What Factors Affect Information Technology Firm Location Choices in Middle America? An Examination of Regional and Industrial Variation in Kansas

Abstract

The importance of information technologies in regional economic performance is widely accepted but only widely studied in more populous regions in the United States. Local and state development practitioners and policy makers actively court and promote Information Technology (IT)—intensive industries in rural areas. Despite various promotional efforts, relatively little is known about the specific location requirements of such industries, however, especially in lower population areas of the United States such as the mostly rural Midwest. In this article, the authors examine the location choice preferences of firms in three IT industry sectors in metro, metro-adjacent, and nonmetro regions using a highly detailed data set of Kansas industries. The modeling framework combines both conditional logit and Poisson regression models. The results indicate that average establishment size, industry clustering, labor intensity, and county employment growth explain variations in industry location choice preferences across regions.

Keywords

urban and regional economic development economic growth and development policy and applications urban and regional spatial structure spatial structure other urban and regional spatial structure regional econometric models economic analysis methods location models spatial analysis

Introduction

Technology has long been considered a critical factor for economic development and a competitive economy. National, state, and local governments devise strategies to promote the use of technology in economic development activities hoping that the contribution of such technology will lead to innovation and technical change in the production and distribution of goods and services. The performance of the technology is considered to be “path dependent” having a set of norms, standards, processes, and procedures leading from potential applications of technology to the solution of a specific market need, while meeting institutional socioeconomic constraints (Corona, Doutriaux, and Mian 2006, 6). This path-dependent nature of the technology has helped catalyze economic activities in the form of science parks, innovation centers, Information Technology (IT) clusters, and technology incubators. Technology-based centers often are publicly funded with the objective to create value through synergy and diffusion or spillover of the technology to other regions. In contrast, some regions in the United States, especially in the less populated Midwest, are without any formal technology-based centers, yet a growing number of technology-based firms locate in the region. In Kansas, for example, there are no major science parks, innovation centers, IT clusters, or technology incubators. However, between 1990 and 2000, IT-producing industries in Kansas grew by 174 percent, 20 percent, and 4 percent in total number of establishments in metro, metro-adjacent, and nonmetro regions, respectively. A recent example is Google’s 2011 announcement of Kansas City, Kansas, as the location for its much sought after gigabyte fiber optic network. Such growth and interest suggest that non-technology center regions also may be attractive to technology-based firms. Yet, while research exists on the location choice of IT firms mostly in the East and West coast regions of the United States, little analysis has been performed for the center of the country. The goal of this article is to examine the location choice decisions of IT firms in a part of the country that has a great deal of rural area and is not typically considered an active IT region. As a case study for this area, we examine the location choices of IT firms in Kansas.

Industry location modeling has been a focus both within academic circles and among state and local policy makers. Researchers have attempted to explain why certain regions are relatively more successful in attracting business. For those interested in rural economic welfare, the fact that larger cities and states with high population densities, concentrations of manufacturing and services business, and the ability to provide incentives and other financial and nonfinancial support are able to attract more businesses than smaller cities and states is nontrivial (e.g., Kim, Barkley, Henry 2000). On the surface, the relative size of the economy would seem crucial for industrial location choice preference. However, this may not hold true for all types of industries. Perhaps, rural areas or smaller cities also have certain characteristics that can successfully attract certain types of industries. For example, rural communities have lower land costs, building costs, housing prices, labor costs, security costs, parking costs, and taxes (Isserman 2001, 45), all of which may be attractive to manufacturing industries. Further, with transportation infrastructure development, advances in telecommunications and data transfer technologies, and the adoption of innovative production and service practices in response to global economic competition, many of the traditional disadvantages of the “rural penalty” (i.e., remoteness) may have weakened. Within this context, we explore whether industry location choices are specific to an economic geography (metro/urban or nonmetro/rural), industry-specific, or some combination of both for a region of the United States that has a large rural area and few of the traditional offerings seen as important for IT investment.

The private sector’s heavy investments and innovations in the telecommunication sector in the 1960s helped to modernize and improve the efficiency of telecommunication networks in the United States (Cronin, Herbert, and Colleran 1992). Some regions effectively made use of these telecommunication infrastructure developments with complementary economic development policies. Many states initiated incentive-based economic development programs targeting specific industries in specialized regions. These incentive-based programs evolved in four waves. The first wave of programs attracted firms from industrial zones to growing regions (Bradshaw and Blakely 1999), while the second wave focused on high-growth firms by providing capital, technology transfer, and workforce development (Ross and Friedman 1990). The third wave of programs had a broader scope in developing human capital, improving income and quality of life (Fosler 1992). According to Fosler (1992), the third wave of programs was “concerned with the way in which workers and businesses interact in networks and clusters. And they are interested in the dynamics among those economic entities and related social and political institutions within the context of specific regions and communities” (p. 5).

The fourth wave of strategies focused on creating value from human capital (entrepreneurship) and clustering technology-based economic development to maximize output among the competing and complementary firms. The famous “Silicon Valley” in northern California, Rout 128 in Boston, and the “Silicon Hills” of Austin are some of the most successful and well-known IT regions that have benefited from this fourth wave of development strategies. The states benefiting from the new wave of economic development strategies were those with higher human capital and larger populations.

The potential for economic development opportunities may be lost as a result of overlooking the desirability of rural areas. In fact, some rural areas may have more potential than urban areas. For example, Hamrick (2001) found that following the 1990-1991 recession, rural labor markets recovered much faster in terms of income and employment growth. Further, in an earlier study, Hamrick (1997) found that “the nonmetro labor market leads metro areas in responding quickly to business cycle movements. Indeed, the nonmetro labor series may be a leading indicator for the metro labor series” (p. 15).

In terms of IT specifically, the relationship between technology and rural development is largely based on anecdotal reports (Steinberg 2003) and most of the studies (Oliner and Sichel 2000) used aggregate data or focused on economic growth in general. Utilizing aggregate data analysis, it is more difficult to establish a clear relationship between IT and economic growth. Micro-data analysis will yield more tenable results. Among the few employing micro-data analysis, Lentz and Oden (2001) examined the relationship between high-technology industries and economic growth in the rural Mississippi delta region. They found that the sparse number of telecom manufacturing and service firms in the area created challenges to economic development. If generally true for other areas of the country, such research suggests that the economic development policies on technology-based industry development in rural areas become more meaningful only when the policy makers know why such firms choose one region or another.

This study contributes to the literature in three ways. First, we investigate the location choice of three technology-intensive industries, namely, IT-producing, IT-using, and E-commerce-intensive industries in Kansas. Modeling the location choice of firms in these industries will advance our understanding of the factors likely to influence their choices, thereby helping inform policies that increase communities' competitiveness in attracting a share of this growth. Second, this study examines three distinct economic regions: metro, metro-adjacent, and nonmetro areas. Mapping the location choice of firms in the rural—urban continuum enriches our understanding of both industry needs and regional development prospects. Many earlier studies of location choice focus on aggregate data (e.g., state level), which requires restrictive assumptions and can foster misleading conclusions. Disaggregate data analysis provides a more accurate picture of regions by matching a firm’s decision choice to a smaller geographical unit, thereby helping to answer why some regions are more attractive to firms than others. Finally, by focusing on an area of the country heretofore ignored in IT-location research, the study will provide greater insight into what works and what does not work in these policies.

The remainder of the article is divided into four sections. Industry Location Choice Modeling section discusses the econometric models employed in the location choice decisions. Data and Variables section describes the data, and the results are presented in the Results and Discussion section. The concluding section presents the policy implications and conclusions.

Industry Location Choice Modeling

Traditional approaches to location choice modeling tended to focus on individual industries in urban/metro (Guimarães, Figueirdo, and Woodward 2004) or rural/nonmetro settings, while others focused on location choice and foreign direct investment (Chung and Alcácer 2002). Given the discrete nature of choice decisions, count data models are commonly used in empirical studies. For example, probit and ordered probit (Basile, Giunta, and Nugent 2003), tobit (Devereux, Griffith, and Simpson 2003), logit and multinomial logit (Gunther et al. 1998), conditional logit (Guimarães, Figueirdo, and Woodward 2004), negative binomial (Coughlin and Segev 2000), Poisson (Guimaraes et al. 2004), and nested logit (Hansen 1987) models have all been used in empirical work.

In classical economic theory, a rational agent of a firm chooses a location or locations that yield the maximum profit for the firm’s operations. However, it is not hard to think of other considerations that may not be profit maximizing but are nonetheless “utility maximizing” and, hence, one can model a monotonic transformation function of profit where firms choose a location by optimizing this function (i.e., the utility of profit and other factors). The economic objective (profit), firm input requirements, proximity to market, proximity to amenities, and other location characteristics are some of the factors affecting location choice decisions.

While McFadden’s (1973) paper on the Random Utility Maximization (RUM) approach provided a foundation for discrete choice modeling, Carlton’s (1979, 1983) application of RUM theory on location choice modeling had greater influence on contemporary location choice modeling. Apart from firm and industry location choice, location choice modeling has been used to study residential location decisions in cities (White 1988), interstate migration patterns (Davies, Greenwood, and Li 2001), and trends in occupational mobility (Dessens et al. 2003).

Location choice models typically begin with the assumption that firms choose locations that maximize their profit or some utility that is in turn a function of profit. Let $π_{i j}$ denote this profit function representing the optimization from firm i’s choice of location j. As such, one should expect location j to be chosen over location k whenever $π_{i j} > π_{i k}$ . However, because of imperfect foresight and/or imperfect measurement, this relationship is bound to be measurable with some error. As such, we can define Equation 1

U_{i j} (π_{i j}) = Z_{i j} (π_{i j}) + ϵ_{i j},

to represent this error-in-measurement function and maintain the assumption that, on average,

π_{i j} > π_{i k}

whenever

U_{i j} (π_{i j}) > U_{i k} (π_{i k})

. Equation 1 also shows that firms that maximize profits also maximize a firm’s utility as well. This equation is likely to hold true where the managers' discretion is limited to a set of decisions and actions that maximize the firms' profit (Misangyi 2002). Further, an increasing number of high-tech firms are also choosing locations that maximize the satisfaction of managers and employees. Goetz and Rupasingha (2002) found that high-tech firms choose amenity-rich regions to accommodate skill workers preferences (Dorfman, Patridge, and Galloway 2008). As is common, we also assume that

Z_{i j} (π_{i j})

is a linear function that may be written as:

Z_{i j} (π_{i j}) = α + β^{'} X_{i j} .

Here,

α

is a parameter and

β

is a parameter vector, each to be estimated,

X_{i j}

is a vector of explanatory variables composed of three categories

X_{i j} \equiv 〈{I N D}_{i j}, {E S T}_{i}, {C O M}_{j}〉

. Specifically, IND_ij is a vector of characteristics particular to the industry of firm i in location j; EST_i is a vector of variables particular to the establishment (firm i) itself and, hence, identical for each location choice, and COM_j is a vector of characteristics particular to the location (i.e., community/region) choice and, hence, the same for all firms, regardless of industry but different for each location.

If the error term in Equation 1 is assumed to be independently and identically distributed (IID) according to the Weibull distribution (see the discussion in Greene 2000, 858), then letting $Y_{i}$ be a random variable indicating the location choice of firm i among j = 1, …, J locations, the probability of firm i choosing location j is $p_{i j}$ = Prob( $Y_{i} = j$ ). This leads to the well-known McFadden (1973) conditional logit model (CLM):

p_{i j} = \frac{exp [α + β^{'} X_{i j}]}{\sum_{j = 1}^{J} exp [α + β^{'} X_{i j}]} .

This model, which also nests the simpler (binary) logit, has been used often in firm location-choice models dating back to the earliest of these in Carlton (1979, 1983) and Bartik (1985).

The CLM of Equation 3 has been widely used in industry location choice modeling based on the merit of linking economic theory to empirical observations (Guimarães, Figueirdo, and Woodward 2004). A well-known feature of the model is that attributes that are firm specific (EST_i) cancel out of the model allowing researchers to focus on the effect of variables particular to the location rather than the firm.¹ In other words, $X_{i j} \equiv 〈{I N D}_{i j}, {C O M}_{j}〉$ since EST_i is a vector common to all location choices. Following the traditional CLM approach, the parameters are estimated through maximization of the likelihood function, $L = \prod_{i = 1}^{I} \prod_{j = 1}^{J} p_{i j}^{y_{i j}}$ where $y_{i j} = 1$ if $Y_{i} = j$ and $y_{i j} = 0$ otherwise.

Unfortunately, researchers in location choice modeling using CLM often confront the Independence of Irrelevant Alternatives (IIA) assumption. Under an IIA assumption, a firm’s location choice within a region is independent of the decision to have chosen another location (e.g., what if A and B are both located in a desirable region, so that the choice of A and the choice of B are conditional on, and therefore not independent of, the choice of the region?). Location choice modelers often find that the CLM implication that choice A really is independent of choice B likely does not hold. Further, the more narrowly defined the alternatives, the more likely it is one will violate the IIA assumption leading to biased coefficient estimates (Guimarães, Figueirdo, and Woodward 2004). One way of overcoming the problem would be to use a nested logit model where the choices are conditional on, perhaps, a regional choice. But, nested logit, too, has its drawbacks as the regional choices may not be obvious and need not be the same for every firm, requiring testing and retesting of the combinations of potential nesting structures to find the structure with the best fit. And for a multitude of choices, this technique becomes time-consuming even for the most powerful computers. Another alternative free of the IIA problem is a multinomial probit, though this technique, too, becomes unwieldy as the number of choices grows.

Guimarães, Figueirdo, and Woodward (2003) provide a solution to the problem inherent in the IIA assumption while avoiding the cumbersome calculations necessitated by other techniques. Guimarães, Figueirdo, and Woodward (2003) prove that an unbiased coefficient can be estimated using a Poisson regression model and that this coefficient is equivalent to the coefficient from a CLM using maximum likelihood estimation. If $y_{j} = \sum_{i = 1}^{I} y_{i j}$ is the number of firms in an industry appearing in location j, then the Poisson model estimates the probability of observing Y firms from this industry in location j:

P r o b (Y = y_{j}) = \frac{e^{- γ_{j}} γ_{j}^{y_{j}}}{y_{j}!} y_{j}, Y = 0, 1, 2, 3, \dots .

Guimarães, Figueirdo, and Woodward (2003) show that the parameterization

γ_{j} = e^{α + β^{'} X_{i j}}

produces the equivalent estimates for

α

and

β

from Equation 3 without the IIA issues.

In their follow-up 2004 article, the authors demonstrate the technique by modeling U.S. manufacturing firms' location choice decisions using industry data from 1989 to 1997. In our research, we apply these techniques to the location choice of firms in IT-producing, IT-using, and E-commerce industries in metro, metro-adjacent, and nonmetro regions of Kansas. Specifically, we first estimate a Poisson regression according to Equations 4 and 5 of the number of firms in each of these industries choosing a particular location. The coefficients from the Poisson regression are then used in the CLM of Equation 3 to predict the location-choice probabilities and the marginal effects of a change in a variable on these probabilities. Specifically, the probabilities are given by Equation 3 and the elasticities are determined by $\frac{\partial ln p_{i j}}{\partial ln X_{i j}} = (Ω (α + β^{^{'}} X_{i j})) (1 - Ω (α + β^{^{'}} X_{i j})) β^{^{'}} X_{i j} / p_{i j},$ where $Ω (α + β^{^{'}} X_{i j})$ is the probability density function of the CLM using the parameter estimates from the Poisson model. (We refer the reader to the detailed discussions on analytical tractability between the CLM and Poisson model in Guimarães, Figueirdo, and Woodward 2003, 2004.) The estimated parameters and their significance levels reveal the variables that are likely to influence the location-choice preference of firms and the elasticities show the direction and magnitude of the effects on the choice probabilities from a one-percent change in the independent variables.

Data and Variables

In our model, we specifically examine the location choices of firms in three IT industries in Kansas. The first category comprises Information Technology-Producing (IT-Pro) industries. IT-Pro industries are defined by the U.S. Department of Commerce (DOC) as industries that either (a) produce, process, or transmit information goods or services as either intermediate or final products or (b) provide the necessary infrastructure for the Internet (Henry et al. 1999). Thirty four-digit SIC industries were selected as IT producers based on the DOC criteria. According to the DOC, Information Technology-Using (IT-Use) industries are defined as those characterized by either (a) the value of the industries' IT capital stock being equal to or more than 50 percent of the value of its total equipment stock or (b) the industry’s IT investment expenditures per worker are equal to or more than $10,000. A total of forty-seven four-digit SIC IT-using industries were included in this second category. Finally, the DOC identifies an E-commerce-intensive industry (Ecom) as one in which E-commerce revenues (sales or shipments) exceeds 15 percent or more of the total revenues (sales or shipments; Buckley, Henry, and Gurmukh 2000). A total of 101 four-digit E-commerce-intensive industries were identified for this category. The industry SIC codes are shown in Table 1 , and fully identified in the Appendix.²

Table 1.

IT- and E-Commerce Industry SIC Codes.

	Industries
	IT-Producing	IT-Using		E-commerce
	SIC Code	SIC Code	SIC Code	SIC Code	SIC Code	SIC Code
1	3571	4612	8049	2082	2392	3669
2	3572	4613	8051	2083	2393	3671
3	3575	4619	8052	2084	2394	3672
4	3577	4812	8059	2085	2395	3674
5	3578	4813	8062	2086	2396	3675
6	3579	4822	8063	2252	2397	3676
7	3651	4832	8069	2253	2399	3677
8	3661	4833	8071	2296	2673	3678
9	3671	4841	8072	2298	3111	3679
10	3672	4899	8082	2299	3131	3695
11	3674	6111	8092	2311	3142	3699
12	3679	6141	8093	2321	3143	3711
13	3695	6153	8099	2322	3144	3713
14	3825	6159		2323	3149	3714
15	3826	6162		2325	3151	3715
16	4812	6163		2326	3161	3716
17	4813	6211		2329	3171	3721
18	4832	6221		2331	3172	3724
19	4833	6231		2335	3199	3731
20	4841	6282		2337	3571	3732
21	5045	6289		2339	3572	3743
22	7371	7812		2341	3575	3751
23	7372	7819		2342	3577	3761
24	7373	7822		2353	3578	3764
25	7374	7829		2361	3579	3769
26	7375	7832		2369	3641	3792
27	7376	7833		2371	3643	3795
28	7377	7841		2381	3644	3799
29	7379	8011		2384	3645	4724
30	8249	8021		2385	3646	4725
31		8031		2386	3651	5013
32		8041		2387	3652	5122
33		8042		2389	3661	5961
34		8043		2391	3663

The Office of Management and Budget (OMB), classifies geographic areas as Metropolitan Statistical Areas (MSA) or nonmetropolitan areas. One or more cities with a population greater than 50,000 are defined as an MSA while the other areas are classified as nonmetropolitan (OMB 2000). In this research, the county or counties in an MSA were identified as metro counties. Counties adjacent to a metro county were grouped as metro-adjacent counties. All other counties were classified as nonmetro, nonadjacent counties. As shown in Figure 1, there are seventeen metro counties, twenty-one metro-adjacent counties, and sixty-seven nonmetro counties in Kansas.

Kansas is a good candidate for a study of how firms in IT industries choose their locations. First, Kansas' three largest counties, Johnson (population 451,086), Shawnee (169,871), and Sedgwick (452,869), are located relatively far from one another so that a firm’s choice to be in or near a large metropolitan area would not likely be blurred if any two of these counties were adjacent. Second, for a mid-sized state (in geographic area) and a low-to-mid-sized state in total population, Kansas has the sixth highest total number of counties of any state. Thus, a county-level analysis of Kansas' 105 counties can be considered a rather fine unit of analysis. Finally and related, because Kansas, which became a state in 1861, developed very quickly in the period leading up to and just after the U.S. Civil War, the legislative divisions of its regions produced a geographic regularity among its counties. The boundaries for the 105 counties were drawn in the short time period between 1855 and 1888, and a map of Kansas looks like a rectangular checker board with counties of roughly equal area. Although the smallest county (Wyandotte) is 155 square miles and the largest (Butler) is 1,428 square miles, the average Kansas county is 779 square miles (median of 730) with a standard deviation of 214 square miles. In fact, only four of Kansas' 105 counties are greater or less than twice the standard deviation from the average. The benefit of such regularity, along with the large number of units examined, is that variables measuring the distance between one county and another or examining the makeup of adjacent counties can be useful measures.

The Kansas Quarterly Census of Employment and Wages (commonly known as ES-202) was used to determine the industry location choices of existing and newly appearing firms from 1990 to 2002. The Bureau of Labor Statistics of the U.S. Department of Labor and the State Employment Security Agencies jointly collect the fully disclosed firm-level employment and wage information for workers covered by Unemployment Insurance laws and federal workers covered by the Unemployment Compensation for Federal Employees program. The ES-202 data consist of monthly establishment-level employment, SIC code, county location, establishment start year, a code identifying the members of multi-establishment firms, a code identifying changes in company business names, and quarterly salaries and wages subject to unemployment compensation insurance reporting requirements. The firm-level employment and location information in the ES-202 system provides a very detailed and robust dataset for the research.

The dependent variable in an industry was the number of establishments in a given county. The number of establishments observed in a given county in a given year is considered a Poisson random variable (stochastic process) and assumed to be independent of other counties and years. As such, the total number of observations for an industry (e.g., IT-producing) within a particular region (e.g., metro) depended on whether industry firms were present in some number of counties during the study period. For example, there were seventeen counties in the metro region and IT-producing firms were present in all seventeen counties for all the years of the study period (thirteen). The total observations for the IT-producing industry in the metro region would be equal to number of metro counties multiplied by the number of years. In this case, 17 counties times 13 years equals 221 observations. For the IT-Pro category, the data consist of 1,327 observations with 221 of these in metropolitan areas, 263 in metro-adjacent areas, and 842 in nonmetropolitan areas. For the IT-Using category, the data consist of 1,365 observations with 221 of these in metropolitan areas, 273 in metro-adjacent areas, and 871 in nonmetropolitan areas. Finally, for the Ecom category, the data consist of 1,110 observations with 212 of these in metropolitan areas; 250 in metro-adjacent areas, and 648 in nonmetropolitan areas.

Most of the previous location choice research focused exclusively on new firm entry. Under the random utility maximization assumption, firms choose locations that maximize their profit/utility. For example, two firms, A and B, operate in location J. Firm A migrates to location K, while firm B continues to operate in location J. In the former case, firm A is considered as a new entry in location K, while firm B is omitted in the location choice modeling. While firm B had an option to migrate, its continuing operation in location J indicates that firm B is able to maximize its profit/utility in location J. Therefore, it is logical to include firm B in the location choice decisions. Exclusion of existing firms in location choice modeling may lead to biased estimates.

The independent variables were the county-level annual estimates for all the Kansas counties (105) from 1990 to 2002 (13 years) from the sources cited in Table 2 . Following the discussion in the Industry Location Choice Modeling section, two sets of independent variables drawn from the research literature were selected in modeling the location choice of firms.³ The industry characteristics (IND_ij) are particular to the industry of firm i in county j and include the average industry establishment size (SIZE), industry clustering (CLUSTER), vertical integration (INTEGRATION), labor intensity (INTENSITY), and the average age of the establishments (AGE).

Table 2.

Definitions of Independent Variables and Data Sources.

Variables	Description	Source/s
ESTABLISHMENT	The number of establishments by industry type in each county.	ES-202
CLUSTER	Location quotient (LQ): $L Q = \frac{\frac{C o u n t y i n d u s t r y e m p l o y m e n t}{T o t a l c o u n t y e m p l o y m e n t}}{\frac{N a t i o n a l i n d u s t r y e m p l o y m e n t}{T o t a l N a t i o n a l e m p l o y m e n t}}$	ES-202 and County Business Patterns
SIZE	Average industry establishment size: Total county employment/ Total number of establishments	ES-202
INTEGRATION	Vertical integration: Indirect output multiplier in million dollars	IMPLAN software (MIG, Inc., 1999)
INTENSITY	Labor intensity: Direct employment multipliers in million dollars output per job	IMPLAN software (MIG, Inc., 1999)
AGE	Average establishment age in years	ES-202
LF-QUALITY	Quality of the labor force: Percent of employees in knowledge industries	ES-202 and Beck (1992)
DENSITY	Population density: population per square mile	Woods and Poole Economics, Inc.
CO-EMP	County employment growth in percentage	ES-202
DISTANCE	Distance to adjacent metro county in miles	U.S. Geological Survey
HIGHWAY	Presence of interstate highway in the county	Kansas State Highway Map

Porter’s (1990) seminal work on industry clusters is a systematic approach to examining regional economic geography. In his more recent work, a cluster is defined as “geographic concentrations of interconnected companies, specialized suppliers and service providers, firms in related industries, and associated institutions in particular fields that compete but also cooperate” (Porter 2000, 253). In a given locale, it is possible to find IT-producing industry establishments but weak clustering in the absence of other related industries and associated institutions. There are several methods available to identify the relative degree of regional industry clustering. The most common approaches include the spatial concentration ratio, spatial Hirschman-Herfindahl index, locational Gini coefficient, the Ellison and Glaeser concentration index (Kim, Barkley, Henry 2000), and location quotient (Miller, Gibson, and Wright 1991). The location quotient was used as measure of industry concentration within counties. The location quotient measures the level of employment of a county industry sector compared to the national industry sector employment.

Vertical integration may arise due to variety of economic reasons (improved efficiency, technological economies, etc.). Vertical integration measures both upstream and downstream relationships. These inter- and intra-industry relationships are likely to affect employment levels and product purchasing relationships between sectors. Regional Social Accounting Matrices (SAM) are used to track inter-industry transactions by tracing the purchase of goods and services through the entire chain of production. Input-output models are used to quantify interactions between firms and industries within an economy (Miller and Blair 1985). The IMPLAN (Impact Analysis for Planning) software (MIG Inc. 1999) is capable of capturing those transactions in a region and estimate the direct and indirect impact of an output change. The resulting impacts are measured as output multipliers for given sectors of the economy. The direct output multiplier measures the direct change in output (millions of dollars) per million dollar change in final demand while the indirect multiplier measures the change in final demand resulting from the interaction of local industries purchasing from other local industries. The indirect multiplier was used as a proxy for vertical integration.

The labor requirements and the inter-industry trading relationships to produce goods and services can be quantified in a SAM. There is direct and indirect employment associated with production through a firm’s own value-added activity and through its purchase of intermediate inputs. It is possible to estimate the change in employment (direct or indirect) in a region as a result of changes in output. IMPLAN software (MIG Inc. 1999) generates direct and indirect employment multipliers that capture the direct and indirect employment for a single county, group of contiguous counties or the entire state for any sector or group of sectors. The direct and indirect employment multipliers measure the direct change in employment (jobs) and indirect change in employment (resulting from interactions from local industries) per million dollar change in final demand. Labor intensity for an industry (i.e., IT-Producing) in every county was estimated using IMPLAN software in three steps. First, the direct employment multipliers for all the sectors were estimated. Then, an average multiplier was estimated for a group of sectors (industry) and labor intensity was estimated by inverting the employment multiplier.

The second set represents the community/regional characteristics (COM_j). These characteristics included population density (DENSITY), labor quality (LF-QUALITY), county employment growth (CO-EMP), a proxy measure of remoteness measured as the miles from the center of a county to the center of the nearest metro county (DISTANCE; equal to zero in the case of a metro county), and the presence of an interstate highway (HIGHWAY).

The quality of the labor force reflects the knowledge and skill level of the community and could be an important factor for IT/EC related industry. High technology industrial clusters are a reflection of such phenomena. Firms are likely to move to locations where the required quality of labor is readily available. Conversely, workers with special knowledge and skills tend to migrate or commute to where their knowledge and skills are rewarded accordingly. Other location factors being equal, the quality of the labor force could be a determining factor influencing firm entry. Among manufacturing industries, a common determinant of “high-tech” or “low-tech” is the skill level of the required labor force (Hackler 2003). It is difficult to directly measure the knowledge and skill level of the local labor force. Often, college educational attainment had been used as a proxy (Pigeon and Wray 1999). In this research, the percentage of workers in high-knowledge industries is used as a proxy for quality of the labor force. An industry is identified as a high-knowledge industry if the industry has more than 40 percent of occupations in managerial, professional, and technical positions (Beck 1992).

The description and the source of the variables are presented in Table 2 and data summary statistics are presented in Table 3 . The summary statistics show the mean and standard deviation of the industry variable (e.g., IT-producing firms/establishments) or common variable (e.g., population density) in a particular regional county. For example, the mean of the ITP establishment variable in the metro region was 77.742. This indicates that on average there were 78 establishments in a metro county.

Table 3.

Data Summary Statistics by Region and Industry, 1990-2003.

	Metro		Metro-Adjacent		Nonmetro
	Mean	Std. Dev.	Mean	Std. Dev.	Mean	Std. Dev.
ITP_ESTABLISHMENT	77.742	213.045	7.534	6.929	4.505	4.709
ITP_CLUSTER	0.200	0.043	0.177	0.042	0.194	0.056
ITP_SIZE	9.547	8.479	9.173	7.718	6.690	7.735
ITP_INTEGRATION	0.029	0.055	0.004	0.003	0.003	0.005
ITP_INTENSITY	0.291	0.368	0.206	0.294	0.130	0.317
ITP_AGE	14.217	2.807	18.494	7.325	18.400	6.789
ITU_CLUSTER	2.697	0.252	2.632	0.216	2.680	0.306
ITU_ESTABLISHMENT	216.647	381.743	41.557	33.089	24.607	29.277
ITU_ SIZE	22.597	9.396	19.718	6.564	20.473	6.541
ITU_INTEGRATION	0.053	0.058	0.020	0.013	0.018	0.017
ITU_INTENSITY	0.651	0.570	0.806	0.526	1.155	1.367
ITU_AGE	17.993	2.027	18.970	2.664	18.895	2.695
ECOM_ESTABLISHMENT	51.811	102.785	7.580	6.004	5.111	6.419
ECOM_CLUSTER	0.318	0.225	0.321	0.219	0.340	0.321
ECOM_ SIZE	23.684	29.056	30.281	20.798	12.378	20.383
ECOM_INTEGRATION	0.029	0.041	0.009	0.009	0.005	0.008
ECOM_INTENSITY	3.776	5.146	3.680	2.370	2.306	2.946
ECOM_AGE	16.795	4.138	17.810	4.517	19.851	8.066
LF-QUALITY	35.952	5.589	33.399	8.461	36.927	7.871
DENSITY	196.233	312.512	27.058	25.563	12.353	14.661
CO-EMP	1.314	1.467	0.742	0.882	0.587	2.260
DISTANCE	24.779	7.178	34.784	12.389	134.453	57.976
HIGHWAY	0.706	0.470	0.286	0.463	0.149	0.359

Results and Discussion

The coefficients from the Poisson model (and their standard errors below in parentheses) are presented for each industry in Table 4 . Table 5 presents the elasticities of these variables on the likelihood of a firm in each industry selecting a given region using the CLM technique as discussed in the Industry Location Choice Modeling section.

Table 4.

Poisson Coefficients for IT-Producing, IT-Using, and E-Commerce Industries by Region.

Parameter	IT-Producing			IT-Using			E-commerce
Parameter	Metro	Metro-Adjacent	Nonmetro	Metro	Metro-Adjacent	Nonmetro	Metro	Metro-Adjacent	Nonmetro
Intercept	−0.302 (0.74)	2.703* (0.179)	1.688* (0.133)	2.353* (0.581)	4.595* (0.206)	3.82* (0.143)	−0.693 (0.727)	2.609* (0.279)	2.322* (0.191)
CLUSTER	−0.397* (0.18)	−0.259** (0.136)	−0.020 (0.067)	0.002 (0.022)	−0.002 (0.014)	−0.017** (0.010)	−0.006 (0.081)	−0.054 (0.040)	0.047 (0.037)
SIZE	0.010** (0.005)	0.000 (0.003)	0.000 (0.002)	−0.012* (0.004)	0.016* (0.004)	−0.006* (0.003)	0.001 (0.001)	0.000 (0.001)	−0.004* (0.001)
INTEGRATION	8.992* (0.918)	7.455* (3.149)	−1.429 (1.977)	9.846* (0.823)	4.193* (1.447)	0.070 (0.408)	18.980* (1.213)	17.905* (3.169)	4.232* (1.007)
INTENSITY	0.214* (0.076)	0.058 (0.042)	0.075* (0.031)	−0.92* (0.16)	0.026 (0.030)	−0.015 (0.010)	−0.08* (0.027)	−0.025* (0.010)	0.022* (0.006)
AGE	−0.048* (0.009)	−0.004 (0.003)	−0.009* (0.002)	−0.011 (0.010)	0.000 (0.005)	−0.005 (0.003)	0.033* (0.008)	0.014* (0.004)	0.010* (0.003)
LF-QUALITY	−0.007 (0.014)	−0.048* (0.005)	−0.020* (0.003)	0.008 (0.009)	−0.062* (0.004)	−0.028* (0.003)	0.030* (0.013)	−0.053* (0.007)	−0.041* (0.004)
DENSITY	0.003* (0.000)	0.026* (0.001)	0.028* (0.001)	0.002* (0.000)	0.027* (0.001)	0.032* (0.001)	0.001* (0.000)	0.020* (0.002)	0.028* (0.001)
CO-EMP	0.021* (0.009)	0.001 (0.005)	−0.001 (0.002)	0.023* (0.007)	0.004 (0.005)	0.001 (0.002)	0.045* (0.007)	0.003 (0.006)	0.007 (0.005)
DISTANCE	0.103* (0.016)	0.000 (0.003)	0.000 (0.000)	0.058* (0.011)	−0.005** (0.002)	0.000 (0.000)	0.018 (0.012)	−0.001 (0.003)	−0.002* (0.001)
HIGHWAY	0.915* (0.280)	−0.079 (0.072)	0.414* (0.042)	0.711* (0.188)	−0.283* (0.063)	0.239* (0.039)	1.102* (0.216)	0.013 (0.089)	0.319* (0.051)
Scale	5.492	1.093	1.025	7.38	2.34	2.286	3.116	1.148	1.156
Log Likelihood	2682.035	2110.086	2781.576	4615.218	6124.266	10585.26	4144.319	1622.281	2490.003

Note: * and ** are level of significance at 5 percent and 10 percent level, respectively.

Table 5.

Variable Elasticities for IT-Producing, IT-Using, and E-commerce Industries by Region.

	IT-Producing			IT-Using			E-commerce
	Metro	Metro-Adjacent	Nonmetro	Metro	Metro-Adjacent	Nonmetro	Metro	Metro-Adjacent	Nonmetro
CLUSTER	−0.439	−0.208	−0.006	0.030	−0.024	−0.067	−0.011	−0.082	0.024
SIZE	0.529	0.004	0.001	−1.501	1.431	−0.181	0.131	0.014	−0.074
INTEGRATION	1.453	0.143	−0.007	2.884	0.386	0.002	3.025	0.786	0.032
INTENSITY	0.345	0.054	0.014	−3.318	0.101	−0.025	−1.672	−0.437	0.076
AGE	−3.778	−0.335	−0.243	−1.096	0.009	−0.139	3.068	1.184	0.296
LF-QUALITY	−1.393	−7.271	−1.086	1.592	−9.352	−1.522	5.993	−8.054	−2.237
DENSITY	3.259	3.191	0.509	2.173	3.313	0.581	1.086	2.675	0.523
CO-EMP	0.153	0.003	−0.001	0.167	0.013	0.001	0.409	0.017	0.006
DISTANCE	14.130	0.016	0.020	7.957	−0.789	0.020	2.469	−0.161	−0.400
HIGHWAY	5.066	−0.359	0.608	3.937	−1.283	0.352	6.100	0.062	0.475

Note: Estimates for Highway show the effect of the presence of a highway on the probability of observing a firm locating in the given region, expressed as a percent for consistency.

Consider the results of the Poisson model shown in Table 4. Clustering generally had a relatively weak negative effect across the industries and regions. In short, industry clustering does not appear to be an important factor in firm-choice location for Kansas. Similarly, Arita and McCann (2000) found that while IT clusters are crucial for location in IT specialized regions, clusters are not important in urban regions. For regional planners in states like Kansas, the implication is that a “Silicon Valley” is not necessarily needed to attract IT firms.

Vertical integration generally had a strong positive effect on industry growth across the industries and regions. This would reflect access to noncompetitive input suppliers and output markets. Average establishment age generally had a negative effect on the IT-Producing sector across the regions, but a positive effect for E-commerce-intensive industries.

One interesting finding was the effect of quality of the labor force. Labor force quality does not seem to have any significant impact either on the industries or regions. This could be due to the fact that this variable has very little variance (Table 3) across counties in Kansas and so is not providing much information. Across all regions, population density had fairly consistent positive impacts on IT-Producing location. Transportation via interstate highway was found to be of importance in location choices in the metro and nonmetro regions for all industries, but of questionable importance for metro-adjacent regions and possibly negative for IT-Using, metro-adjacent firms; a curious finding that certainly warrants more analysis. Overall, the variables that seem to play the most important roles would be vertical integration, labor force intensity, population density, and the presence on an interstate highway.

Considering the elasticities of the variables in Table 5, a consistent impact in all three regions and across the industries would appear to be that of population density. Population density had a consistent positive impact on the probabilities of firms from all industries choosing locations in the three regions.⁴ This result is consistent with previous research, where population density was shown to be an important factor for economic development. In our study, the relative importance and size of the effect associated with this variable varied across the regions, where a one-percent increase in population density had similar and consistently more impact on the metro-adjacent and nonmetro regions compared to the metro region, regardless the industry. Across parts of rural America, population out-migration has been a major factor in lagging economic growth. Lichter et al. (2005, 4) asserts, “The statistical association between economic underdevelopment and out-migration has been unmistakable.” Our results support this assertion and also show that the effect is more pronounced outside of metro counties, as would be expected. For example, a one-percent decrease in population per square mile would be predicted to lower the probability of an IT-Producing firm choosing the metro region by 3.3 percent, while for the metro-adjacent and nonmetro regions, the impacts were stronger at 3.2 and 0.5 percent, respectively. Similar stories hold for the two other industries as well.

Considering the vertical integration variable, with the exception of the nonsignificant negative effect in the nonmetro choice for IT-Producing, a one-percent increase in integration had a positive effect on the likelihood of a firm locating in the metro and metro-adjacent regions. Given the size of these elasticities in the metro regions especially, it would be hard to argue that vertical integration is unimportant. One possible reason that vertical integration shows this positive effect is that our indirect economic multiplier measures the flow of goods between industries within a single four-digit sector. Since the measurement was the flow within a single industry, one would expect to find a strong impact, hence firms in IT industries appear to be making their location choices in large measure because of the economic linkages that exist with input providers within a region.

Likewise, the presence of an interstate highway is mostly positive across industries and counties but appears to have the biggest impact on location choice in metro counties. For example, the presence of an interstate highway in a metro region is likely to increase the location choice of IT-producing, IT-using, and E-commerce firms by 5.1 percent, 3.9 percent, and 6.1 percent, respectively. Highways are known to have influence on the location choice of high-tech firms. Route 128 in Massachusetts was a key factor in the formation of Boston’s high-tech industry clusters (Rosegrant and Lampe 1992). In the case of Kansas, we would conclude that for IT industries, the information superhighway is not a substitute for the old-fashioned kind.

Typically, the quality of the labor force is measured by the percentage of the population aged twenty-five or older with some college. Based on our measure of the percentage of workers in knowledge industries, the quality of the labor force had a consistent negative impact in the metro-adjacent and nonmetro regions for the IT-Using and E-commerce-intensive industries. A higher quality labor force in the metro region may attract more ecommerce and IT-using industries, while it does not appear to be a major asset in metro-adjacent and nonmetro regions. Increasing the quality of the labor force may reflect rising labor costs. In a recent study, Wolff (2008) found that high-tech employment salary increased in double-digits for the fifth consecutive year; however, the employment in those jobs has dropped to levels last seen during the mid-1990s. Nonmetro regions are attractive to the firms seeking relatively lower land and labor cost. In the case of our high-tech firms, this may not necessarily hinder growth opportunities in adjacent or nonmetro areas. Such places would need to position themselves for more of the back-office or routine functions within these industry sectors.

Industry clustering generally had a negative impact on IT-Producing, IT-Using, and E-commerce industries in the metro, metro-adjacent, and nonmetro regions. In the IT-Producing industry, the negative probability of establishment location decreased moving from the metro to the nonmetro region as clustering increased. A one-percent increase in IT-Producing industry clustering decreased the probability of an IT-Producing firm choosing the metro region by 0.44 percent but only 0.01 percent in the nonmetro region. This is interesting in light of the fact that vertical integration proved important. Arguably, IT firms in Kansas choose locations based upon the vertically integrated infrastructure as opposed to the horizontal connections that come in clustering. While industry clustering is generally expected to have a positive impact on firms locating within a region, our results suggest that not all clusters are equal. The type of cluster that may matter in this instance is not within one or several closely related industries, but between more disparate but mutually supporting industry sectors as one might find in vertical coordination.

In the metro region, the IT-Producing industry labor intensity variable had a significant positive impact, while having a significant negative impact on the probabilities of metro IT-Using and E-commerce industry firms. It might be speculated that IT-Producers seek locations closer to centers of innovation wherein they enjoy a relatively high return on labor, while IT-Users have no such positive returns. The coefficient on this variable in the Poisson model was generally not significant in the metro-adjacent and nonmetro regions.

The direction and significance of distance to a metropolitan area and presence of an interstate highway for IT-Producing industries in all regions indicates that these firms are more likely to locate farther from the metro region, where an interstate highway is readily available. However, E-commerce industries would seem to prefer locations that have both an interstate highway and are closer to a metro area. Proximity and presence of an interstate highway would likely reduce the cost of the transportation. Transportation costs would seem to be a key factor for E-commerce-intensive businesses, perhaps evidenced by the giant E-commerce retailer, Amazon.com, who opened distribution warehouses throughout the United States to reduce transportation costs.

The impact of average establishment size of industry firms choosing a given region offers interesting results. A negative effect could be interpreted as a region being relatively more attractive to smaller firms, while a positive effect might suggest the opposite. If this is the case, the metro region would seem attractive to larger IT-Producing and smaller IT-Using firms. The attraction of larger IT-Producing firms may be access to economies of scale and human capital in the metro region. On the other hand, small firms in an IT-Using industry may be attracted to the metro region as they compete in the niche markets within a higher density area. A small independent Web-design firm (IT-using), for example, might locate in the metro region, while the Web software developer (IT-producing) might be employed at a larger firm located in a metro region.

Conclusions

This article focused on the location choice decisions of IT-Producing, IT-Using, and E-commerce industries in metro, metro-adjacent, and nonmetro regions of Kansas. Since little research has been conducted concerning IT firm location choice decisions in the Midwestern United States, this article helps enhance our understanding of what matters in a U.S. IT firm’s location choice. A combination of conditional logit and Poisson techniques were used in modeling location choice decisions to examine the influence of variables thought important to IT-producing, IT-using, and E-commerce firms choosing metro, metro-adjacent, and nonmetro regions. Selected industry and community characteristics were used to explain the location choices of firms. Population density, vertical integration, and highways were three of the most important characteristics affecting the location choice of region. Vertical integration and urbanization had relatively larger impacts, while population density and the presence of an interstate highway had more modest impacts on the probabilities of a firm choosing a location.

Insights related to the location choices of IT and E-commerce firms offer two principle benefits. First, it provides insight into the growth prospects of several sectors thought to have both positive growth prospects as the economy continues to move into the still emerging Information Age, and sectors believed vital to future regional competitiveness as globalization becomes ever more entrenched. In this regard, metro areas would seem to retain an advantage over both adjacent and nonadjacent regions in the Midwest. Rural areas, however, are not without hope or prospects.

In this vein, our combined modeling approach helps inform local and regional policies intended to enhance regional economic welfare in the U.S. midsection. They suggest, for instance, the importance of infrastructure and the need to foster regional integration to promote the welfare of both core and peripheral areas.

Footnotes

Appendix

Authors’ Note

The names of the authors are listed alphabetically. Senior authorship is considered shared.

Declaration of Conflicting Interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project was supported by the National Research Initiative of the Cooperative State Research, Education and Extension Service, USDA, Grant # 2003-35401-13887.

Notes

References

Arita

McCann

. 2000. “Industrial Alliances and Firm Location Behaviour: Some Evidence from the US Semiconductor Industry.” Applied Economics 32:1391–403.

Bartik

T. J.

1985. “Business Location Decisions in the United States: Estimation of the Effects of Unionization, Taxes, and other Characteristics of States.” Journal of Business and Economic Statistics 3:14–22.

Basile

Giunta

Nugent

B. J.

. 2003. “Foreign Expansion by Italian Manufacturing Firms in the Nineties: An Ordered Probit Analysis.” Journal Review of Industrial Organization 23:1–24.

Beck

1992. Shifting Gears: Thriving in the New Economy. Toronto: Harper Collins Publishers.

Bradshaw

T. K.

Blakely

E. J.

. 1999. “What are “Third-Wave” State Economic Development Efforts? From Incentives to Industrial Policy.” Economic Development Quarterly 13:229–44.

Buckley

Henry

Gurmukh

. 2000. Digital Economy 2000. Economics and Statistics Administration Office of Policy Development. Washington, DC: U. S. Department of Commerce.

Carlton

W. D.

1979. “Why New Firms Locate Where They Do: An Econometric Model.” In Interregional Movements and Regional Growth, edited by Wheaton

. Washington, DC: The Urban Institute.

Carlton

W. D.

1983. “The Location and Employment Choices of New Firms: An Econometric Model with Discrete and Continuous Endogenous Variables.” Review of Economics and Statistics 65:440–49.

Chung

Alcacer

. 2002. “Knowledge Seeking and Location Choice of Foreign Direct Investment in the United States.” Management Science 48:1534–54.

10.

Corona

Doutriaux

Mian

A. S.

. 2006. Building Knowledge Regions in North America: Emerging Technology Innovation Poles. Northampton, MA: Edward Elgar Publishing.

11.

Coughlin

C. C.

Segev

. 2000. “Location Determinants of New Foreign-Owned Manufacturing Plants.” Journal of Regional Science 40:323–51.

12.

Cronin

F. J.

Herbert

Colleran

. 1992. “Linking Telecommunications and Economic Competitiveness (Impacts of Network Modernization on the U.S. Economy During the Past 30 Years).” Telephony 223:38–40.

13.

Davies

Greenwood

. 2001. “A Conditional Logit Approach to U.S. State-To-State Migration.” Journal of Regional Science 41:337–60.

14.

Dessens

A. G. J.

Jansen

Ganzeboom

B. G. H.

Van der Heijden

G. M. P.

. 2003. “Patterns and Trends in Occupational Attainment of First Jobs in The Netherlands, 1930-1995: Ordinary Least Squares Regression Versus Conditional Multinomial Logistic Regression.” Journal of the Royal Statistical Society: Series A 166:63–84.

15.

Devereux

P. M.

Griffith

R. R.

Simpson

. 2003. Agglomeration, Regional Grants and Firm Location. WP04/06. University of Warwick: The Institute for Fiscal Studies.

16.

Dorfman

H. J.

Patridge

M. D.

Galloway

. 2008. “Are High-Tech Employment and Natural Amenities Linked?: Answers from a Smoothed Bayesian Spatial Model.” 2008 American Agricultural Economics Association Annual Meeting, July 27-29, Orlando, FL.

17.

Fosler

R. S.

1992. “State Economic Policy: The Emerging Paradigm.” Economic Development Quarterly 6:3–13.

18.

Goetz

S. J.

Deller

Harris

, Eds. 2009. Targeting Regional Economic Development. New York: Routledge.

19.

Goetz

S. J.

Rupasingha

. 2002. “High-Tech Firm Clustering: Implications for Rural Areas. (Principal Paper Sessions).” American Journal of Agricultural Economics 84:1229–36.

20.

Greene

W. H.

2000. Econometric Analysis, 4th ed. Englewood Cliffs, NJ: Prentice-Hall.

21.

Guimarães

Figueirdo

Woodward

. 2003. “A Tractable Approach to the Firm Location Decision Problem.” The Review of Economics and Statistics 85:201–4.

22.

Guimarães

. 2004. “Industrial Location Modeling: Extending the Random Utility Framework.” Journal of Regional Science 44:1–20.

23.

Gunther

F. C.

Liu

W. K.

Bhaskaran

Rothmayer

A. P.

Devereux

P. M.

Griffith

. 1998. “Taxes and the Location of Production: Evidence from a Panel of U.S. Multinationals.” Journal of Public Economics 68:335–67.

24.

Hackler

2003. “High-Tech Location in Five Metropolitan Areas.” Journal of Urban Affairs 25:625–40.

25.

Hamrick

1997. “Rural Labor Markets Often Lead Urban Markets in Recessions and Expansions.” Rural Development Perspectives 12:11–17.

26.

Hamrick

2001. “Displaced Workers: Differences in Non-Metro and Metro Experience in the Mid-1990s.” ERS Rural Development Research Report No. 92. Washington, DC: U.S. Department of Agriculture.

27.

Henry

Cooke

Buckley

Dumagan

Gill

Pastore

LaPorte

. 1999. The Emerging Digital Economy II. Economics and Statistics Administration. Office of Policy Development, Washington, DC: U. S. Department of Commerce.

28.

Hansen

R. E.

1987. “Industrial Location Choice in São Paulo, Brazil: A Nested Logit Model.” Regional Science and Urban Economics 17:89–108.

29.

Isserman

A. M.

2001. “Competitive Advantages of Rural America in the Next Century.” International Regional Science Review 24:38–58.

30.

Kim

Barkley

D. L.

Henry

M. S.

. 2000. “Industry Characteristics Linked to Establishment Concentrations in Nonmetropolitan Areas.” Journal of Regional Science 40:231–59.

31.

Lentz

R. G.

Oden

M. D.

. 2001. “Digital Divide or Digital Opportunity in the Mississippi Delta Region of the U.S.” Telecommunications Policy 25:291–313.

32.

Lichter

D. T.

Garrett

Marshall

Cardella

. 2005. Emerging Patterns of Population Redistribution and Migration in Appalachia. Demographic and Socioeconomic Change in Appalachia Series. Washington, DC: Population Reference Bureau and Appalachian Regional Commission.

33.

McFadden

1973. “Analysis of Qualitative Choice Behaviour.” In Frontiers in Econometrics, edited by Zarembka

, 105–42. New York: Academic Press.

34.

MIG Inc. 1999. IMPLAN Professional, Version 2.0: User’s Guide, Analysis Guide, Data Guide. Stillwater, MN: Minnesota IMPLAN Group, Inc.

35.

Miller

M. M.

Gibson

L. J.

Wright

N. G.

. 1991. “Location Quotient: Basic Tool for Economic Development Analysis.” Economic Development Review 9:65–8.

36.

Miller

R. E.

Blair

P. D.

. 1985. Input-Output Analysis: Foundations and Extensions. Englewood Cliffs, NJ: Prentice-Hall, Inc.

37.

Misangyi

V. F

. 2002. A Test of Alternative Theories of Managerial Discretion. Graduate School, University of Florida, PhD Thesis. http://etd.fcla.edu/UF/UFE1000145/misangyi_v.pdf (accessed April 27, 2011.).

38.

Office of Management and Budget. 2000. “Standards for Defining Metropolitan and Micropolitan Statistical Areas; Notice.” Federal Register 65:82228–38.

39.

Oliner

Sichel

. 2000. “The Resurgence of Growth in the Late 1990s: Is Information Technology the Story?” Journal of Economic Perspectives 14:3–22.

40.

Pigeon

Wray

L. R.

. 1999. “Did the Clinton Rising Tide Lift All Boats?” Challenge 42:14–33.

41.

Porter

M. E.

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

42.

Porter

M. E.

2000. “Locations, Clusters and Company Strategy.” In The Oxford Handbook of Economic Geography, edited by Clark

G. L.

Feldman

M. P.

Gertler

M. S.

, 253–74. Oxford: Oxford University Press.

43.

Rosegrant

Lampe

D. R.

. 1992. Route 128: Lessons from Boston's High-Tech Community. New York: Basic Books, Inc.

44.

Ross

Friedman

R. E.

. 1990. “The Emerging Third Wave: New Economic Development Strategies.” Entrepreneurial Economic Review 90:3–10.

45.

Steinberg

2003. “Information Technology & Development Beyond “either/or”.” The Brookings Review 21:45.

46.

White

1988. “Location Choice and Commuting Behaviour in Cities with Decentralized Employment.” Journal of Urban Economics 24:129–52.

47.

Wolff

M. F.

2008. “Tech Salaries Up, but Jobs Down (PERSPECTIVES: Views and News of the Current Research--Technology Management Scene).” Research-Technology Management 51:5–6.