Municipal Broadband Restrictions and Mothers’ Labor Outcomes
Summary
NBER Working Paper No. 32257, Restrictions on Municipal Broadband Provision and Mothers' Labor Market Outcomes: Evidence from the COVID-19 Pandemic, by Saket S. Hegde and Jessica Van Parys, dated March 2024 and revised June 2026. The paper examines the 17 states that legally restricted local governments from building or providing broadband before the pandemic. Using Federal Communications Commission data, it reports that broadband subscriptions grew between 4.4% and 8.4% slower in restriction states after the pandemic. Using Current Population Survey data from 2018–2022, it reports that mothers' post-pandemic employment decreased by 4.7% in restriction states, while their share of time working from home fell 49% and commute time rose 27%. The paper ends with regression tables for mothers, women without children and married men with children.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Full text
NBER WORKING PAPER SERIES
RESTRICTIONS ON MUNICIPAL BROADBAND PROVISION AND
MOTHERS’ LABOR MARKET OUTCOMES:
EVIDENCE FROM THE COVID-19 PANDEMIC
Saket S. Hegde
Jessica Van Parys
Working Paper 32257
http://www.nber.org/papers/w32257
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
March 2024, Revised June 2026
Corresponding Author: Jessica Van Parys. The authors have nothing to disclose. The views
expressed herein are those of the authors and do not necessarily reflect the views of the National
Bureau of Economic Research or the Federal Reserve System.
NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.
© 2024 by Saket S. Hegde and Jessica Van Parys. All rights reserved. Short sections of text, not to
exceed two paragraphs, may be quoted without explicit permission provided that full credit,
including © notice, is given to the source.
Restrictions on Municipal Broadband Provision and Mothers’ Labor Market Outcomes: Evidence
from the COVID-19 Pandemic
Saket S. Hegde and Jessica Van Parys
NBER Working Paper No. 32257
March 2024, Revised June 2026
JEL No. J2, K2, L86
ABSTRACT
Prior to the COVID–19 pandemic, 17 states had legally restricted local governments from building
their own broadband infrastructure and/or providing broadband internet to their communities. We
use Federal Communications Commission (FCC) data to show that broadband subscribership grew
slower in states with these restrictions compared to states without restrictions post-2020,
particularly in counties with less affordable plans and fewer broadband providers. Then we examine
how municipal restrictions affected the labor supply of married women with children following the
sharp increase in telework after the pandemic. We employ the Current Population Survey (CPS)
from 2018–2022 to estimate event studies and difference–in–differences models. Results show that
mothers’ post-pandemic employment decreased by 4.7% in restriction states relative to non-
restriction states additionally, mothers in restriction states left the labor force and shifted to
caregiving. Among mothers continuing to work, the share of time spent working from home
decreased by 49%, while commute time increased 27%, in states with broadband restrictions versus
states without. The results suggest that mothers were less likely to enter the workforce in states
where remote work opportunities were limited by restrictions on broadband provision.
Saket S. Hegde
Federal Reserve Bank of Philadelphia
saket.hegde@phil.frb.org
Jessica Van Parys
Hunter College of the City University of New York
Department of Economics
and NBER
jv947@hunter.cuny.edu
Introduction
The internet is an important medium for the exchange of information. Most Americans have access
to the internet, but many still lack access to a high–speed broadband connection at home. In 2017, 21.3
million Americans lacked broadband connectivity (FCC 2019). More recent surveys find that over 25% of
Americans do not have home broadband internet (Pew Charitable Trusts 2021). Pew also finds that the lack
of access is twice as high for those without any college education and is even higher in rural and tribal areas.
Non–broadband users say they do not subscribe because service is too expensive (21%), because they use
a smartphone to access the internet (23%), or because service is unavailable or speeds are too slow (7%)
(Anderson 2019). Meanwhile, the benefits of broadband diffusion are significant. For example, it increases
income (Whitacre, Gallardo, and Strover 2014), improves health outcomes (Van Parys and Brown 2024),
and increases employment (Atasoy 2013), partly through improved labor market matching (Bhuller, Kostol,
and Vigtel 2020).
Despite the benefits of broadband access, state–level regulations often block public communications
initiatives and public–private broadband partnerships. As of 2020, 17 states had regulations that explicitly
banned the establishment of municipal broadband (Chamberlain 2021). Most of those 17 states also previ-
ously had roadblocks making it onerous for telephone and electric cooperatives to provide broadband internet
(Whitacre and Gallardo 2020; Morton 2021; Pew Charitable Trusts 2021).1 Yet, cooperative and municipal
broadband providers are prevalent in states without restrictions. Maps from the Institute for Local Self Re-
liance (ILSR) show that nearly 60% of U.S. counties have at least one community (municipal or cooperative)
broadband provider (Parker 2024).
Cooperative and municipal broadband networks operate differently from private internet service providers
(ISPs). For example, as nonprofit membership corporations, rural electric and telephone cooperatives are
exempted from paying taxes under section 501(c)(12) of the Internal Revenue Code (IRC). Similarly, munic-
ipal providers are often oriented towards the local community and they lack a profit motive, which may also
incentivize them to maximize broadband access and pass on significant cost savings to subscribers. Case
studies find that community–owned networks offer prices that are 3–50% lower than private ISPs. Private
1
11 of the 17 restriction states either had significant roadblocks for co-ops to provide broadband, or they had easement ambiguities,
which made it risky for co-ops to provide broadband. However, many of these ambiguities were eliminated by 2019. We explore
this in the Robustness Section.
1
ISPs often offer low promotional rates that later rise sharply (Talbot, Hessekiel, and Kehl 2018). Therefore,
restricting the entry of public or non–profit firms may have consequences for access and prices in broadband
markets. In a country–wide analysis, Whitacre and Gallardo (2020) find that states’ restrictions on munici-
pal broadband provision had reduced the number of people with broadband access by approximately 4.5%
between 2012 and 2018.
Two recent events brought state–level restrictions on municipal broadband to the forefront of public
discourse. First, beginning in March 2020, the COVID–19 pandemic forced millions of people to stay at
home for an extended period of time. The “COVID–19 shut–downs” elevated the importance of having a
high–quality broadband connection for employment, education, entertainment, and social connection. The
increase in remote work persisted well after the end of COVID shut-downs in July 2021 (Kaiser Family
Foundation 2021). Therefore, state–level restrictions on municipal broadband provision may have been es-
pecially limiting after the COVID–19 pandemic when many Americans needed to have high–speed internet
at home to find and perform work. Second, in November 2021, a bipartisan infrastructure bill was signed
into law. The bill included the largest-ever federal investment in broadband. The bill included $42.45 billion
for the Broadband Equity, Access and Deployment (BEAD) program, which gave money to states to fund
broadband deployments in unserved and underserved areas.2 So while some states blocked local govern-
ments from building their own broadband networks, the federal government encouraged states to suspend
their restrictions to spend billions in new grant dollars (NTIA 2022). Thus, there has been an increasing ten-
sion between state–level policies that restrict municipal broadband provision and federal policies that seek
to expand access.
Therefore, it is important to examine whether policies that restrict municipalities from providing
broadband affect peoples’ labor market outcomes. In this paper, we estimate how municipal broadband
restrictions affected the labor force participation of married mothers in the aftermath of the COVID–19 pan-
demic. We link data on state–level broadband restrictions from BroadbandNow to county–level data on
broadband subscribership from the Federal Communications Commission (FCC), and to individual–level
Current Population Survey (CPS) data on labor market outcomes measured between Q1:2018 and Q4:2022.
We focus on municipal restrictions in place prior to the COVID–19 pandemic. Since all municipal broadband
restrictions in our data had been implemented prior to 2018, the empirical strategy exploits the exogeneity
2
As of mid-2025, BEAD funds have still not been spent at the local level. Therefore, the BEAD program does not confound our
estimates of the impacts of broadband restrictions around the COVID-19 pandemic.
2
of the COVID–19 shock. The COVID–19 shock normalized peoples’ use of broadband to work from home,
and arguably increased demand for in-home broadband.
We first use FCC data on the share of broadband subscriptions per county (measured every 6 months)
to show that restriction states had lower broadband adoption and ISP competition leading up to the pandemic.
We also estimate event studies and difference–in–differences models that show broadband subscriptions grew
between 4.4% (at 100Mbps+) and 8.4% (at 25Mbps+) slower in states with municipal restrictions after the
pandemic compared to states without restrictions. We then show that the relative decreases in broadband
subscribership observed after the pandemic in restriction states were likely driven by mechanisms such as
reduced ISP competition, fewer plans, and higher prices.
In our labor supply results, we focus on married women with children because (i) their labor force
participation is responsive to new technology (Greenwood, Seshadri, and Yorukoglu 2005; Nieto 2023), and
(ii) they have a preference for working-from-home, on average (Angrisani, Burke, and Perez-Arce 2023).
Prior work has found that even during periods without widespread telework (i.e., before COVID), broadband
access had direct and substantial effects on maternal labor supply (Dettling 2017; Viollaz and Winkler 2020).3
We also examine the effects of municipal restrictions on women without children and married men with
children.
Trends in maternal labor force recovery after the COVID pandemic also motivate our focus on this
group. Initially, the labor supply of women with children was disproportionately negatively impacted during
the early stages of the pandemic (Albanesi and Kim 2021; Couch, Fairlie, and Xu 2022). However, these
effects diminished over time since female employment prior to the pandemic was more concentrated in jobs
that could be done remotely (Cowan 2023). Recent work by the Hamilton Project at the Brookings Institution
shows that prime-age women with children played an outsized role in the overall post-pandemic rebound
in labor force participation. During the post-COVID recovery, prime-age female labor force participation
reached a record high, driven by a rebound in labor force participation for married women with children
under five (Bauer and Yu Wang 2023). As of mid-2023, full days worked at home continued to account for
28% of paid workdays among Americans 20-64 years old (Barrero, Bloom, and Davis 2023), with married
mothers still more likely to report that they teleworked once a week compared with other workers (Bauer
and Yu Wang 2023).
3
We discuss this research and the role of telework in the section titled, “Telework and Maternal Labor Force Participation.”
3
We use individual-level CPS data to estimate event studies and difference-in-differences models to
show how labor market outcomes differed for married mothers living in states with municipal restrictions
compared to their counterparts in states without restrictions before and after the COVID–19 pandemic. Our
results show that the trends in maternal labor force participation diverged sharply across states with and
without municipal broadband restrictions by Q3-2021. In states with restrictions, mothers reduced their labor
force participation (LFP) and employment by 2.68 percentage points and 3.1 percentage points respectively
(or 3.6% and 4.8%), compared with mothers in states without broadband restrictions.4 For comparison,
Jones and Wilcher (2024) find that paid family leave reduces maternal labor force detachment by more than
5 percentage points in the year of birth. Hence our estimated effects of broadband restrictions on maternal
labor force detachment post-pandemic are about half as large as the estimated effects from paid family leave.5
In contrast, broadband restrictions had no effects on the LFP or employment of women without children or
married men with children – groups with lower labor supply elasticity (Hansen, Sabia, and Schaller 2024;
Heggeness, Suri et al. 2021). Among mothers who continued to work, those living in restriction states
experienced a 49% decrease in the share of time spent working from home and a 27% increase in commute
time. We conclude that state-level restrictions on municipal broadband had important economic implications
for families in states with restrictions after the COVID-19 pandemic.
We offer two contributions to the literature. First, we show that states with restrictions on municipal
broadband experienced slower growth in broadband adoption after the COVID-19 pandemic compared to
states without such restrictions. We also find that growth in broadband adoption was especially slow in coun-
ties in restriction states with non-competitive ISP markets and/or less affordable broadband plans leading up
to the pandemic. Our results are consistent with Carvalho, Hagerman, and Whitacre (2022), who show that
low broadband adoption rates prevented some counties from recovering their COVID-19-related employ-
ment losses. We build on their work by showing why adoption rates were lower in certain counties. Second,
we are the first to show that states without municipal restrictions on broadband experienced larger increases
in maternal labor force attachment. Our results suggest that these “non-restriction states” had more afford-
able high-speed plans, which helped mothers balance work and caregiving in the post-COVID era. Given the
recent discourse around return-to-office policies and the labor-force-detachment of working mothers (Lutz
4
Throughout the paper, we refer to married women ages 25-54 who were living with their children ages 0-17 years old as “married
mothers” or simply “mothers.”
5
We use the estimate for maternal labor force participation of -2.68 percentage points taken from Table 8, panel A, column 1.
Our estimated effect of -2.68 pp is approximately 54% as large as the -5 pp paid family leave effect ( 2.68
5
= 0.536).
4
2025), our paper shows how policies that reduce workplace flexibility can have consequences for working
mothers.
State–Level Restrictions on Municipal Broadband
We rely on the 2020 annual report on the state of municipal broadband from BroadbandNow (Cham-
berlain 2021) – which details restrictions in 2019 – to assess the degree to which state policies restrict the
provision of municipal broadband.6 BroadbandNow describes state-level regulatory barriers that categori-
cally ban the establishment and/or operation of broadband infrastructure by local government or non-profits.
Outright bans mean that counties, towns, cities, municipal governments, and sometimes electric and tele-
phone cooperatives cannot provide broadband to the general public. The 17 states that have explicit bans
on municipal broadband during our sample period (2018-2022) are Alabama, Florida, Louisiana, Montana,
Michigan, Missouri, Minnesota, Nebraska, Nevada, North Carolina, Pennsylvania, South Carolina, Ten-
nessee, Texas, Utah, Virginia, and Wisconsin (Chamberlain 2021). For the rest of the paper, we refer to such
policies as “municipal broadband restrictions.” In other research, such restrictions are sometimes called “pre-
emption barriers,” as in state law “pre-emptively blocks” a local government from providing broadband. The
reasoning behind such restrictions is explained in the text of the proposed 2021 CONNECT Act in the House
Commerce Committee of the U.S. Congress (Republicans, Energy and Commerce Committee 2021). The
bill would create a nationwide ban on establishing and operating municipal networks. The bill’s sponsors say
the ban would promote competition by limiting government-run broadband networks throughout the coun-
try and by encouraging private investment. The spirit of bills that restrict municipal broadband suggest that
sponsors and supporters believe local governments are inefficient at providing broadband, and that public
funds should not be used in an effort to subsidize such an endeavor.
Three other states (Iowa, Oregon, and Wyoming) had implemented other roadblocks, such as requir-
ing the municipal broadband service to be self-sustaining, price matching with incumbent internet service
providers, and/or phantom cost requirements (Chamberlain 2021). We code these states as control states in
our analysis.7 To better understand this minority of 3 states with other roadblocks but not explicit bans, we
6
For more current information on restrictions, we direct readers to the most recent BroadbandNow report on municipal broadband
(Cooper 2024).
7
Results are qualitatively similar if we code them as treatment states or drop them.
5
elaborate on the state of Iowa below.
In Iowa, there are no statutes that specifically ban municipalities from offering broadband services
to residents, but state law requires that all new public utilities must be approved by voter referendum of
51%. If the referendum fails, the municipality cannot hold another referendum vote on the same proposal
or a similar proposal for at least four years. If a municipality wishes to use bonds to finance a public broad-
band network, the measure needs to obtain 60% approval in a referendum. Municipalities are also prevented
from using general fund revenues to support a broadband network, and they must complete a detailed an-
nual audit, subject to open meeting requirements, meaning that aspects of the audit which might contain
commercially sensitive information must be made public. Two Iowa bills that passed in 2004 erected most
of the barriers discussed above (Chamberlain 2021). Despite this, some cities in Iowa have been able to
establish municipal broadband. An example is the city of Waterloo, Iowa (City of Waterloo 2023). Waterloo
voters approved a $20 million General Obligation Bond in a special election in September 2022 to fund the
network. Construction of the network started in 2023 and the city said it would take 3 years to complete.
The population covered by community (municipal or cooperative) broadband is not insignificant –
approximately 20% of the population in non-restriction states lives in a county with municipal broadband in
2023. Previous studies have noted that states with municipal restrictions also tend to impose restrictions on
cooperative broadband (Whitacre and Gallardo 2020; Pew 2019; Trostle et al. 2019; Morton 2021). Among
these, we were able to identify at least 8 states that also restricted cooperatives from participating in the
broadband market as of 2018 - these include Alabama, Louisiana, Michigan, Minnesota, North Carolina,
Pennsylvania, South Carolina and Virginia (Pew Charitable Trusts 2021). Updated maps from ILSR suggest
that nearly 60% of U.S. counties have access to broadband via either municipal broadband or telephone and
electric cooperatives (Parker 2024).
Cooperatives have also played a vital role in state broadband expansion alongside municipal gov-
ernments once municipal restrictions are removed – even if there were never restrictions on cooperatives.
As an example, we describe the evolution of broadband legislation and the interplay between co-ops and
municipalities in providing broadband in the state of Arkansas.
The (Arkansas) Telecommunications Regulatory Reform Act of 1997 prohibited any government en-
tity in the state from offering basic exchange services (Gonzalez 2012). In legal circles, “basic exchange”
means “telephone services.” Therefore, no town in Arkansas could create its own telephone company that
6
offered the traditional concept of telephone service, as defined in statute. In 2011, Arkansas expanded this
prohibition to include data, broadband, video, and wireless. Arkansas towns were prohibited from offering
broadband services to the public (unless they owned community electric utilities or cable television sys-
tems) (Gonzalez 2012). As of 2018, this prohibition remained in place in Arkansas. FCC Form 477 for that
year shows that Arkansas ranked second from last among all U.S. states for broadband access (FCC 2021;
Arkansas State Broadband Manager 2021).8
In 2018, the Arkansas Republican Woman’s Legislative Caucus sought to overturn state barriers to
municipal broadband by introducing Senate Bill SB150 (Gonzalez 2019). However, when ultimately passed
in February 2019, SB150 said government entities are only able to deploy broadband in unserved areas and
only if they apply for and received grant funding. Community-centric providers often have needs that are
too large for foundation grants (Vo and Miller 2025). Initial deployment costs can be high, especially in
unserved areas. Often, broadband expansion is funded by tax revenue, bonds, or in some cases, the members
of a cooperative pitch in to fund the expansion themselves. Since SB150 did not allow grant dollars to be
supplemented with revenue from areas considered served, it effectively left the ban in place.
In February 2021, Arkansas passed SB74. SB74 allows government entities “to acquire, construct,
furnish, or equip facilities for the provision of voice services, data services, broadband services, video ser-
vices, or wireless telecommunications services” so long as they “partner, contract, or otherwise affiliate with
an entity that is experienced in the operation of the facilities,” as well as conduct due diligence, and provide
ten-days’ notice and hold a public hearing (Marcattilio 2021).
Notably, laws prior to 2021 did not prevent cooperatives from participating in Arkansas prior to that.
However, the expansion of broadband by electric co-ops was likely accelerated by the 2021 repeal of munici-
pal restrictions (Bode 2024). Since electric utilities in Arkansas were already authorized to deploy broadband
(Arkansas General Assembly 2007), many local government entities chose to partner with them as the 2021
reform allowed government entities in Arkansas to provide broadband if they “partner with experienced op-
erators.” Some examples include the partnership between the city of Cabot and First Electric Cooperative
(Zhone 2024). Similarly, the Ozarks Electric Cooperative has expanded residential broadband access after
initially partnering with local school districts (Cash 2024a).
8
The measure we used to determine state-level broadband access was the percentage of population with access to 25/3 Mbps (down-
load/upload) speeds.
7
Many cities in Arkansas also operated municipally owned utilities providing electric, water, and
wastewater services for many decades before the reform. The 2021 legislation enabled these municipally
owned utilities to begin providing broadband; many leveraged existing community ties and infrastructure to
expand broadband without coop support. For example, the cities of Conway and Clarksville have recently
expanded broadband through municipally owned utilities (Marcattilio 2021) and did so independently of
cooperatives.
Following SB74, the state of Arkansas saw rapid growth in community broadband. As of November
2024, 17 cooperative broadband providers in Arkansas said they had constructed 40,000 miles of fiber and
connected 1.4 million Arkansans with broadband capability, representing over one-third of the state’s pop-
ulation (Cash 2024b). 80% of this increase was self-funded (by members of the cooperatives themselves).
Only 20% of the expansion was grant-funded.
Using the example of Arkansas, we see how barriers to municipal broadband developed (1997-2011)
and were eventually removed (2018-2021). This trajectory is typical in many states, although the removal of
barriers remains less common. As detailed above, 17 states still had barriers as of 2022 (end of our sample
period), while only 3 states rolled them back between 2018-2022. These states include Arkansas (2021),
California (2018), and Washington (2021).9
Telework and Maternal Labor Force Participation
Research on the COVID-19 pandemic has shown that the pandemic had a disproportionate impact
on the labor force participation of married women (Alon et al. 2020; Faberman, Mueller, and Şahin 2022;
Schroeter, Lalive, and Karunanethy 2023; Cheng et al. 2020). For example, Garcia and Cowan (2022) find
that school closures during the 2020–21 academic year led to a decrease in the probability of women being
at work. Importantly, they find workers in jobs with high telework potential did not experience reductions
in work hours when schools were closed. Hansen, Sabia, and Schaller (2024) study the effect of K–12
reopenings on the employment outcomes for married women with school–age children. They find K–12
reopenings were associated with a 3.3 percentage point increase in employment and a 0.76 increase in (con-
9
Arkansas removed barriers in February 2021. A bill removing all restrictions on public broadband in Washington was signed into
law by Governor Jay Inslee in May 2021. California removed its roadblocks (before COVID) in 2018. Although our results are
not sensitive to including or excluding California, we drop the state as it removed roadblocks to municipal broadband in the year
our sample period begins (2018).
8
ditional) weekly work hours among married women with school–aged children. Hansen, Sabia, and Schaller
(2024) estimate reopenings led to a 3.9 percentage point reduction in remote work among married women
with children, with larger reductions among college–educated mothers. This suggests that broadband access
and remote work during the COVID–19 pandemic helped women balance the competing demands of family
and career.
Research prior to the COVID–19 pandemic has also shown that access to broadband increases the
labor force participation of married women. Dettling (2017) finds that broadband use increases labor force
participation for married women in the U.S. by 4.1 percentage points but has no impacts on single women or
men. She finds that the increase in participation is explained by women using the internet for telework and
time–saving tasks related to home production. Outside the U.S. context, Viollaz and Winkler (2020) also
find that a 1 percentage point increase in internet adoption in Jordan between 2010–2016 increased female
labor force participation by about 0.7 percentage points.
We contribute to the existing literature on how broadband affects maternal labor supply by focusing
on the importance of state–level municipal broadband restrictions during a period when work–from–home
increased in a plausibly exogenous manner for a broad swath of workers. We focus on municipal restrictions
because they create barriers to broadband access despite other state–level policies aimed at increasing broad-
band access (Whitacre and Gallardo 2020). Municipal restrictions have also been shown to affect broadband
access more than other state–level policies such as funding subsidies to private ISPs or opening a broadband
office (Whitacre and Gallardo 2020).
Data
We utilize data from six sources. First, we determine whether states had broadband restrictions in place
as of January 2018 (and through 2022) using reports from BroadbandNow. Second, we analyze broadband
subscribership, including the role of broadband prices and ISP competition using data from the FCC Form
477. Third, to estimate labor supply effects, we employ individual labor force outcomes from the monthly
Current Population Survey (CPS) between January 2018 and December 2022. Fourth, we complement the
labor-force-related outcomes in the CPS with data on time spent working from home from the American
Time Use Survey (ATUS) and the CPS Internet Use Supplement. Fifth, to construct control variables for our
9
analysis, we collect data from the Centers for Disease Control (CDC) on COVID-19 cases and COVID-19
deaths, and finally, we employ data from the school closures database compiled by Parolin and Lee (2022).
Restrictions on Municipal Broadband
Data on state-level broadband policies come from BroadbandNow (Chamberlain 2021; Cooper 2024).
We focus on whether states had outright bans or severe roadblocks on the provision of municipal broadband
at the start of our sample period. These regulations include outright bans on the establishment or operation
of municipal broadband infrastructure or bureaucratic obstacles making it practically infeasible to create a
citywide network. Between 2018 and 2022, 17 states met this criteria. Figure 1 shows a map of the states with
and without municipal restrictions in place as of 2018. The states with broadband restrictions are generally
spread out geographically. With the exception of New England, we find that all regions of the country feature
states with and without restrictions on municipal broadband.
Figure 2 shows the years in which states first implemented restrictions on municipal broadband. Most
restrictions went into place between 1999 and 2005, although two states passed restrictions between 2006-
2011. We find that states rarely revoke their restrictions on municipal broadband once they are put into place.
Indeed, only Arkansas, California and Washington reversed their policy on municipal restrictions between
January 2018 and December 2022. For this reason, we drop Arkansas, California and Washington from our
analysis. Finally, we focus our analysis on the continental U.S. and drop Alaska and Hawaii.
Broadband Adoption/Subscribership and the Number of Broadband Providers
We use the county-level data on Internet Access Services from FCC Form 477 to measure broad-
band adoption (i.e., subscribership) before and after the COVID-19 pandemic. The FCC collects broadband
subscribership data through its bi-annual Form 477 data collection (Federal Communications Commission
2025a). There are 10 data points, one in June and one in December for each of the 5 years in our sample
period (2018-2022). Our primary variables of interest are the number of connections at or exceeding 25Mbps
download per 1,000 households in a county, and at or exceeding 100Mbps download per 1,000 households
in a county. We focus on the 25Mbps and 100Mbps threshold because 25+/3+ 10 was the FCC standard since
10
25+/3+ refers to download speeds of 25Mbps or higher and upload speeds of 3Mbps or higher
10
2015, and 100+/20+ Mbps is the current standard (in effect since 2024). Further, upload speeds of 4Mbps
are often required for common modalities of telework such as video conferencing especially when multiple
household members are streaming video or on calls at the same time (Microsoft 2024; University of Chicago,
Data Science Institute 2022).
FCC measures of potential broadband access or availability lack consistency during our sample period
and overstate access. This includes the older 477 data, which the FCC chair notes overstates access (Rosen-
worcel, Jessica 2023) and the new broadband maps, which received millions of challenges (FCC 2024; ILSR
2023). In contrast, county-level subscriptions provide a more accurate measure of actual broadband use. For
example, the FCC says in their Section 706 Report for 2024 (FCC 2024) that “if the cost of broadband service
is higher than millions of people can afford, service cannot be said to be available.” Further, they state “low
adoption rates in areas where broadband is technically deployed and available, for example, may evidence
that other factors are in play that make it effectively unavailable for some portion of the population.” Another
advantage of the subscriptions/adoption data is that it is consistently coded during our sample period (June
2018-December 2022). Previous work also finds that the effect of broadband on economic growth in the
United States is driven by adoption rather than access - this helps validate the ordinal measures we employ
(Whitacre, Gallardo, and Strover 2014).
Despite this, the FCC subscriptions data have some drawbacks. Granular data on the exact number
of subscriptions in each county are considered confidential data at the FCC, and are not available to the
public.11 Instead of the exact number of subscriptions, the FCC provides external researchers with the num-
ber of connections per 1,000 households in each county in terms of ordinal tier values where 0 represents 0
connections, 1 represents between 1 and 200 connections, 2 represents between 200 and 400 connections, 3
represents between 400 and 600 connections, 4 represents between 600 and 800 connections, and 5 repre-
sents above 800 connections (Federal Communications Commission 2025a). Another drawback is that the
FCC does not consider upload speeds when assigning ordinal tier values. For example, the tier values for
100Mbps+ are determined solely by the download speed of the connections.
Therefore, we supplement the subscribership data with measures of the percentage of the county-level
population with access to different numbers of residential ISPs, using the biannual (June, December) FCC
Area Tables between June 2018 and the last period available, June 2021. Unlike adoption, this is a different
11
The granular data are available internally to FCC researchers – see LoPiccalo (2022) for example.
11
measure that captures the percentage of the county-level population with access to different numbers of
residential ISPs through the FCC Area Tables. For the percentage of county-level population with access to
different numbers of residential ISPs, we use “acfo” for 10+/1+ speeds and only “cfo” for 25+/3+ speeds or
higher.12 The use of acfo follows the FCC’s own definition for fixed wired or wireline connected broadband;
hence we exclude satellite and fixed wireless. We additionally exclude ADSL from 25+/3+ speeds or higher
because the theoretical maximum download speed for the ADSL2+ standard is 24Mbps, and in practice,
ADSL speeds are lower ITU (2003). As an illustrative example for a given county and biannual period, FCC
Area Tables report that 10% of a county’s population had access to 0 residential broadband providers, 40%
had access to 1 provider, 20% had access to 2 providers and the remaining 30% of the county population
had access to 3 or more providers.13 Note that these data are available at different speeds. In contrast to
the subscribership/adoption data, competition measures incorporate upload speeds as well; for example, the
percentage of the county population covered at 25+/3+ speeds. We use this data to show (1) how states
with municipal restrictions differ from states without restrictions in terms of the percent of the population
with access to different numbers of ISPs, and (2) how ISP competition mediates the effects of municipal
restrictions on broadband subscribership.
Tables 1 and 2 include county-level summary statistics for the pre-COVID period (2018–19) by
whether the state had broadband restrictions (column 2) or not (column 1). Before the pandemic, coun-
ties in restriction states had lower levels of adoption (as measured by 60% or higher subscribership) across
all 3 speed tiers. These differences are significant beyond the 5% level for 100Mbps+ (row 1) and close to
the 10% level for 10Mbps+ (row 3, p = 0.11). Counties in restriction states also had a higher percentage of
the population with 0 providers at every speed tier (100/10 Mbps+, 25/3 Mbps+ and 10/1 Mbps+) during the
pre-period (June 2018 to December 2019), as seen in rows 4 through 6. These differences are statistically
significant beyond the 1% level. Finally, counties in restriction states have a lower percentage of the popu-
lation with access to 2 or more providers where the differences are significant beyond the 5% level for 25/3
Mbps+ (second from last row) and 10/1 Mbps+ (last row). These results indicate that pre-COVID levels of
broadband penetration, as well as ISP competition, were lower in restriction states.
12
a=ADSL, c=Cable, f=fiber, o=other
13
These data are available under Area Summary at https://broadband477map.fcc.gov/#/data-download.
12
Labor Market Outcomes
Data on labor market outcomes and respondent characteristics come from the monthly Current Pop-
ulation Survey (CPS) (January 2018– December 2022). The CPS provides information on labor force par-
ticipation, (conditional) hours worked, and employment. For employment, we consider whether a person
is employed (Employed, Any) and whether an employed person is at work the previous week (Employed,
At Work).14 Whether an individual is at work is defined as doing any work for pay or profit or working at
least fifteen hours without pay in a family business or farm in the previous week. This excludes individuals
who are employed but currently absent from work due to illness, childcare problems, or family/personal
obligations. For hours worked, we examine effects on the number of hours the individual worked last week.
These work hours are conditional on being employed.
To examine the channels that affect labor force attachment, we also include measures of non- or under-
employment related to childcare. We use CPS variables to identify respondents who report that they were
“taking care of house and family” while not in the labor force (NILF) or working part time last week due
to “child care problems.” These variables have been used in previous COVID-19-related research to show
that women with school-aged children took on a disproportionate share of childcare responsibilities while
schools were closed (i.e., the “COVID motherhood penalty”) (Garcia and Cowan 2022; Hansen, Sabia, and
Schaller 2024).
The CPS records the demographic characteristics of respondents, such as age, gender, education, race,
marital status, number of children, age of children, metro status, state-of-residence, and survey month. We
characterize the respondent’s racial/ethnic group as either White Non-Hispanic, Black Non-Hispanic, other
Non-Hispanic, or Hispanic. We characterize the respondent’s educational attainment as either less than a
high school degree, high school degree, some college, bachelor’s degree, or advanced degree.
Our empirical approach compares labor market outcomes for people living in states with municipal
broadband restrictions to people living in states without restrictions before and after the COVID-19 pan-
demic. Table 3 reports summary statistics for our CPS sample in restriction and non-restriction states during
the pre-pandemic period (2018-19). Our sample includes prime-age adults (25-54 years old) who were not
14
During the early stages of the pandemic, some workers who were not at work during the entire reference week were misclassified
as employed but absent from work by the BLS. Analyses of the underlying data suggest that this group included some workers
affected by the pandemic who should have been classified as unemployed on temporary layoff (U.S. Bureau of Labor Statistics
2022). For this reason, we use both “employed” and “employed at work” as our outcomes.
13
employed in public administration or education.15 Respondents in both types of states had similar levels
of labor force attachment in the pre-pandemic period. Prime-age respondents in states without restrictions
were more likely to have bachelor’s or advanced degrees. They were also less likely to be Black or Hispanic,
more likely to be White, and more likely to live in a central city. Therefore, we control for education and
individual-level demographic characteristics in our difference-in-differences design.
The final row in Table 3 reports the mean (and standard deviation) for the proportion of teleworkable
jobs in restriction and non-restriction states. We classify jobs as teleworkable based on occupation using
the methodology developed by Dingel and Neiman (2020). During the pre-pandemic period (when telework
was less common), we find no differences between restriction and non-restriction states in the proportion of
prime-age individuals in teleworkable jobs.
Much of our analysis focuses on married women with children. This group of workers might have
benefited the most from using broadband to work from home during the pandemic. Therefore, Appendix
Table A1 shows summary statistics for married women with children from 2018-19 in states with and without
broadband restrictions. As with our full sample, we focus on married women with children aged 25-54 years
old (prime-age) from the CPS. Children are coded as present if a respondent has at least one child in the
household between the ages of 0 and 17. In states without municipal restrictions, married women with
children have slightly higher labor force attachment prior to the pandemic. They are also more likely to go
to college, hold an advanced degree, less likely to be Black or Hispanic, and more likely to be White. As in
the full sample of prime-age workers, married women with children are no more likely to be in teleworkable
occupations across restriction or non-restriction states.
Time-Use and Remote Work
Data on working from home comes from the American Time Use Survey (ATUS). The ATUS is a
24-hour time diary, where respondents report the activities they were doing between 4:00 am of the first
day and 4:00 am of the following day. Respondents are randomly sampled from individuals who completed
the Current Population Survey (CPS). They respond to the ATUS between two and five months after their
15
Consistent with previous work, we exclude workers in public administration and education from our analysis (Althoff et al.
2022). Employment in government and education was hit hard due to the early and direct effects of the pandemic (Dylan
Maloney 2019) (e.g., from school closures). We exclude workers with ind1990 codes 900,932 (public administration) and
842,850,851,852,860,862 (education).
14
final CPS interview. We restrict our sample to prime-age workers (25-54 years old) and exclude time-use on
weekends and holidays.
We follow Cowan, Jones, and Swigert (2024) and aggregate ATUS activity codes into 10 main cat-
egories. Our primary outcome is work-from-home (WFH), defined as work activity performed at one’s
residence. Additionally, we examine total time spent working, share of work time from home, and time
spent commuting. Similar to the CPS, for the ATUS, we follow Cowan (2024) and exclude all of 2020 and
January-July 2021 to avoid the confounding effects of the COVID-19 pandemic and associated policies on
time use.
Control Variables Related to the COVID-19 Pandemic
The COVID-19 pandemic affected labor supply and labor demand in non-random ways across states
(Gupta, Simon, and Wing 2020). We include several control variables in our analysis to account for these
differences. All of our models control for school closures, COVID-19 cases per capita, and COVID-19 deaths
per-capita. These measures vary by county-quarter. We also include COVID-19 controls to assuage concerns
that our results might be driven by state differences in stay-at-home orders and restaurant closures. Stay-
at-home orders were highly correlated with COVID cases, deaths, and school closures – and we argue our
measures better control for such orders given issues such as non-compliance and differences in the duration
of such orders.
We employ the strategy used by Garcia and Cowan (2022) for measuring school closures. They use the
school closures database from Parolin and Lee (2022). These data are aggregated and anonymized mobile
phone records from Safegraph, which track year-over-year changes in the number of visitors to each school
or childcare facility relative to the pre-pandemic baseline in 2019. We consider institutions closed if there
is at least a 50 percent year-over-year decline in the number of in-person visits. We use the share of closed
institutions in each county in each quarter between 2020 and 2022 as the measure of the extent of school
closures in the analysis; thus, the variable is a continuous measure between zero and one. The variable equals
zero for all counties and quarters before January 2020 and after May 2022 (the last month with available data).
The cumulative COVID-19 case rate and the cumulative COVID-19 death rate come from the CDC
(CDC 2022), and the rates are measured in each county-quarter between 2020 and 2022. Prior to Q1:2020,
15
these measures equal 0 in all counties and quarters. We use these rates to control for changes in the severity
of the pandemic, and to proxy for differences in labor demand and supply, across counties over time. For
instance, research shows that COVID-19 cases per capita affected the number of unemployment claims filed
across states in the early pandemic period (Lozano Rojas et al. 2020). In addition, Goda and Soltas (2023)
show that workers with severe COVID-19 infections reduced their labor supply one year later. Coombs
et al. (2022) show that the early withdrawal of enhanced unemployment benefits in 22 states temporarily
increased weekly employment among Unemployment Insurance (UI) beneficiaries. However, enhanced
unemployment benefits ended in all states by September 2021. Therefore, the post-period in much of our
analysis begins in August 2021, after enhanced unemployment benefits expired.
The CPS obscures county identifiers (57% of our observations) so as to not identify persons in smaller
counties. Therefore, including county fixed effects will drop observations for respondents living in small
counties. To overcome this challenge, similar to Hansen, Sabia, and Schaller (2024) we match geographic
areas as follows – if a county identifier is available we assign it as an area identifier for the CPS (or ATUS)
individual. If a county is unavailable but the metro area FIPS is identified (typically covering multiple
counties) we use that instead as the area identifier.16 Finally, if neither county nor metro area is identified,
we combine individuals into “unincorporated geographic areas” within each state. We then include these
area fixed effects in order to control for area-level initial broadband availability and any fixed characteristics
in areas over time.
Our COVID-19 controls – cumulative case and death rates are similarly aggregated. If the CPS iden-
tifies an individual’s county, we match it with cumulative COVID case and death rates for that county and
quarter. For state-wide averages, we calculate within-state averages across counties other than the larger
counties for which the CPS reports the county identifier, rather than overall statewide averages of these data,
since this better represents cases and deaths for counties for which the county identifiers are not available.
Methods
16
Definitions of metropolitan areas along with their codes change over time. However, consistent coding is used between May 2014
and April 2025.
16
Broadband Adoption
For the effects of municipal broadband restrictions on broadband adoption, we estimate results using
a biannual county-panel constructed from FCC Form 477 County Data on Internet Access Services between
June 2018 and December 2022. Our model compares broadband adoption for counties in states with re-
strictions on municipal broadband to adoption in counties in states without restrictions before and after the
COVID-19 pandemic. We estimate our main effects on broadband subscriptions using a standard difference-
in-differences estimator:
yist = β(M unis × P ostt ) + ΩXit + δBBi + λt + ϵist (1)
First, we estimate Equation 1 as an ordered logistic model where yist is an ordinal (0, 1 or 2) biannual
value that we consolidate from the ordinal FCC measure for the number of connections from the FCC County
Broadband Access (corresponding to 0 to 40%, 40 to 60% and over 60% subscribership, respectively) at
various speeds in a county i = 1, . . . , N in state s at time t, where t indicates a biannual time period
between June 2018 and December 2022 (10 periods total). Second, we estimate linear probability models
that use binary outcomes for broadband subscribership of 20%+ at 100Mbps+ speeds, 60%+ subscribership
at 25Mbps+, and 60%+ subscribership at 10Mbps+ speeds.17 M unis is a state-level binary variable that
equals one for counties in states that had restrictions on municipal broadband as of Q1:2018 – since the
panel is biannual (June/December), this corresponds with June 2018. M unis is interacted with an indicator
for the post-pandemic period (P ostt ) that equals 1 for each biannual period after June 2020. β represents
the effect of restrictions on broadband subscribership following the onset of the pandemic. Year×6-month
fixed effects (λt ) help account for unobservables that vary every half-year like economic conditions and
seasonality. To avoid the confounding effects of cross-county differences in initial broadband availability,
we include county-fixed effects in our linear probability models. However, using county-fixed effects is not
possible in our ordered logistic models, so we include initial broadband adoption in the county for 2017,
as a control instead.18 Specifically, we use the December 2017 measure for broadband adoption in that
17
From Table1, 100Mbps+ subscribership is comparatively low – for the median county, adoption is 0 to 20% during the pre-period.
Hence we use the 20% threshold for 100Mbps+ and the higher 60% threshold for other speeds.
18
Because we have a large number of counties, but only use a small number of time periods (FCC data are biannual leading to only
17
county at the same speed as a control. For example, when estimating effects on 100Mbps+ speeds, we use
the December 2017 measure of broadband adoption at 100Mbps+ for that county as a control in lieu of
county fixed-effects. Finally, we control for a suite of time-varying county characteristics known to affect
broadband adoption including total population, population density, % population with a bachelor’s degree
or higher, lagged % employed and median income, and % population below the poverty line, all of which
are described in Table 2.19
We also present the results of event studies estimated as linear probability models over the sample
period (2018–2022) to rigorously estimate changes over time. Our event-study model is defined as follows:
∑
5
yist = βk (M unis × Ik ) + ΩXit + δi + λt + ϵist (2)
k=−4,k̸=−1
In Equation 2, yist is now a binary outcome that indicates whether broadband adoption (as measured
by actual connections per total number of households) is 20% or more, 60% or more, or 60% or more at
download speeds of 100Mbps+, 25Mbps+, or 10Mbps+ respectively in a county i = 1, . . . , N in state s
at time t, where t indicates a biannual time period between June 2018 and December 2022 (10 periods
total). M unis is a state-level binary variable that equals one for counties in states that had restrictions on
municipal broadband as of Q1:2018. We interact the binary M unis variable with binary variables indicating
biannual periods from June 2020 (when the COVID-19 pandemic begins). We omit the interaction between
M unis and December 2019 as the period just before the base period. We estimate Equation 2 separately
for download speeds of 100Mbps+, 25Mbps+, and 10Mbps+. The coefficients of interest, βk , show the
effects of municipal restrictions on broadband adoption for different speeds before and after the COVID-19
pandemic. Xit and λt are defined in the same way as in Equation 1. To avoid the confounding effects of
cross-county differences in initial broadband availability, we include county-fixed effects (δi ).
10 observations for each county between 2018 and 2022), we do not include county-fixed effects in our ordered logistic models.
This helps us avoid the incidental parameters problem that is well known in non-linear panel data models.
19
We employ the 5-year ACS estimates to adequately cover all counties including those with smaller populations. The ACS also
has measures of broadband usage. Employing the 5-year ACS measure on broadband (ACS5), we do not find significant impacts
of municipal restrictions on broadband adoption. This could be because effects are dampened by averaging the overlapping
broadband measures over 5 years and because ACS survey collection was disrupted by the COVID pandemic. Further, ACS
respondents are not queried about broadband speeds. This makes the ACS5 measures prone to errors such as changing broadband
standards over a 5-year survey period. In contrast, our FCC and BLS measures on broadband adoption are consistently coded,
they cover most counties, and they are higher in frequency (i.e., FCC adoption measures are biannual; ATUS telework measures
are monthly), allowing us to identify sharp changes in usage after COVID.
18
Labor Market Outcomes
Next, we estimate the effects of municipal broadband restrictions on labor force outcomes using a
repeated (quarterly) cross-section of data constructed from the CPS between Q1:2018 and Q4:2022. We
compare labor force outcomes for people living in states with restrictions on municipal broadband to labor
force outcomes for people living in states without restrictions before and after the COVID-19 pandemic. We
focus our analysis on married women with children, a group of people whose labor force participation may
have been particularly sensitive to broadband availability during the COVID-19 pandemic. For comparison,
we also show results for two other groups: (i) women without children and (ii) married men with children.
We estimate an event-study model defined as follows:
∑
11
yirst = βk (M unis × Ik ) + Xit Ω + Cct Φ + δs + λtxr + ζBst + ϵirst (3)
k=−8,k̸=−1
We estimate Equation 3 as a linear probability model where yirst is an indicator variable for outcomes
such as labor force participation and employment for individuals i = 1, . . . , N in state s from region r at
time t, where t indicates a quarter between Q1:2018 and Q4:2022 (20 quarters total). M unis is a state-level
binary variable that equals one for people living in states that had restrictions on municipal broadband as of
Q1:2018. We interact the binary M unis variable with binary variables indicating quarters from Q1:2020
(when the COVID-19 pandemic begins). We omit the interaction between M unis and Q4:2019 as the period
just before the base period. Individual-level control variables (Xit ) include age, race (Black Non-Hispanic,
other Non-Hispanic, and Hispanic), education (less than high school, some college, bachelor’s degree, ad-
vanced degree), metro area residency, MSA or CBSA size, number of children in the household, and an
indicator for at least one child under age 13. The δs are state fixed effects, and Cct includes the following
county-quarter-level controls: school closures,20 cumulative COVID-19 cases per 1 million residents, and
cumulative COVID-19 deaths per 1 million residents.21 λt×r are time-by-region fixed effects and Bst is a
20
The data set we use for school closures does not have estimates for counties not found in the CPS; it only provides estimates
for larger counties (Parolin and Lee 2022). Thus, in the case of school closures, we are forced to use statewide averages if the
CPS excludes the county. Helpfully, all 3 of our COVID-19-related measures are highly correlated, so including accurate controls
for cumulative COVID case and death rates should control for a large proportion of the labor supply effects coming from school
closures.
21
Our COVID controls - cumulative case and death rates are similarly aggregated. If the CPS identifies an individual’s county, we
19
Bartik shift-share variable, both of which we use to control for time-varying demand for labor (explained
below). We estimate Equation 3 separately for married women with children, women without children, and
married men with children. We cluster the standard errors at the state-level. The coefficients of interest, βk ,
show the effects of municipal restrictions on labor force attachment for these three groups before and after
the COVID-19 pandemic.
Finally, our main specification for labor market outcomes is Equation 4, where we summarize the
results from Equation 3 with a standard difference-in-differences estimator:
yisrt = β(M unis × P ostt ) + Xit Ω + Cct Φ + δs + λt×r + ζBst + ϵisrt (4)
Equation 4 contains the same control variables as in Equation 3. The difference is that we only include
one interaction between the binary variable for states with municipal restrictions on broadband (M unis ) and
an indicator for the post-pandemic period (P ostt ). In Equation 4, P ostt = 0 from Q1:2018-Q4:2019 and
P ostt = 1 from Q3:2021 (beginning August 2021) to Q4:2022. Data from January 2020 to July 2021 are
excluded to avoid the confounding effects of the early pandemic. Data from the second quarter of 2020
(April to June 2020) and more generally for the entire year of 2020 are also known to have significant noise.
This is because the pandemic impacted CPS data collection and led to a large dip in response rates for all
measures during that period.22 Although we include the full period between 2018 to 2022 in our event stud-
ies, while plotting raw trends and using the most restrictive specification in our robustness Table in Figure
A13, we believe that significant noise and higher than usual sampling error during the entire year of 2020
provides further justification to exclude this period when estimating our main results. Moreover, our event
study estimates showed that broadband restrictions became statistically significantly important for married
mothers’ labor supply starting in Q3:2021, after extended unemployment benefits expired. Finally, we es-
timate Equation 4 for our three subgroups: (i) married women with children, (ii) women without children,
and (iii) married men with children.
match it with cumulative COVID case and death rates for that county and quarter. For state-wide averages, we calculate within-
state averages across counties other than the larger counties for which the CPS reports the county identifier, rather than overall
statewide averages of these data, since this better represents cases and deaths for counties for which the county identifiers are not
available.
22
See Table1 in https://www.bls.gov/osmr/research-papers/2020/pdf/st200030.pdf and the CPS response rate chart in
https://www.bls.gov/cps/methods/response_rates.htm
20
Identifying Assumptions and Limitations
Our identifying assumption is that the labor force outcomes for mothers in states with municipal re-
strictions would have evolved comparably over time to mothers’ outcomes in states without restrictions,
if not for the restrictions. We cannot test this assumption directly, but we can compare differences in
trends in labor force participation across these states prior to COVID-19. To evaluate differences in trends
in the pre-COVID-19 period, we plot our dynamic difference-in-differences estimates. Our estimates for
[β−8 , . . . , β−1 ] show a lack of differential pre-trends in labor force outcomes for married women with chil-
dren. To verify the lack of differential pre-trends, we test whether these parameters are jointly statistically
significant.
One advantage of the COVID-19-era analysis is that nearly all of the states with restrictions on mu-
nicipal broadband implemented those restrictions prior to the COVID-19 pandemic (we drop the exception
states – Arkansas, California and Washington). Moreover, it was less common to work from home prior
to the pandemic. The U.S. Government Accountability Office (GAO) used the ATUS to estimate that the
percentage of workers who teleworked for any portion of an average workday increased from 24 percent to
38 percent from 2019 to 2021 (U.S. Government Accountability Office 2023). Using the ACS instead, GAO
found that the percentage of workers who primarily teleworked in the prior week more than tripled from 5.7
percent in 2019 to 17.9 percent in 2021.
Since it was less common to work from home before the pandemic, it is plausible that municipal
broadband restrictions had small and constant labor market effects before 2020. The fact that we fail to
find differential pre-trends in the labor force outcomes of mothers across states with and without municipal
restrictions prior to the COVID-19 pandemic seems to support this hypothesis. As we show later, labor
market impacts of municipal broadband policy increased after the pandemic due to the increased importance
of telework.
We control for several potential confounders in our analysis. First, we control for confounders related
to the COVID-19 pandemic itself. For example, it is well known that state and local governments took dif-
ferent approaches to in-person K-12 schooling during the pandemic. If states with restrictions on municipal
broadband also closed in-person schools for longer periods of time, then our results for mothers could be
driven by differences in school closures rather than differences in access to broadband. Therefore, we control
21
for school closures. In addition, we control for COVID-19 cases per capita and COVID-19 death rates per
capita at the county-quarter-level in an effort to proxy for the severity of the pandemic. The rationale is that
COVID-19 case/death rates could have affected labor supply/demand differently across counties and/or they
could have been correlated with state policies to address the pandemic.
Second, we use two strategies to control for differences in labor demand across states during our
sample period. First, we include include time-by-region fixed effects to control for time-varying demand
for labor by region over time.23 This approach helps mitigate the concern that differences in women’s labor
force attachment across states reflects regional differences in underlying economic conditions rather than the
effects of state restrictions on municipal broadband. It also means we compare states from the same region
and the same quarter – for example, we compare Alabama and Tennessee (restriction states) with Kentucky
and Mississippi in the same quarter. Second, we use a Bartik shift-share variable (Bst ) (Bartik 1991). In
the literature, it is common to include shift-share variables as controls when estimating models for labor
supply, especially when policy variation is at the state-level (Ne’eman and Maestas 2023; Chodorow-Reich
et al. 2012). This helps to mitigate the concern that our estimated labor supply effects are driven by state
differences in pre-existing shares of workers in industries with teleworkable jobs. We follow Ne’eman and
Maestas (2023) and construct the shift-share variable at the state-level using data from the ACS. We measure
each state’s initial industry share in 201524 and interact that share with the national employment level for that
industry over time, t. The variable is rescaled (divided by 1,000) to produce the estimates in the Appendix
tables.
Finally, we note a key limitation in our analysis. Although we show that prime-age workers in restric-
tion states are less likely to use the internet at home and mothers are especially less likely to use the internet
for calls (using data from the CPS Internet Use Supplement), we do not know the menu of broadband plans
or prices that the CPS respondents face. Therefore, the major limitation of our study is that we cannot prove
directly that the mothers who opted out of the labor force in restriction states post-COVID-19 did so because
they lacked access to a high-speed broadband connection at home. Instead, in the results that follow, we
provide suggestive evidence that lower broadband adoption in restriction states– driven by fewer provider
choices and more expensive plans – may have affected maternal labor supply decisions.
23
The CPS identifies the region (census division) where the respondent’s housing unit is located. Collections of states are categorized
into regions. For example, the East North Central Division includes Illinois, Indiana, Michigan, Ohio and Wisconsin. There are
9 Census divisions in the U.S.
24
Our results are not sensitive to using a different year for calculating initial industry shares.
22
Results
We first present difference-in-differences (DD) and event-study results demonstrating how municipal
broadband restrictions affected broadband adoption post-COVID, and we highlight the role of (higher) prices
and (less) competition. Next we confirm that municipal restrictions reduced internet usage by showing
difference-in-differences results using the CPS Internet Use Supplement. Notably, married women with
children see a substantial decrease in using the internet for calls – a good predictor of high-speed internet
use for a common modality of telework. Then we examine how municipal broadband restrictions affected
maternal labor supply, and we compare pre-trends in maternal labor supply across states with and without
restrictions. We further estimate the effects of municipal broadband restrictions on telework and caregiving –
highlighting their role as potential channels. In robustness tests, we show that our results for both broadband
subscriptions and maternal labor supply are robust to several alternative specifications. Finally, we explore
heterogeneity in the effects of municipal broadband restrictions on post-COVID maternal labor supply by
education, race, and by the age of children in the home.
Broadband Adoption
Table 4 presents the impact of state-level municipal broadband restrictions on broadband subscribership
as estimated in Equation 1.25 Recall that the dependent variable is an ordinal tier (0, 1 or 2) biannual value
for FCC County Broadband adoption (2018-2022) reflecting connections per total housing units in a county
where 0 is 0 to 40%, 1 is 40 to 60%, and 2 is over 60%. The ordered logistic estimate in column 1 implies
that the post-COVID effect of restrictions was a 0.45 increase in the log-odds of being in a lower ordinal tier
of broadband subscribership; for example, it increased the probability of going from over 60% adoption to
between 40 to 60% adoption at 100Mbps or higher. Figure 3 plots the marginal effects on the ordinal tier
value of connections based on the ordered logit estimates from column 1. There is a decrease in the relative
odds at the ordinal values 1 and 2, implying a lower probability of 100Mbps+ subscribership that is 40% or
higher in the county. In column 2, we employ a linear probability model using a county-level binary outcome
that equals 1 if county’s 100Mbps subscriptions is 20% or higher. The estimate in column 2 suggests that
municipal restrictions decreased the probability of at least 20% 100Mbps subscribership by 2.7 percentage
25
Appendix Table A3 presents complete estimates from column 1 of the table correct to 4 decimal places.
23
points.
In columns 3 and 4 of Table 4 we see similar effects for the number of county broadband connections
at 25Mbps download speeds. Column 3 has a negatively signed estimate using the ordinal tier value from
0 to 2 for 25Mbps+ penetration. The marginal effects from the ordered logit model in column 4 are plotted
in Figure 4. There is a decrease in the probability of the county having 60% or more households subscribed
to 25Mbps+ speeds. This is confirmed using the linear probability model estimate in column 4 (negative
2.1 percentage points). Although we do not see a significant estimate employing the ordinal tier value for
10Mbps+ in column 5, we find a significant effect using the LPM at a threshold of 60%+ subscriptions. This
estimate in column 6 shows that even for speeds of 10Mbps or more, restrictions decreased the probability of
at least 60% of the households in a county being connected to broadband by 2.5 percentage points. Scaling
the estimates by the means of the dependent variables shows they are meaningful in magnitude. The LPM
estimates in columns 2, 4, and 6 of Table 4 suggest that municipal broadband restrictions decreased the
probability of 20% or more broadband adoption by 4.4% for 100Mbps+, of 60% or more broadband adoption
by 8.4% for 25Mbps+, and 60% or more broadband adoption by 6.1% for 10Mbps+ in the COVID era.26
Figure 5 presents an event study using a window that begins in June 2018, 2 years before the first
COVID-19-pandemic observation in June 2020, and extends to December 2022 – about two and half years
after. Figure 5 plots OLS coefficients of the dynamic effects for states with municipal broadband restric-
tions compared to states without restrictions over time on county broadband subscribership of 20% or more
at 100Mbps+ speeds. Specifically, the treatment variable for state-level municipal broadband restrictions27
is interacted with dummy variables for each of the 10 semi-annual time periods between June 2018 and
December 2022. Estimated effects of municipal restrictions are insignificant for the 2 periods immediately
before the pandemic, covering December 2018 and June 2019. This suggests that broadband subscribership
rates prior to the pandemic were not trending differently over time according to state-level broadband policy
prior to the pandemic. Instead, the coefficient on municipal restrictions becomes significant only about a
year into the pandemic (December 2020) with the effect rising and peaking at the end of the sample period
(December 2022) at negative 7.57 percentage points. This provides evidence that the estimates in Table 4
are not biased by pre-pandemic trends in broadband adoption. Instead, it suggests that municipal restric-
tions may have had a causal impact on broadband subscribership following the onset of COVID-19 and the
26
From column 2, −0.027
0.61
∗ 100 = 4.4%. From column 4, −0.021 0.25
∗ 100 = 8.0% From column 6, −0.025 0.41
∗ 100 = 5.9%
27
Recall, the treatment variable is a binary variable that equals one for counties in states that had municipal restrictions as of Q1:2018
24
accompanying growth of telework.
Figures 6 and 7 replicate this event study for 25Mbps+ and 10Mbps+ speeds, respectively, with similar
results. Although less precise, the estimates follow a similar trend. By December 2022, there is a 4.79
percentage point lower chance that a county in a restriction state will be 60% or more covered by 25Mbps+
speeds vs. a county in a non-restriction state. For 10Mbps+ speeds, by December 2022, the estimate is
slightly smaller (about 2.32 percentage points) and not significant – indicating that at lower speeds, access
may have begun to converge across states by 2022. This is consistent with the convergence in maternal labor
supply between restriction and non-restriction states in 2022 that we see in Figure 10.
To investigate the channels that drive the post-COVID divergence in broadband penetration across
restriction and non-restriction states, we plot data on the number of broadband plans in each state and the
affordability of the average plan from the FCC Urban Rate Survey. The FCC only provides data at the state-
level over the sample period (Federal Communications Commission 2025b). For plan choice (Figure 8), we
restrict the data to plans with minimum 25/3 (download/upload) speeds, unlimited usage, and costing $80
or less in total monthly charges for each state-year observation. For affordability (Figure 9), we also restrict
to plans with unlimited usage at 25+/3+ speeds and calculate the percent of monthly median household
income spent on total monthly charges for the average plan. From the figures, it is clear that broadband
consumers face far fewer choices (Figure 8) and lower affordability (Figure 9) in restriction states. This
result is consistent with previous work that finds private ISPs often offer low promotional rates that later rise
sharply (Talbot, Hessekiel, and Kehl 2018).
To rigorously evaluate the role of consumer choice and price as channels, we interact the number of
broadband ISPs available (2 or more) with our independent variable (M unis × P ostt ) in our difference-in-
differences model. Table 5 utilizes a county and time-varying measure of the percentage of county population
(0-100%) covered by 2 or more broadband providers from the FCC Area Tables (2018-21). The FCC does
not provide this measure after 2021. The independent variable in row 1 is a dummy that equals 1 for states
with restrictions for biannual periods beginning in June 2020. The estimate on the interaction between the
independent variable and the percentage of the county covered by 2 or more providers is in row 3. The
coefficient on the interaction term in row 3 is positive and significant in columns 2 and 3. This suggests
that counties in states with restrictions and competitive ISP markets have smaller relative differences in
subscribership compared with counties in states without restrictions. Specifically, counties with 0-1 ISPs in
25
restriction states are 9.3 percentage points less likely to have subscribership levels above 60% at 25Mbps (row
1, column 2) compared to counties with 0-1 ISPs in non-restriction states. However, a 10% increase in 2+
provider coverage reduces the difference in the probability of 60%+ subscribership between restriction and
non-restriction states to -6.7 percentage points (−9.3 + 2.6). In other words, the probability of high levels
of subscribership after COVID is lowest in counties in restriction states with non-competitive broadband
markets.
For information on county-level broadband prices, we turn to data from BroadbandNow. Broadband-
Now provides county-level measures of the lowest monthly prices in a report titled, “United States County
Broadband Statistics for 2020.” Unlike the FCC, which provides average prices at the state-level (during
our sample period), BroadbandNow provides the lowest price at the county-level (Tanberk 2020). They de-
fine lowest price as the average of all lowest priced broadband speed plans available in the county at 25/3
speeds. This data is only available for 2020. In Table 6 we interact the independent variable with a dummy
that equals one if the county’s lowest price for broadband is over 2% of the the county’s median household
income.28 Note that because BroadbandNow only provides prices for 2020, the dummy for affordability is
absorbed as a county-fixed effect. In all three columns, we see that the effects of municipal restrictions on
broadband subscriptions are operating through counties in restriction states where plans are priced higher
(row 2). In particular, in restriction-state counties where the lowest priced plan costs over 2% of a house-
hold’s budget, the probability of at least 60% broadband subscribership at 25Mbps or 10Mbps is between
5.1 and 6.2 percentage points lower, respectively, compared with restriction-state counties where the lowest
priced plan is 2% or less of the median household’s budget.
For information on broadband service quality, we use a measure of data caps from the FCC Urban Rate
Survey (Federal Communications Commission 2025b). During the analysis pre-period (2018–19), plans in
restriction states were 11 percentage points more likely to be subject to a data cap. Further, among plans
subject to monthly data caps, the usage allowance was 106GB lower in restriction states.29 Assuming 1 GB
of use per hour, this difference would allow subscribers in non-restriction states to make over 100 hours
of additional video calls every month. There are many other aspects to service quality that we wish we
28
Results are robust to other affordability metrics – such as a $75 per month threshold while controlling for median household
income in the county. Households spent 4.3% of their monthly income on all other utilities in 2025 including electricity, gas,
heating and water (Lending Tree 2025), a figure that is historically high itself. FCC guidance suggests two percent of monthly
household income as a preliminary standard. See https://docs.fcc.gov/public/attachments/FCC-16-38A1.pdf
29
These state-level differences are significant beyond the 1% level. The Urban Rate Survey does not provide data at the county-level
during our sample period.
26
could have explored. However, the FCC acknowledges “..there are additional areas and measurements other
than latency and consistency of service that are likely relevant to the universal service goal of availability.
These include other metrics such as service outages and access to inside wiring. At present, we do not have
comprehensive sources of data, but intend to revisit this issue in future inquiries” (FCC 2024).
Taken together, our results in this section suggest that broadband markets in restriction states had less
competition leading up to the pandemic (Table 1 and Figure 8). We then show that the relative decreases in
broadband subscribership observed after the pandemic in restriction states were likely driven by mechanisms
such as reduced competition, lack of consumer choice, and less affordable prices (Figure 9, Table 5, and Table
6).
Internet Use from the CPS
Next we examine whether municipal broadband restrictions affect how people use internet, by using
the CPS Internet Supplement. The CPS Internet Use data are only collected for a single month (November)
every other year. We use data from November 2017 and 2019 as our pre-period, and from November 2021
and 2023 as our post-period. Note that this time period overlaps with, but is not the same as, the sample
period that we employ for our labor supply outcomes (2018-2022). Internet Supplement data are also much
smaller in sample size (<10% the size of our main CPS sample) – especially for outcomes such as the use of
internet for email, phone and job search – so we should not expect to precisely estimate effects. Finally, the
CPS Internet Use measures that we employ are generally broad; for example, the CPS simply asks whether
someone accesses the internet from home or not. So someone answering “yes” to this question could be
using dial-up internet or a cellular data plan.
We present our difference-in-differences results on CPS Internet use in Table 7. In panel A, we pool
the sample for all prime-age (25-54 year-old) individuals. Prime-age workers in restriction states are 2.5%
(= −0.02
0.81 ∗ 100%) less likely to access the internet at home (column 1). Despite the broader internet use
question in the Internet Use Supplement (and previous caveats), the negative 2.5% estimate in column 1 is
directionally consistent with the 5.9% estimate in column 6 of Table 4. CPS respondents in restriction states
are also less likely to access the internet outside the home (column 2) and they are less likely to use email
(column 3), although these estimates are smaller in magnitude. This suggests that during the post-COVID
period, CPS respondents in restriction states not only experienced a relative decrease in a broad measure
27
of internet use, but they may also have experienced modest reductions in internet use even for ubiquitous,
low-bandwidth activities such as email.
Looking at our subgroups of interest (panels B through D), we see that all 3 subgroups experience a
reduction in home internet use; however, we are only able to estimate this with precision for women without
children (column 1, panel B). However, estimates for outcomes in columns 1 through 3 are not statistically
different across the 3 subgroups.
Column 4 of Table 7 uses a dummy for whether or not the respondent uses the internet for calls.
Because of how the CPS asks this question, we believe this outcome is a good predictor of high-speed internet
access especially for telework. The CPS asks, “In the past six months, (have/has) (you/NAME) participated
in video or voice calls or conferencing over the Internet, such as with SKYPE or FaceTime? (Do you/Does
NAME) participate in video or voice calls or conferencing?” Higher internet speed thresholds are often
required for common modalities of telework such as video conferencing. From column 4 of Table 7, we see
that married women with children (panel B) experience a large relative 5.1% (= −0.034
0.67 ∗ 100%) decrease
in using the internet for telephone calls during the post-pandemic period. This scaled estimate is similar in
magnitude to the 4.7% decrease in employment that we estimate married women experienced in restriction
states during the post-pandemic period (see Table 8, panel A, column 2). The effects for other groups (in
column 4) are not precisely estimated.
Given that there is a large literature on the role of the internet in reducing market frictions and im-
proving labor market matching (Bhuller, Kostol, and Vigtel 2020), we also examine the use of the internet to
search for jobs. In column 4 of Table 7, we see that despite a lack of statistical significance, all workers may
have experienced a modest decrease in the probability of using the internet to search for jobs (panel A). This
effect is driven by married men with children (panel D). The scaled estimate implies that married fathers
may have experienced an 11% decline in the probability of searching for a job using the internet (p = 0.18).
Given the small sample size and lack of statistical significance, it is important to treat this estimate with cau-
tion. However, a decrease in job searching using the internet could help explain the modest 1.0% decrease
in employment that we observe for married men with children (see panel C, column 3 of Table 8).
28
Labor Force Outcomes
Figure 10 plots the dynamic difference-in-differences estimates for three maternal labor force out-
comes: the probability of being in the labor force (LFP), the probability of being employed at all, and the
probability of being employed, at work. We find that married mothers are less likely to be in the labor force,
they are less likely to be employed, and they are less likely to be employed at work in states with restrictions
on municipal broadband compared with states without restrictions, post-2020. The results are most signif-
icant for married mothers in states with municipal restrictions starting in Q3-2021 and extending through
Q4-2022.30 This divergence in maternal labor force participation across states with and without broadband
restrictions coincides with the expiration of extended federal unemployment benefits (Coombs et al. 2022).
In other words, married mothers’ ability to work from home may have become especially consequential for
their labor force participation as unemployment insurance benefits expired.3132
Table 8 summarizes the effects of municipal broadband restrictions on labor force outcomes for mar-
ried women with children (panel A), women without children (panel B), and married men with children
(panel C).33 We find that broadband restrictions only reduce labor force participation among married women
with children. The probability that married mothers are in the labor force is 0.027 lower in states with re-
strictions on municipal broadband compared with states without restrictions during the pandemic (column
1). Additionally, the probability that married mothers are employed-at-work is 0.031 lower in states with
municipal restrictions compared with states without restrictions during the pandemic (column 2). There is
also a reduction in the probability that married mothers are employed (at all) in states with restrictions dur-
ing the pandemic (column 3), but there is no effect on the number of conditional hours worked (column
4).34 Our results suggest that restrictions on municipal broadband primarily affected the extensive margin
30
This divergence in maternal labor force participation rates across restriction and non-restriction states can also be seen in the
raw trends for the same labor force outcomes in Figure A1. Labor force participation and employment decrease in states with
restrictions (dotted line) starting in Q3-2021 while they increase in states without restrictions (solid line).
31
Related to the timing of the effect, we rule out alternative explanations such as changes in sample sizes or school re-openings
(which we control for). We also find similar results when we only examine mothers with children aged 3 and older (results
available upon request).
32
The event studies for women without children and married men with children are plotted in Appendix Figures A2 and A3 respec-
tively.
33
Appendix Tables A5 through A8 present complete regression estimates for all three sub-groups. Appendix Table A5 reports the
estimates for labor force participation, Table A6 reports the estimates for “employed at work,” Table A7 reports the estimates for
employment overall, and Table A8 reports the estimates for hours worked last week conditional on working at all. The COVID
rate, COVID death rate, and Bartik Shift-Share variables are rescaled by 1,000 to produce the numerically meaningful estimates
in these tables.
34
The number of hours worked last week is conditional on working at all, the unconditional hours are 0 for individuals who are not
employed.
29
of employment for married mothers, rather than the intensive margin. panel B shows no relationship be-
tween restrictions on municipal broadband and labor supply for women without children. Married men with
children see modest effects on employment and hours, although the latter is not precisely estimated (panel
C).
Scaling the estimates by the means of the dependent variables shows that married women with children
in states with municipal restrictions were 4% (= −0.0268
0.67 ∗ 100%) less likely to be in the labor force and
4.7% (= −0.0305
0.655 ∗ 100%) less likely to be employed after the COVID-19 pandemic compared with their
counterparts in states without broadband restrictions. Married fathers in restriction states were about 1.0%
(= −0.009
0.93 ∗ 100%) less likely to be employed after the COVID-19 pandemic.
Labor Supply Channels: Caregiving and Telework
Table 9 shows that much of the decrease in maternal labor supply in states with restrictions after
COVID-19 comes from married women with children shifting from employment to childcare (panel A).
0.314 ∗ 100%)
Married women with children in states with restrictions on municipal broadband are 7.3% (= 0.023
0.289 ∗
more likely to work part-time or not-at-all due to “family reasons” (column 2). They are 8.5% (= 0.0246
100%) more likely to “not be in the labor force (NILF)” because they are “taking care of house or family”
(column 3). However, we find no difference in the probability of having “child care problems” for married
mothers in states with broadband restrictions compared with states without restrictions during COVID-19
(column 4). Similar to Table 8, we find no significant effects of restrictions on municipal broadband during
COVID for these outcome variables for women without children or married men with children (panels B and
C, respectively).
We also consider whether an individual works part-time (< 35 hours per week across all jobs) as an
additional measure of the intensive margin of labor supply.35 These estimates are reported in column 1 of
Table 9. In column 1, we see that the effects for switching from full-time to part-time work are small and
not precisely estimated for all 3 groups.
The effects of municipal broadband restrictions on work-related time use appear in Table 10. All
estimates are measured in minutes spent per workday except for column 2, which is a share between 0 and 1
35
Our definition is based on the definition of “usual full-time work” in the CPS published by the Bureau of Labor Statistics :
https://www.bls.gov/cps/definitions.htm#fullparttime.
30
(conditional on working a positive amount of time). The sample in Table 10 includes employed people only.
72 × 100% = 64%) working from
In restriction states, working mothers spend 46 fewer minutes per day ( 45.9
home compared with working mothers in non-restriction states. They also experience an 11.2 percentage
0.226 × 100% = 49%) relative decrease in the share of work time spent at home, and a 7 minute
point ( 0.1115
25 × 100% = 27%) relative increase in commute-time. These results indicate that even among mothers
( 6.66
that continue to work, a lack of broadband access leads to a substantial decrease in work-from-home (about
191 hours per year)36 and a large increase in commute time. The estimated effects in Table 10 are smaller
and lack statistical significance for women without children and married fathers.
Heterogeneity by Family Characteristics
Next, we present results showing how restrictions on municipal broadband affect maternal labor supply
by family characteristics. Comparing labor supply effects for mothers across different types of marital status
we find that the absence of a spouse magnified the effect of municipal broadband restrictions on mothers’
labor supply (effects are 3 times larger compared with if spouse/partner is present). This is reported in Table
A9.37
Table A10 shows results by education level for married women with children. Previous research has
shown that college graduates were more likely to work remotely during the pandemic (Goldin 2022). We
may therefore expect our effects to be concentrated among this group. However, our estimates for college-
educated and non-college educated mothers are not statistically significantly different from one another. We
think there are several reasons for this. A large share of women without a college degree also worked in
teleworkable jobs as of 2022 (Cowan 2023) and would therefore be similarly affected by restrictions on
broadband. Further, an increasing share of consumer spending is coming from high-income workers – who
tend to have a higher education.38 The withdrawal of some of this consumer spending – from college-
educated educated mothers dropping out of the labor force may have effects for mothers with lower levels
of education working in consumer service establishments such as hotels, restaurants, coffee shops, bars, and
36
We assume 250 work days per year.
37
Some may find the large coefficient estimate on Spouse Absent in row 3 of the table surprising. This reflects a large and persistent
gap in the labor supply of married women with a spouse present versus mothers when the spouse is absent. See https://
www.bls.gov/news.release/archives/famee_04192023.pdf
38
See figure 3 from https://www.federalreserve.gov/econres/notes/feds-notes/a-better-way-of-
understanding-the-u-s-consumer-decomposing-retail-spending-by-household-income-20241011.html
31
hair salons. Previous research has found that the location decisions of highly educated workers specializing
in business services affects consumer spending, and therefore the work hours of consumer service workers,
who tend to be less educated (Althoff et al. 2022).
Table A11 shows results by race (white vs. non-white) for married women with children. Here we
find that restrictions on municipal broadband affect women of different races similarly; both white (panel
A) and non-white (panel B) married mothers reduce their labor force participation in states with restrictions
compared with states without restrictions during the COVID-19 pandemic. In other words, our estimated
effects by race are not statistically significantly different. Finally, we divide the sample of married mothers
into groups based on the age of the children in Table A12. Our results are largest in magnitude, and most
precisely estimated, for married mothers whose youngest child is under age 11 (panel A) (vs. 12 or older in
panel B); however, the estimates are not statistically significantly different.
Robustness Checks
There may be concerns that our labor supply results are sensitive to dropping the early portion of the
COVID period. Recall that our main specification in Table 8 drops CPS sample months between January
2020 and July 2021 to avoid measuring the direct effects of the pandemic. Further, cross-county differences
in initial broadband penetration may drive our labor supply results. To assuage these concerns, we present
labor supply results with our most restrictive specification in Table A13. This specification includes geo-
graphic area fixed effects (as described in the Data Subsection on Control Variables Related to the COVID-19
Pandemic) instead of state fixed effects. Additionally, the sample includes all months for years between 2018
and 2022. The results in Table A13 are qualitatively similar to those in Table 8, although slightly smaller in
magnitude.39
We also estimate a robustness test where we specify the “control group” of states as those that did
not have restrictions on municipal broadband and they had municipal broadband initiatives. This alternative
analysis might show that the impacts of municipal restrictions are even larger than we previously estimated
in our primary specification, since the control group in our primary specification included 6 non-restriction
states without any municipal broadband initiatives. The Institute for Local Self-Reliance (ILSR) provided
39
Differences in initial county-level broadband penetration may also be coming from differences in state municipal broadband policy.
Since including area fixed-effects also absorbs some of this variation, our results in A13 could be considered a lower (conservative)
bound.
32
us with a list of municipal broadband networks – data are available only for 2023. According to the ILSR
list, the six non-restriction states that have no municipal broadband initiatives include Arizona, Delaware,
Mississippi, New Mexico, North Dakota, and West Virginia. We begin by dropping these 6 former non-
restriction (i.e. control states from our sample). This leaves us with the same 17 treatment, i.e. restriction
states as before, but our control states now only include 23 non-restriction states that had municipal broad-
band initiatives. We then replicate the analysis using the main difference-in-differences specification for
Table 4 but use our restricted sample. Table A2 shows the results from this specification. Similar to Table
4, we use the same outcomes and estimate both an ordered logit and a LPM model.
Comparing results between Table A2 and Table 4, we see that dropping non-restriction states without
municipal broadband initiatives makes our results stronger. Estimates in columns 2–6 are larger in magni-
tude. In general, we also see that the effects are more precisely estimated. From columns 2, 4, and 6 in Table
A2, we see that restrictions reduced the probability of achieving high broadband subscribership thresholds
by between 2.6 percentage points and 4.3 percentage points when we consider only non-restriction states
with municipal broadband initiatives as control states. Note that although the average effects are larger (af-
ter dropping the 6 states), they are not statistically significantly different. This robustness test suggests that
states with actual municipal broadband initiatives, rather than those without bans on such initiatives, drove
the relative difference in broadband subscribership between restriction and non-restriction states.
Next, we perform a similar analysis for our labor outcomes, dropping the 6 non-restriction states
without municipal broadband initiatives. The results are presented in Table A14. Compared with Table 8,
the effects are slightly larger but not statistically significantly different.
We also consider whether our results are driven by states with municipal restrictions alone or by states
with a combination of municipal and co-op restrictions on broadband provision. Eleven out of seventeen
states with municipal restrictions had some type of roadblock to cooperative broadband provision in 2017.40
However, most of those roadblocks entailed “easement ambiguities.” Easement ambiguity means the instal-
lation of any broadband infrastructure – even a single cable – may exceed the scope of the existing easement
agreement and lead to claims of damages for trespassing from private landowners. Between 2019 and 2022,
40
These include Nebraska and Missouri (explicit bans), North Carolina (ban on use of state funds), Virginia (onerous cost imputa-
tions). Tennessee and South Carolina (limited electrical cooperatives to providing broadband only within their authorized service
territory or allowed a narrow buffer). The remaining states lacked explicit easement authority and included Alabama, Texas,
Michigan, Minnesota and Pennsylvania.
33
roughly 20 states clarified their easement issues, thus allowing electric utilities to install broadband infras-
tructure without securing easements from individual property owners (Lide and Magee 2022). Aside from
the easement ambiguity, 5 states had other significant roadblocks to co-op broadband provision in 2018.
Those states were Mississippi, Missouri, Nebraska, North Carolina, and Virginia. Therefore, in Table A4,
we show how broadband subscribership differed in the COVID-era between states with municipal restric-
tions only and states with municipal and co-op restrictions as of 2018 (MO, NC, NE, VA).41 We find no
statistically significant differences across the two types of states. The presence of municipal restrictions
alone appears to be the limiting factor to higher broadband subscribership; however, we acknowledge that
the history of easement ambiguities in many of these states complicates the analysis.
Next, we explore whether our labor force participation results hold up for all workers in teleworkable
jobs (instead of only married mothers). We replicate the results in Table 8, but for teleworkable jobs only.
We use the methodology from Dingel and Neiman (2020) to categorize occupations into teleworkable or not.
This gives us the result in Table A15.42 We see that results for married women with children in teleworkable
jobs (panel A) are less precisely estimated than our main results, potentially due to the smaller sample size,
but they are not statistically different than our main results. Interestingly, in panels B and C, we now see
comparable decreases in labor supply for women without children and custodial fathers in teleworkable jobs,
although only the result for the former group is precisely estimated (panel B). The results in Table A15 suggest
that labor supply for all 3 groups may have been affected by municipal broadband restrictions if the people
were in teleworkable jobs. This is consistent with previous work (Carvalho, Hagerman, and Whitacre 2022)
showing that low broadband adoption rates prevented counties from recovering their COVID-19 employment
losses.
Finally, we also rule out identification concerns due to drastic and differential changes in demographic
characteristics across restriction vs. non-restriction states around COVID-19. For example, non-restriction
states do not gain relatively large numbers of college-educated or young persons over the sample period.43
41
We drop MS from this analysis since it is the only state with co-op only restrictions.
42
Note that the table retains workers who are unemployed or not in labor force, since we cannot determine their occupation and
therefore cannot use the Dingel and Neiman (2020) methodology to determine if their occupation is teleworkable.
43
Results available upon request.
34
Discussion
This paper shows how restrictions on municipal broadband affected the labor force outcomes of married
women with children before and after the COVID-19 pandemic. In multiple surveys, Americans cite high
prices and limited access as barriers to using broadband. Municipal broadband has the potential to reduce
prices and increase access; for example, Talbot, Hessekiel, and Kehl (2018) show that municipal broadband
networks charge lower prices than private ISPs. Using FCC broadband subscribership data, we show that
broadband subscribership grew faster post-2020 in states that did not restrict local governments and cooper-
atives from competing in broadband markets. We further show that subscribership growth was particularly
low in states with municipal restrictions and non-competitive ISP markets and/or with less affordable plans.
Our paper’s primary hypothesis, therefore, was that reduced access to high-speed internet at home reduced
peoples’ ability to work from home during and after the pandemic (in restriction states). Our analysis focuses
on married women with children because their labor supply may have been especially sensitive to remote
work options around the pandemic. For comparison, we also consider the effects of broadband restrictions
on women without children and married men with children.
Our difference-in-differences estimates reveal that municipal broadband restrictions reduced labor
force participation (LFP) and employment among married women with children. Though maternal labor
supply was recovering in both restriction and non-restriction states during the post-pandemic period, it re-
covered more quickly in states without restrictions on broadband internet. In particular, LFP among women
with children in states with broadband restrictions begins its relative decline around Q3-2021. This time pe-
riod coincided with the expiration of federal extended unemployment benefits. As such, LFP among mothers
began increasing in the country overall in 2021, but LFP increased more in states without restrictions on mu-
nicipal broadband. We posit this divergence across states occurred because mothers had more ability to work
remotely in states without municipal broadband restrictions. In contrast, we find that municipal restrictions
did not affect labor force outcomes for women without children or married men with children, unless they
worked in teleworkable occupations.
This paper highlights the importance of broadband access in the modern economy. Fewer barriers to
entry in broadband markets (i.e., removing municipal/cooperative restrictions) could lead to greater compe-
tition among ISPs. More ISP competition may improve broadband take-up by making plans more affordable
35
and higher quality. In turn, increased broadband adoption by households may foster a flexible work envi-
ronment that can be conducive to mothers and other marginally attached workers.
36
Bibliography
Albanesi, Stefania and Jiyeon Kim. 2021. “The Gendered Impact of the COVID-19 Recession on the US Labor
Market.” Working Paper 28505, National Bureau of Economic Research. URL http://www.nber.org/papers/
w28505.
Alon, Titan, Matthias Doepke, Jane Olmstead-Rumsey, and Michèle Tertilt. 2020. “The Impact of COVID-19 on
Gender Equality.” Working Paper 26947, National Bureau of Economic Research. URL http://www.nber.org/
papers/w26947.
Althoff, Lukas, Fabian Eckert, Sharat Ganapati, and Conor Walsh. 2022. “The geography of remote work.” Regional
Science and Urban Economics 93: 103770.
Anderson, Monica. 2019. “Mobile Technology and Home Broadband 2019.” Tech. rep. URL https://
www.pewresearch.org/internet/2019/06/13/mobile-technology-and-home-broadband-2019/.
Angrisani, Marco, Jeremy Burke, and Francisco Perez-Arce. 2023. “Racial, Ethnic, and Gender Disparities in Labor
Market Outcomes in the Aftermath of the COVID-19 Pandemic.” Ann Arbor, MI. University of Michigan Retirement
and Disability Research Center (MRDRC) Working Paper .
Arkansas General Assembly. 2007. “AN ACT TO ENABLE THE USE OF ELECTRIC UTILITY POWER
LINES TO DEPLOY BROADBAND NETWORKS; AND FOR OTHER PURPOSES.” URL https://
arkleg.state.ar.us/Home/FTPDocument?path=%2FBills%2F2007%2FPublic%2FHB1589.pdf. Arkansas.
Arkansas State Broadband Manager. 2021. “Arkansas State Broadband Manager’s Report.” URL
https://sas.arkansas.gov/wp-content/uploads/2020/06/Broadband-Managers-Activities-
Operations-Report-Reduced-Size.pdf.
Atasoy, Hilal. 2013. “The effects of broadband internet expansion on labor market outcomes.” ILR review 66 (2):
315–345.
Barrero, José María, Nicholas Bloom, and Steven J Davis. 2023. “The evolution of work from home.” Journal of
Economic Perspectives 37 (4): 23–49.
Bartik, Timothy J. 1991. “Who benefits from state and local economic development policies?” .
Bauer, Lauren and Sarah Yu Wang. 2023. “Prime-Age Women Are Going Above and Beyond in the Labor Market Re-
covery.” Brookings Institution, The Hamilton Project (blog) URL https://www.hamiltonproject.org/
publication/post/prime-age-women-are-going-above-and-beyond-in-the-labor-market-
recovery/.
Bhuller, Manudeep, Andreas Ravndal Kostol, and Trond Christian Vigtel. 2020. “How Broadband Internet Affects
Labor Market Matching.” Tech. rep., IZA Discussion Papers.
Bode, Karl. 2024. “Arkansas Electric Cooperatives Pass 1 Million Broadband Connection Milestone.”
https://communitynetworks.org/content/arkansas-electric-cooperatives-pass-1-million-
broadband-connection-milestone.
Carvalho, Mckenzie, Amy Hagerman, and Brian Whitacre. 2022. “Telework and COVID-19 Resiliency in the South-
eastern United States.” Journal of Regional Analysis Policy 52 (1): 19–34.
37
Cash, Cathy. 2024a. “Arkansas Co-ops Build Fiber Broadband Access to 1.4 Million Residents.”
https://www.electric.coop/arkansas-co-ops-build-fiber-broadband-access-to-1-4-million-
residents.
———. 2024b. “Arkansas Co-ops Build Fiber Broadband Access to 1.4 Million Residents.” URL
https://www.electric.coop/arkansas-co-ops-build-fiber-broadband-access-to-1-4-million-
residents.
CDC. 2022. “Centers for Disease Control and Prevention, COVID-19 Response. Weekly United States COVID-
19 Cases and Deaths by State (version date: October 20, 2022).” URL https://data.cdc.gov/Case-
Surveillance/Weekly-United-States-COVID-19-Cases-and-Deaths-by-/pwn4-m3yp.
Chamberlain, Kendra. 2021. “2020 Report: Municipal Broadband Is Roadblocked Or Outlawed In 22 States.” URL
https://broadbandnow.com/report/municipal-broadband-roadblocks-2020.
Cheng, Wei, Patrick Carlin, Joanna Carroll, Sumedha Gupta, Felipe Lozano Rojas, Laura Montenovo, Thuy D Nguyen,
Ian M Schmutte, Olga Scrivner, Kosali I Simon, Coady Wing, and Bruce Weinberg. 2020. “Back to Business and
(Re)employing Workers? Labor Market Activity During State COVID-19 Reopenings.” Working Paper 27419,
National Bureau of Economic Research. URL http://www.nber.org/papers/w27419.
Chodorow-Reich, Gabriel, Laura Feiveson, Zachary Liscow, and William Gui Woolston. 2012. “Does state fiscal relief
during recessions increase employment? Evidence from the American Recovery and Reinvestment Act.” American
Economic Journal: Economic Policy 4 (3): 118–145.
City of Waterloo. 2023. “Waterloo Fiber Project/Municipal Communications Utility.” URL https://
cityofwaterlooiowa.com/government/city_boards_and_commissions/telecommunications_board/
waterloo_fiber_project_municipal_communications_utility_.php.
Coombs, Kyle, Arindrajit Dube, Calvin Jahnke, Raymond Kluender, Suresh Naidu, and Michael Stepner. 2022. “Early
withdrawal of pandemic unemployment insurance: Effects on employment and earnings.” In AEA Papers and
Proceedings, vol. 112. American Economic Association 2014 Broadway, Suite 305, Nashville, TN 37203, 85–90.
Cooper, Tyler. 2024. “Municipal Broadband Remains Roadblocked In 16 States (May 2024).” URL https:
//broadbandnow.com/report/municipal-broadband-roadblocks.
Couch, Kenneth A, Robert W Fairlie, and Huanan Xu. 2022. “The evolving impacts of the COVID-19 pandemic on
gender inequality in the US labor market: The COVID motherhood penalty.” Economic Inquiry 60 (2): 485–507.
Cowan, Benjamin. 2024. “Time use, college attainment, and the working-from-home revolution.” Journal of Popula-
tion Economics 37 (3): 60.
Cowan, Benjamin, Todd R Jones, and Jeffrey Swigert. 2024. “Parental and student time use around the academic year.”
Journal of Economic Behavior & Organization 224: 66–110.
Cowan, Benjamin W. 2023. “Time Use, College Attainment, and The Working-from-Home Revolution.” Working
Paper 31439, National Bureau of Economic Research. URL http://www.nber.org/papers/w31439.
Dettling, Lisa J. 2017. “Broadband in the Labor Market: The Impact of Residential High-Speed Internet on Married
Women’s Labor Force Participation.” Industrial and Labor Relations Review 70.
38
Dingel, Jonathan I and Brent Neiman. 2020. “How Many Jobs Can be Done at home?” Journal of Public Economics
189: 104235.
Dylan Maloney. 2019. “United States Census Bureau Effects of the COVID-19 Pandemic on Private and Pub-
lic Employment.” URL https://www.census.gov/library/stories/2024/11/pandemic-effects-on-
employment.html.
Faberman, R Jason, Andreas I Mueller, and Ayşegül Şahin. 2022. “Has the willingness to work fallen during the covid
pandemic?” Labour Economics 79: 102275.
FCC. 2019. “Federal Communications Commission 2019 broadband deployment report.” URL https://
docs.fcc.gov/public/attachments/DOC-357699A1.pdf.
FCC. 2021. “Fixed Broadband Deployment Data from FCC Form 477.” URL https://www.fcc.gov/general/
broadband-deployment-data-fcc-form-477.
———. 2024. “2024 SECTION 706 REPORT.” Federal Communications Commission. URL https://
docs.fcc.gov/public/attachments/FCC-24-27A1.pdf.
FCC. 2024. “Federal Communications Commission Broadband Data Task Force Seeks Comment on the Broad-
band Data Collection Challenge Process.” URL https://docs.fcc.gov/public/attachments/DA-24-
64A1_Rcd.pdf.
Federal Communications Commission. 2025a. “Form 477 County Data on Internet Access Services.” URL https:
//www.fcc.gov/form-477-county-data-internet-access-services.
———. 2025b. “Urban Rate Survey Data Resources.” URL https://www.fcc.gov/economics-analytics/
industry-analysis-division/urban-rate-survey-data-resources.
Garcia, Kairon Shayne D. and Benjamin W. Cowan. 2022. “The Impact of U.S. School Closures on Labor Market
Outcomes during the COVID-19 Pandemic.” Working Paper 29641, National Bureau of Economic Research. URL
http://www.nber.org/papers/w29641.
Goda, Gopi Shah and Evan J Soltas. 2023. “The impacts of Covid-19 absences on workers.” Journal of Public
Economics 222: 104889.
Goldin, Claudia. 2022. “Understanding the Economic Impact of COVID-19 on Women.” Working Paper 29974,
National Bureau of Economic Research. URL http://www.nber.org/papers/w29974.
Gonzalez, Lisa. 2012. “Explaining Arkansas’ Changed Barriers to Community Broadband.” URL https://
ilsr.org/articles/explaining-arkansas-changed-barriers-to-community-broadband/.
———. 2019. “Arkansas Considers Lifting Ban on Municipal Broadband Networks.” URL https://ilsr.org/
articles/arkansas-considers-lifting-ban-on-municipal-broadband-networks/.
Greenwood, Jeremy, Ananth Seshadri, and Mehmet Yorukoglu. 2005. “Engines of liberation.” The Review of Economic
Studies 72 (1): 109–133.
Gupta, Sumedha, Kosali Simon, and Coady Wing. 2020. “Mandated and voluntary social distancing during the COVID-
19 epidemic.” Brookings Papers on Economic Activity 2020 (2): 269–326.
39
Hansen, Benjamin, Joseph J Sabia, and Jessamyn Schaller. 2024. “Schools, job flexibility, and married women’s labor
supply.” Journal of Human Resources .
Heggeness, Misty, Palak Suri et al. 2021. “Telework, childcare, and mothers’ labor supply.” Opportunity and Inclusive
Growth Institute Working Papers 52.
ILSR. 2023. “Deadline to Challenge FCC Maps Highlights Failed Process.” URL https://ilsr.org/articles/
fcc-mapping-challenge-deadline-highlights-failed-process/.
ITU. 2003. “Asymmetric Digital Subscriber Line (ADSL) transceivers – Extended bandwidth ADSL2(ADSL2+).”
URL https://www.itu.int/rec/T-REC-G.992.5/en.
Jones, Kelly and Britni Wilcher. 2024. “Reducing maternal labor market detachment: A role for paid family leave.”
Labour Economics 87: 102478.
Kaiser Family Foundation. 2021. “State Actions to Mitigate the Spread of COVID-19.” https://www.kff.org/
other/state-indicator/state-actions-to-mitigate-the-spread-of-covid-19/?activeTab=
map¤tTimeframe=0&selectedDistributions=status-of-reopening&sortModel=%7B%22colId%
22:%22Location%22,%22sort%22:%22asc%22%7D.
Lending Tree. 2025. “Households Face Rising Utility Costs — These Metros Have the Highest Bills.” URL
https://www.lendingtree.com/debt-consolidation/utility-costs-study/#households-spend-4-
3-of-median-income-on-utilities.
Lide, Casey and Thomas B Magee. 2022. “Using Electric Utility Easements for Broadband.” URL https:
//www.beyondtelecomlawblog.com/using-electric-utility-easements-for-broadband/.
LoPiccalo, Katherine. 2022. “Impact of broadband penetration on US Farm productivity: A panel approach.” Telecom-
munications Policy 46 (9): 102396.
Lozano Rojas, Felipe, Xuan Jiang, Laura Montenovo, Kosali I Simon, Bruce A Weinberg, and Coady Wing. 2020.
“Is the Cure Worse than the Problem Itself? Immediate Labor Market Effects of COVID-19 Case Rates and School
Closures in the U.S.” Working Paper 27127, National Bureau of Economic Research. URL http://www.nber.org/
papers/w27127.
Lutz, Ashley. 2025. “The new American workplace crisis: Return-to-office mandates lead to a working mom
exodus.” Fortune URL https://fortune.com/2025/08/11/working-moms-leave-american-workforce-
return-to-office/.
Marcattilio, Ry. 2021. “Arkansas Takes a Huge Step Forward in Removing Barriers to Municipal
Broadband.” URL https://communitynetworks.org/content/arkansas-takes-huge-step-forward-
removing-barriers-municipal-broadband.
Microsoft. 2024. “Prepare your organization’s network for Microsoft Teams.” URL https://
learn.microsoft.com/en-us/microsoftteams/prepare-network#bandwidth-requirements.
Morton, Heather. 2021. “Broadband 2020 Legislation.” URL https://www.ncsl.org/technology-and-
communication/broadband-2020-legislation.
40
Ne’eman, Ari and Nicole Maestas. 2023. “How Does Medicaid Expansion Impact Income Support Program Partici-
pation and Employment for Different Types of People with Disabilities?” Tech. rep., National Bureau of Economic
Research.
Nieto, Adrián. 2023. “Television and the Labor Supply: Evidence from the Digital Television Transition in the UK.”
Journal of Human Resources .
NTIA. 2022. “Broadband Equity, Access, and Deployment Program.” URL https://broadbandusa.ntia.doc.gov/
sites/default/files/2022-05/BEAD%20NOFO.pdf.
Parker, Christine. 2024. “ILSR Unveils Community Networks Predictive Map.” URL https:
//communitynetworks.org/content/new-resource-alert-ilsr-unveils-community-networks-
predictive-map.
Parolin, Zachary and Emma Lee. 2022. “U.S. School Closure Distance Learning Database.” URL https://osf.io/
tpwqf/.
Pew. 2019. “How State Policy Shapes Broadband Deployment.” URL https://www.pew.org/-/media/assets/
2019/12/bri_how_state_policy_shapes_broadband_deployment_v3.pdf.
Pew Charitable Trusts. 2021. “Internet/Broadband Fact Sheet.” URL https://www.pewresearch.org/internet/
fact-sheet/internet-broadband/?menuItem=6b886b10-55ec-44bc-b5a4-740f5366a404.
Pew Charitable Trusts. 2021. “State Broadband Policy Explorer.” URL https://www.pew.org/en/research-and-
analysis/data-visualizations/2019/state-broadband-policy-explorer.
Republicans, Energy and Commerce Committee. 2021. “EC GOP Leaders Unveil the Boosting Broadband Con-
nectivity Agenda.” https://republicans-energycommerce.house.gov/news/press-release/ec-gop-
leaders-unveil-the-boosting-broadband-connectivity-agenda/.
Rosenworcel, Jessica. 2023. “National Broadband Map: It Keeps Getting Better.” URL https://www.fcc.gov/
national-broadband-map-it-keeps-getting-better.
Schroeter, Sofia, Rafael Lalive, and Kalaivani Karunanethy. 2023. “School Closures and Parental Labor Supply:
Differential Effects of Anticipated and Unanticipated Shocks.” Industrial labor relations review 76 (1).
Talbot, David, Kira Hessekiel, and Danielle Kehl. 2018. “Community-Owned Fiber Networks: Value Leaders in
America.” Research publication, Berkman Klein Center. URL https://ssrn.com/abstract=3099626.
Tanberk, Julia. 2020. “United States County Broadband Statistics for 2020.” URL https://broadbandnow.com/
research/county-broadband-statistics.
Trostle, H, Katie Kienbaum, Michelle Andrews, and C Mitchell. 2019. “Cooperatives fiberize rural America: A trusted
model for the Internet era.” Institute for Local Self-Reliance .
University of Chicago, Data Science Institute. 2022. “How Much Internet Does Video Conferencing
Need?” URL https://internetequity.cs.uchicago.edu/data-story/how-much-internet-does-
video-conferencing-need/.
U.S. Bureau of Labor Statistics. 2022. “Effects of COVID-19 Pandemic on the Employment Situation News Release
and Data.”
41
U.S. Government Accountability Office. 2023. “Telework: Growth Supported Economic Activity During the Pan-
demic, but Future Impacts Are Uncertain — gao.gov.” https://www.gao.gov/products/gao-23-105999.
Van Parys, Jessica and Zach Y Brown. 2024. “Broadband Internet Access and Health Outcomes: Patient and Provider
Responses in Medicare.” International Journal of Industrial Organization 95.
Viollaz, Mariana and Hernan Jorge Winkler. 2020. “Does the Internet Reduce Gender Gaps? The Case of Jordan.”
Policy Research Working Paper Series 9183, The World Bank. URL https://ideas.repec.org/p/wbk/wbrwps/
9183.html.
Vo, Brian and Clara Miller. 2025. “Financing Broadband in Hard-to-Reach Communities.” Article, Stanford Social
Innovation Review. URL https://ssir.org/articles/entry/financing-broadband-digital-divide#:
~:text=Broadband%20deserts%20exist%20largely%20because,and%20low%2Dincome%20communities%
20behind.
Whitacre, Brian and Roberto Gallardo. 2020. “State broadband policy: Impacts on availability.” Telecommunications
Policy 44 (9).
Whitacre, Brian, Roberto Gallardo, and Sharon Strover. 2014. “Broadband’s contribution to economic growth in rural
areas: Moving towards a causal relationship.” Telecommunications Policy 38 (11).
Zhone. 2024. “The City of Cabot and Connect2First Bridge Digital Divide in Arkansas with Help from DZS
and Irby.” https://zhone.com/news/news-releases/cabot-and-connect2first-bridge-digital-
divide-with-help-from-dzs-and-irby/.
42
Figures
Figure 1: States with Restrictions on Municipal Broadband (2018)
States with Restrictions on Municipal Broadband
Not Restricted
Restricted
Source: Pew State Broadband Policy Explorer
43
Number of Restrictions
1 2 3
19
91
19
92
19
93
19
94
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
Source: Pew State Broadband Policy Explorer, 1999–2022
04
20
05
20
44 06
20
07
20
08
20
09
20
10
20
11
20
12
20
13
20
14
20
15
20
16
20
17
20
Figure 2: Number of Municipal Restrictions Passed in State Legislatures
18
20
19
20
20
20
21
20
22
Figure 3: Impact of Restrictions on FCC County Broadband Access (100Mbps or Higher)
.06
.04
Effects on probability
.02
0
−.02
−.04
0 0 0+
40 60 60
to to
0 0
40
Connections per 1,000 Households (Ordinal)
Notes. Figure plots the marginal effect of restrictions on the probability of the number of connections per 1,000 households. The
dependent variable is an ordinal (0, 1 or 2) biannual value for FCC County Broadband Access reflecting connections per 1,000
housing units in a county with 25Mbps download or higher, where 0 is 0 to 400, 1 is 400 to 600, and 2 is 600+. Model controls
include time-varying log of county population, log density,population with a bachelor’s degree or higher, lagged log employment,
lagged log of household median income, population below poverty income, 2017 broadband access, year-month and state fixed
effects. Robust standard errors are clustered at the county level. Vertical bars are 95% confidence intervals. Source: FCC Form 477
County Data on Internet Access Services and American Community Survey 5-Year Data (2018-2022).
45
Figure 4: Impact of Restrictions on FCC County Broadband Access (25Mbps or Higher)
.03
.02
Effects on probability
.01
0
−.01
−.02
0 0 0+
40 60 60
to to
0 0
40
Connections per 1,000 Households (Ordinal)
Notes. Figure plots the marginal effect of restrictions on the probability of the number of connections per 1,000 households. The
dependent variable is an ordinal (0, 1 or 2) biannual value for FCC County Broadband Access reflecting connections per 1,000
housing units in a county with 25Mbps download or higher, where 0 is 0 to 400, 1 is 400 to 600, and 2 is 600+. Model controls
include time-varying log of county population, log density,population with a bachelor’s degree or higher, lagged log employment,
lagged log of household median income, population below poverty income, 2017 broadband access, year-month and state fixed
effects. Robust standard errors are clustered at the county level. Vertical bars are 95% confidence intervals. Source: FCC Form 477
County Data on Internet Access Services and American Community Survey 5-Year Data (2018-2022).
46
Figure 5: Event Study Estimates of Municipal Broadband Restrictions on FCC County Broadband Access
(100Mbps or Higher)
.05
0
−.05
−.1
18 18 19 19 20 20 21 21 22 22
20 20 20 20 20 20 20 20 20 20
Ju
n Dec
n
Ju Dec
n
Ju Dec
n
Ju Dec
n
Ju Dec
Notes. The dependent variable is a dummy that equals 1 if FCC County Broadband Access is 20% or higher. The figure plots OLS
coefficients of the dynamic effects for states with municipal broadband restrictions compared with states without restrictions over
time on broadband access. Model controls include time-varying log of county population, log density, population with a bachelor’s
degree or higher, lagged log employment, lagged log of household median income, population below poverty income, year-month
and county fixed effects. Robust standard errors are clustered at the county level. Vertical bars are 95% confidence intervals. Source:
FCC Form 477 County Data on Internet Access Services and American Community Survey 5-Year Data (2018-2022).
47
Figure 6: Event Study Estimates of Municipal Broadband Restrictions on FCC County Broadband Access
(25Mbps or Higher)
.02
0
−.02
−.04
−.06
−.08
18 18 19 19 20 20 21 21 22 22
20 20 20 20 20 20 20 20 20 20
Ju
n Dec
n
Ju Dec
n
Ju Dec
n
Ju Dec
n
Ju Dec
Notes. The dependent variable is a dummy that equals 1 if FCC County Broadband Access is 60% or higher. The figure plots OLS
coefficients of the dynamic effects for states with municipal broadband restrictions compared with states without restrictions over
time on broadband access. Model controls include time-varying log of county population, log density, population with a bachelor’s
degree or higher, lagged log employment, lagged log of household median income, population below poverty income, year-month
and county fixed effects. Robust standard errors are clustered at the county level. Vertical bars are 95% confidence intervals. Source:
FCC Form 477 County Data on Internet Access Services and American Community Survey 5-Year Data (2018-2022).
48
Figure 7: Event Study Estimates of Municipal Broadband Restrictions on FCC County Broadband Access
(10Mbps or Higher)
.02
0
−.02
−.04
−.06
18 18 19 19 20 20 21 21 22 22
20 20 20 20 20 20 20 20 20 20
Ju
n Dec
n
Ju Dec
n
Ju Dec
n
Ju Dec
n
Ju Dec
Notes. The dependent variable is a dummy that equals 1 if FCC County Broadband Access is 60% or higher. The figure plots OLS
coefficients of the dynamic effects for states with municipal broadband restrictions compared with states without restrictions over
time on broadband access. Model controls include time-varying log of county population, log density, population with a bachelor’s
degree or higher, lagged log employment, lagged log of household median income, population below poverty income, year-month
and county fixed effects. Robust standard errors are clustered at the county level. Vertical bars are 95% confidence intervals. Source:
FCC Form 477 County Data on Internet Access Services and American Community Survey 5-Year Data (2018-2022).
49
Figure 8: Number of Plans in each State
30
25
Number of Plans
Not Restricted
Restricted
20
15
2017 2018 2019 2020 2021 2022
Year
Notes. Figure plots the raw trends for the count of number of plans with at least 25/3 Mbps Speeds for Fiber to the Home, Cable
and DSL broadband, unlimited usage and costing $80 or less in total monthly charges for each state-year. Source: FCC Urban Rate
Survey for Fixed Broadband Service (2019-2023, reflecting data from 2018-2022).
Figure 9: Average Plan Affordability as Percentage of Monthly Income
2
Average Total Monthly Cost of Plan
1.8
Not Restricted
1.6
Restricted
1.4
1.2
2017 2018 2019 2020 2021 2022
Year
Notes. Figure plots the raw trends for percent of monthly household income spent on total monthly charges on average for plans
with at least 25/3 Mbps Speeds for Fiber to the Home, Cable or DSL broadband with unlimited usage for each year. Estimates
are weighted by sample weights. Source: FCC Urban Rate Survey for Fixed Broadband Service (2018-2023, reflecting data from
2017-2022).
50
Figure 10: Event Study Estimates of Municipal Broadband Restrictions on Labor Force Outcomes for Mar-
ried Women with Children
(a) Labor Force Participation
.05
0
change in LFP
-.05
-.1
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Quarters since Q1 2020
(b) Employment
.05
change in Employment
0
-.05
-.1
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Quarters since Q1 2020
51
(c) Employed, At Work
.05
change in Employed, at work
0
-.05
-.1
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Quarters since Q1 2020
Notes: The dependent variables are a dummy for being in the labor force (Panel a), being employed (Panel b) and being employed,
at work (Panel c) for prime-age married women with children. All models include state fixed effects, region-by-quarter fixed
effects and Bartik shift-share variables. Individual control variables include age, race (Black Non-Hispanic, other Non-Hispanic,
and Hispanic), education (less than high school, some college, bachelor’s degree, advanced degree), metro area residency, MSA or
CBSA size, marital status, number of children in the household and an indicator for at least one child under 13. County COVID-
19 controls include a continuous measure of cumulative cases and deaths per capita (per 1M residents) along with a control for
school closures. Observations are weighted using CPS sample weights. Robust standard errors are clustered at the state-level.
95% confidence intervals are shown.
Source:Monthly Current Population Survey, 2018–2022.
52
Tables
Table 1:
County-Level Summary Statistics, FCC Measures of Broadband Subscribership and Competition (2018-19)
Non-Restriction Restriction Difference
mean/sd mean/sd b/t
% with 60%+ households at 100MBps 3.72 2.67 1.05∗∗
(18.94) (16.12) (3.20)
% with 60%+ households at 25 MBps 12.53 12.46 0.06
(33.10) (33.03) (0.10)
% with 60%+ households at 10 MBps 25.19 23.91 1.27
(43.41) (42.66) (1.58)
% 0 Providers 100Mbps 30.83 35.15 -4.32∗∗∗
(27.40) (30.08) (-8.03)
% 0 Providers 25Mbps 22.48 26.45 -3.97∗∗∗
(21.14) (24.17) (-9.36)
% 0 Providers 10Mbps 14.34 16.93 -2.59∗∗∗
(15.31) (17.99) (-8.29)
% 2+ Providers 100Mbps 18.65 18.35 0.30
(24.56) (24.55) (0.66)
% 2+ Providers 25Mbps 28.72 27.06 1.66∗∗
(27.46) (27.86) (3.21)
% 2+ Providers 10Mbps 45.10 42.07 3.03∗∗∗
(28.93) (29.12) (5.59)
N 5,828 5,616 11,444
Notes. The first two columns display the means with standard deviations in parentheses for a select set of variables for states
without and with restrictions on municipal broadband, respectively. The last column shows the difference in the variables
by state restriction status with t-statistics in parentheses. All broadband estimates are biannual county percentages (0-100%)
between June 2018 and December 2019. In rows 1 to 3, we consider a county covered by broadband for a given speed if the FCC
County-Level Internet Access Services Speed Tier value indicates that over 60% of households are actually connected at that
speed. The remaining rows indicate the population percentage of county covered by the number of fixed residential broadband
providers if broadband is available at that speed – we use “acfo’ for 10+/1+ speeds and only “cfo” for 25+/3+ speeds. a=ADSL,
c=Cable, f=fiber, o=other. Source: County-Level Internet Access Services Speed Tier Data from FCC Form 477 (June 2018-
December 2019), FCC Area Tables (June 2018- December 2019). Means and (Standard Deviations)
53
Table 2:
County-Level Summary Statistics, ACS5 Measures of Socio-Demographic Covariates (2018-19)
Non-Restriction Restriction Difference
County Population 91534.58 98127.53 -6592.95
(262889.01) (255344.57) (-1.36)
Pop/Sq Mile 312.70 233.76 78.94∗
(2458.47) (736.92) (2.31)
Median Age 41.15 41.51 -0.37∗∗∗
(4.84) (5.59) (-3.74)
Percent Female 50.02 49.89 0.13∗∗
(2.02) (2.54) (3.08)
Percent Black 8.01 10.59 -2.58∗∗∗
(14.24) (14.76) (-9.52)
Percent Hispanic 7.42 10.40 -2.98∗∗∗
(10.57) (15.66) (-11.95)
Percent College or More 15.21 14.64 0.57∗∗∗
(6.66) (6.12) (4.78)
Percent Employed 44.63 43.81 0.82∗∗∗
(6.49) (6.12) (6.98)
Median Income 52952.43 51819.67 1132.75∗∗∗
(14055.37) (13111.06) (4.45)
Percent below Poverty Income 14.67 14.66 0.01
(6.31) (5.50) (0.07)
N 2,914 2,808 5,722
Notes. The first two columns display the means with standard deviations in parentheses for a select set of variables for states
without and with restrictions on municipal broadband, respectively. The last column shows the difference in the variables by
state restriction status with t-statistics in parentheses. All demographic estimates are annual county percentages (0-100%).
Source: County-Level American Community Survey 5-Year Data (2018-19). Means and (Standard Deviations).
54
Table 3: Summary Statistics, CPS Labor Force and Socio-Demographic Measures for All Prime-Age Work-
ers 2018–2019
Non-Restriction Restriction Difference
Labor Force Participation 0.805 0.801 0.004∗∗∗
(0.396) (0.399) (20.07)
Employed, Any 0.777 0.776 0.001
(0.416) (0.417) (2.42)
Employed, At Work 0.756 0.755 0.001
(0.429) (0.430) (0.96)
Hours Worked Last Week 40.832 40.864 -0.032
(11.774) (11.692) (1.11)
Not in Labor Force, Family Reasons 0.089 0.095 -0.006∗∗∗
(0.285) (0.294) (90.82)
Part-Time/Not Working, Family Reasons 0.097 0.103 -0.006∗∗∗
(0.296) (0.304) (68.58)
Part-Time, Childcare Reasons 0.008 0.007 0.001∗∗∗
(0.087) (0.083) (11.61)
Less than High-School 0.074 0.085 -0.011∗∗∗
(0.262) (0.279) (337.59)
High-School 0.630 0.669 -0.039∗∗∗
(0.483) (0.471) (1317.88)
Some College 0.261 0.283 -0.022∗∗∗
(0.439) (0.450) (453.50)
College degree 0.251 0.232 0.019∗∗∗
(0.434) (0.422) (402.47)
Advanced Degree 0.119 0.099 0.020 ∗∗∗
(0.323) (0.299) (789.85)
Age 39.054 39.101 -0.047∗∗
(8.805) (8.754) (5.60)
Black 0.131 0.151 -0.020∗∗∗
(0.338) (0.358) (600.56)
White 0.618 0.584 0.034∗∗∗
(0.486) (0.493) (968.00)
Hispanic 0.159 0.196 -0.037∗∗∗
(0.366) (0.397) (1812.17)
Other Race 0.091 0.070 0.021∗∗∗
(0.288) (0.255) (1271.83)
Number of Own Children 1.079 1.098 -0.019∗∗∗
(1.254) (1.271) (42.30)
In Central City 0.286 0.275 0.011∗∗∗
(0.452) (0.447) (110.43)
Teleworkable 0.123 0.122 0.001
(0.328) (0.327) (0.57)
N 445,268 354,090 799,358
Notes: The table shows summary statistics for all prime-age workers (25-54 years old) in states with (column
2) and without (column 1) restrictions on municipal broadband, measured prior to the COVID-19 pandemic
in 2018-2019. Means with standard deviations reported in parentheses. Observations are weighted using CPS
individual sample weights. Hours are estimated conditional on employment. In column 3, the difference in
variables is reported along with F-statistics in parentheses.
Source: CPS Monthly Estimates 2018–2019
55
Table 4:
Impact of Restrictions on FCC County Broadband Subscribership by Download Speeds
(1) (2) (3) (4) (5) (6)
100Mbps 100Mbps 25Mbps 25Mbps 10Mbps 10Mbps
3 Categories 20% or More 3 Categories 60% or More 3 Categories 60% or More
Ologit LPM Ologit LPM Ologit LPM
Restrictions x Post -0.455∗∗∗ -0.027∗∗ -0.207∗∗ -0.021∗ -0.023 -0.025∗∗
(0.102) (0.012) (0.086) (0.012) (0.083) (0.012)
R2 0.524 0.755 0.562 0.699 0.578 0.720
Mean 0.433 0.611 0.817 0.254 1.192 0.408
N 28,592 28,592 28,592 28,592 28,592 28,592
Notes. All outcomes are county-level biannual values for broadband penetration (actual number of connections) between June
2018 and December 2022. In columns 1 and 2, broadband penetration is defined as a minimum download speed of 100Mbps,
in columns 3 and 4 it is 25Mbps, in columns 5 and 6 it is 10Mbps. The dependent variable in columns 1, 3 and 5 is an ordinal
value (0, 1 or 2) for FCC County broadband penetration reflecting actual connections per housing units in a given county
(corresponding to 0 to 40%, 40 to 60% and over 60% subscribership respectively) and is regressed using an ordered logistic
model (Ologit). For columns 2, 4 and 6, the dependent variable is a dummy that equals 1 if FCC County broadband penetration
is 20% or higher, 60% or higher and 60% or higher respectively and is regressed using a Linear Probability Model (LPM).
All specifications include time-varying log of county population, log density (county population divided by land area), percent
population with a bachelor’s degree or higher, lagged log of employment, lagged log of household median income, percent
population below poverty income, 2017 broadband penetration, year-month and state fixed effects. LPM models additionally
include county fixed-effects. Standard errors are clustered at the county-level. Source: FCC Form 477 County Data on Internet
Access Services and American Community Survey 5-Year Data (2018-2022).
56
Table 5:
The Impact of Restrictions on County Broadband Subscribership - The Role of Competition
(1) (2) (3)
100Mbps 25Mbps 10Mbps
∗∗∗
Restrictions 0.0123 -0.0928 -0.0480∗∗∗
(0.0135) (0.0118) (0.0135)
% Pop. Coverage by 2+ providers 0.0025∗∗∗ -0.0006∗ 0.0016∗∗∗
(0.0004) (0.0003) (0.0004)
Restrictions x 2+ Coverage -0.0007∗∗∗ 0.0026∗∗∗ 0.0006∗∗
(0.0002) (0.0003) (0.0003)
2
R 0.826 0.759 0.792
N 20,030 20,030 20,030
Notes. All outcomes are county-level biannual values for broadband penetration (actual number of connections) between June
2018 and December 2022. In columns 1, 2 and 3 respectively, broadband penetration is defined as a minimum download speed of
100Mbps, 25Mbps or 10Mbps. The dependent variable is a dummy that equals 1 if FCC County broadband penetration is 20% or
higher, 60% or higher and 60% or higher respectively and is regressed using a Linear Probability Model (LPM). The independent
variable in row 1 is a dummy that equals 1 for states with restrictions after 2020. Row 2 provides the coefficient estimates on
the variable for the percentage of county population (0-100%) covered by 2 or more providers. In row 3, the independent
variable (for restrictions) is interacted with the percentage of population covered by 2 or more providers. All specifications
include time-varying log of county population, log density, percent population with a bachelor’s degree or higher, lagged log of
county employment, lagged log of household median income, percent population below poverty income, county-fixed effects,
year-month and state fixed effects. Standard errors are clustered at the county-level. Source: FCC Form 477 County Data on
Internet Access Services and American Community Survey 5-Year Data (2018-2022). Percentage of population covered by 2
or more providers comes from FCC Area Tables (2018-21).
57
Table 6:
Do Higher Prices Serve as a Channel for the Impact of Restrictions on County Broadband Subscribership?
(1) (2) (3)
100Mbps 25Mbps 10Mbps
Restrictions -0.022∗ -0.014 -0.016
(0.012) (0.012) (0.013)
Restrictions x Over 2 percent of Budget on Broadband -0.035∗∗ -0.051∗∗∗ -0.062∗∗∗
(0.018) (0.017) (0.020)
2
R 0.755 0.699 0.721
N 28,592 28,592 28,592
Notes. All outcomes are county-level biannual values for broadband penetration (actual number of connections) between June
2018 and December 2022. In columns 1, 2 and 3 respectively, broadband penetration is defined as a minimum download speed
of 100Mbps, 25Mbps or 10Mbps. The dependent variable is a dummy that equals 1 if FCC County broadband penetration is
20% or higher, 60% or higher and 60% or higher respectively and is regressed using a Linear Probability Model (LPM). The
independent variable in row 1 is a dummy that equals 1 for states with restrictions on municipal broadband after 2020 and zero
otherwise. In row 2, the same independent variable is interacted with an indicator that equals one if the county lowest price
for broadband constitutes over 2% of the county’s median monthly household income. All specifications include time-varying
log of county population, log population density, percent population with a bachelor’s degree or higher, lagged log of county
employment, percent population below poverty income, county-fixed effects, year-month and state fixed effects. Standard errors
are clustered at the county-level. Source: FCC Form 477 County Data on Internet Access Services and American Community
Survey 5-Year Data (2018-2022). County-level measures of lowest monthly prices from BroadbandNow’s United States County
Broadband Statistics for 2020. Lowest price is the average of all lowest priced broadband speed plans available in the county.
Note that because BroadbandNow only provides prices for 2020, the dummy for prices is absorbed as a county-fixed effect.
58
Table 7:
Impact of Restrictions on CPS Internet Use by Subgroup
(1) (2) (3) (4) (5)
Home Outside Job
Internet Home Email Phone Search
Panel A: All Prime-Age Individuals
Restrictions ×Post -0.0201* -0.0168** -0.0116* 0.0152 -0.0089
(0.0108) (0.0082) (0.0058) (0.0128) (0.0092)
M ean 0.812 0.736 0.932 0.607 0.235
R2 0.043 0.073 0.071 0.123 0.029
N 122,576 122,576 55,870 55,870 55,870
Panel B: Married Women with Children
Restrictions ×Post -0.0076 0.0031 0.0093 -0.0340* 0.0195
(0.0149) (0.0106) (0.0109) (0.0199) (0.0173)
M ean 0.841 0.754 0.953 0.675 0.206
R2 0.036 0.075 0.096 0.116 0.025
N 21,574 21,574 8,487 8,487 8,487
Panel C: Women without Children
Restrictions ×Post -0.0244** -0.0232* -0.0012 0.0293 -0.0030
(0.0118) (0.0127) (0.0121) (0.0239) (0.0133)
M ean 0.803 0.724 0.937 0.615 0.249
R2 0.044 0.081 0.074 0.135 0.040
N 30,185 30,185 13,964 13,964 13,964
Panel D: Married Men with Children
Restrictions ×Post -0.0120 -0.0080 -0.0190 0.0098 -0.0218
(0.0135) (0.0197) (0.0133) (0.0196) (0.0153)
M ean 0.844 0.785 0.939 0.635 0.200
R2 0.041 0.072 0.098 0.124 0.027
N 21,722 21,722 9,179 9,179 9,179
Notes. This table reports estimates from difference-in-differences models that compare workers in states with restrictions on
municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic. The outcomes in
columns 1 through 5 are respectively - a dummy for whether or not a person accesses the internet from home, a dummy for
whether or not a person accesses the internet outside of the home, whether or not a respondent uses the internet for email,
whether or not a person used the internet for telephone calls and whether or not a person has used the internet to search for
jobs. The treatment variable is a binary variable for persons (aged 25 to 54) living in states that had municipal restrictions as of
Q1:2018. CPS data are from the Computer and Internet Use Supplement for the month of November in 2017, 2019 (pre-COVID)
and 2021, 2023 (post-COVID). All models include state fixed effects and region-by-quarter fixed effects. Individual control
variables include age, race (Black Non-Hispanic, other Non-Hispanic, and Hispanic), education (less than high school, some
college, bachelor’s degree, advanced degree), metro area residency, MSA or CBSA size, marital status, number of children
in the household and an indicator for at least one child under 13. Area COVID-19 controls include a continuous measure of
cumulative cases and deaths per capita (per 1M residents) along with a control for school closures. Observations are weighted
using Computer and Internet Use Supplement Weights or Computer and internet use supplement primary respondent weights
as appropriate. Robust standard errors are clustered at the state-level. Significance at 1%, 5%, and 10% levels are indicated by
***, **, and *, respectively.
59
Table 8: Effects of State-Level Restrictions on Municipal Broadband on Labor Force Outcomes by Sub-
Group
(1) (2) (3) (4)
Labor Force Employed, Employed, Hours
Participation At Work Any Last Week
Panel A: Married Women with Children
Restrictions ×Post -0.0268∗∗ -0.0310∗∗ -0.0305∗∗ 0.1789
(0.0121) (0.0132) (0.0125) (0.2067)
M ean 0.670 0.627 0.655 36.552
R2 0.078 0.074 0.080 0.027
N 222,009 222,009 222,009 139,138
Panel B: Women without Children
Restrictions ×Post 0.0041 0.0036 0.0041 -0.1340
(0.0062) (0.0063) (0.0064) (0.2083)
M ean 0.752 0.707 0.728 39.330
R2 0.069 0.063 0.068 0.021
N 309,196 309,196 309,196 218,528
Panel C: Married Men with Children
Restrictions ×Post -0.0028 -0.0082 -0.0093∗ -0.2065
(0.0041) (0.0050) (0.0046) (0.1561)
M ean 0.947 0.908 0.930 44.035
R2 0.016 0.017 0.020 0.021
N 224,011 224,011 224,011 203,372
Notes: This table reports estimates from difference-in-differences models that compare workers in states
with restrictions on municipal broadband to workers in states without restrictions before and after the
COVID-19 pandemic. The outcome in column 4 is hours worked last week conditional on working at all.
The treatment variable is a binary variable for persons (aged 25 to 54) living in states that had municipal
restrictions as of Q1:2018. CPS data are for all months in 2018,19 (pre-COVID) and August 2021 to
December 2022 (post-COVID). All models include state fixed effects, region-by-quarter fixed effects
and Bartik shift-share variables. Individual control variables include age, race (Black Non-Hispanic,
other Non-Hispanic, and Hispanic), education (less than high school, some college, bachelor’s degree,
advanced degree), metro area residency, MSA or CBSA size, marital status, number of children in the
household and an indicator for at least one child under 13. Area COVID-19 controls include a continuous
measure of cumulative cases and deaths per capita (per 1M residents) along with a control for school
closures. Observations are weighted using CPS sample weights. Robust standard errors are clustered at
the state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
60
Table 9:
Effects of Municipal Restrictions on Labor Force Status (Family Reasons)
(1) (2) (3) (4)
Part-Time NILF, Part-Time,
or Not Working, Taking Care of Child-Care
Part-Time Family Reasons House or Family Problems
Panel A: Married Women with Children
Restrictions ×Post -0.0089 0.0233** 0.0246** -0.0010
(0.0078) (0.0115) (0.0113) (0.0027)
M ean 0.286 0.314 0.289 0.024
R2 0.029 0.086 0.086 0.008
N 139,138 222,009 222,009 222,009
Panel B: Women without Children
Restrictions ×Post 0.0013 -0.0068 -0.0066 -0.0002
(0.0069) (0.0044) (0.0041) (0.0005)
M ean 0.190 0.087 0.085 0.002
R2 0.017 0.056 0.056 0.002
N 218,528 309,196 309,196 309,196
Panel C: Married Men with Children
Restrictions ×Post 0.0066 0.0028 0.0025 0.0004
(0.0044) (0.0030) (0.0027) (0.0006)
M ean 0.089 0.020 0.017 0.003
R2 0.009 0.005 0.005 0.002
N 203,372 224,011 224,011 224,011
Notes: “Part-Time” is defined as working less than 35 hours per week. “Part-Time or Not Working, Family Reasons”
is an indicator for any the following: (i) Main reason not looking for work during last four weeks - “can’t arrange
childcare,” (ii) Major activity (NILF) - “taking care of house or family,” (iii) Reason for working part time last week
- “childcare problems,” or (iv) Reason for absence from work - “childcare problems.” The treatment variable is a
binary variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data
are for all months in 2018,19 (pre-COVID) and August 2021 to December 2022 (post-COVID). All models include
state fixed effects, region-by-quarter fixed effects and Bartik shift-share variables. Individual control variables
include age, race (Black Non-Hispanic, other Non-Hispanic, and Hispanic), education (less than high school, some
college, bachelor’s degree, advanced degree), metro area residency, MSA or CBSA size, marital status, number
of children in the household and an indicator for at least one child under 13. Area COVID-19 controls include a
continuous measure of cumulative cases and deaths per capita (per 1M residents) along with a control for school
closures. Observations are weighted using CPS sample weights. Robust standard errors are clustered at the state-
level. Significance at 1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
61
Table 10: Estimated Effects of Municipal Restrictions on Work-Related Time Use
(1) (2) (3)
Share of Work
WFH from Home Commuting
Panel A: Married Women with Children
Restrictions ×Post -45.8648∗∗ -0.1115∗∗ 6.6612∗
(20.2221) (0.0533) (3.9203)
M ean 72.148 0.226 25.484
R2 0.208 0.222 0.134
N 1,408 1,182 1,408
Panel B: Women without Children
Restrictions ×Post 16.5709 0.0573 5.0953
(23.3363) (0.0403) (4.2571)
M ean 84.357 0.207 30.088
R2 0.231 0.256 0.164
N 1,301 1,124 1,301
Panel C: Married Men with Children
Restrictions ×Post -26.9908 -0.0015 -6.3635
(20.6934) (0.0410) (5.7215)
M ean 72.446 0.168 40.527
R2 0.246 0.267 0.135
N 1,624 1,481 1,624
Notes: All outcomes are measured in minutes spent per workday, except for column 2 which is a share
between 0 and 1. Column 1 is time spent working from home, column 2 is the share of work time
from home (conditional on a positive amount of work), and column 3 is commute time. The sample
includes employed people only. The treatment variable is a dummy for persons (aged 25 to 54) living
in states with municipal restrictions as of Q1:2018. ATUS data are restricted to employed persons
only for all months in 2017-19 (pre-COVID) and August 2021-December 2022 (post-COVID). All
models include state fixed effects, region-by-quarter fixed effects and Bartik shift-share variables.
Individual control variables include age, race (Black Non-Hispanic, other Non-Hispanic, and His-
panic), education (less than high school, some college, bachelor’s degree, advanced degree), metro
area residency, MSA or CBSA size, marital status, number of children in the household and an in-
dicator for at least one child under 13. Area COVID-19 controls include a continuous measure of
cumulative cases and deaths per capita (per 1M residents) along with a control for school closures.
Observations are weighted using CPS sample weights. Robust standard errors are clustered at the
state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
62
Appendix Figures
Appendix Tables
63
Figure A1: Trends in Labor Force Outcomes for Married Women with Children by State Municipal Broad-
band Restrictions, Unadjusted Data
(a) Labor Force Participation
72
Percentage Labor Force Participation
70
No Restrictions
68
66
64
With Restrictions
62
-8 -6 -4 -2 0 2 4 6 8 10 12
Quarters from Q1 2020
(b) Employment
70
Percentage Employed
No Restrictions
65
60
With Restrictions
55
-8 -6 -4 -2 0 2 4 6 8 10 12
Quarters from Q1 2020
64
(c) Employed, At Work
65
No Restrictions
Percentage At Work
60
With Restrictions
55
50
-8 -6 -4 -2 0 2 4 6 8 10 12
Quarters from Q1 2020
Notes: The variables are the percentage of married women with children in the labor force (Panel a), employed (Panel b) and
employed, at work (Panel c) plotted by states with and without restrictions. Observations are weighted using CPS sample weights.
Source:Monthly Current Population Survey, 2018–2022.
65
Figure A2: Event Study Estimates of Municipal Broadband Restrictions on Labor Force Outcomes for
Women without Children
(a) Labor Force Participation
.04
.02
change in LFP
0
-.02
-.04
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Quarters since Q1 2020
(b) Employment
.04
.02
change in Employment
0
-.02
-.04
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Quarters since Q1 2020
66
(c) Employed, At Work
.04
change in Employed, at work
.02
0
-.02
-.04
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Quarters since Q1 2020
Notes: The dependent variables are a dummy for being in the labor force (Panel a), being employed (Panel b) and being employed,
at work (Panel c) for prime-age women without children. All models include state fixed effects, region-by-quarter fixed effects
and Bartik shift-share variables. Individual control variables include age, race (Black Non-Hispanic, other Non-Hispanic, and
Hispanic), education (less than high school, some college, bachelor’s degree, advanced degree), metro area residency, MSA or
CBSA size, marital status, number of children in the household and an indicator for at least one child under 13. County COVID-
19 controls include a continuous measure of cumulative cases and deaths per capita (per 1M residents) along with a control for
school closures. Observations are weighted using CPS sample weights. Robust standard errors are clustered at the state-level.
95% confidence intervals are shown.
Source:Monthly Current Population Survey, 2018–2022.
67
Figure A3: Event Study Estimates of Municipal Broadband Restrictions on Labor Force Outcomes for
Married Men with Children
(a) Labor Force Participation
.05
0
change in LFP
-.05
-.1
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Quarters since Q1 2020
(b) Employment
.04
.02
change in Employment
0
-.02
-.04
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Quarters since Q1 2020
68
(c) Employed, At Work
.04
change in Employed, at work
.02
0
-.02
-.04
-8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11
Quarters since Q1 2020
Notes: The dependent variables are a dummy for being in the labor force (Panel a), being employed (Panel b) and being employed,
at work (Panel c) for prime-age married men with children. All models include state fixed effects, region-by-quarter fixed effects
and Bartik shift-share variables. Individual control variables include age, race (Black Non-Hispanic, other Non-Hispanic, and
Hispanic), education (less than high school, some college, bachelor’s degree, advanced degree), metro area residency, MSA or
CBSA size, marital status, number of children in the household and an indicator for at least one child under 13. County COVID-
19 controls include a continuous measure of cumulative cases and deaths per capita (per 1M residents) along with a control for
school closures. Observations are weighted using CPS sample weights. Robust standard errors are clustered at the state-level.
95% confidence intervals are shown.
Source:Monthly Current Population Survey, 2018–2022.
69
Table A1: Summary Statistics, Married Women with Children 2018–2019
States without Restrictions States with Restrictions
Labor Force Participation 0.664 0.654
(0.472) (0.476)
Employed, Any 0.648 0.638
(0.478) (0.481)
Employed, At Work 0.621 0.613
(0.485) (0.487)
Hours Worked Last Week 36.548 36.511
(11.957) (12.065)
Not in Labor Force, Family Reasons 0.293 0.306
(0.455) (0.461)
Part-Time/Not Working, Family Reasons 0.319 0.328
(0.466) (0.470)
Part-Time, Childcare Reasons 0.026 0.022
(0.159) (0.146)
Less than High-School 0.069 0.081
(0.254) (0.273)
High-School 0.544 0.571
(0.498) (0.495)
Some College 0.256 0.279
(0.437) (0.448)
College degree 0.301 0.292
(0.459) (0.455)
Advanced Degree 0.155 0.137
(0.362) (0.344)
Age 38.869 38.470
(7.242) (7.159)
Black 0.078 0.083
(0.268) (0.276)
White 0.631 0.610
(0.483) (0.488)
Hispanic 0.174 0.213
(0.379) (0.410)
Other Race 0.118 0.094
(0.322) (0.291)
Number of Own Children 2.173 2.201
(1.048) (1.077)
In Central City 0.228 0.227
(0.419) (0.419)
Teleworkable 0.141 0.139
(0.348) (0.346)
N 78,089 63,780
Notes: The table shows summary statistics for prime-age (25-54 years old) married women with children in
states with (column 2) and without (column 1) restrictions on municipal broadband, measured prior to the
COVID-19 pandemic in 2018-2019. Means with standard deviations reported in parentheses. Observations
are weighted using CPS individual sample weights. Hours are estimated conditional on employment.
Source: CPS Monthly Estimates 2018–2019
70
Table A2:
Robustness Check 1– Impact of Municipal Restrictions (States with Municipal Broadband Initiatives) on
Broadband Subscribership
(1) (2) (3) (4) (5) (6)
100Mbps 100Mbps 25Mbps 25Mbps 10Mbps 10Mbps
3 Categories 20% or More 3 Categories 60% or More 3 Categories 60% or More
Ologit LPM Ologit LPM Ologit LPM
Restrictions x Post -0.431∗∗∗ -0.036∗∗∗ -0.267∗∗∗ -0.026∗∗ -0.120 -0.045∗∗∗
(0.107) (0.012) (0.089) (0.012) (0.086) (0.013)
2
R 0.525 0.754 0.560 0.700 0.575 0.719
N 26,256 26,256 26,256 26,256 26,256 26,256
Notes. All outcomes are county-level biannual values for broadband penetration (actual number of connections) between June
2018 and December 2022. In columns 1 and 2, broadband penetration is defined as a minimum download speed of 100Mbps,
in columns 3 and 4 it is 25Mbps, in columns 5 and 6 it is 10Mbps. The dependent variable in columns 1, 3 and 5 is an ordinal
value (0, 1 or 2) for FCC County broadband penetration reflecting actual connections per housing units in a given county
(corresponding to 0 to 40%, 40 to 60% and over 60% subscribership respectively) and is regressed using an ordered logistic
model (Ologit). For columns 2, 4 and 6, the dependent variable is a dummy that equals 1 if FCC County broadband penetration
is 20% or higher, 60% or higher and 60% or higher respectively and is regressed using a Linear Probability Model (LPM). All
specifications include time-varying log of county population, log population density, percent population with a bachelor’s degree
or higher, lagged log of county employment, lagged log of county household median income, percent population below poverty
income, 2017 broadband penetration, year-month and state fixed effects. LPM models additionally include county fixed-effects.
Standard errors are clustered at the county-level. Source: FCC Form 477 County Data on Internet Access Services and American
Community Survey 5-Year Data (2018-2022).
71
Table A3:
Impact of Restrictions on FCC County Broadband Subscribership by Download Speeds – Complete Esti-
mates
(1)
100Mbps
3 Categories
Ologit
Restrictions ×P ost -0.4552∗∗∗
(0.1022)
2017 BB=1 1.0293∗∗∗
(0.3386)
2017 BB=2 2.6436∗∗∗
(0.3566)
2017 BB=3 4.1916∗∗∗
(0.4010)
2017 BB=4 19.8759∗∗∗
(0.8100)
Population 0.5667
(0.5914)
Employment -0.0291
(0.5868)
Density 0.6872∗∗∗
(0.0786)
Percentage College or More 0.0501∗∗∗
(0.0098)
Lagged Ln Median Income 1.3760∗∗∗
(0.4069)
Percent below Poverty Income -0.0395∗∗∗
(0.0152)
State FE Yes
Year-Month FE Yes
2
P seudoR 0.524
Mean 0.433
N 28,592
Notes. This table includes complete regression coefficient estimates from column 1 of Table 4. The outcome is a county-level
biannual value for broadband penetration (actual number of connections) between June 2018 and December 2022. In Column 1,
broadband penetration is defined as a minimum download speed of 100Mbps. The dependent variable in column 1 is an ordinal
value (0, 1 or 2) for FCC County broadband penetration reflecting actual connections per housing units in a given county
(corresponding to 0 to 40%, 40 to 60% and over 60% coverage respectively) and is regressed using an ordered logistic model
(Ologit). The specification includes time-varying log of county population, log density, percent population with a bachelor’s
degree or higher, lagged log employment, lagged log of household median income, percent population below poverty income,
2017 broadband penetration, year-month and state fixed effects. LPM models (not included in this table) additionally include
county fixed-effects. Standard errors are clustered at the county-level. Source: FCC Form 477 County Data on Internet Access
Services and American Community Survey 5-Year Data (2018-2022).
72
Table A4:
Effects of Municipal and/or Cooperative Restrictions on County Broadband Subscribership
(1) (2) (3) (4) (5) (6)
100Mbps 100Mbps 25Mbps 25Mbps 10Mbps 10Mbps
3 Categories 20% or More 3 Categories 60% or More 3 Categories 60% or More
Ologit LPM Ologit LPM Ologit LPM
Muni Only x Post -0.448∗∗∗ -0.028∗∗ -0.185∗ -0.021 -0.014 -0.030∗∗
(0.113) (0.013) (0.095) (0.013) (0.093) (0.014)
Muni+Coop x Post -0.429∗∗∗ -0.027 -0.308∗∗ -0.041∗∗ -0.007 -0.040∗∗
(0.141) (0.017) (0.123) (0.017) (0.120) (0.018)
2
R 0.522 0.755 0.559 0.699 0.574 0.724
Mean 0.443 0.619 0.832 0.259 1.210 0.416
N 27,792 27,792 27,792 27,792 27,792 27,792
Notes. Note that Mississippi is excluded from this table as it is the only state with restrictions on cooperative broadband but
not municipal broadband in 2018. All outcomes are county-level biannual values for broadband penetration (actual number of
connections) between June 2018 and December 2022. In Columns 1 and 2, broadband penetration is defined as a minimum
download speed of 100Mbps, in columns 3 and 4 it is 25Mbps, in columns 5 and 6 it is 10Mbps. The dependent variable in
columns 1, 3 and 5 is an ordinal value (0, 1 or 2) for FCC County broadband penetration reflecting actual connections per
housing units in a given county (corresponding to 0 to 40%, 40 to 60% and over 60% coverage respectively) and is regressed
using an ordered logistic model (Ologit). For columns 2, 4 and 6, the dependent variable is a dummy that equals 1 if FCC
County broadband penetration is 20% or higher, 60% or higher and 60% or higher respectively and is regressed using a Linear
Probability Model (LPM). All specifications include time-varying log of county population, log density, percent population
with a bachelor’s degree or higher, lagged log employment, lagged log of household median income, percent population below
poverty income, 2017 broadband penetration, year-month and state fixed effects. LPM models additionally include county fixed-
effects. Standard errors are clustered at the county-level. Source: FCC Form 477 County Data on Internet Access Services and
American Community Survey 5-Year Data (2018-2022).
73
Table A5:
Effects of Municipal Restrictions on Labor Force Participation
(1) (2) (3)
Married Women Women Married Men
with Children without Children with Children
Municipal Restrictions -0.0268∗∗ 0.0041 -0.0028
(0.0121) (0.0062) (0.0041)
Age 0.0420∗∗∗ 0.0047∗∗∗ 0.0061∗∗∗
(0.0044) (0.0016) (0.0016)
Age-Squared -0.0005∗∗∗ -0.0001∗∗∗ -0.0001∗∗∗
(0.0001) (0.0000) (0.0000)
High School 0.1085∗∗∗ 0.1850∗∗∗ 0.0137∗∗
(0.0115) (0.0114) (0.0068)
Some College 0.1852∗∗∗ 0.2877∗∗∗ 0.0268∗∗∗
(0.0100) (0.0131) (0.0068)
Bachelor’s Degree 0.2050∗∗∗ 0.3364∗∗∗ 0.0526∗∗∗
(0.0116) (0.0130) (0.0071)
Advanced Degree 0.2699∗∗∗ 0.3469∗∗∗ 0.0543∗∗∗
(0.0123) (0.0121) (0.0074)
Child under 13 -0.0592∗∗∗ 0.0006
(0.0072) (0.0028)
COVID Rate (per 1,000) 0.0004∗∗ 0.0002∗∗∗ -0.0000
(0.0002) (0.0001) (0.0000)
COVID Death Rate (per 1,000) -0.0092 -0.0064∗ -0.0042
(0.0072) (0.0038) (0.0029)
Share of School Closures 0.0576∗∗ -0.0741∗∗ -0.0229
(0.0263) (0.0308) (0.0260)
Bartik Shift-Share 0.0161 -0.0081 -0.0079
(0.0152) (0.0062) (0.0056)
State Fixed Effects Yes Yes Yes
Region-Year-Quarter Fixed Effects Yes Yes Yes
Other Indicators Yes Yes Yes
R2 0.078 0.069 0.016
N 222,009 309,196 224,011
Notes: This table reports estimates from difference-in-differences models that compare workers in states with restrictions on
municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic. The treatment variable
is a binary variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are
for all months in 2018,19 (pre-COVID) and August 2021 to December 2022 (post-COVID). Individual control variables in all
models include age, age-squared, education (less than high school, some college, bachelor’s degree, advanced degree) and an
indicator for at least one child under 13. Other indicators include individual race (Black Non-Hispanic, other Non-Hispanic,
and Hispanic), marital status, number of children in the household, metro area residency and a categorical variable (provided
by the IPUMS CPS) for CBSA size. Area COVID-19 controls include a continuous measure of cumulative cases and deaths
per capita (per 1M residents) along with a control for school closures. All models include state fixed effects, region-by-quarter
fixed effects and Bartik shift-share variables. Observations are weighted using CPS sample weights. Robust standard errors are
clustered at the state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
74
Table A6:
Effects of Municipal Restrictions on Employed, At Work
(1) (2) (3)
Married Women Women Married Men
with Children without Children with Children
Municipal Restrictions -0.0310∗∗ 0.0036 -0.0082
(0.0132) (0.0063) (0.0050)
Age 0.0491∗∗∗ 0.0044∗∗∗ 0.0099∗∗∗
(0.0047) (0.0016) (0.0019)
Age-Squared -0.0006∗∗∗ -0.0001∗∗∗ -0.0001∗∗∗
(0.0001) (0.0000) (0.0000)
High School 0.1088∗∗∗ 0.1830∗∗∗ 0.0192∗∗
(0.0108) (0.0114) (0.0088)
Some College 0.1797∗∗∗ 0.2819∗∗∗ 0.0405∗∗∗
(0.0098) (0.0130) (0.0086)
Bachelor’s Degree 0.2014∗∗∗ 0.3403∗∗∗ 0.0727∗∗∗
(0.0112) (0.0132) (0.0087)
Advanced Degree 0.2616∗∗∗ 0.3542∗∗∗ 0.0781∗∗∗
(0.0132) (0.0120) (0.0090)
Child under 13 -0.0628∗∗∗ 0.0009
(0.0065) (0.0029)
COVID Rate (per 1,000) 0.0004∗ 0.0002∗∗∗ -0.0000
(0.0002) (0.0001) (0.0000)
COVID Death Rate (per 1,000) -0.0044 -0.0076∗ -0.0058
(0.0076) (0.0044) (0.0036)
Share of School Closures 0.0521 -0.0805∗∗ -0.0158
(0.0335) (0.0299) (0.0304)
Bartik Shift-Share 0.0083 -0.0083 -0.0074
(0.0142) (0.0050) (0.0085)
State Fixed Effects Yes Yes Yes
Region-Year-Quarter Fixed Effects Yes Yes Yes
Other Indicators Yes Yes Yes
R2 0.074 0.063 0.017
N 222,009 309,196 224,011
Notes: This table reports estimates from difference-in-differences models that compare workers in states with restrictions on
municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic. The treatment variable
is a binary variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are
for all months in 2018,19 (pre-COVID) and August 2021 to December 2022 (post-COVID). Individual control variables in all
models include age, age-squared, education (less than high school, some college, bachelor’s degree, advanced degree) and an
indicator for at least one child under 13. Other indicators include individual race (Black Non-Hispanic, other Non-Hispanic,
and Hispanic), marital status, number of children in the household, metro area residency and a categorical variable (provided
by the IPUMS CPS) for CBSA size. Area COVID-19 controls include a continuous measure of cumulative cases and deaths
per capita (per 1M residents) along with a control for school closures. All models include state fixed effects, region-by-quarter
fixed effects and Bartik shift-share variables. Observations are weighted using CPS sample weights. Robust standard errors are
clustered at the state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
75
Table A7:
Effects of Municipal Restrictions on Employment
(1) (2) (3)
Married Women Women Married Men
with Children without Children with Children
Municipal Restrictions -0.0305∗∗ 0.0041 -0.0093∗
(0.0125) (0.0064) (0.0046)
Age 0.0442∗∗∗ 0.0042∗∗ 0.0086∗∗∗
(0.0044) (0.0016) (0.0018)
Age-Squared -0.0005∗∗∗ -0.0001∗∗∗ -0.0001∗∗∗
(0.0001) (0.0000) (0.0000)
High School 0.1086∗∗∗ 0.1856∗∗∗ 0.0193∗∗
(0.0111) (0.0113) (0.0086)
Some College 0.1888∗∗∗ 0.2900∗∗∗ 0.0402∗∗∗
(0.0100) (0.0131) (0.0088)
Bachelor’s Degree 0.2131∗∗∗ 0.3475∗∗∗ 0.0700∗∗∗
(0.0115) (0.0134) (0.0087)
Advanced Degree 0.2797∗∗∗ 0.3611∗∗∗ 0.0744∗∗∗
(0.0126) (0.0120) (0.0089)
Child under 13 -0.0601∗∗∗ 0.0018
(0.0071) (0.0027)
COVID Rate (per 1,000) 0.0004∗∗ 0.0002∗∗∗ -0.0000
(0.0002) (0.0001) (0.0000)
COVID Death Rate (per 1,000) -0.0082 -0.0092∗∗ -0.0071∗
(0.0070) (0.0042) (0.0041)
Share of School Closures 0.0485∗ -0.0764∗∗ -0.0397
(0.0272) (0.0304) (0.0326)
Bartik Shift-Share 0.0130 -0.0084 -0.0047
(0.0158) (0.0058) (0.0063)
State Fixed Effects Yes Yes Yes
Region-Year-Quarter Fixed Effects Yes Yes Yes
Other Indicators Yes Yes Yes
R2 0.080 0.068 0.020
N 222,009 309,196 224,011
Notes: This table reports estimates from difference-in-differences models that compare workers in states with restrictions on
municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic. The treatment variable
is a binary variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are
for all months in 2018,19 (pre-COVID) and August 2021 to December 2022 (post-COVID). Individual control variables in all
models include age, age-squared, education (less than high school, some college, bachelor’s degree, advanced degree) and an
indicator for at least one child under 13. Other indicators include individual race (Black Non-Hispanic, other Non-Hispanic,
and Hispanic), marital status, number of children in the household, metro area residency and a categorical variable (provided
by the IPUMS CPS) for CBSA size. Area COVID-19 controls include a continuous measure of cumulative cases and deaths
per capita (per 1M residents) along with a control for school closures. All models include state fixed effects, region-by-quarter
fixed effects and Bartik shift-share variables. Observations are weighted using CPS sample weights. Robust standard errors are
clustered at the state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
76
Table A8:
Effects of Municipal Restrictions on Conditional Hours Worked
(1) (2) (3)
Married Women Women Married Men
with Children without Children with Children
Municipal Restrictions 0.1789 -0.1340 -0.2065
(0.2067) (0.2083) (0.1561)
Age 0.5243∗∗∗ 0.4266∗∗∗ 0.0695
(0.1351) (0.0375) (0.0768)
Age-Squared -0.0059∗∗∗ -0.0050∗∗∗ -0.0006
(0.0017) (0.0004) (0.0009)
High School 1.6564∗∗∗ 1.6862∗∗∗ 1.5665∗∗∗
(0.3925) (0.2967) (0.2703)
Some College 1.4098∗∗∗ 2.4808∗∗∗ 1.9407∗∗∗
(0.5163) (0.3099) (0.2977)
Bachelor’s Degree 2.0257∗∗∗ 3.7614∗∗∗ 1.9901∗∗∗
(0.4924) (0.2378) (0.2904)
Advanced Degree 3.4444∗∗∗ 5.3804∗∗∗ 2.9950∗∗∗
(0.3661) (0.2876) (0.2990)
Child under 13 -1.3026∗∗∗ -0.4235∗∗∗
(0.2204) (0.1292)
COVID Rate (per 1,000) -0.0005 0.0015 -0.0033∗
(0.0032) (0.0021) (0.0018)
COVID Death Rate (per 1,000) 0.0310 -0.1854 -0.0292
(0.1799) (0.1248) (0.1322)
Share of School Closures -2.4171∗∗ -0.2506 -0.5360
(1.0277) (0.6041) (0.7449)
Bartik Shift-Share 0.7151∗∗∗ -0.0218 0.0368
(0.2152) (0.2637) (0.1574)
State Fixed Effects Yes Yes Yes
Region-Year-Quarter Fixed Effects Yes Yes Yes
Other Indicators Yes Yes Yes
R2 0.027 0.021 0.021
N 139,138 218,528 203,372
Notes: This table reports estimates from difference-in-differences models that compare workers in states with restrictions on
municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic. The treatment variable
is a binary variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are
for all months in 2018,19 (pre-COVID) and August 2021 to December 2022 (post-COVID). Individual control variables in all
models include age, age-squared, education (less than high school, some college, bachelor’s degree, advanced degree) and an
indicator for at least one child under 13. Other indicators include individual race (Black Non-Hispanic, other Non-Hispanic,
and Hispanic), marital status, number of children in the household, metro area residency and a categorical variable (provided
by the IPUMS CPS) for CBSA size. Area COVID-19 controls include a continuous measure of cumulative cases and deaths
per capita (per 1M residents) along with a control for school closures. All models include state fixed effects, region-by-quarter
fixed effects and Bartik shift-share variables. Observations are weighted using CPS sample weights. Robust standard errors are
clustered at the state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
77
Table A9: Presence/Absence of Spouse in Mediating Maternal Labor Supply Effects
(1) (2) (3) (4)
Labor Force Employed, Employed, Hours
Participation At Work Any Last Week
Restrictions x Spouse Present -0.025∗∗ -0.030∗∗ -0.029∗∗ 0.198
(0.012) (0.013) (0.012) (0.212)
Restrictions x Spouse Absent -0.078∗∗∗ -0.070∗∗∗ -0.064∗∗∗ -1.033
(0.020) (0.018) (0.017) (1.063)
Spouse Absent 0.118∗∗∗ 0.095∗∗∗ 0.096∗∗∗ 0.739∗∗
(0.013) (0.014) (0.012) (0.358)
M ean 0.670 0.627 0.655 36.552
R2 0.078 0.074 0.080 0.027
N 222,009 222,009 222,009 139,138
Notes: Sample restricted to women with children. The outcome in column 4 is hours worked last
week conditional on working at all. The treatment variable is a dummy for persons (aged 25 to 54)
living in states with municipal restrictions as of Q1:2018. All models include state fixed effects,
region-by-quarter fixed effects and Bartik shift-share variables. Individual control variables include
age, race (Black Non-Hispanic, other Non-Hispanic, and Hispanic), metro area residency, MSA or
CBSA size, marital status, number of children in the household and an indicator for at least one child
under 13. Area COVID-19 controls include a continuous measure of cumulative cases and deaths
per capita (per 1M residents) along with a control for school closures. Observations are weighted
using CPS sample weights. Robust standard errors are clustered at the state-level. Significance at
1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
78
Table A10:
Effects of Municipal Restrictions on Labor Force Outcomes by Education
(1) (2) (3) (4)
Labor Force Employed, Employed, Hours
Participation At Work Any Last Week
Panel A: College-Educated Married Women with Children
Restrictions ×P ost -0.0260** -0.0268* -0.0268** 0.1238
(0.0128) (0.0135) (0.0131) (0.3459)
M ean 0.734 0.688 0.723 37.127
R2 0.051 0.049 0.050 0.026
N 100,219 100,219 100,219 68,976
Panel B: Non College-Educated Married Women with Children
Restrictions ×P ost -0.0267* -0.0332** -0.0327** 0.2449
(0.0143) (0.0150) (0.0148) (0.3676)
M ean 0.619 0.576 0.600 35.987
R2 0.062 0.059 0.061 0.028
N 121,790 121,790 121,790 70,162
Notes: This table reports estimates from difference-in-differences models that compare workers in states with re-
strictions on municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic.
The outcome in column 4 is hours worked last week conditional on working at all. The treatment variable is a binary
variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are for
all months in 2018,19 (pre-COVID) and 2020,21 (post-COVID). All models include state fixed effects, region-by-
quarter fixed effects and Bartik shift-share variables. Individual control variables include age, race (Black Non-
Hispanic, other Non-Hispanic, and Hispanic), education (less than high school, some college, bachelor’s degree,
advanced degree), metro area residency, MSA or CBSA size, marital status, number of children in the household
and an indicator for at least one child under 13. Area COVID-19 controls include a continuous measure of cumula-
tive cases and deaths per capita (per 1M residents) along with a control for school closures. The specification used
for this table additionally includes area fixed effects. Observations are weighted using CPS sample weights. Robust
standard errors are clustered at the state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **,
and *, respectively.
79
Table A11:
Effects of Municipal Restrictions on Labor Force Outcomes by Race
(1) (2) (3) (4)
Labor Force Employed, Employed, Hours
Participation At Work Any Last Week
Panel A: White Married Women with Children
Restrictions ×P ost -0.0257** -0.0320** -0.0300** 0.2058
(0.0126) (0.0137) (0.0127) (0.2593)
M ean 0.698 0.654 0.685 36.296
R2 0.077 0.073 0.078 0.032
N 150,613 150,613 150,613 98,476
Panel B: Non-White Married Women with Children
Restrictions ×P ost -0.0274* -0.0285* -0.0317* 0.0436
(0.0149) (0.0160) (0.0158) (0.3478)
M ean 0.613 0.570 0.593 37.174
R2 0.066 0.063 0.067 0.027
N 71,396 71,396 71,396 40,662
Notes: This table reports estimates from difference-in-differences models that compare workers in states with re-
strictions on municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic.
The outcome in column 4 is hours worked last week conditional on working at all. The treatment variable is a binary
variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are for
all months in 2018,19 (pre-COVID) and 2020,21 (post-COVID). All models include state fixed effects, region-by-
quarter fixed effects and Bartik shift-share variables. Individual control variables include age, race (Black Non-
Hispanic, other Non-Hispanic, and Hispanic), education (less than high school, some college, bachelor’s degree,
advanced degree), metro area residency, MSA or CBSA size, marital status, number of children in the household
and an indicator for at least one child under 13. Area COVID-19 controls include a continuous measure of cumula-
tive cases and deaths per capita (per 1M residents) along with a control for school closures. The specification used
for this table additionally includes area fixed effects. Observations are weighted using CPS sample weights. Robust
standard errors are clustered at the state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **,
and *, respectively.
80
Table A12:
Effects of Municipal Restrictions on Labor Force Outcomes by Age of the Youngest Child
(1) (2) (3) (4)
Labor Force Employed, Employed, Hours
Participation At Work Any Last Week
Panel A: Oldest Child Age 0–11
Restrictions ×P ost -0.0264*** -0.0295*** -0.0308*** -0.0951
(0.0074) (0.0077) (0.0075) (0.2499)
M ean 0.655 0.604 0.640 35.966
R2 0.085 0.077 0.087 0.032
N 110,649 110,649 110,649 66,874
Panel B: Youngest Child Age 12–17
Restrictions ×P ost -0.0259 -0.0365** -0.0335* -0.3512
(0.0169) (0.0176) (0.0179) (0.5469)
M ean 0.742 0.706 0.727 37.955
R2 0.056 0.052 0.056 0.022
N 55,642 55,642 55,642 39,281
Notes: This table reports estimates from difference-in-differences models that compare workers in states with re-
strictions on municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic.
The outcome in column 4 is hours worked last week conditional on working at all. The treatment variable is a binary
variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are for
all months in 2018,19 (pre-COVID) and 2020,21 (post-COVID). All models include state fixed effects, region-by-
quarter fixed effects and Bartik shift-share variables. Individual control variables include age, race (Black Non-
Hispanic, other Non-Hispanic, and Hispanic), education (less than high school, some college, bachelor’s degree,
advanced degree), metro area residency, MSA or CBSA size, marital status, number of children in the household
and an indicator for at least one child under 13. Area COVID-19 controls include a continuous measure of cumula-
tive cases and deaths per capita (per 1M residents) along with a control for school closures. The specification used
for this table additionally includes area fixed effects. Observations are weighted using CPS sample weights. Robust
standard errors are clustered at the state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **,
and *, respectively.
81
Table A13:
Robustness Check - Effects on Labor Force Outcomes, Area Fixed Effects, All Months (2018-2022)
(1) (2) (3) (4)
Labor Force Employed, Employed, Hours
Participation At Work Any Last Week
Panel A: Married Women with Children
Restrictions ×Post -0.0188* -0.0243** -0.0216** 0.1492
(0.0105) (0.0105) (0.0104) (0.1637)
M ean 0.668 0.619 0.649 36.488
R2 0.088 0.085 0.092 0.038
N 318,256 318,256 318,256 197,074
Panel B: Women without Children
Restrictions ×Post 0.0054 0.0040 0.0033 -0.0611
(0.0048) (0.0047) (0.0047) (0.1822)
M ean 0.751 0.698 0.720 39.183
R2 0.077 0.073 0.077 0.028
N 442,590 442,590 442,590 308,928
Panel C: Married Men with Children
Restrictions ×Post 0.0026 0.0001 -0.0021 -0.0999
(0.0043) (0.0052) (0.0047) (0.0981)
M ean 0.945 0.899 0.923 43.810
R2 0.027 0.033 0.036 0.031
N 320,575 320,575 320,575 288,279
Notes: This table reports estimates from difference-in-differences models that compare workers in states with re-
strictions on municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic.
The outcome in column 4 is hours worked last week conditional on working at all. The treatment variable is a binary
variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are for
all months in 2018,19 (pre-COVID) and 2020,21 (post-COVID). All models include state fixed effects, region-by-
quarter fixed effects and Bartik shift-share variables. Individual control variables include age, race (Black Non-
Hispanic, other Non-Hispanic, and Hispanic), education (less than high school, some college, bachelor’s degree,
advanced degree), metro area residency, MSA or CBSA size, marital status, number of children in the household
and an indicator for at least one child under 13. Area COVID-19 controls include a continuous measure of cumula-
tive cases and deaths per capita (per 1M residents) along with a control for school closures. The specification used
for this table additionally includes area fixed effects. Observations are weighted using CPS sample weights. Robust
standard errors are clustered at the state-level. Significance at 1%, 5%, and 10% levels are indicated by ***, **,
and *, respectively.
82
Table A14:
Robustness Check - Effects of Municipal Restrictions (States with Municipal Broadband Initiatives) on Labor
Force Outcomes
(1) (2) (3) (4)
Labor Force Employed, Employed, Hours
Participation At Work Any Last Week
Panel A: Married Women with Children
Restrictions ×P ost -0.0290* -0.0333* -0.0322* 0.2215
(0.0132) (0.0144) (0.0135) (0.2086)
M ean 0.673 0.629 0.658 36.506
R2 0.079 0.074 0.081 0.027
N 198,455 198,455 198,455 124,820
Panel B: Women without Children
Restrictions ×P ost 0.0050 0.0050 0.0062 -0.1125
(0.0066) (0.0067) (0.0067) (0.2198)
M ean 0.758 0.712 0.733 39.343
R2 0.068 0.062 0.067 0.021
N 274,887 274,887 274,887 195,766
Panel C: Married Men with Children
Restrictions ×P ost -0.0006 -0.0056 -0.0071 -0.2177
(0.0042) (0.0051) (0.0048) (0.1609)
M ean 0.948 0.909 0.932 43.991
R2 0.015 0.017 0.020 0.021
N 200,616 200,616 200,616 182,311
Notes: This table reports estimates from difference-in-differences models that compare workers in states with re-
strictions on municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic.
The outcome in column 4 is hours worked last week conditional on working at all. The treatment variable is a binary
variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are for
all months in 2018,19 (pre-COVID) and August 2021 to December 2022 (post-COVID). All models include state
fixed effects, region-by-quarter fixed effects and Bartik shift-share variables. Individual control variables include
age, race (Black Non-Hispanic, other Non-Hispanic, and Hispanic), education (less than high school, some college,
bachelor’s degree, advanced degree), metro area residency, MSA or CBSA size, marital status, number of children
in the household and an indicator for at least one child under 13. Area COVID-19 controls include a continuous
measure of cumulative cases and deaths per capita (per 1M residents) along with a control for school closures.
Observations are weighted using CPS sample weights. Robust standard errors are clustered at the state-level. Sig-
nificance at 1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
83
Table A15:
Robustness Check - Effects of Municipal Restrictions (Teleworkable Jobs) on Labor Force Outcomes
(1) (2) (3) (4)
Labor Force Employed, Employed, Hours
Participation At Work Any Last Week
Panel A: Married Women with Children
Restrictions ×P ost -0.0270 -0.0256 -0.0267 0.3493
(0.0167) (0.0151) (0.0159) (0.4326)
M ean 0.316 0.298 0.309 38.023
R2 0.126 0.122 0.128 0.035
N 105,524 105,524 105,524 31,439
Panel B: Women without Children
Restrictions ×P ost -0.0236*** -0.0227** -0.0224** 0.0240
(0.0078) (0.0091) (0.0086) (0.2327)
M ean 0.395 0.374 0.383 40.584
R2 0.142 0.135 0.139 0.026
N 124,499 124,499 124,499 46,537
Panel C: Married Men with Children
Restrictions ×P ost -0.0142 -0.0189 -0.0216 0.0366
(0.0127) (0.0127) (0.0137) (0.3104)
M ean 0.826 0.798 0.816 44.727
R2 0.125 0.116 0.127 0.029
N 66,106 66,106 66,106 52,761
Notes: This table reports estimates from difference-in-differences models that compare workers in states with re-
strictions on municipal broadband to workers in states without restrictions before and after the COVID-19 pandemic.
The outcome in column 4 is hours worked last week conditional on working at all. The treatment variable is a binary
variable for persons (aged 25 to 54) living in states that had municipal restrictions as of Q1:2018. CPS data are for
all months in 2018,19 (pre-COVID) and August 2021 to December 2022 (post-COVID). All models include state
fixed effects, region-by-quarter fixed effects and Bartik shift-share variables. Individual control variables include
age, race (Black Non-Hispanic, other Non-Hispanic, and Hispanic), education (less than high school, some college,
bachelor’s degree, advanced degree), metro area residency, MSA or CBSA size, marital status, number of children
in the household and an indicator for at least one child under 13. Area COVID-19 controls include a continuous
measure of cumulative cases and deaths per capita (per 1M residents) along with a control for school closures.
Observations are weighted using CPS sample weights. Robust standard errors are clustered at the state-level. Sig-
nificance at 1%, 5%, and 10% levels are indicated by ***, **, and *, respectively.
84
File and source
- File
- the-impact-of-municipal-broadband-restrictions-on-covid-19-labor-market-outcomes.pdf
- Size
- 1,959,268 bytes
- SHA-256
- 995ad9d2cbb45613ac9fca469bb553f3cca4057451153032d4688a9d0dc8d3f2
- Original
- doi.org