Professional Documents
Culture Documents
Author(s): Shefali Dahiya, Lila N. Rokanas, Surabhi Singh, Melissa Yang and Jon M. Peha
Source: Journal of Information Policy , 2021, Vol. 11 (2021), pp. 202-221
Published by: Penn State University Press
REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/10.5325/jinfopoli.11.2021.0202?seq=1&cid=pdf-
reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
This content is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0
International License (CC BY-NC-ND 4.0). To view a copy of this license, visit
http://creativecommons.org/licenses/by-nc-nd/4.0/.
Penn State University Press is collaborating with JSTOR to digitize, preserve and extend access to
Journal of Information Policy
Abstract
When COVID-19 hit, many people began working, going to school, and living
much of their lives from home. The Internet was a gateway to the world. This
article uses data from Internet speed tests, consumer complaints, search engine
optimization tools, and logs of Internet use from public libraries to understand
the effects of the pandemic on Internet use and performance. Despite reports
that the Internet handled the surge in traffic well, we find that complaints about
Internet speed nearly tripled, and performance was degraded. Downstream data
rates changed little, but median upstream data rates at midday dropped by about
a third. When discussing Internet performance, people typically focus on down-
stream. This focus should shift. Internet service providers and policymakers should
reduce the asymmetry by changing how infrastructure is designed, how Internet
services are advertised, how regulators write transparency rules, and how govern-
ment defines “broadband” in subsidy programs intended to reduce the digital
divide. We also find significant increases in the use of many important categories
of online content, including those used for work communications, education, gro-
cery shopping, social media, news, and job searches. This shows the importance
of the Internet during the crisis. Many people without Internet at home turned to
public Wi-Fi hotspots during the pandemic. We find that this occurred dispropor-
tionately in neighborhoods with more students. Future distance learning initiatives
should consider the challenges some students face in obtaining Internet access.
Keywords: broadband, COVID-19, digital divide, Internet performance, speed
tests
The outbreak of COVID-19 provided a massive stress test for the Internet.
Before COVID, people might access the Internet from many places,
including schools, workplaces, universities, public libraries, and cafes. As
attempts to prevent the spread of COVID made many of those places inac-
cessible, much of the Internet communications had to shift to the home,
or be lost entirely. Before COVID, we could not know how people would
use the Internet during a crisis of this kind, how well Internet infrastruc-
ture would handle this shift, or how those who lack Internet access from
home would respond. This article sheds light on those mysteries. In the
likely event that people continue to use their home Internet connection
to work, go to school, access healthcare, and engage in other demanding
activities, our results have important implications for the future
It is well established that Internet use increased with COVID-19, but the
impact on quality of service has been debated. One prominent professor in
this area opined that “the coronavirus is breaking the Internet,”1 but that is
not a common view. Internet service providers (ISPs) certainly disagreed.
As Comcast stated on its corporate website, “During COVID-19, network
traffic spiked more than 30% as people rapidly transitioned to working and
learning from home. Comcast’s network continued to deliver fast speeds,
even under the heaviest usage, and even in the areas most severely affected
by COVID-19.”2 Fortune magazine concluded after obtaining quotes from
a number of large US ISPs that “Internet service providers maintain that
their networks are performing well.”3 A former Chief Economist of the
President’s Council of Economic Advisors even published an article in the
nation’s most widely read newspaper asserting in its title that “Internet
providers are handling coronavirus demand just fine.”4
Unfortunately, this is not how many Internet users saw it. Although it
has not been reported as far as we know, this article will show that during
this period when many ISPs were reporting that their services were perform-
ing well, user complaints about Internet performance were skyrocketing.
These facts are not as contradictory as they may first appear. We cannot
know what ISPs were seeing within their own networks, but our analysis of
results from open-source Internet “speed tests” shows that there was little
change in the performance metric that ISPs use most: downstream data
rate. However, there was a substantial change in upstream performance,
1. Meinrath.
2. Comcast.
3. Pressman.
4. Holtz-Eakin.
and this may matter more. We conclude that in a world where people are
doing many of life’s activities from home, we should not be so focused on
downstream data rate as the measure of performance. This observation
has broad implications. We may need to change how Internet infrastruc-
ture is designed, how Internet services are advertised, how those adver-
tisements are regulated by agencies such as the Federal Communications
Commission (FCC), and what kinds of systems government should subsi-
dize to close the digital divide.
When policymakers decide whether to provide those subsidies, they
should also consider how important a role the Internet played in people’s
lives. We therefore quantify any changes in the use of Internet for vari-
ous purposes of interest, including remote work and education, social dis-
tancing during a pandemic, access to credible health information, and job
search and advancement during the accompanying economic downturn.
Given these potentially important uses, those Internet users who merely
saw a reduction in the quality of their home broadband service due to
COVID-19 were lucky, compared to those who had no home Internet
service at all. To support this group in the future, we should better under-
stand who they are and what they need. We examined the use of publicly
accessible Wi-Fi hotspots in 73 Pennsylvania libraries at a time when the
libraries were closed, but Wi-Fi was turned on. One might expect that
usage would be greatest in neighborhoods where the most residents could
not afford Internet at home. We found that income was a factor, but not as
important as the number of students in the region. This may indicate that
online education drove many people to library hotspots even when those
hotspots were especially difficult to access. This has implications for future
distance learning initiatives.
COVID-19 put on vivid display what is best and worst about the
Internet. The Internet made it easy for some people to work, go to school,
and interact with loved ones safely without leaving their homes. Those
lucky individuals maintained an extraordinary degree of normalcy and
safety despite a pandemic that in some regions disrupted nearly every
aspect of daily life. At the other extreme were those who normally used
the Internet outside the home. Their access to the Internet was severed
just when they needed it most. Some could not go to school, earn a pay-
check, apply for a new job, or gather the best information about how to
stay safe during a pandemic. In between were households that did have
Internet access, but not at the quality of service that would make all of
these activities easy. By examining Internet use and performance during
the pandemic, we gain insights into how ISPs might design their networks
to better meet subscriber needs in the future, and how government might
adopt policies that encourages them to do so.
The section “Methods and Data Sources” presents our data sources
for this research, and methods of analysis. Results are presented in the
“Results” section, and conclusions in the “Conclusion” section.
and technology. Within the Internet category, consumers can select the
technology they receive Internet through (Cable, Digital Subscriber Line
[DSL], Fiber, Satellite, and Wireless) and the specific issue (Availability,
Billing, Equipment, Interference, Privacy, and Speed) their complaint is
about. Although complaints can be filed by mail and phone, filing a com-
plaint is easiest online, which probably means complaints from people
with poor or no Internet access are underrepresented.
Results
This section presents the results of our analysis. The section “Content”
shows how extensively different types of online content and applica-
tions were accessed during the pandemic. The section “The Internet User
Experience During COVID” examines Internet performance as observed by
users, as reflected in both direct measurements and consumer complaints.
Finally, the section “Public Access to Internet” addresses the use of public
Wi-Fi hotpots.
Content
In March 2020, when some parts of the country were shutting down in
response to the pandemic, demand for some types of online content and
applications began to grow. Figure 1 shows the trends for several example
categories. Each curve shows the number of unique visitors in each month,
divided by the number of unique visitors in February 2020, which is the
last month before any US city entered lockdown. In many (but not all)
cases, usage increased dramatically from February to May, and then begin
a gradual decline. A spike in usage for any category may indicate that the
Internet was meeting a related need. (We show only a subset of the catego-
ries described in “Analyzing the Use of Online Content” as examples so that
all curves shown will be clearly visible.)
The impact of COVID is perhaps best demonstrated by examining the change
from February 2020 to May 2020. Figure 2 shows the percent change of both num-
ber of visits per month, and the number of unique visitors per month. Clearly the
Internet was important for those working from home, as work fueled some
particularly large changes. Zoom led the way, perhaps because Zoom use
is not limited to work. The number of visits and unique visitors to Zoom
increased over 400% and 700%, respectively. (And this does not include
accesses from the mobile app.) Use of other work communications systems
that we analyzed (Hangouts, Slack) roughly tripled. Some of those people
not busy working from home were improving their work skills, or looking
for jobs. Use of our adult education sites roughly doubled. The number
of unique visitors to job search sites increased by about 40%, presumably
because the Internet was an important resource for those who had lost
their jobs in the economic downturn.
The Internet allowed people to get their shopping done without going
to stores where a virus can spread. The number of unique visitors to gro-
cery sites tripled, although the number of visits did not change very much.
Visits to Amazon and eBay also increased by 28% and 46%, respectively.
Use of Internet for communications also increased, perhaps as an antidote
to the isolation that comes with social distancing. Visits to social media
sites increased by 67%, and visits to email sites increased by 73%. There
was a smaller but still significant increase in visits to entertainment sites,
perhaps as people trapped at home looked for things to do.
The biggest surprise is that despite the need for health information
during a life-threatening pandemic, the number of visits to authoritative
sources of health information such as the Center for Disease Control,
WebMD, and the Mayo Clinic did not change significantly. However, the
number of visits to more general news sites increased by 26%, and COVID-
19 was constantly in the news. Some people also get such information from
friends via social media and email, which saw increases. Websites dedicated
entirely to COVID-19 were not included, since it is meaningless to talk
about how usage of a site increased during the pandemic if the site did not
exist before the pandemic.
Internet Performance
In this section, we show Internet performance as measured using the
M-Lab NDT “speed tests” described in the section “Analyzing Internet
Performance.” Upstream median data rates were affected far more than
downstream. Figure 3 shows the median upstream and downstream data
rates obtained throughout the United States in a given month. The down-
stream rates did drop about 6% from February 2020 before any part of
the country went into lockdown to April of 2020, but rates were already
back above prepandemic levels by May. In contrast, median upstream rates
dropped 22% from February to April, and did not recover until September.
The return to normal throughputs presumably occurred in part because
ISPs made an admirable effort to expand capacity during this crisis.
The difference between upstream and downstream is probably because
of the tremendous growth in use of Zoom and other videoconferencing
applications, as we showed in the section “Content.” These applications
generate a great deal of upstream traffic. Some have argued that the pri-
mary reason for Internet performance degradation is congestion in Wi-Fi
links coming from neighboring households.11 Although the M-Lab mea-
surements do not differentiate congestion in the networks of ISPs and
congestion in Wi-Fi networks inside individual households, if the latter
were the sole problem, we would expect significant performance drops
upstream and downstream. It is therefore likely that in-home networks
were not the sole or primary source of upstream performance degradation,
although they contribute.
These effects vary greatly by time of day. Figure 4 shows median data rate
in the United States upstream and downstream at different times of day
(expressed in the Eastern time zone). Performance did not change much in
the middle of the night when most people were sleeping, but the impact
was great in midday when many were working and going to school from
figure 3 Median Data Rate over a Month. (a) Upstream. (b) Downstream.
11. Ergen.
figure 4 Median Data Rate versus Time of Day (Eastern Time Zone) in Different
Months. (a) Upstream. (b) Downstream.
for services offering downstream rates close to the 10th percentile, which is
under 2 Mb/s.
Consumer Complaints
Another way to measure users’ satisfaction with the Internet is by track-
ing trends in consumer complaints to the FCC. As shown in Figure 6,
nearly all complaints are about either billing, availability, or speed, and
these complaints increased dramatically during the pandemic. Complaints
about billing, availability, and speed increased from February to April by
24%, 85% and 176%, respectively.
The pain was not spread equally across technologies, as shown in
Figure 7. The largest increases by far were in complaints about speed from
subscribers to cable and satellite services, while the speed complaints about
DSL remained almost unchanged during the pandemic. The increase in
complaints about wireless and fiber fell somewhere in between. It is not
whether the service is “faster” or “slower” that led to more complaints, since
fiber typically offers the best performance and DSL among the worst. The
figure 5 Change in Data Rate from February 2020 to given Month at the 10th, 30th,
50th, 70th, and 90th Percentile. (a) Upstream. (b) Downstream.
services that experienced the greatest increase in speed complaints were the
most asymmetric services, that is, the ones where downstream capabilities
most exceed upstream capabilities. This is consistent with the results of the
section “Internet Performance,” which showed that upstream performance
was degraded far more than downstream, and with the results in the sec-
tion “Content,” which showed that usage of more symmetric applications
like Zoom increased greatly during the pandemic.
Individuals who did not have Internet access from home faced challenges
far greater than a reduction in upstream data rates. For some, the best way
to take classes, look for a job, or stay in touch with family was to find a
building that offered free Internet to the public, see whether the Wi-Fi
was still operating, and then sit awkwardly on the sidewalk outside with
a portable or mobile device. Many people obtained this vital service from
their public library.
As discussed in the section “Analyzing Wi-Fi Use at Public Libraries,”
we examined the use of publicly accessible Wi-Fi in 73 libraries in Western
Pennsylvania. Even though all of these libraries closed on March 14 due
to the pandemic, thousands of patrons were using the Wi-Fi in sessions
that lasted 2 to 2.5 hours on average, which is a long time to sit on the
sidewalk. As shown in Figure 8, sessions after lockdown were even longer
than when patrons could sit comfortably inside the library. Sessions in
May and June were still long, although shorter than April. The decrease
may be because universities and public schools broke for summer in May
and June, respectively.
table 1 Correlations between the Use of Library Wi-Fi Hotspots between March 14 and
June 9 and Demographic Characteristics of Zip Code Surrounding Library
Minutes Sessions
Income −0.15 −0.15
Median income from 2018
Adult 0.15 0.09
Fraction of pop. between 18 and 65
Black −0.01 0.08
Fraction of pop. Black
Hispanic 0.01 −0.02
Fraction of pop. Hispanic
Poverty 0.29 0.23
Fraction of pop. in poverty
Family −0.34 −0.23
Fraction of pop. in family
household
Child −0.15 −0.08
Fraction of pop. under 18
Bachelor’s Degree −0.17 −0.10
Fraction of pop. with Bachelor’s
Unemployed 0.17 0.13
Unemployment rate from 2018
In school 0.55 0.31
Fraction of pop. enrolled in school
shows the correlations between minutes of Internet use and the number of
Internet sessions at library hotspots and various demographic factors of the
surrounding neighborhood. By far the best predictor of minutes of Wi-Fi
use during the pandemic is the percentage of the neighborhood popula-
tion that are in school, including everything from kindergarten to college.
This is not because households with students are less likely to have
broadband at home. Households containing a student who uses the
Internet at school are more motivated to subscribe to an Internet service.12
Census data shows that in 2015, 77.2% of US households had broadband,13
whereas 85% of households with K-12 school-age children had broadband
in 2015 according to a Pew Research Center study.14 This percentage is
even greater for households with college students. A recent survey found
that even among college students from low-income households, 86% had
broadband at home.15
Although students are less likely to lack Internet at home than nonstu-
dents, those students who did lack Internet at home probably had a more
desperate need for Internet access. In spring of 2020, K-12 schools and the
universities in this region ended in-person instruction, and began offering
classes online. This presumably brought many students to the sidewalk
near a closed public library. The remote education plan apparently did not
provide students adequate assistance with Internet access. Some of these
students may even have had Internet at home, but not at a quality that met
the needs of everyone in their household.
We also see more use of public Wi-Fi in neighborhoods with a higher
poverty rate, and a larger fraction of the population that do not live with a
family member. This is not surprising, since households in poverty are less
likely to subscribe to an Internet service, and many students in community
colleges and universities do not live with family.
Conclusion
There are many indications that the Internet was a valuable resource during
the pandemic, and the lockdown and economic crisis that it precipitated.
Our results show that people increased their use of the Internet to look
for work, to access educational material, to maintain social ties despite
social distancing, and to buy groceries and other items without entering
stores where the virus can spread. This shows some of the benefits of mak-
ing Internet access widely available. Bridging the digital divide is a public
health issue. By far the biggest increases we saw were in the use of Zoom
and other work-related communications tools. For some, these tools were
essential for employment, or education.
These changes also substantially increased Internet traffic. The Internet
remained usable under this load, but performance was degraded, primar-
ily for upstream traffic. ISPs were accustomed to carrying streaming video
to tens of millions of homes every evening, so there was already capacity
to handle much of the downstream traffic for videoconferencing, but not
the upstream. That is presumably why upstream data rates at midday fell
by roughly a third during the pandemic, and why complaints about speed
increased by 176%. These increases were much greater in services that
offer highly asymmetric service, for example, 291% for cable and 213% for
satellite, whereas the increase for slow but symmetric DSL was just 15%.
Today, people tend to focus primarily on the downstream capabilities of
Internet services, as if the upstream did not matter. The implications for ISPs
are obvious. Even after COVID-19 has been tamed, we will probably see more
people working and going to school from home than before the pandemic. To
prepare for that possibility, ISPs that offer asymmetric services should reevalu-
ate their plans for upstream capacity, or risk becoming less competitive.
The implications for policymakers are at least as important. When
upstream is poor, some of the Internet uses that should be priorities for
policymakers are especially impaired. As too many students and parents
learned the hard way during the pandemic, upstream capacity is often
required for effective distance learning. It is also critical for many tele-
health applications. For example, the pandemic forced policymakers and
insurance companies to allow and pay for mental health services delivered
over the Internet. As a result, some rural areas that have been chronically
underserved by mental health providers for decades could finally access
these services, but effective interactions with a psychologist often require
enough upstream capacity for two-way video. Policymakers can help.
The FCC defines a service as “broadband” if the downstream is at least
25 Mb/s and the upstream is at least 3 Mb/s. The majority of households
that contain people working from home are far better off with a service
that is 20 Mb/s down and 20 Mb/s up than a service that is 100 Mb/s
down and 3 Mb/s up, but only the latter would meet the FCC’s current
acknowledgments
We thank the eiNetwork and the Carnegie Library of Pittsburgh for shar-
ing both data and their many insights related to use of library resources
during the pandemic, and for their great service to the community at a
appendix
The websites for our analysis of use of online content discussed in the
sections “Analyzing the Use of Online Content” and “Content” were selected
as follows. To select specific sites within any given category, we generally
used articles (cited below) about which websites are most popular and
SEMrush website popularity rankings. Listed websites with far fewer vis-
its in the SEMrush traffic analysis than others in the same category were
dropped. As SEMrush can only provide traffic analysis for domains and
subdomains, we also only include websites that provide services from mul-
tiple categories if the service associated with the desired category was given
its own subdomains (e.g., Amazon was not included as a grocery seller
because groceries are not in a separate subdomain.) See Table 2.
17 Lua.
18 “Top 15 Most Popular News Websites | February 2020.”
19 “Who Are the Top 10 Grocers in the United States (2019)?”; Kristic.
20 Gerber.
21 Half.
22 Ross.
bibliography
Anderson, Monica, and Andrew Perrin. “Nearly One-In-Five Teens Can’t Always Finish Their
Homework Because of the Digital Divide.” Pew Research Center, October 26, 2018.
Comcast. “Our Network.” June 10, 2020. https://corporate.comcast.com/our-network.
Dovrolis, C., K. Gummadi, A. Kuzmanovic, and S. D. Meinrath. “Measurement Lab:
Overview and an Invitation to the Research Community.” ACM SIGCOMM Computer
Communications Review 40, no. 3 (July 2010), 53–56.
The Education Trust West. “The Digital Divide in Higher Education,” October 6, 2020. https://
west.edtrust.org/resource/the-digital-divide-in-higher-ed/.
Ergen, Mustafa. “COVID-19 Makes your Internet Slow Down? Real Problem–Your Neighbor is
Interfering You!.” Horasis, May 4, 2020.
Federal Communications Commission, “Consumer Complaint Center.” Accessed July 2020.
https://consumercomplaints.fcc.gov/hc/en-us.
Gerber, Scott. “9 Most Effective Apps for Internal Communication.” Business.com, January 9, 2015.
https://www.business.com/articles/9-most-effective-apps-for-internal-communication.
Half, Robert. “25 Best Job Search Websites.” Robert Half, April 28, 2020. https://www.roberthalf.
com/blog/job-market/10-best-job-search-websites.
Healthcare Consumer Navigator Center. “Top 15 Most Popular Health Websites.” Accessed July 20,
2020. https://healthcareconsumernavigatorcenter.com/consumer-information-navigator/
top-15-popular-health-websites.
Holtz-Eakin, Douglas, “Who Needs Net Neutrality? Internet Providers are Handling
Coronavirus Demand Just Fine.” USA Today, May 11, 2020.
Krstic, Zee, and Amina L. Abdelrahman. “10 Best Grocery Delivery Services to Use in 2020.”
Good Housekeeping, April 14, 2020. https://www.goodhousekeeping.com/food-products/
g28039081/best-grocery-delivery-services.
Lua, Alfred. “21 Top Social Media Sites to Consider for Your Brand.” Buffer. Accessed July 20,
2020. https://buffer.com/library/social-media-sites.
Measurement Lab. “NDT (Network Diagnostic Tool).” https://www.measurementlab.net/tests/ndt.
Meinrath, Sascha. “The Coronavirus Pandemic is Breaking the Internet.” The Hill, May 2, 2020.
Peha, Jon M., and Ning Guan. “How the Public Accesses the Internet from Pennsylvania
Libraries.” Carnegie Mellon University Report, May 2018.
Pressman, Aaron. “Coronavirus Quarantining has Slowed Internet Speeds in Some Cities.”
Fortune Magazine, March 20, 2020.
Ross, Sharen. “You Need To Visit These 20 Websites If You Want To Learn New Skills.”
Lifehack. Accessed July 20, 2020. https://www.lifehack.org/352144/you-need-visit-these-
20-websites-you-want-learn-new-skills.
SEMrush. SEMrush Toolkit for SEO, November 12, 2019.
SEMrush, “What Is the Difference between Traffic Analytics and Domain Analytics?” Accessed
July 18, 2020. https://www.semrush.com/kb/858-traffic-numbers-in-semrush.
SEMrush, “Where Does SEMrush Data Come From.” Accessed July 18, 2020. https://www
.semrush.com/kb/998-where-does-semrush-data-come-from.
Statista. “U.S. households with Internet Subscription 1997-2019.” Statista. https://www.statista
.com/statistics/189349/us-households-home-internet-connection-subscription.
Tengtrakul, P., and J. M. Peha. “Does ICT in Schools Affect Residential Adoption and Adult Utilization
Outside Schools?” Telecommunications Policy 37, no. 6–7 (July–August 2013): 540–562.
“Top 15 Most Popular News Websites | February 2020.” eBiz MBA. Accessed 20 July 2020.
http://www.ebizmba.com/articles/news-websites.
Tucker, Ilyas. “12 Great Reasons to Use SEMrush to Supercharge Your Content Marketing.”
SEMrush Blog, May 20, 2016. https://www.semrush.com/blog/12-great-reasons-to-use-
semrush-to-supercharge-your-content-marketing.
“What Is SEO?.” Moz. Accessed July 18, 2020. https://moz.com/learn/seo/what-is-seo.
“Who Are the Top 10 Grocers in the United States.” Food Industry, 2019. Accessed July 20, 2020.
https://www.foodindustry.com/articles/top-10-grocers-in-the-united-states-2019.