Policy Sciences

, 41:183

Education and economic growth in the United States: cross-national applications for an intra-national path analysis

Authors

    • Univerisity of Alabama
  • Stephen A. Borrelli
    • Univerisity of Alabama
Article

DOI: 10.1007/s11077-008-9062-2

Cite this article as:
Baldwin, N. & Borrelli, S.A. Policy Sci (2008) 41: 183. doi:10.1007/s11077-008-9062-2

Abstract

Do the leading predictors of economic growth found in the cross-national research have a capacity to predict economic growth at the state level in the United States (US)? Are the effects of education spending on economic growth underestimated because research fails to examine the indirect effects of spending on economic growth? This article presents the findings from a study investigating the relationship between education and economic growth in US states while controlling for the effects of the leading predictors of economic growth from the cross-national research. It also utilizes a path model to examine direct and indirect relationships between education spending and economic growth measured as per capita income growth. The results indicate that spending on higher education and highway expenditures demonstrate a positive association with growth in per capita income, while K12 (kindergarten through 12th grade) spending and K12 pupil–teacher ratios demonstrate a negative association with income growth from 1988 to 2005. Moreover, K12 spending and population growth indirectly affect income growth through their relationship with K12 pupil–teacher ratios, and spending on higher education indirectly affects income growth through college attainment rates. Overall, all but one variable from the cross-national research demonstrates a significant direct or indirect relationship with income growth during at least one time-period investigated. Treating K12 pupil–teacher ratios and college attainment as mediating variables also enhances our understanding of the dynamics that explain growth in per capita income at the sub-national level in the US. However, some unexpected findings emerge when the data are analyzed on the basis of two eight-year sub-periods.

Keywords

EducationEconomic growthCross-nationalPer capita incomeUnited States

Introduction

Although scholars and policymakers have made an admirable effort to look beyond income growth to measure economic growth and improvement in the human condition domestically and internationally (United Nations 2005), growth in per capita income remains a major focus of their study. Scholars and policymakers are fixated on finding the right mix of policies that will engineer consistent increases in per capita income. Perhaps the most intensely studied and debated policies thought to enhance growth are those designed to improve the quality of education, or increase human capital.

Before launching another investigation into the impact of education on income growth, we believe that prudence dictates stepping back and speculating as to why and how the previous literature may be confused or incomplete. One concern is the simple absence of effective communication and cumulation among the scholars who have studied this topic. For example, reviews of the cross-national (Temple 1999; Barro and Sala-i-Martin 2004; Nijkamp and Poot 2004), and intra-national (in the United States: see Fisher 1997; Reed 2006; Hungerford and Wassmer 2004; Bell et al.2005) economic growth research reveal little cross-citation or even awareness that the other literature exists. Recognizing the differences between comparing countries and comparing states, regions, or provinces (Brace 1993), we think that the state economic growth research has the potential to advance by applying some of the lessons learned from the cross-national economic growth research. We consequently begin this application by incorporating the most commonly cited control variables from the cross-national literature into a study of income growth in the states of the United States (US).

A second concern has been insufficient discussion of how to conceptualize, measure, and specify educational effects. Early studies of the impact of human capital on growth utilized whatever educational data were available to operationalize human capital, making no distinctions between overall educational spending, per-pupil expenditures, pupil–teacher ratios, teacher salaries, and educational attainment rates (Plaut and Pluta 1983; Nardinelli et al. 1987). However, given the current accessibility of higher quality data, scholars now have a broader range of educational measures from which to choose and must carefully select those measures, as well as carefully discern how they conceptually relate to each other. In recent works, distinguishing educational “quantity” from “quality” is more common (Lee and Barro 2001), but while there appears to be basic agreement that spending measures capture quantity, there is less agreement about which indicators measure quality. For example, some scholars (Card and Krueger 1992; Heckman et al. 1996) consider pupil–teacher ratios to be a measure of the quality of the educational experience, while others (Hanushek et al. 1996) argue that pupil–teacher ratios are just alternative measures of resources invested, that is, more quantity than quality.

Because the conceptual difference between quantity and quality is vague and controversial and because the terms are value-laden (quality is typically construed more positively than quantity), we believe that viewing the measures of human capital in terms of their place in the policy “system” is more useful. If one thinks of educational policy as a simple Eastonian system—with inputs, throughputs, outputs, and outcomes—the causal roles of the different educational variables fall better into place. In this spirit, Hanushek (2003) now calls policies that emphasize sheer spending “input-based” policies, denoting the role of spending as an input in the policy system. Graduation and attainment rates, on the other hand, are more like “outputs”—the direct product of the government educational system and often explicit targets of government policy. Standardized test scores, which Hanushek favors as measures of educational policy success and as targets for policy reforms, also seem to be at the output or outcome end of the policy system. Still ambiguous, however, is the role of pupil–teacher ratios. Hanushek considers pupil–teacher measures as alternative measures of inputs, more a reflection of resources thrown at education than what the educational system is actually producing. However, if one concedes that pupil–teacher ratios are not outputs, they should not be automatically treated as inputs. Pupil–teacher ratios are a reflection of what is done with investments in education, or how educational spending is deployed or implemented, which logically and chronologically follows decisions about how much to spend.1

Thus, we hope to contribute to the scholarship on education and income growth through a sub-national study that imports the leading predictors from the cross-national literature and attends to the selection and causal sequencing of different educational measures. Because previous education and economic growth research on US states does not use a systems approach, we believe that it may underestimate the effects of educational spending by failing to observe possible indirect effects on income growth. Relative to most cross national and intra-national economic growth research, our research also attempts to construct more robust measures of variables through utilizing multi-year averages rather than single-year measures that can include aberrant years. Finally, relative to most economic growth research, our research attends to the question of the time lengths of lagged effects of independent variables. For example, since it usually takes longer for high school graduates to enter the workforce than college graduates, we specify a shorter lag period for college attainment than for high school attainment.

In sum, applying what is known from the enormous body of cross-national economic growth research, our study investigates the relationship between state per capita income growth and education while controlling for the effects of infrastructure spending, savings deposits, and population growth. We also examine the capacity of educational spending to affect economic growth directly and indirectly through the mediating effects of pupil–teacher ratios and educational attainment operationalized as high school and college attainment rates (see Fig. 1). Through testing the path model in Fig. 1, this research adds to the debate over whether smaller pupil–teacher ratios facilitate desirable educational and economic outcomes (e.g., see Card and Krueger 1992; Finn et al. 2001; Hanushek 1999; Krueger 2003).
https://static-content.springer.com/image/art%3A10.1007%2Fs11077-008-9062-2/MediaObjects/11077_2008_9062_Fig1_HTML.gif
Fig. 1

Economic growth model 1997–2005

Theory and previous research

The theoretical foundation for asserting that education has a positive effect on economic growth in US states is derived from human capital theory. Originally presented in the works of Becker (1962) and Schultz (1963), this theory asserts that skills, knowledge, abilities, experience, aptitude, and training are human capital that, like physical capital, accrue a stream of future benefits when developed (Jorgenson and Fraumeni 1992; Mincer 1994). Likewise, when invested in, education is a form of human capital that can affect future benefits over a lifetime (Jorgenson and Fraumeni 1992; Senter 1999). Through generating human capital, education increases the productivity of labor, for example, through introducing technological innovation that enhances efficiencies (Garcia-Mila and McGuire 1992; Quan and Beck 1987). High quality labor is also attracted to states with superior public schools, and more highly educated individuals are better consumers because of their higher wages. Quan and Beck (1987) contend investments in education yield returns from educated parents who migrate to states that offer a quality education for their children. As an engine of economic growth, education enhances both private and public benefits that are ultimately reflected in measures such as per capita income (Evans and Oneal 1995).

Growth in per capita income is one of the most commonly investigated measures of economic growth in the US. The studies of economic growth and education in the US most commonly address the effects of educational expenditures on income. They indicate mixed findings but generally reveal positive relationships between educational spending and per capita income growth. Bensi et al. (2004), Berry and Kaserman (1993), Crain (2003), Dholakia and Harlam (1994), and Moomaw et al. (2002) demonstrate that educational expenditures have a positive relationship with per capita income, while Jones’ (1990) work shows that educational spending has a significant negative association with change in per capita income from 1969 to 1974 (but not during three other five-year periods). Quan and Beck (1987) show that educational expenditures have positive, negative, or insignificant relationships with per capita income depending on the region of the country, the lag period used to determine the effects of spending, and whether one is looking at K12 expenditures or spending on higher education. Vedder’s (2004) research demonstrates a negative relationship between spending on higher education and growth in per capita income, as well as a positive relationship between per capita income growth and the percentage of the population over 25 years holding a college degree. Two additional studies addressing the effects of educational attainment on per capita income (Crihfield et al. 1995; Crown and Wheat 1995) reveal that the log of the fraction of persons 25 years and older with 16 or more years of schooling, the median years of school completed, and the percentage change in college graduates 25 years and older are positively associated with per capita income.

Accessible economic growth research on states, provinces, and regions of countries outside the US disproportionately addresses the relationship between education and various measures of productivity, especially gross domestic product and gross state product (GSP). While this research affirms human capital theory, the research addressing education and income growth is limited and demonstrates mixed results. Chen and Feng (2000) find a positive association between enrollments in higher education and income growth in the provinces of China, while Gonzalez-Paramo and Martinez’s (2003) research concludes that there is an unclear connection between public investment in education and income growth in the states of Spain. Coulombe and Tremblay (2001) find that, like initial per capita income, initial educational attainment in the provinces of Canada demonstrates convergence, or a negative relationship with per capita income growth.

On the basis of human capital theory, we assume that educational spending translates into labor, programs, technology, and innovations that enhance the knowledge, skills, and abilities of students. These enhancements to the human capital of students, in turn, increase labor productivity, innovations, and consumer behavior that contribute to income growth. We consequently assert the following hypothesis:

Hypothesis 1

As state expenditures on education increase, state per capita income growth increases.

We further assume that individuals who graduate from high school and college possess larger amounts of human capital that can be translated into income growth than individuals without high school or college degrees.

We therefore propose the following:

Hypothesis 2

As educational attainment rates increase, growth in state per capita income increases.

Finally, we assume that smaller pupil–teacher ratios allow for more personalized classrooms and enhanced individual attention that augment the development of student knowledge, skills, and abilities. Card and Krueger’s (1992) research demonstrates that pupil–teacher ratios are directly associated with the earnings power of men born between 1920 and 1949. Moreover, meta-analyses and reviews of the research on class size demonstrate diverse findings, interactions, contradictions, and contingencies, but generally show that smaller classes are associated with increases in student achievement (Burr 2001; Fleming et al. 2002; Glass and Smith 1978; McGiverin et al. 1989). We assume that increases in student achievement associated with smaller classes reflect a growth in human capital that leads to income growth. We consequently assert the following:

Hypothesis 3

As state K12 pupil–teacher ratios decrease, state per capita income growth increases.

Although the quantity of studies on education and state growth is impressive, it is easily eclipsed by the volume of cross-national research relating human capital to economic growth (for meta-analyses and reviews, see Barro and Sala-i-Martin 2004; Nijkamp and Poot 2004; Poot 2000; Temple 1999). Beyond demonstrating the importance of education for economic growth, this body of research demonstrates the importance of infrastructure, savings deposits, and population growth in explaining economic growth. Although some of these variables are investigated in state-level studies, no state-level growth study investigates these variables as a body and no state growth studies investigate the predictive capacity of savings. Our research controls for equivalent state-level measures of these variables while investigating the relationship between education and state economic growth.

We utilized controls from the cross-national research because almost half of this research addresses growth in developed countries and an additional 22.8% investigates a combination of developed and less developed countries (Nijkamp and Poot 2004, p. 102). Borrowing cross-national controls took advantage of the largest body of research potentially applicable to the US as a developed country. A precedent for investigating US state economic growth based on cross-national theory and research also dates back to the 1980s (e.g., Gray and Lowery 1988; Olson 1982).

Given that countries in the contemporary global economy enjoy an interdependence beginning to resemble the interdependence of political subdivisions within larger countries, recent cross-national findings have become increasingly relevant to the US. Sharing a common currency, countries within the European Union are particularly inclined to approximate the flow of money and trade that occurs between states in the US. More importantly, regions (Hefner 1990) and many states in the US demonstrate an economic diversity or uniqueness typical of the aggregate economies of countries in the world. That is, many US states are like countries—their aggregate economies do not share a homogeneous foundation.2 This diversity even occurs among neighboring states or regions. For example, in the western US, Wyoming is uniquely a mining giant, Nevada’s economy reflects the arts, entertainment, and recreation, California dominates information services, and Arizona distinguishes itself in the real estate and rental and leasing NAIC classification (North American Industry Classification System) (US Census Bureau 2002a).3 Moreover, the economies of California, Texas, and New York perform on the same level as the top ten most productive countries in the world (Bureau of Economic Analysis 2007; World Bank Group 2007a). And, although national and regional components play an important role in state economic growth, Hendricks and Garand (2001, p. 1093) conclude that “a substantial portion of the variance in state economic growth is unique to each state.”

Spending on infrastructure enhances economic growth through reducing transportation costs and raising the social return to ongoing investments (Temple 1999, p. 146). Infrastructure investments also enhance growth through reducing commuting costs as well as the costs of importing goods that otherwise drive up the cost of living (Crown and Wheat 1995; Garcia-Mila and McGuire 1992). We consequently assert:

Hypothesis 4

As spending on state highways increases, state per capita income growth increases.

Neoclassical economic growth theory suggests that population growth causes the ratio of capital to labor to fall and therefore to diminish returns (Sedgley 1998). High rates of population growth lower the “steady-state level of capital and output per worker,” thereby reducing per capita growth rates (Barro and Sala-i-Martin 2004, p. 21). We therefore contend:

Hypothesis 5

As population growth increases, state per capita income growth decreases.

Neoclassical theory also contends that income grows as a function of labor and capital, with capital growing at a rate dependent upon investment or national savings (Poot 2000, p. 520). Savings allow investments that increase capital and subsequent growth beyond that which can be achieved if all the capital is consumed. However, in contemporary times, consolidation in the banking industry has seen the decline of autonomous state banks and a surge of national and regional banks that take savings to other states (Strahan 2006; Hirtle and Metli 2004).4 US institutional and individual investors are also generally increasingly moving money into stocks, mutual funds, hedge funds, bonds, debt securities, and private equities that take their savings into other states (Investment Company Institute 2008; Jones 2007; US Census Bureau 2003, pp. 755, 759). And, over the last 25 years, US international assets and investments in equities, debt securities, monetary authorities, general government, and foreign banks indicate that US investors have moved their money increasingly overseas (International Monetary Fund 1997, pp. 854–855, 2007, p. 602). Finally, in contrast to the savings habits of the rest of the world, including the developed countries of Europe, Americans save less as a percentage of their gross national income (World Bank Group 2007b). For these reasons, we do not hypothesize a relationship between state savings and income growth, but include it as an exploratory variable to help clarify the applicability of the cross-national findings to state economic growth research.

As stated earlier, our study is the first to test human capital effects using a path model, positing that educational spending leads to higher educational attainment and lower pupil–teacher ratios that subsequently lead to increased growth. Although we have a great deal of guidance from previous literature in specifying models predicting economic growth, the state academic literature provides little guidance in identifying predictors of the intervening variables. Because of the near-universal acceptance of a high school degree as a minimum credential in the modern US economy, the research explaining state variation in high school attainment is understandably historical in nature. Goldin (1998) finds that per capita wealth, agricultural income per agricultural worker, percentage of the work force in manufacturing, and average manufacturing wage all have a significant impact on variation in high school graduation rates across states in 1928. She argues that the state wealth measures represent the tax base available to fund education and that the manufacturing measures represent both the advancement of the states’ economies and the available alternatives to schooling for workers of high-school age. These variables help explain the stark regional differences in high school attainment found by Baier et al. (2004) in their analysis of data dating back to 1840.

The paucity of research on college attainment demonstrates a modest relationship between the number of degrees produced in a state and the number of college-educated workers in a state (Bound et al. 2004). Hendricks' (2004, p. 26) research, in turn, reveals that states with a high percentage of college graduates “employ skillbased technologies and specialize in skill-intensive technologies…. Impediments to labor mobility and other sources of regional wage differences play a minor role” in determining college graduates.

The few state studies examining pupil–teacher ratios as a dependent variable focus on the effects of teacher unionization on class size. Reduced pupil–teacher ratios are a desirable working condition sought by teachers’ unions. Kerchner (1986) contends that unionization is the most important explanation for nationwide decreases in class size between the 1960s and 1980s. Although her primary focus was unionization, Hoxby (2000) examined school district-level data and finds that average income, percent below poverty, percent unemployed, percent black, and urbanization are significantly related to class size. Replicating Hoxby’s model using improved unionization measures but only three states, Lovenheim (2007) finds that unionization and socio-economic control variables have insignificant effects on class size.

Our model assumes that as K12 expenditures decrease fewer financial resources are available to hire more teachers and therefore reduce pupil–teacher ratios. We further assume that state population growth typically involves growth in school-age children and that enrollments grow faster than the hiring of new teachers. In rapidly growing states, pupil–teacher ratios expand, especially in the short-run, to accommodate increased enrollments. We consequently assert the following hypothesis:

Hypothesis 6

As state K12 expenditures increase, state K12 pupil–teacher ratios decrease.

Hypothesis 7

As state population growth increases, state K12 pupil–teacher ratios increase.

We also contend that as K12 expenditures and college expenditures increase, more resources are available to invest in programs, technology, and human resources to increase high school and college attainment rates. The inflationary price of higher education has brought to bear especially strong pressures on institutions of higher learning to develop curricula, standards, and support structures that facilitate successful completion of college degrees.5 We therefore propose the following compound hypothesis:

Hypothesis 8

As state expenditures on education increase, high school and college attainment rates increase.

In sum, although human capital theory suggests that educational attainment has a positive effect on personal income growth, the studies on educational spending and income reveal inconsistent findings that beg for additional research. Beyond the effects of education on economic growth, the enormous body of cross-national research suggests that infrastructure, savings, and population growth also affect economic growth, yet these control variables as a set have not been applied to the states. Finally, previous research that operationalizes human capital as educational spending, attainment, and pupil–teacher ratios has not considered the impact of spending on educational attainment and pupil–teacher ratios. In fact, a paucity of research examines educational attainment and pupil–teacher ratios as dependent variables.

Data and analysis

To assess the economic growth returns to investment in education, we conducted cross-sectional regression analyses of the 48 states in the continental US.6 Although pooled cross-sectional analysis would increase our sample size substantially, we believe multi-year averaging yields more robust results through reducing measurement error and avoiding the need to correct for violations of traditional regression assumptions associated with pooled analysis.

We gathered 18 years of per capita income data from a Bureau of Economic Analysis (2007) website.7 Annual changes (growth) in per capita income were calculated and averaged for the growth occurring from 1988–1989 to 2004–2005 and from the 1988–1989 to 1995–1996 and 1997–1998 to 2004–2005 sub-periods. The measures of the control variables were averaged according to these timeframes, and we utilized lagged timeframes to measure and average the delayed effects of all variables related to human capital.

All financial predictors in our model—educational expenditures, highway disbursements, and savings—are measured on the basis of their per capita effects. Most studies (see Fisher 1997 for review) of the relationship between education and economic growth at the state level in the US address the impact of per capita spending. Moreover, Barro and Sala-i-Martin (2004) assert that people conceive of countries as having certain properties, such as being poor or rich, based on the quality of life for their inhabitants. Per capita measures of capital, consumption, GSP, and income provide a clearer picture of a country because they capture how people are affected. Jones (1990, p. 222), in turn, claims that per capita measures control for the size of a state’s public sector that can be attributed to population differences, thus incorporating “the assumption that more citizens need more policy expenditures.”

Our first measure of public investment in education, the first link in the proposed causal chain linking human capital and growth, is the total state expenditure on public primary and secondary education (K12) per capita. Given the argument that K12 spending requires at least as long as it takes a generation to complete its schooling in order to impact economic growth, we gathered data and calculated means on this variable for the 1975–1976 to 1991–1992 school-year period.8 The K12 spending data were drawn from multiple volumes of the Digest of EducationStatistics (US Department of Education 1980, 1997, 1998c; US Department of Health, Education, and Welfare 1977–78), and the population data used to determine per capita spending were drawn from multiple volumes of Statistical Abstracts of the United States (US Census Bureau 1985–1990 et seq.).

A second measure of human capital input investigated is state operations appropriation per capita for higher education. Because attaining a college degree requires fewer years than completing K12, we assumed a seven-year lag for the impact of higher education spending on per capita income growth. We consequently gathered data on this variable from 1982 to 1998. The data on higher education come from the State Higher Education Executive Officers’ (SHEEO) (2007) website.

To investigate the second link—educational attainment—in the proposed chain connecting education with growth, we gathered data on the percentages of the population 25 years and older who have completed high school and who hold a four-year college degree. Data on high school and college attainment were accessed through the US Census Bureau (2007a) website. Given that high school and college attainment rates are unavailable for 1992, we derived estimates of 1992 rates through 96 regressions between time and the 17 known annual high school and college attainment rates for each state.

Our final education-related measure, a throughput, is K12 pupil–teacher ratios. Measures of this variable come from the Digest of EducationStatistics (US Department of Education 1997 et seq., 1998c) website from which we collected data from 1976 to 1992 under the assumption that improving K12 quality also has a 13-year delayed effect on economic growth. Pupil–teacher ratios in higher education were not investigated because we were unable to locate a reliable source of data covering all 48 states and the relevant time-period.

Given our purposefully limited sample of 48, we necessarily restricted ourselves to a relatively short list of predictors and control variables. As previously indicated, we turned to the enormous body of comparative economic growth research (for reviews, see Barro and Sala-i-Martin 2004; Nijkamp and Poot 2004; Poot 2000; Temple 1999) to select spending on infrastructure, savings deposits, and population growth as control variables.

Data on total state highway disbursements between 1989 and 2005 provided the basis for the measure of spending on infrastructure. Because we assume that the benefits of physical capital occur almost immediately, the measure of infrastructure spending is not lagged. We obtained these data from the US Department of Transportation’s (2007) Office of Highway Policy Information website. Year-end deposits in Federal Deposit Insurance Company (2007) insured commercial banks from 1989 to 2005 provided the basis for the measure of mean savings deposits. We accessed these data through the FDIC’s website. State population data from 1988 to 2005 found in Statistical Abstracts of the United States were used to determine growth in population. To construct per capita measures, the spending variables, savings, and the dependent variables were all divided by the relevant annual state population data found in Statistical Abstracts of the United States. For the variables utilizing school-year data or measuring growth between two years, population data from the most recent year were used to determine per capita measures.9 Table 1 presents the definitions for the variables investigated and the labels for the variables as they appear in the tables and figures that follow.
Table 1

Variable labels and definitions

Label

Definitions

Income

Average growth in current per capita income (1988–1989 to 2004–2005 growth)

K12 expends

Average K12 current state operating expenditures per capita lagged 13 years (1975–1976 to 1991–1992 school years)

HEd expends

Average higher education state expenditures for operations per capita lagged 7 years (1982–1998)

Hway expends

Average state disbursement for highways per capita (1989–2005)

Savings

Average state deposits in FDIC insured commercial banks per capita (1989–2005)

Pop growth

Average growth in state population (1988–1989 to 2004–2005 growth)

Pup-teacher ratio

Average K12 pupil to teach ratio lagged 13 years (1976–1992)

% HS graduates

Average percentage of the population 25 years or older with a high school degree or more (1989–2005)

% College graduates

Average percentage of the population 25 years or older with a bachelors degree or higher (1989–2005)

Taxation

Average general revenue from state taxes per capita (1997–1998 to 2004–2005 fiscal years)

To determine direct and indirect effects on income growth, we conducted path analyses using ordinary least squares multiple regression. We ran initial regressions with per capita income growth as the dependent variable and the five independent variables and three intervening variables as predictors. When initial regressions revealed different intervening variables to be related to income growth, we then tested for indirect effects on income growth through regressions involving the intervening variables as criterion (dependent) variables and the hypothesized independent variables as predictors. In other words, we regressed (1) pupil-teacher ratios on population growth and K12 expenditures, (2) high school attainment on K12 expenditures, and (3) college attainment on spending on higher education. In turn, the predictors that were not significantly related to income growth were eliminated from the path analysis and the regressions were rerun to determine the direct path coefficients to income growth. Ultimately, the path coefficients appearing in Fig. 2 are the standardized beta weights of the (1) significant predictors of income growth and (2) significant predictors of the intervening variables (Meyers et al. 2006). The total effect of an independent variable on income growth is the sum of its direct and indirect effects with indirect effects measured as the product of the effect of the independent variables on the intervening variables times the effect of the intervening variables on income growth.10
https://static-content.springer.com/image/art%3A10.1007%2Fs11077-008-9062-2/MediaObjects/11077_2008_9062_Fig2_HTML.gif
Fig. 2

Revised growth models

Results

We first examine the results with per capita income growth as our dependent variable and then turn to the relationship between the independent and intervening variables. Table 2 reveals that, between 1988 and 2005, spending on higher education and highway spending have significant positive relationships with growth in per capita income, while K12 expenditures and K12 pupil–teacher ratios have significant negative relationships with income growth. Table 2 further reveals moderately large standardized beta coefficients, and the model explains a moderate amount of variance in income growth (R2 = .44). Savings and population growth fail to demonstrate a relationship with income growth, as do two intervening variables—high school attainment, and college attainment.
Table 2

Multiple regressions predicting per capita income growth over three periods

Predictors

Income 1988–2005

Income 1988–1996

Income 1997–2005

Beta

SE

Beta

SE

Beta

SE

K12 expends

−.363*

.000

−.207

.000

−.192

.000

HEd expends

.466***

.000

.199

.000

.350**

.000

Hway expends

.309*

.000

−.020

.000

.445**

.000

Savings

−.106

.000

−.084

.000

−.069

.000

Popn growth

−.156

.065

−.133

.117

−.050

.111

Pupil–teacher ratio

−.272*

.001

−.096

.001

−.343*

.001

% HS graduates

.224

.000

.341*

.000

−.001

.000

% College graduates

−.077

.000

−.549***

.000

.312*

.001

R2

.436

 

.296

 

.444

 

p < .10; ** p < .05; *** p < .01

n = 48

To explore the applicability of the path model for sub-periods, we tested for direct effects on per capita income growth from 1988–1989 to 1996–1997 and from 1997–1998 to 2004–2005 (see Table 2). From 1988–1989 to 1996–1997, high school attainment demonstrates a significant positive relationship with income growth, while college attainment demonstrates a significant negative relationship with income growth. The model explains almost 30% of the variation in income growth and the significant standardized coefficients are moderately large to large. For the 1997–1998 to 2004–2005 period, two of the independent variables—spending on higher education and highways spending—reveal significant positive associations with income growth (see Table 2). In turn, of the intervening variables, pupil–teacher ratio demonstrates a significant negative association with income growth, while college attainment demonstrates a significant positive association with income growth. Overall, the standardized beta coefficients are sizeable and the model explains a moderate amount (R2 = .44) of variance in per capita income growth.

Since the lagged measure of K12 pupil–teacher ratio is significantly related to income growth during the 1988–1989 to 2004–2005 period, we tested for the indirect effects of K12 spending and population growth on income growth through their relationship with K12 pupil–teacher ratios. A regression reveals that both independent variables are significantly related to K12 pupil–teacher ratios, their standardized coefficients are moderately large, and they explain 40% of the variation in K12 pupil–teacher ratios (see Table 3).
Table 3

Multiple regressions predicting the intervening variables significantly related to per capita income growth

Predictors

1976–1992

1989–1996

1989–1996

1985–1992

1998–2005

K12 PT ratio

HS graduates

College graduates

K12 PT ratio

College attainment

Beta

SE

Beta

SE

Beta

SE

Beta

SE

Beta

SE

K12 expends

−.346**

.003

.453***

.005

  

−.382***

.002

  

Popn growth

.456***

26.133

    

.377***

27.234

  

HEd expends

    

−.132

.010

  

−.327**

.001

R2

.402

 

.205

 

.018

 

.373

 

.107

 

** p < .05; *** p < .01

n = 48

Because high school and college attainment—intervening variables—are significantly related to income growth from 1988–1989 to 1996–1997, they were regressed with K12 expenditures and spending on higher education respectively to test for indirect effects on income growth. The results indicate that K12 expenditures are significantly related to high school attainment, while spending on higher education is not related to college attainment (see Table 3).

Because pupil–teacher ratio is significantly related to income growth from 1997–1998 to 2004–2005, it was regressed with K12 expenditures and population growth to test for indirect effects on income growth. The regression demonstrates that these variables are significantly related to K12 pupil–teacher ratios and are therefore indirectly related to income growth (see Table 3). Moreover, since college attainment is related to income growth during this period, it was regressed with spending on higher education to test for indirect effects on income growth. Table 3 reveals that spending on higher education has a significant negative relationship with college attainment and therefore a negative indirect relationship with income growth.

Several scholars (Fisher 1997; Helms 1985; Miller and Russek 1997) emphasize the importance of controlling for the effects of taxation when trying to discern expenditure effects. Fisher (1997, p. 63) specifically asserts:

If the revenue side of the government budget is not fully specified in a study of service effects [such as education effects], the coefficients on the service variables would effectively capture the combined effect of increased spending partly financed by an increase in the omitted taxes which should lead to an underestimate of the services effect.

In light of this concern, we tested the full model including a measure of state taxes per capita as a predictor. A reliable measure of state taxes is available through the US Census Bureau (2007b) website for the 1997–1998 to 2004–2005 fiscal-year time period. We divided the tax measure by the relevant population data described earlier to establish a per capita measure of state taxes.

This model produces the same significant results as the model without taxes (see Table 2) for the eight-year period investigated (see Table 4). More importantly, controlling for the effects of taxation enhances three of the four significant Beta coefficients and reduces one.
Table 4

Multiple regression predicting per capita income growth with taxation in the model

Predictors

Income 1997–2005

Beta

SE

K12 expends

−.094

.000

HEd expends

.356**

.000

Hway expends

.492***

.000

Savings

−.064

.000

Popn growth

.035

.111

Pupil–teacher ratio

−.309*

.000

% HS graduates

−.037

.000

% College graduates

.363**

.000

Taxation

−.183

.000

R2

.463

 

* p < .10; ** p < .05; *** p < .01

n = 48

Discussion

In sum, this study demonstrates that, between 1988 and 2005, state per capita income in the US increases as expenditures on higher education and highways increase, and, contrary to hypothesis one, as K12 expenditures decrease (see Fig. 2).11 As hypothesized, per capita income increases as K12 pupil–teacher ratios decrease, and population growth indirectly affects income growth through its positive association with K12 pupil–teacher ratios. Between 1988 and 1996, high school attainment demonstrates the hypothesized positive relationship with income growth, and college attainment reveals an unpredicted inverse relationship with income growth. K12 expenditures also reveal indirect effects on income growth through their predicted positive relationship with high school attainment. Between 1997 and 2005, higher education expenditures, highway expenditures, college attainment, and K12 pupil–teacher ratios all demonstrate their hypothesized associations with income growth. K12 expenditures and population growth, in turn, reveal hypothesized indirect relationships with income growth through reducing and enhancing K12 pupil–teacher ratios respectively. Contrary to expectations, expenditures on higher education show an indirect relationship with income growth through a negative relationship with college attainment. Finally, when taxation per capita is added to the model, the paths in the 1997–2005 revised model in Fig. 2 do not change, and the size of significant beta coefficients generally increase as predicted.

While the significant findings generally support the hypotheses, three do not. Although Hanushek (2003) and critics of input-based educational policies provide plenty of theory and evidence suggesting that increasing educational expenditures improves neither educational performance nor economic growth, no literature supports our significant negative relationship between K12 spending and growth. Given the small population size (n = 48 states) and the number of independent and intervening variables tested (eight), such findings are always potentially a function of a shortage of cases (Green 1991; Milton 1986). However, bivariate correlations reveal an insignificant relationship between K12 spending and income growth for the 1988–2005 period, and a statistically significant positive relationship between these variables for the 1997–2005 period.

Is it possible that increased educational spending at the K12 level actually impedes growth? Before drawing such a conclusion, we should reiterate that educational spending does significantly reduce pupil–teacher ratios, as predicted, and that smaller pupil–teacher ratios are related to income growth (for supporting evidence, see also Krueger 2003). So, to the extent that increased expenditures are used to hire more teachers and reduce class sizes, they can enhance growth. Otherwise, our results suggest some possible limitations to the effectiveness of additional K12 spending in promoting growth.12 This conclusion reinforces the single point on which both proponents and critics of input-based educational policies seem to agree—that how educational funding is spent is more important than the sheer amount spent (Picus 1995).

The relationship between college attainment and income growth that switches from negative to positive between the 1988–1996 and 1997–2005 periods is difficult to explain, but may also partially reflect the statistical limitations of this study. It may also be a partial function of rapid technological change in the US. Jorgenson and Stirroh (2000, p. 126) assert that the late 1990s experienced “exceptional” economic growth and heightened demand for a highly skilled labor force in contrast to the preceding 25 years. Prior to 1997, cheap labor without college degrees may have been more central to state income growth than college-educated labor that was bearing expenses with less optimal returns. However, by the late 1990s, it may have become clearer to state policymakers that the way to attract businesses in the growing hi-tech sector was to cultivate and retain a skilled workforce through higher education. Galor and Moav (2000, p. 495) show that increases in the rate of technological progress raise the “return to ability,” wage inequality, and the incentive to seek higher education. Last, the positive association between college attainment and income growth in the late 1990s and early 2000s may reflect the growth of the service sector and the globalization of the American economy, both necessitating a more highly skilled and educated labor force.13

Finally, the negative association between spending on higher education and college attainment rates between 1997 and 2005 might reflect the migration of college graduates to states that do not invest as heavily in higher education (Miller and Russek 1997). This negative association may also reflect that states with the lowest investments in public higher education—Massachusetts, Connecticut, Vermont, and New Hampshire—have high per capita enrollments in private systems of higher education (Goldin and Katz 2001) that are graduating a substantial percentage of their state’s population. Regardless, the indirect negative effect of higher education spending on income growth, (.08, or the product of the −.33 and .24 path coefficients) is smaller than its positive direct effect (r = .31) presented in Fig. 2.

In general, human capital explanations for economic growth at the cross-national level may be less applicable to sub-national economic growth in the U.S. because of the mobility of American labor. For example, Rustbelt (i.e., Midwestern industrial states) to Sunbelt (South and Southwest) shifts in population prevent Rustbelt states from enjoying the full benefits of generous education spending because their state-trained human capital often moves to work in warmer climates (U.S. Census Bureau 2007c). Moreover, Jaschik (2007) and Rogers and Heller (2003) note that “brain drain,” college-educated residents leaving their home states, has become salient even in states that enjoyed favorable business climates.14

That spending on higher education is significantly associated with income growth is consistent with contemporary trends and approaches to economic growth in the US. States such as Michigan and Pennsylvania have partnered with universities to develop technology and business incubators (e.g., “Smart Zones,” and life science corridors), while Kentucky has developed programs (e.g., “Bricks for Brains”) to attract better faculty and to retain their best and brightest graduates (Finkle 2002). Georgia and Ohio, in turn, have developed umbrella intermediary organizations to help unite the resources and efforts of higher education with those of business, government, and the non-profit sectors (Plosila 2004). In general, state growth efforts have used institutions of higher learning to sponsor growth-related research, to access cutting-edge technology and equipment, to provide customized training and school-to-work programs, to develop lifelong learning programs, to provide technical assistance and problem solving, and to provide training programs and assistance in entrepreneurship (Finkle 2002; Plosila 2004). State growth efforts have also followed the lead of many Western European countries in applying internet communication technology to enhance human capital through making higher education more accessible (Huisman and Kaiser 2002).

That our findings support the importance of highway expenditures reinforces a traditional approach toward economic growth that does not appear to receive as much attention in the current state growth literature in America as it did in the late 1980s and early 1990s (Gramlich 1994). Finkle (2002) emphasizes the importance of transportation and infrastructure in economic growth efforts aimed at reviving inner cities and managing urban sprawl; yet, contemporary trends reported in the literature focus more on the building of technological infrastructure for economic growth (Finkle 2002; Huisman and Kaiser 2002; Plosila 2004). By contrast, our finding that pupil–teacher ratios are associated with income growth reinforces the contemporary trend in states such as California, Colorado, Georgia, Indiana, Nevada, New York, Tennessee, and Wisconsin to adopt initiatives to reduce their class sizes typically through state subsidies (Jacobson 2001; Murray et al. 2007).

Conclusion

Overall, our findings indicate that the three most consistent predictors of income growth are expenditures on higher education, highway expenditures, and K12 pupil–teacher ratios. They consequently contribute to the debate over the effects of class size, by supporting the body of research asserting that smaller classes make a positive difference (e.g., Burr 2001; Glass and Smith 1978; McGiverin et al. 1989). The findings also reinforce the utility of path analysis for understanding the dynamics of economic growth. All three of the hypothesized mediators demonstrate relationships with independent variables in at least one time-period investigated. Given the paucity of path analyses, the potential complexities of causal relationships, and the need to understand the effects of government spending, our findings provide some support for the continued application of path analysis in economic growth studies within and between countries. Given that we found controlling for taxation generally increases the coefficients of predictors of income growth, we also advocate continued inclusion of taxation measures in economic growth models to further discern where or methodologically how taxation causes the underestimation of service effects.

Our research also points to the value of utilizing predictors from the cross-national economic growth research. Three of the five indicators relevant to the cross-national research explain per capita income growth in at least one of the time-periods explored. Although the phenomena that predict economic growth in the states of the US may be idiosyncratic to specific time-periods and measures of growth, we suspect that cross-national predictors of economic growth have the potential to add to the explanation of economic growth across political subdivisions of other large countries with federated arrangements of power. No doubt, exposure to the enormous body of cross-national research enriched our understanding of state economic growth. We suspect that exposure to the cross-national research on health care, law enforcement, social welfare, environmental affairs, and human rights would similarly benefit scholars engaged in intra-national research in these areas.

In conclusion, efforts to enhance educational spending based on its presumed effect on economic growth will continue to be orchestrated in the chambers and lobbies of state legislatures and gubernatorial offices in the US. Our research lead us to support reductions in K12 pupil–teacher ratios and augmentations to the operating expenses of higher education as means of enhancing growth in per capita income. However, given the escalating cost of higher education in the twenty-first century and the need to burden increasingly students, families, and donors to operate US public institutions of higher learning, the combined effects of multiple revenue sources remain unclear and an important direction for future research (SHEEO 2007; US Census Bureau 2008, p. 183). Finally, consistent with a common refrain in the cross-national research, our research echoes the virtues of investing in highways in order to stimulate economic growth in the states of the US.

Footnotes
1

Developing a theoretical model of education policy from beginning to end, especially one that resolves the disputes over the placement of educational variables, is beyond the scope of this article. However, we believe that efforts to acknowledge that education policy can be depicted as a causal process are productive and that it is time to begin sequencing processes in our empirical models involving the relationship between education and economic growth. So, after our first-cut analysis, which includes K12 spending, spending on higher education, pupil–teacher ratios, and high school and college attainment rates as simultaneous predictors of economic growth, we move to a more realistic path analysis that depicts education spending as a causal antecedent of pupil–teacher ratios and high school and college attainment rates. Even if our path analytic approach does not rehabilitate educational spending as a predictor of economic growth, we think its estimates are more reliable because the specification is more realistic.

 
2

Hefner (1990) found that regional GSP is not cointegrated with national GDP, nor is there an equilibrating mechanism that prevents regions from drifting apart from each other. Moreover, given that Nijkamp and Poots' (2004) meta-analysis demonstrates an absence of discernable effects of national versus regional data, we suspect that the cross-national predictors of economic growth are likely to be relevant to the political subdivisions of geographically large and economically diverse federations such as the US.

 
3

The US demonstrates regional patterns in manufacturing and retail trade, and various regional patterns emerge as NAIC classifications narrow (US Census Bureau 2002a). Various state economies are also quite similar with respect to social assistance, health care assistance, and professional, scientific, and technical services. However, no region of the US dominates in the major industrial classifications of wholesale trade, utilities, or construction.

 
4

Morgan et al. (2004) find that, as bank linkages between two states increase, the fluctuation in a state’s economic growth becomes less volatile and converges with the growth fluctuations of the state in the paired linkage.

 
5

States with a high level of urban poverty may have a difficult time demonstrating impressive high school attainment rates because of the high drop-out rates of poor students in poorly funded inner city public school districts. Recent academic research supports the idea that inequalities in the distribution of funding within the state (Roscigno et al. 2006) and/or inefficiencies in the education bureaucracy (Ragkumar and Swaroop 2007) can disrupt the aggregate correlation between spending and high school attainment. A recent study funded by the Bill and Melinda Gates Foundation and Colin Powell’s America’s Promise Foundation highlights the alarmingly high dropout rates in many of the US’ largest urban school districts (Swanson 2008). That many of these high-dropout districts are located in states with traditions of relatively high state funding for education (e.g., Michigan, Ohio, Minnesota, and New York) would seem to cast doubt on the face validity of the hypothesis that spending is associated with higher attainment rates.

 
6

Hanushek et al. (1996) argue that aggregate, state-level analyses of the impact of educational resources on various outputs and outcomes exaggerate the effects of school resources to a much greater extent at the state level than at the school level because of omitted-variable bias. Just because state-level spending might be correlated with state-level growth does not mean that the impact of spending is causal. Omitted parental or community variables—correlated with both school resources and economic growth—could possibly be distorting our aggregate-level results. However, we conducted an aggregate-level analysis because we mirrored the pattern of most scholars to investigate state-level data and because state-level data for a large number of years are much more readily available than district-level, school-level, classroom-level, or student-level data.

 
7

Per capita income is measured in current dollars instead of constant dollars because the data sources (Statistical Abstracts and The Survey of Current Businesses) provided 2000–2005 per capita income data in 2000 constant dollars, while providing 1999 and 2001 per capita income data in 1996 constant dollars.

 
8

We assume that educational spending enhances the content of what is learned, which has an eventual payoff in terms of economic growth. However, if education systems do not adjust to the nature and demands of the economy, then the explanatory value of expenditure variables that are lagged based on how long it takes to complete a degree will have limitations. For example, if K12 and higher education systems in the Rustbelt states of the Midwest are slow to adjust their curricula to accommodate the shift from a manufacturing-based economy to a more service-based economy, then the lagged effects of education spending will be limited. More appropriate but less identifiable lags might be the number of years that it takes K12 and higher education personnel to adjust their thinking about curricular needs sufficiently enough to override the “politics of inertia” and the lingering appeal of popular programs that no longer work.

 
9

For example, in determining K12 expenditures per capita for the 2004–2005 school year, we divided the 2004–2005 K12 expenditures by the state population in 2005 versus 2004.

 
10

For example, the effect on income growth of spending on higher education is the sum of the beta coefficient for its direct effect on income growth plus the beta coefficient for the relationship between expenditures on higher education and college attainment rates multiplied by the beta coefficient for the relationship between college attainment rates and income growth.

 
11

The path coefficients in Fig. 2 for the relationships between the independent and intervening variables are the significant standardized Beta coefficients from Table 3. The path coefficients for the relationships between the independent variables and income growth are the significant standardized coefficients from Appendix 1.

 
12

This conclusion does not negate the importance of investing in K12 education. High school drop-out rates in the US range from 3.3 to 4.7% between 2000 and 2005, and the earnings of high school drop-outs are substantially lower than those of high school and college graduates (US Census Bureau 2008, p. 171, 453). Spending aimed at preventing high school attrition is undoubtedly a worthwhile effort for enhancing per capita income. Moreover, even when educational spending translates into economic growth, concerns over how dollars are spent and whether those dollars optimize returns to human capital persist. To optimize learning experiences that transfer to contemporary work environments, concerns over investments in curriculum, teaching methods, classroom technology, and methods of assessment are legitimate.

 
13

These speculative explanations focus more on the recent positive association between college attainment and state income growth. They are less adept at explaining the negative association between college attainment and income growth from 1988 to 1996. And, they are not meant to imply that, prior to the years investigated (pre-1988), college attainment has always had a negative association with economic growth in US states because of slower technological progress, a smaller service sector, and a less global economy.

 
14

A general examination of the data reveals that three Rustbelt state—Ohio, Michigan, Illinois—have particularly low income growth rankings, especially during the latter period investigated where they represent three of the four lowest ranking states in per capita income growth (see Appendix 2). No Rustbelt states ranked in the top 10 states for income growth from 1988 to 1996 or from 1997 to 2005. Further, Minnesota is the only Rustbelt state to rank among the top ten states (number eight) in income growth for the entire period investigated. If regional dummy variables are added to the model tested in Table 2 using the Rustbelt states as the referent variable, the South and the West also have significantly larger associations with per capita income growth than do the Rustbelt states from 1997 to 2005. However, from 1988 to 2005, the West emerges as the only regional variable with a significantly more positive relationship with income growth than the Rustbelt states.

 

Copyright information

© Springer Science+Business Media, LLC. 2008