spillover effects of telecommunications infrastructure

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002, pp. 339–360

GEOGRAPHIC AND NETWORK NEIGHBORS: SPILLOVER EFFECTS OF TELECOMMUNICATIONS INFRASTRUCTURE* Serdar Yilmaz The World Bank, 1818 H Street NW, Washington, DC 20433, U.S.A. E-mail: [email protected]

Kingley E. Haynes School of Public Policy, George Mason University, Fairfax, VA 22030, U.S.A. E-mail: [email protected]

Mustafa Dinc The World Bank, 1818 H Street NW, Washington, DC 20433, U.S.A. E-mail: [email protected]

ABSTRACT. This paper tests for spatial spillover effects of state-level telecommunications infrastructure investment on state output, using panel data for 48 U.S. states from 1970 through 1997. As information and communication technologies support more industrial locational freedom, states may use telecommunications infrastructure investment as a competitive tool for attracting factors of production. In a production-function framework, this effect would manifest itself as a negative output spillover effect from telecommunications infrastructure investment. Findings indicate that a state benefits from its own telecommunications infrastructure, but telecommunications investment by other states has a negative impact on its output growth path, and proximity amplifies this negative spillover effect.

1.

INTRODUCTION

In the last two decades, telecommunications infrastructure emerged as an important factor in interregional economic activities. Advancements in information technology have diminished the importance of geographic proximity and created new “network neighborhoods.” As all nonneighbor localities become more accessible and in some sense closer to each other, firms can now establish and maintain contacts with suppliers and customers over greater distances and in remote locations. If information and communication technologies increase

*We thank Yesim Yilmaz, D. Holtz-Eakin, and three anonymous referees for their constructive comments and valuable suggestions. We express our appreciation for the support of NSF/EPA Grant # SES–9976483 “Social Vulnerability Analysis.” The findings, interpretations, and conclusions are entirely our own, and do not necessarily represent the views of the World Bank, its executive directors, or the countries they represent. Received March 2000; revised October 2000 and March 2001; accepted July 2001. © Blackwell Publishing, Inc. 2002. Blackwell Publishing, Inc., 350 Main Street, Malden, MA 02148, USA and 108 Cowley Road, Oxford, OX4 1JF, UK.

339

340

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

accessibility and support more locational freedom, the search by firms for the least-cost location is likely to reshape regional development patterns, which will lead to higher rates of growth for better-endowed regions. Therefore, such advanced technologies may alter location decisions of firms, increasing investments and output in some states, while causing disinvestments and possible job losses for other states. In the U.S., although telecommunications networks are privately owned, states have a major impact on the level of telecommunications investment through their regulatory power.1 State regulatory agencies set the return rate for telecommunications investments in pursuit of various policy goals. In this rate-setting process, some states emphasize equitable access to telecommunications networks whereas others emphasize cost effectiveness or increased capacity. For example, after the breakup of AT&T, in order to preserve universal access, the New Jersey state utility commission did not allow intrastate competition in the telecommunications market and tried to keep local access rates relatively low.2 In some other states, governors have formed task forces to make recommendations for building state-of-the-art networks (for example Michigan, Wisconsin, Washington, Kansa, Idaho, and Nebraska) or policymakers have enacted master plans (Tennessee, Florida, and Texas). For example, in order to ensure investments in telecommunications infrastructure Tennessee adopted a master plan in 1993 to accelerate telecommunications technology deployment throughout the state, and the availability of new services, such as ISDN, has significantly increased since then. Florida and Texas followed Tennessee in enacting similar legislation. All these policies had an immediate impact on the growth of telecommunications infrastructure and access to telecommunications services by residential and commercial customers. Given that states have such a policy tool that directly affects the availability of telecommunications infrastructure, it is reasonable to expect that state policymakers may use this tool to attract factors of production from other regions. States may use their regulatory power over telecommunications infrastructure to induce new businesses within the state in order to generate new employment and output growth. However, when a firm in search of better telecommunications infrastructure moves to a better-endowed state, the increase in this state’s output will, at least initially, come at the expense of its former location. In a production-function framework, this effect would manifest itself as a negative output spillover effect from telecommunications infrastructure investment. 1

Federal regulations by FCC generally deal with equity issues such as universal service. It is the state regulatory commissions’ responsibility to ensure reasonable price levels and to determine levels of return to telecommunications investment (National Governors’ Association, 1994). For a more detailed discussion see Dinc et al. (1998). 2 To this end, New Jersey regulators approved only 26 percent of the local exchange operator’s rate increase requests. In response to criticisms of the New Jersey Board of Public Utilities’ lack of professionalism and economic expertise the state legislature enacted a bill that specified investment goals in telecommunications infrastructure (Teske, 1990). © Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

341

This paper tests for spatial spillover effects of state-level telecommunications infrastructure investment on state output. Using a pooled data set of output, labor, public-sector capital, private-sector capital, and telecommunications capital for 48 contiguous U.S. states between 1970 and 1997, we find that a state benefits from its own telecommunications infrastructure investment, but telecommunications investment in other states has a negative effect on its output growth. Geographic proximity amplifies this negative spillover effect. The results suggest that the “network neighborhood” has become an important factor in interregional economic activities, but geographic proximity still plays a key role in regions competing with other regions or states in attracting factors of production. This study differs from the earlier literature on two fronts: To the best of our knowledge, it is the first attempt to investigate the role of telecommunications infrastructure investment from a regional growth standpoint. Although the role of infrastructure on economic performance and productivity has been studied extensively in the regional development literature, studies on the spatial distribution of infrastructure systems have focused almost exclusively on publicly owned infrastructure, and ignored telecommunications as a viable competitive force (Crandall, 1997).3, 4 In fact, telecommunications networks exhibit all the characteristics of an infrastructure network such as immobility, large initial investment, and presence of external economies (Lakshmanan, 1989; Youngson, 1967). On the other hand, the growth literature focusing on telecommunications either concentrates on cross-country analysis or examines the impact of telecommunications infrastructure on national productivity and growth.5, 6 These studies provide little insight about the use of telecommunications investment as a regional development tool. The telecommunications industry is heavily regulated in order to ensure socially desirable outcomes, so our analysis of the impact of telecommunication infrastructure on output is relevant in guiding policymakers. Second, this paper offers an alternative view of the impact of telecommunications on interregional economic patterns. The popular view is that new developments in telecommunications services have reduced the importance of geographic proximity in regional development because firms can now compete for customers and raw materials across greater distances. However, recent studies show that availability of a skilled labor force, level of state and local taxes, regional business climate, quality of life, and availability and accessibility of basic infrastructure have become very important factors in business location 3

See Gramlich (1994) and Mikelbank and Jackson (2000) for a detailed survey of the literature. Aschauer (1989a, 1989b, 1989c), Boarnet (1997a, 1997b), Eisner (1991), Evans and Karras (1994), Holtz-Eakin (1994), Hulten and Schwab (1997), Munnell (1990, 1992). 5 Leff (1984), Norton (1992), Staranczak et al. (1994), Baer (1995), Antonelli (1993), Madden and Savage (1998). 6 Cronin et al. (1991), Cronin et al. (1993a), Cronin et al. (1993b), Cronin, Gold, Mace, Sigalos (1994), Dholakia and Harlam (1994), Cronin et al. (1995), Nadiri and Nandi (1997), Cronin, Colleran, and Gold (1997), Resende (1999). 4

© Blackwell Publishing, Inc. 2002.

342

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

decisions (Blair and Premus, 1993). Therefore, availability of labor and capital are important factors that affect location decisions of firms, and states will compete to attract these factors of production. Geographic proximity may indeed amplify the impact of state-level policies designed to increase the attractiveness of the state for business location and relocation. We test this proposition in terms of telecommunications infrastructure investment by investigating the impact of geographic and network distances on regional output growth. This paper draws heavily on the tools and techniques used in recent work on spatial econometrics by Cliff and Ord (1972), Anselin and Florax (1995), Anselin (1999), and Rey and Montouri (1999), as well as the work on the spatial distribution of infrastructure systems in the regional science literature (HoltzEakin and Schwartz, 1995; Kelejian and Robinson, 1997; Boarnet, 1998). The existing literature on the spatial distribution of public infrastructure is not conclusive, and offers contradictory results on both the existence and the sign of the spillover effects. For example, Boarnet (1998) examines the output spillover effects of public streets and highways in California counties from 1969 to 1988 and reports a negative spillover effect. On the other hand, Holtz-Eakin and Schwartz (1995) study state highways and their impact on the productivity benefits beyond the state borders, and find no evidence of quantitatively important productivity spillovers. We expect that this study will make a contribution to this ongoing debate. The organization of the paper is as follows: Section 2 describes the conceptual framework and specifies the econometric model; Section 3 discusses data and presents the results. Section 4 presents test results for robustness of the empirical findings and Section 5 draws conclusions. 2.

THE MODEL SPECIFICATION

Conceptual Framework This section outlines a classical production model, where telecommunications capital stock and public infrastructure stock are quasi-fixed inputs for gross state production. In our model we consider states as the unit of analysis because the state regulatory agencies set the return rate for telecommunications investment within their geographic boundaries.7 Following Boarnet (1998), the model treats a state as a single entity whose output depends on public infrastructure, telecommunications infrastructure, capital, and labor.8 The output in each state is produced according to

7 We recognize that the effect of telecommunications investment may vary across a state. However, we are primarily interested in the impact of an individual state’s telecommunications policy on the other states. Because telecommunications policy is implemented uniformly within a state, we believe it is appropriate to treat states as the unit of analysis. 8 The discussions about model specification and expected outcome of migration of mobile factors are drawn heavily from Boarnet (1998) who developed it in an intrastate intercounty context; readers interested in the theoretical basis of the model are encouraged to read Boarnet (1998).

© Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

343

Q = α(G)β(TK)f(K, L)

(1)

where Q is output, G is public capital stock, TK is telecommunications capital stock, L is labor input, and K is private capital stock and meet the following conditions α′(G) > 0 β′(TK) > 0 fK > 0; fKK < 0 and fL > 0; fu < 0 In the model, both public infrastructure and telecommunications infrastructure are assumed to be complementary to labor and capital. Because the purpose of this paper is to test spillover effects of telecommunications infrastructure and the role of geographic distance in the magnitude of the spillover, we particularly focus on the possible outcomes of an increase in telecommunications capital. If the markets are competitive and the factors of production are mobile, then each input is paid its marginal revenue product, which depends on G and TK

∂Q = α G β TK f L L, K ∂L

g

∂Q = α G β TK f K L, K ∂K

g

b gb g b

(2) and

b gb g b

The factor prices in each state i then become wi = pα(Gi)β(TK)fL(LiKi) and (3)

ri = pα(Gi)β(TK)fK(LiKi)

where p, w, and r are the prices of output, labor, and capital, respectively. As state i increases its investment in the telecommunications stock, from Equation (3) the price of labor and capital increase accordingly. With fully mobile labor and capital, one would expect to see factors of production move from other states to state i in the short run. Therefore, after factor migration, labor, and capital increase in state i and decrease in other states, and the output in state i would be (4)

Qi = α(Gi)β(TKi + ∆TK) f(Li + ∆L, Ki + ∆K)

© Blackwell Publishing, Inc. 2002.

344

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

If the increase in capital and labor is proportional, and the original factor input ratio can be preserved, state i shifts to a higher level of production relative to other states and the economy returns to the long-run equilibrium with w1 = w2.... = wi, and r1 = r2... = ri. Therefore, ceteris paribus, migration of factors would result in output increases in states with well-developed telecommunications networks and output losses for others. In cases where the increase in capital or labor is not proportional, then the marginal revenue products of labor or capital may either increase or decrease relative to the initial level in state i, and the final capital-labor ratio would depend on the marginal rate of technical substitution in that particular state. However, there would still be in-migration of factors of production from other states, which is sufficient for our spillover analysis. A plausible case can be made even if imperfect mobility of factors of production and possibility of substitution of inputs are taken into consideration. Following the literature, suppose capital can move with lower cost than labor (Greenwood and Hunt, 1986). With telecommunications infrastructure as a complementary input in state i, an increase in telecommunications capital would result in an increase in the marginal revenue products of both labor and capital. If labor supply is fixed, compared to the case with mobile factors of production, more capital is expected to move into state i. The final increase in output in state i would depend on the elasticity of substitution between capital and labor; However, the sign of the spillover effect would be preserved even in cases when factors of production are fairly immobile. Interested readers can find more about equalization of returns and factor substitution in Borts (1960), Arrow et al. (1961), Kendrick and Sato (1963), Clague (1969), and Blackorby and Russell (1989). Finally, if geographic proximity increases the mobility of labor, then the in-migration of capital would be accompanied by in-migration of labor from neighboring states. Therefore, the output would increase in state i at the expense of neighboring states. In other words, geographic proximity is expected to amplify the spillover effects of telecommunications investment. The Model Boarnet (1998) argues that regions with the best infrastructure stock would bid mobile factors of production away from other regions. Even if capital or labor migration does not immediately draw production away from other locations, competitive advantage from previous infrastructure investments would change the path of future investments in a region’s economy, boost its output, and increase its competitiveness. In addition, when efficiently supplied, infrastructure services have an inherent role in improving access to markets, reducing unit costs of production and generating consumer surplus by reducing the cost of consumption. Therefore, availability and quality of infrastructure services influence the economic performance of regions. Similarly, if telecommunications infrastructure investment in a state enhances factors of production, the total output in that state would depend positively on its stock of telecommunications © Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

345

capital, and negatively on the stock of telecommunications in other states. Thus, we can rewrite the production function for the state after including the spatial spillover as (5)

Q = f (L, G, K, TK, OTH)

where Q is state output, L is labor, G is public sector capital stock, K is private sector capital stock (excluding telecommunications), TK is telecommunications capital stock in the state, and OTH is telecommunications capital stock in all other states. In Equation (5), OTH is the spillover variable, and it captures the impact of the network neighborhood on state output. The paper hypothesizes that interconnectivity of the systems diminishes the role of physical borders and geographic distance, so we consider all states to be adjacent to each other in terms of access to communications. In our representation, the network neighborhood is defined as the weighted sum of telecommunications capital in all other states. In order to test the importance of geographic proximity, we allow for a traditional representation of geographic distance through the geographic neighbor variable. The geographic neighbor variable FOTH represents adjacency with physical borders. For each state, the geographic neighborhood is defined as the weighted sum of telecommunications capital in all adjacent states. In this case, the production function given in Equation (5) changes to (6)

Q = f (L, G, K, TK, FOTH)

Both neighbor variables are calculated by using the formulas OTHit = Wt TKjt = All Neighbors (network neighbors) FOTHit = Wt TKjt = Adjacent first-order neighbors (geographic neighbors) where W is the weight matrix with elements wi,j for each year, i indexes the state under investigation and j indexes all other states in OTH, and first-order contiguous or adjacent neighbors in FOTH. Before introducing each of these neighbor variables we ran an F-test to assess whether or not inclusion of neighbor variables into the model has additional explanatory power. The procedure and the results of these tests will be discussed in the next section. The matrix W is constructed so that the weight wi,j is larger for states that are more similar. For our purposes, the weight matrix should reflect similarities in industrial composition of states. In a recent study, Yilmaz, Haynes, and Dinc (2000) examined the impact of telecommunications capital on individual sectors at the one-digit SIC level, and found that the impact was positive and statistically significant on service related sectors, whereas it was statistically insignificant on others. Because of this greater interaction between telecommunications and service related sectors, the share of total service sector employment (represented by the sum of all service related sectors) in total state employment is © Blackwell Publishing, Inc. 2002.

346

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

used as a measure of cross-state similarity for a given state. It is assumed that states with similar sectoral structures are potential rivals in competing for the same mobile factors of input. Therefore, the weight matrix reflects similarity of these locational characteristics of states for telecommunications infrastructure. The general formula for the weights in these weight (similarity) matrices is wi,j = (1/ |Xi – Xj|) / Si Si = Σ 1/ |Xi – Xj| where Xi is proportion of service employment to total employment in state i and Xj is all states in the larger neighbor definition (OTH = network neighbor) and adjacent states in the second geographic neighbor definition (FOTH = geographic neighbor). The term Si normalizes wi,j such that the sum of the weights for any state is equal to unity. See Boarnet (1998) for similar treatment. The regression model is based on the log-linear Cobb-Douglas aggregate production function for states, including a spillover variable, as presented in Equations (5) and (6) (7)

log(Qit) = α + β1log(Lit) + β2log(Git) +β3log(Kit) + β4log(TKit) + β5log(OTHit) + εit

Previous research has already established the importance of controlling for both regional differences and business cycle effects in a pooled data set (GarciaMila, McGuire, and Porter, 1996; Kelejian and Robinson, 1997; Lall and Yilmaz, 2001). Heterogeneity of state specific characteristics, such as location, climate, and initial endowments can be controlled through the use of state dummy variables or, equivalently, by transforming the data into deviations from state means. In addition to inclusion of state dummies, business cycles and other nationwide output fluctuation effects can be controlled by the use of year-specific intercepts. Therefore, the regression model of Equation (7) becomes (8) log(Qit)= β1log(Lit)+ β2log(Git)+ β3log(Kit)+ β4log(TKit)+ β5log(OTHit) + γt + f + εit where γ is a vector of year specific intercepts and f is a vector of time-invariant state effects. Equation (8) represents the base model for our spillover analysis. In the analysis of spillover effects of telecommunications infrastructure, OTH and FOTH variables are introduced to the base model separately in order to examine the role of geographic and network proximity. The anticipated direction of labor, public capital, private capital, and telecommunications capital variables are apparent; they should have positive coefficients. OTH and FOTH variables represent spillover effects. A negative sign for these variables would imply that locations with better telecommunication infrastructure will experience output growth at the expense of others. A positive sign can be interpreted as an indicator © Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

347

of the existence of positive externalities; that is, the expansion of telecommunication capital in a state will increase output in other states. 3.

THE DATA SET AND EMPIRICAL ANALYSIS

Data The output variable is represented by the total output of private industries. The labor variable represents the total employment of private industries by place of work in a state for a given year. There is no readily available public capital stock data for individual states, so we estimated this variable by apportioning the national total to states. To estimate state public capital, we used the ratio of state total capital outlay expenditure to the national total for each year and apportion total U.S. public capital to respective states. In this process, we examined the state expenditure patterns from 1970 to 1997 and found that the capital outlay variable for all states and the U.S. followed the same trend. This confirms that each state’s share in national capital outlay is a good proxy for the size of its public capital. Therefore, the ratio of state total capital outlay to national total for each year is used to apportion total U.S. public capital to respective states. Similar to public capital there is no readily available private capital stock data at the state level. The estimation of private capital stock is more problematic than public capital because with the exception of the manufacturing sector there is no annual data for private capital investment at the state level. Therefore, state fixed private capital stock Ki is estimated by using the following procedure Ki = [(VADDi – WSi) / (VADDn – WSn)] Kn

(9)

where i indexes state and n indexes the nation. VADD is total value added (output) of private industries, and WS is total wages and salaries for private industries. In this equation, (VADD – WS) represents returns to capital, which is assumed to be an indicator of the size of the private capital stock in a state.9 The estimation of telecommunications capital stock in each state is similar to the methodology specified by Resende (1999) and Shin and Ying (1992). Telecommunications capital stock in each state is obtained by using the automated reporting management information system (ARMIS) of the Federal Communications Commission (FCC).10 The real capital stock is obtained by 9 The ratio of (VADD i – WSi)/(VADD n – WS n) has been steady over the study period. It is expected that in a perfectly competitive environment the long-run equilibrium of VADD i − WSi K i = VADD n − WSn K n holds for each year. We know Kn, so Ki can be derived by

b

g

b

g

using Equation (9). 10 The Automated Reporting Management Information System (ARMIS) was initiated in 1987 for collecting financial and operational data from the largest carriers. A carrier is required to file the ARMIS report if its revenues exceed the ARMIS filing threshold (currently $112 million), or if it is a price-cap carrier, regardless of revenues. Nearly 95 percent of the local exchange service industry is covered in the ARMIS reports. © Blackwell Publishing, Inc. 2002.

348

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

subtracting accumulated depreciation from the gross communications plant figure for each local exchange provider.11 Output, labor, public and private fixed capital stock, and telecommunications stock data are from the Bureau of Economic Analysis (BEA). Both the state and national public capital outlay data are from the Government Finance files of the Bureau of the Census. All monetary values are in per capita constant 1996 dollars. One final note concerns the BEA estimates of capital stock. Since the BEA recently revised its public and private fixed capital stock data to reflect productive capital stock, our data set is not prone to depreciation and measurement related problems. Fraumeni (1997) describes the theoretical and empirical literature that supports the new BEA methodology. Results Access to panel data gives enough degrees of freedom to estimate more robust results. However, previous research on production functions and pooled data shows that in this type of analysis the estimation results might be subject to econometric problems. One such problem is the issue of serial correlation.12 In the case of serial correlation, panel data estimates may result in even more biased estimates than simple ordinary least-squares estimations using cross sectional data alone. Bhargava, Franzini, and Narendranathan (1982) provide a test for serial correlation in panel data sets. The BFN test gives a modified Durbin-Watson statistic.13 We first applied the BFN test to the level form 11

The ARMIS reports cover the period 1987–1997. In this period, the total telecommunications capital stock in a state is calculated as the sum of reporting local exchange carriers’ capital stock in that state. For 1970–1986, the data are available for total local exchange carriers in the U.S. in annual reports of the Statistics of Common Carriers of the FCC. We used the percentage share of a state in total U.S. telecommunications capital stock for years 1987–1997 as a proxy to estimate state-level telecommunications capital stock in previous years. Then, the estimation results for each year are controlled by three different indicators of telecommunications infrastructure reported in the Statistics of Common Carriers: miles of wire, number of central office switches, and total access lines. The correlation coefficients for each year vary between 0.97–0.99 for miles of wire, 0.68–0.89 for number of central office switches, and 0.96–0.99 for total access lines. 12 We thank an anonymous referee for highlighting potential serial correlation problems. In the earlier version of this paper we did not test for serial correlation and indeed our estimation results were subject to serial correlation problems. 13 The Durbin-Watson statistic for panel data is estimated by using the model n

yit = α i

∑x

ijtβ j

+ uit

j =1

uit = ρuit–1 + eit where eit is independently normally distributed with mean zero and variance σ2, the αs are the fixed effects (or dummy variables), the xs are the nonstochastic regressors. The null hypothesis that ρ = 0 H

against the alternative that |ρ| < 1 is tested by d p =

T

∑ ∑ cu~

it

i=1 t=1

~ −u it − 1

H

T

h ∑ ∑ u~ 2

2 it

~ are where u it

i=1 t=1

the residuals from estimating the fixed-effect model, H and T are the number of individual units and the number of time periods, respectively. In our analysis T = 27 and H = 48 therefore it is expected © Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

349

estimation of the model. The BFN statistic of dP = 0.015 is well below the critical value, which suggests that the variables should be transformed into differences. So we employed the first-difference form of the model, in which the BFN statistics is d P = 1.99. Hence the hypothesis that residuals follow a random walk can be rejected. Therefore, to alleviate the serial correlation problem the first-difference form of the regression model is employed in this analysis. Table 1 presents the results of generalized least-squares (GLS) estimation of the production function specified in Equation (8) without a spillover variable.14 The model treats telecommunications capital stock as a factor input in the production function framework. The coefficient on telecommunications capital is positive and statistically significant, suggesting that states benefit from an increase in their telecommunications capital stock. This finding is consistent with the existing literature on telecommunications infrastructure and growth. Private capital and labor variables also have expected signs (indicating that all positively contribute to output) and are significant at the one percent level. The magnitude of private capital stock and labor is at the higher end of the estimates reported in the literature. Note that aggregate level studies usually find high elasticity coefficients compared to studies with disaggregated data sets. The results also provide a reasonable estimate for the coefficient of public capital stock (which is positive and statistically significant) and give some support to the hypothesis that public capital has a positive effect on state output. The time period covered in this analysis coincides with an important event in the history of the U.S. telecommunications system. In 1984, the AT&T monopoly was broken up into regional companies. This is seen as a milestone in telecommunications policymaking that marked the beginning of a new era. After the break up, not only is AT&T divided into seven regional “Baby Bells,” but also telecommunications policymaking power begins shifting increasingly towards the states. Since then, states have become more active in telecommunications regulation and have followed a variety of different telecommunications policy models (Teske, 1995). The empirical evidence and the history of the industry suggest that there may exist a structural break in the analysis.15 Hsiao (1986) suggests an F-test to examine parameter constancy. The F-test result of F (188, 1104) = 7.9 is well above the critical value; therefore the null hypothesis that the parameter vector,

that the dp value will be similar to the Durbin-Watson statistic. As in the case of Durbin-Watson statistic, the null hypothesis of serial independence is rejected if dP is less than dPL, the null is accepted if dP is greater than dPU, the test is inclusive for dPL < dP dPU. 14 The test results for heteroskedasticity indicates that the OLS results suffer from the heteroskedasticity problem: Breusch-Pagan Test (435.12); White Test (294.97). Therefore, we used GLS estimators, which minimize a weighted sum of residual squares. 15 We thank an anonymous referee for bringing this point to our attention. © Blackwell Publishing, Inc. 2002.

350

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

TABLE 1: Regression Results Variable

Public Capital Labor Private Capital Telecommunications Capital Percentage Manufacturinga Adjusted R2 n RSS F Dp

Basic Model

0.001** (0.000) 0.500* (0.009) 0.519* (0.003) 0.013* (0.003) 0.013* (0.001) 0.98 1296 0.038 2809* 1.99

* indicates one percent significant; ** indicates five percent significant. Standard errors are in parentheses. The regression model includes a set of time dummy variables. a The variables are in difference form, so one cannot use state specific dummies to account for the heterogeneity problem. Without including a variable that captures state specific effects on omitted variables, F-test results rejected, the homogeneity hypothesis therefore the variable percentage of employment in the manufacturing sector is included into the analysis to account for heterogeneity. The F-test result with this variable is below the critical value that cannot reject the homogeneity hypothesis; F (240, 1056) = 1.36.

βT, is constant over cross-sectional units in the same period is rejected. The presence of a structural break suggests that it is necessary to fit the model separately for data for the pre-divestiture and post-divestiture periods. We tested the negative spillover hypothesis by using the econometric specification in Equation (8). In specifications 2 and 3, an additional variable representing different neighbor definitions is included in the basic model (specification 1). Specifications 2 and 3 have an additional variable, so specification 1 is the restricted, and specifications 2 and 3 are the unrestricted forms of the econometric model. To test the validity of relaxing these restrictions (i.e., if spillover variables should be introduced to the models), we performed F-tests.16 The computed F scores are presented in the Table 2. The F scores for each of the models are greater than the critical F value except for the model of geographic neighbors in the pre-divestiture period. As seen in the second column, inclusion of the geographic neighbor variable has no explanatory power. The test result for the geographic neighbor is less than the critical value; FG = 0.89. The null 16

The computed F-statistics are given by Fr, n-k = [(SSEr – SSEu)/ r] / (SSEu/ n-k) where SSEr, SSEu, r, n, and k stand for sum of squared errors in the restricted specification and unrestricted specification, number of restrictions, number of observations, and number of estimated parameters, respectively. © Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

351

TABLE 2: F-Test Values for Neighbor Variables Pre-Divestiture Period

SSEr SSEu Fcritical (10 percent level) Numerator Degrees of Freedom Denominator Degrees of Freedom

Post-Divestiture Period

FN = 11.02

FG = 0.74

FN = 2.83

FG = 2.72

0.016107 0.015845 2.71 1 667

0.016107 0.016089 2.71 1 667

0.023952 0.023850 2.71 1 667

0.023952 0.023855 2.71 1 667

hypothesis that the spillover variable has no explanatory value is rejected for all of the other models. Tables 3 and 4 present estimation results for two different time periods, 1970–1984 and 1984–1997, respectively. Specifications 2 and 3 reported in Tables 3 and 4 test the main hypothesis of this paper. In specification 2, the network neighborhood variable is included in the analysis in order to test for spillover effects of telecommunications infrastructure in all other states. As seen in Table 3, during the pre-divestiture period the sign of the network neighbor variable is negative and significant, supporting the negative spillover argument. Specification 3 tests for the impact of the geographic neighborhood. The estimation results for geographic neighbors are reported in the third column of Table 3. The geographic neighbor variable is insignificant in the pre-divestiture period, but insignificance of the variable is expected during this period. In the pre-divestiture period, telecommunications policy was determined to a great degree at the federal level and states did not actively pursue different regulatory policies. Therefore, the impact of negative spillovers, if any, was not amplified at the regional level. However, as seen in Table 4, the coefficient of the geographic neighbor variable in the post-divestiture period is statistically significant, which suggests that geographic proximity amplifies the negative spillover effects. Estimation results in Table 4 support our negative spillover hypothesis. The network neighborhood concept seems to diminish the importance of geographic proximity. However, adjacency still has a stronger effect because of the existing regional linkages. Therefore, it is reasonable to expect the magnitude of the negative spillover effect from telecommunications to be larger at the regional level than at the national level. A given state’s own telecommunications capital stocks have a positive impact on the state’s output whereas neighbors’ telecommunications capital stocks have a negative impact. Overall, the elasticity of telecommunications capital stock could be interpreted as follows—when all states increase their rate of telecommunications investment by one percent, the net change in output growth rate in any state is the sum of coefficients on the telecommunications capital stock and neighbor © Blackwell Publishing, Inc. 2002.

352

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

TABLE 3: Regression Results for Pre-Divestiture Period (1970–1984) Variable

Public Capital Labor Private Capital Telecommunications Capital Percentage Manufacturing

Basic Model

0.005 (0.000) 0.543* (0.08) 0.529* (0.003) 0.023* (0.004) 0.008* (0.001)

Network Neighbor

Specification 2

Specification 3

(0.000) 0.000 (0.008) 0.543* (0.003) 0.528* (0.003) 0.023* (0.001) 0.008* –0.0007** (0.0003)

0.001*** (0.000) 0.544* (0.008) 0.528* (0.003) 0.023* (0.003) 0.0008* (0.001)

Geographic Neighbor Adjusted R2 n RSS F dp

–0.0007 (0.0008) 0.99 672 0.016 12479* 2.163

0.99 672 0.015 9525* 2.179

0.99 672 0.016 11412* 2.173

* indicates one percent significant; ** indicates five percent significant; *** indicates ten percent significant. Standard errors are in parentheses. The regression models include a set of time dummy variables.

TABLE 4: Regression Results for Post-Divestiture Period (1984–1997) Variable

Public Capital Labor Private Capital Telecommunications Capital Percentage Manufacturing

Basic Model

0.001** (0.000) 0.456* (0.013) 0.512* (0.005) 0.010* (0.004) 0.017* (0.003)

Network Neighbor

Specification 2

(0.000) 0.001* (0.013) 0.457* (0.005) 0.512* (0.004) 0.010* (0.003) 0.018* –0.0009** (0.004)

Geographic Neighbor Adjusted R2 n RSS F dp

Specification 3

0.001** (0.005) 0.458* (0.012) 0.511* (0.005) 0.010* (0.002) 0.018* (0.003)

–0.001* (0.000) 0.98 672 0.023 2594* 1.61

0.98 672 0.023 2487* 1.61

0.98 672 0.023 2389* 1.61

* indicates one percent significant; ** indicates five percent significant; *** indicates ten percent significant. Standard errors are in parentheses. The regression models include a set of time dummy variables. © Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

353

variables.17 For example, a one percent increase in the rate of telecommunications investment in all states will result in a net output effect of 0.009 percent on a given state in the post-divestiture period. 4.

ROBUSTNESS OF EMPIRICAL ANALYSIS

Two common problems in the infrastructure literature that contaminate the estimation results are reverse causality and spatial interaction problems. In this section, we perform additional tests to assess the robustness of our findings. Reverse Causality To test the reverse causal link between output and telecommunications capital stock, we use vector autoregressive techniques described in HoltzEakin, Newey, and Rosen (1986). We specify a vector autoregression containing two, three, and four lags of output growth and growth in telecommunications capital stock and cannot reject the null hypothesis that the coefficients on output growth equal zero in the equation for growth in telecommunications capital stock. To control for possible endogeneity, the impact of telecommunications capital stock on output is estimated using the interstate communications portion of local exchange carriers’ networks as an instrumental variable.18 The motivation for this is that regulation of interstate communications is a federal jurisdiction and state regulatory agencies do not set the rates for interstate communications; therefore, the rate of return to this portion of network is exogenous to the system. Thus, investment in the interstate communications portion of telecommunications networks must be exogenous to short-term changes in output of states. The estimation 47

17

OTHi =

∑ w TK ij

j

for the network neighborhood variable, so the impact of changes in

j =1

47

∑ TK

telecommunications infrastructure of state j on state i’s output is ∂ log Qi ∂TK j = βwij

j

. If

j =1

F I GG ∑ w TK JJ H K 47

β1 i

β2 i

β3 i

we write the production function as Qi = αL G K TK

β4 i

ij

j

β5

, the elasticity of output in

j =1

state i with respect to the telecommunications capital in state j is

ε Q ,TKj i

LM = MαL G MN β1 i

β2 i

F I β G ∑ w TK J GH JK 47

β3 i

K TK

β4 i

5

ij

j

j =1

β5−1

OPLM w P MTK PQMN ij

F I β G ∑ w TK J GH JK 47

j

β1 i

β2 i

β3 i

αL G K TK

β4 i

5

ij

j =1

j

β5−1

OP PP Q

47

which is equal to ε Q ,TK = β5 wij TK j 1

j

∑ TK

j

. If all states increase telecommunications capital by

j =1

one percent, the effect on state i will be the coefficient of its own telecommunications capital stock (β4) plus the sum of its network (or geographic) neighbors (β5). 18 The data on investment in interstate communications of local exchange carriers are from the ARMIS Report 43-01. © Blackwell Publishing, Inc. 2002.

354

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

results for the weighted two-stage least-squares show that state telecommunications infrastructure still has a positive and significant effect on output even after filtering out reverse causality impacts. Spatial Interaction Rey and Montouri (1999) indicate that although regional analysis has geographic components, spatial effects have been largely ignored in the literature. The presence of spatial dependence may lead to misspecification of models and spatial heterogeneity can cause instability of behavioral relationships. In this section, we conduct additional tests to address spatial interaction issues and to assess the robustness of our estimates. Spatial Error Dependence One of the main reasons for spatial correlation or dependence in the error terms of regional econometric models is omitted variables that may relate to the connectivity of neighboring regions (Kelejian and Robinson, 1997). In a properly specified model, it is quite likely that spatial dependence would be reduced or eliminated. An important issue in an empirical analysis is to detect the presence of spatial effects and distinguish between cases when spatial dependence is a nuisance and cases when spatial dependence is a substantive spatial process. The null hypothesis in a standard linear regression model is that the error terms are i.i.d. A well-known test for spatial autocorrelation in the regression error term is Moran’s I (Cliff and Ord, 1972). To control for spatial error dependency, a Moran’s I test on residual values of each year has been employed. The results of the Moran’s I test (presented in Table 5) are highly insignificant for all the years in both periods for each specification; hence, spatial error dependence is not a concern in this empirical analysis. Spatial error dependency is not the only source of the spatial interaction problems. In addition to the substantive form of spatial error dependency, a mismatch between spatial boundaries of explanatory variables and administrative boundaries used to compile data might be a source of nuisance dependency. In most cases, the spatial clustering of variables raises questions about model specification. The data employed in the analysis may exhibit a strong spatial interrelation because of geographic proximity, and the results may be misleading. The data for the U.S. states were tested for spatial autocorrelation using Moran’s I statistic with a contiguity based on a first-order spatial weights matrix. According to Moran’s I statistics, none of the variables exhibit considerable spatial autocorrelation throughout the study period. The results for Moran’s I tests are given in Table 6 for selected years. 5.

CONCLUSION

In this study, we examined the impact of telecommunications infrastructure on state output using a pooled data set for 48 states in the U.S. Our results from the fixed-effect model indicate that a state’s output growth rate is positively © Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

355

TABLE 5: Moran’s I Tests for Error Correction Pre-Divestiture Period Year

Moran’s I

1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984

–0.06 –0.01 –0.06 –0.04 –0.05 –0.06 –0.04 –0.03 0.13 0.10 –0.05 –0.06 –0.06 –0.04

Basic Model z-value Probability

–0.36 0.13 –0.45 –0.18 –0.30 –0.42 –0.22 –0.09 1.50 1.23 –0.29 –0.37 –0.44 –0.19

0.72 0.89 0.65 0.85 0.76 0.67 0.82 0.92 0.13 0.22 0.77 0.71 0.66 0.85

Specification 2 Moran’s I z-value Probability

–0.06 –0.01 –0.07 –0.04 –0.05 –0.06 –0.04 –0.03 0.13 0.10 –0.05 –0.06 –0.06 –0.04

–0.36 0.13 –0.45 –0.18 –0.30 –0.42 –0.22 –0.10 1.50 1.23 –0.29 –0.37 –0.44 –0.19

0.72 0.89 0.66 0.86 0.77 0.68 0.83 0.93 0.13 0.22 0.78 0.71 0.66 0.85

Specification 3 Moran’s I z-value Probability

–0.06 –0.01 –0.07 –0.04 –0.05 –0.06 –0.04 –0.03 0.13 0.10 –0.05 –0.06 –0.06 –0.04

–0.36 0.13 –0.45 –0.18 –0.30 –0.42 –0.22 –0.10 1.50 1.23 –0.29 –0.37 –0.44 –0.19

0.72 0.89 0.66 0.86 0.77 0.68 0.83 0.92 0.13 0.22 0.78 0.71 0.66 0.85

Post-Divestiture Period Year

Moran’s I

1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997

–0.06 –0.01 –0.07 –0.04 –0.05 –0.06 –0.04 –0.03 0.13 0.10 –0.05 –0.06 –0.64 –0.04

Basic Model z-value Probability

–0.36 0.13 –0.45 –0.18 –0.30 –0.42 –0.22 –0.10 1.51 1.23 –0.29 –0.37 –0.44 –0.19

0.72 0.89 0.65 0.86 0.76 0.67 0.83 0.92 0.13 0.22 0.78 0.71 0.66 0.84

Specification 2 Moran’s I z-value Probability

–0.06 –0.01 –0.07 –0.04 –0.05 –0.06 –0.04 –0.03 0.13 0.10 –0.05 –0.06 –0.06 –0.04

–0.36 0.13 –0.45 –0.18 –0.30 –0.42 –0.22 –0.10 1.51 1.23 –0.29 –0.37 –0.44 –0.19

0.72 0.90 0.66 0.86 0.77 0.68 0.83 0.92 0.13 0.22 0.78 0.71 0.66 0.85

Specification 3 Moran’s I z-value Probability

–0.06 –0.01 –0.07 –0.04 –0.05 –0.06 –0.04 –0.03 0.13 0.10 –0.05 –0.06 –0.06 –0.04

–0.36 0.13 –0.45 –0.18 –0.30 –0.42 –0.22 –0.10 1.50 1.23 –0.28 –0.37 –0.44 –0.19

0.72 0.89 0.65 0.86 0.77 0.68 0.83 0.92 0.13 0.22 0.78 0.71 0.66 0.85

related to its rate of telecommunications investment, and negatively related to the rate of telecommunications investment by other states. These findings suggest that telecommunications investment is an important factor for a state’s output growth, but it has a negative spillover effect for other states. According to our estimates, the negative spillover reduces the growth rate of output in any state by 0.0009 percent if all states increase their investment rates by 1 percent. The findings suggest that the larger network neighborhood is now an important factor in regional analysis, but geographic proximity continues to play an important role in interregional economic activities. We tested the popular © Blackwell Publishing, Inc. 2002.

356

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

TABLE 6: Moran’s I Test for Spatial Autocorrelation (for selected years) Year

Variable

Moran’s I

1970

Private Capital Public Capital Labor Telecommunications

–0.102 0.013 –0.052 –0.090

–0.832 0.352 –0.318 –0.709

z-value

Probability

0.40 0.72 0.75 0.47

1975

Private Capital Public Capital Labor Telecommunications

–0.108 0.012 –0.054 –0.090

–0.898 0.347 –0.345 –0.711

0.37 0.73 0.73 0.47

1980

Private Capital Public Capital Labor Telecommunications

–0.117 0.008 –0.055 –0.090

–0.992 0.307 –0.348 –0.709

0.32 0.76 0.73 0.48

1985

Private Capital Public Capital Labor Telecommunications

–0.118 0.011 –0.060 –0.088

–1.000 0.335 –0.404 –0.690

0.31 0.74 0.68 0.49

1990

Private Capital Public Capital Labor Telecommunications

–0.126 0.010 –0.064 –0.089

–1.077 0.328 –0.441 –0.695

0.28 0.74 0.66 0.48

1995

Private Capital Public Capital Labor Telecommunications

–0.130 0.004 –0.066 –0.088

–1.127 0.261 –0.456 –0.688

0.26 0.79 0.64 0.49

view of the declining role of geographic proximity and found that geographic proximity amplifies the negative spillover effects of general telecommunications investments. The impact of negative spillover is stronger from geographic neighbors than from network neighbors. Furthermore, the results are unchanged when the spatial interaction problem is corrected for spatially correlated error structure and for spatial autocorrelation. Overall, the evidence supports the idea that telecommunications capital has a significant positive impact on a state’s output and states with similar telecommunications infrastructures compete for mobile factors of production. These findings have important implications for policymakers. First, even though the magnitude of the negative spillover effect is small, its presence calls into question the current regulatory emphasis on subsidizing private telecommunications investment. Current regulatory framework considers telecommunications infrastructure as a public good with only positive network externalities and prescribes subsidies to achieve a socially desirable outcome. Our findings suggest that viewing telecommunications as a purely © Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

357

public good—without taking into account its use as a factor of production—may result in overestimation of the positive externalities. It is important to note that the magnitude of negative spillovers is small so the overestimation of positive externalities would be minor.19 However, the existence of negative spillover effects marks not only the sins of omission, that is, that the current policies fail to ensure that all externalities are internalized; but also the sins of commission, that the existing subsidy scheme may be inefficient. Second, the presence of a negative spillover effect suggests that states may use telecommunications policy as a competitive tool to enhance their own output growth rather than setting socially optimal returns on telecommunications investment. The downside of such behavior is that regions may enter into a “beggar-thy-neighbor” competition where each region tries to provide more infrastructure than it would have otherwise provided. In this context, the agency problem between the federal government and the state government may result in suboptimal levels of telecommunications investment if state governments and the federal government pursue different goals (Jensen and Meckling, 1976). For example, in our second specification, by increasing its own telecommunications capital stock by 1 percent, each state can increase its output growth rate by 0.010 percent in the post-divestiture period. From the agency theory point of view, this is the only relevant variable for the state’s telecommunications investment decision. However, the same action results in a 0.0009 percent reduction in overall output growth, reducing the social benefit of telecommunications investment to 0.0091 percent. If states ignore the negative spillovers, they would invest at higher than socially desirable rates. This interpretation is consistent with the finding that under rate-of-return regulation, firms would choose too much capital relative to other inputs (Averch and Johnson, 1962). Further empirical research on the welfare effects of telecommunications infrastructure and rate-of-return regulation would greatly benefit our understanding of the locational effects of telecommunications infrastructure. Overall, the empirical evidence presented in this paper supports the negative spillover hypothesis. Some interesting questions for further research are whether the same results hold for individual states and whether similarities in regulatory practices amplify the negative spillover effects. Finally, the exact nature of the relation between the marginal products of physical capital, labor, and telecommunications investments is an empirical question, and it is very hard to obtain data on the marginal rate of substitution between these inputs. Therefore, in order to estimate the impact of an increase in telecommunications capital on output, a cost function framework might be used to estimate shadow values where a negative shadow value for telecommunications capital stock would reinforce our findings. This provides the motivation for pursuing further research on the impact of telecommunications infrastructure investments using a cost-function framework.

19

We thank an anonymous referee for bringing up this point.

© Blackwell Publishing, Inc. 2002.

358

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

REFERENCES Anselin, Luc. 1999. Spatial Econometrics. Anselin, Luc and Raymond J. G. M. Florax. 1995. “Small Sample Properties of Tests for Spatial Dependence in Regression Models: Some Further Results,” in Luc Anselin and Raymond Florax (eds.), New Directions in Spatial Econometrics, Berlin: Springer, pp. 56–97. Antonelli, Cristiano. 1993. “Investment, Productivity Growth and Key-Technologies: The Case of Advanced Telecommunications,” The Manchester School, 60, 386–397. Arrow, Kenneth Joseph, Hollis Burnley Chenery, Bagicha Singh Minhas, and Robert Merton Solow. 1961. “Capital-Labor Substitution and Economic Efficiency,” The Review of Economics and Statistics, 43, 225–250. Aschauer, David Alan. 1989a. “Is Public Infrastructure Productive?” Journal of Monetary Economics, 23, 177–200. ———. 1989b. “Public Investment and Productivity Growth in the Group of Seven,” Economic Perspectives, 13, 17–25. ———. 1989c. “Does Public Infrastructure Crowd Out Private Capital?” Journal of Monetary Economics, 24, 171–188. Averch, Harvey and Leland Johnson. 1962. “Behavior of the Firm under Regulatory Constraint,” American Economic Review, 52, 1053–1069. Baer, Walter S. 1995. “Telecommunications Infrastructure Competition The Cost of Delay,” Telecommunications Policy, 19, 351–363. Blackorby, Charles and R. Robert Russell. 1989. “Will the Real Elasticity of Substitution Please Stand Up?” The American Economic Review, 79, 882–888. Blair, John P. and Robert Premus. 1993. “Location Theory” in Richard D. Bingham and Robert Mier (eds.), Theories of Local Economic Development, California: Sage Publications, pp. 121–144. Bhargava, Aruna, Leonard Franzini, and Wiji Narendranathan, 1982. “Serial Correlation and the Fixed Effects Model,” Review of Economic Studies, 49, 533–549. Boarnet, Marlon G. 1997a. “Highways and Economic Productivity: Interpreting Recent Evidence,” Journal of Planning Literature, 11, 4–12. ———. 1997b. “Infrastructure Services and the Productivity of Public Capital: The Case of Streets and Highways,” National Tax Journal, 50, 39–57. ———. 1998. “Spillovers and The Locational Effects of Public Infrastructure,” Journal of Regional Science, 38, 381–400. Borts, George H. 1960. “The Equalization of Returns and Regional Economic Growth,” The American Economic Review, 50, 319–347. Clague, Christopher K. 1969. “Capital Labor Substitution in Manufacturing in Underdeveloped Countries,” Econometrica, 37, 528–537. Cliff, Andrew and Keith Ord. 1972. “Testing for Spatial Autocorrelation Among Regression Residuals,” Geographical Analysis, 4, 267–284. Crandall, Robert W. 1997. “Are Telecommunications Facilities ‘Infrastructure?’ If They Are,So What?” Regional Science and Urban Economics, 27, 161–179. Cronin, Francis J., Elisabeth Colleran, and Mark A.Gold. 1997. “Telecommunications, Factor Substitution and Economic Growth,” Contemporary Economic Policy, XV, 21–31. Cronin, Francis J., Elisabeth Colleran, Paul L. Herbert, and Steve Lewitzky. 1993a. “Telecommunications and Growth The Contributions of Telecommunications Infrastructure Investment to Aggregate and Sectoral Productivity,” Telecommunications Policy, 17, 677–690. Cronin, Francis J., Mark A. Gold, Paul L. Hebert, and Steve Lewitzky. 1993b. “Factor Prices, Factor Substitution, and the Relative Demand for Telecommunications Across U.S. States,” Information Economics and Policy, 5, 73–85. Cronin, Francis J., Mark A. Gold, Beth B. Mace, and John L. Sigalos. 1994. “Telecommunications and Cost Savings in Educational Services,” Information Economics and Policy, 6, 53–75. Cronin, Francis J., Patricia M. McGovern, Michael R. Miller, and Edwin B. Parker. 1995. “The Rural Economic Development Implications of Telecommunications Evidence from Pennsylvania,” Telecommunications Policy, 19, 545–559. © Blackwell Publishing, Inc. 2002.

YILMAZ, HAYNES, & DINC: GEOGRAPHIC AND NETWORK NEIGHBORS

359

Cronin, Francis J., Edwin B. Parker, Elisabeth Colleran, and Mark A. Gold. 1991. “Telecommunications Infrastructure and Economic Growth An Analysis of Causality,” Telecommunications Policy, 15, 529–535. Dholakia, Ruby R. and Bari Harlam. 1994. “Telecommunications and Economic Development Econometric Analysis of The U.S. Experience,” Telecommunications Policy, 18, 470–477. Dinc, Mustafa, Kingsley E. Haynes, Roger R. Stough, and Serdar Yilmaz. 1998. “Regional Universal Telecommunication Service Provision in the U.S.” Telecommunications Policy, 22, 541–553. Eisner, Robert. 1991. “Infrastructure and Regional Economic Development,” New England Economic Review, (September/October), 15–27. Evans, Paul and Georgios Karras. 1994. “Are Government Activities Productive? Evidence from a Panel of U.S. States,” Review of Economics and Statistics, 76, 1–11. Fraumeni, Barbara M. 1997. “The Measurement of Depreciation in the U.S. National Income and Product Accounts,” Survey of Current Business, (July), 7–23. Garcia-Mila, Teresa, Therese J. McGuire, and Robert H. Porter. 1996. “The Effect of Public Capital in State-Level Production Functions Reconsidered,” Review of Economics and Statistics, 78, 177–180. Gramlich, Edward M. 1994. “Infrastructure Investment: A Review Essay,” Journal of Economic Literature, XXXII, 1176–1196. Greenwood, Michael J. and Gary. L. Hunt. 1986. “Migration and Employment Change: Empirical Evidence on the Spatial and Temporal Dimensions of the Linkage,” Journal of Regional Science, 26, 223–234. Haynes, Kingsley E., D. H. Good, and T. Dignan. 1988. “Discrete Spatial Choice and the Axiom of Independence from Irrelevant Alternatives,” Socio-Economic Planning Sciences, 22, 241–251. Holtz-Eakin, Douglas. 1994. “Public-Sector Capital and the Productivity Puzzle,” Review of Economics and Statistics, 76, 12–21. Holtz-Eakin, Douglas, Whitney Newey, and Harvey S. Rosen. 1988. “Estimating Vector Autoregression with Panel Data” Econometrica, 56, 1371–1395. Holtz-Eakin, Douglas and Amy Ellen Schwartz. 1995. “Spatial Productivity Spillovers from Public Infrastructure: Evidence from State Highways,” International Tax and Public Finance, 2, 459–468. Hsiao, Cheng. 1986. Analysis of Panel Data. Cambridge: Cambridge University Press. Hulten, Charles R. and Robert M. Schwab. 1997. “A Fiscal Federalism Approach to Infrastructure Policy,” Regional Science and Urban Economics, 27, 139–160. Jensen, Michael C. and William H. Meckling. 1976. “Theory of Firm: Managerial Behavior, Agency Costs and Ownership Structure,” Journal of Financial Economics, 3, 305–360. Kelejian, Harry H. and Dennis P. Robinson. 1997. “Infrastructure Productivity Estimation and Its Underlying Econometric Specifications: A Sensitivity Analysis,” Papers in Regional Science, 76, 115–131. Kendrick, John W. and Ryuzo Sato. 1963. “Factor Prices, Productivity, and Economic Growth,” The American Economic Review, 53, 974–1003. Lall, Somik and Serdar Yilmaz. 2001. “Regional Economic Convergence: Do Policy Instruments Make a Difference?” Annals of Regional Science, 35, 151–166. Lakshmanan, Tiruvarur Ramaiyer. 1989. “Infrastructure and Economic Transformation” in Åke E. Andersson (ed.) Advances in Spatial Theory and Dynamics, Amsterdam: North-Holland. Leff, Nathaniel H. 1984. “Externalities, Information Costs, and Social Benefit-Cost Analysis for Economic Development: An Example from Telecommunications,” Economic Development and Cultural Change, 32, 255–276. Madden, Gary and Scott J. Savage. 1998. “CEE Telecommunications Investment and Economic Growth,” Information Economics and Policy, 10, 173–195. Mikelbank, Brian A. and Randall W. Jackson. 2000. “The Role of Space in Public Capital Research,” International Regional Science Review, 23, 235–259. Munnell, Alicia H. 1990. “How Does Public Infrastructure Affect Regional Economic Performance?” New England Economic Review, (September/October), 11–32.

© Blackwell Publishing, Inc. 2002.

360

JOURNAL OF REGIONAL SCIENCE, VOL. 42, NO. 2, 2002

———. 1992. “Infrastructure Investment and Economic Growth,” Journal of Economic Perspectives, (Fall), 189–198. National Governors’ Association. 1994. Telecommunications: The Next American Revolution, Washington DC: NGA. Nadiri, M. Ishaq and Banani Nandi. 1997. “The Changing Structure of Cost and Demand for the U.S. Telecommunications Industry,” Information Economics and Policy, 9, 319–347. Norton, Seth W. 1992. “Transaction Costs, Telecommunications, and the Microeconomics of Macroeconomic Growth,” Economic Development and Cultural Change, 41, 175–196. Resende, Marcelo. 1999. “Productivity Growth and Regulation in U.S. Local Telephony,” Information Economics and Policy, 11, 23–44. Rey, Sergio and Brett Montouri. 1999. “U.S. Regional Income Convergence: A Spatial Econometric Perspective,” Regional Studies, 33, 143–156. Shin, Richard T. and John S. Ying. 1992. “Unnatural Monopolies in Local Telephone,” Rand Journal of Economics, 23, 171–83. Staranczak, Genio A., Edgardo R. Sepulveda, Peter A. Dilworth, and Shafi A. Shaikh. 1994. “Industry Structure, Productivity and International Competitiveness: The Case of Telecommunications,” Information Economics and Policy, 6, 121–142. Teske, Paul. 1990. After Divestiture, New York: SUNY Press. ———. 1995. “Introduction and Overview,” in Paul Teske, (ed.) American Regulatory Federalism and Telecommunications Infrastructure. New Jersey: Lawrence Erlbaum. Yilmaz, Serdar, Kingsley E. Haynes, and Mustafa Dinc. 2000. “The Role of Telecommunications Infrastructure on Regional Growth,” Working Paper No. 2000-5. School of Public Policy, George Mason University, Fairfax, Virginia. Youngson, Alexander J. 1967. Overhead Capital, A Study in Development Economics, London: Edinburg University Press.

© Blackwell Publishing, Inc. 2002.