Content uploaded by Bento J. Lobo
Author content
All content in this area was uploaded by Bento J. Lobo on May 31, 2019
Content may be subject to copyright.
1
Broadband Speed and Unemployment Rates: Data and
Measurement Issues
May 2019
Bento J. Lobo*
The University of Tennessee at Chattanooga,
Department of Finance and Economics,
615 McCallie Avenue, Chattanooga TN 37403, USA.
Bento-Lobo@utc.edu
Md. Rafayet Alam
The University of Tennessee at Chattanooga,
Department of Finance and Economics,
615 McCallie Avenue, Chattanooga TN 37403, USA.
Rafayet-Alam@utc.edu
Brian E. Whitacre
Oklahoma State University,
Department of Agricultural Economics,
504 Ag Hall, Stillwater, OK 74078, USA.
Brian.Whitacre@okstate.edu
*Contact author: Bento-Lobo@utc.edu; +1-423-425-1700
2
Broadband Speed and Unemployment Rates: Data and
Measurement Issues
Abstract
We examine the effects of broadband speed on county unemployment rates in the U.S. state of
Tennessee. We merge the older National Broadband Map dataset and the newer FCC dataset in
lengthening our broadband access data over the period 2011-2015. Extending the dataset improves
the precision of the estimates. Our panel regressions control for potential selection bias and reverse
causality and show that broadband speed matters: unemployment rates are about 0.26 percentage
points lower in counties with high speeds compared to counties with low speeds. Ultra-high speed
broadband also appears to reduce unemployment rates; however, we are unable to distinguish
between the effects of high and ultra-high speed broadband. We document beneficial effects of the
early adoption of high speed broadband on unemployment rates. Better quality broadband appears
to have a disproportionately greater effect in rural areas.
JEL Classification: E24, O18, J64, C23, D12
Keywords: Broadband speed; unemployment rates; selection bias; endogeneity; rural counties
3
Introduction
As internet connectivity becomes ubiquitous, attention in the U.S. has shifted from expanding
access to the internet towards improving the connections of users. The Federal Communications
Commission (FCC) reports that over 92 percent of the U.S. population has access to a fixed
broadband connection with download speeds of at least 25 megabits per second (Mbps) (FCC,
2018a). However, only one in four U.S. households has access to state-of-the-art technologies such
as optical fiber which delivers the fastest symmetrical internet access. The National Broadband
Plan specifically addresses the need for speed by stipulating that 100 million households or more
should have download speeds of at least 100 Mbps and upload speeds of at least 50 Mbps (Mack,
2014).
Academic research has shown that broadband penetration has important economic impacts.
However, Middleton (2013) argues that features other than penetration, such as speed and quality
of service are also important determinants of the effects of broadband. Kongaut and Bohlin (2017)
point out that academic studies on speed are sparse and suggest the need for micro-level studies to
serve as guides for policy-makers since they provide more detail on a specific area. We pick up on
this suggestion in attempting to further contribute to our understanding of the effects of broadband
speed on local economic outcomes. While others have used national and foreign data sets, we
examine the effects of broadband speed on unemployment rates in the U.S. state of Tennessee.
Our study is motivated by the fact that a Tennessee city, Chattanooga, was the first city in the
Western hemisphere to offer gigabit speed broadband to households and businesses in 2011. This
study examines the effects of high speed (defined as 100 Mbps) and ultra-high speed (1,000 Mbps)
broadband on county unemployment rates from 2011 to 2016. Our approach circumvents inter-
state differences in broadband policy and is the first study, to our knowledge, to merge broadband
access data from the National Broadband Map (2011 to 2014) and the FCC (post 2014).
Why is broadband speed important?1 Presumably, broadband facilitates efficiency, heightens
productivity and, likely, fosters innovation. The move from dial-up to broadband internet
connectivity has shown positive impacts on productivity, growth, employment and poverty levels
(Katz et al. 2010; Czernich et al. 2011; Rohman and Bohlin, 2012; Whitacre et al. 2014b).
Middleton (2013) and Mack (2014), however, argued that not merely the presence of broadband
but the speeds available are likely to be an important consideration when developing strategic plans
to enhance the business climate of locations. Similarly, a former FCC Commissioner emphasized
the need for gigabit speeds to encourage U.S. innovation (Genachowski, 2013). One analysis
concluded that high-bandwidth applications tend to overwhelm mobile data plans and slow
connections. This limits or even cuts off many families from e-commerce, banking, health care
and other services (Coren, 2016). Lobo (2015) found that high speed broadband not only had
impacts on employment and income in Hamilton county, Tennessee, but also in applications to
1 To clarify, broadband speed is determined by bandwidth and latency. Bandwidth is the amount of data that can be
transferred in a second and is measured in bits per second. It is usually the same as speed when downloading files, but
may not always be the same in real time applications (e.g. videoconferencing). Speed is also dependent on (low)
latency or delay, i.e. how much time it takes for a packet of information to travel from source to destination. Latency
is measured in milliseconds and is a function of the electrical characteristics of the circuit. From a practical standpoint,
broadband speed is most often characterized in bits per second, i.e. in terms of bandwidth. Latency, by contrast, is not
systematically recorded or reported in the same fashion as bandwidth by the FCC or any other agency. Due to this
data limitation, our study uses the terms speed and bandwidth interchangeably.
4
medicine, banking, education, and in the evolution of an entrepreneurial ecosystem, bearing out
the narratives in Koutrompis (2009) and Qiang et al (2009). Other positive effects pertained to
disaster recovery services for businesses and benefits stemming from intelligent traffic systems. A
major benefit cited in the study stemmed from reduced outage minutes due to weather disruptions
which was made possible by smart grid technology. Despite these findings, the effects of
broadband speed on the broader economy have been understudied. This is mostly because speed
innovations in broadband have emerged in a short time span and consequently, the length of time
series data on higher broadband speed access is short. Furthermore, the dearth of adoption/use data
has constrained quantitative analysis of the conditions under which broadband has an economic
effect.
The FCC defines broadband as, “…high-speed Internet access that is always on and faster than the
traditional dial-up access” (FCC, 2014). Over time, our perceptions of what constitutes a “fast”
Internet connection have changed.2 As consumer and business uses of the Internet evolve, and new
applications become more deeply embedded into everyday life, higher speeds frequently shift from
being a luxury to a requirement for many users. In fact, as broadband access becomes ubiquitous,
it becomes expected infrastructure and other factors (such as the quality of that access) become
more influential drivers of economic growth. Since broadband serves as an enabler of remote
information technology access, its implementation could affect organizational and process changes
in local enterprises, and thus could have indirect effects on economic outcomes (International
Telecommunication Union, 2012). It is also possible that broadband speed serves as a proxy for
other factors, such as agglomeration benefits (Mack, 2014). Holt and Jamison (2009) point out that
measuring the economic effects of such an investment is difficult because of the need to isolate
the unique effects associated with this infrastructure.3
Moreover, it is unclear whether there is a linear relationship between broadband speeds and
economic impact. It is conceivable that higher bandwidth may be associated with declining returns
to scale, i.e. an inverted-U curve (see e.g. Kongaut and Bohlin, 2017 - Table 9, and Stocker and
Whalley, 2016). Yet, anecdotal evidence points to the converse: gigabit broadband will allow the
development and deployment of high-value applications which cannot be delivered in any other
way, suggesting additional bandwidth carries considerable returns. Atkinson et al. (2009) point to
four main areas that benefit from enhanced bandwidth: file transfer, video streaming, real-time
communication, and the simultaneous use of multiple applications.
This study focuses on the potential incremental impacts of broadband speed on county
unemployment rates. We study 95 counties in the state of Tennessee over the period 2011 to 2016.
According to BroadbandNow statistics, Tennessee is the 23rd most connected state in the U.S. with
172 internet providers (the average for the U.S. is 153). The major ISPs in Tennessee are AT&T
Internet, CenturyLink, Xfinity from Comcast, Charter Spectrum and EPB (BroadbandNow, 2019).
2 For example, in 2000 the Federal government defined broadband as any service with a download speed of 200
kilobits per second (kbps) or faster. In 2010, the FCC redefined “basic” broadband service as a connection with speeds
of at least 4 Mbps downstream – 20 times faster than the 2000 definition – and at least one Mbps upstream. The current
FCC benchmark for broadband is Internet download speed of 25 Mbps or faster and upload speed of 3 Mbps or faster
(BroadbandNow, 2018).
3Lobo (2015), for instance, asks in the context of Hamilton county, Tennessee, “..how do we attribute [investments
and] jobs to particular features [e.g. broadband] of a location when such answers are not elicited from relocating
firms?” (p.13) In his qualitative analysis, Lobo reports that ultra-high speed broadband in the county accounted for at
least 2,800 new jobs.
5
A comparison of current broadband coverage in Tennessee relative to the U.S. as a whole suggests
that Tennessee’s broadband access appears to be fairly similar to the nationwide averages with
respect to wired broadband coverage, average download speed, population access to 25 Mbps, 100
Mbps and gigabit speed service, and underserved population (i.e. those with access to less than 2
wired providers).
We use data on broadband access from the FCC Form 477 and National Broadband Map (NBM)
databases and estimate a two-way fixed effects or difference-in-difference model. Our research is
similar to some work previously done in this area (Shideler and Badaysan, 2012; Mack, 2014;
Lapoint, 2015). However, we add to the literature by examining current levels of high speed and
ultra high speed broadband, and by studying the effects of incremental speed levels available in
various counties. We merge the older National Broadband Map (NBM) dataset (2011-2014) and
the newer FCC dataset (2015 on), identify data and measurement issues involved in this process,
and demonstrate the value of lengthening the time series of broadband access on the estimated
coefficients. We also examine the benefits of early adoption of fast broadband on unemployment
rates. Our empirical specification controls for selection bias and for an aspect of endogeneity, i.e.
reverse causality. We use panel data models with county and year fixed effects. We use the lags
of the explanatory variables to address possible endogeneity and cluster the standard errors at the
county level to address the potential issues related to heteroscedasticity and autocorrelation in the
error terms.
We find that high speed broadband has significant effects on county-level unemployment rates;
however, we were unable to distinguish between the effects of high and ultra-high speed tiers. We
also find measurable benefits to early adoption of high speed broadband. Compared to urban areas,
the benefits of better quality broadband are disproportionately greater in rural areas.
The rest of this paper is set out as follows: Section II briefly describes broadband efforts in the
state of Tennessee; section III summarizes the relevant literature; section IV describes the
methodology and data; section V presents the empirical results, and section VI concludes.
II. Broadband Efforts in Tennessee
Broadband access initiatives in the U.S. have mostly stemmed from Federal, rather than state,
efforts. In particular, On March 16, 2010, the Federal Communications Commission (FCC)
released “Connecting America: The National Broadband Plan.” Mandated by the American
Recovery and Reinvestment Act of 2009, the FCC’s National Broadband Plan (NBP) is mandated
to “seek to ensure that all people of the United States have access to broadband capability.” The
NBP identified significant gaps in broadband availability and adoption in the United States, and in
order to address these gaps and other challenges, the NBP set six specific goals to be achieved by
the year 2020 (FCC, 2010). Speeds of 100 Mbps in 100 million homes (popularly referred to as
“100 squared”) would constitute next-generation broadband in most U.S. households. As a
milestone, the FCC set an interim goal of 100 million homes with actual download speeds of 50
Mbps and actual upload speeds of 20 Mbps by 2015. The FCC noted that it was likely that 90% of
the country would have access to advertised peak download speeds of more than 50 Mbps by 2013.
Regarding broadband adoption, the NBP set an adoption goal of “higher than 90%” by 2020.
6
Tennessee, like many other states, uses a variety of grant and incentive programs to build out
broadband coverage, especially in rural areas. This usually takes the form of giving Connect
America Fund (or “CAF”) grants to ISPs that promise to provide a certain level of service to an
area. The CAF grant pays for construction so the ISP can logistically afford to serve a sparsely
populated or otherwise challenging area. This effort is funded by an industry tax that is generally
passed on to consumers in broadband bills often as a “regulatory fee” or “universal service fee”
BroadbandNow, 2017). Since 2010, Connected Tennessee has been awarded about $4.5 million in
federal grants for Tennessee’s Broadband Initiative. Other federal incentive programs include
Community Connect broadband grants, Rural Broadband Access loans, and the Distance Learning
/ Telemedicine program – all run by the United States Department of Agriculture’s (USDA’s)
Rural Utilities Service. Kruger (2018) reports that Tennessee was awarded over $240 million
across these grants from 2009-2016, accounting for 3.3% of all USDA broadband awards.
In 2005, EPB (formerly, The Electric Power Board), a city-owned utility in Chattanooga,
developed a strategic plan to build out a new fiber optic infrastructure in the community to
modernize the electric system and provide fiber-to-the-home (FTTH) telephone/internet/TV
service to residential and commercial customers. At the time, internet service was provided to the
area primarily by BellSouth, AT&T and Comcast. Shortly after receiving approval from
Chattanooga’s City Council in 2007, EPB made a bond offering of $220 million to fund the build
out of fiber optic infrastructure that would support a Smart Grid and provide TV, internet and
phone service to residents and businesses in their footprint. In November 2009, in the wake of the
recession of 2008-2009, EPB received a federal stimulus matching grant in the amount of $111.6
million from the Department of Energy to expedite the build-out and implementation of the Smart
Grid.4 The first broadband customers were connected in the fall of 2009 and the build out was
completed roughly 6 years ahead of schedule. In September 2010, EPB made available residential
symmetrical internet connection speeds of up to one gigabit per second - the fastest Internet not
merely in the country, but in the entire Western hemisphere (Micheli, 2013). Competitive pressures
among ISPs resulted in gigabit internet service costs to customers dropping from about $300 per
month in 2010 to $70 per month by 2013 (EPB, 2019). Dubbed the “gig city”, Chattanooga has
received numerous accolades from global sources and has become a model for publicly-owned
fiber deployments in the country.
Neighboring counties would like to have the same service as that provided in EPB’s footprint.
However, Tennessee House Bill 1045, which proposed to allow counties and municipalities to
make use of their infrastructure to provide high-speed Internet access to surrounding cities where
only low performance services are available, failed to pass a state senate committee. Tennessee
remains one of at least 20 states with anti-municipal broadband state laws (BroadbandNow, 2019).
III. Literature Review
With the advent of broadband services in the 1990s, researchers started to examine the economic
impact of such internet connectivity. Much of the early work done in this area focused on the
impact of simple internet availability (Katz et al., 2010; Czernich et al., 2011). Important studies
such as Crandall et al. (2003), Lehr et al. (2005), and Ford and Koutsky (2006) concluded that
4 This grant was matched $111.5 million in cash by EPB and the City of Chattanooga, and $3.57 million by EPB’s
private partners, Alcatel-Lucent, Tantalus, and Medium.
7
communities with broadband experienced faster job and firm growth than non-broadband
communities. Kolko (2010, 2012) added that the positive relationship between broadband
expansion and economic growth is even stronger in industries with a greater reliance on
information technology and in areas with low population densities. Atasoy (2013) concluded that
gaining access to broadband services in a county is associated with approximately a 1.8 percentage
point increase in the employment rate, with larger effects in rural and isolated areas. She found
that broadband technology is complementary to skilled workers, with larger effects among college-
educated workers and in industries and occupations that employ more college-educated workers.
The relatively larger impacts on rural areas is also reflected in Whitacre et al. (2014b) who
concluded that high levels of broadband adoption in rural areas positively (and potentially,
causally) impacted income and employment growth. Jayakar and Park (2013) show a significantly
negative relationship between broadband deployment and the unemployment rate at the county
level in their cross-sectional specification for the year 2011. However, they fail to uncover a
meaningful relationship when they extend their analysis to changes over time, and suggest that
more work is needed when additional broadband deployment data becomes available.
Regarding the effects of broadband speed, Kongaut and Bohlin (2017) rightly point out that
academic research has been sparse despite the fact that “… broadband speed has started to become
more recognized by the authorities and included in the targets of national broadband plans in
several countries.” (p. 15). The literature in this area has taken the form of studying the effects of
broadband speed primarily on output (GDP), firm presence and employment. Kongaut and Bohlin
(2017) study the impact of average national download speed (sourced from Ookla) on GDP per
capita in a sample of 33 OECD countries. They find robust positive effects of broadband speed on
GDP, a result that is stronger for lower-income countries relative to higher-income countries.
Rohman and Bohlin (2013) measured the impact of broadband speed on economic growth in 34
OECD countries. They studied a quarterly balanced panel dataset during the period 2008-2010
using a two-stage fixed effects panel model. They found that doubling the broadband speed (from
8.3 to 16.6 Mbps) contributed 0.3% to GDP growth compared with the growth rate in 2008 (the
base year). Conversely, Gruber et al. (2014) found in a study of 27 EU countries from 2005-2011
that there is a growth impact from moving away from basic broadband (≤ 0.75 Mbps), but then the
incremental speed impact appears to level off, i.e. going from 1 Mbps to 2 Mbps. Most of these
studies use speed thresholds that are considerably lower than those currently available in the U.S.
– in fact, the FCC finds that 87% of all U.S. Census Blocks have access to 25 Mbps, and 56% have
access to 100 Mbps (FCC, 2018b).
Broadband speed effects on firm presence were studied by Mack (2014). She modeled broadband
speed as a binary dummy variable that indicates whether a census tract had at least one high-speed
provider. A high-speed provider was defined as one that provided broadband at speeds of at least
3 Mbps downstream and 768 Kbps upstream. She found that broadband speed is particularly
important for firm presence in rural locations suggesting that broadband speed substitutes for the
agglomerative benefits of urban locations and enables firms to carry out operations in rural areas.
She found a lack of significance of the broadband speed variable across several sectors predicted
to show an effect (such as health care or public administration), possibly because the dummy
variable for speed failed to capture specific nuances of broadband quality such as speed tiers.
Importantly, the insignificant results for speed could be because the threshold speed limits may
not have been high enough (only 3 Mbps) to be a differentiating factor.
8
Results for broadband speed effects on employment are presented in Hasbi (2017) who shows
evidence of a positive relationship between municipalities in France with access to 30 Mbps or
higher speed networks and the growth of new companies and entrepreneurship. She also found
that such “very high-speed” networks helped reduce the unemployment rate by between 7% to 9%.
By contrast, Ford (2018) found no evidence that U.S. counties with 25 Mbps broadband speed as
of 2013 outperformed those counties with only 10 Mbps speed in terms of average growth in jobs,
personal income and labor earnings over the next two years (i.e. from 2013 to 2015). The study
controls for selection bias (or covariate imbalance) using a Coarsened Exact Matching (CEM)
technique matched on population, population density, the percentage of adults with a college
education, and household size. Notably, he considers effects of incremental speed, but confines
the analysis to a single increment of 15 Mbps (i.e. moving from 10 to 25 Mbps). Bai (2017) used
a pooled first-differenced regression to report a positive relationship between access to broadband
speed and the county-level employment rate for eight states between 2011 and 2014. However,
Whitacre et al. (2018) found an econometric error with the study, which nullified this finding.
They suggested more work was needed to refine existing models and units of analysis to uncover
the relationship between broadband speed and employment.
Other studies of broadband speed include Molnar et al. (2015) who report that single-family homes
in census block groups with the ability to upgrade to a one gigabit per second Internet connection
have a transaction price that is about 1.8% higher than similar homes in neighborhoods where a
100 Mbps connection is available. Previously, Ahlfeldt et al. (2016) estimated the impact of
broadband availability on property prices in the UK during 1995–2010 can add up to 5% to a house
price.
On the whole, the broadband literature has begun to shift from analyzing the impact of simple
availability to assessing the degree to which faster speeds matter. Early findings on this topic are
mixed, with positive results found for employment (Hasbi, 2017) and housing (Molnar et al. 2015;
Ahlfeldt et al. 2016). Other results suggest that no such employment effect exists, including Ford’s
(2018) analysis of U.S. counties and Whitacre et al.’s (2018) failed validation attempt of Bai’s
(2017) work.
IV. Methodology
How might broadband impact employment levels and unemployment rates? Holt and Jamison
(2009) argue that broadband applications can potentially substitute for labor, make the use of labor
more efficient and change the way work is done and products are produced. They point out that
while it seems reasonable that broadband adoption should improve productivity and economic
growth, the effects on job growth could depend on employment and demographic trends.
Broadband adoption could decrease frictional and structural unemployment by improving the
efficiency of labor, but it may also increase structural unemployment by causing changes in the
demand for particular labor skills, at least in the short run. They conclude that, “one of the
difficulties learned from studies of the effects of ICT is that impacts evolve, perhaps even going
through periods of negative growth, while businesses experiment with applications and reorganize
their operations.” (p. 580) These trade-offs suggest heterogeneous effects that must be empirically
sorted out.
9
Our empirical strategy involves starting with a simple stratification of counties by high versus low
broadband speed; then decomposing high speed broadband into high and ultra-high speed
categories. In particular, we estimate panel models with fixed effects of the following form:
𝑦 𝛽
𝐵𝐵
,𝛽𝑋
𝛾𝛿
𝛿
𝜀
(1)
𝑦 𝛽
𝐵𝐵
,, 𝛽𝐵𝐵
,, 𝛽𝑋
,
𝛾𝛿
𝛿
𝜀
(2)
In (1) and (2), the dependent variable is the unemployment rate in county i at time t. Following
Hasbi (2017), the right-hand side variables are lagged one period to control for endogeneity arising
from potential reverse causality. Several approaches have been proposed to tackle endogeneity in
the broadband literature (see Table 2 in Kongaut and Bohlin, 2017). This lagged explanatory
variables approach is a common strategy to handle endogeneity in the social sciences, with many
articles in political science, economics, and sociology journals arguing that such an approach
alleviates endogeneity (Bellemare et al. 2017).5 Nonetheless, estimates of (1) and (2) using
contemporaneous terms are reported in the appendix. X′ is a vector of control variables; δt and δi
account for time and county fixed effects, respectively, and εit is an error term.
The broadband speed variables of interest are dummy variables that capture the percentage of
households in a county with access to particular broadband speeds. We examine two speed
categorizations:
Speed Categorization 1 for Model (1):
Low speed: less than 100 Mbps; High speed: 100 Mbps and higher
Speed Categorization 2 for Model (2):
Low speed: less than 100 Mbps; High1 speed: 100 Mbps and less than 1000 Mbps; High2 (or
Ultra-high) speed: 1000 Mbps and above
In (1), the dummy variable follows speed categorization 1 and is binary so that it takes a value of
one if the majority of the population in county i had access to high speed broadband in year t; the
omitted category is no/low speed. In (1), the coefficient β1 captures the effect of high speed
broadband relative to low speed broadband.
In (2), we explore additional speed effects by separating speeds of 100 to 1000 Mbps (which we
now call High1) and speeds of 1000 Mbps and higher (now called High2). We define two binary
dummy variables, BBH1 and BBH2 in line with speed categorization 2. A county is classified as
Low, High1, or High2 based on the speed that the highest percentage of their residents have access
to. In this specification, the effect of moving from Low to High1 speed is captured by the
coefficient β1, and the effect of moving from Low to High2 (or ultra-high speed) is captured by
the coefficient β2. The incremental effect of going from High1 to High2 broadband speed is
captured by the difference (β2 – β1).
5Other studies (e.g. Czernich et al, 2011; Atasoy, 2013; Kandilov and Renkow, 2010; Lee et al,
2015) also use a lag structure in examining similar research questions.
10
Several variables are used to control for factors, other than broadband, that independently affect
unemployment rates and mitigate the selection bias discussed in Ford (2018). These include county
education levels, household income, population diversity, population density and working age
population as a fraction of the total population, all sourced from the U.S. Census Bureau. In
particular, we expect that unemployment rates should be lower in counties with more educated
people, and a larger working age population. Conversely, unemployment is likely to be positively
related to population diversity (i.e. percent of the population that is non-white). The effects of
population density on unemployment are less clear: for instance, Oded and Murphy (2003) find a
negative effect, but Kolko (2012) finds a positive relationship.
We also examine the effects of broadband speed on rural and urban counties. Using the 2013 Rural-
Urban Continuum Codes (RUCC) provided by the U.S. Department of Agriculture, counties
classified as “metro” (RUCC codes 1-3) were deemed to be urban, and all others were classified
as rural (RUCC codes 4-9). By this classification, there are 42 urban and 53 rural counties in the
state of Tennessee.
To estimate differential rural/urban effects, we estimate the following models:
𝑦 𝛽
𝐵𝐵
,𝛽𝐷 ∗ 𝐵𝐵,𝛽 𝑋
𝛾𝛿
𝛿
𝜀
(3)
𝑦 𝛽
𝐵𝐵
,𝛽𝐵𝐵,𝛽𝐷∗𝐵𝐵
,𝛽 𝐷 ∗ 𝐵𝐵,𝛽 𝑋
𝛾𝛿
𝛿𝜀
(4)
In both (3) and (4), D is a binary dummy variable that takes a value of 1 for rural counties, 0
otherwise. In (3), which stems from (1), 𝛽 captures the effect of high speed broadband in urban
counties, (β1+βRH) is the effect of high speed broadband in rural counties, and βRH captures the
additional effects of high speed broadband in rural counties compared to urban counties.
Analogously, in (4), which stems from (2), β1 and β2 measure the effects of high speed and ultra-
high speed broadband in urban counties; (β1+βRH) is the effect of high speed access and (β2+βRU)
is the effect of ultra-high speed access in rural counties; βRH and βRU capture the additional effects
of high speed and ultra-high speed broadband in rural counties compared to urban counties.
Data
We create a panel dataset of the 95 counties in the state of Tennessee over the period 2011 to 2016.
Data on the dependent variable (county unemployment rates) for this model are from the Local
Area Unemployment Statistics of the Bureau of Labor Statistics (BLS).
The independent variables of interest in our study are the percent of the population with access to
particular broadband speeds (by any wireline or fixed wireless technology, excluding satellite).
We use broadband availability rather than adoption data in this study for two reasons. While
availability data has limitations as pointed out by Ford (2011) and others, Kolko (2012) points out
that adoption rates can be influenced by economic growth more so than availability, thereby
exacerbating the endogeneity problem. Importantly, from the perspective of our study, adequate
11
adoption data does not exist. Currently available adoption data only includes information on speed
thresholds for broadband which are very slow in the current environment. 6
To populate the speed categories used in this analysis, data from various versions of the NBM
were used. From 2010-2014, these data were collected by state entities as part of a National
Telecommunications and Information Administration (NTIA) effort funded by the American
Recovery and Reinvestment Act (ARRA). The mapping entity in each state gathered data from
Internet Service Providers (ISPs), including the type of technology used, maximum upload and
download speeds offered, and a list of all Census blocks served. During this period, the NTIA
summarized this data into an “Analyze Table” listing the percentage of residents in a county with
access to various speed thresholds and number of providers. However, this program was
discontinued after 2014 when ARRA funding expired. In 2015, the FCC took over the task of
gathering broadband data for the country. The FCC does not provide a summary table similar to
the NTIA’s “Analyze Table”; however, the same data can be constructed using the underlying
provider-based information.
Merging NBM and FCC data
To extend the broadband access dataset beyond 2014, annual provider records for Tennessee were
compiled, with over 420,000 entries (i.e. one for each census block served by each provider).
These were aggregated to a single observation for each census block, after calculating the
maximum download speed available. Next, these data were then meshed with all census blocks in
the state (over 240,000) and the population of those blocks using 2010 data from the U.S. Census.
We then use the maximum speed available to residents of each census block to calculate county-
level percentages similar to those presented in the NBM Analyze Table. To verify that our data
collection procedure worked appropriately, we successfully replicated the 2014 NBM Analyze
Table entries using the underlying census-block level provider data at that time. In Table 1, we
provide summary data for two states, Alabama and Tennessee. Our replication of the 2014
Analyze Table produces minor mean errors. Thus, we believe that our calculations for 2015 are an
appropriate extension of the NBM data. We note, however, that there are some differences
between the two datasets – namely, the NBM data was based on voluntary participation by
providers (and was gathered by a variety of organizations), while the FCC requires providers to
file the underlying Form 477. In particular, the NBM data was gathered by the organization
Connected Tennessee, while the FCC itself gathers the more recent data for all states.7
Additionally, the FCC data, unlike the NBM, differentiates between availability for consumers and
businesses. In compiling the FCC data, we elected to only include those entries that were
6The FCC began collecting adoption data on 10Mbps or faster connections in 2015 via their Form 477 efforts. Prior
to that, only slower speeds were considered.
7 Data from the entity that collected broadband data for TN as part of the NBM includes the following blurb: “The
first submission of mapping data under the State Broadband Data and Development grant program represented 77%
provider participation in Tennessee. In each subsequent submission, staff was able to increase provider participation
in the voluntary program. The final data update, submitted in October 2014, included datasets for 98.82% of the
provider community” (Connected Tennessee, 2015).
12
designated as being available to consumers. This means that if a county was categorized as high
speed in 2015, this was due solely to broadband availability for consumers.
We use county access to particular speed levels based on advertised download speeds reported by
ISPs. We categorize a county as either low, high or ultra-high speed based on the highest
percentage of the population with access to the speed categories defined earlier. Note that
implausible access data led us to drop 10 counties (Benton, Fayette, Hancock, Lewis, Macon,
Meigs, Morgan, Polk, Scott and Shelby) from the analysis.8 This resulted in an adjusted sample of
85 counties: 37 urban and 48 rural.
V. Empirical Findings
Summary Statistics
Table 2 shows the percent of county population with access to broadband by speed tier. Panel A
shows that the access has varied over the years with a sharp rise in access to 100+ Mbps broadband
in 2012, ostensibly because of Federal stimulus funding in the wake of the credit crisis of 2008-
2009. Additionally, flexible technologies may have facilitated speedy upgrades (especially in the
case of municipal deployments) in the wake of Google’s announced Gigabit plans around that
time. Some nearby private Incumbent Local Exchange Carriers (or ILEC’s) may also then have
had pressure to upgrade.9In the overall sample (2011-2015), 48% of the population had access to
100 Mbps or higher speed on average. However, some 6% of the population still had no broadband
or very low speed (< 3 Mbps) broadband (not shown). Even fewer people (5%) had access to
gigabit speed and faster broadband.
In Panel B, we show the distribution of counties by speed tiers based on the population access
distribution in Panel A. Note that counties are placed in a speed tier based on the highest percentage
of the population that had access to a particular speed.10 Notably, only six counties had gigabit
speed broadband in 2015. As previously noted, in 2011, Hamilton County became the first county
in the state (and indeed, in the Western hemisphere) to receive ultra-high speed broadband.
Ford (2018) cautions that counties with higher-speed broadband are unlike those with lower-speed
broadband, i.e. selection bias could be serious. In Table 3, we present summary statistics on key
socio-economic variables for high and low speed counties. We find that low speed counties are
characterized by higher unemployment rates relative to high speed counties. Also, low speed
counties have smaller populations and population density, lower household income and a slightly
smaller proportion of people with at least a high school diploma. The RUCC score for high speed
counties is lower suggesting a greater likelihood of being urban/metro counties relative to low
8 In each case, the percentage of population with access to low speed/no broadband was reported as significantly
higher than in the previous year, which is likely implausible, suggesting a possible problem.
9 We are grateful to Mike Render of RVA LLC for this insight.
10 As an example of how a county was classified, consider Wilson county in 2014: 6% of the population had access
to less than 100 Mbps, 86% had access to 100 – 1000 Mbps and 8% had access to 1000+ Mbps speed. The county
was placed in the “high” speed tier.
13
speed counties. These statistics indicate that econometric methods must control for potential
selection bias. Our regression models use the controls suggested by Ford.
Early Adoption Effects
Does earlier adoption of high speed broadband make a difference? To answer this question, we
examine the average unemployment rates in 2016 for those counties that were classified as high
speed in 2011-2015 relative to those that were low speed. Mack and Wentz (2017) note that places
served by earlier roll-out of high speed broadband are more likely to receive upgrades, which in
turn facilitate the efficient transfer of large volumes of information, the transfer of analytical tools
from desktops to online web services, and the intensified use of cloud computing for data storage
and retrieval. How does this translate into county employment?
In Table 4, we see that in 2011, only eight counties had high speed (100 Mbps+) access. By 2012,
as many as 49 counties had access to high speed. The number of new counties accessing high
speed broadband dropped sharply to 7 and 1 in 2013 and 2014. By 2015, the total number of
counties with high speed access had risen to 61.
Importantly, the data suggest that high speed counties were characterized by roughly one percent
lower unemployment rates in 2016 than low speed counties on average (see URL - UR
H).
Furthermore, the unemployment differential is greater for counties that adopted high speed earlier.
For instance, counties that adopted high speed in 2011 had the largest differential in unemployment
rates in 2016 (1.57%) relative to counties with low speed broadband.
The last three columns show the effect of high speed broadband on only counties that newly moved
to high speed in a particular year. Here, the differential relative to low speed counties is smaller
on average, reflecting the missing effects of early adoption of high speed. For instance, in 2012,
counties that newly adopted high speed broadband had 0.66% lower unemployment in 2016
compared to low speed counties (URL - URH*). However, when we add the effects of the eight
counties that had already adopted high speed in 2011, the differential widens to 0.86%. The
average benefit to early adoption (i.e. the difference between URL - URH and URL - URH*) from
2011 to 2015 appears to be about 0.16 percentage points (if we exclude the single 2014 data point).
This positive association between broadband speed and labor market outcomes does not indicate
causality. In particular, it could simply mean that faster broadband was deployed in the “best”
locations first – those that already had low unemployment rates. To explore this issue further, we
next present our regression estimates in which we explicitly control for various socio-economic
factors that help mitigate potential selection bias and reverse causality.
Regression Results
Estimates of (1) and (2) are in Table 5.11 Note that the Hausman test rejected a random effects
model in favor of a fixed effects model.Our models are estimated with standard errors that are
11 On the suggestion of a referee, the model is estimated without the household median income variable. Adding the
variable to the model does not alter our results.
14
clustered at the county level to address the potential issues related to heteroscedasticity and
autocorrelation in the error terms.Estimates of (1) show that broadband speed matters: counties
with fast broadband (i.e. 100 Mbps or higher) have unemployment rates roughly 0.26 percentage
points lower than counties with low speed broadband. In model (2), we separate ultra-high speed
broadband to examine the incremental effects of broadband speed. We find that compared to low
speed broadband, ultra-high speed broadband appears to lower unemployment rates - however, the
estimate for ultra-high speed broadband is not statistically significant, perhaps due to the relatively
small number of observations in the ultra-high speed category.12
Our finding runs contrary to those who have not found broadband access effects on unemployment
rates (e.g. Jayakar and Park (2013), Czernich (2014), Whitacre et al. (2014a, 2014b)). This is
possibly because access alone is not a sufficient determinant of economic outcomes; broadband
quality matters as well. We believe that more data and updated speed levels help clarify the
relationship between speed and economic outcomes. Later, in Table 8, we show the sensitivity of
our estimates to the sample period.
The education and population density controls are statistically significant. The estimates suggest
that a one percentage point increase in the educated population results in a decline in
unemployment rates by 0.06 percentage points. Likewise, a 10 percent increase in population
density results in an increase of 1.018 percentage points in the unemployment rate.
Relative to the unemployment rate differential reported in Table 4, we find that even after
controlling for potential selection bias and other fixed effects, the benefits of high speed broadband
are significant, albeit smaller. Moreover, our Table 5 estimates do not uniquely capture the early
adoption effects (approximately on the order of 0.16 percentage points on average) which could
further enhance the effects of broadband speed on employment as we demonstrate shortly.
Robustness test #1: Pseudo-treatment effects
A concern with estimates using data where low and high speed counties exhibit sharply different
features is that the effects of other (possibly missing) variables might be confused for effects due
to the treatment (i.e. broadband access). Following Ford (2018), we report a simple test of the
“unconfoundedness” or “conditional independence” assumption. We estimate the causal effect of
the treatment on an outcome that is determined prior to the availability of the treatment effect,
based on the logic that the future cannot determine the past. In the context of our study, we regress
lagged values of the unemployment rate on current values of broadband access (i.e. the pseudo-
treatment) during our sample period. Our results, available on request, show no effects of the
pseudo-treatment on historical unemployment rates, supporting the conditional independence
assumption.13
Robustness test #2: The broadband speed dummy variable
12 Regressions run on all 95 counties produce qualitatively similar results to those reported in this section.
13 Additionally, we ran this “pseudo-treatment” test with 21 regressions using five-year rolling blocks of
unemployment data from 1990 to 2010. We found no cases where the broadband speed coefficient was significant at
the one percent level, and two cases where the coefficient was significant at the five percent level.
15
The dummy variable for county broadband speed is based on the highest percentage of the
population that had access to a particular speed. Following Ford (2018), we consider an alternative
specification of the dummy variable in re-testing model (1). Here, a county is classified as high-
speed if at least 80% of the population had access to 100 Mbps or higher broadband; a county is
classified as low-speed if at least 80% of the population had access to less than 100 Mbps
broadband, and less than 20% of the population had access to 100 Mbps or higher broadband. Our
results for model (1) are reported in the Appendix as model (1′′). Our sample shrinks from 510 to
190 with this experiment; however, this re-definition of the dummy variable has no material impact
on the broadband speed coefficient reported in Table 5.
A Rough Approximation of Jobs Created/Saved
The broadband effects reported in Table 5 can be roughly translated into the annual number of jobs
saved/created in each county. To do so, we first characterize counties by their average broadband
speed over the sample period by sorting counties by their speed tier score each year (i.e. low=0,
high=1, ultra-high=2). We then average a county’s score over the period 2011-2015. Counties with
average scores of 1.5 or higher were classified as ultra-high speed, those with scores between 0.5
and 1.5 were classified as high speed, and counties with scores less than 0.5 were deemed low
speed counties. Accordingly, only one county (Hamilton) was classified as ultra-high speed, 53
were classified as high speed, and 31 were classified as low speed, on average over the sample
period.
We would then multiply the appropriate coefficient (β1 or β2) from Model 2 in Table 5 with the
working age population of that county in 2015 to calculate the annual number of jobs saved/created
in that county. However, since β2 was not statistically significant, we use only the β1 estimate. As
an example, consider Hamilton County, TN, which had an average working age population of
187,992 in 2015. By applying a 0.26 percentage point reduction in the unemployment rate to the
working age population, we get an estimate of 489 jobs saved/created each year, or roughly 2,444
jobs over a five-year period. A further adjustment of 0.16 percentage points for early adoption
would raise the five-year estimate of jobs created/saved for Hamilton county to 3,948. These
computations are directly proportional to the size of the working age population, and therefore
largest among the biggest counties (Davidson, Knox, Hamilton and Rutherford).
In Table 6, we see that among all the high speed or ultra-high speed counties, the median
incremental jobs created/saved is about 59 per year relative to low speed counties. We also show
computations for the top 10 and bottom 10 counties arranged by size of the working age
population. We caution that these estimates are based on the assumption that the effects of high
speed broadband are evenly distributed among counties belonging to a particular speed category.
Clearly, this need not be the case. Counties differ in important ways such as demographics, skill-
bias, industry structure, and along the rural/urban continuum. Nonetheless, the findings could be
instructive for policy purposes especially if studying the effects of a technology on jobs is
paramount.
Rural Effects
16
In Table 7, we present estimates of (3) and (4) to study the differential effects of broadband speed
on rural and urban counties. We find that the broadband interaction terms for rural counties are
highly significant with larger coefficients than those observed for the full sample. From the model
(4) estimates, it appears that rural counties with high speed broadband (100 Mbps or higher) have
roughly 0.38 percentage points lower unemployment rates than high speed urban counties; rural
counties with ultra-high speed broadband might benefit even more. In such cases, however, the
ultra-high speed variable is estimated with substantial error possibly because the number of rural
counties with ultra-high speed broadband is small (i.e. only 3 by 2015), and we caution this is a
finding that requires further investigation in a larger dataset. The average effects for urban counties
as gauged by the coefficients β1 and β2, while indicative, are not statistically significant, suggesting
likely offsetting effects of technology on job gains and losses, on average.
This (lack of an) urban effect requires further research. Our results are in contrast to those in
Whitacre et al. (2014a), who found a positive relationship between broadband availability and
employment levels in urban areas (but not rural ones) in 2011. It may be the case that, by the early
2010s, most urban areas had reasonably fast broadband access. As a consequence, rural areas with
faster-speed providers had a competitive advantage in finding employment opportunities for their
residents, perhaps in jobs emphasizing telework or real-time interaction with urban firms. A recent
review of the literature emphasizes the variety of categories in which broadband could impact rural
locations – including entrepreneurship, telehealth, and “big data” opportunities for agriculture
(Gallardo et al., 2018).
Model Sensitivity Analysis
In Table 8, we show the sensitivity of our estimates to the sample period and to the definition of
rural counties. We examine estimates for the 2012-2015 sample period and compare them to our
estimates in this paper for the sample period 2012-2016. To be clear, the lag structure used in this
paper means that we effectively use speed access data for the period 2011-2015. The comparison
sample in Table 8 uses speed access data for the period 2011-2014. Note that all previous studies
(to our knowledge) of U.S. broadband speed effects use some version of the NBM Analyze data,
restricting their analysis to speed access data in the window from 2011-2014 (e.g. Lapoint, 2015;
Mack, 2014; Ford, 2018). This paper adds more speed access data to the analysis by merging the
NBM and FCC datasets as previously explained.
The effects of adding more data to the analysis are clear: the first two columns show that more
data sharpens the estimates, i.e. the coefficients are estimated with smaller standard errors. Also,
we find that the size of the coefficients typically increases. In the next 4 columns, we show the
effects of marginally altering the classification of counties as rural or urban. RUCC A is the
definition used in the current paper, i.e. urban counties are RUCC 1-3 and rural counties are RUCC
4-9. RUCC B classifies urban counties as RUCC 1-4 and rural counties as RUCC 5-9. This
reclassification moves counties that are adjacent to a metro area and have urban populations over
20,000 (i.e. RUCC code 4) from rural into urban. In the columns titled RUCC A and RUCC B, we
show that the results are impervious to marginal changes in the rural/urban continuum; the rural
effect remains strong and is more precisely estimated with the longer data set.
17
VI. Conclusion
We attempt to decipher the incremental effects of broadband speed on county unemployment rates
in the U.S. state of Tennessee. We use panel data that captures the percent of the population served
by different broadband speeds over the period 2011-2015. Our panel regression results show that
high broadband speed matters and results in approximately 0.26 percentage points lower
unemployment in counties with high speed compared to counties with low speed broadband; ultra-
high speed broadband appears to also reduce unemployment rates but our limited sample of such
counties prevents us from generating efficient estimates. Additionally, early adoption of high
speed broadband could reduce unemployment rates by an average of 0.16 percentage points per
year. The results also show that compared to urban areas, the benefits of better quality broadband
are disproportionately greater in rural areas.
From a policy standpoint, our research shows that investments in faster broadband can have
significant employment effects, especially in rural areas. Our findings are at odds with claims that
“…the definition of “broadband” is typically immaterial to economic outcomes…” (Ford, 2018,
p.2). While it may make little difference to move from 10 Mbps to 25 Mbps, it could (and our
results suggest that it does) make a significant difference to move to 100 Mbps or higher speeds,
a view reflected in Mack (2014). Our results consistently show that access to faster speed results
in a decrease of 0.2 – 0.3 percentage points in unemployment, which can be in the 100s of jobs for
some counties. We believe many local policymakers would consider these effects to be
economically meaningful. Our research shows that it can be valuable to examine larger datasets
across longer periods of time using current speed offerings. More work is required, however, in
understanding speed effects on employment in urban areas. To this end, we reiterate a
recommendation in Mack and Wentz (2017) that in studying broadband effects, it is important to
consider how and why broadband is used. Thus, a key need in the study of broadband effects on
business, for instance, is collecting usage data so that researchers and policymakers are better able
to assess multiple aspects of access and use of broadband internet connections simultaneously.
This research should be considered in the broader context of ongoing enquiries into the effects of
broadband quality on economic outcomes. Governments worldwide have adopted measures to
support the rapid diffusion of broadband and to reduce digital divides. Perhaps the most
controversial issue regarding telecom policy pertains to the role of local governments in providing
broadband to smaller communities. To that end, Chattanooga’s successful fiber deployment is
perhaps a useful model for further study.
We find that high speed broadband has a significant effect on county-level unemployment rates;
however, we were unable to distinguish between the effects of high and ultra-high speed tiers. Our
empirical specification is linear in parameters; other specifications should be explored to
investigate the potentially nonlinear relationship between broadband speed and labor market
outcomes. The exact mechanism by which broadband infrastructure adds value to communities
remains an open question. Anecdotal evidence and case studies (e.g. Lobo, 2015), for instance,
show diverse effects across various facets of community life. To the extent that the impacts of such
infrastructure might take years to be fully realized, the search for stable, long-term, and possibly
nonlinear effects must continue.
18
References
Ahlfeldt, G. M., Koutroumpis, P. and Valletti, T. M. (2016). Speed 2.0 - Evaluating Access to
Universal Digital Highways. CEPR Discussion Paper No. DP11046.
Available at https://ssrn.com/abstract=2717595
Atasoy, H. (2013). The Effects of Broadband Internet Expansion on Labor Market Outcomes.
Industrial and Labor Relations Review, 66 (2), 315-345.
Atkinson, R., Castro, D., Ezell, S., & Ou, G. (2009). The need for speed: The importance of next
generation broadband networks. The Information Technology and Innovation Foundation
(ITIF), Washington, D.C.
Bai, Y. (2017). The Faster, the Better? The Impact of Internet Speed on Employment.
Information Economics and Policy 40: 21-25.
Bellemare, M., Masaki, T., and Pepinsky, T. (2017). Lagged Explanatory Variables and the
Estimation of Causal Effect. The Journal of Politics, 79(3): 949-963.
BroadbandNow. (2019). Broadband Service in Tennessee. Available at:
(https://broadbandnow.com/Tennessee)
BroadbandNow. (2018). FCC Broadband Definition Has Changed Before and Will Change
Again. Available at https://broadbandnow.com/report/fcc-broadband-definition/
BroadbandNow. (2017). Broadband Bill Autopsy: The Origin and Purpose of Every Single Fee.
Available at: https://broadbandnow.com/report/broadband-bill-autopsy-origin-purpose-every-
single-fee/
Connected Tennessee. (2015). Final Grant Report. Available at:
http://connectedtn.org/sites/default/files/connected-
nation/tennessee_final_grant_report_2015_04_20_complete.pdf
Coren, M. (2016). Broadband service tends to stop at the poverty line in the US. Quartz.
Available at https://qz.com/684388/broadband-service-tends-to-stop-at-the-poverty-line-in-the-
us/
Crandall, R., Jackson, C. & Singer, H. (2003). The Effects of Ubiquitous Broadband Adoption
on Investment, Jobs, and the US Economy. Criterion Economics, L.L.C. Released by New
Millennium Research Council. Available at
http://newmillenniumresearch.org/archive/bbstudyreport_091703.pdf.
Czernich, N. (2014). Does broadband internet reduce the unemployment rate? Evidence for
Germany. Information Economics and Policy, 29, 32-45.
Czernich, N., Falck, O., Kretschmer, T., and Woessmann, L. (2011). Broadband infrastructure
and economic growth. The Economic Journal, 121 (May), 505-532.
19
EPB – Our History. (2019). Available at: https://epb.com/about-epb/our-history.
Federal Communications Commission (FCC). (2010). The National Broadband Plan. Available
at: https://transition.fcc.gov/national-broadband-plan/national-broadband-plan.pdf
Federal Communications Commission (FCC). (2014). Types of Broadband Connections.
Available at https://www.fcc.gov/general/types-broadband-connections
Federal Communications Commission (FCC). (2018a). 2018 Broadband Deployment Report.
Available at https://www.fcc.gov/reports-research/reports/broadband-progress-reports/2018-
broadband-deployment-report
Federal Communications Commission (FCC). (2018b). Internet Access Services: Status as of
Decmember 31, 2016. Industry Analysis and Technology Division. Available at
https://transition.fcc.gov/Daily_Releases/Daily_Business/2018/db0207/DOC-349074A1.pdf
Ford, G.S. (2018). Is faster better? Quantifying the relationship between broadband speed and
economic growth. Telecommunications Policy, https://doi.org/10.1016/j.telpol.2018.05.006
Ford, G.S. (2011). Challenges in using the national broadband Map's data. Phoenix Center
Policy Bulletin, 27, pp. 1-21
Ford, G.S. & Koutsky, T.M. (2006). Broadband and Economic Development: A Municipal Case
Study from Florida. Applied Economic Studies, April 2005.
Gallardo, R., Whitacre, B., & Grant, A. (2018). Broadband’s Impact: A Brief Literature
Review. Purdue University Center for Regional Development Research & Policy Insights series.
Available online: https://www.pcrd.purdue.edu/files/media/Broadbands-Impact-Final.pdf
Genachoswki, J. (2013). Faster, Sooner: Why the U.S. Needs ‘Gigabit Communities’. Forbes
Magazine. Available at https://www.forbes.com/sites/ciocentral/2013/01/18/faster-sooner-why-
the-u-s-needs-gigabit-communities
Gruber, H., Hätönen, J. & Koutroumpis, P. (2014). Broadband access in the EU: An assessment
of future economic benefits. Telecommunications Policy, 38, 1046-1058.
Hasbi, M., (2017). Impact of Very High-Speed Broadband on Local Economic Growth:
Empirical Evidence. 14th International Telecommunications Society (ITS) Asia-Pacific Regional
Conference: “Mapping ICT into Transformation for the Next Information Society,” Kyoto,
Japan, 24-27 June 2017, International Telecommunications Society (ITS), Kyoto. Available at
https://www.econstor.eu/bitstream/10419/168484/1/Hasbi.pdf
Holt, L. & Jamison, M. (2009). Broadband and contributions to economic growth: Lessons from
the US experience. Telecommunications Policy, 33, 575-581.
20
International Telecommunication Union. (2012). The Impact of Broadband on the Economy.
Available at http://www.itu.int/ITU-D/treg/broadband/ITU-BB-Reports_Impact-of-Broadband-
on-the-Economy.pdf
Jayakar, K. & Park, E. (2013). Broadband Availability and Employment: An Analysis of
County-Level Data from the National Broadband Map. Journal of Information Policy 3: 181-
200.
Kandilov, I., and Renkow, M. 2010. Infrastructure Investment and Rural Economic
Development: An Evaluation of the USDA’s Broadband Loan Program. Growth and Change,
41(2): 165-191.
Katz, R., Vaterlaus, S., Zenhausern, P., & Suter, S. (2010). The impact of broadband on jobs and
the German economy. Intereconomics, 45(1), 26-34.
Kolko, J. (2010). Does Broadband Boost Local Economic Development? Public Policy Institute of
California. Available at http://www.ppic.org/content/pubs/report/R_110JKR.pdf
Kolko, J. (2012). Broadband and local growth. Journal of Urban Economics, 71, 100-113.
Kongaut, C. & Bohlin, E. (2017). Impact of broadband speed on economic outputs: An empirical
study of OECD countries. Economics and Business Review 3 (17), No. 2, 12-32.
Koutroumpis, P. (2009). The economic impact of broadband on growth: A simultaneous approach.
Telecommunications Policy, 33(9), 471-485.
Kruger, L. (2018). Broadband Loan and Grant Programs in the USDA’s Rural Utilities Service.
CRS Report RL33816. Available at https://fas.org/sgp/crs/misc/RL33816.pdf
Lapoint, P. (2015). Does speed matter? The employment impact of increasing access to fiber internet.
Mimeo. Washington Academy of Sciences, Spring, 9-28.
Lee, S., Kim, D., and Son, H. 2015. The Impact of Mobile Broadband Infrastructure on
Technological Innovation: An Empirical Analysis. International Telecommunications Policy
Review, 22(2): 93-108.
Lehr, W. H., Osorio, C. A., Gillett, S. E., & Sirbu, M. A. (2005). Measuring broadband’s
economic impact. Proceeding of the 33rd Research Conference on Communication, Information,
and Internet Policy (TPRC), September 23-25, Arlington, VA.
Lobo, B.J. (2015). The realized value of fiber optic and smart grid infrastructure in Hamilton
County, Tennessee, 2011-2015. Mimeo. University of Tennessee at Chattanooga. Available at
http://media-
cdn.timesfreepress.com/news/documents/2015/09/15/realizedvalueoffiberlobofinaljune18201515
650443634.pdf
21
Mack, E. and Wentz, E. (2017). Industry variations in the broadband business nexus. Socio-
Economic Planning Sciences 58, 51-62.
Mack, E. (2014). Businesses and the need for speed: The impact of broadband speed on business
presence. Telematics and Informatics, 31, 617-627.
Micheli, R. (2013). Rebooting Chattanooga’s Fortunes. CNBC.com. Available at:
https://www.cnbc.com/2013/11/18/hattanoogas-fortunes.html
Middleton, C. (2013). Beyond broadband access: What do we need to measure and how do we
measure it? In R. D. Taylor, & A. M. Schejter (Eds.), Beyond broadband access: Developing
data based information policy strategies (pp. 9-22). New York: Fordham University Press.
Molnar, G., Savage, S.J. and Sicker, D.C. (2015). Reevaluating the Broadband Bonus: Evidence
from Neighborhood Access to Fiber and United States Housing Prices. Available at:
https://www.lightwaveonline.com/content/dam/lw/documents/FTTH_Report_06_26_2015.pdf
Oded, I. and Murphy, K.J. (2003). The effect of industrial diversity on state unemployment rate
and per capita income. The Annals of Regional Science, 37(1), 1-14.
Rohman, I. K. and Bohlin, E. (2012). Does broadband speed really matter as a driver of
economic growth? Investigating OECD countries. International Journal of Management and
Network Economics, 2(4), 336-356.
Shideler, D. and Badasyan, N. (2012). Broadband impact on small business growth in Kentucky.
Journal of Small Business and Enterprise Development, 19(4), 589-606.
Stocker, V., and Whalley, J. (2016). Speed isn’t everything: a multi-criteria analysis of
broadband access speeds in the UK. Proceeding of the 27th European Regional Conference of
the International Telecommunication Society (ITS), September 7-9, Cambridge.
Qiang, C., Rossotto, C. and Kimura, K. (2009). Economic impacts of broadband, Information
and Communications for Development: Extending Reach and Increasing Impact, 3, 35–50.
Whitacre, B.E., Gallardo, R. and Strover, S. (2014a). Does rural broadband impact jobs and
income? Evidence from spatial and first-differenced regressions. Annals of Regional Science
53(3): 649-670.
Whitacre, B.E., Gallardo, R. and Strover, S. (2014b). Broadband’s Contribution to Economic
Growth in Rural Areas: Moving towards a Causal Relationship. Telecommunications Policy
38(11): 1011-1023.
Whitacre, B.E., Alam, Md. R. and Lobo, B. J. (2018). Econometric error nullifies finding of the
impact of broadband speed on county-level employment. Information Economics and Policy,
44, 58-60.
22
Table 1. Replicating NBM Data on Population Access to Broadband in 2014
Alabama Counties [N=67] Tennessee Counties [N=95]
Speed Tier Our
Calculation
NBM
Analyze
Table
Mean Error Our
Calculation
NBM
Analyze
Table
Mean Error
Low 33 34 -1% 35 35 -1%
High 34 33 +1% 55 55 0%
Ultra-high 0 0 0% 5 5 0%
Notes:
Our calculation is based on NBM Census Block-level data as explained in the text. Mean Error is based on the
difference between our calculation of the percent of the population with access to a speed tier relative to the NBM
Analyze table. A positive (negative) value indicates that our calculation overestimated (underestimated) the speed
tier access across all counties, relative to the NBM Analyze table.
23
Table 2. Summary of Broadband Access
Panel A. Population Access to Broadband Speed
< 100
Mbps
100 - 1000
Mbps
≥ 1000
Mbps
Average
2011-2015 52% 43% 5%
2011 86% 13% 1%
2012 51% 48% 1%
2013 43% 50% 7%
2014 43% 50% 8%
2015 36% 55% 9%
Panel B. Classification of Counties by Broadband Speed Categories
Broadband Speed Category for Models 1 and 3 [N=85]
No/Low Speed
< 100 Mbps
High Speed
≥ 100 Mbps
2011 77 8
2012 36 49
2013 29 56
2014 28 57
2015 24 61
Broadband Speed Category for Models 2 and 4 [N=85]
No/Low Speed
< 100 Mbps
High Speed
100 Mbps - 1000 Mbps
Ultra-High Speed
≥ 1000 Mbps
2011 77 7 1
2012 36 48 1
2013 30 52 3
2014 29 51 5
2015 24 55 6
Rural (Non-Metro) Counties [N=48]
2011 48 0 0
2012 28 20 0
2013 24 23 1
2014 23 24 1
2015 21 24 3
24
Table 3. Summary Statistics on Dependent and Control Variables
Panel A. Full Sample of 85 Counties
Obs Mean Median Max Min Std. Dev
Unemployment rate (%) 510 7.96 7.85 13.80 3.40 2.18
Population 510 62,632 32,292 667,885 5,094 94,578
Population density 510 140.31 73.94 1346.92 18.90 188.99
Working age population (18 to 64)
(%)
510 53.92 52.48 71.88 44.43 4.84
Education: High school or higher
(%)
510 80.43 80.40 95.60 65.00 5.13
Median household income ($) 510 41,102 39,414 104,367 26,101 9,390
Population Diversity (% Non-
White)
510 11.08 7.70 55.30 0.71 10.15
RUCC (1– most urban; 9 – most
rural)
510 4.47 4.00 9.00 1.00 2.65
Panel B. Sub-Samples by Broadband Speed (Mean Values)
All
Counties
Low Speed
Counties
High Speed
Counties
No. of Observations 510 216 294
Unemployment rate (%)7.96 9.18 7.07
Population62,632 29,975 86,626
Population density140.31 71.47 190.88
Working age population (18 to 64) (%) 53.92 52.96 54.62
Education: High school or higher (%)80.43 78.10 82.13
Median household income ($)41,102 37,381 43,836
Population Diversity (% Non-White)11.08 11.17 11.02
RUCC (1– most urban; 9 – most rural) 4.47 5.76 3.52
25
Table 4. Early Adoption Effects? Impact of Broadband Speed on 2016 Unemployment Rates
Speed tier
in Low Speed Counties High Speed Counties Difference Counties that newly accessed
High Speed
N UR
L
(%)
N URH
(%)
URL - URH
(%)
N* URH*
(%)
URL - URH*
(%)
2011 77 5.62 8 4.05 1.57 8 4.05 1.57
2012 36 5.97 49 5.11 0.86 41 5.31 0.66
2013 29 6.09 56 5.15 0.93 7 5.49 0.60
2014 28 6.05 57 5.19 0.87 1 7.00 -0.95
2015 24 6.14 61 5.21 0.93 4 5.50 0.64
Average 5.97 4.94 1.03 5.47 0.87§
Notes: § excludes the 2014 data point.
26
Table 5. Effects of Broadband Speed on Unemployment Rates
Models:
(1) 𝑦 𝛽
𝐵𝐵
,𝛽𝑋
𝛾𝛿
𝛿
𝜀
(2) 𝑦 𝛽
𝐵𝐵
,, 𝛽𝐵𝐵
,, 𝛽𝑋
,
𝛾𝛿
𝛿
𝜀
Model (1) Model (2)
Speed: High (β1) -0.2634**
(0.1009)
NA
Speed: High1 (β1) NA -0.2575**
(0.1006)
Speed: High2 (β2) NA -0.2668
(0.2428)
Education -0.0600*
(0.0318)
-0.0585*
(0.0323)
Working Age Population 0.0580
(0.0921)
0.0572
(0.0923)
Diversity 0.0052
(0.0559)
0.0030
(0.0570)
Ln Population Density 10.1579***
(3.3307)
10.1772***
(3.3361)
N 425 425
Degrees of freedom 331 330
County Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
R2 0.9679 0.9678
Notes:
***, **, * are significant at 1%, 5% and 10%, respectively. Standard errors (in parentheses) are clustered at the
county level to address the potential issues related to heteroscedasticity and autocorrelation in the error terms.
Speed categorization for Model (1): Low: < 100 Mbps; High: ≥ 100 Mbps
Speed categorization for Model (2) and (3): Low: < 100 Mbps; High1: 100 Mbps to 1000 Mbps; High2: ≥ 1000
Mbps
27
Table 6. The Effect of High Speed Broadband on Jobs Saved/Created
County Average
Speed Tier
2011-2015
Average
Speed
Category
Working Age
Population
(2015)
Incremental
Annual Jobs Saved / Created
Relative to Low Speed
Counties
MEDIAN 0.8 High 23,453 59
Top 10 Counties
Davidson 1.0 High 392,554 1,010
Knox 0.8 High 245,968 633
Hamilton 2.0 Ultra-high 187,992 489
Rutherford 1.0 High 163,029 420
Montgomery 1.0 High 107,281 276
Williamson 1.0 High 104,576 269
Sumner 1.0 High 90,071 232
Sullivan 1.4 High 78,948 203
Washington 0.6 High 67,641 174
Wilson 0.8 High 65,065 167
Bottom 10 Counties
White 1.0 High 13,044 34
Hardin 0.8 High 12,576 32
Grainger 0.8 High 11,732 30
Smith 0.8 High 10,155 26
Union 0.8 High 10,022 26
DeKalb 0.8 High 9,748 25
Johnson 0.8 High 9,603 25
Unicoi 0.6 High 9,223 24
Crockett 1.0 High 7,290 19
Grundy 0.8 High 6,589 17
Notes:
Top 10 and Bottom 10 counties arranged according to size of the working age population. Counties were placed
in a speed category based on their average speed tier for the period 2011-2015 as follows: Low: ≤ 0.5; High: >
0.5 and less than 1.5; Ultra-high: ≥ 1.5. MEDIAN refers to the median of all counties classified as high or ultra-
high speed counties on average.
28
Table 7. Effects of Broadband Speed on Unemployment Rates in Rural Counties
Models:
(3): 𝑦 𝛽
𝐵𝐵
,𝛽𝐷 ∗ 𝐵𝐵,𝛽 𝑋
𝛾𝛿
𝛿
𝜀
(4): 𝑦 𝛽
𝐵𝐵
,𝛽𝐵𝐵,𝛽𝐷∗𝐵𝐵
,𝛽 𝐷 ∗ 𝐵𝐵,𝛽 𝑋
𝛾𝛿
𝛿𝜀
Model (3) Model (4)
Speed: High (β1) -0.0659
(0.1153)
NA
Speed: High1 (β1) NA -0.0665
(0.1126)
Speed: High2 (β2) NA 0.0970
(0.3429)
Rural*High Speed (βRH) -0.3889**
(0.1567)
NA
Rural*High1 Speed (βRH) NA -0.3776**
(0.1561)
Rural*High2 Speed (βRU) NA -0.6470
(0.4229)
Education -0.0515
(0.0332)
-0.0490
(0.0335)
Working Age Pop 0.0425
(0.0905)
0.0405
(0.0910)
Diversity 0.0044
(0.0547)
0.0018
(0.0561)
Ln Population Density 9.9765***
(3.2053)
10.0047***
(3.2164)
N 425 425
Degrees of freedom 330 328
County Fixed Effects Yes Yes
Year Fixed Effects Yes Yes
R2 0.9691 0.9691
Notes:
***, **, * are significant at 1%, 5% and 10%, respectively. Standard errors (in parentheses) are clustered at the
county level to address the potential issues related to heteroscedasticity and autocorrelation in the error terms.
Speed categorization for Model (3): Low: < 100 Mbps; High: ≥ 100 Mbps
Speed categorization for Model (4): Low: < 100 Mbps; High1: 100 Mbps to 1000 Mbps; High2: ≥ 1000 Mbps
Urban = RUCC 1-3, Rural = RUCC 4-9
29
Table 8. Model Sensitivity to Sample Period and Rural/Urban Classification
Models:
(1) 𝑦 𝛽
𝐵𝐵
,𝛽𝑋
𝛾𝛿
𝛿
𝜀
(2) 𝑦 𝛽
𝐵𝐵
,, 𝛽𝐵𝐵
,, 𝛽𝑋
,
𝛾𝛿
𝛿
𝜀
(3) 𝑦 𝛽
𝐵𝐵
,𝛽𝐷 ∗ 𝐵𝐵,𝛽 𝑋
𝛾𝛿
𝛿
𝜀
(4) 𝑦 𝛽
𝐵𝐵
,𝛽𝐵𝐵,𝛽𝐷∗𝐵𝐵
,𝛽 𝐷 ∗ 𝐵𝐵,𝛽 𝑋
𝛾𝛿
𝛿
𝜀
Sample RUCC A RUCC B
Period Coeff. Model (1) Model (2) Model (3) Model (4) Model (3) Model (4)
2011-2015
β1 -0.2053*
(0.1255)
-0.1939
(0.1305)
-0.0387
(0.1277)
-0.0203
(0.1319)
-0.0816
(0.1194)
-0.0682
(0.1233)
β2 -0.2319
(0.3829)
0.1823
(0.3231)
0.1541
(0.3231)
βRH -0.3194*
(0.1865)
-0.2930
(0.1891)
-0.3079
(0.2213)
-0.2688
(0.2280)
βRU -1.1625***
(0.3295)
-1.1356***
(0.3296)
2011-2016
β1 -0.2634**
(0.1009)
-0.2575**
(0.1007)
-0.0659
(0.1153)
-0.0665
(0.1126)
-0.1006
(0.1094)
-0.0977
(0.1082)
β2 -0.2668
(0.2428)
0.0970
(0.3429)
0.0940
(0.2894)
βRH -0.3889**
(0.1567)
-0.3776**
(0.1561)
-0.4107**
(0.1976)
-0.3899*
(0.1984)
βRU
-0.6470
(0.4229)
-0.8756***
(0.3240)
Notes:
***, **, * are significant at 1%, 5% and 10%, respectively. Standard errors clustered at the county level are reported
in parentheses.
RUCC A: Urban = 1-3, Rural = 4-9;
RUCC B: Urban = 1-4, Rural = 5-9
30
Appendix.
Contemporaneous Effects of Broadband Speed on Unemployment Rates
(1′) 𝑦 𝛽
𝐵𝐵
,𝛽𝑋
𝛾𝛿
𝛿
𝜀
(2′) 𝑦 𝛽
𝐵𝐵
,, 𝛽𝐵𝐵
,, 𝛽𝑋
,
𝛾𝛿
𝛿
𝜀
Effects of Broadband Speed on Unemployment Rates – New Dummy Variable
(1′′) 𝑦 𝛽
𝐵𝐵
,𝛽𝑋
𝛾𝛿
𝛿
𝜀
Model (1′) Model (2′) Model (1′′)
Speed: High (β1) -0.2226*
(0.1196)
NA -0.2678*
(0.1546)
Speed: High1 (β1) NA -0.2103*
(0.1160)
NA
Speed: High2 (β2) NA -0.1822
(0.2249)
NA
Education -0.0646**
(0.0280)
-0.0638**
(0.0285)
-0.0669
(0.0420)
Working Age Population 0.0700
(0.0702)
0.0711
(0.0706)
0.0421
(0.1261)
Diversity 0.0125
(0.0338)
0.0120
(0.0339)
-0.0442
(0.0878)
Ln Population Density 7.1340**
(2.8947)
7.1513**
(2.8955)
6.8296*
(3.5909)
N 510 510 190
County Fixed Effects Yes Yes Yes
Year Fixed Effects Yes Yes Yes
R2 0.9620 0.9620 0.9828
Notes:
***, **, * are significant at 1%, 5% and 10%, respectively. Standard errors (in parentheses) are clustered at the
county level to address the potential issues related to heteroscedasticity and autocorrelation in the error terms.
Speed categorization for Model (1′): Low: < 100 Mbps; High: ≥ 100 Mbps
Speed categorization for Model (2′): Low: < 100 Mbps; High1: 100 Mbps to 1000 Mbps; High2: ≥ 1000 Mbps
In model (1′′): High: ≥ 80% population has access to 100 Mbps or higher; Low: ≥ 80% population has access to
less than 100 Mbps AND < 20% has access to 100 Mbps or more
