بهره وری در ارائه خدمات پهنای باند : تجزیه و تحلیل مکانی

With the 2008 Federal Communications Commission (FCC) order amending both the definition of broadband and its data collection practices, the problems associated with data integrity and ZIP code aggregations in the United States will soon be forgotten. However, between 1999 and 2007, FCC Form 477 data remain the only viable, nationwide database of broadband provision in the United States. While broadband data from 1999–2004 and 2005–2007 are not directly comparable due to a modification in collection procedure, there is an absence of objective empirical analysis for the latter time period. Interestingly, although the FCC made the 2005–2007 data publically available on the Internet, password protected files largely prevented analysts from accessing, manipulating and analyzing these data. The purpose of this paper is three-fold. First, the process utilized for liberating these data from their protected format and integrating them into a geographic information system (GIS) is outlined. Second, the spatial distribution of broadband provision in the United States for 2005–2007 is explored. A mathematical programming approach is also utilized for comparing the relative efficiencies of ZIP code areas in acquiring broadband service given their demand-side socio-economic and demographic determinants. Finally, implications for public policy, particularly those associated with developing local and regional benchmarks for broadband provision, competition and access, are addressed.

In June 2008, the Federal Communications Commission (FCC) released a Report and Order and Further Notice of Proposed Rulemaking (Form 477 Order) that modified several important elements of the Form 477 data collection program for broadband (FCC, 2008). First, providers are required to document and report broadband speed data (upload and download) for customers and categorize them into one of three tiers, ranging from 200 kbps to 10 mbps. Second, the Form 477 Order requires that mobile wireless broadband providers report the number of subscribers whose data plans permit access to the Internet. Third, and perhaps most importantly, Form 477 is modified to require that broadband providers report the number of broadband connections at the Census tract level. These changes represent a marked departure from existing broadband data collection efforts by the FCC. Between 1999 and 2007, there were two basic versions of the broadband provision data available for public use. Between 1999 and 2004, the FCC required that all facilities-based providers with at least 250 high-speed lines in a single state file a Form 477 report.1 However, beginning in June 2005, the FCC modified its reporting requirements, mandating that all providers with at least one high-speed line in a state file a Form 477 report. As noted by Kolko (2007), while the 2005–2007 data are more complete, comparing them to 1999–2004 data is impossible. One of the major caveats associated with all of the Form 477 data was the use of ZIP codes as the reporting units. As noted by Grubesic, 2006 and Grubesic, 2008, Flamm and Chaudhuri (2007) and many others (GAO, 2006; Flamm, Friedlander, Horrigan, & Lehr, 2007), ZIP codes are highly dynamic and fail to provide an empirically sound unit for reporting broadband provision. As a result, the change in policy mandating that providers document broadband connections at the Census tract level is a welcome one. While the problems associated with broadband data collected and reported at the ZIP code level will soon be forgotten, the FCC Form 477 data remain the only viable, nationwide database of broadband provision in the United States between 2005 and 2007. Interestingly, although the FCC made the 2005–2007 data publically available on the Internet, their format differed from the 1999–2004 data. Instead of making broadband provider counts available in a user-friendly, Microsoft Excel worksheet (.xls), the FCC released the 2005–2007 data in a password protected Adobe Acrobat (.pdf) document. This format effectively prevented analysts from accessing, manipulating and analyzing these data. In several email exchanges with officials at the FCC, their response to requests for obtaining these data in an alternative format or the existing format without encryption was, “we have not prepared an .xls or other easily sortable/searchable version of the recent files… [and] there's no intention to revisit the decision to allocate available staff time to producing only one version of the list” (Burton, 2007). It is not surprising, therefore, that there is not a single published researched article dealing with the 2005–2007 data. This would require an analyst to manually enter over 32,000 records (one for each ZIP code in the United States) for each reporting period using the password protected FCC documents. With these limitations in mind, the purpose of this paper is three-fold. First, it develops and outlines an easily repeatable and automated process for liberating these data from their protected format and integrating them into a geographic information system (GIS). Second, the spatial distribution of broadband provision in the United States for 2005–2007 is explored and a mathematical programming approach is utilized for comparing the relative efficiencies of ZIP code areas in acquiring broadband service given their demand-side socio-economic and demographic determinants. Finally, implications for public policy, particularly those associated with developing local and regional benchmarks for broadband provision, competition and access, are addressed.

There are several aspects of the results worth further discussion. First, the spatial distribution of broadband provision in the United States continues to be relatively dynamic, although between 2005 and 2007 it slowed to some extent, particularly when compared to the period between 1999 and 2004 (Grubesic, 2008). For example, while there are fractional changes in the composition and number of ZIP code areas provided with service, a net increase of 0.97% points is not remarkable. That said, the overall intensity of broadband provision is increasing, with the aggregate count of non-unique providers offering service between 2005 and 2007 increasing by 58,869. Again, while this suggests that the addition of new markets is slowing, levels of competition in existing markets are increasing. It is also important to note that these conclusions are somewhat tempered by the problems associated with measuring broadband provision at the ZIP code level. As noted previously, although broadband providers may be present in a ZIP code area, this does not suggest that availability or competition is ubiquitous throughout. Fortunately, these problems with spatial resolution will largely be resolved when broadband provision data becomes available at the Census tract level. Issues associated with the efficiency of broadband provision, as determined by demand-side metrics, are also intriguing. Perhaps the most important information revealed by this analysis is that broadband core areas and their constituent components are not equal in their ability to attract advanced services. While there is little doubt that all of the ZIP code areas located in the broadband core have a relatively strong selection of broadband providers, the results of the DEA models suggest that some locales are more efficient than others. For example, the broadband core region associated with Atlanta, GA had three technically efficient ZIP code areas and several more considered highly efficient during 2007 for Model 2. Areas not achieving technical efficiency were determined to have excess levels of demand-side inputs—often displaying decreasing returns to scale. Simply put, one would expect these areas to have more providers than they have, when compared to technically efficient ZIP code areas. In part, this is why both DEA models recommended a reduction in demand-side inputs for improving non-technically efficient locales. More importantly, while the models were consistent in their recommendations for demand-side inputs, it is clear that Model 2 provided a more realistic snapshot of broadband efficiency, particularly for suburban areas. As noted previously, one of the problems with the more naïve model (Model 1) and its use of only three demand-side variables, is that it did not account for many of the less obvious dynamics associated with broadband provision—particularly commercial (e.g., small- and medium-sized enterprises) versus residential demand. The naïve demand-side DEA model suggested that the central portions of metropolitan areas in the broadband core were most efficient—with a gradual decay of efficiency as one moved toward suburban and exurban areas. While this is a reasonable result when using three variables for inputs, it is likely that this model was simply capturing the overall spatial dynamics associated with US population distributions (i.e., higher in the center of metropolitan areas with decreasing densities and counts as one moves to suburban and exurban areas). Obviously, this fails to account for the influence of suburban employment centers, exemplified by Alpharetta, GA. Conversely, when the more complex DEA model that incorporated additional demand-side parameters (including metrics for industrial composition) was run, the results suggested that many suburban areas were, in fact, technically efficient. These conclusions were reinforced by Fig. 8, which displayed efficiency clusters for the broadband core. In most cases, metropolitan portions of the broadband core are defined by a relatively good mix of high–high (efficient) and low–low (less than efficient) clusters, although there was no systematic pattern for all regions in the US Denver, Salt Lake City, Tucson and several other metropolitan areas provide good counterpoints in this instance, displaying a more homogenous composition than many of the other metropolitan regions with broadband core areas. While there may be some debate concerning the choice of demand-side determinants for modeling efficiency, there is little doubt that incorporating a combination of demand and supply-side determinants would likely provide the most complete and accurate picture of efficiencies associated with broadband provision. Fortunately, the inclusion of these variables can be readily accommodated by DEA. This is reinforced by the significant differences in the spatial distribution of modeled efficiencies between Models 1 and 2. As noted earlier, the incorporation of metrics for industrial composition made a big difference in differentiating between locales that were considered efficient in attracting broadband provision and those that were not. Finally, there are several policy-related issues worth mentioning. As illustrated by the results of this study and others, broadband provision is an inherently local problem (Gillett, Lehr, & Osorio, 2004). While the provision of broadband in the United States is largely influenced by federal policies (e.g. Telecommunications Act of 1996) (Gabel, 2007; Grubesic & Murray, 2004; Prieger, 2003), evaluating its associated spatial outcomes is scale-dependent. This paper opted for a global comparison of efficiencies between all constituent units within the broadband core—in effect, a cross-sectional analysis of efficiency. While this provides a good set of benchmarks against which all broadband core areas can be compared, the results also revealed significant heterogeneities within core areas. It is likely that if the same analysis was conducted for the periphery or any of the other cluster groups in the spatial taxonomy of regions, heterogeneities would also appear. This leaves policy analysts at a crossroads. Does one evaluate performance between areas or within them? The most likely answer to this question is both. On a global scale, policy analysts, economic development officials and local governments are often interested in determining how their supply of essential infrastructure (e.g. broadband) compares to other regions. This type of information is important for regional development efforts as allied agencies attempt to deploy the appropriate mix of resources to attract businesses in an increasing competitive global economy (Gulati, Nohria, & Zaheer, 2000; Mack & Grubesic, 2009). As the results of this paper indicate, the use of DEA allows one to benchmark regional performance in attracting broadband provision—providing a more complete picture of regional status and competitiveness. In turn, this can be used to more effectively develop policies or related interventions that attempt to mitigate infrastructure deficiencies because a reliable performance benchmark is in place. Not surprisingly, the same thing can be accomplished at a more localized level within core areas, although the impetus for asking questions concerning local broadband provision can differ. While issues concerning regional development persist at the local level, issues of equity and availability of broadband for all residents within a community are salient. As highlighted in Section 3, the broadband landscape is far from homogenous, particularly for smaller and more geographically remote communities. In this instance, DEA could be used to evaluate which portions of a community are more efficient in attracting broadband services. With these results in hand, local governments may wish to contact service providers to determine a course of action for mitigating any gaps or deficiencies in broadband provision. Information generated from the DEA models would also be helpful in this regard. Recall that there were some locales where the DEA models recommended a reduction in median income to improve efficiency. Again, while this should not be viewed as a literal course of action, it would provide community development officials some assurance that the lack of provision is not necessarily related to supply-side factors. Given this information, if some type of mitigation was still deemed appropriate, local government broadband initiatives to deploy publicly owned communications infrastructure could be a viable option (Gillett et al., 2004). In conclusion, this paper makes three contributions to the existing literature. First, it describes a methodology for extracting a difficult-to-use database of broadband provision from the FCC. Second, using a variety of statistical approaches and a geographic information system, the spatial distribution of broadband provision is documented for the United States between 2005 and 2007, highlighting major changes during this time frame. Results suggest that while the spatial diffusion of broadband is slowing, the intensity of competition within areas already served is increasing. Finally, this paper presents a data envelopment analysis of the broadband core areas in the United States for determining how efficient they are in attracting providers in June 2007. Based on the most sophisticated DEA model and its utilized demand-side determinants, the results indicate a relatively heterogeneous distribution of efficiency within core areas. Specifically, while many central cities and suburban communities with high levels of information intensive industries are the most efficient, there are many spatially proximal suburban and exurban areas which are not. As broadband technologies continue to evolve, issues of provision, affordability, competition and access will remain important. As illustrated by this paper, both spatial analytical techniques and data envelopment analysis are extremely useful approaches for monitoring the rollout of advanced services and benchmarking local and regional efficiencies in attracting them. It is hoped that this paper will serve as a starting point for additional analyses on these topics and that both academics and practitioners will benefit from the methods and results presented.