Regional Impact of Public Transportation Infrastructure

Abstract

This research examines the regional impact of public transportation infrastructure in the northeast megaregion of the United States: public highways, railways, transit, and airports. Infrastructure stock is valued in monetary terms from 1991 to 2009. A spatial panel approach with fixed effects is adopted to test the hypothesis of spillovers by controlling for spatial dependence. The result suggests that transportation infrastructure, in general, does have a significant impact on regional economic growth, most of which is from spillover effects. Highways have an overwhelming influence through local effects and spillover effects. The impacts from public railways and airports are significant, but transit impacts are insignificant, although a positive spillover effect is found.

Keywords

regional impact multimodal transportation spatial econometrics northeast megaregion

In recent years, there has been an increasing debate about how to effectively allocate public funding for U.S. transportation infrastructure. Some favor a traditional approach that focuses on continued investment in the primary highway network. Others support an alternative mix mode approach, represented by high-speed intercity passenger rail but including expansion of public transit and air transport. This debate is particularly heated in regions such as the U.S. northeast megaregion where passenger intercity rail and public transit is widely utilized. With passage of the Federal Aviation Administration (FAA) Modernization and Reform Act of 2012 and the Moving Ahead for Progress in the 21st Century (MAP-21), the U.S. transportation infrastructure received additional financial resources for coming years. How to allocate these funds or future public funding wisely and efficiently so that a higher level of socioeconomic return can be achieved is a critical challenge for decision makers. It is critical to establish an understanding of the regional economic impacts of the transportation infrastructure system. To achieve this objective, this study is conducted to add a piece of new evidence to the literature.

This study follows the goals and questions concerning the impact of transportation investment in regional output explored in an earlier article using different methodology and data sources (Chen & Haynes, 2013). The results are similar, which gives us some confidence in the robustness of findings presented here. The study differs from previous studies in the following respects.

First, the focus of the investigation is on multiple modes of transportation infrastructure in the northeast megaregion in the United States, which includes highways, public railways,¹ public transit,² and public airports.³ This research design allows us to evaluate the relative importance of different transportation modes in terms of output elasticity.

Second, the study considers a mature transportation system. A mature transportation system refers to a system that is mostly completed and is currently at a developed rather than at a construction stage. At the stage of construction, economic impacts of transportation infrastructure are not primarily derived from network effects such as a better connectivity and accessibility. Instead, impacts are mostly derived from construction-related activities, and job creation is driven by the demand for raw materials, equipment, engineering design, and technology installation. At the developed stage, because transportation networks are completed, impacts from added transportation infrastructure can further be achieved through network effects as well as the effects from maintenance, operation, and continuous technology-upgrading activities. Furthermore, user behavior is stabilized in mature systems. Because of these different effects, the marginal effects of transportation infrastructure are not constant across stages as the system is built out.

Recognizing the different stages of transportation infrastructure development is important, especially when doing project comparison or modal comparison analysis, as it helps avoid the mistake of “comparing apples and oranges.” Meanwhile, such a consideration helps improve the appreciation of the economic impacts of critical infrastructures under similar conditions.

Third, this study focuses on the northeast megaregion in the United States, which is also referred to variously as megalopolis (Gottmann, 1961) and BOSWASH (Kahn & Wiener, 1967). This is an important region but lacks sufficient investigation in terms of economic impact of transportation infrastructure. The metropolitan statistical area (MSA) is selected as the scale of this analysis because most economic activities occur in MSAs. In addition, the northeast corridor primarily refers to the railway corridor between Boston and Washington, D.C.⁴ The rail system is owned by the National Railroad Passenger Corporation (Amtrak), which is a public transportation agency that serves between Boston in the north via New York City and Washington, D.C., in the south. Conducting the study at the MSA level is useful for capturing the scale effects of transit and public rail infrastructure on regional economic performance. A list and a map of all MSAs are illustrated in Appendices A and B, respectively. In total, the study covers 30 MSAs and 2 micropolitan statistical areas (microSAs).

This study addresses the following research questions: What are the regional impacts of transportation infrastructure in the northeast megaregion? How do such impacts vary across public transportation infrastructure modes: highways, rail, transit, and airports? Are there any spillover effects from public transportation infrastructure investment? And if yes, how do these effects vary across transportation modes?

Following the research discussion, the study lays a theoretical foundation for the assessment based on literature. Data and methodology are then introduced, with an interpretation of the empirical results and a conclusion closing the study.

Literature Review

The regional impact of transportation infrastructure is reviewed from the perspective of traditional economic theory and the new economic geography. Meanwhile, the need for a multimodal investigation is also discussed.

Traditional Theory

Given that a series of articles by Aschauer (1989, 1990, 1994) argued that enhancing public infrastructure expenditures will help regions to achieve their economic potential, a large number of studies on public infrastructure were conducted following a neoclassical approach using various forms of aggregated production functions (Bhatta & Drennan, 2003; Boarnet, 1997; Boarnet & Haughwout, 2000; Duffy-Deno & Eberts, 1991; Fernald, 1999; Gramlich, 1994, 2001; Harmatuck, 1996; Mattoon, 2002; Nadiri & Mamuneas, 1996). Given the nature of different evaluation methods, time periods, measures of economic outcome, and control variables, findings are inconsistent. Some argue that transportation infrastructure, such as the highway system, had a positive and a large impact on productivity (Fernald, 1999; Harmatuck, 1996; Keeler & Ying, 2008); others suspect the magnitude of such an impact (Boarnet, 1997; Button, 1998).

Specifically, Aschauer (1989) and Munnell and Cook (1990) analyzed the relationship between public capital and economic performance from 1970 to 1986 at the national and state levels, respectively. The output elasticity of public capital stock was found to be 0.38 to 0.56 (Aschauer, 1989) and 0.15 (Munnell & Cook, 1990), respectively, with highways alone contributing 0.06 (Munnell & Cook, 1990). By focusing on nonmilitary public capital in the period between 1949 and 1985, Harmatuck (1996) found the economic impact from such investment to be positive and significant. However, Lau and Sin (1997) found different returns to public capital of around 0.1, which was much smaller than what Aschauer and Munnell had found. These earlier studies (Aschauer, 1989; Munnell & Cook, 1990) were determined to suffer from a series of statistical issues that may have jeopardized their results (Jorgenson, 1991).

Scholars also argue that the scale of analysis matters in this kind of research. The rate of return becomes less significant when investigated at the state level as opposed to the national level. By using a general equilibrium model as well as state-level public capital data, Holtz-Eakin and Lovely (1996) found that public capital did not significantly affect output. Garcia-Mila, McGuire, and Porter (1996) also found no positive relationship between public capital and private output.

Those studies using econometric estimation have been subjected to a variety of criticisms. Gramlich (1994) summarized these problems as follows:

Unclear causal relationship between infrastructure and economic performance

Poor use of policy variables that were inconsistent in terms of level with relevant infrastructure variables

Lack of isolation of factors influencing macroeconomic performance: mixing transport with higher level variables and often leaving out issues in soft infrastructure including law, education, business services, and defense

Different methodologies are applied to different types of data sets, resulting in implications that attribute an imprecise weight to quantitative estimates

Spatial Perspective

Most early studies did not consider spatial interactions among units across geographic locations. Those studies normally assumed spatial homogeneity. According to spatial theory, estimation outcomes may vary significantly if spatial dependence is not considered in regional analysis (LeSage & Pace, 2009). Munnell (1992) indicated that the estimated impact of public capital becomes smaller as the geographic focus narrows. She suggested this is because of the effects of leakages from an infrastructure investment could not be captured at a small geographic level. Although this hypothesis may not be entirely accurate, as indicated by Boarnet (1998), it does suggest that the spatial dimension influences estimation and should not be neglected.

LeSage (1999) emphasized that traditional econometrics has largely ignored the spatial dimension of sample data. When data contain geographic information, the issue of spatial dependence between observations may violate the Gauss–Markov theorem. Without considering this spatial issue, the estimation results may be statistically biased.

Thanks to the development of spatial econometric techniques by Paelinck and Klaassen (1979), Cliff and Ord (1981), Anselin (1988), LeSage and Pace (2009), Elhorst (2014), and many others, a number of empirical spatial analytical methods were developed. One of the dominant functions is to allow for the measurement of spatial spillover effects. These effects refer to situations in which the input in one sector or region influences changes in neighboring local economies through trade linkages and market relationships (Bo, Florio, & Manzi, 2010). Transportation infrastructure may have a spillover effect on regional economic growth because the benefits generated from infrastructure would not be confined to that specific region (Moreno & López-Bazo, 2007). To test the hypothesis empirically, different types of spatial models were adopted (Cohen & Morrison Paul, 2003, 2004; Holtz-Eakin & Schwartz, 1995; Kelejian & Robinson, 1997).

Reviews of the existing literature regarding spatial impact of transportation infrastructure (Boarnet, 1998; Cohen, 2007; Cohen & Morrison Paul, 2003, 2004; Holtz-Eakin & Schwartz, 1995; Kelejian & Robinson, 1997; Mohammad, 2009; Moreno & López-Bazo, 2007; Ozbay, Ozmen-Ertekin, & Berechman, 2007) indicate that conclusions are not consistent given the fact that different data, methods, regions, and periods were used. Despite the development of spatial econometric techniques enabling scholars to investigate spillover effects of infrastructure in a more comprehensive way, most studies have considered spatial lag dependence or spatial error dependence alone. In the circumstance when no solid evidence indicates whether a spatial dependence or a spatial error model is preferred, LeSage and Pace (2009) recommend a spatial Durbin model be applied. Without adequate interpretation of the reasons for why a specific spatial model is used, it is likely that these studies may have estimation bias because of the neglect of a specific spatial issue.

Unimodal Versus Multimodal

Another common feature of infrastructure impact studies is that many of them investigate transportation from a unimodal perspective. Some focus on public capital or transportation infrastructure in aggregate (Berndt & Hansson, 1992; Duffy-Deno & Eberts, 1991; Kelejian & Robinson, 1997), whereas others only focus on a specific mode such as highways or airports (Cohen, 2007; Cohen & Morrison Paul, 2003, 2004; Holtz-Eakin & Schwartz, 1995; Ozbay et al., 2007). Very few studies investigate the issue from a multimodal and comparative perspective (Anderson, Anderstig, & Harsman, 1990; Blum, 1982; Cantos, Gumbau-Albert, & Maudos, 2005).

Theoretical Motivation

The theoretical motivation of this study is to follow the path of the new economic geography to test the spatial effects of public transportation infrastructure. As one of the world’s largest agglomerations, the northeast megaregion functions as one of the primary engines of economic growth and development in the world economy. Transportation plays a central role on regional economic activities, both interregionally and intraregionally. On one hand, completion of transportation infrastructure networks among two regions may benefit each other because of better connectivity and accessibility. Regional economic growth could be achieved because of the significant reduction of transportation costs of both goods delivery and labor mobility (Krugman, 1991). On the other hand, economic agglomeration may happen because of declining of spatial and temporal distance. Labor and raw material may start to flow into one region from other regions (Fujita, Krugman, & Venables, 1999). Consequently, the growth of one region may be achieved while leaving other regions stagnant when assuming the existence of scant resources in the society.

This unequal regional impact of transportation infrastructure may also happen because of the competitive nature of public investment. In other words, positive economic growth is likely to be achieved when a heavy public investment occurs in one region relative to other regions. This may induce a negative impact on regional growth in other regions because of insufficient public investment. In sum, whether spillover effects exist among different modes of transportation deserves a thorough investigation.

As Fingleton and López-Bazo (2006) pointed out, many regional studies modeled externalities in a somewhat ad hoc manner that often fails to consider the causes of externalities. For example in Boarnet’s (1998) path-breaking work, spatial dependence was only considered for the variable of streets and highways. The externalities of regional output as well as labor and private capital were not mentioned. Given the nature of his modeling structure and his finding of a negative spillover of public streets and highways, this may be an issue.

In this study, we test Boarnet’s hypothesis of negative output spillovers from public transportation infrastructure. To make the analysis consistent with Boarnet’s study, a neoclassical model in the form of a Cobb–Douglas production function is established. In addition, the study expands Boarnet’s work in the following ways. First, public infrastructure includes not only highways and streets but also public rail, public transit, and airports. This multimodal focus will enable us to differentiate the relative importance of transportation infrastructure by mode and their impact on regional output. Second, a test of endogeneity is conducted to identify the direction of potential causation between output and capital input. Third, a systematic spatial modeling selection approach is introduced to achieve a rigorous estimation.

Data and Methodology

This section outlines the analytical process, which is divided into five steps: data selection, data refining, spatial modeling, preliminary tests, and final analysis.

Data Selection

Variables are selected with consideration to research questions as well as availability. Because of data limitations, it is a challenge to find transportation capital stock data measured in monetary terms. Many studies use approximate variables as alternatives, such as miles of highways and rail lines, numbers of air passengers (Anderson et al., 1990; Jiwattanakulpaisarn, Noland, Graham, & Polak, 2009), or numbers of freight rail stations (Blum, 1982), to quantify infrastructure stock. These physical unit substitutes for financial data, although relevant in engineering terms, are less satisfactory in economic terms.

Whereas most of the economic variables are publicly available, transportation variables measured in monetary terms are difficult to find. This is an even greater challenge for investigation at a disaggregated level both spatially and by mode. In the United States, most capital stock data are only available at the aggregated level provided by the U.S. Bureau of Economic Analysis (BEA). The most relevant transportation capital stock data by mode can be found in the Transportation Statistics Annual Reports published by the U.S. Bureau of Transportation Statistics (BTS).

These data, however, are not adopted in this study because the data are only available for years 1998 to 2008. Earlier years of data were not collected. In addition, majority of the data reflect privately owned capital stock levels. Public capital stock of rail and transit are unfortunately combined in the category “other publicly owned transportation.” A further follow-up with officials in BTS and BEA confirmed the impossibility of disaggregating these data. Therefore, the public rail, transit, and airport capital stock in the northeast megaregion are estimated based on the financial information gathered from the U.S. Census Bureau and Amtrak.

Data Refining

All variables are converted into per capita measured in 2005 real dollar terms. Variables at a per capita level instead of at a gross level help to reduce the influence of demographic variations and the size effect of each MSA. The details of the data refining process are illustrated in Table 1. The variables are refined through several steps. The first step is to aggregate or disaggregate the original data to the unified MSA level based on an apportioning weight. Private fixed asset was disaggregated from the national level by using industrial earnings as a weight (the ratio of total industrial earnings of each MSA divided by the national level industrial earnings). Data of gross metropolitan product (GMP) per capita (adjusted to 2005 dollars) and employment are directly retrieved from the BEA website (www.bea.gov).

Table 1.

Data Refining Process.

	Original data	Sources	Initial stock	Method
GMP	Regional Data, GDP by State	BEA	—	—
Private fixed asset	Table 6.1. Current-Cost Net Stock of Private Fixed Assets by Industry Group and Legal Form of Organization	BEA	—	Following Garofalo and Yamarik (2002), the national capital stock estimates is apportioned to each metro area using annual private industry earning as a proxy.
Private fixed asset	Table 6.2. Chain-Type Quantity Indexes for Net Stock of Private Fixed Assets by Industry Group and Legal Form of Organization	BEA	—
Employment	CA04 Total Employment (number of jobs)	BEA	—
Public rail capital stock	Amtrak NEC Infrastructure Expenditure	Amtrak	Federal grant to NEC project from 1972 to 1990^b	Perpetual inventory method (PIM), infrastructure expenditure is apportioned by ridership. Capital procurement is based on geographic location.
	Amtrak Fact Sheets of States VA, DC, MD, DE, PA,NJ, NY, CT, RI, MA, NH, and ME	Department of Engineering
		Department of Public Affairs^a
Highway capital stock	Annual Survey of State and Local Government Finances and Census of Governments (1970-2009)^c	U.S. Census Bureau	PIM estimated based on 1970-1990	Aggregated from individual government units based on their jurisdictional locations.^d
Transit capital stock^e	Annual Survey of State and Local Government Finances and Census of Governments (1970-2009)	U.S. Census Bureau	PIM estimated based on 1970-1990	Aggregated from individual government units based on their jurisdictional locations.
Airport capital stock	Annual Survey of State and Local Government Finances and Census of Governments (1970-2009)	U.S. Census Bureau	PIM estimated based on 1970-1990	Aggregated from individual government units based on their jurisdictional locations.
Population	BEA CA1-3 Person Income Summary^f	U.S. Census Bureau	PIM estimated based on 1970- 1990	Aggregated from individual government units based on their jurisdictional locations.

Note. BEA, Bureau of Economic Analysis; GMP, gross metropolitan product; GDP, gross domestic product.

Northeast corridor capital infrastructure expenditures include data from 1990 to 2009 for the main line and is obtained from the Department of Engineering at Amtrak. Data only include capital expenditures on safety and reliability, high-speed rail facilities. Data include federal, state, and local government expenditures. ^bThe initial railway infrastructure stock of the northeast corridor contains two parts. The majority of the corridor assets were purchased from the Consolidated Rail Corporation (Conrail) during 1976-1980 as part of the disposition of the Penn Central Transportation Company’s assets (U.S. Government Accountability Office, 2004, p. 7). The second part was formed from the Northeast Corridor Improvement Project from 1976 to 1990 (Federal Railroad Administration, 1998). ^cData include federal, state, and local government expenditures. ^dDetails of calculation can be provided on request. ^eTransit modes include bus, commuter rail, light rail, and personal transit. ^fU.S. Census Bureau midyear population estimates.

Second, the focus of our assessment is on transportation capital stock. The concept represents stock rather than flow. Transportation capital stock is adopted as the indicator of transportation infrastructure inputs. Since no disaggregated transportation infrastructure stock data are publicly available, they have to be estimated manually. Before the estimation, two assumptions were made. First, to simplify the estimation process, all transportation capital stock are assumed to be static and nonmovable for the period of assessment; and second, all movable transportation stock such as bus and railway rolling stock, are assumed to be possessed by the geographic locations where their headquarters are located. The stock of each mode is calculated through the following function using the perpetual inventory method (PIM):

T K_{t} = (1 - δ) T K_{t - 1} + T I_{t},

where TK and TI indicate transportation capital stock and transportation capital expenditure, respectively, and δ denotes the geometric depreciation rate of transportation stock. The geometric depreciation rate has been regarded as the appropriate value for infrastructure asset studies and has been commonly adopted by BEA (Katz & Herman, 1997). As far as the value of δ is concerned, the rate of 4.1% is adopted in accordance with other studies (Holtz-Eakin, 1993; Hulten & Wykoff, 1981; Ozbay et al., 2007). The initial capital stock levels (TK₁₉₉₀) of each mode are estimated using PIM based on public capital expenditure from 1970 to 1990.

The trade-off between regional level of analysis and accuracy of data is an issue that to which particular attention should be paid for the impact assessment of capital stock. Although the assessment may be easier to implement at the aggregated level (such as the national level or state level) given the data availability, the assessment becomes challenging at the disaggregated level (especially at the MSA or county level) as data are usually not available and capital stock has to be estimated. As a result, it should be noted that as the data being estimated move away from the “ideal,” so does the level of confidence in the results.⁵ Hence, it is important for any user to carefully review and assess such disaggregation for the level of confidence needed.

Next, all stock per capita is converted to real 2005 dollars to eliminate the influence of inflation. The World Bank gross domestic product (GDP) deflator for the United States is applied to all transportation stock variables. The temporal distributions of all variable means are illustrated in Figure 1. After the logarithmic transformation, the means of all variables are stable between 1991 and 2009, which suggests that the per capita-based variables are stationary. The descriptive statistics for the distributions of all variables are described in Table 2.

Figure 1.

Temporal variations of the variable means.

Table 2.

Descriptive Statistics.

Variables	M	SD	Min	Max	Unit
GMP per capita (gmppc)	41,387	9,341	26,408	89,688	2005$
Employment (emp)	874,057	1,846,585	41,942	1.10e+07	No. of jobs
Private capital per capita (pfapc)	95,733	23,109	37,527	220,121	2005$
Highway capital per capita (hwyspc)	714	330	108	2,168	2005$
Public rail capital per capita(amspc)	71	103	0	668	2005$
Transit capital per capita (traspc)	281	649	0	3,871	2005$
Airport capital per capita (airspc)	95	148	0	928	2005$
Total transport capital per capita (ttspc)	1,161	896	329	5,935	2005$

Note. Total number of observations: 608. gmppc = GMP per capita, emp = employment, pfapc = private fixed asset per capita, hwyspc = highway capital stock per capita, amspc = public rail capital stock per capita (Amtrak Northeast Corridor), traspc = transit capital stock per capita, airspc = public airport capital stock per capita, ttspc = total transportation (highways + public rails + transits + public airports) capital stock per capita. Source. Bureau of Economic Analysis and U.S. Census Bureau.

The distribution of transportation infrastructure in the northeast megaregion is uneven in terms of both mode and geography. Highway has the highest value of average stock per capita whereas public rail has the lowest. With regard to geographic comparison, regional differences in stock are very large. For instance, the highest highway stock per capita is $2,168, which is for the Ocean City MSA in New Jersey in 2009. The lowest amount is $108, which is in the Boston–Cambridge–Quincy MSA in 1991. The Trenton–Ewing MSA in New Jersey has the highest amount of public rail stock per capita, whereas the amount equals zero in MSAs such as Willimantic microSA in Connecticut, the Lebanon MSA in Pennsylvania, and the Lewiston–Auburn MSA in Maine where there is no public rail service. Transit stock has a similar distributional pattern. The New York–Northern New Jersey–Long Island MSA has the highest amount of transit stock per capita, whereas MSAs such as the Vineland–Millville–Bridgeton MSA in New Jersey and the Willimantic microSA in Connecticut have no public transit stock. The Washington–Arlington–Alexandria MSA has the highest public airport capital stock per capita, although some regions have almost no public airport capital stock because of negligible amounts of public airport expenditures. In sum, the distribution of transportation infrastructure in the northeast megaregion is quite uneven. Most stocks are clustered in urbanized and population-dense MSAs along the northeast corridor, such as Washington D.C., Baltimore, Philadelphia, New York, and Boston. This may further imply the existence of a positive spatial autocorrelation.

Spatial Modeling Structure

To test Boarnet’s hypothesis of negative spillover effects of public infrastructure, the study follows the same neoclassical model structure in a Cobb–Douglas production function form. The basic equation is defined as

Y_{i, t} = f (L_{i, t}, K_{i, t}, T_{i, t}) + u_{i, t},

where Y denotes the economic output in region i at period t, which is measured by GMP per capita, and L and K denote level of employment and private capital asset per capita, respectively. T is the key policy variable, which denotes transportation infrastructure stock per capita, either in total or by mode. All variables are converted in the logarithmic term, so the coefficients from the log-linearized estimation can be interpreted as elasticity. The disturbance term u_i,t is specified as

u_{i, t} = μ_{i} + ν_{i, t},

where µ_i represents MSA-specific effect assumed to be independent and ν_i,t is a classical random disturbance, which is assumed iid $(0, σ_{ν}^{2})$ . u_i,t can be modeled as either fixed or random effect. The MSA-specific effect includes regional specific factors for output such as “endowment of natural resources, the quality of public infrastructure, physical characteristics of a MSA, the ability to attract and utilize foreign investment and network effects” (Pinnoi, 1994, p. 130).

Preliminary Tests

In the fourth step, three preliminary tests are implemented so as to provide supportive information for model selection.

The first is to test whether all the variables are stationary. This is an important prerequisite of regional impact analysis as any use of nonstationary data may lead to a spurious estimation. In addition, given the spatial nature of data being used, a panel unit root test with a consideration of cross-sectional dependence is also needed. The cross-sectional augmented Dickey Fuller (ADF)/covariate-augmented Dickey Fuller (CADF) statistics method proposed by Pesaran (2007) is adopted and implemented in STATA using the pescadf command (Lewandowski, 2007). The results of the standard panel stationary tests (Levin, Lin, & Chu, 2002), as summarized in Table 3, suggest that all the variables are statistically significantly stationary. Although variables are generally significant at the 10% level in the CADF test, the variable representing the total transportation capital per capita is insignificant even at the 10% level. This is possible because of the strong spatial dependence of transportation infrastructure across the northeast megaregion.⁶

Table 3.

Panel Unit Root Test (32 Metropolitan Statistical Areas, 1991-2009).

	Levin, Lin, and Chu (2002)			Pesaran’s CADF
Variable name	Stat.	Prob.	Obs	Stat.	Prob.	Obs
GMP per capita (lgmppc)	−9.279	.000	561	−4.256	.000	544
Employment (lemp)	−7.311	.000	533	−1.519	.064	544
Private capital per capita (lpfapc)	−5.894	.000	521	−1.574	.058	544
Total transport capital per capita (lttspc)	−4.414	.000	570	−1.228	.110	544
Highway capital per capita (lhwyspc)	−8.989	.000	564	−1.832	.033	544
Public rail capital per capita (lamspc)	−1.944	.026	403	—	—	544
Public airport capital per capita (lairspc)	−3.203	.000	425	—	—	544
Public transit capital per capita (ltraspc)	−13.675	.000	425	—	—	544

Note. All variables were measured in level and in logarithmic term. Automatic lag length selection based on SIC: 0 to 3. “—” indicates no test result is generated because of containing zero numbers; CADF = covariate-augmented Dickey Fuller.

An important issue regarding the regional impact analysis of transportation infrastructure is the endogeneity between transportation stock and economic output. On one hand, transportation investment enhances the connection of the regional transportation network, which subsequently facilitates both freight and passenger movement by reducing the generalized transportation cost. On the other hand, the improvement of economic performance may, as a consequence, lead to an increase in demand for both freight and passenger mobility, which thus requires more investments for transportation infrastructure improvement. Failure to recognize this endogenous issue may severely jeopardize the outcome of investigation and may further lead to mistaken policy implication. Therefore, the issue of endogeneity must be addressed before any concrete impact analysis is attempted.

In general, the classic Granger causality test (Granger, 1969) can be used to test the temporal directional linkage between regional output and economic input. Of course “causality” is a difficult and complex issue but the granger test establishes the necessary but insufficient condition for causality (i.e., which comes first in time). This is critical for identifying and understanding endogeneity, among other things. As illustrated in Table 3, the stationary test indicates that all economic output and input variables are stationary, which implies the data are cointegrated and can be utilized directly for a Granger test. Using the generalized method of moments (GMM) estimator by controlling for the influences of labor and private capital, the results of the Granger test suggest that endogeneity between transportation inputs and the regional economic outputs is not a serious issue in our assessment. Transportation infrastructure is found to “Granger cause” the GMP per capita, whereas the reversed effects were not found (Table 4).⁷

Table 4.

Panel Granger Causality Tests.

	GMPPC-TTSPC		GMPPC-HWYSPC		GMPPC-AMSPC		GMPPC-TRASPC		GMPPC-AIRSPC
Dependent Variable	gmppc	ttspc	gmppc	hwyspc	gmppc	amspc	gmppc	traspc	gmppc	airspc
Independent Variables
gmppc (-1)	1.094***	0.003	1.067***	−0.015	1.059***	−0.513*	1.047***	−0.586	1.011***	0.168
ttspc
ttspc (-1)	−0.02***	0.992***
hwyspc					−0.013**	0.013	−0.011***	0.038	−0.007**	−0.105**
hwyspc (-1)			−0.014**	0.988***
amspc			−0.005**	0.000			−0.005***	0.002	−0.005**	−0.030**
amspc (-1)					−0.005***	0.997***
traspc			−0.001	0.000	−0.002	−0.026*			−0.002	−0.017
traspc (-1)							−0.002**	0.989***
airspc			−0.003	−0.007**	−0.004**	0.011	−0.004***	−0.009
airspc (-1)									−0.006**	0.938***
emp	0.012***	0.009**	0.014***	−0.006	0.015***	0.010	0.014**	0.015	0.016***	0.135***
pfapc	−0.086***	−0.032*	−0.065**	−0.061	−0.030	0.325	−0.048**	0.026	−0.019	−0.212**
Arellano–Bond test for AR(2), (p value)	0.682	0.147	0.074	0.123	0.07	0.366	0.100	0.117	0.101	0.526
Wald test (H0: lags = 0)	179.85***	0.01	4.67**	0.24	35.53***	3.54*	5.57**	0.42	36.22***	3.48
Number of observations	576	576	306	306	300	300	305	305	306	305

Note. All variables were measured in level and in logarithmic term. All models are estimated using the Arellano and Bond dynamic panel system generalized method of moments (GMM) estimations. gmppc = GMP per capita, emp = employment, pfapc = private fixed asset per capita, hwyspc = highway capital stock per capita, amspc = public rail capital stock per capita (Amtrak Northeast Corridor), traspc = transit capital stock per capita, airspc = public airport capital stock per capita, ttspc = total transportation (highways + public rails + transits + public airports) capital stock per capita.

Significant at the 10% level. **Significant at the 5% level. ***Significant at the 1% level. All other tests assume asymptotic normality.

To further confirm the results of the Granger test, the Hausman test is conducted. The rationale of the Hausman test is to compare instrumental variable (IV) estimates using the two-stage GMM estimator to ordinary least squares (OLS) estimates. If a significant difference were found between the two estimates, then the test suggests that endogeneity does exist and the two-stage IV-GMM estimator is preferred.

The results of the Hausman test for endogeniety and the Hansen J test for overidentifying restrictions are reported in Table 5. The insignificance of the Hansen J tests suggests that all instrumental variables are valid. The first total transportation stock per capita (TTSPC) model tests whether the total public transportation capital per capita (lttspc) is endogenous in the model, where GMP per capita is the dependent variable and labor and private capital per capita are exogenous variables. The null hypothesis is that lttspc is properly exogenous in the model. The test statistic has a p value of .177, suggesting that the test cannot reject the null hypothesis. In other words, the total public transportation infrastructure input variable is exogenous in the model. Likewise, in the four modes model, highway capital per capita (lhwyspc), public airport capital per capita (lairspc), public rail capital per capita (lamspc), and public transit capital per capita (ltraspc) are treated as endogenous regressors. The Hansen J test shows all instrumental variables are valid to estimate coefficients at the 5% level. The Hausman test suggests that the four transportation variables are statistically significantly exogenous to the modeling structure. Based on the results of both the Granger causality test and the Hausman test, the endogenous issue of transportation infrastructure is not considered in this assessment.

Table 5.

Hausman Test for Endogeniety.

	TTSPC model		Four modes model
	lttspc		lhwyspc, lamspc, ltraspc
Regressors tested	χ²	p value	χ²	p value
Hansen J statistics (overidentification test of all instruments)	0.000	.989	8.518	.074
Endogenous test of endogenous regressors	1.041	.308	4.336	.363

Note. The four transportation infrastructure input variables are tested separately following the per capita–based production function form. Lag variables of the endogenous regressor are treated as instrumental variables at the first stage. Employment and private fixed-asset per capita are treated as exogenous control variable in the second stage. Regional output variable (lgmppc) is the dependent variable. The null hypothesis for the Hansen J test is that instrumental variables are valid. The null hypothesis for endogenous test is that the tested regressor is exogenous. TTSPC = total transportation stock per capita.

The third test is to check whether spatial autocorrelation exists. The spatial autocorrelation, measured by the value of Moran’s I, is tested for all variables by the software GeoDa. The universal global Moran’s I is defined as (Cliff & Ord, 1981; Moran, 1950)

I = \frac{n}{\sum_{i = 1}^{n} \sum_{j = 1}^{n} w_{i}_{j}} \cdot \frac{\sum_{i = 1}^{n} w_{i j} (x_{i} - \bar{x}) (x_{j} - \bar{x})}{\sum_{i = 1}^{n} w_{i j} {(x_{i} - \bar{x})}^{2}},

where n is the number of MSAs, which in our case equals 32, x and $\bar{x}$ denote the specific MSA and the mean of x, respectively. w_ij is the spatial weight matrix, representing the spatial relationships between region i and j. This spatial relationship in the study is defined as being contiguous to each other; thus the spatial weight matrix is generated using the Queen contiguity method.⁸

The global Moran’s I of all the variables are displayed in Table 6. Interestingly, except for employment, the Moran’s I values of a few variables are significant, which indicate spatial autocorrelations exist across most of the variables. Negative values of Moran’s I are found for the employment and transit capital, which indicates a tendency toward dispersion. With respect to GMP per capita, private fixed-asset per capita, total transportation capital, public highway capital, and rail capital, positive and significant Moran’s I values are found, indicating a tendency toward clustering. The existence of spatial dependence among both the dependent variable and independent variables implies a complicated spatial issue for this analysis.

Table 6.

Descriptive Statistics of Spatial Dependence (Global Moran’s I).

	gmppc	emp	pfapc	ttspc	hwyspc	airspc	amspc	traspc
1991	0.214**	−0.073	0.214**	0.249**	0.084	0.070	0.209*	−0.003
1992	0.195*	−0.074	0.195*	0.233*	0.106	0.100	0.199*	−0.006
1993	0.185**	−0.075	0.185*	0.221*	0.131	0.099	0.197*	−0.011
1994	0.188**	−0.075	0.188**	0.205**	0.162	0.099	0.190**	−0.020
1995	0.189***	−0.075	0.189*	0.198*	0.200**	0.112	0.187*	−0.023
1996	0.182**	−0.076	0.182*	0.190**	0.224**	0.101	0.187*	−0.033
1997	0.171*	−0.076	0.171*	0.187	0.210*	0.135	0.191	−0.043
1998	0.148	−0.076	0.148*	0.197**	0.215**	0.152*	0.192*	−0.047
1999	0.154*	−0.075	0.154*	0.195**	0.225*	0.126	0.199*	−0.050
2000	0.157**	−0.075	0.157**	0.189*	0.214**	0.121	0.200**	−0.056
2001	0.168	−0.074	0.168*	0.185*	0.216*	0.092	0.199**	−0.056
2002	0.160*	−0.072	0.160	0.176*	0.219**	0.083	0.200**	−0.054
2003	0.151	−0.070	0.151	0.163	0.219**	0.079	0.195**	−0.055
2004	0.150*	−0.069	0.150**	0.138	0.218**	0.085	0.195**	−0.067
2005	0.153	−0.068	0.153*	0.121	0.219*	0.088	0.192**	−0.073
2006	0.160*	−0.068	0.160*	0.114	0.240**	0.095	0.197*	−0.077
2007	0.175*	−0.068	0.175*	0.101	0.237***	0.101	0.192**	−0.083
2008	0.182**	−0.069	0.182*	0.102	0.252**	0.109	0.185*	−0.084
2009	0.170*	−0.068	0.170*	0.089	0.257**	0.112	0.185**	−0.086

Note. ***, **, and * denote that coefficients are significant at 1%, 5%, and 10% statistical levels, respectively. The spatial weight matrixes are generated based on Queen contiguity method.

Spatial Analysis

From a theoretical perspective, it is reasonable to assume spatial autocorrelation may possibly exist among both dependent and independent variables. Metropolitan statistical areas such as Washington D.C., Baltimore, Philadelphia, New York, and Boston function as regional growth poles that may have strong economic relations with their neighboring MSAs. Thus, a spatial dependence may exist in the regional output variable. On the other hand, spatial autocorrelation may also exist among independent variables such as employment, private capital, and transportation infrastructure across different MSAs in the northeast region because of the agglomeration effect and integrated urbanization.

To avoid ad hoc modeling specification and also to adequately accommodate various potential spatial externalities, a spatial Durbin model (SDM) that includes the spatial lags of both dependent and independent variables is adopted as the initial model form. The general form of SDM recommended by LeSage and Pace (2009) is denoted as

Y_{i t} = ρ \sum_{j = 1}^{n} W_{i j} Y_{j t} + X_{i t} β + \sum_{j = 1}^{n} (W_{i j} X_{j t}) θ + μ_{i} + ν_{i t},

ν_{i, t} \sim N (0, σ_{i, t}^{2}),

where Y and X denote the dependent and explanatory variables, respectively. WY and WX denote the spatial lag terms of dependent variable and explanatory variables, respectively. ρ, β, and θ denote coefficients that would be estimated.

To help identify the appropriate spatial panel model in a systematic way, we follow the Elhorst (2014) spatial model testing procedure to test which spatial model is preferred. Table 7 displays the test results of Elhorst’s routine. Although the Lagrange multiplier (LM) test suggests a spatial lag model is preferred, the general test (with consideration of a likelihood ratio [LR] test) recommends that a spatial Durbin model is more efficient. The Hausman test confirms that fixed effect is preferred.

Table 7.

Lagrange Multiplier (LM) Test and Likelihood Ratio (LR) Test.

	TTSPC model	Four modes model
	Test statistic (p value)	Test statistic (p value)
LM lag	94.753 (.000)	118.959 (.000)
LM error	27.861 (.000)	49.379 (.000)
LM lag robust	71.251 (.000)	70.866 (.000)
LM error robust	4.358 (.037)	1.306 (.253)
H0: θ = 0
LR value	130.092 (.000)	138.604 (.000)
H0: θ + ρβ = 0
LR value	217.746 (.000)	233.772 (.000)

One of the key functions of spatial analysis is to investigate the spatial effects among different MSAs. Because the spatial information of neighboring regions is added in the form of a spatial weight matrix, the SDM is endowed with the capacity to separate spatial effects from total effects (LeSage & Pace, 2009). As a result, three types of impacts can be estimated through the spatial model: average direct impact, average indirect impact, and average total impact (LeSage & Pace, 2009).The first impact measures the influences of the explanatory variables that come from the same geographic unit as the dependent variable. The second impact measures influences of explanatory variables that come from different geographic units. The third impact consists of both the direct impacts and indirect impacts. Although some studies (for instance, Mohammad, 2009) give these impacts different names, such as long-term local effect, long-term neighbor effect, and long-term total effect, respectively, they represent the same classified effects as specified here.

To make the analysis consistent with Boarnet’s study, this study focuses on mature transportation infrastructure only, thus a constant regional impact of transportation infrastructure is considered in this investigation. In addition, only spatial fixed effect is considered in our investigation. This makes the results comparable to previous studies. Given the fact that the panel unit root test confirms that public transportation infrastructure and the economic variables are stationary during the research period, the inclusion of potential time fixed effect may add influences that may contaminate the model estimation and purpose.

Empirical Results

Some studies, for example, Cohen and Morrison Paul (2007), suggest that higher order spatial weight matrix are important as transportation infrastructure may have spillover effects not only on its adjacent neighbors, but that its impact may spill over further to higher order neighbors.⁹ To demonstrate the robustness of our analysis, different spatial weight specifications including inverse distance and various k nearest neighbors are adopted. Table 8 displays estimation results of the regional impacts of the aggregated transportation infrastructure capital stocks of four modes using different spatial weight specifications. The spatial lags of both dependent and independent variables are highly statistically significant in the SDM, which indicates that both spatial dependence and spatial autocorrelation are captured. After controlling for spatial autocorrelation, employment is found to be the most important factor for regional output. The output elasticity of public transportation infrastructure capital stock ranges between 0.11% and 0.22%.

Table 8.

Estimation Results From Spatial Fixed Panel Models (TTSPC Model).

	Model 1	Model 2	Model 3	Model 4
Spatial weight	Inverse distance	1-nearest neighbor	3-nearest neighbors	5-nearest neighbors
Direct effect
emp	0.437*** (15.754)	0.486*** (14.375)	0.398*** (11.592)	0.339*** (9.453)
pfapc	0.268*** (4.194)	0.268*** (13.454)	0.354*** (18.816)	0.343*** (18.093)
ttspc	0.052*** (4.168)	0.066*** (6.190)	0.059*** (5.583)	0.031*** (2.984)
Indirect effect
emp	0.570*** (15.754)	0.422*** (10.636)	0.705*** (11.809)	0.893*** (11.494)
pfapc	−0.140*** (4.194)	−0.127*** (−5.812)	−0.407*** (−11.468)	−0.375*** (−9.641)
ttspc	0.057*** (4.168)	0.083*** (6.469)	0.157*** (7.502)	0.098*** (3.362)
Total effect
emp	1.007*** (15.754)	0.912*** (17.715)	1.102*** (16.546)	1.234*** (14.676)
pfapc	0.128*** (4.194)	0.140*** (4.957)	−0.053 (−1.387)	−0.033 (−0.793)
ttspc	0.109*** (4.168)	0.149*** (7.578)	0.216*** (8.138)	0.129*** (3.844)
Spatial variables
W*dep.var.	0.325*** (8.573)	0.251*** (8.223)	0.358*** (8.965)	0.389*** (8.126)
W*emp	0.299*** (6.055)	0.235*** (6.070)	0.358*** (6.430)	0.458*** (6.286)
W*pfapc	−0.195*** (−8.890)	−0.174*** (−8.453)	−0.417*** (−15.231)	−0.381*** (13.725)
W*ttspc	0.027* (1.872)	0.053*** (4.728)	0.090*** (6.161)	0.053*** (2.833)
ML	1367.884	1352.509	1404.307	1403.687
Sigma²	0.0007	0.0007	0.0006	0.0006
R²	0.985	0.985	0.987	0.987
Corr²	0.937	0.932	0.948	0.948

Note. Numbers in parentheses are t-statistics. *, **, and *** denote significant level at 10%, 5%, and 1%, respectively. All variables were transformed into logarithmic terms. TTSPC = total transportation stock per capita.

In the four modes model, the total transportation variable is constituted as highways, public rail, transit, and airport variables. The estimation results are displayed in Table 9. Most of the spatial components are statistically significant. After controlling for the effects of employment and private capital, highway infrastructure has a positive effect on the regional economic output. The elasticity ranges between 0.05 and 0.1, which can be interpreted as a 1% increase in highway infrastructure per capita is associated with a 0.05% to 0.1% increase in GMP per capita, ceteris paribus. Public rail is also significant, although with much smaller impacts than highways.

Table 9.

Estimation Results From Fixed Panel Model and Spatial Fixed Panel Models (Four Modes Model).

	Model 1	Model 2	Model 3	Model 4
Spatial weight	Inverse distance	1-nearest neighbor	3-nearest neighbors	5-nearest neighbors
Direct effect
emp	0.410*** (15.754)	0.461*** (17.234)	0.377*** (12.454)	0.340*** (9.944)
pfapc	0.273*** (4.194)	0.266*** (13.755)	0.322*** (18.079)	0.341*** (18.424)
hwyspc	0.036*** (4.168)	0.053*** (5.207)	0.025** (2.240)	0.025* (1.926)
amspc	0.004* (1.865)	0.006*** (2.482)	0.003 (1.326)	0.001 (0.264)
traspc	0.002 (−0.542)	−0.001 (−0.254)	0.005 (1.232)	0.001 (−0.276)
airspc	0.006*** (3.816)	0.007*** (4.378)	0.005*** (3.090)	0.005*** (3.355)
Indirect effect
emp	0.546*** (9.961)	0.429*** (10.911)	0.708*** (12.246)	0.778*** (9.932)
pfapc	−0.126*** (−5.096)	−0.100*** (−4.581)	−0.374*** (−10.401)	−0.338*** (−8.605)
hwyspc	0.016*** (0.688)	0.047*** (3.345)	0.044 (1.550)	0.068 (1.274)
amspc	0.018*** (3.503)	0.013*** (4.300)	0.030*** (4.940)	0.026*** (2.958)
traspc	0.013* (1.694)	0.010*** (2.060)	0.055*** (5.737)	0.012 (0.832)
airspc	0.001 (0.165)	0.007** (2.720)	−0.017** (−2.609)	−0.010 (−1.391)
Total effect
emp	0.955*** (14.361)	0.891*** (17.234)	1.086*** (16.763)	1.118*** (13.695)
pfapc	0.147*** (4.194)	0.166*** (6.070)	−0.052 (−1.392)	0.003 (0.080)
hwyspc	0.052*** (4.168)	0.100*** (5.014)	0.069* (1.890)	0.094 (1.469)
amspc	0.023*** (3.370)	0.019*** (4.160)	0.033*** (4.443)	0.027*** (2.582)
traspc	0.011 (1.059)	0.009 (1.200)	0.060*** (5.129)	0.011 (0.630)
airspc	0.007 (1.441)	0.015*** (4.041)	−0.012 (−1.621)	0.004 (−0.541)
Spatial variables
W*dep.var.	0.328*** (8.707)	0.246*** (8.045)	0.357*** (9.362)	0.395*** (8.342)
W*emp	0.291*** (6.039)	0.253*** (6.757)	0.372*** (6.884)	0.379*** (5.274)
W*pfapc	−0.186*** (−8.399)	−0.150*** (−7.213)	−0.383*** (−14.244)	−0.357*** (−12.528)
W*hwyspc	−0.001 (−0.026)	0.027** (2.331)	0.023 (1.310)	0.034 (1.04)
W*amspc	0.012*** (3.428)	0.010*** (3.891)	0.020*** (4.981)	0.018*** (3.140)
W*traspc	0.011** (1.973)	0.009** (2.215)	0.038*** (5.959)	0.008 (0.909)
W*airspc	−0.002 (0.565)	0.005** (2.028)	−0.014*** (9.362)	−0.008** (−2.001)
ML	1388.719	1376.741	1454.386	1424.986
Sigma²	0.0006	0.0006	0.0005	0.0005
R²	0.986	0.986	0.989	0.987
Corr²	0.942	0.938	0.954	0.951

Note. Numbers in parentheses are t-statistics. *, **, and *** denote significant level at 10%, 5%, and 1%, respectively. All variables were transformed into logarithmic terms.

In the SDM with spatial fixed effect, transit and airport capital stock per capita are found insignificant when an inverse distance and 5 nearest neighbors spatial weights are adopted. Airport capital is significant when 1 nearest neighbor is introduced and transit capital is significant when 3 nearest neighbors spatial weight is adopted. A possible explanation for such differences may be because of the different levels of spatial autocorrelation being captured in each specification. Spatial dependence of each transportation mode varies substantially because of their different service area and network characteristics. This sensitivity issue that is found in our study further confirms the finding of Kelejian and Robinson (1997), who point out that the estimation of regional infrastructure productivity involving spatial spillover is very sensitive to model specifications.

Spillover effects of transportation infrastructure were estimated using various spatial weight specifications. Both a general form and a modal comparative form are estimated, respectively. Table 8 illustrates different regional effects of the general transportation infrastructure. The result suggests that its direct, indirect, and total effects are all significant. In terms of the magnitude, the direct effects range between 0.031 and 0.066, whereas the indirect effects range between 0.057 and 0.157 depending on the specification of spatial weights. The results indicate that public transportation infrastructure in the northeast megaregion, in general, has both a positive local effect and a positive spillover effect. Most of these effects are achieved through spillover effects on neighboring economies.

The results suggest spillover effects of employment and private capital stock also exist. The total elasticity effects of labor range between 0.912 and 1.234, indicating labor plays a pivotal role in stimulating regional economic growth in the northeast region. On the other hand, private capital was found to have a positive and significant contribution to output. The direct elasticity effect is around 0.27 to 0.35, meaning that a 1% increase of private capital stock per capita is associated with a 0.27% to 0.35% increase in regional output. On the contrary, the indirect effect of private capital is −0.127 to −0.407, meaning a 1% increase of private capital stock per capita in a metropolitan area is associated with a 0.127% to 0.407% decrease of regional output in its neighboring metro areas. This suggests the competitive nature of private capital investment in this regional economy.

Table 9 also displays the spillover effects of transportation infrastructure by mode. In terms of a modal spillover effect comparison, the highway variable has both positive direct and indirect effects. This implies after controlling for labor, private capital, and public capital of rail, transit, and airport, highway infrastructure has a positive impact to both the local economy and its neighboring economies.

A positive effect of public rail capital stock is found to exist in both the local effect and the spillover effect. A higher magnitude of the spillover effect indicates that intercity passenger rail plays a role in facilitating interregional passenger flow, which may possibly result in regional economic growth through labor mobility and knowledge spillover.

Public transit has a relatively small effect on regional output compared to highway and public rail. Although significant direct effects are found across estimations using various spatial weights, significant indirect effects are found when lower order of spatial weights (smaller than 5) are used. The spillover effects range between 0.01 and 0.055, which can be interpreted as a 1% change in transit stock per capita is associated with a 0.01% to 0.055% change in its neighboring economy.

Public airport capital stock also has different influences on regional output through both direct and indirect effects. Generally, the local effect of public airport capital ranges between 0.005 and 0.007, suggesting that a 1% increase in public airport capital stock per capita is associated with a 0.005% to 0.007% increase of output in the local economy. The spillover effect of airport declines when higher order spatial weights are adopted.

Conclusion

This study improves the understanding of the linkages between regional output and transportation infrastructure within the context of the U.S. northeast megaregion. Unlike traditional studies, estimations are improved mainly through two considerations: (a) adopting financial data measured in monetary terms with a focus on the period between 1991 and 2009 and (b) introducing spatial analysis with a cautious approach to model specification. The regional impact of different transportation modes can be summarized in Figure 2.

Figure 2.

Regional impact of transportation infrastructure.

The result confirms that transportation infrastructure in the northeast megaregion during this period has a positive impact on regional economic output, most of which is achieved through regional spillover effects. In terms of modal comparison, highway infrastructure dominates with a larger impact than other transportation modes. Public railway infrastructure has the second largest impact, whereas public airport and transit impacts rank third and fourth.

Our findings reject Boarnet’s hypothesis that public capital has a negative spillover effect on regional growth. Instead, positive spillover effects are observed both for total transportation capital stock and for highway capital in particular after controlling for the spillover effects of labor and private capital. The finding is robust across different weight specifications. In addition, positive spillover effects are found to exist in other transportation infrastructure modes including public rail and transit. The spillover effects of public airport vary when different spatial weights are specified. The findings also differ from Holtz-Eakin and Schwartz (1995), who found no evidence of positive output spillover across states for the case of highway capital. The results confirm that transportation infrastructure has both positive local effects and positive output spillovers.

The research findings provide two implications. First, transportation infrastructure in the northeast megaregion has a significant impact on regional economic growth, at least in the period between 1991 and 2009. Because of the nature and maturity of the transportation systems, the major benefits are likely achieved through the network effects of transportation systems, which often stimulate regional growth through spillovers instead of local effects.

Second, our multimodal investigation reveals that the total effect of each transportation infrastructure investment varies substantially given the differences of capital stock per capita by mode. It is clear that in the northeast megaregion, highways have the most dominant impact. Passenger rail and airports function as complements to other modes and also have significant impacts to regional output. Although transit receives a comparatively high level of public investment, it has a relativly small influence on regional output.

It should be noted that the findings in the studies are subject to the specific data and methodology being used. Thus, the conclusion and implication should not be generalized and may vary if any specification is altered. It should also be noted that the assessment results of the public transportation infrastructure may vary given the existence of the potential data issue, specific panel effect being considered, and the characteristics of spatial econometrics. These issues, although important caveats of the study, should be addressed in future research.

Another future research direction is to develop a better approach to evaluate the impact of railway systems. Given the nature of mixed infrastructure usages for both intercity rail and commuter rail services in the northeast megaregion, one possible direction is to measure the effects of the passenger rail infrastructure as a whole. Another approach to evaluate the impact of public railway infrastructure is to develop an analytic framework under general equilibrium. In that case, the micro impact of transportation infrastructure on the change of social welfare can be assessed. A strategy for doing this is presented in Chen and Haynes (2014).

Footnotes

Appendix A

Regions in the Analysis.

CBSA	Name	Type	ID
10900	Allentown–Bethlehem–Easton, PA–NJ	MSA	17
12100	Atlantic City, NJ	MSA	26
12580	Baltimore–Towson, MD	MSA	31
12700	Barnstable Town, MA	MSA	9
14460	Boston–Cambridge–Quincy, MA-NH	MSA	5
14860	Bridgeport–Stamford–Norwalk, CT	MSA	14
18180	Concord, NH	MicroSA	2
20100	Dover, DE	MSA	28
25420	Harrisburg–Carlisle, PA	MSA	20
25540	Hartford–West Hartford–East Hartford, CT	MSA	10
28740	Kingston, NY	MSA	7
29540	Lancaster, PA	MSA	24
30140	Lebanon, PA	MSA	21
30340	Lewiston–Auburn, ME	MSA	0
31700	Manchester–Nashua, NH	MSA	3
35300	New Heaven–Milford, CT	MSA	15
35620	New York–Northern New Jersey–Long Island, NY–NJ–PA	MSA	18
35980	Norwich–New London, CT	MSA	13
36140	Ocean City, NJ	MSA	29
37980	Philadelphia–Camden–Wilmington, PA–NJ–DE–MD	MSA	22
38860	Portland–South Portland–Biddeford, ME	MSA	1
39100	Poughkeepsie–Newburgh–Middletown, NY	MSA	8
39300	Providence–New Bedford–Fall River, RI-MA	MSA	16
39740	Reading, PA	MSA	19
42540	Scranton–Wikes–Barre, PA	MSA	12
44140	Springfield, MA	MSA	4
45940	Trenton–Ewing, NJ	MSA	23
47220	Vineland–Millville–Bridgeton, NJ	MSA	27
47900	Washington–Arlington–Alexandria, DC–VA–MD–WV	MSA	30
48740	Willimantic, CT	MicroSA	11
49340	Worcester, MA	MSA	6
49620	York-Hanover, PA	MSA	25

Note. CBSA = core-based statistical area; MSA = metropolitan statistical area; MicroSA = micropoltian statistical area.

Source: Bureau of Economic Analysis.

Appendix B

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Zhenhua Chen is a postdoctoral research associate in the Price School of Public Policy at the University of Southern California. His research interests include economic geography, regional science, transportation planning and policy, and public finance.

Kingsley E. Haynes is a professor in the School of Policy, Government, and International Affairs at George Mason University. He is a national scholar in regional economic development and infrastructure.

References

Anderson

A. E.

Anderstig

Harsman

(1990). Knowledge and communications infrastructure and regional economic change. Regional Science and Urban Economics, 20, 359-376.

Anselin

(1988). Spatial econometrics: Methods and models. New York, NY: Springer.

Aschauer

D. A.

(1989). Is public expenditure productive? Journal of Monetary Economics, 23, 177-200.

Aschauer

D. A.

(1990). Why is infrastructure important? In Proceedings of the Federal Reserve Bank of Boston Conference on “The Third Deficit: The Shortfall in Public Capital Investment” (Vol. 34, pp. 21-68). Boston, MA: Federal Reserve Bank of Boston.

Aschauer

D. A.

(1994). Infrastructure and macroeconomic performance: Direct and indirect effects. Paper presented at the OECD Conference on Capital Formation and Employment, Amsterdam, Netherlands.

Berndt

E. R.

Hansson

(1992). Measuring the contribution of public infrastructure capital in Sweden. Scandinavian Journal of Economics, 94(Suppl.), S151-S168.

Bhatta

S. D.

Drennan

M. P.

(2003). The economic benefits of public investment in transportation. Journal of Planning Education and Research, 22, 288-296.

Blum

(1982). Effects of transportation investiments on regional growth: A theoretical and empirical investigation. Papers of the Regional Science Association, 49, 169-184.

C. D.

Florio

Manzi

(2010). Regional infrastructure and convergence: Growth implications in a spatial framework. Transition Studies Review, 17, 475-493.

10.

Boarnet

M G.

(1997). Infrastructure services and the productivity of public capital: The case of streets and highways. National Tax Journal, 50, 39-57.

11.

Boarnet

M. G.

(1998). Spillovers and the locational effects of public infrastructure. Journal of Regional Science, 38, 381-400.

12.

Boarnet

M. G.

Haughwout

A. F.

(2000). Do highways matter? Evidence and policy implications of highways’ influence on metropolitan development (Discussion Paper). Washington, DC: Brookings Institution Center on Urban and Metropolitan Policy.

13.

Button

(1998). Infrastructure investment, endogenous growth and economic convergence. Annals of Regional Science, 32, 145-162.

14.

Cantos

Gumbau-Albert

Maudos

(2005). Transport infrastructures, spillover effects and regional growth: evidence of the Spanish case. Transport Reviews, 25, 25-50.

15.

Chen

Haynes

(2013). Public surface transportation and regional output: A spatial panel approach. Papers in Regional Science. Advance online publication. doi:10.1111/pirs.12092

16.

Chen

Haynes

(2014). Transportation capital in the US: A multimodal general equilibrium analysis. Public Works Management and Policy, 19, 97-117.

17.

Cliff

A. D.

Ord

J. K.

(1981). Spatial processes: Models and applications. London, England: Pion.

18.

Cohen

J. P.

(2007). Economic benefits of investments in transport infrastructure (Discussion Paper No. 2007-13). Paris, France: OECD/ITF.

19.

Cohen

J. P.

Morrison Paul

C. J.

(2003). Spatial and supply/demand agglomeration economies: State- and industry-linkages in the U.S. food system. Empirical Economics, 28, 733-751.

20.

Cohen

J. P.

Morrison Paul

C. J.

(2004). Public infrastructure investment, interstate spatial spillovers, and manufacturing costs. Review of Economics and Statistics, 86, 551-560.

21.

Cohen

J. P.

Morrison Paul

C. J.

(2007). The impacts of transportation infrastructure on property values: A higher-order spatial econometrics approach. Journal of Regional Science, 47, 457-478.

22.

Duffy-Deno

K. T.

Eberts

(1991). Public infrastructure and regional economic development: A simultaneous equations approach. Journal of Urban Economics, 30, 329-343.

23.

Elhorst

J. P.

(2014). Matlab software for spatial panels. International Regional Science Review, 37, 389-405.

24.

Federal Railroad Administration. (1998). Privatization of intercity rail passenger service in the United States. Washington, DC: Author.

25.

Fernald

J. G.

(1999). Roads to prosperity? Assessing the link between public capital and productivity. American Economic Review, 89, 619-638.

26.

Fingleton

López-Bazo

(2006). Empirical growth models with spatial effects. Papers in Regional Science, 85, 177-198.

27.

Fujita

Krugman

P. R.

Venables

A. J.

(1999). The spatial economy: Cities, regions, and international trade. Cambridge: MIT Press.

28.

Garcia-Mila

McGuire

T. J.

Porter

R. H.

(1996). The effect of public capital in state-level production functions reconsidered. Review of Economics and Statistics, 78, 177-180.

29.

Garofalo

G. A.

Yamarik

(2002). Regional convergence: Evidence from a new state-by-state capital stock series. Review of Economics and Statistics, 84, 316-323.

30.

Gottmann

(1961). Megalopolis: The urbanized northeastern seaboard of the United States. New York, NY: Twentieth Century Fund.

31.

Gramlich

E. M.

(1994). Infrastructure investment: A review essay. Journal of Economic Literature, 32, 1176-1196.

32.

Gramlich

E. M.

(2001). Infrastructure and economic development. Remarks by Governor Edward M. Gramlich at the Texas Trade Corridors New Economy Conference, San Antonio, TX.

33.

Granger

C. W. J.

(1969). Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 37, 424-438.

34.

Harmatuck

D. J.

(1996). The influence of transportation infrastructure on economic development. Logistics and Transportation Review, 32, 63-76.

35.

Holtz-Eakin

(1993). State-specific estimates of state and local government capital. Regional Science and Urban Economics, 23, 185-209.

36.

Holtz-Eakin

Lovely

M. E.

(1996). Scale economies, returns to variety, and the productivity of public infrastructure. Regional Science and Urban Economics, 26, 105-123.

37.

Holtz-Eakin

Schwartz

A. E.

(1995). Spatial productivity spillovers from public infrastructure: Evidence from state highways (Working Paper No. 5004). Cambridge, MA: National Bureau of Economic Research.

38.

Hulten

C. R.

Wykoff

F. C.

(1981). The estimation of economic depreciation using vintage asset prices: An application of the Box-Cox power transformation. Journal of Econometrics, 15, 367-396.

39.

Jiwattanakulpaisarn

Noland

R. B.

Graham

D. J.

Polak

J. W.

(2009). Highway infrastructure and state-level employment: A causal spatial analysis. Papers in Regional Science, 88, 133-159.

40.

Jorgenson

D. W.

(1991). Fragile statistical foundations: The macroeconomics of public infrastructure investment. Washington, DC: American Enterprise Institute for Public Policy Research.

41.

Kahn

Wiener

A. J.

(1967). The year 2000: A framework for speculation on the next thirty-three years. New York, NY: Macmillan.

42.

Katz

A. J.

Herman

S. W.

(1997). Improved estimates of fixed reproducible tangible wealth. Survey of Current Business, 77(5), 69-92.

43.

Keeler

T. E.

Ying

J. S.

(2008). Measuring the benefits of a large public investment: The case of the U.S. Federal-Aid Highway System. Journal of Public Economics, 36, 69-85.

44.

Kelejian

H. H.

Robinson

D. P.

(1997). Infrastructure productivity estimation and its underlying econometric specifications: A sensitivity analysis. Papers in Regional Science, 76, 115-131.

45.

Krugman

(1991). Increasing returns and economic geography. Journal of Political Economy, 99, 484-499.

46.

Lau

S.-H. P.

Sin

C.-Y.

(1997). Public infrastructure and economic growth: Time-series properties and evidence. Economic Record, 73(221), 125-135.

47.

LeSage

J. P.

(1999). The theory and practice of spatial econometrics. A manual to accompany the Spatial Econometrics Toolbox. Retrieved from http://www.spatial-econometrics.com

48.

LeSage

J. P.

Pace

R. K.

(2009). Introduction to spatial econometrics (1st ed.). Boca Raton, FL: Chapman & Hall/CRC Press.

49.

Levin

Lin

C. F.

Chu

C. S.

(2002). Unit root tests in panel data: asymptotic and finite-sample properties. Journal of Econometrics, 108, 1-24.

50.

Lewandowski

(2007). PESCADF: Stata module to perform Pesaran’s CADF panel unit root test in presence of cross section dependence (Statistical Software Component No. S456732). Boston College Department of Economics.

51.

Mattoon

(2002). Midwest infrastructure: Assessing the contribution of basic infrastructure to economic growth. Chicago Fed Letter, 184b (Special Issue December).

52.

Mohammad

(2009). Infrastructure investment in Indonesia: process and impact (Unpublished doctoral dissertation). University of Groningen, Netherlands.

53.

Moran

P A P.

(1950). Notes on continuous stochastic phenomena. Biometrika, 37, 17-23.

54.

Moreno

López-Bazo

(2007). Returns to local and transport infrastructure under regional spillovers. International Regional Science Review, 30, 47-71.

55.

Munnell

A. H.

(1992). Policy watch: Infrastructure investment and economic growth. Journal of Economic Perspectives, 6, 189-198.

56.

Munnell

A. H.

Cook

L. M.

(1990). How does public infrastructure affect regional economic performance? New England Economic Review, 86, 1095-1111.

57.

Nadiri

M. I.

Mamuneas

T. P.

(1996). Contribution of highway capital to industry and national productivity growth (No. BAT-94-008). Report Prepared for Apogee Research, Inc., for the Federal Highway Administration Office of Policy Development.

58.

Ozbay

Ozmen-Ertekin

Berechman

(2007). Contribution of transportation investments to county output.Transport Policy, 14, 317-329.

59.

Paelinck

J. H. P.

Klaassen

L. H.

(1979). Spatial econometrics. Farnborough, England: Saxon House.

60.

Pesaran

M. H.

(2007). A simple panel unit root test in the presence of cross-section dependence. Journal of Applied Econometrics, 22, 265-312.

61.

Pinnoi

(1994). Public infrastructure and private production measuring relative contributions. Journal of Economic Behavior & Organization, 23, 127-148.

62.

U.S. Government Accountability Office. (2004). Intercity passenger rail: Amtrak’s management of Northeast Corridor improvements demonstrates need for applying best practices (No. GAO-04-94). Washington, DC: Author.