Hostname: page-component-74656f9884-rdgjk Total loading time: 0 Render date: 2026-11-08T13:32:27.630Z Has data issue: false hasContentIssue false
  • English
  • Français

Human Capital and Growth in the Postbellum South: A Separate but Unequal Story

Published online by Cambridge University Press:  09 June 2004

MICHELLE CONNOLLY
MICHELLE CONNOLLY
Affiliation:
Assistant Professor, Department of Economics, Duke University, 213 Social Sciences, Box 90097, Durham, NC 27708. E-mail: connolly@econ.duke.edu
Rights & Permissions [Opens in a new window]

Abstract

This article tests the importance of human capital in explaining convergence across the states from 1880 to 1950. Human capital matters to a state's income level and to its growth rate through technological diffusion. The South, whose overwhelmingly agricultural society relied more heavily on work experience than formal education, and whose racial discrimination in school resource allocation lowered human capital accumulation of both blacks and whites, presents a unique pattern. The South's low human capital levels following the Civil War and its active postbellum resistance to education reduced its speed of conditional convergence toward the rest of the nation.

Information

Type
ARTICLES
Information
The Journal of Economic History , Volume 64 , Issue 2 , June 2004 , pp. 363 - 399
Copyright
© 2004 The Economic History Association

Both the convergence literature on regions within the United States and the literature on the evolution of the Southern economy have highlighted the need for an explanation of the slow convergence of Southern per capita income to that of the rest of the United States after the Civil War.

Kris Mitchener and Ian McLean identify the West and the South as key regions in explaining the convergence pattern of the United States from 1880 to 1980.1

Mitchener and McLean, “U.S. Regional Growth.”

The West began as the highest income region due to resource abundance and recent settlement. The South began as the lowest income region due to the negative effects of the Civil War and slavery. Together, these two regions accounted for a majority of the initial income inequality in the United States in 1880. Much of the initial convergence in income in the United States after 1880 is accounted for by the West, which simply grew more slowly, thereby allowing other regions to catch up to it from 1880 to 1940.2

Ibid., p. 1021.

Conversely, Mitchener and McLean find that the South did not contribute significantly to convergence in the United States until after 1940.3

Of the observed convergence between Southern and Northern average service income between 1880 and 1950, Caselli and Coleman (“U.S. Structural Transformation”) show that up to 81 percent is attributable to structural transformation. For the 1940 to 1990 period, structural transformation is empirically less important than in the earlier period, but still accounts for approximately 57 percent of the measured convergence between the two regions. The remainder is explained by within-sector North-South wage convergence. Caselli and Coleman therefore focus on modeling the Southern structural transformation caused by labor movement out of agriculture and the subsequent rise in agricultural wages relative to manufacturing wages. This insightful model is able, in positive terms, to explain the role of structural transformation in the observed pattern of service income convergence. The key assumption generating this structural transformation is a long-run decline in the relative costs of acquiring nonagricultural skills in both regions.

Further, they suggest that “…the slow catch-up of the South (relative to the national average) appears attributable to changes in productivity rather than to price or labor input effects.”4

Mitchener and McLean, “U.S. Regional Growth,” p. 1030.

In other words, price or gender and age characteristics, which the literature has found are important in explaining convergence patterns both across regions and across countries, are not sufficient to explain the convergence pattern of Southern states to the rest of the nation after the Civil War.5

Mitchener and McLean, “U.S. Regional Growth”; Williamson, “Growth”; and Williamson and Lindert, American Inequality.

Instead, one must explain the relatively low productivity in the South in the postbellum period as well as gradual convergence in productivity to that of the rest of the nation to explain Southern convergence.6

Margo, “North-South Wage Gap.”

Clearly, many factors contributed to this relatively low productivity.7

A wide range of factors have been proposed to explain the South's relative stagnation in the postbellum period. Wright (“Cotton Competition”) focuses on poor world demand for cotton, Ransom and Sutch (One Kind) on the reduced agricultural labor supply, and Carlton and Coclanis (“Capital Mobilization”) on the lack of adequate capital for industrialization.

Within a significant literature focusing on this question, Gavin Wright stresses the lack of a sufficiently skilled labor force and generally low levels of education as major obstacles to the process of industrialization in the South.8

Wright, Old South.

In this article, I focus empirically on human capital levels to explain the relatively low productivity in the postbellum South. To accomplish this, I create both education-based and experience-based decadal human capital estimates for 48 states of the United States from 1880 to 1950. The education-based measure controls not only for interstate migration but also for relative price levels across states. The importance of human capital, particularly in explaining the lack of Southern convergence from 1880 to 1950, can be seen in Figure 1. There we see the strong positive contemporaneous relationship between education-based human capital per worker and income per worker in the North Atlantic (NA), North Central (NC), South Atlantic (SA), South Central (SC) and Western (W) states of the United States in 1880, 1900, 1920, and 1950.9

The regions are defined as follows: North Atlantic: Connecticut, Maine, Massachusetts, New Hampshire, New Jersey, New York, Pennsylvania, Rhode Island, and Vermont; North Central: Illinois, Indiana, Iowa, Kansas, Michigan, Minnesota, Missouri, Nebraska, North Dakota, Ohio, South Dakota, and Wisconsin; South Atlantic: Delaware, Florida, Georgia, Maryland, North Carolina, South Carolina, Virginia, and West Virginia; South Central: Alabama, Arkansas, Kentucky, Louisiana, Mississippi, Oklahoma, Tennessee, and Texas; and Western: Arizona, California, Colorado, Idaho, Montana, Nevada, New Mexico, Oregon, Utah, Washington, and Wyoming.

We also see that the conditional convergence of the South Atlantic and South Central regions appears to have depended heavily on convergence in their human capital levels towards those in the rest of the United States.

FIGURE 1 EDUCATION-BASED HUMAN CAPITAL

Given the strong relationship present in Figure 1, I test the contribution of human capital to both income levels and growth in 48 states of the United States from 1880 to 1950. Controlling for possible reverse causality, I find that along with physical capital, a state's human capital stock significantly contributes to both the level and growth of its income. Moreover, I am able to consider the theory that racial discrimination in Southern education was a primary contributor to the low levels of human capital, not only for Southern blacks, but for Southern whites as well.

GROWTH THEORY

Robert Solow's neoclassical growth model predicts convergence in income per capita among countries conditional on identical production functions, savings rates, and labor force growth rates, without requiring factor mobility.10

Solow, “Contribution.”

Robert Barro and Xavier Sala-i-Martin show that the South did conditionally converge in personal income to the rest of the nation from 1840 to 1963.11

Barro and Sala-i-Martin, “Convergence across States” and “Convergence.”

However, the question central to this article concerns the speed with which the South converged with the rest of the nation, and how Southern educational policies affected the steady-state income level to which the South was converging.

The answer to this question may lie in human capital differences between the states. In human capital models of endogenous growth, growth is driven by the accumulation of human capital, broadly defined as an individual's skill level accumulated through formal education or through on the job learning-by-doing.12

Uzawa, “Optimum Technical Change”; Lucas, “On the Mechanics”; and Romer, “Increasing Returns” and “Endogenous Technical Change.”

If there are any positive externalities due to human capital that are not taken into account in the individual's utility maximization, market equilibrium levels of investment in human capital will fall short of socially optimal levels.

Because human capital accumulation is the driving force in these models, absolute convergence will not necessarily occur between countries with different initial levels of human and physical capital, unless factor mobility forces convergence. Between 1880 and 1900, Southern income per capita grew at about the national rate. It did not greatly converge with the rest of the nation until after the turn of the century, when labor mobility increased and educational and skill differentials began narrowing.13

Wright, Old South.

A majority of the convergence actually occurred after World War II. Thus, while both the South and the nation grew at approximately the same rate until the 1900s, level differences were roughly maintained, due to level differences in human capital. Southern labor market segmentation and low levels of education and industrial skill in the postbellum South did little to improve the South's human capital stock and had much to do with the slow convergence of the South to national standards.14

Wright, ibid.; and Rosenbloom, “One Market” and “Was There.”

HISTORICAL SETTING

Four factors contributed to the South's emergence at the turn of the century as a low-wage, low-skill region specializing in labor-intensive industries. They include the South's plantation legacy, its active resistance to educating its workforce, both black and white, the relative isolation of its labor market, and “backwash” effects of being a latecomer to industrialization.

The legacy of slavery and reliance on a plantation economy left the South without a significant industrial structure or labor force capable of quickly adapting to industrialization after the Civil War. David Weiman argues that slavery and labor-intensive plantation technology in the South resulted in both reduced investment in physical capital and a reduced labor supply for prospective Southern manufacturing industries prior to the Civil War.15

Weiman, “Staple Crops.”

Moreover, Wright and Bess Beatty point to the lack of an indigenous technological community to explain the South's reliance on Northern technology and its inability to quickly industrialize after the Civil War.16

Wright, Old South; and Beatty, “Lowells.”

The plantation legacy also left the South with very little human capital relative to the rest of the nation immediately after the Civil War. The reasons for this are quite straightforward. By 1835, there existed a uniform legal proscription across the South against the formal education of slaves as a result of slave insurrections earlier in the nineteenth century. Consequently, 95 percent of the black Southern population was completely illiterate at the beginning of the Civil War. Furthermore, the lagged effect of older generations tended to slow educational improvements for younger generations.17

Smith, “Race”; and Margo, Race.

As Robert Margo explains, children of illiterate parents not only could not seek help from their parents when learning to read and write, but also tended to be drawn away from school to help their families on the farm or in the market.18

Using data from the 1910 Census, Margo (Race and Schooling) finds that school attendance rates rose 4.2 percent with a 10-percent reduction in adult illiteracy.

Figure 2 shows the strong racial divide in literacy rates within the South continuing into the 1930s.

FIGURE 2 SOUTHERN LITERACY (percentage of the population over the age of 10)

Moreover, although state funds were allocated on a per pupil basis independently of race, within the segregated schooling system (and aided by disfranchisement after 1877), states diverted funds at the local district level from black schools to white schools.19

Bond, Education; Margo, “Race, Educational Attainment” and Race; and Gerber, Southern White Schooling.

Because redistribution occurred at the county or school district level, the severity of redistribution depended on the concentration of blacks in that county. In a county with fewer blacks, fewer resources were available for diversion to white pupils. Conversely, in predominantly black counties, significant funds could be redirected towards relatively few white students. For example, in white counties in Mississippi in 1907, $3.50 per member of the school age population was spent on black children relative to $5.60 on white children. In black counties, $2.50 was spent on black children, versus $80.00 on white children.20

Gerber, Southern White Schooling, pp. 9, 13.

This discrimination continued well into the 1930s, as seen in the ratio of school expenditures per black pupil relative to the total state allocation per pupil in Southern states (Table 1). The 12 Southern states on average spent only 37 percent of what should have been spent on black students had an equal distribution of educational expenditures been made across all students regardless of race.21

Bond, Education.

Black schools in the South were marked by lower teacher salaries, higher student to teacher ratios, shorter terms, and lower educational levels of teachers.22

For example, in 1911 in Georgia, black teachers earned less than half of what white teachers earned, black schools generally had twenty more students per class than white schools, the school term was three to four weeks shorter and only 16 percent of black teachers had at least two years of postsecondary education, compared to 35 percent of white teachers (Donohue, Heckman and Todd, “Schooling,” p. 229).

John Donohue, James Heckman, and Petra Todd find that absolute improvements in the quality of Southern black schools occurred from 1910 to the mid 1930s (due largely to Northern philanthropy).23

Donohue, Heckman and Todd, “Schooling.”

Still, there were few gains relative to white schools, except in attendance rates. Conversely, from the late 1930s to 1960, there were both absolute and relative improvements in black school quality in the South (due primarily to legal actions on the part of the NAACP). Looking at Southern-born men born between 1900 and 1949, David Card and Alan Krueger find that these improvements in the relative quality of black schools explain 20 percent of the reduction in the overall black-white earnings gap between 1960 and 1980.24

Card and Krueger, “School Quality.”

In the prewar era, there is ample evidence of selective migration of more educated blacks out of the South.25

Margo, “Schooling” and Race; and Vigdor, “Pursuit.”

This is evident in Table 2, which shows that Southern-born blacks with above average education levels for their birth and state cohort were between two to almost five times more likely to have left the South then their peers with below average education levels.26

Vigdor, ibid.

Using U.S. Census microdata on blacks, Jacob Vigdor demonstrates positive intergenerational effects between the education level of parents who migrated between 1940 and 1970 and educational attainment of their children in 1970 and 1990.27

Vigdor, ibid.

Although his study focuses on a later time period than that of this article, Vigdor's evidence of positive intergenerational effects, as well as James Smith's and Margo's evidence of intergenerational effects in earlier periods, suggest that selective migration patterns of blacks from the South not only led to an immediate drop in black human capital levels in the South, but also had compounding intergenerational effects because the South lost the positive intergenerational effects that more educated blacks would have yielded had they remained in the South.28

Smith, “Race and Human Capital”; and Margo, Race.

The large fraction of the labor force composed of blacks (about 46 and 58 percent of the South Central and South Atlantic populations respectively in 1890), combined with severe racial discrimination in schooling and selective migration of more educated blacks out of the South, were sufficient to lower Southern human capital levels relative to the rest of the nation.29

Abstract of the Twelfth Census, 1900, (1904), p. 41.

Still, educational levels, even for whites, were much lower in the South than in the North. Of Southern whites born between 1870 and 1880, fewer than 20 percent ever received a high school diploma and only 30 percent ever went beyond elementary school.30

Smith, “Race and Human Capital.”

Even after the Civil War, large North-South educational differences were maintained and even increased as a result of Southern educational policies. This is evident in the divergence in North-South schooling levels for both races after 1870 (Figure 3).

FIGURE 3 MEAN SCHOOLING LEVELS IN MONTHS

Even aside from issues of racial discrimination, the South had a historical legacy of low education norms. James Gerber points to the lack of property tax use for public education to explain the limited resources for public education in the South prior to Reconstruction.31

Gerber, Southern White Schooling.

So while most of the United States was moving towards public education in the antebellum period, the South stood apart from this movement. The Reconstruction period in the South led to several improvements in the education system in the South between 1865 and 1877. They included the establishment of school revenue systems based primarily on state land taxes, laws to establish centralized state administrations for the schools, and the mandate that both black and white students have access to schools. Unfortunately, after the Reconstruction period a period of backlash occurred, severely curtailing support for public education, including legislation in many Southern states that prohibited local taxation for schools.32

Ibid., pp. 5–8.

Using 1940 U.S. census data, Gerber finds that between 20 and 25 percent of lower individual incomes of all Southern men were attributable to lower educational attainment.33

Ibid.

When considering only white men, education accounted for 5 to 17 percent of the lower individual income levels in Southern states.34

Ibid, pp. 30, 33–34.

Hence, whereas much of the 1940 income differences due to educational disparities were borne by Southern blacks, Southern whites also suffered from lower educational attainment.

Still, it could be argued that the widespread discrimination against blacks in the South may have resulted in depressed support for public education in general because of the inability of lower-class whites to join political forces with blacks in support for public education. Gerber explains:

In the case of school finance, the gulf between black and white Southerners was exacerbated by the racist propaganda that attempted to convince lower-class whites that state school revenues derived from white citizens were going to black schools. Given the inadequacy of white schools in many areas, this argument raised racial tensions and prevented, or at least hindered, a unified front of poor whites and blacks from demanding better schools. Those whites who benefited saw no need to support campaigns for higher taxes for schools: they could always raid the funds intended for black schooling when they desired more revenue…. Even though this type of discrimination was aimed ostensibly at the black population, it may have affected lower-class whites as well, much in the same manner that disfranchisement removed the vote from significant numbers of poor and illiterate whites.35

Ibid, p. 99.

Along these lines, Gerber finds that, for Southern states, the higher the percentage of blacks in the state population, the higher the level of white schooling inequality.36

This was not the case for non-Southern states. Gerber, ibid., pp. 115, 126.

Using the education-based measure of human capital per worker, it is possible to indirectly consider the hypothesis that, within the South, higher percentages of blacks in the population would have led to easier redistribution of school resources towards white pupils, and therefore to lower incentives for politicians to argue in favor of greater overall resource allocation towards public education. Because the education-based measure of human capital is based on school expenditures, it lends itself well to this question. I therefore run a state-level panel regression for 1880, 1900, 1920, and 1950 of education-based human capital per worker on income levels and the percentage of the state population that was black.37

Data on the percentage of state populations that are black come from Gibson and Jung, “Historical Census Statistics.”

If higher percentages of the population being black led to greater opportunities for diversion of schooling resources and hence lower incentives to raise overall schooling expenditures, this discrimination variable should enter negatively in Southern states.

This admittedly simple regression is intended to see if there is support for the notion that racial discrimination contributed significantly to lower educational standards for the average Southerner (rather than just for the average black Southerner). Because state income is clearly a primary determinant in resources potentially available for education, it is included to control for the fact that Southern states were also lower income states. The results from this simple regression are intriguing. Column 1 of Table 3 presents the results using fixed effects (FE) estimation for all states of the United States.38

FE estimation treats unobservable latent individual effects as fixed and focuses on deviations of states over time from their individual means. This is identical to having a dummy for each state in the regression. Further descriptions of this estimation technique are provided in a later section.

We see first that higher state incomes did indeed lead to higher state human capital levels. Still, controlling for income, a higher percentage of blacks in the state population (implying greater opportunities for discriminatory school resource allocation) led to lower human capital levels. Separating the coefficient on this discrimination variable into a general component and the marginal change when considering a Southern state (see column 2), we see that the finding of column 1 is purely driven by the Southern states. Specifically, the marginal contribution to this elasticity when a state is in the South is –0.117 and highly significant, whereas the general component is not statistically significant. Hence, the finding that higher percentages of blacks in the population led to overall lower levels of human capital (for the average pupil, and not just the black pupil) holds only in the South.39

When the regression is run only for the South, the discrimination variable is statistically significant and negative, whereas when run only for non-Southern states, the discrimination variable is not significant.

These results support the argument that racial discrimination occurring at the local level in Southern schooling led not only to lower human capital levels for Southern blacks, but also for Southern whites through its depressing effects on overall support for public education.

Finally, there was strong opposition from both industrial employers and planters to educating the common laborer, whether black or white, for fear that educated workers would leave the South.40

Wright, Old South.

The mere fact that the South was still primarily a plantation economy within a traditional (versus modernizing) environment led to active resistance to education. A traditional environment is defined as using primitive technology or traditional farming practices and crops and either little innovation or little exposure to innovation.41

Evidence, such as the slowness with which the South adopted mechanization in cotton, suggests that the Southern plantation system was relatively traditional and not modernizing relative to the rest of the nation, even compared to other agricultural regions (Gerber, Southern White Schooling).

Consequently, among the plantations there was high demand for unskilled labor, little demand for skilled labor, and a fear that increases in education would drive workers out of the plantation sector, thus threatening its labor supply.42

Gerber, Southern White Schooling, pp. 91–93.

Although school expenditures always rely upon the contemporaneous economic situation, this negative attitude towards education driven by employer pressures and racial discrimination was largely responsible for the fall in real Southern school expenditures per pupil in the post-bellum period (Figure 4). No concerted effort was made to change the quality of the Southern educational system at least until the turn of the century. There was significant improvement in the percentage of school-aged children in the South who attended school.43

U.S. Bureau of Education, Report, 1893, p. 36, and 1911, p. 694.

Still, these students were attending schools with shorter school terms and less well trained teachers. The relative scarcity of skilled labor in the South, as well as the isolation of the Southern labor market (particularly for unskilled workers), is reflected in the large and increasing North-South real wage gaps for unskilled workers between 1880 and 1914.44

Wright (Old South) argues that these wage gaps for unskilled labor were not merely the result of racial discrimination. Although racial discrimination occurred, and racial wage gaps existed, the North-South wage gap for unskilled labor was significantly larger than the racial gap.

FIGURE 4 REAL ANNUAL SCHOOL EXPENDITURE PER PUPIL (1967 dollars)

A final consideration for the South during this period is its position as a periphery region relative to the core of the North. Specifically, the appearance of a more national product market during this time period offered new opportunities, but also placed the South in a latecomer position relative to the North and even the West. One can view the Southern experience of industrialization as suffering from both Albert Hirschman's “polarization” and Gunnar Myrdal's “backwash” effects, whereby in early periods of development market forces accentuate initial disparities across regions.45

Hirschman, Strategy; and Myrdal, Economic Theory.

A core region develops initially and becomes well endowed with skilled labor and capital. The high productivity of the core relative to the periphery leads to further reallocation of skills and capital towards the core and away from the periphery (polarization) and overtaking of the national market by more efficient producers in the core (backwash).46

Williamson, “Regional Inequality”; Carlton, “Revolution”; and Carlton and Coclanis, “Capital Mobilization.”

As Jeffrey Williamson suggests within the context of cross-country patterns of initially increasing regional inequality, the take-off of one region often leads to the selective interregional migration of the more skilled, educated, entrepreneurial and young, and to reallocation of capital towards the core region.47

Williamson, “Regional Inequality.”

This is due to agglomeration effects in the core, in contrast to high risk premiums, a lack of entrepreneurial abilities, and immature capital markets in the periphery. Only later in the development process do internal factor flows (including technological flows, greater and less selective labor migration, and the development of more efficient capital markets) occur sufficiently to offset the polarization and backwash effects that tend to increase regional inequality in earlier stages.48

Ibid., pp. 5–9.

From this perspective, the South of the United States may have been following a more general regional development pattern observed in many countries.

Evidence of these effects was present in the South, as demonstrated by selective migration patterns for blacks that later become less selective, and selective choices for industrialization. David Carlton describes the Southern choices for industrialization as limited by the lack of entrepreneurial expertise and labor skills, leading the region “…to compensate by developing or attracting industries at advanced stages in the product cycle, industries in which skills have been largely ‘built in’ to their basic technology and structure.”49

Carlton, “Revolution,” p. 473.

This also implies that capital-goods industries will locate in areas with a sufficient stock of skilled labor and demand for specialized products. In turn, once situated, these industries will attract and expand the local mass of skilled workers.50

Carlton and Coclanis, “Capital Mobilization.”

Using patent data, Carlton and Peter Coclanis try to explain the relative lack of innovating activity in the South.51

Ibid.

Controlling for urbanization, the percentage of the workforce in capital-goods manufacturing, and school attendance, they find that the South did not generally appear to have a distinct cultural pattern separating itself from the nation in terms of inventiveness. Moreover, they emphasize that the South's education gap and patenting gap were reflections of these backwash effects: “If the South, as a technologically backward region beset by ‘backwash’ effects, chose to industrialize by importing its technology, thus minimizing its investment in invention, the region also tended to choose technology that would minimize the need to develop the skills of its workforce.”52

Carlton and Coclanis, “Uninventive South,” pp. 321–22.

In line with the theories of polarization and backwash effects, Sukkoo Kim empirically demonstrates the importance of factor proportions to the location of industries in the United States.53

Kim, “Expansion.”

He shows that changes in the relative mobility of factors and changes in scale economies can explain U.S. patterns of regional specialization in manufacturing. As transportation costs fell in the late nineteenth and early twentieth centuries, firms adopted large-scale production that was intensive in relatively immobile resources. As a consequence, regions became more specialized. However, as factors of production later became more mobile, regional factor proportions became more similar, leading to reduced specialization after World War II. In turn, Kim demonstrates that differences in regional industrial structures played a key role in the patterns of U.S. regional income divergence and convergence between the nineteenth and twentieth centuries.54

Kim, “Economic Integration.”

In other words, growing economic integration in the United States initially led to greater regional specialization in manufacturing in the mid-nineteenth and early twentieth centuries, and then with greater factor mobility, the trend reversed itself in the second half of the twentieth century.55

Ibid., pp. 660–61.

With greater differences in factor proportions, initial divergence in industrial structures led to initial divergence in income levels that were later reversed as factor mobility increased.56

Kim notes that the primary cause of income divergence in the U.S. between 1840 and 1900 was the relative decline in Southern per capita income “…caused by the region's growing unfavorable industry-mix and lower wages relative to other regions.” He calculates that Southern income per capita would have risen by as much as 20 percent had its industrial structure converged towards that of the national average (Kim, ibid., p. 672).

Both issues—low investment in education, primarily for blacks but also for whites, and core and periphery—rely on increased labor mobility for final resolution of the inequalities across regions. Hence, greater levels of labor movement between the South and the rest of the nation were necessarily part of the final conditional convergence of the South to national norms.

A key part of the migration story relies on the migration pattern of Southern blacks. Immediately after emancipation blacks began to shift locations, but the movement was generally local (i.e., within the county of origin or between contiguous counties).57

To the extent that there was longer distance migration of blacks it tended to occur in a westerly direction within the South (Higgs, Competition, pp. 24, 26).

Hence, there was general stability in the regional distribution of the black population from 1865 through 1914.58

In both 1860 and 1910 approximately nine-tenths of the black population lived in the South. This held despite positive net migration of blacks to the North because of higher fertility and lower mortality in the South than in the North (Higgs, Competition, p. 28).

Only after the drastic reduction of foreign immigration to the United States beginning around 1920 and peaking in the late 1920s and early 1930s did employment opportunities in the North increase for blacks, leading to greater (nonselective) black migration northward.59

On immigration, see Lebergott, Manpower, pp. 29, 163; and Easterlin, Population, pp. 187–88. On black migration, see Higgs, Competition, p. 26.

Marshall Colberg further argues that it is the exporting of abundant unskilled labor and the importing of scarce educated labor in the 1940s and 50s that led to the final convergence of the South with the rest of the nation.60

Colberg, Human Capital.

In summary, the plantation legacy, combined with low educational standards, labor market isolation, and periphery effects, left the South to industrialize almost completely without the aid of an indigenous technical community. Consequently, Southern firms were dependent upon Northern technology, but lacked a sufficiently educated labor force to aid in the adoption or possible adaptation of Northern technology to Southern needs.61

For striking examples of this in the textile industry see Beatty, “Lowells.”

As suggested by Richard Nelson and Edmund Phelps, human capital increases the rate at which existing technology can be applied for practical uses.62

Nelson and Phelps, “Investment.”

Hence, if Southern workers' educational levels had been higher, the speed of adoption of Northern technology in the South, as well as the general productivity of the workers, would likely have been greater. The ultimate result of Southern educational policies was the emergence of the South as a low-wage, low-skill region, characterized by labor-intensive industries with low value-added.63

Wright, Old South.

In turn, specialization in low-value-added industries led to less human capital accumulation by the workers employed in Southern industries.

GROWTH ACCOUNTING

The regressions undertaken here draw from growth accounting. I first consider level regressions similar to those of Gregory Mankiw, David Romer, and David Weil and then comparable growth regressions along the lines of Jess Benhabib and Mark Spiegel's. 64

Mankiw, Romer, and Weil, “Contribution”; and Benhabib and Spiegel, “Role.”

Growth accounting regressions are generally based on the following Cobb Douglas production function:

where Y is real state income, A is the technology level, K is the real capital stock, H is the human capital stock, and L is the number of employed workers in the state. In per worker terms, equation 1 becomes

where all lower case letters are the original variables expressed in per worker terms. In natural logarithmic form, for each state, i, equation 2 becomes

This equation likely neglects certain unobservable state specific effects. Hence, the regressions to follow test equation 3 with a one-factor error term, μit, where μit=αi+εit, i=1…48, and t=1…4. Following Jerry Hausman and William Taylor, αi represents an unobservable latent individual state effect.

65

Hausman and Taylor, “Panel Data.”

The αi are assumed to be time-invariant, and independently distributed across individual states with variance, σ2α. The εit are assumed to be identically, independently distributed with zero mean and constant variance, σ2ε, conditional on the explanatory variables. While the εit are assumed to be uncorrelated with the explanatory variables, there may be correlation between the latent individual effects, αi, and the explanatory variables kit and hit. For example, the capital stock of a state should vary according to the type of industry present in the state, in addition to depending on whether the state is primarily agricultural or industrial. Similarly, a state's culture and policy environment affect funding and administrative decisions for public schools. Since such state specific characteristics are not included as independent variables, their effects will be captured in the latent individual effects, αi, and are likely to be correlated with the independent variables included in the regression. Hence, there is an a priori reason to think that fixed effects estimation is the appropriate specification.66

Estimation by Ordinary Least Squares (OLS) assumes no latent individual effects. Hence, OLS will be inefficient if such effects are present. Fixed effects (FE) estimation treats latent individual effects as fixed, focusing on deviations of states over time from their individual means. FE estimation also yields inefficient estimates. However, by treating the αι as fixed, the FE estimates will be unbiased and consistent regardless of whether or not there is correlation between the individual effects and the explanatory variables. Random effects (RE) estimation allows for random latent individual effects and represents a weighted average of both cross-sectional and within-group variance. In the absence of correlation between the αι and the explanatory variables, RE estimation will be BLUE. If however, such correlation is present, then its results will not be consistent. Because such correlation affects FE and RE estimators differently, differences in the estimated coefficients suggest possible correlation (Hausman, “Specification Tests”). Accordingly, rejection of the null hypothesis of no correlation suggests that the FE estimates are the only consistent estimates.

In addition to the basic relationship in equation 3, growth regressions are run on equation 3 using natural log differences. These growth regressions are further supplemented following a methodology similar to that of Benhabib and Spiegel.

67

Benhabib and Spiegel, “Role.”

They specify a model, based on the Nelson and Phelps model of technological diffusion, in which total factor productivity growth depends on the level of human capital in a country.68

Nelson and Phelps, “Investment.”

Specifically, Nelson and Phelps argue that human capital increases the rate at which applied technology catches up with theoretical knowledge. This concept can easily be adapted to consider the diffusion of technology from more advanced states to less advanced states within the United States. I therefore adopt the structural specification of Benhabib and Spiegel, in which the growth rate of total factor productivity in state i is

where c reflects exogenous technological progress, g(hi) reflects endogenous technological process within the state, and m[hi(ymax−yi)/yi] embodies the notion of technological diffusion or catch-up to the most advanced state. Both g([bull ]) and m([bull ]), are nondecreasing functions of hi. We can regroup the hi terms and write equation 4 as

The catch-up term in equation 4′ should depend on relative technology levels, rather than relative income levels. However, Benhabib and Spiegel use y as a proxy for A in the catch-up term, as the technology level is inherently difficult to measure. The technology level is even more difficult to quantify for individual states from 1880 to 1950. I therefore also use y as a proxy for A in the diffusion term. Applying equation 4′, the growth accounting specification with catch-up becomes

where I use natural logs for the level of hi and the catch-up term.

DATA

The model is tested using panel data from 48 states in 1880, 1900, 1920, and 1950. These are the only years for which both capital stock and income data are available by state. Data on each state's capital stock (except for 1950), workforce, and personal income for these years are from estimates by various authors in Population Redistribution and Economic Growth: United States, 1870–1950, edited by Simon Kuznets and Dorothy Swaine Thomas. I create a 1950 capital stock estimate to supplement the existing series using data from various issues of the Annual Survey of Manufactures, and from works by Thomas Romans and John Kendrick.69

Romans, Capital Exports; and Kendrick, Productivity Trends.

These series are described in the Appendix.

I also create two estimates for the human capital stock in each state from 1880 to 1950. The first reflects human capital accumulated through formal education, and the second reflects human capital accumulated through work experience. I first consider the two types of human capital separately and then interact the two measures to create a more comprehensive human capital term.

Formal education can be measured in terms of quantity of education (for example, years of education) or in terms of quality of education (i.e., school expenditures, teacher salaries, teacher education, and teacher-pupil ratios). However, Margo argues that yearly school attainment data before 1910 are biased because of the interpretation that a year of schooling in an ungraded Southern school, with shorter school terms and less trained teachers, was measured as being equal to one grade level in the rest of the nation.70

Margo, “Race and Human Capital.”

Furthermore, as summarized by Ronald Rizzuto and Paul Wachtel, the existing schooling literature has found that the quantity and quality of schooling can be considered substitutes and that “…societies' marginal rate of return to investment in school quality is at least as large as its marginal return to investment in additional years of schooling.”71

Rizzuto and Wachtel, “Further Evidence,” p. 241.

Additionally, school expenditures per pupil have been found to positively affect a student's educational attainment level. For these reasons, I focus on measures of school quality rather than quantity.

When examining the effect of school quality on individual earnings, two issues arise. The first is whether school inputs, as a measure of school quality, actually have any effect on earnings. Julian Betts provides a comprehensive survey of the relevant empirical studies and shows that studies focusing on individuals educated after 1960 tend to find little or no role for school inputs in explaining student earnings.72

Betts, “Is There a Link.”

However, studies that focus on school resources measured at the state level or on individuals educated before the 1960s find a strong link between school inputs and individual earnings. The second issue is which of the school input measures best reflect the quality of schooling. According to Betts, studies that use state data from the Biennial Survey of the Commissioner of Education on average find that teacher education, expenditures per pupil, length of school year, teacher salary, and teacher-pupil ratios are respectively significant in 100, 69, 54, 54, and 19 percent of the surveyed regressions.73

These are simple cross-study averages of the percentage of the reported regressions in each study where the stated school input is significant at the 5-percent confidence level. Betts, “Is There a Link,” table 6-1, pp. 144–45.

Data on teacher education are not available during the period under consideration. Hence, as a measure of investment in human capital, I rely on total school expenditures. This measure also has the advantage of incorporating other measures of school quality. For example, higher teacher salaries, longer school terms, and higher teacher-pupil ratios all lead to greater total school expenditures. I therefore estimate the education-based human capital stock per worker for each state from real annual school expenditures based on the perpetual inventory model.74

The school expenditure data come from annual reports from 1870 to 1915 and from biennial reports until the 1950s by the U.S. Commissioner of Education. A detailed description of the perpetual inventory method is given in the appendix.

These estimates are then adjusted for interstate migration, as derived by Everett Lee.75

Lee, “Migration Estimates.”

For the experience-based human capital measure, I use data on the age profile of the workforce for each state from Ann Miller and Carol Brainerd's labor force estimates, along with years of schooling and literacy data for annual birth cohorts by state of residency created from the 1920, 1940, and 1960 U.S. Censuses to create a proxy for the average work experience of a worker in each state.76

Miller and Brainerd, “Labor Force Estimates,” pp. 362–517, presented in Kuznets and Thomas, Population Redistribution. Ruggles, Sobek et. al., IPUMS.

The primary drawback to this measure is the fact that the years-of-schooling data necessary to create a work-experience measure are not available for the earliest birth cohorts I consider. Because literacy data are available for these birth cohorts and there are overlapping data on literacy and years of schooling for many birth cohorts, I use the literacy data to predict the years of schooling for the early cohorts that do not report years of schooling. A detailed description of the creation (and limitations) of this measure is presented in the Appendix.

All data (except for the experience-based human capital measure) are first converted to real terms (1967 dollars) using a national consumer price index (CPI).77

U.S. Department of Commerce, Historical Statistics, pp. 210–11.

They are then adjusted using state relative prices constructed by Mitchener and McLean to better reflect differences in costs of living across states.78

Mitchener and McLean, “U.S. Regional Growth.” I thank Kris Mitchener and Ian McLean for providing me with their data.

In interpreting the results that follow it is also important to stress the fact that the data used are estimates constructed from survey or census data from 1870 to 1950. Given the dates of the surveys, there is likely a great deal of measurement error in the raw data, as well as possible biases due to the procedures used to create time-consistent series.

RESULTS

Level Regressions

In Table 4, fixed effects (FE) estimation results are presented for equation 3.79

OLS regressions with robust standard errors were also undertaken for the four time periods separately. The results are consistent with the panel regression results of Table 2 and suggest a fair amount of stability over time.

These results are compared with random effects (RE) estimation results using the Hausman specification test.80

Hausman, “Specification Tests.”

When looking only at the South, the Hausman test fails to reject the hypothesis of no correlation between the independent variables and the latent individual effect. In that case, the Hausman test suggests that FE estimation is not necessary, implying that the more efficient RE estimation is preferable. For all other regressions, however, the Hausman test favors FE estimation. For consistency, the FE estimates are presented in the body of the article, and the RE estimates are given in Appendix Tables 1 and 2.

The first three columns of Table 4 present results from FE estimation of equation 3 respectively for the United States, the South, and the Non-South, in 1880, 1900, 1920, and 1950 jointly, when the two human capital measures are considered separately. All variables are in natural logs. The results are good. In the three level regressions, all of the explanatory variables have the expected signs and are statistically significant, with the exception of the experience-based human capital measure, which is not significant for the United States and is negative for non-Southern states. For the United States as a whole, the coefficient estimates for education based human capital and physical capital are 0.35 and 0.06 respectively. The coefficient estimate for education-based human capital is statistically significant in all three regressions and yields similar estimates for the two regions and the nation as a whole. On the other hand, findings for the experience-based human capital measure vary greatly across regions. Only in the South does this proxy for on-the-job training show up positively and significantly. Moreover, for the Southern states, the coefficient estimate for experience based human capital is 1.27, relative to the 0.39 coefficient on education-based human capital. This result is in line with the observation that a majority of the workforce in Southern agricultural society had little to no formal education, and hence output relied heavily on learning gained through work experience. The result further underscores a potential inflexibility within Southern states in shifting to new types of industry, because on-the-job training is less likely to be useful for a worker shifting from an agricultural job to an job in industry than is learning based on formal education that will serve the worker in all work settings.

The effects of migration are captured within the education-based human capital estimate, which is adjusted for migration. In order to directly consider the effects of migration, the same regression was run including the education-based human capital measure without the migration adjustment and the natural log difference in this measure when migration is taken into account. Both the raw human capital term and the migration adjustment are statistically significant. The elasticity of real income with respect to the raw education based human capital stock is 0.17 and with respect to the migration adjustment is 0.26. This suggests that migration, or more specifically migration of the human capital embodied in individuals, played a very important role in determining per worker income levels across states. Moreover, because the migration adjustment used here assigns the average education-based human capital stock to an individual based on his or her state of birth, the measured effect of migration is likely underestimated. Given the previously discussed evidence of educational selectivity of individuals choosing to migrate, the true effects of migration are likely larger than those documented here.

Columns 4 through 6 of Table 4 present the same regression for the United States, the South, and the Non-South using an interacted education and experience human capital stock measure.81

To control for possible technological progress, the national regression was run first using a time trend and then using time dummies. The time trend is not significant and leaves the remainder of the regression results basically unchanged. Only the time dummy for 1950 is significant (and positive) in the FE regression. Again the results presented in Table 4 remain.

This interacted term reflects the notion that the benefit of experience may be greater with higher education levels (or, similarly, that greater experience enhances the benefits of education). Because these measures are at the state level, this may not mean that the benefit from the laborer's experience depends necessarily on his own education, but perhaps on that of his manager or average co-worker with whom he works and who likely disseminate information to him. The coefficient estimate for the interacted human capital measure ranges from 0.3 for the Non-South to 0.4 for the South. Relative to the coefficient estimates for human capital from formal education, which were not statistically different across the regions, the regional estimates for the interacted human capital term are statistically significantly different from that for the United States as a whole. In particular, the Southern coefficient is significantly higher than the U.S. coefficient estimate. This is perhaps not surprising considering the extremely different findings for experience in the regions. Taking the U.S. coefficient estimate for interacted human capital of 0.34 implies that a 10 percent increase in a state's per capita stock of human capital would have led to a 3.4 percent increase in its per capita income level during this time period, all else equal. For the South, the increase in income would have been over 4 percent. This evidence suggests that the South would have gained even more than the nation as a whole from marginal investments in human capital.

There is, of course, an issue of possible reverse causality between income and investment in human capital, which could be driving the results. To control for the possibility of dual directions of causation, three different regressions are considered. First, a panel instrumental variable regression was run using lagged human capital levels as an instrument for the human capital term. The results from this regression follow those of column 4, Table 4 extremely closely. The only differences are a slightly higher R2(0.93), a larger coefficient (0.44) on the interacted human capital term, and fewer observations (because the use of a lagged variable as an instrument prevents the use of observations from 1880). Second, the regression was again run for the 1900 to 1950 time periods, including 1880 state income per worker as an explanatory variable. Third, in a slightly stronger test, lagged income levels are included as independent variables. If high investment in human capital (through school expenditures) is solely the consequence of high state wealth, then the human capital stock coefficient should become insignificant once initial income, or lagged income, is included. In the separate regressions, both initial 1880 income and lagged income are statistically significant and positive, with coefficient estimates of 0.07 and 0.2 respectively.82

The RE estimate for 1880 income is presented because FE estimation cannot estimate a constant variable.

Still in both regressions the overall results of Table 4 stand. Specifically, the coefficient estimates for the interacted human capital term are again significant at the 1-percent confidence level and have even slightly higher coefficient estimates of 0.44 and 0.43 in the two regressions. Hence, the findings for human capital are not driven purely by income levels.

A final issue to consider is that of omitted variables. It is possible that there are omitted variables that may be correlated with human and physical capital stock variables.83

For example, a state's policy environment is likely to be correlated with its investment in human and physical capital, leading to a positive bias on the human and physical capital coefficients if ignored.

However, FE estimation explicitly controls for such state-specific factors so long as they are time-invariant. Additionally, the inclusion of initial 1880 income in the regression mentioned above captures many of the state-specific characteristics that might otherwise positively bias the estimated human and physical capital coefficients.

Growth Regressions

Table 5 presents growth regression results for the United States, the South, and the Non-South from 1880 to 1950, following the Benhabib and Spiegel catch-up specification of equation 5. It is worth highlighting the interpretation of the coefficient estimates on hi and the catch-up term. Specifically, g reflects possible endogenous technological progress, while m reflects the catch-up component due to technological diffusion. If one looks at national patent data, we see that innovative activity is concentrated in only a handful of countries. This pattern is likely to also hold within a country. That is, if most endogenous technological progress is occurring in a few lead states and then diffusing to the rest of the nation, one should expect that m will be greater than g. Note also that the concept of conditional convergence is embodied in the notion of technological diffusion: conditional convergence, if present, is occurring through technological diffusion and will be evidenced by a positive and statistically significant estimate for m. Moreover, it is important that the estimate for g be less than that for m, because if g were greater than m throughout the United States, endogenous technological progress would exceed diffusion, implying divergence in technology and income levels across states. This implies that the coefficient estimate (g−m) on ln hi should be negative.

The results are presented in Table 5. Both the growth of physical and interacted education and experience human capital contribute to income growth per worker, with statistically significant coefficients of 0.08 and 0.4 respectively.84

The finding that human capital growth matters to income growth demonstrates an important role for human capital as an input into production itself, which Benhabib and Spiegel (“Role”) do not find in their 1965–1980 cross-sectional country regressions.

For the South, the estimated coefficient for human capital growth is 0.58, although the difference from the United States as a whole is not statistically significantly. The results also confirm that in addition to the growth of human capital, the level of human capital in a state is crucial to its growth because of its contribution to technological diffusion. For the United States as a whole, the statistically significant coefficient estimate for m is 0.03, suggesting that technological diffusion was occurring. The coefficient estimate for (g−m) is −0.014, suggesting an estimate for g of 0.016. The finding that m is much larger than g is not surprising because it is likely that very little endogenous technological progress was occurring in a majority of the states. Taken together, these results are consistent with the notion of conditional convergence through technological diffusion.

Capital Stock

The lack of available capital is often cited as having limited the speed of Southern industrialization.85

Carlton and Coclanis, “Capital Mobilization”; and Wright, Old South.

However, it is likely that the scarcity of human capital able to productively use and maintain physical capital may have significantly contributed to the lack of physical capital accumulation in the South.86

Using a cross-country regression for 1965, Benhabib and Spiegel (“Role”) find that the human capital level of a nation positively affects its physical capital accumulation.

Running a regression for the determinants of physical capital accumulation based on human capital levels, physical capital levels, and workforce levels yields interesting results. For the South, the level of interacted human capital positively affects physical capital accumulation with a statistically significant coefficient of 0.012.87

Labor is positive but marginally insignificant and the current capital stock enters negatively and significantly as expected with diminishing returns to capital.

More interestingly, if education- and experience-based human capital are considered separately for the South, the education-based human capital term is positive (0.014) and significant at the 1-percent confidence level, whereas the experience-based human capital coefficient is not statistically significant. This is consistent with the notion that although the South relied heavily on experience-based human capital for production, this is not necessarily the type of human capital which encourages investment, especially in industries in which the South did not have great previous experience. Instead, human capital gained through formal education appears to have been a positive determinant of physical capital accumulation in the South, as one might expect given the transferability and applicability of this type of human capital to a broader range of industries.

CONCLUSION

This article finds two crucial roles for human capital in explaining the growth and convergence pattern of states after the Civil War. In particular, human capital levels are shown to matter not only to a state's income level, but also to its growth rate, both directly as an input into production and indirectly through technological diffusion.

Still, there is a unique pattern in the South, whose overwhelmingly agricultural society relied more heavily on work experience than formal education, whose racial discrimination in school resource allocation played a crucial role in lowering human capital accumulation of both blacks and whites, and whose investment in physical capital is found to have depended on human capital accumulated through formal education rather than through work experience.

Not only is this last aspect of the Southern experience consistent with the notion that experience-based human capital may be more job specific and therefore less useful than education-based human capital when considering switching to new industries, but it also provides support to the argument that the South's lack of emphasis on formal education slowed both investment and growth.

In levels, the elasticity of state income per worker in terms of interacted human capital per worker is found to be approximately 0.34. In growth terms, the elasticity of per worker income growth with respect to growth of human capital is 0.4 and with respect to the role of human capital in technological diffusion it is 0.03 times the percentage income gap relative to the richest state.

The fundamental question is this: why did the South not industrialize rapidly after the abolition of slavery? The lack of available capital is often cited as having limited the speed of Southern industrialization.88

Carlton and Coclanis, “Capital Mobilization”; and Wright, Old South.

Still, the results presented here suggest that the lack of human capital (especially that formed through formal education) may have had as much, if not more, to do with the slow pace at which the South industrialized. Racial discrimination appears to have played a crucial role in the choice of Southern politicians to not support public education for either race. Furthermore, it could be argued that the lack of capital (especially Northern capital) available in the South, may itself have been a consequence of the scarcity of education-based human capital in the South.

After the Civil War, the South did not catch up with the educational standards of the rest of the nation. So long as pronounced educational and skill differences existed between the South and the rest of the United States, real Southern per capita income did not quickly converge towards the national norm. This was due to both level (productivity) effects and growth (technological diffusion) effects. Had the South been able to more quickly increase its human capital levels, particularly through formal education, it would have benefited both from a greater speed of convergence with the rest of the nation, as well as from a greater absolute convergence in income levels with the rest of the nation.

Appendix: Data

Education Based Human Capital

Annual estimates for the stock of human capital per worker generated from formal education in each state from 1870 to 1950 are based on total school expenditures data (from annual reports from 1870 to 1915 and from biennial reports until the 1950s by the U.S. Commissioner of Education) and on migration data from Lee.

89

Lee, “Migration Estimates.” Data from Dakota is used for North and South Dakota in 1880 because this territory had not yet been separated into two states at that date. Oklahoma is missing all 1880 data.

First, non-migration-adjusted human capital stock estimates,

, are constructed from real annual school expenditure data, St, as a measure of investment in education, using the perpetual inventory model:

where δ is a depreciation (or obsolescence) rate reflecting the fact that human capital disappears as people forget or die. A depreciation rate of 10 percent is used because this implies that only 1 percent of initial human capital is left after 44 years.

90

Other rates such as 5 percent and 20 percent are also tested yielding generally similar regression results.

An initial human capital stock value for 1870 is estimated according to

where g is the average annual growth rate (in ln differences) of real school expenditures from 1870 to 1950 and S* is a weighted estimate of the steady-state value of investment based on the following formula:

The raw human capital stock estimates are then adjusted for interstate migration using data derived by Lee.

91

Lee, “Migration Estimates.”

The Lee data decompose the adult population in a state at a given time according to their state of origin. Hence, to create a migration-adjusted measure of a state's human capital, Ht, I took a weighted sum of the human capital stocks from the states of origin of the adult population. This assumes that an individual was educated in their state of origin before migrating. Education-based human capital estimates for each decade from 1870 to 1950 are available upon request in the technical appendix.

Experience Based Human Capital:

Estimates of the average work experience of laborers in each state are derived by first estimating the average work experience of people born within a specific birth cohort by state of residence using data on average years of schooling from the 1940 and 1960 U.S. Censuses.92

All of the census data come from the Integrated Public Use Microdata Series: Version 2.0 (Ruggles, Sobek, et al., 1997: http://www.ipums.org).

This yields schooling estimates for birth cohorts from as early as 1840 to 1940. Schooling data are available for a majority of states from 1840 to 1850, with almost all states having observations starting with the 1850 birth cohort. For early birth cohorts not recorded in the 1940 Census, I use data on literacy from the 1920 census to predict years of schooling by running the years of schooling data from the 1940 Census on the 1920 literacy responses by annual birth cohorts and state of residency.93

This panel regression with 3,233 observations yields a coefficient estimate of 3.98 for the literacy index (which ranges from 1 (completely illiterate) to 4 (completely literate)) with an R2 of 0.36 and an Wald statistic of 964.

The predicted values for specific state cohorts are then used to replace the missing observations. This measure on years of schooling for each annual cohort by state is then used to estimate the average experience of the cohort by state using the following estimate:

Average annual cohort experience in year x=age of worker in year x−average years of schooling for that birth cohort−6

where six is assumed to be the earliest age that people either go to school or work. Data on the age profile of workers in each time period and state are then used to created a weighted measure of the years of work experience of the workforce in each state in each time period. Finally, the data are put into per worker terms to reflect the work experience of the average laborer in a state. The workforce data come from Ann Miller and Carol Brainerd's labor force estimates presented in Kuznets and Thomas.94

Miller and Brainerd, “Labor Force Estimates,” pp. 372–80, table 2.4.

The need for some predicted values for years of schooling is a strong drawback to this experience measure. Because the earliest birth cohorts rely more heavily on predicted values (i.e., generally prior to 1849), the importance of the predicted values is significant for 1880 estimates but diminishes quickly with time. For example, in the 1880 estimate of average work experience, approximately 26 percent of the U.S. labor force was born before 1835 (71 percent before 1855).95

These two cutoff dates are reported here since these are the two end years for relevant birth cohort groupings reported in the age profile labor data.

By 1900 only 4 percent of the labor force was born before 1835 (24 percent before 1855) and by 1920 no workers were born before 1835 (with 4 percent born before 1855). This gives some sense of the strength or weakness of the experience measure. Still, the article's results when this experience term is interacted with the education-based human capital term are extremely consistent with the results obtained using only the education-based human capital term.

Capital Stocks

Easterlin provides estimates for the capital stock in manufacturing for each state in 1880, 1900, and 1920. These estimates are based on census reports on the gross assets of manufacturing establishments, including land, buildings, machinery, and cash, but excluding rented capital.

96

A detailed explanation of the construction of this series is available in Kuznets and Thomas, Population Redistribution, pp. 675–78.

I supplement the Easterlin data by creating 1950 estimates according to

where g is the average annual growth rate (in ln differences, using the three year averages to smooth out the endpoints) of real investment in new plant and equipment in the manufacturing sector from 1951 to 1976 based on data from the Annual Survey of Manufactures (ASM), various issues, and δ is a depreciation rate of 8 percent (chosen to match Easterlin's assumed depreciation rate). I1950 should ideally be investment data in 1950. However, no data for investment in 1950 are available by state. Hence I consider three proxies for this measure. The first is Romans's estimates of investment in new plant and equipment by state in 1953, which uses ASM data appropriately adjusted to match aggregate U.S. manufacturing data from various issues of the Survey of Current Business.

97

Romans, Capital Exports.

The second is 1947 investment data on investment in new and used plant, equipment, and land in manufactures from the 1947 Census of Manufactures. The third is a steady-state measure of investment (weighted in similar fashion to equation A3) using ASM data on investment in new plant and equipment by manufacturers from 1951 to 1976. The three alternative measures are highly correlated and yield extremely similar results in the regressions. The results presented in the article come from the first measure, which uses Romans's 1953 data as a proxy for initial investment.

Finally, estimates for the capital stock in manufacturing in each state in 1880, 1900, 1920, and 1950 are proportionately scaled up to match the time series behavior of the aggregate U.S. physical capital stock using data on the real capital stock for the U.S. private nonfarm nonresidential sector from John Kendrick.98

Kendrick, Productivity Trends.

Personal Income

Personal income data come from estimates by Richard Easterlin in Kuznets and Thomas.99

Easterlin, “Regional Growth.”

Easterlin uses annual estimates from the National Bureau of Economic Research (NBER) and the Department of Commerce for the 1919–1921 data and 1949–1951 time periods, respectively. He then constructs estimates for 1880 and 1900 using estimates of service and property income. A lengthy description of the estimation procedure is given in Kuznets and Thomas.100

Kuznets and Thomas, Population Redistribution, pp. 703–27.

Labor Force

Ann Miller and Carol Brainerd estimate labor force data in Kuznets and Thomas using decennial censuses of the population for the period 1870–1950.101

Miller and Brainerd, “Title,” pp. 364–411.

A description of the estimation techniques, as well as the estimates themselves, is available in Kuznets and Thomas's Population Redistribution.

I thank Erica Groshen, Vassilis Hajivassiliou, James Harrigan, Sebnem Kalemli-Ozcan, Thomas Meyer, Carol Osler, Gabriel Perez-Quiros, Xavier Sala-i-Martin, David Weiman, Kirk White, Jeffrey Willamson, Eric van Wincoop, Gavin Wright, Kei-Mu Yi, and two anonymous referees for valuable comments. Sairah Burki, Joseph Davis III, and Jenessa Gunther provided valuable research assistance. I remain responsible for any errors.

References

Barro Robert, and Xavier Sala-i-Martin. 1991 “Convergence across States and Regions.” Brookings Papers on Economic Activity, no. 1: 107– 82.Google Scholar
Barro Robert, and Xavier Sala-i-Martin. 1992 “Convergence.” Journal of Political Economy 100, no. 2: 223– 51.Google Scholar
Beatty Bess. 1987 “Lowells of the South: Northern Influences on the Nineteenth-Century North Carolina Textile Industry.” Journal of Southern History 53, no. 1: 37– 62.Google Scholar
Benhabib Jess, and Mark Spiegel. 1994 “The Role of Human Capital in Economic Development: Evidence from Aggregate Cross-Country Data.” Journal of Monetary Economics 34, no. 2: 143– 73.Google Scholar
Betts Julian. 1996. “Is There a Link between School Inputs and Earnings? Fresh Scrutiny of an Old Literature.” In Does Money Matter?: The Effect of School Resources on Student Achievement and Adult Success, edited by Gary Burtless, chap. 6 141– 91. Washington, DC: The Brookings Institution
Bond Horace Mann. 1934. The Education of the Negro in the American Social Order. New York: Prentice Hall
Card David, and Alan Krueger. 1992 “School Quality and Black-White Relative Earnings: A Direct Assessment.” The Quarterly Journal of Economics 107, no. 1: 151– 200.Google Scholar
Carlton David. 1990 “The Revolution from Above: The National Market and the Beginnings of Industrialization in North Carolina.” Journal of American History 77, no. 2: 445– 75.Google Scholar
Carlton David, and Peter Coclanis. 1989 “Capital Mobilization and Southern Industry, 1880–1905: The Case of the Carolina Piedmont.” This JOURNAL 49, no. 1: 73– 94.Google Scholar
Carlton David, and Peter Coclanis. 1995 “The Uninventive South? A Quantitative Look at Region and American Inventiveness.” Technology and Culture 36: 302– 26.Google Scholar
Caselli Francesco, and Wilber John Coleman II. 2001 “The U.S. Structural Transformation and Regional Convergence: A Reinterpretation.” Journal of Political Economy 109, no. 3: 584– 616.Google Scholar
Colberg Marshall. 1965. Human Capital in Southern Development, 1939–1963. Chapel Hill: The University of North Carolina Press
Donohue IIIJohn, James Heckman, and Petra Todd. 2002 “The Schooling of Southern Blacks: The Roles of Legal Activism and Private Philanthropy, 1910–1960.” Quarterly Journal of Economics 117, no. 1: 225– 68.Google Scholar
Easterlin Richard. 1957. “Regional Growth of Income: Long Term Tendencies.” In Population Redistribution and Economic Growth: United States, 1870–1950, Vol. 2, edited by Simon Kuznets, Ann Miller, and Richard Easterlin, 702– 59. Philadelphia: American Philosophical Society
Easterlin Richard. 1960. “Interregional Differences in Per Capita Income, Population, and Total Income, 1840–1950.” In Trends in the American Economy in the Nineteenth Century, Conference on Research in Income and Wealth, National Bureau of Economic Research. Princeton, NJ: Princeton University Press
Easterlin Richard. 1968. Population, Labor Force, and Long Swings in Economic Growth. New York: NBER, Columbia University Press
Gerber James. 1986. Southern White Schooling, 1880–1940. University of California Davis Ph.D. diss.,
Gibson Campbell, and Kay Jung. 2002. “Historical Census Statistics on Population Totals by Race, 1790 to 1990.” Washington, DC: U.S. Census Bureau. Population Division Working Paper No. 56,
Hausman Jerry. 1978 “Specification Tests in Econometrics.” Econometrica 46, 6: 1251– 72.Google Scholar
Hausman Jerry, and William Taylor. 1981 “Panel Data and Unobservable Individual Effects.” Econometrica. 49, no. 6: 1377– 99.Google Scholar
Higgs Robert. 1977. Competition and Coercion: Blacks in the American Economy, 1865–1914. Cambridge: Cambridge University Press
Hirschman Albert. 1958. The Strategy of Economic Development. New Haven, CT: Yale University Press
Kim Sukkoo. 1995 “Expansion of Markets and the Geographic Distribution of Economic Activities: the Trends in U.S. Regional Manufacturing Structure, 1869–1987.” Quarterly Journal of Economics 110, no. 4: 881– 908.Google Scholar
Kim Sukkoo. 1998 “Economic Integration and Convergence: U.S. Regions, 1840–1987.” This JOURNAL 58, no. 3: 659– 83.Google Scholar
Kendrick John. 1961. Productivity Trends in the United States. NBER. Princeton University Press
Kuznets Simon, and Dorothy Swaine Thomas, eds. 1957. Population Redistribution and Economic Growth: United States, 1870–1950, Vol. 1: Methodological Considerations and Reference Tables. Philadelphia: American Philosophical Society
Lebergott Stanley. 1964. Manpower in Economic Growth: the American Record since 1800. New York: McGraw-Hill
Lee Everett. 1957 “Migration Estimates.” In Population Redistribution and Economic Growth: U.S. 1870–1950, vol. 1, ed. by Simon Kuznets and Dorothy Swaine Thomas. Philadelphia: American Philosophical Society: table P-3, 249– 348.
Lucas Robert Jr. 1988 “On the Mechanics of Economic Development.” Journal of Monetary Economics 22, no. 1: 3– 42.Google Scholar
Mankiw Gregory, David Romer, and David Weil. 1992 “A Contribution to the Empirics of Economic Growth.” Quarterly Journal of Economics 107, no. 2: 407– 38.Google Scholar
Margo Robert. 1986 “Race, Educational Attainment and the 1940 Census.” This JOURNAL 46, no. 1: 189– 98.Google Scholar
Margo Robert. 1986 “Race and Human Capital: Comment.” The American Economic Review 76, no. 5: 1221– 24.Google Scholar
Margo Robert. 1988. “Schooling and the Great Migration.” NBER working paper #2679,
Margo Robert. 1990. Race and Schooling in the South, 1880–1950: An Economic History. Chicago: University of Chicago Press
Margo Robert. 2002. “The North-South Wage Gap, Before and After the Civil War.” NBER working paper #8778, Forthcoming in the festschrift for Stanley Engerman.
McCuistion Fred. 1930. “Financing Schools in the South: Some Data Regarding Sources, Amounts, and Distribution of Public School Revenue in the Southern States.” Nashville,
Miller Ann, and Carol Brainerd. 1957. “Labor Force Estimates.” In Population Redistribution and Economic Growth: U.S. 1870–1950, vol. 1, edited by Simon Kuznets and Dorothy Swaine Thomas, 362– 517. Philadelphia: American Philosophical Society
Mitchener Kris, and Ian McLean. 1999 “U.S. Regional Growth and Convergence, 1880–1980.” This JOURNAL 59, no. 4: 1016– 42.Google Scholar
Myrdal Gunnar. 1957. Economic Theory and Underdeveloped Regions. London: G. Duckworth
Nelson Richard, and Edmund Phelps. 1966 “Investment in Humans, Technological Diffusion, and Economic Growth.” American Economic Review 56, no. 2: 69– 75.Google Scholar
Ransom Roger, and Richard Sutch. 1977. One Kind of Freedom. Cambridge University Press: Cambridge
Rizzuto Ronald, and Paul Wachtel. 1980 “Further Evidence on the Returns to School Quality. Journal of Human Resources 15, no. 2: 240– 54.Google Scholar
Romans J. Thomas. 1965. Capital Exports and Growth Among US Regions. Middletown, CT: Wesleyan University Press
Romer Paul. 1990 “Endogenous Technological Change.” Journal of Political Economy 98, no. 5: S71– S102.Google Scholar
Romer Paul. 1986 “Increasing Returns and Long-Run Growth.” Journal of Political Economy 94, no. 5: 1002– 37.Google Scholar
Rosenbloom Joshua. 1990. “One Market or Many? Labor Market Integration in the Late Nineteenth-Century United States.” This JOURNAL 50, no. 1: 85– 107.Google Scholar
Rosenbloom Joshua. 1996. “Was There a National Labor Market at the End of the Nineteenth Century? New Evidence on Earnings in Manufacturing.” This JOURNAL 56, no. 3: 626– 56.Google Scholar
Ruggles Steven, Matthew Sobek, et al. Integrated Public Use Microdata Series: Version 2.0 Minneapolis: Historical Census Projects, University of Minnesota (1997): http://www.ipums.org.
Smith James. 1984 “Race and Human Capital.” The American Economic Review 74, no. 4: 685– 98.Google Scholar
Smith James. 1986 “Race and Human Capital: Reply.” The American Economic Review 76, no. 5: 1225– 29.Google Scholar
Solow Robert. 1956 “A Contribution to the Theory of Economic Growth.” Quarterly Journal of Economics 70, no. 1: 65– 94.Google Scholar
U.S. Bureau of Education. 1870–1950. Report of the Commissioner of Education. Washington DC: GPO
U.S. Bureau of the Census. 1947–1976. Annual Survey of Manufactures. Washington, DC: GPO
U.S. Bureau of the Census. 1947. 1947 Census of Manufactures. Washington, DC: GPO
U.S. Department of Commerce. 1989. Historical Statistics of the United States: Colonial Times to 1970. White Plains, NY: Kraus International Publications
Uzawa Hirofumi. 1965 “Optimum Technical Change in an Aggregative Model of Economic Growth.” International Economic Review 6, no. 1: 18– 31.Google Scholar
Vigdor Jacob. 2002 “Locations, Outcomes, and Selective Migration.” Review of Economics and Statistics 84, no. 4: 751– 55.Google Scholar
Vigdor Jacob. 2002 “The Pursuit of Opportunity: Explaining Selective Black Migration.” Journal of Urban Economics 51, no. 3: 391– 417.Google Scholar
Weiman David. 1990. “Staple Crops and Slave Plantations: Alternative Perspectives on Regional Development in the Ante-bellum Cotton South.” In Agriculture and National Development, edited by Louis Ferleger, 141– 45. Ames: Iowa State University Press
Williamson Jeffrey. 1965 “Regional Inequality and the Process of National Development: A Description of the Patterns.” Economic Development and Cultural Change 13, no. 4: 3– 47.Google Scholar
Williamson Jeffrey. 1998 “Growth, Distribution, and Demography: Some Lessons from History.” Explorations in Economic History 35, no. 3: 241– 71.Google Scholar
Williamson Jeffrey, and Peter Lindert. 1980. American Inequality: A Macroeconomic History. New York: Academic Press
Wright G. 1974 “Cotton Competition and the Post-Bellum Recovery of the American South.” This JOURNAL 34, no. 3: 610– 35.Google Scholar
Wright G. 1986. Old South, New South: Revolutions in the Southern Economy Since the Civil War. New York: Basic Books
Figure 0

FIGURE 2 SOUTHERN LITERACY (percentage of the population over the age of 10)

Figure 1

INEQUALITY IN SPENDING BY RACE IN SOUTHERN STATES IN 1930

Figure 2

THE PROBABILITY OF NORTHERN RESIDENCE IN 1940 FOR SOUTHERN-BORN BLACKS

Figure 3

FIGURE 3 MEAN SCHOOLING LEVELS IN MONTHS

Figure 4

EFFECTS OF RACIAL DISCRIMINATION: STATE-LEVEL FIXED-EFFECTS REGRESSIONS, 1880–1950

Figure 5

FIGURE 4 REAL ANNUAL SCHOOL EXPENDITURE PER PUPIL (1967 dollars)

Figure 6

FIXED-EFFECTS LEVEL REGRESSIONS, 1880–1950

Figure 7

FIXED EFFECTS GROWTH REGRESSIONS, 1880–1950

Figure 8

APPENDIX TABLE 1 RANDOM-EFFECTS ESTIMATION, 1880–1950

Figure 9

APPENDIX TABLE 2 RANDOM-EFFECTS ESTIMATION, 1880–1950