Spatial agglomeration and location determinants: Evidence from the US communications equipment manufacturing industry

Abstract

This paper examines the spatial agglomeration of the communications equipment manufacturing (CEM) industry in the US metropolitan statistical areas. We examine the influence of vertical industrial linkages, horizontal industrial linkages and corporate taxation on the locational choice of CEM establishments using discrete count data regression models. Panel data regression models are used for sensitivity tests. Our results suggest that both types of linkages have significant positive impacts on the spatial agglomeration of the CEM industry, indicating that input-output connections are geographically localised. Our findings support the hypothesis that the presence of vertical linkages suppliers and horizontal linkages in a metropolitan area could facilitate the spatial agglomeration of CEM establishments there. We also find that higher state corporate taxes can impede the choice of location of CEM establishments within a state. These findings suggest that state policymakers can improve the pro-business environment and attract CEM establishments in their state by lowering corporate taxes and by increasing potential horizontal and vertical industrial linkages.

Keywords

communications equipment manufacturing corporate taxation industrial linkages location determinants spatial agglomeration

Introduction

This paper investigates the impact of vertical and horizontal industrial linkages as well as corporate taxation on the location choice of the communications equipment manufacturing (CEM) industry in the United States (US) metropolitan statistical areas (MAs).¹ Vertical linkages with the same or intra-industry linked sectors capture localisation economies while horizontal linkages with inter-industry concentrations of different industries take account of urbanisation economies. Both theoretical and empirical literature suggest that the spatial agglomeration of economic activity is a complex phenomenon. Evidence shows, for example, that economic activity is not distributed evenly at the country, state, provincial or even district levels (World Bank, 2009). An important high-tech manufacturing sector, the CEM sector is highly concentrated in a few US regions. The eastern and western coasts have experienced a rapid increase in the concentration of high-tech industries, with the emergence of a large number of small and medium-sized firms. Most of the CEM establishments have located in those regions, with only a small percentage of them located in the south of the USA. About half of the states contain more than 90% of CEM establishments. The CEM industry also demonstrates unique locational patterns. According to the County Business Patterns data, the geographical distribution of the CEM sector varies significantly across MAs: the top 20 MAs accounted for more than 60% of CEM establishments in 2003 and such concentration remains stable in 2010.

Understanding the locational determinants behind such agglomeration is an important research question, because sector agglomeration is a crucial driver of economic development and the CEM industry represents one of the high-tech manufacturing industries characterised by cutting-edge technologies associated with major companies such as Apple, Cisco Systems and Qualcomm. A thorough investigation of these determinants entails a comprehensive analysis of how particular characteristics of an area affect firms’ propensity to locate their establishments in that area.

In this paper we focus on three such key characteristics, namely, the presence of vertical industrial linkages, the presence of horizontal industrial linkages and the level of corporate taxation. Specifically, we study whether connections of vertically linked suppliers and horizontally linked manufacturing activity are geographically based and localised, and whether such linkages encourage or impede the locational decision of establishments. Regarding the effect of taxation on investment, Baldwin and Krugman (2004) argue that the presence of agglomeration forces makes the tax competition less attractive and can mitigate ‘race-to-the-bottom’ tax competition. Brulhart et al. (2012) claim that agglomeration economies can reduce the sensitivity of firm location choice to tax differentials based on the evidence of Swiss municipalities. It is not known, however, if state corporate tax policy has any impact on the location choice of CEM establishments specifically, in the presence of agglomeration economies generated by industrial linkages. Although a growing literature links agglomeration economies with high concentrations of industrial activity, little is known about such associations specifically for the CEM sector. CEM is characterised by rapid technological changes, a large proportion of skilled labour, high capital intensity and a high ratio of material costs to final products. Companies in this industry make equipment used in telephone, TV broadcasters, transmitters and receivers, data and wireless communications networks. The CEM sector needs a significant R&D investment and depends on large basic material purchases and costs. The sector has a variety of applications and has the potential to change world development patterns. For instance, the telephone, as it merges into the Internet world, begins to offer large data-based features and visual materials. An important trend is the machine-to-machine communication at a distance (Lucky and Eisenberg, 2006). Nevertheless, the sector has experienced a decline in the USA, from approximately 1900 establishments employing nearly 250,000 people in 2003 to only 1351 establishments with half that employment by 2010. Considering the importance of the industry for economic growth and security, it is crucial to understand its locational preferences and determine key location factors that could revive investment in the CEM industry and increase employment.

Many factors determine the location choice of industrial activity. Marshall (1920) first explained three channels influencing firm co-location: intermediate input sharing, labour market pooling and knowledge spillovers. Porter (1990) argues that proximity to suppliers can enhance innovation by increasing knowledge flows and reducing transaction costs. Puga and Venables (1996), Amiti (2005), Amiti and Cameron (2007) and Helsley and Strange (2007) claim that strong forward and backward linkages between industries may foster the clustering of firms. Strauss-Kahn and Vives (2009) also find that same-sector activity agglomeration and industry specialisation have an important impact on location choice, based on US headquarters data. On the other hand, as Coughlin and Segev (2000) suggest, different manufacturers weigh location factors differently because of the different production chains and sector characteristics; while Dumais et al. (2002) do not find the presence of input suppliers for manufacturing activity significant. For the CEM industry specifically we have little information on whether the geographical presence of vertical linkages in an area can also influence its location choice and agglomeration behaviour. Based on the national input-output table from the Bureau of Economic Analysis, we can identify vertically linked sectors that represent inputs to the CEM sector. Our objective is to test whether such input-output linkages are geographically based or localised. So, based on this discussion, we develop our first hypothesis: a stronger presence of vertically linked establishments tends to attract more CEM establishments.

Besides localisation economies generated by the concentration of same-industry firms, the impact of urbanisation economies also has been widely discussed in the literature. Based on the growth of innovative activities of 170 US cities between 1956 and 1987, Glaeser et al. (1992) find that regional diversity offers more advantages for firms than specialisation. Correspondingly, Feldman and Audretsch (1999) and Audretsch (2003) argue that diversity across complementary economic activities is more conducive to innovation than specialisation of economic activities. It is possible that firms act rationally to seek urbanisation economies built up from firms across different industries (Alcacer and Chung, 2007). Such firms would thus be expected to co-locate and form industrial clusters. Based on empirical evidence from the UK, Devereux et al. (2004) find that most geographically concentrated industries appear to reinforce agglomeration with both low entry and exit rates.

On the other hand, many firms with advanced technologies may risk knowledge leakage by co-locating with competitors. In addition, Henderson (2003) finds no evidence of urbanisation diversity or Jacobs’ economies in the machinery industry. Therefore, we aim to examine whether the horizontal linkage or presence of different high-tech manufacturing firms has a positive influence on the location choice of CEM establishments. The discussion generates the second hypothesis for this research: a stronger presence of horizontally linked establishments tends to attract a larger number of CEM establishments.

This study also attempts to examine the impact of corporate taxation on the location and agglomeration of CEM establishments in MAs. Tax rate differences across jurisdictions may lead to distortions of firms’ investment decisions. However, empirical evidence on tax-induced location decisions and subsequent economic development is still inconclusive. One view is that corporate taxation has a decisive effects on firms’ site decisions. Feld and Kirchgassner (2003), Hanson and Rohlin (2011) and Barrios et al. (2012) demonstrate that high corporate taxation deters firms from locating in a region with high tax rates. As such, a low corporate tax rate would be a very important location factor for firms and their headquarters (Strauss-Kahn and Vives, 2009). But another view is that agglomeration economies can offset the negative effects of high tax rates (Baldwin and Krugman, 2004; Brulhart et al., 2012). Furthermore, some scholars argue that the impact of tax rates depends on various conditions. Rathelot and Sillard (2008), based on French micro data, uncover that higher local taxes actually tend to deter firms from setting up in a given zone, but that the effect is weak. Devereux et al. (2007) argue that grants only have a small effect in attracting plants and that firms are less responsive to government subsidies in areas where there are few existing plants of the same industry. Lee (2008) also finds that the magnitude of some incentive programmes is relative small and the use of public funds as incentives to attract industrial plants is not very effective. These debates raise the question whether corporate taxes have a negative impact on the location decision of the CEM sector. Following this line of thought, we derive our third hypothesis: higher corporate taxes may not necessarily drive CEM establishments away.

We believe that our study has useful policy implications by extending the existing literature in several directions. To our knowledge, it is the first systematic analysis of the locational behaviour of this important but understudied sector, the CEM industry. CEM is a major producer of infrastructure and equipment for telecommunication services. Communications equipment companies employed nearly 250,000 people and telecommunications services employed 1 million US workers in 2002, representing 1.1% of the total private workforce. In the long run, the need for humans and machines to communicate indicates that telecommunications will continue to grow apace, as evidenced by the increasing worldwide expansion of wireless and broadband access services. The CEM sector provides a technological foundation for social communications and vital infrastructure for national security and it constitutes a foundation for global competitiveness. Lucky and Eisenberg (2006) even argue that US industrial competiveness would be affected if American telecommunications infrastructures fall significantly behind those of the rest of the world.

However, there is little location analysis of the CEM sector in US MAs. Indeed, there is a dearth of empirical location research on a single sector at the detailed-area level. Only a few papers on a single sector can be found in the literature: Carlton (1983), Beardsell and Henderson (1999), Frenkel (2001) on the high-tech industry, Syverson (2004) on the concrete sector, Cohen and Morrison Paul (2005) and Haddad et al. (2010) on the food manufacturing industry. Also, research on the distribution of a single sector in a country allows us to differentiate the influence of both sector and country variances from the impact of metropolitan area characteristics on the location choice of firms over time.

The purpose of this paper is not to give a final word on the locational behaviour of a particular industry across regions in a country. Instead, the sector-based study shows how upstream intermediate suppliers, horizontal linkages in a manufacturing base and corporate taxation influence the location choices of the CEM sector, while controlling other factors such as professional high-tech manufacturing services, transportation, ICT (information, communication and technology) services, education services and other metropolitan characteristics that might affect spatial distribution variation in disaggregated geographical areas. The primary benefit of analysing a single industry within a country is that it minimises the influence of sector differences on location selection heterogeneity and isolates the impacts of sector and country characteristics.

In addition, our research employs metropolitan level data that are based on economic integration rather than on mere administrative boundary. The relatively independent, integrated and disaggregated MAs are based on natural economic linkages and have the advantage of avoiding abrupt administrative divisions. The availability of data for small localities, coupled with detailed metropolitan information, allow us to get unbiased estimates on location determinants.

Finally, we test three hypotheses by using count data models. Panel data models are used to deal with identification issues. The empirical evidence is consistent with our three hypotheses that the presence of vertical and horizontal linkages in a metropolitan area exerts positive influence on the CEM industry to locate there, and that high corporate taxes have the opposite effect on the CEM sector; i.e. that for the CEM sector, the presence of agglomeration economies cannot offset the effect of high corporate taxation.

Distribution of the CEM establishments

Before presenting the methodology for the paper, we briefly review the geographical distribution of the CEM sector. The sector is not evenly distributed among MAs. Of a total of 363 US metropolitan statistical areas, 147 had no CEM establishments in 2003, while the top 20 MAs accounted for around two-thirds of the total CEM establishments. Figure 1 demonstrates the geographical distribution of log CEM establishments across MAs. The CEM sector mainly clusters in four primary regions: Los Angeles, San Jose and San Francisco in the Far West region, Boston, New York and Washington DC in the Eastern coast, Chicago in the Midwest region, and Dallas and Palm Bay in the Southern region.

Figure 1.

The geographic distribution of log CEM establishments in US metropolitan areas in 2003.

Another interesting aspect of the geographical distribution of the CEM sector is in the difference between larger MAs and smaller ones. In the left panel of Figure 2, we plot a kernel distribution of log CEM establishments per 100,000 people of MAs above 1 million people (solid), and the establishment distribution of those below 1 million people (dashed). We can immediately see that log CEM establishments per capita in larger population areas do not have significant density externalities. It seems that smaller MAs have a higher number of establishments per capita.

Figure 2.

The distribution of log CEM establishments (left side) and CEM employment (right side) per 100,000 people by metropolitan area population size.

In contrast to the left panel of Figure 2, the distribution of log CEM employment per capita in the right panel of Figure 2 conveys a much different picture. It shows that MAs with a population of more than 1 million people have higher CEM employment rates per capita than smaller MAs, which indicates the existence of agglomeration economies. The different results in Figure 2 are attributed to the fact that CEM establishments in larger MAs are larger in size and hence have more employment. Therefore, large size MAs have more CEM employment or jobs per capita while having fewer CEM establishments per capita.

Methods, variables and data sources

Methods

Considering the discrete distribution of CEM establishments across MAs, count data models are applied to test the three hypotheses presented in the Introduction. Two main count models, the Poisson model and the negative binomial model, are usually used in discrete count data analysis (e.g. Cheng and Stough, 2006; Coughlin and Segev, 2000). Poisson regression, as one of the main models for count data (Hilbe, 2007), is derived from the Poisson distribution by allowing the intensity parameter to depend on explanatory variables. The Poisson regression model would be more suitable and would fit count data well when the variables could not be well estimated by OLS regression.

While Poisson models fit general count data well, the negative binomial regression model can fit count data better with over-dispersed distribution. Over-dispersion tests show that CEM establishments across MAs are over-dispersed, so negative binomial models would be better suited than Poisson models for them. According to Hausman et al. (1984) and Cameron and Trivedi (1986), the general negative binomial model is given as:

p (Y_{θ} = y) = \frac{Γ (y + θ)}{Γ (y + 1) Γ (θ)} {(\frac{θ}{μ + θ})}^{θ} {(\frac{μ}{μ + θ})}^{y}

(1)

where $Γ (\cdot)$ is the gamma function, $θ$ measures over-dispersion and is called the dispersion parameter, and $μ$ is the mean of the negative binomial distribution (like Poisson) but the variance is $\frac{μ + μ^{2}}{θ}$ . When $θ = α^{- 1}$ then the unconditional distribution of $y_{i}$ is negative binomial:

\begin{matrix} f (y_{i} | μ, α) = \frac{Γ (y_{i} + α^{- 1})}{Γ (y_{i} + 1) Γ (α^{- 1})} \\ {(\frac{α^{- 1}}{α^{- 1} + μ})}^{α^{- 1}} {(\frac{μ}{α^{- 1} + μ})}^{y_{i}} \\ α \geq 0, y_{i} = 0, 1, 2, \dots \end{matrix}

(2)

As $θ$ increases and $μ$ is fixed, the negative binomial model converges to a Poisson model, i.e. var $(Y_{θ}) \to μ .$ The negative binomial model will be transformed into the Poisson model if $a = 0$ . The function $Γ (\cdot)$ is the gamma function $Γ (y + α) / Γ (α) = Π_{j = 0}^{y - 1} (j + α)$ , and $y$ is an integer:

\ln (\frac{Γ (y_{i} + α^{- 1})}{Γ (α^{- 1})}) = \sum_{j = 0}^{y - 1} \ln (j + α^{- 1})

(3)

Substituting equation (3) into equation (2), the log-likelihood function for exponential mean $μ_{i} = \exp (x'_{i} β)$ is therefore:

\ln L = \sum_{i = 1}^{n} [\sum_{j = 0}^{y - 1} \ln (j + α^{- 1}) - {lny}_{i}! - (y_{i} + α^{- 1}) \ln (1 + α \exp (x'_{i} β)) + y_{i} lna + y_{i} x'_{i} β]

(4)

where the log-likelihood function $lnL = \ln L (α, β)$ . For building a regression model, it is natural to express the negative binomial distribution in terms of the parameter $μ_{i} = \exp (x'_{i} β)$ .

For regression purposes, we typically assume:

y_{i} ~ Negbin (μ_{i}, 1 / α)

(5)

and apply a log link, so that:

\log μ_{i} = η_{i} = o_{i} + x_{i}^{T} β

(6)

Using cross-sectional data from 2003, we build a simple empirical model using negative binomial regression:

\begin{matrix} \log cem = & β_{1} + β_{2} \log up + β_{3} \log im \\ + β_{4} \log tax + ρ Z + ε \end{matrix}

(7)

where the dependent variable $cem$ captures the spatial agglomeration of the CEM establishments, while $\ln up$ , $\ln im$ , and $\ln tax$ represent the three factors to be tested. Variable $Z$ refers to a number of control variables in the model discussed below, and $ε$ is the error term. $β_{1}$ , $β_{2},$ and $β_{3}$ are coefficients to be estimated through negative binomial regression.

Variables

The US national input-output tables (such as the total requirements tables) and IBIS world industry reports on the CEM sector for various years provide ideas on factors that influence the CEM sector most.² The input-output table shows that the CEM sector is closely related to its component suppliers such as semiconductor and other electronic component manufacturing (NAISCS 3344), manufacturing and reproducing magnetic and optical media (NAISCS 3346), which are the same with the CEM sector in terms of three-digit NAICS code industries. The tables also indicate that the CEM sector makes use of machinery manufacturing (NAICS 333) or industrial machinery manufacturing (NAICS3332), which are different from the CEM sector in terms of three-digit NAICS code industries.

To find the impact of vertical linkages and horizontal linkages on the location choice of the CEM sector, we first test whether those factors are geographically located. In order to test Hypothesis 1, we first review the presence of vertical linkages or upstream intermediate input suppliers (MFGUP). Following the method of Cainelli and Iacobucci (2012), this research measures the vertical linkages as the total number of semiconductor manufacturing establishments (NAICS 3344) in a MA.

To examine Hypothesis 2, we use the presence of horizontal linkages or manufacturing base as inter-industry activity or horizontal linkage (MFGIM). We choose industrial machinery manufacturing services (NAICS 3332) as a proxy for horizontal manufacturing services. Regarding Hypothesis 3 on corporate taxation, the MA tax rate is directly available for only two years (2004 and 2005) from the Tax Foundation. To bypass this shortage of data, we used as proxy for the MA corporate tax burden the average state corporate burden where the MA is primarily located. When one MA involves several states, we choose the corporate tax burden of the state where the first city of a MA located as the MA tax rate. There are no particular tax rates or investment incentives that target the CEM sector specifically. Following the method of Coughlin and Segev (2000), we choose the general measurement of taxation by calculating the state and local tax burden as a share of gross state product: the state tax burden divided by the gross state product (TAX/GSP).

As for control variables for a simple cross-section count data model, we consider several major control factors that have been widely discussed in the literature. The first one is GDP per capita (PERCAP), measuring the development level of an area that may attract CEM industry. GDP per capita also has significant correlation with wages, which may deter investment with high GDP per capita. The second control variable, population (POP), is used to control for the metropolitan area size effect because areas with higher population may result in more manufacturing activity. Some researchers (e.g. Coughlin and Segev, 2000) have used the absolute number of new plants for each region as their dependent variable to examine the location determinants of manufacturing plants, but such an approach needs population or GDP of a metropolitan area to control for the scale effect. In our count data model, we use population to control for the scale effect (for details, please see Wooldridge, 2013). Other control variables include financial services (FINANCE) represented by the credit intermediation services and university knowledge spillovers and services (UNIV) exemplified by the number of colleges and universities.

Furthermore, we also use a dummy variable to control for whether a metropolitan area is located in a pro-business state (RTW = 1 if the state adopts right-to-work law), as Holmes (1998) found that there is more manufacturing activity in right-to-work states. A truck-port-rail hub dummy (TRSPHUB = 1 if an area has a truck-port-rail hub) is applied to measure the convenience of transportation access level of a metropolitan area. A Far West region dummy to account for unobserved heterogeneity (FARWEST = 1 if the metropolitan area is in the Far West region).

Data sources

Metropolitan area level data sets are taken from several sources. The Metropolitan Business Patterns (MBP) database provides establishment data for the dependent variable, CEM establishments. The independent variables from the MBP data set include variables regarding the presence of vertical and horizontal linkages (MFGUP, MFGIM, MFGMO and MFGMA), professional scientific and technological services (MFGSVS), equipment wholesalers (EQPWHL), air transport (AIRTSPT), credit intermediation (FINANCE), information, communication and technology (ICT) and number of colleges (UNIV) from 2003 to 2010. Data on population are taken from the US Census for the period 2003–2010. The real GDP per capita data set comes from the Bureau of Economic Analysis. State corporate income tax rates (CORPTAX) are available from Tax Foundation. Table 1 lists our data sources and contains descriptive statistics for the main variables. While there is no missing data issue for establishment data, some CEM industry employment data used in Figure 2 are missing. Following the practice of Isserman and Westervelt (2006) and Zhang and Guldmann (2009), we estimate the missing employment data through averaging the employment by MAs across years from 2003 to 2010.

Table 1.

Descriptive summary of variables.

Definition	Variables	N	Mean	Median	Std.dev.	Min	Max	Sources
# Communications equipment manufacturing (CEM) establishments (NAICS3342)	CEM	1728	7.418	2	16.649	0	145	MBP
# Semiconductor and other electronic component manufacturing establishments (NAICS 3344)	MFGUP	1728	19.397	4	50.882	0	525	MBP
# Industrial machinery manufacturing establishments (NAICS 3332)	MFGIM	1728	13.517	5	26.174	0	224	MBP
# Manufacturing and reproducing magnetic and optical media establishments (NAICS 3346)	MFGMO	1728	3.5877	5	9.871	0	122	MBP
# Machinery manufacturing (NAICS 333)	MFGMA	1728	87.461	32	173.186	1	1462	MBP
State corporate tax rates (2003–2010)	CORPTAX	1728	9.825	9.810	1.184	7.188	13.080	TF
State corporate tax rates (2000–2007)	CORPTAX_t-3	1728	9.638	9.611	1.140	7.078	12.422	TF
# Professional, scientific, and technical services establishments (NAICS 541)	MFGSVS	1728	3202.576	993	7098.651	91	69,953	MBP
The real average annual wages of the computer and electronic product manufacturing sector (NAICS 334) divided by metropolitan area GDP per capita	WAGERT	1728	1.443	1.380	0.476	0.396	4.749	MBP
The real average annual wages of the computer and electronic product manufacturing sector (NAICS 334) (in US$1000)	WAGEAV	1728	52.497	53.985	14.481	13.718	121.721	MBP
# Professional and commercial equipment and supplies merchant wholesalers establishments (NAICS 4234)	EQPWL	1728	144.869	40	332.037	1	3212	MBP
# Air transportation establishments (NAICS 481)	AIRTSPT	1728	19.577	6	43.305	0	388	MBP
# Credit intermediation and related activities (NAICS 522)	FINANCE	1728	723.153	285	1296.365	31	11,707	MBP
# Wireless telecommunications carriers (NAICS 5172)	ICT	1728	44.097	18	76.518	1	784	MBP
# Colleges, universities, and professional schools (NAICS 6113)	UNIV	1728	14.459	5	29.697	0	276	MBP
Real GDP per capita in 2001 US$	PERCAP	1728	38,266	36,512	11,177	11,793	89,350	BEA
# Population in metropolitan areas	POP	1728	1,014,852	409,022	1,935,934	55,212	1.89e+07	Census

Notes: All variables in the table are in their original values. The data sources MBP, TF, BEA and Census refer to Metropolitan Business Patterns, Tax Foundation, Bureau of Economic Analysis, and US Census data.To save space some variables such as dummy variables are not presented in the table.

Results

We begin with the results for count data model specifications and check the robustness of results by looking at different samples and specifications. We first conduct both Poisson and negative binomial regression analyses. The over-dispersion tests³ show that the data is over-dispersed and thus we adopt the negative binomial regression analysis for the CEM establishments. Table 2 presents the results of the negative binomial regressions. The baseline model in Model 1 based on 363 observations shows that both lnMFGUP and lnMFGIM are positive and statistically significant at the 1% level. Model 2 extends the baseline model by including a dummy variable ‘right-to-work’ (RTW), which has a value of 1 if the metropolitan area in a state has such regulation.

Table 2.

Negative binomial regression results for CEM establishments.

	CEM establishments (N = 363)			CEM establishments (N = 216)
Variables	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
lnMFGUP	0.626***	0.637***	0.606***	0.458***	0.441***	0.408***
	(0.0616)	(0.0673)	(0.0662)	(0.0633)	(0.0703)	(0.0687)
lnMFGIM	0.969***	0.945***	1.003***	0.693***	0.731***	0.768***
	(0.192)	(0.201)	(0.197)	(0.196)	(0.208)	(0.202)
lnCORPTAX	−0.549	−0.551	−0.733*	−0.902**	−0.902**	−1.093***
	(0.398)	(0.397)	(0.389)	(0.394)	(0.395)	(0.385)
lnPERCAP	−0.133	−0.130	−0.0712	0.543*	0.544*	0.615**
	(0.305)	(0.304)	(0.299)	(0.306)	(0.306)	(0.300)
lnPOP	0.943***	0.944***	0.984***	0.802***	0.801***	0.831***
	(0.0438)	(0.0438)	(0.0463)	(0.0453)	(0.0455)	(0.0478)
lnFINANCE	−0.157	−0.153	−0.245	−0.359**	−0.367**	−0.469***
	(0.166)	(0.166)	(0.166)	(0.173)	(0.174)	(0.173)
lnUNIV	0.0147	0.0294	0.159	−0.0556	−0.0766	0.0760
	(0.149)	(0.153)	(0.157)	(0.153)	(0.158)	(0.161)
RTW		0.0389	0.0496		−0.0548	−0.0101
		(0.0971)	(0.0987)		(0.0992)	(0.0995)
TRSPHUB			−0.223**			−0.144
			(0.0974)			(0.0975)
FARWEST			0.244**			0.328***
			(0.115)			(0.114)
Constant	−21.68***	−21.74***	−23.51***	−19.72***	−19.66***	−21.37***
	(2.243)	(2.243)	(2.280)	(2.257)	(2.266)	(2.313)
Lnalpha constant	−2.109***	−2.115***	−2.263***	−2.186***	−2.178***	−2.343***
	(0.300)	(0.302)	(0.326)	(0.253)	(0.252)	(0.276)
Observations	363	363	363	216	216	216
Pseudo R ²	0.305	0.305	0.309	0.294	0.295	0.301
AIC	1188.1	1190.0	1186.0	971.86	973.55	968.88
BIC	1223.2	1228.9	1232.8	1002.2	1007.3	1009.4
Log-likelihood	−585.06	−584.98	−581.01	−476.93	−476.78	−472.44
$χ^{2}$	512.73	513.89	520.83	397.85	398.15	406.82

Notes: Standard errors in parentheses.

***

p < 0.01, **p < 0.05, *p < 0.1.

Modes 1–3 include all 363 MAs and Models 4–6 have the same structure as Models 1–3 but use a subsample of 216 MAs which have at least one CEM establishment.

Model 3 further extends Model 2 by adding two dummy variables, truck-port-rail hub (TRSPHUB) and a regional dummy (FARWEST). As shown in Model 3 of Table 2, the coefficient of the vertical linkage (lnMFGUP) is 0.606 and statistically significant at the 1% level. The negative binomial regression coefficients can be interpreted as follows: for a one unit change in the independent variable, the difference in the log of expected counts of the dependent variable is expected to change by the corresponding regression coefficient, given that other independent variables in the model are held constant (Cameron and Trivedi, 2010). With a one unit increase in the lnMFGUP and lnMFGIM, the difference in the logs of expected CEM establishments is expected to increase by 0.606 and 1.003, respectively. When a MA is to increase its corporate tax by one unit, the difference in the logs of expected establishments in a MA is likely to decrease by a factor of 0.733 while holding all other variables in the model constant.⁴ Therefore, the higher a MA’s corporate tax, the fewer the predicted CEM establishments. But one unit of lnCORPTAX increase is likely to decrease the number of log CEM establishments by 0.73, which is a pretty strong effect on the location choice of the CEM. We also can interpret the result in Model 3 of Table 2 in terms of incidence rate ratios for the negative binomial regression model shown earlier.

The IRR coefficients of MFGUP, MFGIM and CORPTAX in Model 3 are 1.83, 2.73 and 0.48, respectively.⁵ For the estimated rate ratio, for a one establishment increase in the lnMFGUP, and lnMFGIM, the CEM establishment in a MA would be expected to increase by 1.83 and 2.73 time, respectively, but an increase of CORPTAX would be expected to decrease the number CEM establishments by a factor 0.48.

Models 4–6 have the same structure as Models 1–3 but use a subsample of 216 MAs which have at least one CEM establishment. The results in Table 2 Variable MFGUP are also statistically significant in our robustness checks for Models 4–6 with lower AIC and BIC, indicating a better model fit. It seems that the influence magnitudes of vertical linkages (MFGUP) and horizontal linkages (MFGIM) have been reduced but the negative corporate tax effects become stronger with statistical significance at the 5% or lower level.

In particular, the corporate income tax per capita (CORPTAX) has a negative impact on the location of CEM establishments with statistical significance at the 10% level when controlling for right-to-work legislation, transportation hub and primary region dummies. Models 1–2 in Table 2 are not statistically significant, but Model 3 becomes significant when we control for the regional dummy. This result seems to confirm the finding by Devereux et al. (2007) who claim that firms are less responsive to government incentives in areas where there are fewer existing firms. Corporate tax incentives may not work significantly in non-Far West regions, but do deter the location of CEM establishments in the Far West region, including the states of California and Washington, where around one-third of the total CEM establishments are located. Nevertheless, results of the subsample analysis shown in Models 4–6 of Table 2 suggest that firm income tax burden has a negative and statistically significant effect on CEM establishment concentrations regardless of regional difference. This result is echoed by the finding of Hanson and Rohlin (2011) and Barrios et al. (2012). It is often claimed that local tax policy is relevant to firm’s location choice (Greenstone et al., 2010; Guimaraes et al., 2004). The subsample results of Models 4–6 of Table 2 clearly show that the CEM sector locational choices are strongly influenced by state corporate tax levels.

In summary, the location choice of CEM establishments is greatly influenced by the existing manufacturing establishments through both vertical linkages and horizontal linkages. We do not find that the intra-industry concentration has a stronger influence on the location choice of the CEM establishments than the inter-industry concentration. In addition, increases in government taxation programmes can deter the location decision of the CEM sector. In other words, the presence of suppliers with vertical linkages and of manufacturing services with horizontal linkages, and the tax hospitality of a metropolitan area (corporate tax burden) indeed influence CEM location selection. To save costs and improve competitiveness, each CEM establishment prefers to locate in a metropolitan area where a large amount of intermediate inputs and manufacturing services are available. Given other things equal, the preferred metropolitan area is the one with the largest manufacturing service agglomeration. CEM firms could benefit from being close to a large number of upstream firms because of value chain linkages: the more upstream firms in a metropolitan area the lower the cost of intermediate inputs. The reason for this is due to increased competition (Amiti, 2005). Meanwhile, the vertical linkages reinforce location proximity and lead to the spatial agglomeration of CEM establishments. The results also show that investment hospitality, represented by a lower corporate tax burden or by investment incentives or stimuli, could reduce production costs and thus attract CEM establishments; while a higher tax burden can chase them off.

Robustness checks

While the results of this analysis have been robust, there could be a concern about the causal relationship between the CEM sector and the vertical linkage variable. It is possible that some upstream suppliers may be attracted by the presence of CEM sector establishments, and that would result in a reverse causality issue and a spurious correlation between the CEM sector and upstream suppliers. To address this identification concern we adopt three different identification strategies: fixed effect panel data model, dynamic panel data model and alternative proxies for variables of interest.

This section attempts to conduct robustness checks for the results generated in Table 2. Especially in the panel data model, the fixed effects methods have the ability to control for all stable characteristics of the individuals in the study and thereby eliminating potential bias. Furthermore, we use lagged independent variables to test the causality. In particular, we first choose a couple of different manufacturing industries (NAICS 3346 and NAICS 333) in the regression as alternatives that are plausibly exogenous. As shown in Model 3, the fixed effect regression results of Models 6–7 show consistent results with those of Model 5 in Table 3.

Table 3.

Panel data regression results.

		Random effect (N=1728)				Fixed effect (N=1728)
Variables	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8
laMFGUP	0.235***		0.240***	0.229***	0.0738*		0.0741*	0.0587
	(0.0279)		(0.0278)	(0.0278)	(0.0393)		(0.0394)	(0.0398)
laMFGIM	0.105***	0.146***		0.103***	0.0985**	0.102***		0.0997**
	(0.0293)	(0.0305)		(0.0293)	(0.0395)	(0.0390)		(0.0398)
CORPTAX	−0.0440***	−0.0432***	−0.0420***		−0.0831***	−0.0808***	−0.0807***
	(0.0109)	(0.0115)	(0.0109)		(0.0148)	(0.0148)	(0.0150)
laMFGSVS	0.150***	0.226***	0.155***	0.152***	0.306**	0.315***	0.312**	0.313**
	(0.0548)	(0.0595)	(0.0551)	(0.0549)	(0.121)	(0.121)	(0.121)	(0.122)
WAGERT	−0.0282	−0.0370	−0.0227	−0.0314	−0.0538**	−0.0580**	−0.0498*	−0.0628**
	(0.0225)	(0.0234)	(0.0226)	(0.0225)	(0.0270)	(0.0267)	(0.0271)	(0.0270)
laEQPWHL	0.0603**	0.0987***	0.0623**	0.0611**	0.147***	0.154***	0.149***	0.149***
	(0.0272)	(0.0277)	(0.0271)	(0.0272)	(0.0312)	(0.0309)	(0.0311)	(0.0314)
laAIRTSPT	0.114***	0.125***	0.117***	0.115***	0.158***	0.154***	0.163***	0.158***
	(0.0259)	(0.0266)	(0.0260)	(0.0260)	(0.0293)	(0.0292)	(0.0292)	(0.0295)
laFINANCE	−0.0495	−0.0938*	−0.0574	−0.0412	−0.0919	−0.111*	−0.105*	−0.0966*
	(0.0513)	(0.0520)	(0.0515)	(0.0516)	(0.0573)	(0.0574)	(0.0578)	(0.0578)
laICT	0.0340	0.0315	0.0367	0.0362	0.0492**	0.0436*	0.0519**	0.0489**
	(0.0227)	(0.0229)	(0.0227)	(0.0228)	(0.0235)	(0.0235)	(0.0235)	(0.0237)
laUNIV	−0.0212	−0.00719	−0.0221	−0.0220	0.0555	0.0620	0.0543	0.0574
	(0.0324)	(0.0337)	(0.0324)	(0.0325)	(0.0389)	(0.0388)	(0.0389)	(0.0390)
laMFGMO		0.222***				0.288***
		(0.0641)				(0.0814)
laMFGMA			0.0743***				0.0912**
			(0.0235)				(0.0390)
CORPTAX_t–3				−0.0349***				−0.0764***
				(0.0119)				(0.0177)
Constant	−0.213	−0.432	−0.323	−0.349	−0.666	−0.654	−0.803	−0.746
	(0.343)	(0.368)	(0.350)	(0.346)	(0.672)	(0.668)	(0.676)	(0.685)
Observations	1728	1728	1728	1728	1728	1728	1728	1728
R–squared	0.2635	0.0973	0.2626	0.2696	0.110	0.115	0.109	0.102
Number of MAs	216	216	216	216	216	216	216	216
F value					18.50	19.50	18.41	17.08
Chi–square	244.5	181.6	241.7	236.3

Notes: Standard errors in parentheses.

***

p < 0.01, **p < 0.05, *p < 0.1.

Variable with prefix la- indicates that the variable is in a natural logarithm form in the regression. To avoid metropolitan area size effect in the panel data regression analysis, all those variables with prefix la- have been averaged by metropolitan area population before the natural logarithm transformation. The two baseline Models 1and 5 report the effects of vertical linkage or upstream intermediate input (MFGUP), horizontal linkage or manufacturing base (MFGIM) and corporate taxation (CORPTAX) on the spatial agglomeration of CEM establishments while controlling variables discussed in the literature including professional high-tech manufacturing services (MFGSVS), wage ratio (average wage of computer and electronic manufacturing industry divided by per capita GDP to reflect a relative wage rate) (WAGERT), equipment wholesalers (EQPWHL), air transportation (AIRTRSP), financial services (FINANCE), information, communication and technology services (ICT) and number of universities (UNIV). Models 2 and 6 change the baseline model by using magnetic and optical media manufacturing (which is the same category with the CEM sector in terms of three-digit NAICS334) (MFGMO) to replace the MFGUP variable to check whether the upstream intermediate input variable is sensitive to a proxy. Similarly, Models 3 and 7 extend the baseline model by considering alternative proxy, machinery manufacturing (MFGMA) which is different from the CEM category, for the horizontal linkage variable. Models 4 and 8 include both of the alternative proxies.

We start with the conjecture that such agglomeration of CEM sector establishments in an area would be influenced by the three factors mentioned above while controlling a wide range of variables, including professional scientific and technological services, wholesaler services, transportation, information, education, skilled labour and financial services. Hence we can write an empirical equation as follows:

\begin{matrix} y_{i, t} = & α + γ m_{i, t} + φ h_{i, t} + θ τ_{i, t} \\ + X_{i, t}' β + u_{t} + v_{i} + ε_{i, t} \end{matrix}

(8)

In equation (8),⁶ i indexes individual metropolitan areas $(i = 1, \dots, N)$ ; t indexes time $(t = 1, \dots, T)$ ; $y_{i, t}$ represents the natural logarithm of the number of CEM establishments per capita in metropolitan area i at time t; $α$ is the constant; $m_{i, t}$ , $h_{i, t},$ and $τ_{i, t}$ is the simplified expression of $\ln M, φ \ln H,$ and $\ln T,$ and $X_{i, t}$ is a vector of control variables; $u_{t}$ denotes the time effect, controlling for all effects affecting the establishments in the same year; $v_{i}$ is the metropolitan area fixed effect, which controls for the unobserved time-invariant metropolitan area-specific characteristics; and $ε_{i, t}$ is the idiosyncratic error term. The $v_{i}$ and $ε_{i, t}$ are assumed to be independent and identically distributed (i.i.d.) random variables, with zero means and constant variances. Standard error is clustered at the metropolitan area level to deal with the potential heteroscedasticity.

While the random effect models show results consistent with the fixed effect models, we adopt the results of fixed effect models. The Hausman test (Chi-square (10) = 104.57 and Prob > chi-square = 0) rejects the null hypothesis that the difference between fixed effects and random effects in coefficients is not systematic and suggests that the fixed models should be selected. The fixed effect models have advantages in reducing the possibility of simultaneity bias arising from natural geographic features in MAs that may explain the spatial agglomeration of CEM, for example, the eastern coast and western coast locations for many CEM establishments. All the results of Models 6–8 in Table 3 are very similar to those of the baseline estimate in Model 5 in Table 3.

The fixed effect results shown in the baseline Model 5 in Table 3 have consistent results with the results in Table 2. Table 3 suggests that the presence of vertically linked establishments (MFGUP) has a positive impact on the concentration of the CEM sector with statistical significance at the 0.1 level. The result indicates that a rise of 10% in upstream intermediate inputs, such as semiconductor and other electronic component manufacturing establishments (NAICS 3344) in a MA, would result in an increase of approximately 0.7% in CEM establishments. Such a result is stable with the same sign and similar magnitude even with other controls changed, as shown in Models 7–8. Model 6 in Table 3 adopts an alternative vertically linked proxy (MFGMO) and shows that the upstream suppliers magnetic and optical media manufacturing establishments (NACS3346) have the same sign with Model 5 and are statistically significant at the 0.01 level. The intermediate input establishments have an even stronger influence on the location choice of the CEM sector than semiconductor. A 10% rise of the sector is likely to increase the number of CEM establishments by 2.9%.

Model 5 in Table 3 also shows a positive impact of horizontal linkage or manufacturing base services on the location decision of CEM establishments. The proxy for such variable, presence of industrial machinery manufacturing establishments in a metropolitan area (MFGIM), shows that the horizontal connection is positive and statistically significant at the 0.05 level. It indicates that a 10% increase in horizontally linked establishments (such as industrial machinery manufacturing) in a metropolitan area is likely to cause an increase of CEM establishments by 0.9%. The result remains stable and its coefficient magnitudes are consistent when control variables change, as shown in Models 6–8.The alternative proxy (MFGMA) for the variable also has the same sign, but with a coefficient magnitude slightly decreased.

The corporate taxation variable (CORPTAX) is measured by the average state corporate tax rate. It is shown that corporate taxation has a negative impact on the attraction of the CEM sector with statistical significance at the 0.01 level. Thus, a 10% increase in the corporate tax burden would decrease the number of CEM establishments by 0.8%. The result nearly remains the same with different controls and a three-period lagged corporate tax rate shown in Models 6–8. Therefore, our main finding on corporate taxation is that higher corporate taxation in a metropolitan area has a strong negative effect on the probability of location choice of CEM establishments in that area.

In order to further check whether the results in Table 3 are sensitive to model specifications and endogeneity issues such as reverse causality, we conduct fixed effect estimate for robustness checks in Table 4. Models 1–3 have each of the three variables of interest (vertical linkage, horizontal linkage and tax) one-period lagged, respectively, and Model 4 has all three variables one-period lagged. Model 5 extends Model 4 by having an alternative wage measurement. All five models in Table 4 show similar results with those of Model 5 in Table 3. In addition, given our narrow definition of the CEM sector, the intensity of spatial concentration of vertical and horizontal manufacturing sectors and number of establishments in those industries are far stronger and higher than the spatial concentration of the CEM sector. The related manufacturing industries have outnumbered the CEM sector in number of establishments as shown in Table 1, which could reduce the possibility of reverse causality. A typical area such as the San Jose-Sunnyvale-Santa Clara metropolitan area in California had 319 establishments of NAICS3344 but only 73 CEM establishments in 2010.

Table 4.

Panel data fixed effect estimates for robustness checks.

Variables	Model 1	Model 2	Model 3	Model 4	Model 5
laMFGUP_t–1	0.0802*			0.0887**	0.0950**
	(0.0412)			(0.0415)	(0.0415)
laMFGIM_t–1		0.0342		0.0310	0.0322
		(0.0406)		(0.0405)	(0.0405)
CORPTAX_t–1			−0.0697***	−0.0705***	−0.0745***
			(0.0176)	(0.0176)	(0.0175)
laMFGUP		−0.0123	−0.0228
		(0.0425)	(0.0427)
laMFGIM	0.0760*		0.0925**
	(0.0428)		(0.0433)
CORPTAX	−0.0738***	−0.0763***
	(0.0150)	(0.0150)
laMFGSVS	0.475***	0.464***	0.433***	0.451***	0.472***
	(0.136)	(0.136)	(0.137)	(0.137)	(0.136)
WAGERT	−0.0349	−0.0400	−0.0424	−0.0375
	(0.0280)	(0.0280)	(0.0281)	(0.0282)
laEQPWL	0.148***	0.159***	0.157***	0.155***	0.155***
	(0.0324)	(0.0324)	(0.0326)	(0.0325)	(0.0325)
laAIRTSPT	0.114***	0.120***	0.122***	0.123***	0.123***
	(0.0303)	(0.0303)	(0.0304)	(0.0303)	(0.0303)
laFINANCE	−0.0138	−0.000448	−0.0142	−0.0158	0.00128
	(0.0585)	(0.0585)	(0.0590)	(0.0590)	(0.0581)
laICT	0.0436*	0.0456*	0.0500**	0.0499**	0.0541**
	(0.0247)	(0.0247)	(0.0247)	(0.0247)	(0.0247)
laUNIV	0.0345	0.0345	0.0362	0.0285	0.0224
	(0.0411)	(0.0412)	(0.0413)	(0.0413)	(0.0414)
laWAGEAV					0.0267
					(0.0453)
Constant	−1.981***	−1.871**	−1.755**	−1.878**	−2.196***
	(0.747)	(0.747)	(0.776)	(0.777)	(0.777)
Observations	1512	1512	1512	1512	1512
R–squared	0.109	0.104	0.100	0.101	0.100
Number of MAs	216	216	216	216	216

Notes: Standard errors in parentheses.

***

p < 0.01, **p < 0.05, *p < 0.1.

As explained earlier, we use the average state corporate tax rate as the tax variable for a metropolitan area in that state. It is unlikely that there is a reverse causality issue related to the tax variable. In addition, we use the three-period lagged corporate tax rates to represent the tax burden as an alternative that is plausibly exogenous. The lagged corporate taxation variable is still statistically significant with a negative sign and with similar magnitude shown in Model 8; which is consistent with the result of Model 5 in Table 3. The comparisons indicate that no statistically significant difference exists between baseline model and alternative models, thus ensuring that the change in the number of CEM establishments is caused only by the exogenous variables. In summary, our robustness checks have shown consistent signs and statistical significances with the results of Table 2.

Conclusion

This article examines the influence of industrial linkages and corporate taxation on the locational choices of CEM sector establishments. In this paper, industrial linkages have been further categorised as vertical linkages or intra-industry linkages to take account of localisation economies, and horizontal linkages or inter-industry connections to capture urbanisation economies. We show that the input-output connections of the CEM sector are geographically localised. The estimate results confirm that the level of supplier access in a metropolitan area influences the degree of spatial agglomeration of CEM activity. This finding indicates that spatial proximity still matters for vertical linkages between firms because close to the intermediate inputs, the CEM establishments could benefit from externalities such as cheap raw materials, multiple choices for supplier selection and the sharing of intangible information.

In contrast to previous empirical literature (e.g. Amiti, 2005; Amiti and Cameron, 2007 for vertical linkages only), we also consider inter-industry horizontal linkages and find a positive relationship between the industrial machinery manufacturing and the CEM sector. This confirms the role of high-tech manufacturing base and diversity in influencing the location choice of CEM. The concentration of related high manufacturing industries in an area is likely to attract firms that are engaged in related production activity.

Besides the impact of industrial linkages, this research reveals that low corporate taxes also have an important and significant impact on the decision of where to locate CEM establishments. The result illustrates that the agglomeration economies generated by industrial linkages cannot dilute the impact of tax differentials on firms’ location choices.

The dependent variable, number of CEM establishments, is assumed to be distributed discretely and its values to be non-negative integers. The CEM establishment data is over-dispersed because many MAs only have a few CEM establishments while a few MAs have many CEM establishments. The negative binomial regression model can fit the data better than the Poisson regression model in dealing with over-dispersion, which may lead to unobserved heterogeneity that cannot be captured by the Poisson model (Hilbe, 2007). To address possible identification concerns, we have further adopted several approaches by using subsample observations for the analysis, fixed effect panel data models and one-period lagged variables. These alternative approaches have produced results consistent with those of the negative binomial regression analysis. Therefore, the negative binomial model fits the count data well and holds promise for applications on research focusing on other industrial sectors for its simplicity and directness.

Overall, the results confirm that the geographical agglomeration of CEM sector establishments is remarkable and is sensitive to the strong presence of related manufacturing activity and corporate taxation. Both the intra-industry (localisation) and inter-industry spillovers (urbanisation) show a significant positive impact on such location, while higher corporate tax rates are likely to deter the locational selection of CEM industry investments. The novelty of this article lies in the consideration of the geographical significance of industry production chains and the detailed analysis of the effect of corporate taxation on CEM location choice while controlling for a number of factors discussed in the literature. In addition, while industrial linkages and taxes in general have been investigated for overall locational decisions, it was not clear that the CEM sector was sensitive to them. This paper reveals the significant influence of those factors on the CEM investment by focusing on the sector based on detailed metropolitan area level data.

The externalities and strength of spatial vertical linkages have important policy implications for regional development. These results have more general policy implications. Given the US experience with the CEM sector, there are some important lessons for developing countries such as China, where manufacturing dominates the economy. In order to attract investments on high-tech manufacturing such as CEM and eventually form a spatial agglomeration (cluster) of high-tech industries, policymakers should give priority to the strengthening of a manufacturing base, encourage the rapid growth of a large number of suppliers, and reduce the corporate tax burden, either at the metropolitan or at the state level.

Future research can further examine the impact of localisation and urbanisation on firms’ location choice through different proxies. Alternative measurement on CEM investment or employment, and payroll tax at the firm level can be further used to test the significance of the industrial linkages or interconnection and tax effects.

Footnotes

Acknowledgements

We thank the managing editor and five anonymous reviewers for their helpful suggestions.

Funding

The research is supported by the National Natural Science Foundation of China (41301631), Fundamental research funds for the central universities (JBK130148; JBK140402), and the Funds for overseas talents.

Notes

References

Alcacer

Chung

(2007) Location strategies and knowledge spillovers. Management Science 53: 760–776.

Amiti

(2005) Location of vertically linked industries: Agglomeration versus comparative advantage. European Economic Review 49: 809–832.

Amiti

Cameron

(2007) Economic geography and wages. The Review of Economics and Statistics 89: 15–29.

Audretsch

(2003) Innovation and spatial externalities. International Regional Science Review 26: 167–174.

Baldwin

Krugman

(2004) Agglomeration, integration and tax harmonisation. European Economic Review 48: 1–23.

Barrios

Huizinga

Laeven

. (2012) International taxation and multinational firm location decisions. Journal of Public Economics 96: 946–958.

Beardsell

Henderson

(1999) Spatial evolution of the computer industry in the USA. European Economic Review 43: 431–456.

Brulhart

Jametti

Schmidheiny

(2012) Do agglomeration economies reduce the sensitivity of firm location to tax differentials? The Economic Journal 122: 1069–1093.

Cainelli

Iacobucci

(2012) Agglomeration, related variety, and vertical integration. Economic Geography 88: 255–276.

10.

Cameron

Trivedi

(1986) Econometric models based on count data: Comparisons and applications of some estimators and tests. Journal of Applied Econometrics 1: 29–53.

11.

Cameron

Trivedi

(2010) Microeconometrics Using Stata. College Station, TX: Stata Press.

12.

Carlton

(1983) The location and employment choices of new firms: An econometric model with discrete and continuous endogenous variables. The Review of Economics and Statistics 65: 440–449.

13.

Cheng

Stough

(2006) Location decisions of Japanese new manufacturing plants in China: A discrete-choice analysis. Annals of Regional Science 40: 369–387.

14.

Cohen

Morrison Paul

(2005) Agglomeration economies and industry location decisions: The impacts of spatial and industrial spillovers. Regional Science and Urban Economics 35: 215–237.

15.

Coughlin

Segev

(2000) Location determinants of new foreign-owned manufacturing plants. Journal of Regional Science 40: 323–351.

16.

Devereux

Griffith

Simpson

(2004) The geographic distribution of production activity in the UK. Regional Science and Urban Economics 34: 533–564.

17.

Devereux

Griffith

Simpson

(2007) Firm location decisions, regional grants and agglomeration externalities. Journal of Public Economics 91: 413–435.

18.

Dumais

Ellison

Glaeser

(2002) Geographic concentration as a dynamic process. The Review of Economics and Statistics 84: 193–204.

19.

Feld

Kirchgassner

(2003) The impact of corporate and personal income taxes on the location of firms and on employment: Some panel evidence for the Swiss cantons. Journal of Public Economics 87(1): 129–155.

20.

Feldman

Audretsch

(1999) Innovation in cities: Science-based diversity, specialization and localized competition. European Economic Review 43: 409–429.

21.

Frenkel

(2001) Why high-technology firms choose to locate in or near metropolitan areas. Urban Studies 38: 1083–1102.

22.

Glaeser

Kallal

Scheinkman

. (1992) Growth in cities. Journal of Political Economy 100: 1126–1152.

23.

Greenstone

Hornbeck

Moretti

(2010) Identifying agglomeration spillovers: Evidence from winners and losers of large plant openings. Journal of Political Economy 118: 536–598.

24.

Guimaraes

Figueiredo

Woodward

(2004) Industrial location modelling: Extending the random utility framework. Journal of Regional Science 44: 1–20.

25.

Haddad

Taylor

Owusu

(2010) Locational choices of the ethanol industry in the Midwest corn belt. Economic Development Quarterly 24: 74–86.

26.

Hanson

Rohlin

(2011) Do location-based tax incentives attract new business establishments? Journal of Regional Science 51: 427–449.

27.

Hausman

Hall

Griliches

(1984) Econometric models for count data with an application to the patents-R&D relationship. Econometrica 52(4): 909–938.

28.

Helsley

Strange

(2007) Agglomeration, opportunism and the organization of production. Journal of Urban Economics 62: 55–75.

29.

Henderson

(2003) Marshall’s scale of economies. Journal of Urban Economics 53: 1–28.

30.

Hilbe

(2007) Negative Binomial Regression. Cambridge: Cambridge University Press.

31.

Holmes

(1998) The effect of state policies on the location of manufacturing: Evidence from state borders. Journal of Political Economy 106: 667–705.

32.

Isserman

Westervelt

(2006) 1.5 million missing numbers: Overcoming employment suppression in County Business Patterns data. International Regional Science Review 29: 311–335.

33.

Lee

(2008) Geographic redistribution of the U.S. manufacturing and the role of state development policy. Journal of Urban Economics 64: 436–450.

34.

Lucky

Eisenberg

(2006) Renewing U.S. Telecommunications Research. National Research Council Report, Washington, D.C.: The National Academies Press.

35.

Marshall

(1920) Principles of Economics. London: MacMillan and Co., Ltd.

36.

Porter

(1990) The Competitive Advantage of Nations. Free Press: New York.

37.

Puga

Venables

(1996) The spread of industry: Spatial agglomeration in economic development. Journal of the Japanese and International Economics 10: 440–464.

38.

Rathelot

Sillard

(2008) The importance of local corporate taxes in business location decisions: Evidence from French micro data. Economic Journal 118: 499–514.

39.

Strauss-Kahn

Vives

(2009) Why and where do headquarters move? Regional Science and Urban Economics 39: 168–186.

40.

Syverson

(2004) Market structure and productivity: A concrete example. Journal of Political Economy 112: 1181–1222.

41.

Wooldridge

(2013) Introductory Econometrics: A Modern Approach. 5th edition. South-Western Cengage Learning: Florence.

42.

World Bank (2009) World Development Report 2009: Reshaping Economic Geography. Washington, D.C.: World Bank.

43.

Zhang

Guldmann

(2009) Estimating suppressed data in regional economic databases: A goal-programming approach. European Journal of Operational Research 192(2): 521–537.