Federal competition and economic growth

Journal of Public Economics 97 (2013) 144–159

Contents lists available at SciVerse ScienceDirect

Journal of Public Economics journal homepage: www.elsevier.com/locate/jpube

Federal competition and economic growth☆ John William Hatfield ⁎, Katrina Kosec Graduate School of Business, Stanford University, United States

a r t i c l e

i n f o

Article history: Received 8 April 2011 Received in revised form 31 July 2012 Accepted 6 August 2012 Available online 30 August 2012 JEL classification: H77 H11 Keywords: Economic growth Inter-jurisdictional competition

a b s t r a c t This paper exploits exogenous variation in the natural topography of the United States to estimate the causal impact of inter-jurisdictional competition on income growth. We find that doubling the number of county governments in a metropolitan area leads to a 17% increase in the average annual growth rate of earnings per employee over 1969–2006, and a 10% increase in 2006 income per employee. Decomposing income effects using 2000 Census worker-level data, we find that approximately half of the effect stems from making workers more productive, while the other half comes from changing the composition of the workforce and inducing workers to work more hours. We also present evidence that inter-jurisdictional competition leads local governments to raise more in taxes, spend more, and issue more debt (per capita), but does not help them obtain more inter-governmental transfers. However, the additional cost from this increase in expenditures to a median-wage employee is much smaller than the increase in that employee's wages due to greater inter-jurisdictional competition. © 2012 Elsevier B.V. All rights reserved.

1. Introduction Decentralization is a key component of institutional reform around the world. Dillinger (1994) reports that all but twelve of the world's seventy-five largest countries claim to be devolving political power to local governments, motivated by the goals of economic growth and a higher standard of living. However, the economic effects of such devolution of power are hotly debated.1 Most empirical work on decentralization is comprised of cross-country studies, and faces at least two methodological problems: first, defining a measure of decentralization that can be consistently measured for all cross-sectional units, and second, addressing the endogeneity of institutional choice to economic outcomes like growth. As a result, previous work has not reached firm conclusions.2

☆ We are grateful to Lakshmi Ayer, Matilde Bombardini, Daniel Elfenbein, Witold Henisz, Saumitra Jha, Claire Lim, Neil Malhotra, John Maxwell, Felix Oberholzer-Gee, Andrew Postlewaite, Jonathan Rodden, Dean Stansel, Francesco Trebbi, and Romain Wacziarg for helpful discussions. We would also like to thank Caroline Hoxby and Jesse Rothstein for providing us with their own data on streams and rivers in U.S. metropolitan areas. ⁎ Corresponding author. E-mail addresses: hatfi[email protected] (J.W. Hatfield), [email protected] (K. Kosec). 1 Kim et al. (1995), Huther and Shah (1998), Akai and Sakata (2002), Stansel (2005) and Hammond and Tosun (2006) find a positive effect of decentralization on growth, while Davoodi and Zou (1998) and Zhang and Zou (1998) find a negative effect. See Boadway and Shah (2009) for a summary of this literature. 2 See the discussion of Oates (1993), Ebel and Yilmaz (2002), and Rodden (2004) below. 0047-2727/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jpubeco.2012.08.005

In this paper, we consider how inter-jurisdictional competition—a defining feature of decentralization—affects economic growth. The presence of inter-jurisdictional competition is perhaps the most important difference between local and national policymaking environments: local governments, much more than national governments, face competition for investment and residents. We study metropolitan areas (MSAs) in the United States and use the number of county governments in them as our central measure of the degree of inter-jurisdictional competition. We find that such competition is a powerful determinant of growth. Specifically, doubling the number of county governments in an MSA—such as by going from one to two—leads to an approximate 0.15 percentage point increase in the average annual growth rate of earnings per employee over 1969–2006. This effect is relatively large and meaningful, amounting to an average annual growth rate in earnings per employee that is 17% higher than average. These results are robust to controlling for county founding years as well as pre-period values of income, population, and racial composition. There is a large theoretical literature on the effects of decentralization on growth. First, Hayek (1945) argues that local officials have better information on optimal provision levels, and thus supply publicly-provided goods more efficiently. Similarly, Tiebout (1956) argues that having many jurisdictions allows individuals to sort by taste, leading to more efficient provision. Second, work by Besley and Case (1995) and Seabright (1996) on “yardstick competition” suggests that having many jurisdictions allows voters to measure outcomes against those in similar jurisdictions, facilitating monitoring of political agents. Finally, several scholars have emphasized the effects of competition for resources like residents and investment. Brennan and Buchanan (1980) emphasize the ability of inter-jurisdictional competition to restrain the power of monopoly local governments over citizens,

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

while Zodrow and Mieszkowski (1986) argue that inter-jurisdictional competition may yield a “race to the bottom” whereby productivityenhancing goods are under-provided by sub-national governments attempting to attract taxable but mobile factors of production.3 Hatfield (2010), by contrast, provides a model where competition for capital drives districts to provide productive public goods at levels which maximize economic growth.4 This debate motivates our empirical analysis. The empirical literature on decentralization and growth can be roughly divided into two categories. The first is cross-country analysis, where economic growth is regressed on a measure of decentralization such as local revenue share or local expenditure share. The findings are mixed, as Kim et al. (1995), Huther and Shah (1998), and Iimi (2005) find a positive effect of decentralization on growth, while Davoodi and Zou (1998) find a negative one, and Woller and Phillips (1998) do not find any significant relationship. However, these studies face some methodological concerns. For instance, Rodden (2004) shows that a unidimensional measure of federalism cannot quantify how the relationship between local and national governments varies across countries.5 The second category studies outcomes in one country, considering growth as a function of inter-jurisdictional competition within a sub-national unit. By concentrating on one country, we can more confidently measure competition the same way across localities. For example, Stansel (2005) considers inter-jurisdictional competition in the U.S., and finds a positive effect on growth. However, all empirical work in this area faces a significant challenge to identification, described by Oates (1993): “Is decentralization a ‘cause’ or an ‘effect’ of economic development?”6 Causality has not been well-established by the existing literature. Some papers, such as Panizza (1999), even estimate the effect of income (among other factors) on decentralization, rather than considering income to be determined by the level of decentralization.7 There are also several possible sources of omitted variable bias that might downward-bias any growth or income benefits of inter-jurisdictional competition. For example, being racially heterogeneous or having poor weather and mountainous terrain might make an MSA both poor and slow to grow, but also more likely to have more jurisdictions. To overcome these empirical difficulties, we focus on one nation— the United States—and consider how variation in the number of competing jurisdictions across metropolitan statistical areas (MSAs) affects growth. To address threats to identification, we implement an instrumental variables strategy inspired by Hoxby (2000): we use the total miles of small streams in an MSA to instrument for its number of county governments. We argue that, while small streams are unlikely to directly affect growth in the modern era, more streams increased the number of natural “break-points” between counties at the

3 Wilson (1984) and Wildasin (1987) provide similar insights. Wilson (1999) summarizes this literature. 4 Similarly, Brueckner (2006) provides a model where inter-jurisdictional competition enhances incentives to invest in human capital, which boosts economic growth. Weingast (1995) and Hatfield and Padró i Miquel (2012) argue more generally that inter-jurisdictional competition can enhance incentives for long-term productive investments. Montinola et al. (1995) apply these ideas to China's recent economic growth. 5 Ebel and Yilmaz (2002) additionally point out that the International Monetary Fund's Government Finance Statistics—frequently used in this line of empirical work – poorly measure the degree of decentralization. They do not differentiate between discretionary and non-discretionary spending by local governments, fail to capture great variance in the level of local borrowing authority enjoyed by sub-national units, and fail to capture differing beliefs over whether debts are, in the end, the responsibility of the national government. 6 A similar point is made in Bardhan (2002). 7 Recent work by Calabrese et al. (forthcoming) approaches this identification issue by instead using parameters estimated from a computational model to assess the welfare implications of one facet—Tiebout sorting—of the effect of having multiple competing jurisdictions.

145

time of an MSA's founding, leading to more counties and thus more inter-jurisdictional competition. 8 We also investigate whether our finding that inter-jurisdictional competition enhances growth may be due to MSAs with many county governments having relatively low incomes in the 1960s; if this were true, then our results might be due to conditional convergence. By contrast, we find that doubling inter-jurisdictional competition is associated with a 1969 income per employee that is 5% higher, but with a 2006 income per employee that is 10% higher. Greater inter-jurisdictional competition was already associated with higher incomes in 1969, and the disparity only grew over the next 37 years. Investigating the causes of these different outcomes, we find that approximately one-third of the effect comes from the fact that MSAs with more competition attract more productive workers. Areas with more competition also induce workers to work longer hours: approximately one-fifth of the effect is due to this factor. The remaining portion comprises the direct effect of competition on hourly wages for a given worker: doubling inter-jurisdictional competition leads to 5% higher wages. We also show that MSAs with more competition have a greater percentage of economic activity in more remunerative industries like finance, insurance, real estate, management, and information, and a smaller percentage in less remunerative industries like entertainment, recreation, and retail trade. Finally, we show that MSAs with more competition raise more revenues via taxes, have greater expenditures, and issue more debt, but do not receive more transfer payments from state and federal governments. However, the additional costs to a median-wage employee from additional taxes (and debt) are much smaller than the associated increase in that employee's annual wages due to greater inter-jurisdictional competition. The paper is organized as follows. Section 2 describes the dataset and our empirical approach. Section 3 presents the main empirical results; Section 4 presents a variety of robustness checks. Section 5 investigates possible causal channels. Finally, Section 6 concludes. 2. Empirical strategy We investigate the effect of inter-jurisdictional competition on economic growth using data from Metropolitan Statistical Areas (MSAs) and Consolidated Metropolitan Statistical Areas (CMSAs) in the United States. We refer to the collection of MSAs and CMSAs as MSAs. MSAs are comprised of an urbanized nucleus with a population of at least 50,000 and the collection of adjacent communities that have a high degree of integration with that nucleus (as evidenced by commuter patterns). 9 Geographically, MSAs are defined by the set of counties of which they are comprised. 10 See the Appendix for more information on MSAs. We address identification and the measurement of economic growth and inter-jurisdictional competition within an MSA below, in Section 2.1. For now, assume that these variables are accurately measured and that all variation in inter-jurisdictional competition is exogenous. We estimate the following empirical specification: g i ¼ β0 þ βN logðNi Þ þ γXi þ α j þ εi

ð1Þ

8 Our analysis circumvents criticisms related to measuring streams and using them as an IV (see, e.g., Rothstein, 2007) by using GIS data to ensure objective and consistently-applied definitions of streams. 9 MSAs and CMSAs are mutually exclusive entities. CMSAs are relatively larger than MSAs and contain multiple urbanized nuclei (called Primary Metropolitan Statistical Areas, or PMSAs). Since PMSAs are integrated with one another, MSAs and CMSAs are our units of analysis (as opposed to MSAs and PMSAs). 10 An exception is New England, where MSAs cross county boundaries and contain only portions of some counties. As data on many of our covariates are not available at more disaggregated levels than the county, we exclude New England MSAs from our analysis. Additionally, we had to exclude three other MSAs for which the boundaries of the counties within them changed over time, preventing the collection of comparable data over time. In total, we have 222 MSAs in our sample for which we have data on all of our covariates.

146

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

where i indexes MSAs. We denote by gi the average annual growth rate of income per employee over 1969–2006, by Ni the number of competing jurisdictions in MSA i, and by αj state group fixed effects.11 Xi is a vector of controls for MSA i, described in Section 2.2.3. 2.1. Identification There are several possible sources of omitted variable bias. First, areas with ethnic, racial, or political tensions may self-segregate into many different, internally-homogenous jurisdictions. However, these same tensions may lead to violence and inefficiently low levels of interaction, lowering incomes and growth. Second, places with bad weather or rugged terrain (e.g., frequent snow and rain, or mountains and valleys) might be hard to move across quickly, reducing income and growth. However, uncooperative weather patterns and terrain might also make it more sensible to have many different governments, since operating and administering a given-sized area may be relatively more difficult. Finally, older MSAs might have more jurisdictions since travel was more difficult hundreds of years ago when their boundaries were drawn. However, older MSAs may be the seats of prosperous industries that have since declined given changes in the world economy—making it harder for them to grow past their 1969 income levels. All of these sources of omitted variable bias would likely downward bias any growth benefits of inter-jurisdictional competition. Inter-jurisdictional competition might also be a response to economic growth—a problem of reverse causality. MSAs with growing incomes might attract such a large collection of people and industries that more local jurisdictions develop to deal with the influx. This would tend to upward bias the coefficient on inter-jurisdictional competition. To address these threats to identification, we exploit exogenous variation in the natural topography of the U.S. Specifically, we instrument for the number of jurisdictions in an MSA with the number of miles of small streams, computed using Geographic Information System (GIS) data from the Environmental Systems Research Institute's Data and Maps (2008). This source allows us to count all small streams, but omit major rivers and manmade features (like canals and aqueducts). See the Appendix for more information on the measurement of streams. Small streams should lead to more jurisdictions (and thus more competition), but should be uncorrelated with unobserved factors that increase the potential for commerce and income growth. The history of county formation in the U.S. motivates our IV strategy. The median metropolitan county was founded in 1848, and boundaries have changed little since. When most counties were founded, geographic obstacles like streams were focal “break-points” between jurisdictions because they were easy to describe and communicate (something especially important before GPS and GIS). Historian Farris Cadle underscores their importance: Georgia's twelve parishes, which existed before the Revolution, were arrayed in a single upright tier stretching from the St. Mary's River on the south to the Broad River on the north. The Atlantic Ocean and the Savannah River formed the eastern limits, while their northern and southern boundaries were delineated by creeks flowing into these bodies of water…Following the Revolution, extensive migration into the hinterland made it necessary to define with precision the buttings and boundings of the growing proliferation of counties…The acts providing for the creation of these new counties relied heavily on well-known roads and natural features such as rivers, streams, and ridges to demarcate boundaries… The acts generally described the artificial lines as running from

11 Because some MSAs cross state lines, we created state dummies not only for each of the 50 states, but also for each of the “state groups” (collections of two or more states) created by multi-state MSAs.

one landmark to another. Bearings or distances were specified only in extremely rare cases. (Cadle, 1991, p. 145) Fig. 1 provides an example of how small streams have contributed to the demarcation of county boundaries in the greater Houston area today. County boundaries frequently coincide with streams. An example of the opposite effect is the Phoenix, AZ MSA, where few streams – and only two county governments — can be found. Our first stage equation states that inter-jurisdictional competition, measured as log Ni, is a function of the number of miles of small streams, si: logðNi Þ ¼ δ0 þ δs si þ θXi þ π j þ ηi

ð2Þ

where πj denotes state group fixed effects. In the next section, we demonstrate that this instrument satisfies the inclusion restriction: it is associated with more inter-jurisdictional competition in MSA i. We argue that the exclusion restriction holds since small streams are easy to get around with modern transportation options and technology. By virtue of being small, they are unlikely to affect modern commerce, economic activity, or growth. Hoxby (2000) presents a similar IV methodology, using a count of small streams as an instrument for the degree of Tiebout choice over schools. She argues that having more streams implies more natural school district boundaries, particularly since these boundaries were chosen long ago, when streams increased travel time to school. However, she maintains that streams are exogenous to modern-day school productivity. 12 There has been some controversy regarding whether a hand count is appropriate and objective (see Rothstein, 2007), but our use of GIS data to measure miles of streams circumvents this criticism. 2.2. Variable measurement and data We use several data sources, matched geographically at the MSA level. Table 1 summarizes the data; an observation is an MSA. Independent variables are measured in the 1960s (so as to pre-date outcome measures), with the year used based on data availability. Dependent variables are measured in the 2000s, with the year used again based on data availability. 2.2.1. Inter-jurisdictional competition The sub-national general-purpose governments in an MSA include county, municipal, and township governments. Counties are the primary legal divisions of most states, and are thus an especially important component of sub-national governance. 13 The roles and jurisdictions of municipal versus township governments vary widely across states, and often governments with the same powers are called municipalities in one state and townships in another. County governments are relatively uniformly-defined local governments, present in all MSAs. Thus, we capture inter-jurisdictional competition using the number of county governments in an MSA. Results instead using the number of municipal plus township governments, or the number of school districts, are presented in Table C.6 in the Appendix, and yield similar findings. When all three types of local government (county, municipal and township, and school district) are included in the same OLS specification, only the coefficient on logged county 12 A few papers since Hoxby (2000) conduct empirical analysis leveraging off of the natural boundaries formed by streams. Baqir (2001) uses streams as an IV for the number of electoral districts in U.S. counties. Cutler and Glaeser (1997) use streams as an IV for the level of segregation in MSAs. 13 We include functional county governments in our count of total county governments. Functional county governments include consolidated city-counties and independent cities. They are local governments with county-like control over an area that is not under the jurisdiction of any county government. An example of the former is San Francisco, California; an example of the latter is St. Louis, Missouri.

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

147

Fig. 1. Map of the counties comprising the Houston–Sugar Land–Baytown MSA, and the MSA's streams.

governments is significant, bolstering the case for focusing on county governments: see Table C.6 in the Appendix. These data come from the Census of Governments, put out by the U.S. Census Bureau (1962). 2.2.2. Income We measure income using data on earnings by place of work per employee for each year during 1969–2006, from the Bureau of Economic Analysis (2008). The BEA defines earnings by place of work as “the sum of wage and salary disbursements (payrolls), supplements to wages and salaries, and proprietors' income.” We divide this by the number of employed persons in an MSA—also available from BEA (2008)—to estimate how much each worker produces. We

found this to be the most appropriate way to measure income since we are primarily interested in how inter-jurisdictional competition affects working and production decisions. We estimate the effects of competition on both the level of income per employee in each year during the period 1969–2006 and on the average annual growth rate of income per employee over 1969–2006. We also estimate the effects of competition on average annual growth during each 10-year interval contained in 1969–2006. Unless otherwise specified, when we refer to income, we are referring to income by place of work per employee. Another reason for focusing on income per employee rather than personal income per capita or another measure was to avoid skewing

148

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

Table 1 Summary statistics. Variable

Mean

Std. dev.

Average annual growth rate of earnings per employee, 1969–2006 Earnings by place of work per employee, 1969 Earnings by place of work per employee, 2001 Earnings by place of work per employee, 2006 Gross municipal product (GMP) per employee, 2001 Number of county governments Log number of county governments Number of municipal and town governments Log municipal and town governments Number of school districts Log number of school districts 100s of miles of streams (ESRI GIS Data) Hoxby (2000) number of small streams Rothstein (2007) total streams Dummy — coastal Dummy — on pacific ocean Dummy — on Atlantic ocean Dummy — on great lakes Dummy — on major river Log land area (1970, 1000s square miles) Cooling degree days, 1970–2000 average Heating degree days, 1970–2000 average Hours of sunshine, average in January (1940–1970) Monthly rainfall, 1970–2000 average Standard deviation of elevation Earliest founding year of a county in this MSA, over 100 Share of workers that are blue collar, 1970 Share of population that is white, 1970 Share of population that is black, 1970 Share of population that is native American, 1970 Share of population that is of another race, 1970 Racial fractionalization index (uses shares of four races, above) Log area of water (1000s square miles) Log earnings by place of work (in 1969) Log population (in 1960) Taxes per capita, 1997–2002 average Income taxes per capita, 1997–2002 average Property taxes per capita, 1997–2002 average Sales and gross receipts taxes per capita, 1997–2002 average Transfer revenues per capita, 1997–2002 average Utility revenues per capita, 1997–2002 average Expenditures per capita, 1997–2002 average Expenditure on central government staff per capita, 1997–2002 Implied interest rate on debt, 1997–2002 average Debt per capita, 1997–2002 average Interest paid on debt per capita, 1997–2002 average Average per capita deficit, 1997–2002 average Number of observations

0.85 25.74 33.78 35.25 50.46 2.53 0.63 37.36 2.84 35.36 2.76 2.92 86.73 154.16 0.40 0.04 0.12 0.07 0.23 0.10 1.17 3.71 1.52 3.32 0.11 18.30 0.35 0.89 0.10 0.00 0.00 0.17 −3.34 3.24 12.10 0.93 0.03 0.68 0.18 1.14 0.29 3.17 0.04 0.06 2.82 0.16 0.03 223

0.37 3.47 5.16 5.14 7.87 2.50 0.71 55.56 1.29 53.70 1.44 2.68 92.47 163.43 0.49 0.20 0.33 0.25 0.42 0.80 0.81 1.95 0.39 1.19 0.17 0.40 0.08 0.10 0.10 0.01 0.01 0.13 1.92 0.14 1.03 0.28 0.07 0.26 0.15 0.41 0.37 0.81 0.02 0.01 2.93 0.16 0.12

our results due to local variation in workforce participation patterns.14 For example, during 1969–2006, women began having fewer children and entering the workforce in significant numbers, but they did so unevenly across the country. Also, retirees became more likely to migrate to particular places—like Arizona and Florida—leading to an influx of non-working individuals in some areas but not in others. 2.2.3. Potential correlates of miles of small streams A possible concern is that small streams are correlated with other natural geographic features that impact growth. We address such concerns by including a variety of topographic and climatic controls in Eq. (2), captured by Xi. These include: an indicator for having one or more coastal counties (National Oceanic and Atmospheric Administration, 2008); indicators for bordering the Pacific Ocean, the Atlantic Ocean, one of the Great Lakes, or a major river; the standard deviation of elevation (Environmental Systems Research Institute's Data and Maps, 2008); the logged land area in thousands 14 The average annual growth rate of real income per employee during 1969–2006 is 0.85. This is far lower than the mean of a more traditional measure: growth in personal income per capita (with a mean of 2.0).

of square miles in 1969 (U.S. Census Bureau, 1970); and various weather controls including average monthly precipitation during 1970–2000 (in inches), average hours of sunshine in January during 1940–1970 (in 100s), and average heating degree days and cooling degree days per month during 1970–2000 (in 100s). 15,16 The inclusion of these controls explicitly allows them to have a direct impact on growth and ensures that any power we get from our instrument is not due to its correlation with them. A second possible concern is that water is simply an attractive feature of an MSA (drawing people and businesses with its natural beauty), and that places with many small streams have more water area. We address this concern in two ways. First, we control for the logged surface area of MSA i covered in water (in 1000s of square miles). We thus rely only on that part of miles of small streams that is uncorrelated with water area. Second, we control for the 1969 value of logged earnings by place of work per employee (Bureau of Economic Analysis, 2008) and the 1960 value of logged population (U.S. Census Bureau, 1962). We thus compare within places with similar initial incomes and populations in the 1960s. Our results are robust to inclusion of these variables. In fact, controlling for 1960s values of population and income per employee reveals a potentially larger impact of competition on growth. A third possible concern is that miles of small streams is simply correlated with an MSA's founding year, or with its racial makeup— factors which might directly impact income and growth. We thus estimate specifications accounting for the share of the population in MSA i that is black, Native American, and “other (non-white) race” (white is the base group); a racial fractionalization index; 17 and for the founding year of the earliest-founded county in MSA i. Race data come from U.S. Census Bureau (1970), while data on county founding years come from the National Association of Counties (2009). A final, remaining concern is that small streams have a direct impact on economic activity. For example, navigating from point A to point B might be less costly without a small stream in the middle. Also, construction costs may be lower without small streams in the way. We argue that this is unlikely simply because streams are small and thus easy to build around. We think it is especially unlikely in specifications that already control for weather, climate, ruggedness of terrain, surface area covered in water, and state group fixed effects—which alone probably capture most of the variation in navigation and construction costs. 18 2.3. Estimating causal channels As our empirical analysis suggests inter-jurisdictional competition leads to higher income and higher income growth, we also consider what mechanisms may help explain the results. We first examine how inter-jurisdictional competition affects the skill mix and demographic characteristics of the workforce, as well as the hours workers supply. We next examine which industries are attracted to places with more inter-jurisdictional competition, and explore whether these differences help explain differences in income and income growth. Finally, we examine how local public finance is affected by the level of inter-jurisdictional competition. 15 Weather data are from National Climatic Data Center (2008). Heating and cooling degree days are used to estimate the energy required to maintain a comfortable indoor temperature. Each day, heating degree days equal max{0, 65− mean temperature} and cooling degree days equal max{0, mean temperature − 65}. 16 We hoped to include measures of soil quality, but the National Soils Database at the U.S. Department of Agriculture (the best source) does not cover highly-urbanized areas, including many counties in MSAs. 17 This is a Herfindahl–Hirschman Index (HHI) that sums the squared population shares of each race. 18 Nonetheless, we note that if streams have a negative, direct impact on growth (while simultaneously leading to more jurisdictions), then this would tend to mute the growth-promoting benefits of inter-jurisdictional competition, making our baseline IV estimates lower bounds.

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

2.3.1. Imputed income decomposition We use detailed data on 4.8 million workers from the 5% sample of the 2000 Census (U.S. Census Bureau, 2000) to decompose the effects of inter-jurisdictional competition into three channels: 19 1. inter-jurisdictional competition could directly cause workers in the MSA to be more productive with each hour of work; 2. inter-jurisdictional competition could lead workers to work more hours (e.g., because each hour is more productive); 3. inter-jurisdictional competition could attract a more productive set of workers to the MSA (i.e., lead to a change in the composition and demographic characteristics of the work force that raises income per employee). To empirically separate these three channels, we first run three worker-level regressions: 1. logged annual income on MSA fixed effects, 2. logged annual income (i.e., salary) on MSA fixed effects and a vector of worker-level demographic controls 20, and 3. logged hourly income (i.e., wage) on MSA fixed effects and a vector of worker-level demographic controls. The MSA fixed effects from the first regression equal the average annual earned income per worker in each MSA. The MSA fixed effects from the second regression are the income differentials (“imputed income”) that capture the effect of living in an MSA on annual income per worker, independent of what types of firms and workers the MSA attracts. And the MSA fixed effects from the third regression are wage differentials (“imputed wages”) which net out not only firm and worker characteristics, but also the effect of the MSA on how many hours of labor workers supply. By regressing the different MSA fixed effects from these three worker-level regressions on inter-jurisdictional competition, we can decompose the effects of inter-jurisdictional competition between the three channels. Briefly, the coefficient on competition in regressions of imputed hourly wages on competition is the direct impact of competition on worker productivity. Taking the difference between the coefficients on competition in regressions with imputed hourly wages and imputed annual income obtains the effect of competition on hours of work. Finally, taking the difference between the coefficients on competition in regressions with imputed annual income and average annual income obtains the effect of competition on wages through the channel of attracting more productive workers to the MSA. 2.3.2. Effects of inter-jurisdictional competition on industrial composition We also examine how inter-jurisdictional competition affects the share of MSA workers in each 1-digit industry (U.S. Census Bureau, 2000). This sheds light on how competition influences industrial composition. We further analyze whether changes in industrial composition may help explain the effects of inter-jurisdictional competition on wages and wage growth. 19 We carry out this analysis in levels and not indifferences (e.g., changes between the 1980 and 2000 Censuses) due to data limitations. The U.S. Census only provides the MSA of residence in their public microdata release—not the county of residence. Furthermore, the delineation of MSAs changes over time: over 40% of the MSAs in 1980 changed their boundaries by 2000. Hence, an analysis in differences would be an “apples-to-oranges” comparison; it could only use MSAs whose boundaries did not change over the period, which would be a special selection of MSAs (likely those with the lowest growth outcomes). However, the analysis in levels is informative about how the effects of competition can be decomposed at a given moment in time, while it also avoids these comparability issues. 20 This vector includes age, age squared, a male dummy, a veteran dummy, a dummy for immigrating to the U.S. in the last 5 years, race dummies, a dummy for being married, education level dummies, 5 occupation dummies, 13 industry dummies, and the interactions of all of these variables with gender. Our methodology follows that of Albouy (2008) and Notowidigdo (2010).

149

2.3.3. Effects of inter-jurisdictional competition on local public finance Finally, we examine how inter-jurisdictional competition affects local public finance decisions related to the size of revenues, expenditures, the deficit, and the debt. This allows us to examine both how local government decision-making responds to competitive pressures as well as how these decisions influence income and income growth. Data on local public finance are average values from the U.S. Census Bureau (1997, 2002). 3. Results 3.1. OLS results Table 2 presents the results from OLS regressions of average annual growth in income per employee over 1969–2006 on the logged number of county governments. We find evidence of a robust correlation between these variables. The coefficient on the logged number of county governments varies little according to which topographic and climate controls are included. Our baseline specification, in column (6), indicates that a 1 unit increase in the log of the number of county governments results in a 0.16 percentage point increase in the average annual growth rate of income per employee. Doubling the number of county governments—e.g., going from one to two, or from two to four—implies increasing the logged number of county governments by ln(2). Thus, doubling the number of county governments is associated with a ln(2) × 0.16 = 0.11 percentage point increase in the average annual growth rate of income per employee. This result is statistically significant at the 1% level. Since the average annual growth rate in income per employee is 0.85, this is equivalent to a 13% increase in the average annual growth rate during 1969–2006. 21 3.2. IV first stage results Table 3 presents estimates of the first stage regressions. The logged number of county governments is robustly positively correlated with the number of miles of small streams. Both the strength of the first stage and the size and magnitude of the coefficient on miles of small streams vary little according to which topographic and climate controls are included. Our baseline specification, in column (6), indicates that an additional 100 miles of small streams in an MSA is associated with an approximate 22% increase in the number of county governments. The t-statistic on total miles of small streams is 13.3 (implying an F-statistic of 175.8), revealing a strong excluded instrument. In Table C.1 in the Appendix, we explore how the effects of miles of small streams on inter-jurisdictional competition vary according to MSA-level characteristics. This sheds light on the particular type of treatment effect we are estimating. Is a subset of MSAs disproportionately driving our IV estimates of the returns to inter-jurisdictional competition? To estimate an average treatment effect (ATE), we would need identification to come from all MSAs, equally-weighted. If there is underlying heterogeneity in returns to inter-jurisdictional competition, and if our IV estimates recover the returns for a subset of MSAs with relatively high (low) returns, then our IV estimates might be upward-biased (downward-biased) estimates of the average marginal return to inter-jurisdictional competition. We would be estimating a local average treatment effect (LATE) instead of an ATE. To explore where identification comes from, we interacted miles of small streams with four variables, in turn: the share of the workforce 21 A few notes about the selected functional form are warranted. We ran several Box– Cox regressions on our OLS model. In our baseline OLS model, we reject the null hypothesis that a reciprocal transformation of total county governments would maximize the likelihood of observing the data we did, and we reject that no transformation maximizes the likelihood. We cannot reject that a log transformation is appropriate.

150

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

Table 2 OLS results, showing the effect of the logged number of county governments on growth in earnings per employee during 1969–2006. Dep. var.: average annual growth of income per employee, 1969–2006

Log number of county governments

(1)

(2)

(3)

(4)

(5)

(6)

0.12 (0.03)***

0.15 (0.03)***

0.15 (0.03)*** 0.08 (0.05)*

0.15 (0.03)***

0.15 (0.03)***

0.16 (0.03)***

0.23 (0.11)** −0.07 (0.09) 0.01 (0.06) −0.13 (0.05)***

0.23 (0.11)** −0.07 (0.09) 0.01 (0.06) −0.13 (0.05)*** 0.006 (0.03)

Yes 223 0.56

Yes 223 0.56

0.13 (0.13) −0.14 (0.10) 0.01 (0.05) −0.13 (0.05)** −0.002 (0.03) −0.05 (0.14) −0.16 (0.09)* 0.03 (0.14) −0.01 (0.05) −0.04 (0.14) Yes 223 0.59

Dummy—coastal Dummy—on Pacific Ocean Dummy—on Atlantic Ocean Dummy—on Great Lakes Dummy—on major river Log land area Cooling degree days Heating degree days Hours of sunshine Monthly rainfall Standard deviation of elevation State fixed effects Observations R-squared

No 223 0.06

Yes 223 0.53

Yes 223 0.54

Notes: each observation is an MSA (or a CMSA in the case of larger metropolitan areas). Robust standard errors appear in parentheses below the coefficient. *** indicates p b .01; ** indicates p b .05; * indicates p b .10. The same sample is used for all regressions. Dummy—coastal is a dummy variable for having one or more counties classified as coastal by NOAA. Dummy—on Pacific Ocean, Dummy—on Atlantic Ocean, and Dummy—on Great Lakes are indicators for bordering the Pacific Ocean, Atlantic Ocean, and Great Lakes, respectively. Dummy—on major river is an indicator for bordering a major river. Log land area is the log of the area of the (C)MSA in 1969, excluding area covered with water (in 1000s of square miles). Each day, heating degree days equal max{0, 65−mean temperature}, cooling degree days equal amx{0, mean temperature−65}, and the heating and cooling degree variables are the average monthly total during 1970–2000, over 100. Hours sunshine is the average hours of sunshine in January, during 1941–1970 (in 100s). Monthly rainfall is average monthly precipitation during 1970–2000, in inches. The standard deviation of elevation is in 1000s of feet. Sources: BEA (1969–2006), Census of Governments (1962), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

that is blue-collared (to take into account the average skill level of an MSA), the founding year of the first county founded in the MSA (to take into account the age of the MSA), logged 1969 earnings by place

of work per employee (to take into account initial income), and the logged 1960 population (to take into account the initial MSA size). As Table C.1 in the Appendix shows, these interactions are insignificant

Table 3 IV first stage results, showing the effect of miles of small streams on the logged number of county governments. Dep. var.: log number of county governments

100s of miles of small streams

(1)

(2)

(3)

(4)

(5)

(6)

0.15 (0.02)***

0.21 (0.02)***

0.22 (0.01)*** −0.11 (0.08)

0.21 (0.01)***

0.22 (0.02)***

0.22 (0.02)***

−0.00 (0.23) 0.00 (0.14) −0.01 (0.19) 0.05 (0.09)

0.00 (0.23) 0.01 (0.14) −0.02 (0.19) 0.06 (0.09) −0.02 (0.07) Yes No 223 0.65 188.85

−0.33 (0.23) 0.08 (0.15) −0.05 (0.19) 0.09 (0.08) 0.02 (0.07) Yes Yes 223 0.68 175.82

Dummy—coastal (NOAA) Dummy—on Pacific Ocean Dummy—on Atlantic Ocean Dummy—on Great Lakes Dummy—on major river Log land area State fixed effects? Weather and elevation controls? Observations R-squared First stage F-stat, excluded instrument

No No 223 0.33 65.59

Yes No 223 0.65 201.63

Yes No 223 0.66 215.83

Yes No 223 0.65 205.47

Notes: each observation is an MSA (or a CMSA in the case of larger metropolitan areas). Robust standard errors appear in parentheses below the coefficient. *** indicates p b .01; ** indicates p b .05; * indicates p b .10. The same sample is used for all regressions. 100s miles of small streams comes from a computation using ESRI (2008) GIS data that show all streams not classified as major national rivers (there are 55 major rivers, spread across 7 major river systems) as line features on a map. Dummy—coastal is a dummy variable for having one or more counties classified as coastal by NOAA. Dummy—on Pacific Ocean, Dummy—on Atlantic Ocean, and Dummy—on Great Lakes are indicators for bordering the Pacific Ocean, Atlantic Ocean, and Great Lakes, respectively. Dummy—on major river is an indicator for bordering a major river. Log land area is the log of the area of the (C)MSA in 1969, excluding area covered with water (in 1000s of square miles). Sources: BEA (1969–2006), Census of Governments (1962), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

151

Table 4 IV results, showing the effect of the logged number of county governments on growth in earnings per employee during 1969–2006. Dep. var.: average annual growth of income per employee, 1969–2006

Log number of county governments

(1)

(2)

(3)

(4)

(5)

(6)

0.22 (0.06)***

0.19 (0.04)***

0.19 (0.03)*** 0.08 (0.05)*

0.19 (0.04)***

0.19 (0.04)***

0.21 (0.03)***

0.22 (0.10)** −0.08 (0.08) 0.00 (0.05) −0.14 (0.04)***

0.22 (0.10)** −0.08 (0.08) 0.00 (0.05) −0.14 (0.04)*** 0.00 (0.03)

Yes 223

Yes 223

0.12 (0.11) −0.14 (0.09)* −0.01 (0.05) −0.14 (0.05)*** −0.01 (0.03) −0.02 (0.13) −0.15 (0.08)* 0.05 (0.13) −0.01 (0.04) −0.06 (0.13) Yes 223

Dummy—coastal Dummy—on Pacific Ocean Dummy—on Atlantic Ocean Dummy—on Great Lakes Dummy—on major river Log land area Cooling degree days Heating degree days Hours of sunshine Monthly rainfall Standard deviation of elevation State fixed effects Observations

No 223

Yes 223

Yes 223

Notes: each observation is an MSA (or a CMSA in the case of larger metropolitan areas). Robust standard errors appear in parentheses below the coefficient. *** indicates p b .01; ** indicates p b .05; * indicates p b .10. The same sample is used for all regressions. The instrumental variable is hundreds of miles of streams, intermittent streams, falls, and intracoastal waterways in the MSA. Dummy—coastal is a dummy variable for having one or more counties classified as coastal by NOAA. Dummy—on Pacific Ocean, Dummy—on Atlantic Ocean, and Dummy—on Great Lakes are indicators for bordering the Pacific Ocean, Atlantic Ocean, and Great Lakes, respectively. Dummy—on major river is an indicator for bordering a major river. Log land area is the log of the area of the (C)MSA in 1969, excluding area covered with water (in 1000s of square miles). Each day, heating degree days equal max{0, 65 − mean temperature}, cooling degree days equal max{0, mean temperature − 65}, and the heating and cooling degree variables are the average monthly total during 1970–2000, over 100. Hours sunshine is the average hours of sunshine in January, during 1941–1970 (in 100s). Monthly rainfall is average monthly precipitation during 1970–2000, in inches. The standard deviation of elevation is in 1000s of feet. Sources: BEA (1969–2006), Census of Governments (1962), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

in all cases except the founding year. We find no evidence that treatment effects are larger in more highly skilled places, in places with higher initial income, or in places with higher initial population. Since all three of these characteristics (MSA skill mix, MSA income, and MSA size) are likely to affect the returns to inter-jurisdictional competition, it is encouraging that treatment effects do not vary with them. It is unsurprising that the effects of streams are larger in older MSAs. Streams were especially useful in creating county boundaries when mapping techniques were more primitive, and salient “natural” boundaries were accordingly more useful. At the 25th percentile of founding year, another 100 miles of small streams leads to a 31% increase in county governments, while at the 75th percentile of MSA founding year, another 100 miles of small streams leads to a 25% increase in county governments. However, this may still be an ATE rather than a LATE if returns to inter-jurisdictional competition are invariant to MSA founding year. Table 5, Column (5) reveals that MSA founding year itself has no statistically significant effect on income growth during 1969–2006, and an OLS regression which interacts founding year with competition (available upon request) leads us to reject the null hypothesis that the effects of inter-jurisdictional competition vary according to founding year. This is evidence in favor of the interpretation of our IV second stage results as an ATE. 3.3. IV second stage results Table 4 presents IV estimates of the effect of inter-jurisdictional competition on average annual growth in income per employee over 1969–2006. Again, the results vary little according to the set of controls included: the coefficient on the logged number of county governments is positive and statistically significant at the 1% level in all specifications. However, it is now somewhat larger in magnitude than in the baseline

OLS regression. This gap is consistent with the channels of downward bias discussed in Section 2.1. In the baseline specification of column (6), the coefficient on logged county governments is 0.21. Hence, doubling inter-jurisdictional competition is associated with a ln(2)×0.21=0.15 percentage point increase in the average annual growth rate of income per employee over 1969–2006. As the mean annual growth rate of income per employee during this period is 0.85, this amounts to an average annual growth rate over 1969–2006 that is about 17% higher than in the case of half the amount of competition. We take this as evidence of a robust, causal effect of increased competition on growth, which we may underestimate by not accounting for the endogeneity of the number of governments. A more modest, 50% increase in competition—such as would result in going from two to three county governments—leads to an approximate 0.09 percentage point increase in average annual growth over the same period. This is nearly a 10% increase in the average annual growth rate—an economically large effect.22 Several other factors seem to affect growth in income per employee. In particular, being on a major river is associated with lower growth of 0.14 annual percentage points (significant at the 1% level) and being on the Atlantic Ocean or being in a cold climate (as measured by the number of heating-degree days) is also associated with lower growth (significant at the 10% level). Being on the Pacific 22 Population is an omitted variable likely to affect both economic growth and the number of jurisdictions. For example, Duranton and Puga (2004) and Rosenthal and Strange (2004) describe how urban increasing returns—or agglomeration economies—come about through labor market pooling, input sharing, knowledge spillovers, home market effects, consumption opportunities, and rent-seeking. Nonetheless, we do not include it in our baseline specification as it is likely to be endogenous. Our IV strategy makes this (arguably) unproblematic. In Section 4, we show that its inclusion does not appreciably affect our main results, and even suggests that we underestimate the growth benefits of competition (though this analysis has caveats).

152

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

Ocean, being on the Great Lakes, the land area of the MSA (in square miles), average hours of sunshine in January, average monthly rainfall, the standard deviation of elevation (i.e., ruggedness of terrain), and a warm climate (as measured by cooling-degree days) do not appear to have a statistically significant impact on growth.

3.4. Income levels A natural question is whether our findings surrounding the effects of inter-jurisdictional competition on economic growth simply reflect convergence to a mean income level. That is, does inter-jurisdictional competition just happen to be higher in places that initially had a relatively low 1969 income per employee, suggesting that inter-jurisdictional competition's apparent effect on income growth over 1969–2006 is just convergence to the mean? To explore this possibility, we run separate regressions of income by place of work per employee (in constant 2000 dollars) on the logged number of county governments for each individual year during the period 1969–2006. Table C.2 in the Appendix presents a subset of these 37 regressions (others are available upon request). In Fig. 2, we then plotted the coefficients on the logged number of county governments from each of these 37 regressions, and also plotted the 95% confidence bounds on each of the coefficients. Fig. 2 reveals two things. First, increasing the number of county governments is associated with a higher 1969 income. Specifically, a one unit increase in the logged number of county governments is associated with a 1969 income per employee that is 7% higher, (so doubling the amount of inter-jurisdictional competition is associated with a 1969 income that is ln(2) × 0.07 = 5% higher). Second, the effect of the number of county governments on income levels has steadily increased over time since 1969; a one unit increase in logged county governments is associated with a 2006 income per employee that is 15% higher (so doubling the amount of inter-jurisdictional competition is associated with a 2006 income that is ln(2) × 0.15 = 10% higher). These results imply that higher inter-jurisdictional competition was already associated with higher income at the beginning of the window over which we measure growth, and that the disparity only grew over the next 37 years.

Fig. 2. Coefficient on logged number of county governments for each year in 1969–2006, from IV regression of logged income per employee in that year on logged number of county governments. Notes: this line graph is drawn using a set of 37 coefficients on the logged number of county governments, from 37 separate regressions. Each regression is an IV specification in which the dependent variable is logged income per employee (in thousands of constant, 2000 $) for each year during 1969–2006, and it is regressed on the logged number of country governments and also on geographic controls. The excluded instrument for the logged number of county governments is the number of miles of small streams (ESRI, 2008). Table C.2 in the Appendix presents the full regressions for the years 1969, 1979, 1989, 1999, 2001, and 2006, and presents the full list of controls included in all specifications. Sources: author's calculations based on data from BEA (1969–2006), Census of Governments (1962, 1997, 2002), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

3.5. The effects of inter-jurisdictional competition on growth over time A natural question is whether the effects of inter-jurisdictional competition on economic growth have been constant or have intensified over time. There are reasons to believe that its effects have intensified: modern transportation and communications technologies have made factors of production increasingly mobile. We explore this possibility by running separate regressions of growth in income by place of work per employee (in constant 2000 dollars) during every 10-year interval contained in 1969–2006 on the logged number of county governments (e.g., 1969–1979, 1970–1980, 1981–1991, etc.). Table C.3 in the Appendix presents a subset of these 28 regressions (others are available upon request). In Fig. 3, we then plotted the coefficients on the logged number of county governments from each of these 28 regressions, and also plotted the 95% confidence bounds on each of the coefficients. Fig. 3 reveals two things. First, since 1974, having more county governments has led to a statistically significantly higher (at the 5% level) average annual growth rate over the next 10 years. Second, the effect of the number of county governments on average annual growth in income per employee increased for most of the next 20 years. The line connecting coefficients in Fig. 3 is generally upward-sloping. Only in the last 5 years has the effect of inter-jurisdictional competition on growth leveled off and declined slightly. This provides some evidence that the growth impacts of competition have increased over time. 4. Robustness 4.1. Specification checks As discussed in Section 2, one possible concern is the potential correlation of miles of small streams with other features of an MSA that predispose it to higher incomes or higher income growth over 1969–2006. As a result, we introduced several specification checks by adding additional control variables to the baseline specification of Table 4, column (6). Table 5 presents regressions with these additional controls. Columns (1)–(3) introduce controls for log earnings per employee in

Fig. 3. Coefficient on logged number of county governments for each 10-year interval in 1969–2006, from IV regression of growth in income per employee in that 10-Year interval on logged number of county governments. Notes: this line graph is drawn using a set of 28 coefficients on the logged number of county governments, from 28 separate regressions. Each regression is an IV specification in which the dependent variable is growth in income per employee (in thousands of constant, 2000 $) for each 10-year period contained within 1969–2006, and it is regressed on the logged number of country governments and also on geographic controls. The excluded instrument for the logged number of county governments is the number of miles of small streams (ESRI 2008). Table C.3 in the Appendix presents the full regressions for the years 1969–1979, 1979–1989, 1989–1999, and 1996–2006, and presents the full list of controls included in all specifications. Sources: author's calculations based on data from BEA (1969–2006), Census of Governments (1962, 1997, 2002), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

153

Table 5 IV second stage results, showing that the effect of the logged number of county governments on growth in earnings per employee during 1969–2006 is robust to inclusion of numerous controls. Dep. var.: average annual growth of income per employee, 1969–2006

Log number of county governments Log earnings per employee in 1969

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

0.31 (0.04)*** −1.30 (0.25)***

0.26 (0.13)**

0.28 (0.13)** −1.32 (0.24)*** 0.02 (0.08)

0.26 (0.05)***

0.20 (0.04)***

0.20 (0.03)***

0.19 (0.03)***

0.36 (0.15)** −1.26 (0.24)*** −0.02 (0.08) −0.05 (0.02)** −0.07 (0.09) −1.30 (1.08) 2.01 (4.25) 7.68 (4.36)* 1.69 (0.97)* 0.28 (0.15)* 0.05 (0.10) 0.20 (0.10)** −0.06 (0.05) 0.03 (0.03) 222 17.38

−0.04 (0.08)

Log population

−0.05 (0.02)**

Log water area

−0.11 (0.11)

Founding year, over 100 Share black

0.68 (0.34)** 4.39 (4.00) 12.63 (4.72)***

Share native American Share other race Racial fractionalization index Dummy—on Pacific Ocean Dummy—on Atlantic Ocean Dummy—on Great Lakes Dummy—on major river Log land area Observations First stage F-stat, excluded instrument

0.11 (0.11) −0.08 (0.08) 0.05 (0.05) −0.10 (0.05)** 0.01 (0.03) 222 105.71

0.14 (0.13) −0.15 (0.09)* 0.01 (0.06) −0.13 (0.05)*** −0.00 (0.03) 222 20.14

0.10 (0.12) −0.08 (0.08) 0.04 (0.06) −0.10 (0.05)** 0.00 (0.03) 222 20.06

0.23 (0.13)* −0.04 (0.10) 0.14 (0.08)* −0.12 (0.05)** 0.01 (0.03) 222 85.08

0.13 (0.12) −0.15 (0.09)* 0.00 (0.05) −0.14 (0.05)*** −0.01 (0.03) 222 162.59

0.10 (0.12) −0.13 (0.09) −0.02 (0.05) −0.13 (0.04)*** 0.00 (0.03) 222 170.36

0.73 (0.30)** 0.14 (0.12) −0.10 (0.09) −0.04 (0.05) −0.13 (0.05)*** −0.01 (0.03) 222 175.76

Notes: each observation is an MSA (or a CMSA in the case of larger metropolitan areas). Robust standard errors appear in parentheses below the coefficient. *** indicates p b .01; ** indicates p b .05; * indicates p b .10. All specifications include state group fixed effects, controls for heating and cooling degree days (1970–2000 averages), hours of sunshine (avg. in Jan.), monthly rainfall (1970–2000 average), and the standard deviation of elevation. The same sample is used for all regressions. 100s of miles of small streams comes from a computation using ESRI (2008) GIS data that show all streams not classified as major national rivers (there are 55 major rivers, spread across 7 major river systems) as line features on a map. Dummy—coastal is a dummy variable for having one or more counties classified as coastal by NOAA. Dummy—on pacific ocean, Dummy—on Atlantic ocean, and Dummy—on great lakes are indicators for bordering the Pacific Ocean, Atlantic Ocean, and Great Lakes, respectively. Dummy—on major river is an indicator for bordering a major river. Log land area is the log of the area of the (C)MSA in 1969, excluding area covered with water (in 1000s of square miles). Sources: BEA (1969–2006), Census of Governments (1962), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

1969 and log population in 1960. MSAs with higher 1969 incomes grow more slowly during 1969–2006, and when initial income is included as a control, the effect of inter-jurisdictional competition is greater. It may be that MSAs with more inter-jurisdictional competition grew more quickly pre-1969, making growth in the 1969–2006 period harder to achieve, and so we may be underestimating the effect of inter-jurisdictional competition when this control is not included. By contrast, initial population does not seem to have a significant impact on growth. When we control for both initial income and population conditions (and thus compare MSAs at similar starting points), we see a larger impact of inter-jurisdictional competition on economic growth; the coefficient on the logged number of county governments, which was 0.21 in the baseline specification, is now 0.28. Doubling inter-jurisdictional competition leads to a ln(2) × 0.28 = 0.19 percentage point increase in the average annual growth rate of income per employee over 1969–2006. This amounts to an average annual growth rate over 1969–2006 that is about 23% higher (this figure was 17% in the baseline specification of Table 4). Column (4) controls for logged water area in the MSA. It has a statistically significant negative effect on income growth, and we see that the coefficient on inter-jurisdictional competition is now 0.26 (compared with 0.21 in the baseline). Column (5) reveals that the founding year of the first founded county in the MSA has no statistically significant impact on income growth, and controlling for it has little effect on the coefficient on inter-jurisdictional competition.

Columns (6) and (7) reveal that controlling for the share of the population of each race, and for racial fractionalization, leads to a slightly lower coefficient on inter-jurisdictional competition (it is now between 0.18 and 0.19, compared with 0.21 in the baseline). Column (8) includes all of these controls, and the coefficient on interjurisdictional competition is now 0.36 (though the coefficient from the baseline regression, 0.21, is within the 95% confidence interval around this estimate). Many of these control variables may be endogenous to economic growth, and hence these empirical specifications may suffer from an endogeneity bias. Nevertheless, it is encouraging that our results vary little according to the particular set of controls included. We take this as additional evidence of the robust, positive effect of inter-jurisdictional competition on growth in income per employee. Table C.4 in the Appendix presents additional specifications controlling for changes in population between 1960 and 2006. One might be concerned that a large number of county governments occurs precisely in areas that have seen high population growth and the development of agglomeration economies since the 1960s. While not a threat to identification, this may affect the interpretation of our results.23 From column (2), we see that the inclusion of the logged 2006 population minus the

23 Of course, it may also be the case that the population increases are due to the higher productivity of those areas.

154

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

logged 1960 population has no appreciable impact on our main result; the coefficient on inter-jurisdictional competition is 0.20 when we control for population changes—close to the 0.21 of the baseline specification, and significant at the 1% level. The coefficient on competition is similarly significant in column (5), where we additionally control for logged earnings per employee in 1969.24

4.2. Alternate measures of streams Both Hoxby (2000) and Rothstein (2007) use streams as instrumental variables. In Table C.5 in the Appendix, we use their measures to check the robustness of our results. 25 We find that our results are substantially unchanged when we instead rely on their measures. When we use the streams variable from Hoxby (2000) with our baseline specification, the coefficient on logged county governments is 0.23, which is slightly larger than it is in the baseline specification (where it is 0.21). The F-statistic on the excluded streams instrument is 75, suggesting no problems of weak instruments. When we instead use the streams variable from Rothstein (2007), the coefficient on logged county governments is 0.20. The F-statistic on the excluded streams instrument in this specification is 193, again suggesting a strong first stage. 4.3. Alternate measures of inter-jurisdictional competition We have measured inter-jurisdictional competition in a specific way: using the number of county governments. Table C.6 in the Appendix shows that our main findings are robust to instead using either the number of municipal plus township governments or the number of school districts in the MSA. Columns (1) and (8) show our baseline OLS and IV specifications but use the logged number of municipal plus township governments in place of the logged number of county governments. In both specifications, the coefficients on inter-jurisdictional competition are positive and statistically significant at the 1% level, though they are slightly smaller in magnitude than in the baseline. The OLS coefficient is 0.11 (compared with 0.16 for county governments), and the IV coefficient is 0.17 (compared with 0.21 for county governments). The results are quite similar, and the coefficients from our baseline regressions are within the 95% confidence intervals around these coefficients. Column (6) presents the corresponding first stage specification for this IV result, revealing that it is strong. Columns (2) and (9) show our baseline OLS and IV specifications using the logged number of school districts instead of counties. Once again, the coefficients on competition are positive and statistically significant at the 1% level. The OLS coefficient is 0.14 (compared with 0.16 for counties), and the IV coefficient is 0.22 (compared with 0.21 for counties). Once again, the coefficients from our baseline regressions are within the 95% confidence intervals around these coefficients. Column (7) presents the first stage specification for this IV result, again showing that it is strong. 24 In column (3), when we control for both the 1960 population and the change in the population between 1960 and 2006, the standard errors around the coefficient on competition grow much larger, and the coefficient becomes insignificant. This is also true in column (6), which additionally controls for logged 1969 income per employee. However, the point estimate is 0.20 and 0.21 in columns (3) and (6), respectively— nearly identical to its value of 0.21 in the baseline specification. Multicollinearity is a likely problem here, given high correlations between population, population change, and the number of jurisdictions. Further, the first-stage F-Stat is much lower in these two specifications—around 13, compared with 176 in the baseline—indicating possible problems of weak instruments that would make the confidence intervals wrong. Nevertheless, it is encouraging that the point estimates are so close to the baseline estimates. 25 Note that the streams variables in Hoxby (2000) and Rothstein (2007) pertain to a slightly different set of geographic areas than does ours; they use the 1990 boundaries of MSAs while we use the 2000 boundaries.

Overall, our results do not seem to be driven by how interjurisdictional competition is defined. However, we also wanted to shed light on the relative effects of each of these three measures by including them in the same specification. With only one excluded instrument (miles of small streams), we were unable to estimate IV specifications of this type. We thus estimated several OLS specifications: municipal plus township governments with county governments (column 3), school districts with county governments (column 4), and municipal plus township governments with county governments and school districts (column 5). Logged county governments is the only measure of competition that is statistically significant in all three specifications. Logged school districts is never significant. In the final specification, which includes all three measures of competition, only logged county governments is statistically significant (at the 5% level). This makes us confident that we have chosen a relatively meaningful measure of inter-jurisdictional competition. 4.4. An alternative measure of income: gross municipal product As we have described above, there are several important reasons to use earnings by place of work per employee as our primary dependent variable. However, we wanted to explore the implications of this choice of outcome variable. One particular concern is that rather than increasing productivity, inter-jurisdictional competition changes the equilibrium returns to capital and labor. In particular, models such as those of Zodrow and Mieszkowski (1986) suggest that an increase in inter-jurisdictional competition will lead to lower wages and higher returns to capital. To explore this possibility, we consider as an outcome variable gross municipal product (GMP) per employee over time. Table C.7 in the Appendix, column (6) presents a regression of 2001 GMP per employee on logged county governments. This is the counterpart of our IV specification that instead uses income by place of work per employee in 2001 (Table C.2, column 5). The coefficient on logged county governments is 0.15 for 2001 GMP (it is 0.16 for 2001 earned income by place of work per employee). Doubling inter-jurisdictional competition is thus associated with a 2001 GMP per employee that is 10% higher. Inter-jurisdictional competition has nearly as great of an effect on GMP per employee as it does on income per employee. This also implies that inter-jurisdictional competition not only increases MSA productivity, but also that these productivity gains are largely captured by employees. 5. Mechanisms 5.1. Income decomposition To examine the mechanisms through which inter-jurisdictional competition leads to higher income, we first use worker-level data to distinguish three separate channels: 1. inter-jurisdictional competition could directly cause workers in the MSA to be more productive with each hour of work; 2. inter-jurisdictional competition could lead workers to work more hours (e.g., because each hour is more productive); 3. inter-jurisdictional competition could attract a more productive set of workers to the MSA (i.e., lead to a change in the composition and demographic characteristics of the workforce that raises income per employee). Table 6 presents the results from regressions of six different dependent variables on logged county governments: actual and imputed versions of log annual earned income (salary), actual and imputed versions of log hourly income (wage), and actual and imputed versions of log hours worked per year. Actual values are MSA-level averages from the 2000 Census worker-level data. Imputed values are

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

155

Table 6 IV results, showing the effect of the logged number of county governments on actual and imputed values of log annual earned income, log hourly earned income, and log hours worked per year in 2000. Dependent variable:

Log number of county governments Dummy—on Pacific Ocean Dummy—on Atlantic Ocean Dummy—on Great Lakes Dummy—on major river Log land area Cooling degree days Heating degree days Hours of sunshine Monthly rainfall Standard deviation of elevation Observations

Log annual earned income (salary)

Log hourly earned income (wage)

Log hours worked per year

Actual

Imputed

Actual

Imputed

Actual

Imputed

(1)

(2)

(3)

(4)

(5)

(6)

0.139 (0.012)*** 0.131 (0.047)*** 0.015 (0.025) −0.001 (0.028) −0.050 (0.020)** 0.006 (0.010) 0.054 (0.051) 0.043 (0.036) 0.119 (0.060)** 0.010 (0.020) −0.148 (0.056)*** 199

0.094 (0.009)*** 0.061 (0.032)* 0.010 (0.019) 0.007 (0.020) −0.040 (0.014)*** 0.000 (0.007) 0.045 (0.042) 0.016 (0.027) 0.113 (0.048)** 0.005 (0.015) −0.090 (0.036)** 199

0.096 (0.009)*** 0.084 (0.032)*** −0.004 (0.017) 0.015 (0.018) −0.025 (0.015)* 0.006 (0.007) 0.028 (0.034) 0.022 (0.024) 0.091 (0.038)** 0.026 (0.012)** −0.111 (0.042)*** 199

0.066 (0.006)*** 0.029 (0.021) 0.001 (0.011) 0.019 (0.013) −0.018 (0.010)* 0.001 (0.005) 0.025 (0.026) 0.002 (0.016) 0.084 (0.029)*** 0.018 (0.009)** −0.065 (0.025)*** 199

0.042 (0.006)*** 0.048 (0.021)** 0.019 (0.015) −0.016 (0.013) −0.025 (0.008)*** −0.001 (0.005) 0.025 (0.024) 0.021 (0.017) 0.028 (0.029) −0.016 (0.011) −0.037 (0.020)* 199

0.028 (0.004)*** 0.032 (0.015)** 0.009 (0.012) −0.012 (0.009) −0.022 (0.007)*** −0.000 (0.004) 0.020 (0.019) 0.014 (0.013) 0.029 (0.023) −0.013 (0.008) −0.025 (0.016) 199

Notes: each observation is an MSA (or a CMSA in the case of larger metropolitan areas). Robust standard errors appear in parentheses below the coefficient. *** indicates p b .01; ** indicates p b .05; * indicates p b .10. The same sample is used for all regressions. All specifications include state group fixed effects. The instrumental variable is hundreds of miles of streams, intermittent streams, falls, and intracoastal waterways in the MSA (the first stage F-stat on it is 185.22). Dummy—coastal is a dummy variable for having one or more counties classified as coastal by NOAA. Dummy—on Pacific Ocean, Dummy—on Atlantic Ocean, and Dummy—on Great Lakes are indicators for bordering the Pacific Ocean, Atlantic Ocean, and Great Lakes, respectively. Dummy—on major river is an indicator for bordering a major river. Log land area is the log of the area of the (C)MSA in 1969, excluding area covered with water (in 1000s of square miles). Each day, heating degree days equal max{0, 65−mean temperature}, cooling degree days equal max{0, mean temperature−65}, and the heating and cooling degree variables are the average monthly total during 1970–2000, over 100. Hours sunshine is the average hours of sunshine in January, during 1941–1970 (in 100s). Monthly rainfall is average monthly precipitation during 1970–2000, in inches. The standard deviation of elevation is in 1000s of feet. Sources: BEA (1969–2006), Census of Governments (1962), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

MSA-level averages after controlling for the vector of worker-level characteristics described in Section 2.3.1. Column (1) indicates that a one unit increase in the logged number of county governments is associated with an annual earned income (salary) that is 13.9% higher. This is the total effect of interjurisdictional competition on earned income from all three channels. It is similar to the findings using BEA data on income by place of work per employee in 2000. Column (2) indicates that, after taking into account the characteristics of the workforce, a one unit increase in interjurisdictional competition leads to an annual earned income (salary) that is a more modest 9.4% higher. Finally, Column (4) indicates that, after taking into account the characteristics of the workforce and also the number of hours its members work (by considering hourly wage rather than annual salary), a one unit increase in inter-jurisdictional competition leads to earned income that is only 6.6% higher. These estimates tell us that the 13.9% total effect of interjurisdictional competition on earned income comes from three sources: 6.6% is the direct impact of making an MSA's workers more productive, 2.8% (i.e., 9.4%–6.6%) is the effect of inducing workers to work more hours, and 4.5% (i.e., 13.9%–9.4%) is the effect of attracting more productive workers to the MSA. In short, just under one-half of it is a direct impact, just over one-fifth comes from workers working more hours, and just under one-third comes from attracting more productive workers.26 Column (6) regresses logged hours worked per worker on 26 It is possible that individuals might also sort on unobservables. If sorting on unobservables goes in the same direction as sorting on observables, then our estimate of the direct impact of competition (i.e. just under one-half of the total effect) is an upper bound for the impact of more local governments on individual productivities.

inter-jurisdictional competition. Even after controlling for worker characteristics and demographics, doubling inter-jurisdictional competition leads the average worker to work ln(2)× 2.8 =1.9% more hours. This is equivalent to adding 47 minutes onto a 40-hour work week. 5.2. Effects on Industrial composition Table 7 presents the effects of inter-jurisdictional competition on the share of MSA workers that are in each industry in the 2000 Census (there are 13 industries in total). MSAs with greater inter-jurisdictional competition have larger shares of their workforces engaged in three relatively highly-remunerated industrial categories: finance, insurance, and real estate; professional, scientific, management, and administrative services; and information and communications. MSAs with greater interjurisdictional competition also have smaller shares of their workforces engaged in three less highly-remunerated industries: arts, entertainment, recreation, accommodations, and food services; education, health, and social services; and retail trade. We take this as evidence that inter-jurisdictional competition has had an impact on the industrial composition of the workforce, and that it has tended to lead to more employment in relatively high-skill, high-remuneration jobs. Table 8 examines whether the effects of inter-jurisdictional competition vary when we control for the share of MSA workers in each industry in the 2000 Census. If the coefficient on inter-jurisdictional competition is still positive but smaller in magnitude, this would suggest that part of the effect of inter-jurisdictional competition comes from its tendency to draw relatively high-skill, high-paying industries to an MSA (as suggested by the results in Table 7). We regress both growth in income per employee during 1969–2006 and the level of

156

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

Table 7 IV second stage results, showing the effect of the logged number of county governments on the share of workers in each industry in 2000. Dependent variable:

Mean of dependent variable:

Log number of county governments Dummy—on Pacific Ocean Dummy—on Atlantic Ocean Dummy—on Great Lakes Dummy—on major river Log land area Cooling degree days Heating degree days Hours of sunshine Monthly rainfall Standard deviation of elevation Observations

Construction

Wholesale trade

Retail trade

Transport/ warehous./ utilities

Information

Finance/ insurance/ real estate

Professnl./ scientific/ manag.

Education/ health/soc. serv.

Arts/ entertain./ recreation

Other services (non-admin.)

0.07

0.03

0.13

0.05

0.02

0.06

0.07

0.21

0.08

0.05

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

0.004 (0.001)*** −0.001 (0.005) 0.015 (0.004)*** −0.002 (0.002) 0.002 (0.002) 0.002 (0.001) 0.010 (0.005)* 0.001 (0.003) 0.004 (0.006) 0.005 (0.002)** −0.010 (0.006) 226

0.004 (0.001)*** −0.005 (0.005) −0.004 (0.002)** −0.000 (0.002) −0.001 (0.001) 0.002 (0.001)** 0.006 (0.005) 0.001 (0.003) 0.003 (0.005) −0.003 (0.002) 0.001 (0.008) 226

−0.003 (0.001)** 0.006 (0.006) 0.006 (0.004)* −0.002 (0.003) −0.002 (0.002) 0.001 (0.001) 0.012 (0.005)** 0.008 (0.003)** 0.011 (0.005)** 0.003 (0.002) −0.005 (0.006) 226

0.006 (0.001)*** −0.008 (0.005)* 0.002 (0.004) 0.000 (0.004) 0.002 (0.002) 0.003 (0.001)** 0.007 (0.006) −0.001 (0.003) −0.007 (0.009) −0.002 (0.002) −0.009 (0.005)** 226

0.004 (0.001)*** 0.007 (0.002)*** −0.002 (0.002) −0.002 (0.002) −0.004 (0.001)*** −0.000 (0.001) −0.002 (0.003) −0.001 (0.002) 0.005 (0.003)* −0.001 (0.001) −0.005 (0.003) 226

0.008 (0.002)*** 0.011 (0.004)*** −0.000 (0.003) −0.003 (0.005) −0.005 (0.003) 0.004 (0.001)*** 0.015 (0.005)*** 0.010 (0.004)*** 0.018 (0.005)*** 0.003 (0.002) −0.019 (0.008)** 226

0.009 (0.002)*** 0.017 (0.009)* 0.007 (0.003)** 0.001 (0.003) −0.001 (0.003) 0.001 (0.002) −0.003 (0.008) −0.003 (0.005) 0.004 (0.008) −0.002 (0.003) −0.014 (0.010) 226

−0.022 (0.003)*** −0.018 (0.013) −0.024 (0.011)** 0.006 (0.008) 0.002 (0.006) 0.003 (0.003) −0.014 (0.015) −0.003 (0.011) −0.012 (0.019) 0.007 (0.006) −0.000 (0.013) 226

−0.008 (0.002)*** 0.017 (0.005)*** 0.022 (0.009)** 0.006 (0.002)*** −0.000 (0.002) 0.002 (0.001) −0.012 (0.005)** −0.002 (0.003) 0.000 (0.005) −0.002 (0.002) 0.008 (0.010) 226

−0.002 (0.000)*** 0.003 (0.002) 0.001 (0.001) −0.001 (0.001) −0.001 (0.001) 0.001 (0.000)** 0.001 (0.002) 0.002 (0.001)* 0.003 (0.003) 0.000 (0.001) −0.000 (0.003) 226

Notes: each observation is an MSA (or a CMSA in the case of larger metropolitan areas). Robust standard errors appear in parentheses below the coefficient. *** indicates p b .01; ** indicates p b .05; * indicates p b .10. The same sample is used for all regressions. All specifications include state group fixed effects. The instrumental variable is hundreds of miles of small streams (the first stage F-stat on it is 189.75). Dummy—coastal is a dummy variable for having one or more counties classified as coastal by NOAA. Dummy—on Pacific Ocean, Dummy—on Atlantic Ocean, and Dummy—on Great Lakes are indicators for bordering the Pacific Ocean, Atlantic Ocean, and Great Lakes, respectively. Dummy—on major river is an indicator for bordering a major river. Log land area is the log of the area of the (C)MSA in 1969, excluding area covered with water (in 1000s of square miles). Data on the share of MSA workers in each industry come from the 2000 Census, where there are 13 industries. Given space considerations, we have omitted results from three industries present: agriculture/forestry/fishing and hunting/minerals (2% of workforce, on average); manufacturing (15% on average); and public administration (5% on average). The logged number of county governments did not have a statistically significant effect (at conventional levels) on the share of workers in any of these three industries. These results are available upon request. Sources: BEA (1969–2006), Census of Governments (1962), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

2006 income per employee on the logged number of county governments. Columns (2) and (3) reveal a small decrease in the effect of inter-jurisdictional competition on income growth (the coefficient drops from 0.21 to 0.20) and on 2006 income (the coefficient drops from 0.15 to 0.13), respectively, when we control for industrial composition. While these differences are not statistically significantly different, this suggests that some of the effect of inter-jurisdictional competition may be channeled through the promotion of a more productive industrial composition. 5.3. Effects on local public finance One important mechanism by which inter-jurisdictional competition may enhance economic outcomes is by affecting the revenue and expenditure decisions of local governments. Column (1) of Table 9 indicates that competition is associated with higher taxes; a 1 unit increase in logged county governments is associated with an additional $158 in tax revenue per capita. However, other forms of local revenue—including transfers from state and national governments—are lower on average, though the effect is not statistically significant. In sum, local governments in MSAs subject to relatively more competition raise more revenue, and they do so locally (not by attracting more transfers from higher levels of government). Column (5) indicates that a 1 unit increase in logged county governments is associated with an additional $162 in expenditures per capita. This exceeds the additional revenue being brought in by $69, and we indeed see that the deficit is $69 higher (column 6). This same increase in logged county governments is associated with a debt per capita that is $752 higher (column 7), which leads to an additional $40 per year,

per capita, of interest payments on the debt (column 8). Hence, MSAs that are subject to more inter-jurisdictional competition do spend more per capita, but the additional cost of this to taxpayers ($162) is small compared to the gains in income per employee. 27 To investigate the effects of inter-jurisdictional competition on interest rates, we take the ratio of interest payments on the debt over the size of the debt to obtain an implied interest rate. We find no evidence that having more county governments is associated with a higher implied interest rate (column 9). We also find no evidence that MSAs with more county governments spend more money per capita on central staff expenditures (column 10); administrative costs per capita— at least as captured by government staffing—are unaffected by the degree of inter-jurisdictional competition. Table 8 explores whether the effects of inter-jurisdictional competition change when we control for the local public finance variables described above. If the coefficient on competition is still positive but smaller in magnitude, this would suggest that part of the effect of competition is channeled through its effect on local public finance in an MSA. We regress both growth in income per employee during 1969–2006, and the level of 2006 income per employee, on the logged number of county governments. Columns (5) and (6) reveal a small decrease in the effect of inter-jurisdictional competition on

27 The average personal earned income per worker (including part-time workers) in the 2000 Census is $17,300/year. As doubling inter-jurisdictional competition leads to a 6.6% increase in the hourly income of a given worker, this implies that an average worker would earn an extra $1140/year (if he does not change hours worked in response to the wage increase).

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

157

Table 8 IV results, showing the effect of the logged number of county governments on growth in earnings per employee during 1969–2006 and income per employee in 2006 when controlling for industrial composition and characteristics of local public finance. Dependent variable:

IJC

Growth

Income

IJC

Growth

Income

IV 1st

IV 2nd

IV 2nd

IV 1st

IV 2nd

IV 2nd

(1)

(2)

(3)

(4)

(5)

(6)

0.20 (0.05)***

0.13 (0.02)***

0.17 (0.04)***

0.12 (0.02)***

−0.03 (0.15) −0.04 (0.12) −0.00 (0.09) 0.13 (0.06)** 0.40 (0.18)** −0.19 (0.60) −0.50 (2.48) −0.00 (0.04) −0.37 (1.58) 222 108.42

0.14 (0.06)** 0.02 (0.05) −0.03 (0.03) 0.03 (0.02) 0.07 (0.07) −0.23 (0.24) 0.19 (1.07) 0.01 (0.01) −0.97 (0.67) 222 108.42

Log number of county governments 100s miles of small streams Share—construction Share—wholesale trade Share—retail trade Share—transport./warehous./utilities Share—information and communications Share—finance/insurance/real estate Share—professnl./scientific/manag. Share—educational/health/soc. serv. Share—arts/entertain/recre./accomm. Share—other services (non-public admin.)

0.18 (0.02)*** 6.83 (3.56)* −2.85 (5.20) −4.76 (3.07) 3.06 (3.52) −5.23 (4.76) 0.98 (2.26) 8.39 (2.78)*** −2.55 (1.65) −5.56 (3.24)* −3.93 (6.33)

0.20 (0.02)*** −6.23 (2.13)*** −3.54 (3.06) 0.55 (2.12) −2.27 (1.53) −0.42 (3.52) 1.25 (1.07) 0.98 (1.59) −1.71 (1.12) −1.26 (1.40) −8.24 (3.96)**

−2.16 (0.86)** −0.62 (1.16) −1.00 (0.75) −0.03 (0.59) −0.38 (1.16) 0.55 (0.45) 2.39 (0.56)*** −0.54 (0.46) 0.09 (0.57) 0.37 (1.36)

Total taxes per capita Transfer revenue per capita Utility revenue per capita Expenditure per capita Deficit per capita Interest paid on debt per capita Implied interest rate on debt Debt per capita Central staff exp. per capita Observations First stage F-stat R-squared

223

223 84.97

0.77

223 84.97

0.78 (0.25)*** −0.08 (0.19) 0.05 (0.13) −0.04 (0.08) −0.36 (0.32) −1.29 (1.57) 1.78 (5.57) 0.09 (0.09) −2.20 (2.54) 222 0.73

Notes: each observation is an MSA (or a CMSA in the case of larger metropolitan areas). Robust standard errors appear in parentheses below the coefficient. *** indicates p b .01; ** indicates p b .05; * indicates p b .10. The same sample is used for all regressions. IJC is inter-jurisdictional competition, defined as the logged number of county governments. Growth in the average annual growth rate in income by place of work per employee during 1969–2006. Income is the 2006 level of income by place of work per employee. All specifications include the following, full set of topographic and climatic controls: dummy variables for being on the Pacific Ocean, Atlantic Ocean, the Great Lakes, and a major river; the log of the land area of the (C)MSA in 1969 (in 1000s of square miles); average hours of sunshine in January, during 1941–1970 (in 100s); average monthly precipitation during 1970–2000 (in inches); the standard deviation of elevation (1000s of feet), and the average monthly total of heating degree days and of cooling degree days during 1970–2000 (over 100) (each day, heating degree days equal max{0, 65 − mean temperature} and cooling degree days equal max{0, mean temperature − 65}). Sources: BEA (1969–2006), Census of Governments (1962), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

income growth (the coefficient drops from 0.21 to 0.17) and on 2006 income (the coefficient drops from 0.15 to 0.12), respectively, when we include local public finance controls. While these differences are not statistically significantly different, this suggests that some of the effect of competition may be channeled through a more efficient system of local public finance. 6. Conclusion This paper exploits exogenous variation in the number of miles of small streams in the United States to estimate the causal impact of inter-jurisdictional competition on income growth. We study metropolitan areas (MSAs) and use the number of county governments as

our central measure of inter-jurisdictional competition. We find that doubling the number of county governments in an MSA—such as by increasing the number of county governments from 1 to 2 —leads to an approximate 0.15 percentage point increase in the average annual growth rate of earnings per employee over 1969–2006. This effect is relatively large and meaningful, amounting to an average annual growth rate in income per employee that is 17% higher than average. We take this as evidence that inter-jurisdictional competition has a robust, positive impact on an MSA's growth potential. These results are robust to controlling for the founding year of the MSA's counties, and for the pre-period values of income, population, and racial composition. The results are also robust to alternative measures of inter-jurisdictional competition, income, and miles of small streams.

158

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

Table 9 IV second stage results, showing the effect of the logged number of county governments on revenue sources, expenditures, the deficit, and the debt. Dependent variable:

Log number of county governments Dummy—on Pacific Ocean Dummy—on Atlantic Ocean Dummy—on Great Lakes Dummy—on major river Log land area Cooling degree days Heating degree days Hours of sunshine Monthly rainfall Standard deviation of elevation Observations

Total taxes

Transfer revenue

Utility revenue

Other revenue

Expenditure

Deficit

Debt

Interest on debt

Implied interest

Central staff exp.

Per capita

Per capita

Per capita

Per capita

Per capita

Per capita

Per capita

Per capita

Rate (debt)

Per capita

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

0.158 (0.025)*** 0.272 (0.079)*** 0.123 (0.056)** 0.108 (0.043)** −0.059 (0.034)* 0.005 (0.021) 0.134 (0.090) 0.059 (0.046) 0.304 (0.103)*** 0.073 (0.031)** −0.174 (0.084)** 222

−0.033 (0.025) −0.305 (0.126)** −0.067 (0.046) 0.030 (0.043) 0.079 (0.035)** 0.015 (0.020) 0.105 (0.099) −0.002 (0.051) −0.343 (0.123)*** −0.117 (0.034)*** 0.337 (0.181)* 222

0.019 (0.041) −0.165 (0.244) −0.281 (0.083)*** −0.048 (0.051) 0.076 (0.080) −0.014 (0.039) −0.038 (0.196) −0.093 (0.131) −0.407 (0.234)* −0.082 (0.115) −0.274 (0.289) 222

−0.051 (0.038) 0.125 (0.146) 0.140 (0.120) 0.109 (0.047)** −0.032 (0.062) −0.101 (0.041)** −0.059 (0.189) −0.073 (0.104) 0.171 (0.201) −0.036 (0.073) 0.213 (0.241) 222

0.162 (0.065)** 0.012 (0.371) −0.071 (0.182) 0.157 (0.109) 0.064 (0.128) −0.094 (0.064) 0.164 (0.313) −0.112 (0.205) −0.283 (0.375) −0.149 (0.171) 0.027 (0.456) 222

0.069 (0.013)*** 0.084 (0.053) 0.013 (0.028) −0.042 (0.027) 0.001 (0.018) 0.000 (0.012) 0.023 (0.050) −0.003 (0.027) −0.009 (0.048) 0.014 (0.015) −0.075 (0.055) 222

0.752 (0.277)*** −2.328 (2.125) −0.222 (0.495) −0.441 (0.438) 0.698 (0.722) −0.203 (0.252) −1.617 (1.843) −1.781 (1.323) −3.868 (2.298)* −1.603 (1.227) −2.175 (2.183) 222

0.040 (0.015)*** −0.111 (0.106) −0.020 (0.028) −0.019 (0.023) 0.029 (0.037) −0.007 (0.013) −0.054 (0.093) −0.074 (0.066) −0.187 (0.118) −0.074 (0.061) −0.127 (0.108) 222

−0.001 (0.001) −0.002 (0.003) −0.002 (0.001) 0.001 (0.002) −0.001 (0.001) 0.000 (0.001) 0.008 (0.003)*** 0.006 (0.002)*** 0.007 (0.003)** 0.003 (0.001)** −0.011 (0.005)** 222

−0.0002 (0.002) 0.016 (0.007)** 0.008 (0.005)* −0.004 (0.003) −0.001 (0.002) −0.002 (0.001) −0.006 (0.006) −0.000 (0.004) −0.004 (0.006) −0.002 (0.002) −0.004 (0.005) 222

Notes: each observation is an MSA (or a CMSA in the case of larger metropolitan areas). Robust standard errors appear in parentheses below the coefficient. *** indicates p b .01; ** indicates p b .05; * indicates p b .10. The same sample is used for all regressions. All specifications include state group fixed effects. The instrumental variable is hundreds of miles of streams, intermittent streams, falls, and intracoastal waterways in the MSA (the first stage F-stat on it is 172.91). Dummy—coastal is a dummy variable for having one or more counties classified as coastal by NOAA. Dummy—on Pacific Ocean, Dummy - on Atlantic Ocean, and Dummy—on Great Lakes are indicators for bordering the Pacific Ocean, Atlantic Ocean, and Great Lakes, respectively. Dummy—on major river is an indicator for bordering a major river. Log land area is the log of the area of the (C)MSA in 1969, excluding area covered with water (in 1000s of square miles). Each day, heating degree days equal max{0, 65− mean temperature}, cooling degree days equal max{0, mean temperature − 65}, and the heating and cooling degree variables are the average monthly total during 1970–2000, over 100. Hours sunshine is the average hours of sunshine in January, during 1941–1970 (in 100s). Monthly rainfall is average monthly precipitation during 1970–2000, in inches. The standard deviation of elevation is in 1000s of feet. Sources: BEA (1969–2006), Census of Governments (1962), NCDC (2002), ESRI (2008), NCDC (2008), and NOAA (2008).

We also investigate whether our findings are due to MSAs with many county governments having relatively low incomes before 1969; if this were true, then our results might be due to conditional convergence. To the contrary, we find that doubling inter-jurisdictional competition is associated with a 1969 income per employee that is 5% higher, but a 2006 income per employee that is 10% higher. Thus, more inter-jurisdictional competition was already associated with higher incomes at the beginning of the window over which we measure growth, and this disparity only grew over the next 37 years. Inter-jurisdictional competition can make an MSA's workers more productive (a direct impact), induce workers to work more hours, or attract more productive workers to the MSA. We decompose the effects of inter-jurisdictional competition into these three channels using worker-level data from the 2000 Census. We find that just under one-half of its impact is direct, just over one-fifth comes from workers working more hours, and just under one-third comes from attracting more productive workers. We also find that MSAs with greater inter-jurisdictional competition have larger shares of their workforces engaged in relatively highly-remunerated industries. We also present evidence that inter-jurisdictional competition affects the behavior of local governments. Local governments raise more in taxes when subject to more inter-jurisdictional competition, but do not obtain more inter-governmental transfers or collect more revenue from other sources. They do issue more debt, but the costs of servicing this debt as well as the additional taxes paid are minuscule compared to average income gains. Our results here show that metropolitan areas could increase local productivity by increasing the level of inter-jurisdictional competition. However, there seems to be a great deal of hysteresis in political

boundaries in the United States 28, and our work suggests that this is not because political boundaries as currently drawn are efficient. Future research could help us understand what determines the drawing and (redrawing) of political boundaries.29 Another natural question is why people do not move between MSAs so as to mitigate productivity differences. We hypothesize that, as suggested by Roback (1982) and Albouy (2008), these wage differences may be compensated for by higher rents or lower levels of amenities.30 The results presented here also suggest many other avenues for further research. In particular, future research could verify whether the effects of inter-jurisdictional competition identified here hold in other contexts. This may be done by using a similar methodology to the one presented here, i.e., using the variation in the natural topography to instrument for the number of competing jurisdictions, to explore the effects of inter-jurisdictional competition in other countries. By doing so, such work may help to identify the powers and responsibilities that should be given to local governments in order for inter-jurisdictional competition to lead to better economic outcomes.

28 Out of over 3000 counties in the continental United States, only six new functional counties have formed since 1970, and only six functional county governments have dissolved since 1970. 29 However, see work by Alesina et al. (2004) on the drawing of local boundaries in the United States, and more generally work by Alesina and Spolaore (1997, 2005) and Alesina et al. (2000, 2005) on the size of nation-states. 30 For instance, in ongoing work, Hatfield and Kosec (2012) show that air quality is, in fact, lower in MSAs with greater inter-jurisdictional competition.

J.W. Hatfield, K. Kosec / Journal of Public Economics 97 (2013) 144–159

Appendix A. Supplementary data Supplementary data to this article can be found online at http:// dx.doi.org/10.1016/j.jpubeco.2012.08.005. References Akai, N., Sakata, M., 2002. Fiscal decentralization contributes to economic growth: evidence from state-level cross-section data for the United States. Journal of Urban Economics 52 (1), 93–108. Albouy, D., 2008. Are Big Cities Really Bad Places to Live? Improving Quality-of-life Estimates Across Cities. Alesina, Alberto, Spolaore, Enrico, 1997. On the number and size of nations. Quarterly Journal of Economics 112 (4), 1027–1056. Alesina, Alberto, Spolaore, Enrico, 2005. The Size of Nations. The MIT Press. Alesina, Alberto, Spolaore, Enrico, Wacziarg, Romain, 2000. Economic integration and political disintegration. American Economic Review 90 (5), 1276–1296. Alesina, Alberto, Baqir, Reza, Hoxby, Caroline, 2004. Political jurisdictions in heterogeneous communities. Journal of Political Economy 2, 348–396. Alesina, Alberto, Spolaore, Enrico, Wacziarg, Romain, 2005. Trade, growth and the size of countries. Handbook of economic growth, pp. 1499–1542. Baqir, Reza, 2001. Government spending, legislature size, and the executive veto. International Monetary Fund Working Paper. Bardhan, Pranab, 2002. Decentralization of governance and development. Journal of Economic Perspectives 185–205. Besley, Timothy, Case, Anne, 1995. Incumbent behavior: vote-seeking, tax-setting, and yardstick competition. American Economic Review 85 (1), 25–45. Boadway, R., Shah, A., 2009. Fiscal Federalism: Principles and Practice of Multi-Order Governance. Cambridge University Press. Brennan, Geoffrey, Buchanan, James M., 1980. The Power to Tax: Analytic Foundations of a Fiscal Constitution. Cambridge University Press. Brueckner, Jan K., 2006. Fiscal federalism and economic growth. Journal of Public Economics 90 (10–11), 2107–2120. Bureau of Economic Analysis, 2008. Regional Economic Accounts. Cadle, F.W., 1991. Georgia Land Surveying History and Law. University of Georgia Press. Calabrese, Stephen, Epple, Dennis, Romano, Richard, forthcoming. Inefficiencies from metropolitan political and fiscal decentralization: failures of tiebout competition. Review of Economic Studies. Cutler, D.M., Glaeser, E.L., 1997. Are ghettos good or bad? Quarterly Journal of Economics 112 (3), 827–872. Davoodi, Hamid, Zou, Heng-Fu, 1998. Fiscal decentralization and economic growth: a cross-country study. Journal of Urban Economics 43 (2), 244–257. Dillinger, William, 1994. Decentralization and Its Implications for Urban Service Delivery. World Bank Publications. Duranton, Gilles, Puga, Diego, 2004. Micro-foundations of urban agglomeration economies. In: Vernon Henderson, J., Thisse, Jacques-François (Eds.), Handbook of Regional and Urban Economics, vol. 4. Elsevier, pp. 2063–2117. Ebel, Robert D., Yilmaz, Serdar, 2002. On the measurement and impact of fiscal decentralization. World Bank Policy Research Working Paper No. 2809. Environmental Systems Research Institute's Data and Maps, 2008. DVD 1: Data and Maps and StreetMap USA. Hammond, G., Tosun, M.S., 2006. Local decentralization and economic growth: evidence from US metropolitan and non-metropolitan regions. Working Papers. Hatfield, John William, 2010. Federalism, taxation, and economic growth. Working Paper. Hatfield, John William, Padró i Miquel, Gerard, 2012. A political economy theory of partial decentralization. Journal of the European Economic Association 10 (3), 605–633. Hatfield, John William, Kosec, Katrina, 2012. “Federal competition and economic growth”, Mimeo, Stanford University Graduate School of Business.

159

Hayek, F.A., 1945. The use of knowledge in society. American Economic Review 35 (4), 519–530. Hoxby, Caroline M., 2000. Does competition among public schools benefit students and taxpayers? American Economic Review 90, 1209–1238. Huther, J., Shah, A., 1998. A simple measure of good governance and its application to the debate on the appropriate level of fiscal decentralization. World Bank Working Paper Series, 1894. Iimi, Atsushi, 2005. Decentralization and economic growth revisited: an empirical note. Journal of Urban Economics 57 (3), 449–461. Kim, Sang Lo, Dept. of Economics, and University of Maryland at College Park, 1995. “Fiscal decentralization, fiscal structure, and economic performance: three empirical studies”. PhD dissertation, research directed by Dept. of Economics. University of Maryland at College Park. Montinola, Gabriella, Qian, Yingyi, Weingast, Barry R., 1995. Federalism, Chinese style: the political basis for economic success in China. World Politics 48 (1), 50–81. National Association of Counties, 2009. List of Counties and Founding Years. National Climatic Data Center, 2002. Climate Atlas of the U.S., Version 2.0. National Climatic Data Center, 2008. Climate Atlas of the United States, Version 2.0. National Oceanic and Atmospheric Administration, 2008. List of Coastal Counties for the Bureau of the Census. Notowidigdo, M.J., 2010. The Incidence of Local Labor Demand Shocks. Oates, Wallace, 1993. Decentralization and economic development. National Tax Journal 46, 237–243. Panizza, U., 1999. On the determinants of fiscal centralization: theory and evidence. Journal of Public Economics 74 (1), 97–139. Roback, Jennifer, 1982. Wages, rents, and the quality of life. Journal of Political Economy 1257–1278. Rodden, Jonathan, 2004. Comparative federalism and decentralization: on meaning and measurement. Comparative Politics 481–500. Rosenthal, Stuart S., Strange, William C., 2004. Evidence on the nature and sources of agglomeration economies. In: Vernon Henderson, J., Thisse, Jacques-François (Eds.), Handbook of Regional and Urban Economics, vol. 4. Elsevier. Rothstein, Jesse, 2007. Does competition among public schools benefit students and taxpayers? Comment. American Economic Review 97 (5), 2026–2037. Seabright, P., 1996. Accountability and decentralisation in government: an incomplete contracts model. European Economic Review 40 (1), 61–89. Stansel, D., 2005. Local decentralization and local economic growth: a cross-sectional examination of US metropolitan areas. Journal of Urban Economics 57 (1), 55–72. Tiebout, Charles M., 1956. A pure theory of local expenditures. Journal of Political Economy 64 (5), 416. U.S. Bureau of Economic Analysis. 1969–2006. Regional Economic Accounts Data. http://www.bea.gov/regional/index.htm. U.S. Census Bureau, 1962. Census of Governments. U.S. Census Bureau, 1970. Integrated Public Use Microdata Series (IPUMS) USA, 1 Percent Metro Sample. U.S. Census Bureau, 1997, 2002. Census of Governments. U.S. Census Bureau, 2000. Integrated Public Use Microdata Series (IPUMS) USA, 5 Percent Sample. Weingast, Barry R., 1995. The economic role of political institutions: market-preserving federalism and economic development. Journal of Law, Economics, and Organization 11 (1), 1–31. Wildasin, David E., 1987. Nash Equilibria in Models of Fiscal Competition. Center, Univ. Wilson, John D., 1984. A Theory of Inter-regional Tax Competition. Social Systems Research Institute, University of Wisconsin, Madison. Wilson, John D., 1999. Theories of tax competition. National Tax Journal 52, 269–304. Woller, Gary M., Phillips, Kerk, 1998. Fiscal decentralisation and IDC economic growth: an empirical investigation. Journal of Development Studies 34 (4), 139–148. Zhang, T., Zou, H., 1998. Fiscal decentralization, public spending, and economic growth in China. Journal of Public Economics 2, 221–240. Zodrow, George R., Mieszkowski, Peter, 1986. Pigou, Tiebout, property taxation, and the underprovision of local public goods. Journal of Urban Economics 19 (3), 356–370.