lable at ScienceDirect

Socio-Economic Planning Sciences 58 (2017) 51e62

Contents lists avai

Socio-Economic Planning Sciences

journal homepage: www.elsevier .com/locate/seps

Industry variations in the broadband business nexus

Elizabeth A. Mack a, *, Elizabeth Wentz b

a Department of Geography, Environment, and Spatial Sciences, Michigan State University, United Statesb School of Geographical Sciences and Urban Planning, Arizona State University, United States

a r t i c l e i n f o

Article history:Received 15 July 2015Received in revised form18 October 2016Accepted 19 October 2016Available online 20 October 2016

Keywords:ICTsBroadbandBusinessIndustryInternetSpatial effects

* Corresponding author.E-mail address: emack@msu.edu (E.A. Mack).

http://dx.doi.org/10.1016/j.seps.2016.10.0070038-0121/© 2016 Elsevier Ltd. All rights reserved.

a b s t r a c t

Although broadband Internet infrastructure is acknowledged as a key ingredient to competitiveness, anunfortunate aspect of current work is the dominant focus on households. Given the need for moreresearch on the multidimensional relationship between broadband and businesses, or the broadband-business nexus, this study estimates econometric models to evaluate the impact of early broadbandavailability on future levels of business activity. Model results suggest regions with an early advantage inbroadband provision had more business growth than other regions. Model results also highlight long-lasting spatial effects on business activity stemming from broadband spillovers from core hubs toneighboring areas.

© 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Technological change continues to be a fundamental driver ofeconomic growth [36]. A landmark achievement in recent decadesis high-speed broadband Internet connections, which have openedthe door to unprecedented economic opportunity associated with alarge number of innovations in products and services [5]. Due to theimportance of Internet connectivity to job creation [38] and theeconomic development of regional and national economies [33],considerable attention has been dedicated to understanding thesocial and economic implications of uneven access to this criticalinfrastructure. Studies to this point have emphasized the socio-economic and demographic drivers of the demand for and supplyof broadband [24]. Geographic perspectives on the distribution ofthe physical infrastructure necessary to access the Internet havehighlighted uneven Internet access between and within metro-politan areas [18].

While early evaluations of uneven access divided people andregions into the haves and have-nots, this binary perspective onbroadband infrastructure has evolved into a much more nuancedand complex issue related to variations in cost, platform choice,provider choice, and infrastructure reliability and performance [15].

Recognizing these inequalities, the case is being made that broad-band provision functions like a utility and is as vital as electricity acentury ago; thus emphasizing the importance of municipalbroadband initiatives in filling service gaps [14]. The recent subsidyplan initiated by the Federal Communications Commission (FCC) tohelp provide low-income households with Internet access [23] onlyhighlights the continued relevance of Internet infrastructure in anetworked and technologically advanced digital economy.

Although broadband infrastructure is acknowledged as a keyingredient to competitiveness, an unfortunate aspect of currentwork is the dominant focus on households. This emphasis ignoresbusiness adoption and use, which account for a large proportion ofthe economic impacts associated with this infrastructure in termsof jobs, productivity impacts and GDP growth [25]. While recentwork on broadband provision and business activity has uncoveredindustrial [22] and geographic [29,32] variations in this relation-ship, more research is needed on the multi-dimensional relation-ship between broadband access and business activity. The findingsof prior work suggest that locales without broadband or poorquality broadband infrastructure, may be at a disadvantage inretaining and attracting businesses [30,32]. If true, this disadvan-tage impacts the ability of economic development entities to pur-sue cluster-oriented development strategies (i.e., high technology).To complicate matters, urban locales that were a point of initialbroadband deployment efforts may have derived a first-moveradvantage in retaining and attracting businesses due to the

Fig. 1. Dimensions of the broadband business nexus (BBN).Source: Grubesic and Mack [16].

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e6252

inertia behind infrastructure upgrades, which follow initialdeployment locations because they represent the most favorabledemand conditions and highest chances of recouping upgrade costs[3]. This suggests further limitations of places outside of corebroadband deployment areas to overcome setbacks in businessretention and attraction efforts stemming from initial disparities inInternet access.

Given the amount of money spent on deploying Internet infra-structure by the federal government, and the dearth of informationabout the linkages between broadband availability and businesses(the broadband business nexus), the goal of this study is to evaluatethe impacts of early levels of broadband availability on future levelsof business activity. In this regard, the goals of this paper are a) toinvestigate industrial variations in the long-term impact of earlylevels of broadband availability on business activity; and b) toexamine the utility of various instruments to econometricallymodel this relationship. To examine industrial variations in thisrelationship, a national county-level dataset is used to constructeconometric models to examine the impact of early levels ofbroadband availability on future levels of business activity. Resultshighlight long-lasting impacts of broadband market dynamics onneighboring areas of initial hubs of broadband availability. Theselong-lasting impacts are also specific to particular industries. Whilerevealing, this look at one aspect of the broadband business nexussuggests new paths for research. In this regard, a research agenda isoutlined at the close of the paper to highlight key areas for newresearch to better understand the dimensionality of this relation-ship and encourage more widespread participation across researchdisciplines.

2. Information and communication technologies (ICTs) andeconomic activity

Investments in telecommunications are positively linked toeconomic growth [6], and the development trajectory of regionaleconomies [9]. More recently, studies have found similar positivelinkages between broadband enabled telecommunications tech-nologies and economic activity including gross domestic product[33] and employment [38]. Taken together, this research suggeststhat broadband is a part of nations' telecommunications infra-structure [13].

While research specific to broadband finds positive impacts oneconomic activity, studies suggest the need to consider nuances inthe size of these impacts stemming from subtle industrial charac-teristics of the workforce and the quality of human capital at theindividual and regional level [12], [4]; [31]. For example, Yilmaz,Haynes, and Dinc [40] uncovered that telecommunications infra-structure investments had the strongest impact on the wholesaletrade, finance, insurance and real estate (FIRE) and services sectors.Kolko [26] found stronger impacts of broadband on economicgrowth for information intensive industries. In addition toindustry-based variations in the economic impacts of broadband;studies of the quality of human capital and ICTs find evidence ofskills-biased technological change related to computer-basedtechnologies [12] and broadband presence at the regional level[31].

3. Evolutionary dynamics of ICTs

The evidence of skills biased technological change suggests thatthe provision of infrastructure alone is an insufficient solution foreconomic advancement [29]. Along these lines, several studies havesuggested that telecommunications be one of many investments inregional economies. Dholakia and Harlam [9] for example, recom-mend that telecommunications be one of many economic

development investments amongst other types of physical infra-structure (i.e. roads) as well as education and training. While in-vestments in telecommunications alone are insufficient to theachievement of positive economic development outcomes, partic-ularly in rural areas with a plethora of development issues [20,30],the provision of infrastructure is an essential precondition toachieving positive outcomes in non-rural locales. BroadbandInternet connections are considered a general-purpose technology(GPT), which means that they provide opportunities for down-stream innovational complementarities in other economic sectorsbesides telecommunications [5]. Further, Internet technologieshave network effects [33], which means that the amount of eco-nomic benefits depend upon the number of users; the greater thenumber of users, the greater the economic benefits.

Unfortunately, the distribution of this infrastructure remainsuneven in terms of basic geographic availability as well as the skillsneeded to unravel the economic benefits associated with broad-band connections. This divide in basic availability and use is alsoknown as the “digital divide”which refers generically to differencesin access and use of information and communications technologies(ICTs). The dynamism associated with ICTs suggests that the phrase“digital divide” be viewed from an evolutionary perspective. Thisdynamism is related to technological change and the latest tech-nologies that are considered ICTs. While fax machines and tele-phones were classified as ICTs in the 1970's, this definition hasrapidly evolved to embody a new set of technologies that representthe convergence of telephone-oriented and computer-orientedtechnologies in smartphones and tablet-based access devices. Asregards the diffusion of ICTs, studies find that the adoption rates ofnew ICTs are accelerating [7]. A recent study found it took justthirty-one years for one-quarter of the U.S. population to adopt theradio, which became commercially available in 1897, as comparedto just seven years for the Internet, which became commerciallyavailable in 1991 [7]. The speeds at which people access theInternet have also increased tremendously since its initial years ofcommercial availability, from dial-up speeds between 14.4 and28.8kbps [2], to 4G LTE speeds with download speeds of 6.5 Mbpsand upload speeds of 5.0 Mbps [37].

4. The broadband-business nexus (BBN)

The importance of these speed increases, as well as many otherfacets of broadband adoption and use specific to businesses, re-mains largely unstudied. As mentioned previously, the volume ofresearch about business adoption and use of broadband pales incomparison to the research about households. Given this lack ofattention to the business side of Internet research, a conceptualframework for understanding this relationship or the BroadbandBusiness Nexus (BBN) has been proposed. The BBN is defined as thereciprocal relationship between broadband availability and busi-nesses [16]. Fig. 1 outlines the four dimensions of this nexus. The

1 Per the precedent set by prior studies, the number of providers in suppressedZIP codes areas is assumed to be one. Previous investigations of this assumptionhave highlighted it has no impact on econometric model results [35].

2 FCC data for this time period does not contain information about access speeds.

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e62 53

first dimension speaks to business to business interactions and theagglomerative benefits of co-located businesses in physical space.The second dimension examines the association between busi-nesses and telecommunications which is the dimension of interestin this study. The third dimension of the nexus is designed tocapture the function of the Internet as a business service. Here,telecommunications providers target areas with a high density ofbusinesses that have high demand for Internet services. The finaldimension deals with interactions between telecommunicationsproviders. One example of this type of interaction is sharingnetwork access points or peering to minimize transactions costs[16].

In the context of this study, the second dimension of the BBN,which speaks to the ability of telecommunications to stimulategrowth, is of principle interest. However, much more work re-mains to unpack this interaction. From an economic develop-ment standpoint, this relationship is important to understandgiven the high costs of broadband deployment and the centralrole of businesses in economic development initiatives. Whilestudies highlight that broadband is not a cure-all for the devel-opment ills of remote locales [20,30], a more nuanced under-standing of the importance of and impact of broadbandinfrastructure on business activity is necessary. To date, much ofthe work on the digital divide has emphasized spatial aspects[18], as well as factors impacting individual adoption and use[24], without a clear understanding of the socioeconomic andphysical drivers.

Early work in this area is important because studies havesuggested that the uneven distribution of ICT infrastructure mayinhibit the ability of firms to move away from central city loca-tions [34]. Other work suggests that two related aspects of firms,the skill level of workers [39] and firm industry membership [11],play a role in the importance and impact of ICTs on firms. Anevaluation of the adoption of dial-up Internet connections forbasic and advanced uses found industry specific adoption rates ofbroadband for advanced purposes [11]. Later work on broadbandand business has also uncovered evidence of industry specificimpacts [22].

More recent work on broadband and business has foundbroadband to be an important explanatory factor of businesspresence at both the ZIP code area [32,35] and county levels [29].This relationship does show marked variation across metropolitanareas that is likely related to metropolitan area size and the in-dustrial legacy of these locales [35]. In an effort to tease out wherewithin the urban hierarchy broadband matters most in terms ofurban, suburban, and rural locations, the Mack [30] study foundthat the definition of the urban hierarchy does matter and thatbroadband mattered most to businesses at intermediate localeswithin this hierarchy which are suburban and exurban locales thatare located between core urban areas and remote rural areas.Combined, this body of work suggests that space and industrymembership matter in the relationship between broadband infra-structure and businesses. Given the need for additional work aboutthese aspects of the BBN however, the present study will examinethe duration of the impact of early broadband levels on future levelsof business activity.

5. Data

5.1. Broadband data

Broadband provision data were obtained from the historicaldata provided by the Federal Communications Commission (FCC)via their Form 477 database. Specifically, broadband informationfor the years 1999 and 2000 were used from this database per the

precedent set by prior studies1 [13] [21]; [26]. These data wereaggregated to the county level using a commercial geographic in-formation system (GIS) to obtain a count of the number of providersin every county across the continental United States. This measureof broadband availability is of interest because it not only speaks tothe availability of broadband within a county, but it also speaks tocompetitive dynamics of broadband markets [27]. These dynamicsinclude more provider choice, lower prices, and better servicequality [15].2

5.2. Business data

Businesses associated with agriculture, manufacturing, retail,knowledge work, healthcare, and rural businesses were selectedfor further examination. These selections was made based on thefindings of prior work, and information about the potential useintensity of broadband Internet connections. Agriculture,manufacturing and retails were selected based on Forman et al.[11] and Mack [29]. Health care establishments were selectedbecause of the emphasis on healthcare in telecommunicationspolicy including the Telecommunications Act of 1996 Section254b [1] and the National Broadband Plan [10]. The Internet isused for provider-to-provider and patient-to-provider consulta-tions in addition to the transfer of images [1]. The present studyalso considers businesses in rural areas to understanding thedegree to which broadband deployment efforts help retain andattract businesses in rural locations. While broadband is likely nota cure all infrastructure for lagging urban areas, the strategicincorporation of deployment, upgrading, and adoption initiativesare now a critical aspect local economic development efforts inthe Internet age [20].

Establishment level data are used as the measure of businesspresence within U.S. counties. These data were obtained fromCounty Business Patterns from the U.S. Census Bureau. This data-base contains industry level information on establishment activitydating back to 1998. Seven different breakdowns of establishmentsare used in the model results that will be reported. These break-downs are as follows:

1. All establishments2. Agricultural establishments (NAICS, 11)3. Manufacturing establishments (NAICS 31e33)4. Retail Trade (NAICS 44e45)5. Knowledge intensive establishments (NAICS 51, 52, 54, 55, 61)6. Healthcare establishments (NAICS 62)7. Rural establishments

5.3. Model covariates

Table 1 contains a description of the dependent variables ofinterest in the models to be estimated, as well as the primary in-dependent variable of interest. Also included in this table areseveral controls that describe a variety of county characteristicsranging from labor force characteristics to business characteristicsto climatic amenities. Personal income, a large component of whichis salaries and labor, is designed to capture the cost of doing busi-ness. In terms of business characteristics, both the Herfindahl indexof business diversity and average establishment size are meant to

Table 1Variable descriptions.

Variable Definition Source

Dependent variablesChange total establishments Change in the natural logarithm of total establishments

between 2013 and 2000County Business Patterns, U.S Census Bureau

Change agricultural establishments Change in the natural logarithm of agriculturalestablishments between 2013 and 2000

County Business Patterns, U.S Census Bureau

Change manufacturing establishments Change in the natural logarithm of manufacturingestablishments between 2013 and 2000

County Business Patterns, U.S Census Bureau

Change retail Trade establishments Change in the natural logarithm of retail tradeestablishments between 2013 and 2000

County Business Patterns, U.S Census Bureau

Change knowledge establishmentsa Change in the natural logarithm of knowledgeestablishments between 2013 and 2000

County Business Patterns, U.S Census Bureau

Change health care establishments Change in the natural logarithm of healthcareestablishments between 2013 and 2000

County Business Patterns, U.S Census Bureau

Change rural establishments Change in the natural logarithm of all establishments inrural locations between 2013 and 2000

County Business Patterns, U.S Census Bureau

Primary independent variables of interestLn Broadband Providers Natural logarithm of the total number of broadband

providers in 2000Federal Communications Commission (FCC) Form 477database

Spatial Lag of Broadband The average of the natural logarithm of broadbandproviders in 2000 in neighboring counties as defined via afive nearest neighbor weights matrix

Auhor's calculation in GeoDa

Control variablesResearch University Dummy variable that indicates whether a county contains a

research I university, as defined by the CarnegieClassification System in 2000

Integrated Postsecondary Education Data System (IPEDS)

Ln of highway miles Ln of the total number of highwaymiles in a county in 2000. National Transportation Atlas Database (NTAD), Bureau ofTransportation Statistics (BTS)

Percent white Percent of the population that is white in 2000 U.S. Census BureauNatural amenity index Natural amenity index in 1999 US Department of Agriculture, McGranahan andWojan [28]Percent bachelor's or higher Percent of the population with a bachelor's degree or higher

in 2000U.S. Census Bureau

Per Capita personal income 30 year time lag of personal income per capita in 1970 Bureau of Economic Analysis (BEA)Herfindahl index Indicator of industrial diversity created from Standard

Industrial Classification (SIC) System employment in 2000Regional Economic Information System (REIS) of the Bureauof Economic Analysis

Average establishment size Total number of employees/total number of establishmentsin 2000

U.S. Census Bureau

Agglomeration economies A dummy interaction variable computed by multiplying adummy variable indicating county urban areab membershipand employment density in 1970

U.S. Census Bureau and Regional Economic InformationSystem (REIS) of the Bureau of Economic Analysis

Population density Number of people per square mile in 1970 U.S. Census Bureau

a The number of knowledge establishments in a county is the sum of establishments in the following two-digit NAICS industries: Information (51), Finance and Insurance(52), Professional, Scientific, and Technical Services (54), Management of Companies and Enterprises (55), and Educational Services (62). It excludes TelecommunicationsProviders (NAICS 517).

b An urbanized area consists of core census block groups or blocks that have a population density of at least 1000 people per square mile and surrounding census blocks thathave an overall density of at least 500 people per square mile. For more information, see http://www.census.gov/geo/www/ua/ua_2k.html.

3 For additional information about this index, please see http://www.ers.usda.gov/data-products/natural-amenities-scale.aspx.

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e6254

capture the degree of specialization or diversity in regionalemployment and the size of the businesses within counties (interms of number of employees) respectively. With regard to theHerfindahl index, it is important to note that high values of thisindex indicate specialization while lower index values indicate di-versity. This is important to note for the interpretation of regressionresults. A positive sign on this variable indicates a positive rela-tionship between specialization and business presence while anegative sign indicates diversity has a positive relationship onbusiness activity.

Other controls designed to capture labor force characteristicsinclude the educational attainment of area residents (bachelor'sdegree or higher), percent white and population density. Percentwhite is considered an indicator of the diversity of the local laborforcewhile a deep time lag of population density is ameasure of thesize of the potential labor force. The control for agglomerationeconomies is computed so as to capture the thickness of labormarkets in urban areas within counties. As is the case with thepersonal income measure and population density, a deep time lagfor employment is used to mitigate endogeneity issues betweenthis variable and other covariates. Other covariates included in the

models are designed to control for the level of infrastructurewithincounties and the natural environment. A dummy variable isincluded to capture the presence of a major research universitywithin county boundaries. The number of highway miles is alsoincluded as a measure of physical infrastructure besides telecom-munications infrastructure (in terms of broadband providers). Thenumber of highway miles includes rural arterials, urban principalarterials, and National Highway System routes (BTS). Finally, thenatural amenities index [28] is included in the models as a measureof climatic amenities [17]. The index is composed of seven types ofnatural environment characteristics: land surface topography(plains, hills, mountains or tablelands), relative rurality of thecounty, mean temperature in January, mean temperature in July,mean hours of sunlight, mean relative humidity, and percent ofarea that is water. This composite index has a range between 1 and7 and is incorporated in the regression models in terms of de-viations from the mean.3

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e62 55

5.4. Instrumental variables

Aside from deep time lags to mitigate endogeneity issues be-tween businesses and several control variables including populationdensity, personal income, and the agglomeration variable; anotherconcern is the simultaneous relationship between broadband pro-viders. In other words, does broadband stimulate business growth,or do businesses stimulate broadband growth? One mechanism fordealing with this type of endogenous relationship is to control forthe simultaneity to understand one dimension of the relationship. Inthe context of this study, the question is: How do early levels ofbroadband availability impact business activity? To answer thisquestion, instruments, which are variables that are correlated withthe endogenous variable (broadband in this case) but not correlatedwith the error term, are used in an instrumental variables estima-tion process in place of the original endogenous broadband variable.

One of the instruments considered is the slope of the terrain,which has been used in prior studies. Kolko [26] suggests that slopemeasures the steepness of the terrain, which is an indication of thelevel of difficulty associated with deploying broadband; the steeperthe terrain in mountainous and semi-mountainous areas, the moredifficult and the more expensive will be deployment efforts. Giventhe cost aversion of private providers, this means locales withsteeper terrains will have fewer broadband providers. To calculatethis instrument, elevation data were obtained from the NationalElevation Dataset (NED) with a 1 arc-second spatial resolution(~30 m cells) from the United States Geological Survey (USGS)National Map Viewer. The data are in geographic coordinates(decimal degrees) in North American Datum 1983 with elevationreported in meters and referenced by the North American VerticalDatum of 1988. While higher spatial resolution elevation data areavailable (1/3 and 1/9 arc second) in portions of the study area, thelarger resolution cells were used for consistency across the wholestudy area. The elevation data were processed using Esri's ArcGISVersion 10.1 software. The elevation data were merged into one filerepresenting continuous elevation across the contiguous U.S. Thesedata were used to calculate a slope value per cell. By intersectingthe slope datawith the U.S. counties, it was possible to calculate theminimum,maximum, mean, and standard deviation values of slopefor all U.S. counties.

6. Estimation strategy

To provide greater resolution on the relationship between earlylevels of broadband availability and business presence in lateryears, multiple industry level econometric models are estimated.This relationship is an important connection to understand if weare to unravel the impacts of broadband on the growth trajectory ofregional economies. A secondary objective is to evaluate whatmight serve as viable instruments for evaluating this simultaneousrelationship. This will serve to improve and streamline futurestudies evaluating the broadband-business nexus. To evaluate thisconnection, two types of models will be estimated, onewithout andone with a spatial lag:

Dyi, 2013-2000 ¼ a þllnBbandi, 2000þbXþssþεi (1)

Dyi, 2013-2000 ¼ aþlBbandi, 2000þrBbandLagi,2000þbXþssþεi (2)

where:Dyi, 2013-2000 represents the change in businesses calculatedas the difference between the natural logarithm of businesses in2013 and the natural logarithm of businesses in the year 2000.

l is the coefficient estimate for the natural logarithm of broad-band in the year 2000b is a matrix of coefficient estimates on the covariates describedin section 4r is a coefficient estimate for the spatial lag of broadband in theyear 2000ss are state fixed effectsεi is the residual

These models are a variation on a growth regression specifica-tion, which is designed to understand how the initial physical,economic, and demographic conditions of counties, impact busi-ness growth in future years (between 2000 and 2013). Both types ofmodels will be estimated with White heteroskedastic robuststandard errors. State fixed effects are included to account for theunique business environment within a state, including regulationsthat impact business operations. These fixed effects are alsoimportant to capture the state specific regulatory environment forbroadband, which likely impacts the number of providers andspatial distribution of providers within each state. For example, notall states allow municipalities to be involved in building and/oroperating broadband networks, and 22 states restrict or prohibitthe involvement of municipalities in such activity [8].

Specification 2, includes r, which corresponds to the coefficientestimate on a spatial lag of broadband. Including a spatial lag forbroadband is an effort to control for spatial autocorrelation that hasbeen identified in prior studies [21]. This spatial lag was con-structed with a five nearest neighbor weights matrix. The weightsmatrix ensures that each county has five neighbors, which isimportant when dealing with irregularly based spatial units. Inthese instances, a nearest neighbor weights matrix is preferable toa contiguity based weights matrix because it ensures that eachobservation has neighbors. Multiple models of types one and twowill be estimated for each of the seven breakdowns of businessactivity described in section four. In addition to industry variationsin the relationship between businesses and broadband, the esti-mation of these models will also consider the best instruments tomodel the endogenous relationship between broadband andbusinesses. With respect to businesses, several studies have sug-gested simultaneity between business demand for broadband andthe supply of broadband by providers [32,35]. As discussed in thedata section, the slope of the terrain has been offered as an in-strument in previous work [26], as have the combined use ofhousehold density and a time lag of broadband provision [32,35]. Apotential drawback of the use of slope however is the time intensityinvolved with computation, which may not make it as readily orfeasible to the scientific community as other instruments. Thus,models will be estimated to compare the relative validity andperformance of these two recommended sets of instruments.

7. Results

7.1. Spatial analysis of broadband and businesses

Figs. 2 and 3 display the geographic distribution of broadbandprovision in 2000 and counties with a positive change in estab-lishments for the industries of interest between 2000 and 2013.Counties with positive changes in the level of agricultural estab-lishments include Montana, Wyoming, Idaho, Nebraska and Iowa,as well as Mid-Western states such as Illinois, Ohio, Indiana, andMichigan. Locations with positive changes in manufacturing es-tablishments are located primarily in upper plains states such asNorth Dakota and South Dakota and western states including

Fig. 2. Broadband provision in 2000 and change in establishments by industry (2000e2013).

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e6256

Nevada, Utah, Colorado, and Idaho. Fig. 3 highlights that growth inknowledge establishments for this time frame also has a distinctspatial pattern evident in Western (Phoenix, Utah, Colorado andWyoming) and Southeastern states (Texas, Louisiana, Florida, andGeorgia). In this time frame however, few locations exhibiteddistinct positive changes in retail establishments with the excep-tion of a few counties in California, Texas, Utah, and Montana. Of allthe industries of interest in this study, healthcare and social assis-tance saw the most widespread growth across the country;numerous counties in nearly every state display positive changes inbusinesses within this sector.

The map of providers in Fig. 2 highlights a core peripheryframework to broadband provision, which has been highlighted inprevious studies [21]. This pattern means that core county areaswithin major metropolitan areas contain the highest levels ofprovision. Examples of these locales include the San Francisco andLos Angeles portions of California; Dallas, Austin and HoustonTexas, as well as major cities on the east coast such as Boston andNew York. Rural portions of the country contain visibly lower levelsof provision. Particular states also contain notably lower number ofbroadband providers; these include states in the middle portion ofthe country, as well as several states in the Southeast such asMississippi, Arkansas, and Louisiana.

To evaluate the statistical significance in the spatial patternshighlighted above, a bivariate local Moran analysis was conducted

using a 5 nearest neighbor weights matrix to match the weightsmatrix used in the construction of the broadband lag variableincluded in the econometric models. The bivariate Moran wascomputedwith the natural logarithm of broadband provision as theindependent variable of interest and the natural logarithm of eachof the business types as the dependent variable of interest. Theresults of this analysis are found in Table 2. All statistics in this tableare significant at the 1% level. A standardized z-value of the originalstatistic is also included in this table because it provides an indi-cation as to the direction of the autocorrelation. A positive z-valueindicates a positive spatial association between broadband andbusinesses while a negative z-value indicates a negative associationbetween broadband and businesses. In other words, high levels ofbroadband are not associated with large positive changes in busi-ness activity. In this table, broadband is positively associated withpositive changes in all types of business activity with the exceptionof agriculture and manufacturing. The extent of this associationalso varies across industries. For example, retail trade and healthcare have higher levels of autocorrelation than do businesses inrural areas. These results are important to the model resultsdescribed in the next section.

7.2. Model results

Results in the form of coefficient estimates and their associated

Table 2Bivariate Moran analysis.

Moran's I statistic Z-value

All businesses 0.1089*** 13.5188Agriculture, forestry, fishing and hunting �0.0392*** �4.8853Manufacturing �0.0752*** �9.2329Retail trade 0.1818*** 22.6476Knowledge businesses 0.1251*** 15.155Health care and social assistance 0.1454*** 17.2889Rural businesses 0.0254*** 3.2222

***Significant at the 1% level.**Significant at the 5% level.*Significant at the 10% level.

Fig. 3. Change in establishments by industry (2000e2013).

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e62 57

standard errors for the models for each of the seven divisions ofbusinesses are reported in Table 3. The top row of Table 3 lists theinstruments used in the estimation of results. One instrument isused in the models with no spatial lag, this is a time lag of broad-band. Two instruments are used in the models with a spatial lag; atime lag of broadband and a spatial lag of counties that arecontiguous to a point-of-presence location in 1999. Appendix Bcontains the ordinary least squares (OLS) model results forcomparative purposes. A comparison of these tables highlights that

endogeneity is an issue in assessing the relationship between his-torical levels of broadband provision and future levels of businessactivity.

Model results in Table 3 highlight nuanced industrial and spatialvariations in the association between historic broadband provisionlevels and future business levels. Across many industries (agricul-ture, manufacturing, retail trade, and businesses in rural areas)historic levels of broadband have no association with the change inbusiness level over the thirteen-year study period. These results arenot necessarily unexpected for several reasons. First, prior work hashighlighted that several industries including retail trade and agri-culture were less likely to use dial-up Internet connections foradvanced purposes [11]. This is likely a trend that translates to theadoption of higher-speed broadband Internet connections. Second,while competitive broadband markets may have been important tothe location of these businesses in the initial years of availability,this importance likely waned as the infrastructure became morewidespread and the costs associated with broadband accessdeclined. Third, it has been suggested that broadband is unable toovercome the inertia associated with the most remote locales [30].Thus, it is not surprising that the early availability of broadband inremote areas of the country are not associated with higher businesslevels in later years.

Table 3Regression results.

All businesses Agriculture,forestry, fishingand hunting

Manufacturing Retail trade Knowledgebusinesses

Health care andsocial assistance

Rural businesses

Instrumental variables Natural logarithmof broadband in1999, spatial lagof countiescontiguous topoint of presence

Natural logarithmof broadband in1999

Natural logarithmof broadband in1999, spatial lagof countiescontiguous topoint of presence

NaturalLogarithm ofbroadband in1999, spatial lagof countiescontiguous topoint of presence

Natural logarithmof broadband in1999

Natural logarithmof broadband in1999, spatial lagof countiescontiguous topoint of presence

Natural logarithmof broadband in1999

Ln broadband providers 2000 �0.0250 (0.0591) 0.0113 (0.0205) �0.0270 (0.0601) 0.0163 (0.0429) 0.1235***(0.0317)

�0.0090 (0.0529) 0.0021 (0.0047)

Spatial lag broadband 0.1663***(0.0562)

e 0.0867 (0.0887) 0.0740 (0.0587) e 0.1974***(0.0777)

e

Research University �0.0020 (0.0542) �0.0765*(0.0436)

�0.0343*(0.0197)

�0.0420***(0.0162)

�0.0037 (0.0467) �0.0776***(0.0203)

0.0024 (0.0147)

Ln of highway miles 0.0288** (0.0137) �0.0464*(0.0250)

�0.0038 (0.0257) 0.0170 (0.0148) 0.1675***(0.0363)

0.0419* (0.0228) �0.0086*(0.0046)

Percent white 0.0006 (0.0006) �0.0002 (0.0008) 0.0018***(0.0006)

0.0010** (0.0004) 0.0018* (0.0010) 0.0010** (0.0005) 0.0000 (0.0003)

Natural amenity index 0.0083** (0.0042) �0.0341***(0.0082)

0.0019 (0.0058) 0.0040 (0.0034) �0.0276**(0.0118)

0.0044 (0.0045) 0.0051***(0.0018)

Percent bachelor's or higher 0.0097** (0.0044) 0.0164***(0.0032)

0.0068***(0.0018)

0.0102***(0.0015)

0.0156***(0.0042)

0.0130***(0.0018)

0.0024***(0.0006)

Per capita personal income �0.0213 (0.0172) �0.0191 (0.0228) �0.0491***(0.0136)

�0.0264***(0.0082)

0.0297 (0.0316) �0.0864***(0.0143)

0.0022 (0.0047)

Herfindahl index 0.2028 (0.3143) �0.5831***(0.2113)

�0.4776**(0.2457)

0.0821 (0.1550) �0.3071 (0.4375) 0.0073 (0.1289) �0.0559 (0.0475)

Average establishment size �0.0064 (0.0071) 0.0032 (0.0027) 0.0001 (0.0020) 0.0009 (0.0011) �0.0041 (0.0046) �0.0017 (0.0015) 0.0001 (0.0007)Agglomeration economies �1.29E-06 1.15E-

06�8.40E-06***2.52E-06

�2.83E-06***1.05E-06

2.39E-05***1.44E-06

�1.45E-06 2.71E-06

3.21E-06***1.08E-06

e

Population density �3.23E-06 2.03E-06

�1.04E-05***2.39E-06

�3.97E-06***1.55E-06

7.27E-07 2.60E-06

�3.09E-06 3.31E-06

�2.14E-06 1.68E-06

�1.23E-08 2.21E-07

Constant �0.4295***(0.0999)

0.0513 (0.1057) �0.0170 (0.1911) �0.5173 (0.1112) �1.4268***(0.2049)

�0.2163 (0.1603) 0.0362 (0.0355)

F-statistic 19.35 6.74 11.07 85.65 9.07 11.42 9.30Degrees of freedom 2956 2957 2956 2956 2957 2956 753Root MSE 0.2513 0.5415 0.3146 0.1893 0.7117 0.2665 0.0537

Robust standard errors are in italics.***Significant at the 1% level.**Significant at the 5% level.*Significant at the 10% level.

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e6258

Interestingly, early broadband provision levels are important tofuture business activity in future years for select industries, such asthe knowledge sector. These results support the findings of priorwork, which find a strong association between broadband provi-sion levels and knowledge oriented businesses [30,35]. The financeindustry for example, has long been noted to rely on the latestadvances in telecommunications [39]. It is also noted that placesserved by the initial roll-out of telecommunications are more likelyto receive upgrades [19]; [21]. These upgrades are particularlysalient for the efficient transfer of large volumes of information, thetransfer of analytical tools from desktops to online web services,and the intensified use of cloud computing for data storage andretrieval. Thus, the initial presence of competitive broadbandmarkets is likely associated with competition between providersfor infrastructure upgrades to improve the speed and quality ofaccess, which are important to knowledge businesses that relyintensively on higher Internet speeds for these types of Internetuses.

Aside from the subtle industry variations in the linkagesbetween broadband and businesses, model results also high-light important spatial nuances to this relationship. The modelresults for all businesses and health care are indicative of this.While the coefficient for broadband is not significant in either ofthese models, the spatial lag of broadband is significant. This

means that the diffusion of broadband to nearby locationspositively impacted business levels in future years. Thesespillover effects indicate that the enduring effects of broadbandon businesses are not perhaps realized in the communities thatreceive initial outlays of infrastructure. Rather, they are realizedin neighboring communities. This is important to note given thebody of work that highlights telecommunications as anenabling technology that permit business activity to decen-tralize from central city locations [34]. That said, this enablingcapacity does not extend to all areas given the limited capabilityof broadband to transform the business trends in the mostremote rural areas [29,30].

7.3. Sensitivity analysis

Aside from examining industrial variations in the relationshipbetween broadband and business presence, the second goal of thisstudy is to examine the utility of various instruments for broad-band. In this study, the instruments of interest are the slope of theterrain as proposed in Kolko [26] and a temporal lag of broadbandprovision levels [32]. From a theoretical standpoint, both in-struments are valid. Kolko [26] suggests that the slope of the terrainis a measure of the difficulty in deploying broadband provision dueto the complexity of the physical environment. As mentioned

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e62 59

previously, areas with steeper slopes will be more expensive todeploy broadband and will therefore have lower numbers of pro-viders. A temporal lag of broadband provision is also theoreticallyvalid because prior work has demonstrated that historical levels ofprovision are related to current provision levels [21]. Two otherinstruments considered are related to historical provision levels.These instruments are a dummy variable indicating a county con-tains a point-of-presence and a spatial variable indicating a countyis contiguous to a count with a point-of-presence. This lag variableis designed to capture any diffusion of telecommunications fromthis initial access point. Points of presence are critical access pointson the Internet backbone that serve as routing points for traffic andact as on-ramps to the Internet superhighway for local and regionaltraffic [19].

That said, there are substantial computational differences inthe use of these instruments. The temporal lag of broadband andpoint-of-presence information require considerably less time tocompute than does the slope of the terrain. Although it is possibleto obtain slope information from digital elevation models (DEM),the amount of time required to compute the slope of the terrain isconsiderable. The processing of the 30 m cell data in this studytook several weeks. Given the differences in each of the in-struments of interest from a computational perspective, a sensi-tivity analysis was conducted to compare the model results,focusing on the coefficient for broadband. In this sensitivityanalysis a total of 56 models were estimated to assess the utility ofthe instruments for the two types of models of interest in thisstudy; the model without a spatial lag of broadband and themodel with a spatial lag of broadband.

Table 4 presents the results of this sensitivity analysis focusingon the coefficient estimates for broadband and the spatial lag ofbroadband. Each business breakdown in this table contains tworows, one row for the sensitivity analysis related to the coefficienton broadband and one row for the sensitivity analysis of the spatial

Table 4Sensitivity analysis.

OLS 2SLS 2SLS OLSSpatial labroadban

Instrumental variables e Mean slope Provision 1999 e

All businesses 0.0613*** �1.9824 0.0763** 0.0475*Broadband lag all

business modele e e 0.0242***

Agriculture, forestry,fishing and hunting

0.0077 �4.2921 0.0113 �0.0054

Broadband lag ag. model e e e 0.0228Manufacturing 0.0139 �1.6116 0.0258* �0.0073Broadband lag manuf. model e e e 0.0371***Retail trade 0.0538*** �1.4984 0.0614*** 0.0412***Broadband lag retail model e e e 0.0221***Knowledge businesses 0.1252*** �4.0263 0.1235*** 0.1233***Broadband lag

knowledge modele e e 0.0034

Health care andsocial assistance

0.0902*** �2.1847 0.1113*** 0.0708***

Broadband lag health model e e e 0.0339***Rural businesses 0.0045 �0.0697 0.0021 0.0015Broadband lag rural model e e e 0.0064

***Significant at the 1% level.**Significant at the 5% level.*Significant at the 10% level.

lag of broadband. In models where the lag of broadband was notsignificant in the ordinary least squares models, further testing wasnot performed because this indicated spatial effects were notrelevant for that particular business type. The strength of in-struments presented in this table was cross-validated using severaldiagnostic tests including the F-test on excluded instruments, theKleibergen-Paap LM statistic for underidentification, and theKleibergen-PaapWald F-fest for weak identification. Shaded cells inthis table indicate a weak instruments issue. In Table 4, both meanslope and household density were identified as weak instruments.For the slope instrument, its poor performance is likely related tothe fact that the slope of terrain is also correlated with businessactivity, and it is difficult to build on very steep slopes. The poorperformance of this instrument could also stem from little variationacross counties. The original slope variable was computed fromfine-scale elevation 30-m cell information in a digital elevationmodel. As this information is aggregated to larger spatial scales,counties in this case, subtle variations in slope information aresmoothed. As regards the poor performance of household density,this is likely related to a similar spatial process operating on thedistribution of households and broadband which results in littlevariation in this variable for this scale of analysis. At this scale ofanalysis (i.e., counties) the temporal lag of broadband and thepoint-of-presence derived variables were the best instruments ofthe five examined.

8. Discussion and conclusion

The goal of this study was to examine the association betweenearly levels of broadband provision on future business activityacross a range of industries in the U.S. This is an important questionto consider because locales with competitive broadband marketslikely gained a distinct advantage in retaining and attracting busi-nesses. A secondary goal of this study was to evaluate the strength

gd

IV withspatial lagbroadband

IV withspatial lagbroadband

IV withspatial lagbroadband

IV withspatial lagbroadband

Mean slope,Provision 1999

Household Density,Provision 1999

Point of Presence,Provision 1999

Spatial LagPoint ofPresence,Provision 1999

�8.7893 �0.0895 0.0505 �0.025014.5469 0.2721 0.0425 0.1663***

e e e e

e e e e

�7.0255 0.4917 �0.0293 �0.027011.5698 �0.7644 0.0904 0.0867�6.6557 �0.3943 �0.0147 0.016311.0214 0.7477 0.1247** 0.0740e e e e

e e e e

�9.7759 �0.0880 �0.0525 �0.0090

16.2230 0.3270 0.2687*** 0.1974***e e e e

e e e e

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e6260

of instruments for broadband given the simultaneity betweenbroadband and business presence indicated in other studies [32];[26]. Model results highlighted industrial and spatial nuances tothis relationship. For most of the industries examined, early levelsof broadband provision had little impact on future business levels.The exception to this trend was knowledge intensive businesses.For this industry, initial broadband levels have a long-lasting rela-tionship on future levels of business activity. This likely stems froma need for frequent infrastructure updates in terms of platformsand speeds.

In terms of spatial nuances in the long-term impact ofbroadband on business levels, for all businesses and healthcarebusinesses, provision levels in neighboring counties (asmeasured by the spatial lag of broadband) have an enduringeffect on business activity. Overall then, these results do suggestsome first-mover advantage in broadband deployment and theability to retain and attract businesses. This is particularly trueregarding knowledge businesses and healthcare businesses thatrequire the latest, fastest broadband speeds for transmittinginformation. Results of the analysis of instruments also high-lighted variations in the different instruments proposed forbroadband. Mean slope and household density performed poorlycompared to time lags and the point-of-presence derivedinstruments.

This study represents a first look at the impact of early broad-band market dynamics on business activity over a decade later.That said, there are recommended extensions to this study toprovide additional resolution on the relationship between broad-band and businesses. The first extension is to examine the functionof broadband from a business service perspective, in addition to theevaluation of broadband from the infrastructure perspective pro-vided in this paper. In other words, do we see an association be-tween broadband and businesses as a result of business demand forthis service? Examining broadband from this side would helpevaluate whether broadband infrastructure helps promote growth,or is simply a result of growth.

The second extension is to subdivide the thirteen-year studyperiod into shorter sub-periods to understand when the advantagefrom the novelty of broadband ended for the industries examinedin this paper. This type of analysis will help pinpoint the momentin time when the rapid diffusion of Internet access rendered

Appendix ADescriptive statistics.

Variable Observations Mean

Change total establishments 3016 �0.00Change agricultural establishments 3016 �0.10Change manufacturing establishments 3016 �0.11Change retail trade establishments 3016 �0.13Change knowledge establishments 3016 �0.25Change health care establishments 3016 0.156Change rural establishments 808 �0.00Ln broadband providers 2000 3016 2.597Ln of highway miles 2000 3016 4.723Percent white 2000 3016 85.06Natural amenity index 1999 3016 0.049Percent bachelor's or higher 2000 3016 10.89Per capita personal income 1970 3016 3.144Herfindahl index in 2000 3016 0.153Average establishment size in 2000 3016 12.04Agglomeration economies in 1970 3016 78.87Population density 1970 3016 244.3

broadband, as a differentiating location factor between places,irrelevant. It is recommended that this be done as a series of cross-sectional models given changes in the entities required to reportinformation for the Form 477 database, and the varied spatial scalesat which broadband data is made available from the FederalCommunications Commission (FCC) and the National Telecommu-nications and Administration (NTIA).

A third extension is an evaluation of the impact of the level ofdetail about elevation on the performance of mean slope as aninstrument for broadband. This is necessary because the level ofspatial resolution of the initial slope information could impactthe processed value of slope; the less initial detail provided, thegreater the level of smoothing in the final slope calculation. Thisstudy used 30-m cell data to compute slope information as aninstrument for broadband. More detailed 1:1,000,000 scaleelevation data with 100-m spatial resolution can be downloadedas a single file from the United States Geological Survey (USGS).While potentially valuable, there are significant time tradeoffs tousing finer-grained slope information. As discussed previously,the time to compute slope information for the 30-m data wasextensive and took several weeks. While access to the USGSslope data have improved and computational tools such asCyberGIS have the potential to streamline the data processing, itremains a “Big Data” problem that is subject to considerableinvestigation [41].

Aside from the use of finer-grained slope data, morework is alsoneeded to understand how and why businesses use the Internet inday-to-day operations. While model results speak to a statisticalrelationship between broadband and businesses, the relationshipbetween the two is much broader than what is discussed in thepresent paper. Other issues to consider include the type of businessexamined, and the basic skills and business experience of em-ployees. They also include how and why businesses use broadbandin their business processes. Thus, a key facet of the broadband-business nexus is additional work to collect usage data so re-searchers and policymakers are better able to assess multiple as-pects of access and use of broadband Internet connectionssimultaneously. In this regard, the social science community hasmuch to offer in terms of the range of theories and techniques thatmay be brought to bear on issues of use from social, economic, andpsychological perspectives.

Std. dev. Min Max

52011 0.2639395 �0.6359887 11.272518438 0.5701 �2.302585 2.079440715 0.3342 �2.397895 1.958725304 0.2159 �1.609438 2.395985421 0.7809 �4.043051 2.53266 0.2877 �1.386294 2.582564 0.0580 �0.2542165 0.84790 1.0687 0 7.6639198 0.6315 0 7.917944 15.8083 4.5083 99.73928 2.2939 �6.4 11.17779 4.857 2.47 40.022 0.8277 1 8.53662957 0.0517 0.014696 0.811104 4.8185 0 93.3015037 1739.3880 0 95001.6064 2510.4910 0.1961 1,08,975

Appen

dix

BOrdinaryLe

astSq

uares

(OLS

)resu

lts. A

llbu

sinesses

Agriculture,forestry,

fish

ingan

dhunting

Man

ufacturing

Retailtrad

eKnow

ledge

businesses

Hea

lthcare

and

social

assistan

ceRuralb

usinesses

Lnbroa

dba

ndproviders20

000.06

13***(0.022

1)0.00

77(0.016

3)0.01

39(0.010

5)0.05

38***(0.008

7)0.12

52***(0.023

5)0.09

02***(0.010

2)0.00

45*(0.002

8)Resea

rchUniversity

�0.008

8(0.061

3)�0

.074

8*(0.043

5)�0

.036

0**(0.017

6)�0

.044

6***

(0.017

1)�0

.004

4(0.046

5)�0

.084

1***

(0.018

3)�0

.001

6(0.013

0)Ln

ofhighway

mile

s�0

.000

1(0.011

0)�0

.044

4*(0.024

2)�0

.016

6(0.014

4)0.00

46(0.009

3)0.16

66***(0.035

0)0.00

94(0.013

9)�0

.009

5*(0.004

9)Pe

rcen

twhite

0.00

10(0.000

7)�0

.000

2(0.000

8)0.00

20***(0.000

6)0.00

12***(0.000

4)0.00

18*(0.001

0)0.00

14***(0.000

5)0.00

00(0.000

3)Naturala

men

ityindex

0.00

72*(0.004

4)�0

.034

0***

(0.008

3)0.00

15(0.005

8)0.00

36(0.003

5)�0

.027

6**(0.011

9)0.00

32(0.004

3)0.00

51***(0.001

9)Pe

rcen

tba

chelor'sor

higher

0.01

00***(0.003

7)0.01

66***(0.003

2)0.00

72***(0.001

7)0.01

04***(0.001

4)0.01

56***(0.004

0)0.01

36***(0.001

7)0.00

23***(0.000

6)Pe

rcapitapersonal

inco

me

�0.016

2(0.018

2)�0

.018

5(0.023

0)�0

.045

8***

(0.013

6)�0

.024

0***

(0.008

3)0.02

94(0.031

7)�0

.079

9***

(0.014

0)0.00

22(0.004

8)Herfindah

lindex

0.21

56(0.298

9)�0

.586

7***

(0.212

7)�0

.474

9*(0.247

4)0.08

70(0.152

3)�0

.305

4(0.440

5)0.01

93(0.131

7)�0

.054

2(0.048

8)Ave

rage

establishmen

tsize

�0.005

4(0.006

9)0.00

34(0.002

7)0.00

08(0.002

1)0.00

13(0.001

1)�0

.004

1(0.004

6)�0

.000

4(0.001

4)0.00

00(0.000

7)Agg

lomerationecon

omies

�1.86E

-06*

(1.01E

-06)

�8.41E

-06***

(2.52E

-06)

�3.14E

-06***

(8.68E

-07)

2.37

E-05

***(1.65E

-06)

�1.45E

-06(2.74E

-06)

2.53

E-06

*(1.47E

-06)

e

Population

den

sity

�1.45E

-06(1.88E

-06)

�1.03E

-05***

(2.37E

-06)

�2.94E

-06***

(1.06E

-06)

1.54

E-06

(2.93E

-06)

�3.14E

-06(3.32E

-06)

5.84

E-08

(2.43E

-06)

�3.44E

-08(2.35E

-07)

Con

stan

t�0

.182

9***

(0.065

5)0.04

42(0.170

8)0.10

36(0.111

1)�0

.409

2***

(0.064

2)�1

.423

5***

(0.202

7)0.07

01(0.087

9)0.03

81(0.037

7)F-statistic

22.65

6.77

12.41

75.89

9.28

12.91

e

Deg

rees

offree

dom

2957

2957

2957

2957

2957

2957

753

Roo

tMSE

0.24

710.54

690.31

740.19

020.71

880.26

090.05

56

Rob

ust

stan

darderrors

arein

italics.

***Significantat

the1%

leve

l.**Sign

ificantat

the5%

leve

l.*Significantat

the10

%leve

l.

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e62 61

References

[1] Allen JC, Koffler ELV. The telecommunications act of 1996: its implementationin the U.S. South. Southern Rural Development Center (SRDC); 1999. http://www.ext.msstate.edu/srdc.

[2] Atkinson R. Market structure for ultrabroadband. Communications & strate-gies. 2008. Special Issue, November 2008, http://papers.ssrn.com/sol3/papers.cfm?abstract_id¼1374334.

[3] Augereau A, Greenstein S. The need for speed in emerging communicationsmarkets: upgrades to advanced technology at Internet Service Providers. Int JInd. Organ 2001;19(7):1085e102.

[4] Autor DH, Levy F, Murnane RJ. The skill content of recent technologicalchange: an empirical exploration. Q J Econ 2003;118:1279e333.

[5] Bresnahan TF, Trajtenberg M. General purpose technologies ‘Engines ofgrowth’? J Econ 1995;65(1):83e108.

[6] Datta A, Agarwal S. Telecommunications and economic growth: a panel dataapproach. Appl Econ 2004;36:1649e54.

[7] Desilver D. Chart of the Week: the ever-accelerating rate of technologyadoption. Pew Research Center; 2014. March 14, 2014, http://www.pewresearch.org/fact-tank/2014/03/14/chart-of-the-week-the-ever-accelerating-rate-of-technology-adoption/.

[8] CBN. Community broadband networks. Institute for Local Self-Reliance; 2015.http://www.muninetworks.org/communitymap.

[9] Dholakia RR, Harlam B. Telecommunications and economic development.Telecommun Policy 1994;18(6):470e7.

[10] FCC. National broadband plan. Federal Communications Commission; 2010.https://www.fcc.gov/general/national-broadband-plan.

[11] Forman C, Goldfarb A, Greenstein S. Which industries use the Internet? AdvAppl MicroeconInnov 2003;12:47e72.

[12] Entorf H, Kramarz F. The impact of new technologies on wages: lessons frommatching panels on employees and on their firms. Econ Innovation NewTechnol 1998;5(2e4):165e98.

[13] Gillett S, Lehr W, Osorio C, Sirbu MA. Measuring Broadband's economicimpact. National Technical Assistance Training, Research, and EvaluationProject; 2006. Technical Report 99-07-13829, http://cfp.mit.edu/publications/CFP_Papers/Measuring_bb_econ_impact-final.pdf (accessed July 20, 2011).

[14] Goldsmith Stephen, Crawford Susan. The responsive city: engaging commu-nities through data-smart governance. John Wiley & Sons; 2014.

[15] Grubesic TH. The broadband provision tensor. Growth Change 2015;46(1):58e80.

[16] Grubesic TH, Mack EA. Broadband telecommunications and regional devel-opment. London: Routledge; 2016.

[17] Glaeser EL, Tobio K. The rise of the sunbelt (No. w13071). National Bureau ofEconomic Research; 2007.

[18] Grubesic TH, Murray AT. Constructing the digital divide: spatial disparities inbroadband access. Pap Reg Sci 2002;81(2):197e221.

[19] Grubesic TH, O'Kelly ME. Using points of presence to measure accessibility tothe commercial Internet. Prof Geogr 2002;54(2):259e78.

[20] Hales B, Gieseke J, Vargas-Chanes D. Telecommunications and economicdevelopment: chasing smokestacks with the Internet. In: Korsching PF,Hipple PC, Abbott EA, editors. Having all the right connections: telecommu-nications and rural viability. London: Praeger; 2000. p. 257e76.

[21] Grubesic TH. A spatial taxonomy of broadband regions in the United States. InfEcon Policy 2006;18:423e48.

[22] Kandilov IT, Renkow M. Infrastructure investment and rural economicdevelopment: an evaluation of USDA's Broadband Loan Program. GrowthChange 2010;41:165e91.

[23] Kang. F.C.C. Approves broadband subsidy for low-income households. 2016.The New York Times March 31, 2016 http://www.nytimes.com/2016/04/01/technology/fcc-approves-broadband-subsidy-for-low-income-households.html?_r¼0.

[24] Hu WM, Prieger JE. The timing of broadband provision: the role of competi-tion and demographics. 2007. AEI-Brookings Joint Center Working Paper,(07e06).

[25] Katz R. The impact of broadband on the economy: research to date and policyissues. Broadband Ser 2012. https://www.itu.int/ITU-D/treg/broadband/ITU-BB-Reports_Impact-of-Broadband-on-the-Economy.pdf.

[26] Kolko J. Broadband and local growth. J Urban Econ 2012;71:100e13.[27] Kolko J. A new measure of US residential broadband availability. Telecommun

Policy 2010;34:132e43.[28] McGranahan D, Wojan T. Recasting the creative class to examine growth

processes in rural and urban counties. Reg Stud 2007;41(2):197e216.[29] Mack EA. Variations in the broadband-business connection across the urban

hierarchy. Growth Change 2015;46(3):400e23.[30] Mack EA. Broadband and knowledge intensive firm clusters: essential link or

auxiliary connection? Pap Reg Sci 2014;93(1):3e29.[31] Mack EA, Faggian A. Productivity and broadband: the human factor. Int Reg

Sci Rev 2013;36(3):392e423.[32] Mack EA, Anselin L, Grubesic TH. The importance of broadband provision

to knowledge intensive firm location. Reg Sci Policy Pract 2011;3(1):17e35.

[33] Koutroumpis P. The economic impact of broadband on growth: a simulta-neous approach. Telecommun Policy 2009;33:471e85.

[34] Sohn J, Kim TJ, Hewings GJD. Information technology and urban spatial

E.A. Mack, E. Wentz / Socio-Economic Planning Sciences 58 (2017) 51e6262

structure: a comparative analysis of the Chicago and Seoul regions. Ann RegSci 2003;37(3):447e62.

[35] Mack EA, Rey SJ. An econometric approach for evaluating the linkages be-tween broadband and knowledge intensive firms. Telecommun Policy2014;38(1):105e18.

[36] Solow RM. A contribution to the theory of economic growth. Q J Econ1956;70(1):65e94.

[37] Sullivan M. 4G wireless speed tests: which is really the Fastest? PCWorld.2011. http://www.pcworld.com/article/221931/4g_speed_tests.html.

[38] Thompson H, Garbacz C. Broadband impacts on state GDP: direct and in-direct impacts. In: Paper presented at the international telecommunicationssociety 17th biennial conference. Montreal, Canada: United States Geolog-ical Survey (USGS) National Map Viewer; 2008. http://viewer.nationalmap.gov/viewer/.

[39] Warf B. Telecommunications and the globalization of financial services. ProfGeogr 1989;41(3):257e71.

[40] Yilmaz S, Haynes KE, Dinc M. The Impact of telecommunications infra-structure investment on sectoral growth. Australas J Reg Stud 2001;7:383e97.

[41] Wang S. A CyberGIS framework for the synthesis of cyberinfrastructure, GIS,

and spatial analysis. Ann Assoc Am Geogr 2010;100(3):535e57.

Elizabeth Mack is an Assistant Professor in the Department of Geography at MichiganState University (MSU) and adjunct faculty at Arizona State University and the Centerfor Global Change and Earth Observations at MSU. Her research is focused on under-standing the impact of broadband infrastructure deployment on businesses and newentrepreneurial ventures. This includes issues stemming from basic access as well asthe use of Internet applications and social media in business processes.

Elizabeth A. Wentz is Dean of Social Science in the College of Liberal Arts and Sciences,Associate Director for the Institute of Social Science Research, and Professor in theSchool of Geographical Sciences and Urban Planning at Arizona State University. Herresearch focuses on the development and implementation of geographic technologiesdesigned to establish better understanding of the urban environment. She earned herPhD in Geography from the Pennsylvania State University, her MA in Geography fromThe Ohio State University, and her BS in Mathematics from The Ohio State University.