.64) Unemployment 0.314*** 0.471*** 0.431*** 0.362*** (4.70) (5.33) (7.00) (4.40) Log Wages *** * *** *** (-5.43) (-1.89) (-3.75) (-9.64) Married Head 0.478*** ** (3.16) (1.65) (2.34) (0.28) High School *** * *** (-4.41) (-1.69) (0.55) (-2.93) College *** *** (-4.68) (-2.75) (-0.79) (0.82) Non Whe *** 0.250*** (-0.68) (0.54) (6.36) (10.34) Single Female 1.078*** 0.736*** (5.08) (3.35) (-1.18) (0.25) Central Cy 0.174*** 0.181*** ** *** (6.57) (7.08) (-2.10) (-3.45) Elderly * 0.166* (-0.68) (0.62) (1.91) (1.93) Year Effects No Yes No Yes Region Effects No No No No Adjusted R # of Obs the 1%, 5%, and 10% levels, respectively. ***,**,* denote significance at 17

18 TABLE 5 Chow Test Variable Wh Intercept Whout Intercept College (0.2635) (0.2869) Non Whe (0.0020) (0.0001) Single Female (0.0103) (0.0201) Elderly (0.8936) (0.7856) Poverty Status (0.0223) (0.0225) Note: p-value (significance level for two-tail test) in 18