## ABSTRACT

The COVID−19 pandemic brought the digital divide to center stage. This article investigates whether the crisis disrupted mobile broadband infrastructure, taking Georgia as a case study. We hypothesize that the pandemic could have slowed down ongoing infrastructure provision initiatives, as in other segments of the economy, or spurred them by bringing renewed attention and resources to overcoming the digital divide. We find that the per capita antenna gap between rural and micropolitan areas as compared to metropolitan has drastically reduced during the pandemic. Long−Term Evolution expansion was positively associated with the presence of vulnerable populations with variation across areas.

Although the vital role of broadband in contemporary society is not a novelty, the outbreak of the COVID−19 pandemic in 2020 brought the challenge of unequal access to center stage. Stay−home restrictions moved working, learning, and other daily activities (such as grocery shopping, going to the doctor, or requesting governmental services) to the virtual space. In this context, access to the internet became a necessary precondition for engagement in economic and social life.

Studies from the initial stages of the pandemic raised the concern that the crisis could reinforce already−known racial, age, and spatial dimensions of unequal access to broadband, regarding infrastructure availability and quality of service.1 As such, the pandemic may have amplified and resignified long−known aspects of the digital divide, including access and digital literacy. Consequently, understanding which changes occurred and what populations were more affected has direct implications from the equity perspective.

In this article, we investigate whether the outbreak of the pandemic and the increased attention to providing adequate internet access was accompanied by increases in the provision of mobile broadband in Georgia. Mobile broadband is an increasingly important source of broadband access, becoming more common than home access to low−income families in urban areas2 and filling fixed−broadband gaps in rural ones.3 Such a role became even more pronounced during the pandemic, although there are constraints in the extent to which mobile broadband allows the complete execution of activities such as remote learning and working.4

Georgia makes an interesting case of study as it portrays a mix of urbanized and rural areas—translating into spatially driven digital divides—and is also one of the most diverse US contiguous states, with roughly 60 percent of non−White population.5 Further, bridging the digital divide has been on Georgia’s agenda in recent years, leading, for instance, to the launch of the Georgia Development Broadband Initiative (GDBI) to promote broadband deployment in unserved areas and improve access measurement. 6 In 2020, the state also passed a new regulation that reduced fees paid by cable companies to attach wires and cables to Electric Membership Cooperatives’ (EMCs) utility poles in rural areas considered unserved by the GDBI map, among other regulatory changes.7 These conditions are expected to stimulate long−run solutions for addressing digital divides in the state.

However, the direction of changes in broadband infrastructure during the pandemic is, in principle, indetermined and requires empirical exploration. On the one hand, federal and other investments in broadband expansion in response to COVID−19 strongly suggest that we should see substantial growth. On the other hand, infrastructure expansion may have been slowed down by the pandemic, in consonance with several other sectors of the economy. To distinguish between both possibilities in practice, we combine data on cellular infrastructure antennas from Skyhook—a location services company—with the American Community Survey (ACS) for 1,956 census tracts in Georgia and examine patterns of change before and during the pandemic outbreak. The direct observation of visible infrastructure contributes to overcoming the well−known difficulties of mapping mobile broadband, including the overstatement of infrastructure availability.

Specifically, we use a difference−in−differences approach to study the patterns of change in Long−Term Evolution broadband (LTE or 4G) visible infrastructure within and between metropolitan, micropolitan, and rural8 census tracts in Georgia from March 2019 to September 2020. Furthermore, we study the association of these changes with the sociodemographic characteristics of the populations in each area, as a strategy to identify how potentially vulnerable groups were affected. Our empirical strategy does not enable one to make causal claims but provides insights into which population groups enjoyed improved access and which ones were left behind.

Our results suggest that mobile broadband expansion continued in 2020, although at a slower speed as compared to 2019. The gaps in LTE antennas per thousand population between rural and micropolitan areas as compared to metropolitan areas have significantly reduced, especially in the former. By September 2020, the total antennas per capita was not that different across areas, although we still observe significant discrepancies when normalizing antennas by area—which has implications for broadband quality. Overall, we interpret these changes as positive signs toward bridging the digital divide in Georgia. Further, we find that, within areas during the pandemic, there has been a more significant expansion in census tracts where some vulnerable groups concentrate, such as the elderly (all areas) and populations without any internet subscription (rural). At the same time, we also observe an increase of antennas in areas with higher income in micropolitan tracts.

This article is organized into five sections that follow this introduction. In the next section, we briefly discuss the digital divide literature and the multidimensionality of the access concept, emphasizing the role of mobile broadband. Next, we further contextualize our discussion by introducing pandemic−motivated initiatives aimed at addressing the digital divide. We introduce our data and methods in the section “Data and Methods.” The “Results” section presents our main results, which are discussed in “Discussion and Conclusions.”

## Literature Review

### Digital Divide: Access Dimensions

Digital inequalities existed before the pandemic, but the crisis exacerbated the problem dramatically. Lack of access became a major vulnerability factor, with uneven predicted impacts across the population.9 For example, during the pandemic, elderly and rural populations faced increased difficulty accessing health services because of the lack of digital literacy or broadband access.10 Likewise, children in underserved areas and low−income families felt more substantial impacts on their schooling experience and learning process.11 Indeed, of the 50 million students sent home by school closures, roughly 18 percent lack internet access at home.12 Therefore, the need for broadband access during this moment put the understanding of digital inequalities into a natural starting point for this study.

The discussion on digital divides originated as information and communication technologies became widespread. At first, the debate focused on access, in which a binary outcome provides a direct indicator of exclusion—have versus have not. There was significant enthusiasm regarding the potential socioeconomic benefits that could result from granting access to excluded populations, such as increased economic growth,13 employment,14 and productivity.15

The access dimension speaks to two sides of the market in which the supply side entails infrastructure provision and the demand side adoption or subscription. Supply is a necessary but not sufficient condition to access, as infrastructure without subscription is ineffective. However, the incentives to increase access from both sides differ. On the one hand, infrastructure deployment involves cost considerations from the companies, such as geographic remoteness, population density, and the economic status of the residents—which in turn affects the likelihood of subscription to services.16 Estimates of deployment costs in rural and remote areas are usually much higher than in urban clusters, sometimes making the end price prohibitive.17

On the other hand, adoption depends on consumers’ decision to subscribe to services, which ultimately depends on their budgetary constraints, willingness to pay, quality, and service reliability.18 Further, lower−income households are typically more price−sensitive,19 and a recent study finds that half of the households without broadband subscriptions do not subscribe because prices are too high.20 Spatial dimensions such as concentrated poverty and neighborhood segregation also explain reduced adoption in certain areas.21 Studies show that increasing subscription rates can be difficult even in the presence of specific programs designed to do it.22

Once access is granted, a follow−up point of concern regards the quality of service. Internet speed, especially, is a common challenge to rural residents. There is evidence of a moderately negative relationship between increases in rurality and download speeds,23 and broadband performs worse in rural and tribal areas than in congested urban contexts.24 Technology−wise, the rural–urban digital divide shows up for faster forms of fixed and mobile broadband.25 At least partially, lack of competition explains the observed differences in quality.26

Another point often discussed in the digital divide literature—at the user’s end—regards literacy. Digital literacy refers to one’s ability to use technology efficiently and effectively and encompasses elements such as access to adequate equipment, autonomy, and experience with the technology.27 The COVID−19 pandemic has made literacy especially critical because people who were not acquainted with the digital world suddenly had to rely on it to access basic services. Elderly populations, for instance, became particularly vulnerable.28

In summary, the digital gap is a multilayered challenge, and the access dimension has several components, including infrastructure availability (provision/subscription), quality of service, and digital literacy. Inability to access the network by a particular group due to any combination of these components contributes to the social exclusion of said group,29 and access disparities are not even across the population. Instead, reduced access highly correlates with race/minority status, rural areas, lower income, and others.30 Therefore, these groups become particularly vulnerable during unprecedented pandemic times. Disparities in access may compound other structural disadvantages and exacerbate health inequalities.31 Hence, uncovering patterns of change and their association with sociodemographic factors during the pandemic is needed to understand these dynamics better.

### The Role of Mobile Broadband

As cellphone ownership grew, mobile broadband became an increasingly important access option. Currently, most of the US adult population owns a cellphone (97 percent) or a smartphone (85 percent)—much higher ownership rates than that of other devices such as desktop/laptop computers (77 percent) or tablets (53 percent).32

The extent to which mobile complements or substitutes fixed broadband varies across income and geography. For example, mobile broadband is known for filling gaps in rural areas where fixed broadband is unavailable, lower−speed suppliers predominate, or with difficult terrain.33 Within urban clusters, mobile and public broadband have been central (though not the only) access option to the less connected populations, notably in poor Black and Latin neighborhoods.34 Consequently, the expansion of mobile broadband can disproportionately benefit low−income and remote populations.

Nevertheless, there are limits to how much mobile broadband may substitute fixed connectivity, especially when computer or tablet devices are unavailable or inadequate. Exclusive mobile access reduces the breadth of possible online activities and limits the adequate performance of activities such as working or schooling.35 In the context of stay−home restrictions, these limitations highlight the problem of under−connectivity beyond the binary disconnectivity.

Under−connectivity speaks to several dimensions of the digital divide, including mobile dependency, broadband quality, data plans and prices, adequate per capita device ownership, and others. Unsurprisingly, vulnerable populations are more likely to experience problems related to under−connectivity. For example, one year into the pandemic, a study found that roughly one−quarter of families in poverty (headed by Latino immigrants or someone with less than a high school degree) relied solely on mobile, dial−up, or had no internet access. Further, 34 percent of those with exclusive mobile access reported hitting data limits, and 16 percent had the service interrupted due to lack of payment. Those with computers and home broadband also reported under−connectedness issues.36

Furthermore, the pandemic imposed additional challenges on the operator’s end. Changes in mobility caused by stay−home restrictions and lockdown measures quickly translated into impacts on how people use telecommunications networks, increasing traffic volume and modifying patterns of use. In the United States, evidence suggests that internet service providers have successfully responded to increase demand by aggressively adding capacity at interconnects.37 In the United States and Europe, part of the successful response stems from advances in network automation and deployment, for instance, by relying on automated configuration management and robots to install cross−connects in internet exchange points.38 In the United Kingdom, changes in mobility patterns seem to have reduced overall traffic in mobile networks, as people at home relied more heavily on fixed broadband, varying by density.39 Yet, given the already discussed heterogeneous reliance on cellular infrastructure by different population groups in the United States, the United Kingdom pattern may not readily apply to all groups.

Therefore, while celebrating any mobile expansion during the pandemic—given its essential role as an access source—we acknowledge that it will most likely be a partial solution to the current digital divide challenges. Fully addressing the issue will require attention to socioeconomic dimensions beyond pure infrastructure.

## Fostering Access in a Pandemic World

The attention that the pandemic brought to the digital inequalities and their social implications have spurred initiatives and resources to bridge the gap. Such initiatives combine efforts from all government levels and the private sector.

For example, as part of the Coronavirus Aid, Relief, and Economic Security (CARES) Act, the Coronavirus Relief Fund (CRF) designated $150 billion for local governments to cover pandemic−related costs throughout the end of 2020, including broadband expansion.40 As of November 2020, states had used these sources to expand broadband in four primary areas: “increasing access to online learning for K−12 and postsecondary students, supporting telehealth services, deploying more public Wi−Fi access points, and investing in residential broadband infrastructure, especially in rural and underserved areas.”41 The Keep Americans Connected Pledge was another initiative proposed by the Federal Communications Commission (FCC) and signed by over 800 telecommunication companies and associations. In the pledge, companies committed not to terminate residential or small business services and waive late fees caused by their inability to pay bills because of COVID−19 disruptions from March 13 to June 20. Companies also committed to opening their Wi−Fi hotspots to facilitate access.42 Beyond the pledge, operators have individually worked to improve network performance and expand capacity, as discussed.43 Local governments also implemented novel solutions for improving access. Some school districts across the nation have equipped school buses with routers or mobile hotspots to bring connectivity to unserved populations.44 Likewise, public libraries and other public institutions have opened their Wi−Fi networks to be accessible from parking lots in these facilities.45 Further, several schools provided laptops and tablets to students in 2020. Equipment provision is especially valuable in rural areas without adequate access and where families rely on mobile internet data plans to connect—which significantly limits the possibility of accessing higher bandwidth content, such as videos. Although such initiatives paint a hopeful portrait of change in access, evidence documenting any changes in infrastructure during the pandemic is still lacking. This article helps fill this gap by bringing a dynamic picture of the mobile broadband infrastructure during the pandemic, leveraging proprietary data from the location services company, Skyhook, which will be discussed further in the Data section. Before concluding this section, we acknowledge that efforts at several government levels to bridge the digital divide were already in place before the pandemic. For example, in 2011, the Federal Government approved the Connect America Fund—CAF,46 a large−scale mechanism to subsidize the deployment of fixed and mobile broadband to unserved areas in the United States—mainly rural and tribal. CAF’s initial budget was$4.5 billion per year for six years.47 In 2020, then, the FCC announced the Rural Digital Opportunity Fund allocating another $20.4 billion over ten years to finance high−speed broadband networks in unserved rural areas48 At the state level, efforts include establishing programs, creating funding mechanisms, setting service speed and goals, redefining who can provide broadband service (to foster competition), and defining how internet service providers can access public−owned infrastructure to build their systems, including sidewalks, roads, and telephone poles.49 In Georgia, as previously mentioned, initiatives to foster access—originating before and during the pandemic—included legislation changes, data measurement improvement, and other incentives. As was the case for several sectors of the economy, it is plausible that the pandemic had slowed down the implementation of these initiatives. If that is true, then the net effect of the pandemic−related expansion initiatives—which we try to capture in our analysis—may have been “canceled” by a reduction in the implementation rate of pre−pandemic programs. Although disentangling these effects is beyond the scope of this article, measuring the net effect empirically is an important endeavor of its own as it speaks to a dimension of potential social exclusion. ## Data and Methods ### Data Our analysis combines two main data sources. The first comes from Skyhook—a geo−positioning service provider that logs anonymized records of users’ mobile device locations. Skyhook’s Global Cell ID Location Database Archive provides monthly snapshots of LTE cell ids (henceforth “antennas”) as seen by mobile devices, therefore, allowing us to observe changes in the existing infrastructure.50 Antenna data include geographical coordinates, enabling the linkage with the census tracts defined by the US Census Bureau. We restricted each monthly count to antennas that Skyhook had observed in the previous 90 days to avoid double−counting or counting infrastructure that was no longer in use.51 Skyhook has provided monthly snapshots of LTE visible infrastructure from March 2019 to September 2020.52 By using visible infrastructure data, our article speaks to the provision dimension of access. Skyhook data enables us to observe the landscape of the mobile infrastructure in a timelier fashion than other traditional sources, which are often not readily available in real−time. Sociodemographic information of each census tract comes from the ACS 5 years estimates 2019, our second main data source.53 The ACS provides the population’s racial, age, income composition, and household broadband subscription rates. By linking broadband infrastructure with sociodemographic information, not only are we able to identify where there has been expansion, but also who benefitted from it. Our dependent variable is LTE antennas per thousand population, generated by dividing the total antennas in each census tract and month by the population (as per 2019 ACS). While this variable provides a good measure of coverage, we acknowledge that other factors not included in this analysis also affect access, including distance, terrain, and others. Some of these factors (e.g., distance) are potentially more relevant in rural areas, where census tracts are typically larger. Additional data sources include the GDBI data on unserved areas and the US COVID−19 Cases and Deaths by USA Facts.54 The former illustrates the access dimension from the provision perspective, whereas the latter helps contextualize how the different areas were affected by the global pandemic in the health realm. ### Methods We initiate our empirical analysis with a description of changes in the broadband scene in Georgia. At this point, we present the spatial differences across metropolitan, micropolitan, and rural areas to show the spatial disparity in access. We label census tracts based on where they fall in the Office of Management and Budget’s (OMB) classification of metropolitan and micropolitan statistical areas. Metropolitan statistical areas comprise agglomerations of one or more counties that contain at least one urbanized area of at least fifty thousand people. Micropolitan statistical areas consist of at least a single urban cluster with a population between ten and fifty thousand. The remaining counties are classified as rural.55. In a second step, we use difference−in−differences models to describe whether there were significant changes in patterns between areas during the pandemic compared to the prior year. Our first specification is the linear regression shown in equation 1. $yi,t^=αt+βpandemict+γstatusi+θpandemict*statusi+δXi$ (1) The dependent variable, $yi,t^$, represents LTE antennas per thousand population (as of 2019) in each census tract i in month t from March 2019 to September 2020. Pandemict is a dummy equal to one for the months following the outbreak (March 2020 onwards) and zero otherwise (March 2019 to February 2020). Statusi indicates if the census tract lies in a metropolitan, micropolitan, or rural area. The interaction coefficient, $θ$, directly allows us to answer the main question that motivated this work, namely: whether rural and micropolitan areas—the most likely to be unserved or underserved—experienced a distinct trend in broadband expansion following the pandemic outbreak.The term $αt$ represents a vector of month dummies to capture time trends common to all areas and Xi comprises a vector of census tract sociodemographic and economic characteristics gathered from the 2019 ACS, including the share of elderly (age above 65) and kids (age below 15), the share of Black and Latino population, median household income, and the share of households without any internet subscription. These variables capture characteristics that could represent increased vulnerability during pandemic times. The vector Xi also includes population density per square mile to control for differences in the potential consumer base. Finally, we cluster standard errors at the tract level to account for potential heteroskedasticity.56 Following this model, we stratify the analysis by area type and introduce interaction terms between the pandemic indicator and the population sociodemographic characteristics, as in equation 2. This strategy enables us to observe whether any changes are associated with the concentration of demographic groups within areas. Given the increased vulnerabilities imposed upon certain groups during the pandemic, identifying these differences is of particular interest. As in equation 1, the coefficient $θ$ reveals if there were significant changes in the pattern of expansion during the pandemic as compared to prior to it, except that in 2 we focus on their association with demographics rather than area type. $yi,t^=αt+βpandemict+δXi+θpandemict*Xi$ (2) We note that our differences−in−differences strategy captures differences in trends across and within area types. It does not provide a traditional difference−in−differences setting in which one of the areas is untreated. This is the case because the pandemic has impacted all areas, even if to different degrees. Given such a limitation, our results aim at describing differences rather than making causal claims. ## Results ### Descriptive Results Georgia has 1,956 populated census tracts in 2010s Census definition, of which 78.8 percent fall within metropolitan, 11.1 percent within micropolitan, and 10.2 percent within rural counties. The digital divide in the state has marked spatial dimensions: Unsurprisingly, broadband access reduces as one moves from metropolitan to rural areas in both access dimensions—provision and subscription. Figure 1 provides a spatial visualization of these variations at the census tract level, revealing an overlapping but incomplete relationship between provision and access. On the provision side, the GBDI data shows that 9.9 percent of census blocks within metropolitan tracts are considered unserved. However, these figures are much higher for micropolitan (35.8 percent) and rural areas (51.7 percent). On the subscription end, ACS 2019 data reveals a similar story: 18.0 percent of the households in metropolitan tracts had no internet subscription, whereas the figures for micropolitan and rural areas were 30.2 percent and 34.6 percent, respectively.57 Further, 13.8 percent of the households in rural areas rely exclusively on cell phone data plans to access the internet (12.8 percent in micropolitan areas and 10.6 percent in metropolitan areas). Cell phone data plans are often associated with more limited use than other types of broadband. FIGURE 1 Access to broadband in Georgia tracts. FIGURE 1 Access to broadband in Georgia tracts. Close modal However, these access numbers refer to a period before the COVID−19 pandemic, and there is reason to contend that the crisis might have disrupted the broadband scene, especially in unserved or underserved areas. Table 1 illustrates the state of the LTE infrastructure in Georgia during our period of analysis, according to Skyhook cellular data. TABLE 1 LTE Cell ids Distribution in Georgia in March and September 2020 MetroMicroRural MonthTotalPer 1k pop.Per areaTotalPer 1k pop.Per areaTotalPer 1k pop.Per area Mar−19 90,842 10.54 3.65 8,227 8.13 0.41 6,449 8.31 0.51 Apr−19 97,652 11.33 3.93 8,861 8.76 0.44 7,002 9.02 0.56 May−19 102,333 11.88 4.12 9,387 9.28 0.47 7,425 9.57 0.59 Jun−19 103,255 11.98 4.15 9,485 9.37 0.47 7,516 9.69 0.60 Jul−19 104,455 12.12 4.20 9,531 9.42 0.47 7,636 9.84 0.61 Aug−19 105,941 12.30 4.26 9,968 9.85 0.50 8,042 10.36 0.64 Sep−19 108,298 12.57 4.36 10,209 10.09 0.51 8,265 10.65 0.66 Oct−19 111,271 12.91 4.48 10,504 10.38 0.52 8,618 11.11 0.69 Nov−19 111,949 12.99 4.50 10,627 10.50 0.53 8,729 11.25 0.70 Dec−19 110,580 12.83 4.45 10,584 10.46 0.53 8,728 11.25 0.70 Jan−20 109,782 12.74 4.42 10,606 10.48 0.53 8,686 11.19 0.69 Feb−20 110,791 12.86 4.46 10,748 10.62 0.54 8,773 11.31 0.70 Mar−20 112,355 13.04 4.52 10,976 10.85 0.55 9,001 11.60 0.72 Apr−20 113,661 13.19 4.57 11,140 11.01 0.56 9,270 11.95 0.74 May−20 113,389 13.16 4.56 11,127 11.00 0.55 9,202 11.86 0.73 Jun−20 113,282 13.15 4.56 11,237 11.11 0.56 9,354 12.05 0.75 Jul−20 111,351 12.92 4.48 11,125 11.00 0.55 9,257 11.93 0.74 Aug−20 113,410 13.16 4.56 11,518 11.38 0.57 9,792 12.62 0.78 Sep−20 114,148 13.25 4.59 11,649 11.51 0.58 9,927 12.79 0.79 Growth (Mar−19 to Feb−20) 1.82% 2.46% 2.84% Growth (Mar−20 to Sep−20) 0.26% 1.00% 1.65% Difference (pp) −1.56 −1.46 −1.19 Census tracts 1,540 217 199 Population 8,616,036 1,011,786 776,025 Land area (sq. feet) 24,859 20,068 12,554 MetroMicroRural MonthTotalPer 1k pop.Per areaTotalPer 1k pop.Per areaTotalPer 1k pop.Per area Mar−19 90,842 10.54 3.65 8,227 8.13 0.41 6,449 8.31 0.51 Apr−19 97,652 11.33 3.93 8,861 8.76 0.44 7,002 9.02 0.56 May−19 102,333 11.88 4.12 9,387 9.28 0.47 7,425 9.57 0.59 Jun−19 103,255 11.98 4.15 9,485 9.37 0.47 7,516 9.69 0.60 Jul−19 104,455 12.12 4.20 9,531 9.42 0.47 7,636 9.84 0.61 Aug−19 105,941 12.30 4.26 9,968 9.85 0.50 8,042 10.36 0.64 Sep−19 108,298 12.57 4.36 10,209 10.09 0.51 8,265 10.65 0.66 Oct−19 111,271 12.91 4.48 10,504 10.38 0.52 8,618 11.11 0.69 Nov−19 111,949 12.99 4.50 10,627 10.50 0.53 8,729 11.25 0.70 Dec−19 110,580 12.83 4.45 10,584 10.46 0.53 8,728 11.25 0.70 Jan−20 109,782 12.74 4.42 10,606 10.48 0.53 8,686 11.19 0.69 Feb−20 110,791 12.86 4.46 10,748 10.62 0.54 8,773 11.31 0.70 Mar−20 112,355 13.04 4.52 10,976 10.85 0.55 9,001 11.60 0.72 Apr−20 113,661 13.19 4.57 11,140 11.01 0.56 9,270 11.95 0.74 May−20 113,389 13.16 4.56 11,127 11.00 0.55 9,202 11.86 0.73 Jun−20 113,282 13.15 4.56 11,237 11.11 0.56 9,354 12.05 0.75 Jul−20 111,351 12.92 4.48 11,125 11.00 0.55 9,257 11.93 0.74 Aug−20 113,410 13.16 4.56 11,518 11.38 0.57 9,792 12.62 0.78 Sep−20 114,148 13.25 4.59 11,649 11.51 0.58 9,927 12.79 0.79 Growth (Mar−19 to Feb−20) 1.82% 2.46% 2.84% Growth (Mar−20 to Sep−20) 0.26% 1.00% 1.65% Difference (pp) −1.56 −1.46 −1.19 Census tracts 1,540 217 199 Population 8,616,036 1,011,786 776,025 Land area (sq. feet) 24,859 20,068 12,554 Source: Skyhook (2020) and American Community Survey 2019 (5 years estimates). Author’s tabulations. Note: Compounded growth rates. The absolute number of antennas varies across areas, being most concentrated in metropolitan census tracts. These differences are significantly reduced when normalized by population. Distinctly, the normalization by land area reveals a much higher concentration of antennas in metropolitan areas as compared to micro and rural tracts. The lower density of antennas in rural and micropolitan areas reduces the accessibility and affects service quality. Interestingly, by September 2020, rural areas had higher access to LTE than micropolitan centers both in terms of antennas per population and per area. Over the period, there has been a continuous LTE expansion, especially in micro and rural tracts, which is expected given the current incentives in place. Between March 2019 and February 2020, there was, on average, an increase of 1.8 percent antennas in metro areas per month, 2.5 percent in micro, and 2.8 percent in rural (as per compounded growth rates). From March, however, all areas experienced speed reductions of 1.6 percentage points (pp), 1.5 pp, and 1.2 pp in metro, micro, and rural, respectively. Unfortunately, we cannot distinguish whether the expansion speed reductions would have naturally occurred absent the pandemic—for instance, in the case providers judged they had achieved a saturation point. We note that the expansion in mobile broadband infrastructure during the first months of 2020—especially in nonmetropolitan areas—is consistent with the observed increase in internet service providers’ capacity in the first quarter of 2020.58 For a clearer visualization of trends, Figure 2 portrays the mean antennas per capita of census tracts by area type. Rural Georgia shows a trend of increasing provision that is distinct from both metropolitan and micropolitan areas. In fact, when it comes to mobile LTE broadband, rural areas seem to be catching up with metropolitan areas, especially after June 2020. Meanwhile, micropolitan census tracts show a seemingly parallel trend compared to metropolitan ones, without any obvious inflection points. While this is good news for rural residents, it does not necessarily imply that provision across rural and metropolitan areas is equal, given the differences in density and the availability of other forms of broadband, such as fiber. FIGURE 2 Mean LTE antennas per 1,000 population by area type (Mar 2019–Sep 2020). FIGURE 2 Mean LTE antennas per 1,000 population by area type (Mar 2019–Sep 2020). Close modal Indeed, we must acknowledge that differences across areas go further beyond access. Substantial sociodemographic variation distinguishes rural, micropolitan, and metropolitan areas, and accounting for such differences allows for identifying the populations that benefit from any infrastructure improvement. As per Table 2, rural and micropolitan tracts are much less dense and have higher proportions of White and elderly populations than metropolitan ones while also having higher poverty rates. These areas have also been unevenly affected by the pandemic itself. Looking at the county level—for which COVID-19 data is available—we observe that both rural and micropolitan areas in Georgia suffered disproportionately more from the pandemic, as illustrated by the higher COVID-19 case and death rates. TABLE 2 Sociodemographic Composition and Covid−19 Incidence AreaPop. per sq. feetNon−White (%)Children (%)Elderly (%)Poverty rate (%)COVID−19 cases per thousand pop.*COVID−19 deaths per thousand pop.* Metropolitan 2,182.8 50.7 19.6 13.3 16.3 28.7 0.7 Micropolitan 350.5 36.2 19.5 16.3 21.3 37.1 1.2 Rural 60.9 33.9 17.4 20.6 21.5 34.1 1.1 AreaPop. per sq. feetNon−White (%)Children (%)Elderly (%)Poverty rate (%)COVID−19 cases per thousand pop.*COVID−19 deaths per thousand pop.* Metropolitan 2,182.8 50.7 19.6 13.3 16.3 28.7 0.7 Micropolitan 350.5 36.2 19.5 16.3 21.3 37.1 1.2 Rural 60.9 33.9 17.4 20.6 21.5 34.1 1.1 Source: American Community Survey 2019 (5−years estimates) and USA Facts, COVID−19 deaths and Cases up to September 30, 2020. Author’s tabulations. Note: * County average of COVID-19 cases and deaths per thousand population. ### Regressions: Between Areas This section examines our research question from an inferential approach. Table 3 portrays the between analysis results, in which we pool metropolitan, micropolitan, and rural tracts together. The results suggest a higher prevalence of mobile LTE during the pandemic, in which there were 2.8 more antennas per thousand population as compared to before the outbreak (pandemic coefficient). Before the pandemic, micropolitan and rural areas had, respectively, 3.4 and 2.4 fewer expected LTE antennas per population as compared to metropolitan areas (columns 1). However, such a difference was reduced during the pandemic as shown by the positive interaction coefficients, a conclusion robust to the inclusion of census tracts’ sociodemographic characteristics. Specifically, during the pandemic, the difference decreased to 2.7 (-3.37+0.70) and 1.0 (-2.36+1.41) in micropolitan and rural areas, respectively. Therefore, and according to the descriptive section, the antennas per capita gap between metropolitan and rural areas during the pandemic has nearly disappeared. TABLE 3 OLS Regressions on Antennas per Thousand Population Variables(1)(2) Pandemic 2.83*** 2.83*** (0.22) (0.22) Micropolitan −3.37*** −4.14*** (0.72) (0.64) Rural −2.36*** −4.36*** (0.75) (0.75) Pandemic * micropolitan 0.70*** 0.70*** (0.26) (0.26) Pandemic * rural 1.41*** 1.41*** (0.27) (0.27) Log. population density −0.58* (0.33) Log. median household income 0.89 (1.50) Elderly share −0.25** (0.12) Child share −0.62*** (0.14) Black share 0.04** (0.02) Latino share −0.02 (0.04) Households with no subscription share 0.06 (0.08) Variables(1)(2) Pandemic 2.83*** 2.83*** (0.22) (0.22) Micropolitan −3.37*** −4.14*** (0.72) (0.64) Rural −2.36*** −4.36*** (0.75) (0.75) Pandemic * micropolitan 0.70*** 0.70*** (0.26) (0.26) Pandemic * rural 1.41*** 1.41*** (0.27) (0.27) Log. population density −0.58* (0.33) Log. median household income 0.89 (1.50) Elderly share −0.25** (0.12) Child share −0.62*** (0.14) Black share 0.04** (0.02) Latino share −0.02 (0.04) Households with no subscription share 0.06 (0.08) Notes: 1. Standard errors clustered by census tract in parentheses. 2. All models include month fixed effects. 3. N = 37,088. 4. ***p < 0.01, **p < 0.05, *p < 0.1 By adding the demographic characteristics in column 2, the main conclusions remain similar. However, we can now observe that LTE antennas per thousand population correlate with specific area characteristics. Each additional percentage point in the share of children and elderly is associated with decreases in expected antennas in the order of 0.62 and 0.025, respectively, holding the remaining variables in the model constant. Hence, the results suggest more extensive provision in areas with working−age adults. Further, areas with larger shares of Black population are expected to have more antennas, a characteristic more common in metropolitan areas in Georgia. Distinctly, we find no convincing evidence that provision correlates with income or service subscription. ### Regressions: Within Area As most census tracts in Georgia fall within metropolitan areas and given the substantial differences in sociodemographic composition—even in census tracts’ sizes—we now turn to within area analysis. Tables 4 to 6 present the specifications from equation 2 stratified by area type. These models allow us to identify if there have been any changes in how LTE infrastructure correlates with population characteristics during the crisis. For each interaction model, the coefficient on the stand-alone (noninteracted) independent variable refers to its association with antennas per capita prior to the pandemic, whereas the coefficient interacted with pandemic tells whether there has been a change in that association during it.59 TABLE 4 Within Metropolitan Areas: OLS Regressions on Antennas per 1,000 Population Variables(1)(2)(3)(4)(5)(6)(7)(8) Pandemic 2.68*** 7.03*** 0.20 3.25*** 2.74*** 1.24 3.62*** 2.86*** Log. population density −0.34 −0.11 −0.34 −0.34 −0.34 −0.34 −0.34 −0.34 Log. median household income 1.43 1.43 1.35 1.43 1.43 1.43 1.43 1.43 Black share 0.04* 0.04* 0.04* 0.04* 0.04* 0.04* 0.04* 0.04* Latino share −0.03 −0.03 −0.03 −0.03 −0.02 −0.03 −0.03 −0.03 Elderly share −0.33** −0.33** −0.33** −0.33** −0.33** −0.37** −0.33** −0.33** Child share −0.70*** −0.70*** −0.70*** −0.70*** −0.70*** −0.70*** −0.68*** −0.70*** No subscription share 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 Pandemic* log. density −0.62*** Pandemic* log. income 0.23 Pandemic* Black share −0.02 Pandemic* Latino share −0.01 Pandemic* elderly share 0.11* Pandemic* child share −0.05 Pandemic* no subscription share −0.01 R−squared 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 Variables(1)(2)(3)(4)(5)(6)(7)(8) Pandemic 2.68*** 7.03*** 0.20 3.25*** 2.74*** 1.24 3.62*** 2.86*** Log. population density −0.34 −0.11 −0.34 −0.34 −0.34 −0.34 −0.34 −0.34 Log. median household income 1.43 1.43 1.35 1.43 1.43 1.43 1.43 1.43 Black share 0.04* 0.04* 0.04* 0.04* 0.04* 0.04* 0.04* 0.04* Latino share −0.03 −0.03 −0.03 −0.03 −0.02 −0.03 −0.03 −0.03 Elderly share −0.33** −0.33** −0.33** −0.33** −0.33** −0.37** −0.33** −0.33** Child share −0.70*** −0.70*** −0.70*** −0.70*** −0.70*** −0.70*** −0.68*** −0.70*** No subscription share 0.10 0.10 0.10 0.10 0.10 0.10 0.10 0.10 Pandemic* log. density −0.62*** Pandemic* log. income 0.23 Pandemic* Black share −0.02 Pandemic* Latino share −0.01 Pandemic* elderly share 0.11* Pandemic* child share −0.05 Pandemic* no subscription share −0.01 R−squared 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 Notes: 1. Standard errors clustered by census tract in parentheses. 2. All models include month fixed effects. 3. N = 29,184. 4. *** p < 0.01, ** p < 0.05, * p < 0.1 TABLE 5 Within Micropolitan Areas: OLS Regressions on Antennas Per 1,000 Population Variables(1)(2)(3)(4)(5)(6)(7)(8) Pandemic 3.74*** 6.42*** −4.06 3.96*** 3.86*** 2.66*** 4.59*** 3.74*** Log. population density −1.89*** −1.68*** −1.89*** −1.89*** −1.89*** −1.89*** −1.89*** −1.89*** Log. median household income 1.25 1.25 0.98 1.25 1.25 1.25 1.25 1.25 Black share −0.00 −0.00 −0.00 0.00 −0.00 −0.00 −0.00 −0.00 Latino share −0.03 −0.03 −0.03 −0.03 −0.03 −0.03 −0.03 −0.03 Elderly share 0.04 0.04 0.04 0.04 0.04 0.01 0.04 0.04 Child share −0.06 −0.06 −0.06 −0.06 −0.06 −0.06 −0.05 −0.06 No subscription share −0.02 −0.02 −0.02 −0.02 −0.02 −0.02 −0.02 −0.02 Pandemic* log. density −0.55*** Pandemic* log. income 0.73** Pandemic* Black share −0.01* Pandemic* Latino share −0.02 Pandemic* elderly share 0.07*** Pandemic* child share −0.04* Pandemic* no subscription share −0.00 R−squared 0.26 0.26 0.26 0.26 0.26 0.26 0.26 0.26 Variables(1)(2)(3)(4)(5)(6)(7)(8) Pandemic 3.74*** 6.42*** −4.06 3.96*** 3.86*** 2.66*** 4.59*** 3.74*** Log. population density −1.89*** −1.68*** −1.89*** −1.89*** −1.89*** −1.89*** −1.89*** −1.89*** Log. median household income 1.25 1.25 0.98 1.25 1.25 1.25 1.25 1.25 Black share −0.00 −0.00 −0.00 0.00 −0.00 −0.00 −0.00 −0.00 Latino share −0.03 −0.03 −0.03 −0.03 −0.03 −0.03 −0.03 −0.03 Elderly share 0.04 0.04 0.04 0.04 0.04 0.01 0.04 0.04 Child share −0.06 −0.06 −0.06 −0.06 −0.06 −0.06 −0.05 −0.06 No subscription share −0.02 −0.02 −0.02 −0.02 −0.02 −0.02 −0.02 −0.02 Pandemic* log. density −0.55*** Pandemic* log. income 0.73** Pandemic* Black share −0.01* Pandemic* Latino share −0.02 Pandemic* elderly share 0.07*** Pandemic* child share −0.04* Pandemic* no subscription share −0.00 R−squared 0.26 0.26 0.26 0.26 0.26 0.26 0.26 0.26 Notes: 1. Standard errors clustered by census tract in parentheses. 2. All models include month fixed effects. 3. N = 4,123. 4. *** p < 0.01, ** p < 0.05, * p < 0.1 TABLE 6 Within Rural Areas: OLS Regressions on Antennas Per 1,000 Population Variables(1)(2)(3)(4)(5)(6)(7)(8) Pandemic 5.20*** 8.98*** 4.17 5.33*** 5.36*** 4.51*** 5.66*** 4.34*** Log. population density −2.81*** −2.43*** −2.81*** −2.81*** −2.81*** −2.81*** −2.81*** −2.81*** Log. median household income 0.53 0.53 0.49 0.53 0.53 0.53 0.53 0.53 Black share 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 Latino share −0.02 −0.02 −0.02 −0.02 −0.01 −0.02 −0.02 −0.02 Elderly share 0.20** 0.20** 0.20** 0.20** 0.20** 0.19** 0.20** 0.20** Child share −0.00 −0.00 −0.00 −0.00 −0.00 −0.00 0.01 −0.00 No subscription share −0.01 −0.01 −0.01 −0.01 −0.01 −0.01 −0.01 −0.02 Pandemic * log. density −1.02*** Pandemic * log. income 0.10 Pandemic * Black share −0.00 Pandemic * Latino share −0.03 Pandemic * elderly share 0.03* Pandemic * child share −0.03 Pandemic * no subscription share 0.02** R−squared 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 Variables(1)(2)(3)(4)(5)(6)(7)(8) Pandemic 5.20*** 8.98*** 4.17 5.33*** 5.36*** 4.51*** 5.66*** 4.34*** Log. population density −2.81*** −2.43*** −2.81*** −2.81*** −2.81*** −2.81*** −2.81*** −2.81*** Log. median household income 0.53 0.53 0.49 0.53 0.53 0.53 0.53 0.53 Black share 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.02 Latino share −0.02 −0.02 −0.02 −0.02 −0.01 −0.02 −0.02 −0.02 Elderly share 0.20** 0.20** 0.20** 0.20** 0.20** 0.19** 0.20** 0.20** Child share −0.00 −0.00 −0.00 −0.00 −0.00 −0.00 0.01 −0.00 No subscription share −0.01 −0.01 −0.01 −0.01 −0.01 −0.01 −0.01 −0.02 Pandemic * log. density −1.02*** Pandemic * log. income 0.10 Pandemic * Black share −0.00 Pandemic * Latino share −0.03 Pandemic * elderly share 0.03* Pandemic * child share −0.03 Pandemic * no subscription share 0.02** R−squared 0.22 0.22 0.22 0.22 0.22 0.22 0.22 0.22 Notes: 1. Standard errors clustered by census tract in parentheses. 2. All models include a running month time trend. 3. N = 3,781. 4. *** p < 0.01, ** p < 0.05, * p < 0.1 Within metropolitan areas, there were more LTE per capita antennas in areas with lower shares of children and elderly. During the pandemic, interaction terms suggest that there has been a reduced expansion in areas with lower density and an increase associated with the share of elderly, although none of the coefficients is large in magnitude (Table 4). For the remaining variables, there is no convincing evidence of changes in the prior trends, namely expansion concentrated in areas with higher shares of Black people and lower shares of children and elderly. The within micropolitan models (Table 5) reveal a very distinct picture. Before the pandemic, the population density was the only variable significantly related to the number of LTE antennas per thousand population—in a negative direction. All the other demographic variables did not seem to explain antenna deployment, as shown by the insignificant coefficients without interactions. However, during the pandemic, expansion was significantly higher in tracts with lower density, larger income, and higher concentrations of elderly populations (as per interaction coefficients). During the pandemic, a 1 percent increase in median household income in a micropolitan census tract was associated with additional 0.02 antennas per capita ((0.98+0.73)/100). Finally, in rural tracts before the pandemic, density and concentration of elderly were the only sociodemographic variables associated with LTE antennas, both trends reinforced during the pandemic (Table 6). Furthermore, during the pandemic, there has been a larger expansion associated with need: a one pp increase in the share of households without any internet subscription was associated with an expected increase of 0.02 antennas per capita. While the coefficient is not large, the continuity of this trend has the potential to close the existing digital inequities. ## Discussion and Conclusions In this article, we portray changes in broadband infrastructure following the pandemic outbreak in 2020, hypothesizing that it could have either slowed down ongoing infrastructure provision initiatives or spurred them by bringing renewed attention and resources to the digital divide. Georgia was selected as a case study given its unique population and spatial composition, which features demographic diversity and large nonmetropolitan areas. Rural and micropolitan areas have higher proportions of unserved populations when compared to metropolitan ones. Our results show that these differences have reduced during the pandemic, especially in rural census tracts, which seem to be catching up faster to metropolitan levels of antennas per capita. Specifically, the gap between rural and metropolitan areas shrunk from −1.44 antennas per capita in March 2020 to −0.46 in September 2020, while the micropolitan-metropolitan difference reduced from −2.19 to 1.74. These changes represent great news for the populations in those areas, especially given the stronger reliance of certain groups on mobile service, including lower-income and those in remote locations. Within areas, our findings uncover changes in the patterns of expansion during the pandemic regarding the population socioeconomic characteristics. Overall, there has been an increased expansion in areas with elderly concentration. Rural populations with higher shares of households without internet subscriptions also saw a disproportional growth in antennas. In micropolitan areas, expansion during the pandemic was larger in areas with higher income. Yet, most of these interactions reflect marginal changes that do not immediately translate into disruptive economic changes (but could do so if continued). While revealing positive changes during a challenging time, our study incurs some limitations. First, we provide descriptive data on how infrastructure availability changed in Georgia during a crisis, but we cannot causally attribute these changes to the pandemic itself. However, and reassuringly, studies documenting internet service providers’ responses suggest that network improvements represented breakpoints with prior trends.60 Second, we acknowledge that antennas per capita may not signify the same access levels across areas, given differences in census tract size, population density, and broadband infrastructure availability beyond LTE. Furthermore, the limits of mobile broadband to the performance of specific activities such as working or studying have been well-documented during the pandemic. Under-connectedness remains a key challenge to be addressed to assure real equity in access.61 Finally, given that our data is specific to Georgia, an unfortunate limitation is not being able to observe if the positive findings we uncover apply to other states. Based on recent studies,62 however, it does seem that there have been substantial advancements towards closing the digital divide in other parts of the country as well. ### APPENDIX FIGURE A1 Infrastructure Composition and Change in March and September 2020. FIGURE A1 Infrastructure Composition and Change in March and September 2020. Close modal ## FOOTNOTES 1. Beaunoyer; Bronzino et al.; Lai and Widmar. 2. Fernandez. 3. Prieger, “The Broadband Digital Divide and the Economic Benefits of Mobile Broadband for Rural Areas” 4. Katz and Rideout. 5. U.S. Census Bureau, “American Community Survey 5 Years Estimates 2015−2019. Prepared by Social Explorer.” 6. Georgia launched the GDBI initiative in 2018. Among others, GDBI aims to provide more accurate (and far less optimistic) measures of broadband provision than the traditionally used Form 477 by the FCC. While the FCC defines a census block as served if at least one residential or business location has the target broadband service available, GDBI’s served definition requires availability on at least 80 percent of the locations in each block.” 8. We label census tracts in metropolitan, micropolitan, and rural counties based on the Office of Management and Budget’s (OMB) classification. See "Methods" section for precise definitions. 9. Beaunoyer. 10. Seifert; Ramsetty and Adams; Lai; Marston. 11. Goldschmidt. 12. Wheeler. 13. Crandall and Jackson. 14. Pociask. 15. Ferguson. 16. Reddick et al. 17. Rendon Schneir and Xiong; Oyana. 18. Flamm and Chaudhuri; Lyons. 19. Liu, Prince, and Wallsten, “Distinguishing Bandwidth and Latency in Households Willingness−to−Pay for Broadband Internet Speed.” 21. Mossberger et al., “Unraveling Different Barriers to Internet Use.” 22. Hauge and Prieger; Manlove and Whitacre; Whitacre and Gallardo. 23. Lai and Widmar (2021) 24. Adarsh et al. 25. Prieger (2013) 26. Ibid. 27. Beaunoyer. 28. Seifert. 29. Ibid. 30. Ramsetty and Adams; Fitzhugh et al.; Prieger, “The Supply Side of the Digital Divide: Is There Equal Availability in the Broadband Internet Access Market?”; Seifert; Adarsh et al.; Parker. 31. Eruchalu et al. 32. Pew Research Center, “Mobile Fact Sheet” 33. Prieger (2013) 34. Mossberger, Tolbert, and Anderson, “The Mobile Internet and Digital Citizenship in African−American and Latino Communities.” 35. Fernandez. 36. Katz and Rideout 37. Liu, Schmitt, Bronzino, and Feamster, “Characterizing Service Provider Response to the COVID−19 Pandemic in the United States” 38. Feldmann et al. 39. Lutu. 40. U.S. Department of the Treasury, “Assistance for State, Local, and Tribal Governments.” 41. De Wit and Read, “States Tap Federal CARES Act to Expand Broadband: Coronavirus Relief Funding Supports Access and Infrastructure.” p. 1 42. Federal Communications Commission (FCC), “Keep Americans Connected Pledge.” 43. Cellular Telecommunications Industry Association (CTIA), “The Wireless Industry Responds to COVID−19.” 44. Reardon. 45. Kang. 46. Federal Communications Commission (FCC), “Order 11−161.” 47. This figure equals the budget allocated to the High−Cost program – CAF’s antecessor – in 2011. Essentially, CAF reformed the Universal Service Fund, which was designed in a time when FCCs priority was to guarantee access to telephone services to the population. See Nazareno and Jose (2021) for a detailed discussion of the CAF reform. 48. Federal Communications Commission (FCC), “Auction 904: Rural Digital Opportunity Fund”; In its first phase, the Rural Digital Opportunity Fund will allocate$16 billion to census blocks where existing data shows there is no broadband (defined at a minimum of 25/3 Mbps) available whatsoever. Later, a second phase will allocate $4.4 billion to target partially served areas. The funding will be allocated, similarly to CAF−phase II auction, in a competitive reversed auction, yet the speed requirements were more than double than previously required. 49. De Wit and Read. 50. Skyhook. 51. Once a cell id is observed, it is retained in the Skyhook database. When an id is not seen for an extended period, this may indicate that the infrastructure was reclassified (to a new id), deactivated, or not seen due to changes in device mobility. After discussing how to best minimize these potential measurement errors with Skyhook experts, we studied the possibility of using three different “last time seen” thresholds: 30 days, 90 days, and 180 days. In our analysis, we report the 90 days threshold, which seems more conservative than 180 days, but also less likely to be affected by pandemic mobility changes such as the 30 days. The main conclusions remain consistent across the different thresholds. 52. LTE is the largest and fastest expanding air interface across all areas, as illustrated in Figure A1 in the appendix. 53. U.S. Census Bureau, “American Community Survey 5 Years Estimates 2015−2019. Prepared by Social Explorer.” 54. COVID−19 deaths and cases were collected at the county level from January 22 to September 30, 2020.” 55. U.S. Census Bureau, Understanding and Using American Community Survey Data: What Users of Data for Rural Areas Need to Know; We note that the metropolitan, micropolitan, and rural classification is not equivalent to rural versus urban areas, as counties classified as metropolitan or micropolitan may have rural areas. 56. Bertrand. 57. No access to broadband of any type (including cellular data plan, cable, fiber optic, DSL, and satellite) follows a close pattern: 18.2 percent in metropolitan areas, 30.5 in micropolitan, and 35.2 in rural. 58. Bronzino et al. 59. For example, in Table 4, column 6, we find that before the pandemic each pp increase in the share of elderly in metropolitan areas was associated with a decrease in antennas per capita of 0.37 all else being equal. During the pandemic, however, there was a reduction in that difference of 0.11, such that each pp increase in the share of elderly was associated with 0.26 fewer antennas per capita (−.037+0.11), therefore, a reduction in the negative association. 60. Liu, Schmitt, Bronzino, and Feamster (2020), Feldmann et al. 61. Fernandez. 62. Katz and Rideout ## BIBLIOGRAPHY Adarsh, Vivek et al, “Coverage Is Not Binary: Quantifying Mobile Broadband Quality in Urban, Rural and Tribal Contexts.” In IEEE ICCCN . 2021 . https://doi.org/10.1109/ICCCN52240.2021.9522152. Beaunoyer, Elisabeth, Dupéré Sophie, and Matthieu J. Guitton. “COVID-19 and Digital Inequalities: Reciprocal Impacts and Mitigation Strategies.” Computers in Human Behavior 111 , no. May ( 2020 ). https://doi.org/10.1016/j.chb.2020.106424. Bertrand, Marianne, Esther Duflo, and Sendhill Mullainathan. “How Much Should We Trust Difference-in-Differences Estimates?” The Quarterly Journal of Economics . no. February ( 2004 ): 249 75 . https://doi.org/10.1162/003355304772839588. Bronzino, Francesco et al, TPRC48: The 48th Research Conference on Communication, Information and Internet Policy . 2021 . https://ssrn.com/abstract=3786158. Crandall, Robert W, and Charles L Jackson. “The$500 Billion Opportunity: The Potential Economic Benefit of Widespread Diffusion of Broadband Internet Access.”
2001
.
De Wit, Kathryn De, and Anna Read.
“States Tap Federal CARES Act to Expand Broadband: Coronavirus Relief Funding Supports Access and Infrastructure
. ”
Issues Brief. The Pew Charitable Trusts
.
November
2020
Eruchalu, Chukwuma N. et al,
“The Expanding Digital Divide: Digital Health Access Inequities during the COVID−19 Pandemic in New York City.”
Journal of Urban Health
98
(
2021
):
183
86
. https://doi.org/10.1007/s11524−020−00508−9.
Federal Communications Commission (FCC)
.
“Auction 904: Rural Digital Opportunity Fund.”
2020
. https://www.fcc.gov/auction/904.
________.
“Keep Americans Connected Pledge.”
2020
. https://www.fcc.gov/keep−americans−connected.
Feldmann, Anja et al,
“The Lockdown Effect: Implications of the COVID−19 Pandemic on Internet Traffic.”
In
ACM Internet Measurement Conference (AMC’20)
. Vol.
20
.
2020
. https://doi.org/10.1145/3419394.
Ferguson, Charles H.
“The United States Broadband Problem: Analysis and Policy Recommendations.”
2002
Fernandez, Laleah, Bianca C. Reisdorf, and William H. Dutton.
“Urban Internet Myths and Realities: A Detroit Case Study.”
Information Communication and Society
23
, no.
13
(
2020
):
1925
46
. https://doi.org/10.1080/1369118X.2019.1622764.
Fitzhugh, Earl et al,
“Covid−19: Investing in Black Lives and Livelihoods.”
2020
. https://www.mckinsey.com/~/media/mckinsey/industries/public. and social sector/our insights/covid 19 investing in black lives and livelihoods/covid−19−investing−in−black−lives−and−livelihoods−report.pdf
Flamm, Kenneth, and Anindya Chaudhuri.
“An Analysis of the Determinants of Broadband Access.”
Telecommunications Policy
31
(
2007
):
312
26
. https://doi.org/10.1016/j.telpol.2007.05.006.
Georgia Department of Community Affairs
.
2020
Goldschmidt, Karen.
“The COVID−19 Pandemic: Technology Use to Support the Wellbeing of Children.”
Journal of Pediatric Nursing
53
(
2020
):
88
90
. https://doi.org/10.1016/j.pedn.2020.04.013.
Hauge, J A, and J E Prieger.
“Evaluating the Impact of the American Recovery and Reinvestment Act’s BTOP on Broadband Adoption.”
2015
. https://doi.org/10.1080/00036846.2015.1080810.
Kang, Cecilia.
“Parking Lots Have Become a Digital Lifeline.”
The New York Times
May
5
,
2020
. https://www.nytimes.com/2020/05/05/technology/parking−lots−wifi−coronavirus.html.
Katz, Vikki, and Victoria Rideout.
“Lower−Income Families During the COVID−19 Pandemic Learning at Home While Under−Connected.”
2021
. www.vjrconsulting.com.
Lai, John, Nicole O. Widmar, and Courtney Bir.
“Eliciting Consumer Willingness to Pay for Home Internet Service: Closing the Digital Divide in the State of Indiana.”
Applied Economic Perspectives and Policy
42
, no.
2
(
2020
):
263
82
. https://doi.org/10.1002/aepp.13000.
Lai, John, and Nicole O. Widmar.
“Revisiting the Digital Divide in the COVID−19 Era.”
Applied Economic Perspectives and Policy
43
, no.
1
(
2021
):
458
64
. https://doi.org/10.1002/aepp.13104.
Liu, Shinan, Paul Schmitt, Francesco Bronzino, and Nick Feamster.
“Characterizing Service Provider Response to the COVID−19 Pandemic in the United States.”
In
International Conference on Passive and Active Network Measurement (PAM 2021), 12671 LNCS
,
20
38
.
Cham
:
Springer
,
2021
. https://doi.org/10.1007/978−3−030−72582−2_2.
Liu, Yu−Hsin, Jeffrey Prince, and Scott Wallsten.
“Distinguishing Bandwidth and Latency in Households Willingness−to−Pay for Broadband Internet Speed.”
Information Economics and Policy
45
(
2018
):
1
12
. https://doi.org/10.1016/j.infoecopol.2018.07.001.
Lutu, Andra, Diego Perino, Marcelo Bagnulo, et al; , .
“A Characterization of the COVID−19 Pandemic Impact on a Mobile Network Operator Traffic.”
In
ACM Internet Measurement Conference (IMC’20)
,
19
33
. (Association for Computing Machinery,
2020
. https://doi.org/10.48550/arxiv.2010.02781.
Lyons, Sean.
Empirical Economics
47
(
2014
):
1341
63
. https://doi.org/10.1007/s00181−013−0790−6.
Manlove, Jacob, and Brian Whitacre.
Telecommunications Policy
43
(
2019
). https://doi.org/10.1016/j.telpol.2019.02.003.
Marston, Hannah R., Charles Musselwhite, and Robin Hadley.
“COVID−19 vs Social Isolation: The Impact Technology Can Have on Communities, Social Connections and Citizens.”
The British Society of Gerontology Blog
2020
. https://ageingissues.wordpress.com/2020/03/18/covid−19−vs−social−isolation−the−impact−technology−can−have−on−communities−social−connections−and−citizens/?fbclid=IwAR1sUsffKNd_G5u6d_oc0Z56u4Es7HyoCJYKr0qSnqFxX68pD3PY5JaSl7g.
Mossberger, Karen, Caroline J. Tolbert, Daniel Bowen, and Benedict Jimenez.
“Unraveling Different Barriers to Internet Use: Urban Residents and Neighborhood Effects.”
Urban Affairs Review
48
, no.
6
(
2012
):
771
810
. https://doi.org/10.1177/1078087412453713.
Mossberger, Karen, Caroline J. Tolbert, and Christopher Anderson.
“The Mobile Internet and Digital Citizenship in African−American and Latino Communities.”
2016
. https://doi.org/10.1080/1369118X.2016.1243142.
Nazareno, Luísa, and Justina Jose.
““The Effects of Broadband Deployment in Rural Areas: Evaluating the Connect America Fund Program.”
In. ”
TPRC49: The 49th Research Conference on Communication, Information and Internet Policy
.
2021
. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3897867.
Oyana, Tonny J.
“Exploring Geographic Disparities in Broadband Access and Use in Rural Southern Illinois: Who’s Being Left Behind?”
Government Information Quarterly
28
(
2011
):
252
61
. https://doi.org/10.1016/j.giq.2010.09.003.
Parker, Edwin B.
“Closing the Digital Divide in Rural America.”
Telecommunications Policy
24
(
2000
):
281
90
. https://doi.org/10.1016/S0308−5961(00)00018−5.
Pew Research Center
.
2019
________.
“Mobile Fact Sheet.”
April
7
,
2021
. https://www.pewresearch.org/internet/fact−sheet/mobile/.
“Building a Nationwide Broadband Network: Speeding Job Growth.”
2002
.
Prieger, James E.
“The Broadband Digital Divide and the Economic Benefits of Mobile Broadband for Rural Areas.”
Telecommunications Policy
37
, no.
6–7
(
2013
):
483
502
. https://doi.org/10.1016/j.telpol.2012.11.003.
________.
“The Supply Side of the Digital Divide: Is There Equal Availability in the Broadband Internet Access Market?”
Economic Inquiry
41
, no.
2
(
2003
):
346
63
. https://doi.org/10.1093/ei/cbg013.
Prieger, James E, and Wei−Min Hu.
“The Broadband Digital Divide and the Nexus of Race, Competition, and Quality.”
Information Economics and Policy
20
(
2008
):
150
67
. https://doi.org/10.1016/j.infoecopol.2008.01.001.
“Impact of the Digital Divide in the Age of COVID−19.”
Journal of the American Medical Informatics Association
27
, no.
7
(
2020
):
1147
48
. https://doi.org/10.1093/jamia/ocaa078.
Reardon, Marguerite.
“Coronavirus Lockdown Spotlights Digital Divide in Distance Learning.”
CNET
April
24
,
2020
. https://www.cnet.com/news/coronavirus−lockdown−spotlights−digital−divide−in−distance−learning/.
Reddick, Christopher G, Roger Enriquez, Richard J. Harris, and Bonita Sharma.
“Determinants of Broadband Access and Affordability: An Analysis of a Community Survey on the Digital Divide.”
Cities
106
(
2020
). https://doi.org/10.1016/j.cities.2020.102904.
Rendon Schneir, Juan, and Yupeng Xiong.
“A Cost Study of Fixed Broadband Access Networks for Rural Areas.”
Telecommunications Policy
40
(
2016
):
755
73
. https://doi.org/10.1016/j.telpol.2016.04.002.
Seifert, Alexander.
“The Digital Exclusion of Older Adults during the COVID−19 Pandemic
. ”
Journal of Gerontological Social Work
63
, no.
6−7
(
2020
):
674
76
. https://doi.org/10.1080/01634372.2020.1764687.
Skyhook
.
“Global CellID Location Database Archive.”
2020
.
________.
Understanding and Using American Community Survey Data: What Users of Data for Rural Areas Need to Know
Washington, DC
:
US Government Printing
,
2019
. https://www.census.gov/content/dam/Census/library/publications/2019/acs/ACS_rural_handbook_2019.pdf.
USAFacts
.
“Georgia Coronavirus Cases and Deaths.”
2020
US Census Bureau
.
“American Community Survey 5 Years Estimates 2015−2019. Prepared by Social Explorer.”
2019
.
US Department of the Treasury
.
“Assistance for State, Local, and Tribal Governments.”
2020
. https://home.treasury.gov/policy−issues/coronavirus/assistance−for−state−local−and−-tribal−governments.
Wheeler, Tom.
“5 Steps to Get the Internet to All Americans.”
2020
. https://www.brookings.edu/research/5−steps−to−get−the−internet−to−all−americans/.
Whitacre, Brian, and Roberto Gallardo.
“State Broadband Policy: Impacts on Availability.”
Telecommunications Policy
44
(
2020
): 102025. https://doi.org/10.1016/j.telpol.2020.102025.