Before the Public Utilities Commission of the State of California Filed 07/02/21 04:01 Pm

BEFORE THE PUBLIC UTILITIES COMMISSION OF THE STATE OF CALIFORNIA FILED 07/02/21 04:01 PM

Order Instituting Rulemaking Regarding Broadband Infrastructure Deployment and to R. 20-09-001 Support Service Providers in the State of Filed September 10, 2020 California

OPENING COMMENTS OF THE UTILITY REFORM NETWORK ON THE ASSIGNED ADMINISTRATIVE LAW JUDGE’S RULING REGARDING REDLINING STUDIES

Brenda D. Villanueva, Telecom and Regulatory Attorney

Regina Costa, Telecommunications Director

The Utility Reform Network 1620 5th Ave., Ste. 810 San Diego, CA 92101 (619) 398-3680 [email protected] [email protected]

July 2, 2021

1 / 67 Table of Contents

I. Introduction ...... 1

II. Contextual Background Present in Redlining as Applied to Broadband ...... 5

A. The Term “Redlining” and its Roots in Other Contexts ...... 5

B. Defining “Digital Redlining” to Recognize Inequitable Broadband Deployment Practices ..... 6

III. The Studies Identified in the ALJ Ruling and Others Presented Provide Strong Evidence of Digital Redlining in California ...... 8

A. The Greenlining Study ...... 9

B. The CWA & NDIA Study ...... 12

C. The Annenberg Study...... 14

1. The Original Annenberg Study ...... 14

2. The Updated Annenberg Study ...... 16

D. The 2021 Update of the CETF Survey ...... 18

E. The Studies Reviewed Demonstrate that the Commission Must Expand Universal Service Programs to Support Broadband Adoption ...... 20

IV. Table 1 - City and Census Designated Place Analysis ...... 21

V. Response to ALJ Ruling Questions ...... 25

A. ALJ Ruling Question 1 ...... 25

B. ALJ Ruling Question 2 ...... 27

C. ALJ Ruling Question 3 ...... 28

D. ALJ Ruling Question 4 ...... 30

E. ALJ Ruling Question 5 ...... 32

1. Digital Redlining Should be Identified Based on the Availability of Wireline Broadband Service ...... 33

2. Digital Redlining Should Consider Affordable and Reliable Service ...... 34

F. ALJ Ruling Question 6 ...... 35

G. ALJ Ruling Question 7 ...... 37

2 / 67 1. The Commission Should Include the 2019 Network Exam Study of AT&T as a Data Source in this Proceeding, and Update the Study ...... 37

2. Cable Business Models Contribute to the Redlining Problem ...... 43

H. ALJ Ruling Question 8 ...... 44

VI. Conclusion ...... 45

Appendix A

Appendix B

3 / 67 I. Introduction

Pursuant to the Assigned Administrative Law Judge’s May 28, 2021, ruling in R.20-09-

001 (“ALJ Ruling” or “Ruling”),1 The Utility Reform Network (TURN) hereby submits opening comments on the general and specific questions set forth in the ALJ Ruling. TURN welcomes the opportunity to provide comment regarding the broadband deployment deficiencies in

California, present evidence of digital redlining,2 and suggest how to investigate and address these issues. For the reasons discussed in detail below, including in the response to Question 5,

TURN believes that digital redlining should be identified as occurring in areas where residents do not have two wireline broadband service providers that offer downstream broadband service speeds of at least 100 Mbps.3

In opening comments on the Order Instituting Rulemaking in this proceeding TURN and

Center for Accessible Technology observed that:4

• Market forces have failed to meet California's broadband needs. • Broadband service is unaffordable for many Californians. • The Commission needs adequate data about key aspects of broadband, ranging from the location of broadband-capable infrastructure to the price impediments that deter adoption. • The Commission also needs to address upload speeds, in addition to download speeds. • The Commission needs to address the telecommunications surcharge contribution base to support the Commission’s universal service goals.

1 Assigned Administrative Law Judge’s Ruling, R. 20-09-001 (May 28, 2021) (“ALJ Ruling”). 2 The ALJ Ruling uses the term “redlining” and explains that this is a practice where ISPs “are refusing to serve certain communities or neighborhoods within their service or franchise areas,” TURN will refer to these practices as “digital redlining.” TURN will refer to redlining practices in the housing industry as “housing redlining” as appropriate. These terms are explained in detail in these comments in Section II. 3 Affordability Metrics Framework Staff Proposal, R.18-07-006, at 22 (January 24, 2020). 4 Comments of The Utility Reform Network and the Center for Accessible Technology on the Commission’s Order Instituting Rulemaking, R. 20-09-001, at 4 (October 12, 2020).

4 / 67 TURN continues to support these observations and reiterates them here because in addition to the background provided below related to the term “redlining,” they serve to shape the circumstances in which California communities struggle to obtain high-speed broadband service that is necessary for everyday life.

The ALJ Ruling sets within the scope of this proceeding the timely and necessary question, whether “Internet service providers (ISPs) are refusing to serve certain communities or neighborhoods within their service or franchise areas, a practice commonly called redlining.”5

As discussed below, there is no question that broadband ISPs6 have refrained from investing to serve all Californians in their service areas or have otherwise selected service areas that target customers who generate high-revenues and who have relatively low deployment costs.

Based on our review of the studies and data referred to in the ALJ Ruling, and other analysis provided below, TURN believes that there is significant evidence of digital redlining in

California. According to data reviewed by TURN, digital redlining impacts lower income communities, including those associated with inner-city neighborhoods and rural areas of the state. In fact, a number of these communities have been shaped and formed due to historic housing redlining practices, helpful context as the Commission seeks to investigate digital redlining practices. The studies and data reveal broadband ISPs persistent lack of investment in lower income communities, among others. This poor investment undermines opportunity for the residents of those communities. TURN urges the Commission to find that these broadband ISPs past and current practices have led to digital redlining and significant impacts in many

5 ALJ Ruling at 1. 6 The ALJ Ruling uses the term “Internet service providers (ISPs).” ALJ Ruling at 1. TURN will use the term “broadband ISPs” for the service providers, and “broadband service” for the service provided.

5 / 67 communities, and further urges the Commission to promptly establish a process in which broadband ISPs can also be part of the solution.

TURN also urges the Commission to think broadly regarding whether Californians across the state have broadband truly available to them. Even in the areas of the state where reliable high-speed broadband is deployed and service offerings are provided by multiple broadband

ISPs, high prices arising from the absence of downward price pressures from competition, make that service unaffordable for many California households.

TURN believes that digital redlining warrants the exercise of Commission authority to investigate and mitigate the inequitable impacts felt by communities across the state. We believe the Commission is pursuing the correct approach in this docket by building a record to support actions to address digital redlining, consistent with its authority.

The Commission has plenary authority over telephone corporations pursuant to both the

California Constitution and the Public Utilities Code,7 that includes oversight of the service and the facilities.8 The Commission has held that broadband internet access is essential and “the

Legislature contemplated a significant role for the Commission in closing the digital divide in

California and in bringing advanced communications services, including broadband internet access, to all Californians.”9 When addressing the issue of network resiliency regulation, the

7 Const. Article XII, sec. 5. This grant of plenary power is further expressed in Pub. Util. Code sec. 701, which allows the Commission to do all things “necessary and convenient” for regulation of public utilities, see also Decision Adopting Wireline Provider Resiliency Strategies, D.21-02-029 (February 11, 2021). 8 Decision Adopting Wireline Provider Resiliency Strategies, D.21-02-029, at 9 (February 11, 2021) (citing Cal. Const. Art. XII, Section 1-6; Section 701). 9 See also, D.20-07-032 (R.18-07-006) at 5, 25.

6 / 67 Commission clarified that its jurisdiction includes using its police powers over all essential utility network service which the Commission has recognized encompasses interconnected voice over internet protocol (VoIP).10 In a more recent decision, the Commission reaffirmed its authority over broadband in response to parties’ assertion that the FCC preempts state law.

There, the Commission cited Mozilla v. FCC, 940 F. 3d 1, 76 (D.C. Cir. 2019), as “recognizing the role of the states in regulating broadband.”11 TURN supports the Commission’s decisions and believes that the Mozilla Court12 opens a door for state commissions to fill the gap left by the

FCC’s reclassification decision. TURN believes that the Commission must take responsibility for consumers’ needs for broadband services through universal service policies, consumer protection rules, and exercise of police powers.

These comments present a contextual overview about digital redlining, then respond to requests for comments on the three studies identified in the ALJ Ruling and other work, and finally offer responses to the questions contained in the ALJ Ruling.

10 Pub. Util. Code sec. 451, 584, 701, 761, 768, and 1001. See also Decision Adopting Wireline Provider Resiliency Strategies, D.21-02-029, February 11, 2021, at 12-13, 97-98 (Conclusions of Law). 11 D.21-04-005, Order Instituting Rulemaking into the Review of the California High Cost Fund-A Program, Decision Adopting Broadband Imputation in the General Rate Cases of the Small Independent Local Exchange Carriers (April 15, 2021) at 13-14 (“We accordingly find no impediment to this Commission’s legal authority to adopt and implement broadband imputation rules set forth in this decision”). 12 Mozilla Corp. v. FCC, 940 F.3d 1 at 81 (D.C. Cir. 2019) (Not only is the FCC lacking its own statutory authority to preempt, but its effort to kick the States out of intrastate broadband regulation also overlooks the Communications Act’s vision of dual federal-state authority and cooperation in this area specifically citing numerous federal law sections that preserve state authority under police powers and consumer protection and affordability).

7 / 67 II. Contextual Background Present in Redlining as Applied to Broadband

A. The Term “Redlining” and its Roots in Other Contexts

The ALJ Ruling uses the term “redlining” when posing the question of “whether Internet service providers (ISPs) are refusing to serve certain communities or neighborhoods within their service or franchise areas.”13 TURN provides a detailed analysis of the history of redlining in

Appendix A to these comments. Redlining was one of a slew of strategies including, but not limited to, exclusionary zoning ordinances, deed restrictions, racial restrictive covenants, blockbusting, subprime lending and violence deployed by the government, private companies, and individuals.14 Redlining propagated and perpetuated government-endorsed racial residential segregation to encourage and support homeownership for people who are White in all-White suburbs and to exclude people of color—especially people who are Black—and foreigners from renting and owning homes in those suburbs. People of color and foreigners were relegated to ethnic and racial ghettos, and housing contracts excluded them from the same benefits of building equity and developing intergenerational wealth reserved for White Americans.15

The role racial residential segregation played, and continues to play, in structuring the life chances and outcomes of different communities cannot be understated. Everything from health, life expectancy, poverty rates, educational attainment, home values and rents, income, wealth,

13 ALJ Ruling at 1. 14 For a compelling account of the enduring legacy of these strategies of racial residential segregation in the California context, see Jesus Hernandez, “Redlining Revisited: Mortgage Lending Patterns in Sacramento 1930–2004,” Volume 33.2 June 2009 (pp. 291–313), International Journal of Urban and Regional Research. 15 See Appendix A for historical context regarding the origins of redlining.

8 / 67 and even political polarization16 can be mapped onto the legacy of racial residential segregation;17 broadband deployment is no exception. As a result, over the last 60 years, the term “redlining” has been applied as a catch-all phrase often used to describe myriad forms of racial discrimination from food deserts in low-income neighborhoods to the difficulty of obtaining student loans for residents of certain communities. While formerly redlined neighborhoods “have been historically bypassed by investments,”18 including broadband infrastructure investment, lending itself to further social justice impediments, the multiple and varied forms of racial discrimination are not all analogous and are not equivalent to the original meaning and historical spatial context of the term “redlining.”

B. Defining “Digital Redlining” to Recognize Inequitable Broadband Deployment Practices

For the reasons described above, TURN strongly cautions the Commission with respect to the use of the term “redlining” alone, and instead encourages the Commission to adopt a definition for “digital redlining” that more accurately reflects the government and private policies which have led to minimal broadband and communications investment—if any—in communities of color, and or, rural areas. Specifically, TURN believes that digital redlining should be identified as occurring in areas where residents do not have two providers of wireline broadband services that offer downstream services of at least 100 Mbps.

16 For a brief but informative survey of the social science literature on the harms of racial residential segregation as these apply to, please see Menendian, et al. (pp-8-9). 17 Stephen Menendian, Arthur Gailes, and Samir Gambhir, The Roots of Structural Racism: Twenty-First Century Racial Residential Segregation in the United States (Berkeley, CA: Othering & Belonging Institute, 2021). https://belonging.berkeley.edu/roots-structural-racism. 18 Galperin et al., “Who Gets Access to Fast Broadband? Evidence from Los Angeles County,” Government Information Quarterly, May 2021, https://doi.org/10.1016/j.giq.2021.101594.

9 / 67 Digital redlining recognizes the role redlining practices and the other strategies used to create, enforce, and perpetuate racial segregation have played in creating and exacerbating the racial and economic inequalities the Commission now strives to dismantle. Light-touch broadband regulation at the federal, state, and municipal levels have fostered the environment in which broadband ISPs deploy infrastructure seemingly “race-neutral,” yet, in practice, have disparate impacts on communities of color—especially Black communities.19 Indeed, for example, a forthcoming May 2021 study analyzing digital inequity and urban segregation found that less affluent and minority communities saw a “slow rollout” of fiber deployment and broadband only “offered by a single provider (typically the incumbent cable TV provider).”20

Disparate impacts exist even when the intent of a policy is race-neutral, but the impact is not.21

Therefore to address Digital Redlining, the Commission must act to address the impacts of the government and private actions in allowing these inequitable practices to develop and persist.

However, while the analysis of digital redlining certainly must include the impact of the lack of broadband ISP investment on communities of color, TURN also believes that the

Commission must consider the lack of investment in rural areas of the state, including tribal areas. These areas suffer the same consequences of digital redlining as urban and suburban communities of color in the form of lost economic and social opportunities. The absence of

19 Galperin, et al. 2021. Unpublished manuscript, at p. 2 (“This strengthens the case for improving the collection and reporting of broadband data at the federal and state level. Further, the findings also bear on the debate over the classification of broadband as an information service (and therefore more lightly regulated under Title I of the Communications Act) or as an essential communication service (and thus subject to more stringent obligations of Title II”) (internal citations omitted). 20 Galperin et al. 2021 at p. 8. 21 See, e.g., U.S. Equal Employment Opportunity Commission, Race/ Color Discrimination, https://www.eeoc.gov/racecolor-discrimination (For example, a “no beard” employment policy that applies to all workers without regard to race may still be unlawful if it is not job-related and has a negative impact on the employment of African-American men (who have a predisposition to a skin condition that causes severe shaving bumps”)).

10 / 67 affordable and high-quality fixed broadband, which is a “required element of essential communications service for Californians to be able to participate fully in society,”22 all but ensures that rural communities that have been subject to digital redlining are experiencing an ongoing crisis of health and safety and will otherwise remain economically disadvantaged due to their inability to fully participate in society.

III. The Studies Identified in the ALJ Ruling and Others Presented Provide Strong Evidence of Digital Redlining in California

The ALJ Ruling requested comment on three studies by—The Greenlining Institute;23 the

Communications Workers of America (CWA) and the National Digital Inclusion Alliance

(NDIA);24 and the USC Annenberg Research Network for International Communication

(ARNIC) and the USC Price Spatial Analysis Lab.25 TURN provides brief comment on the studies identified in the ALJ Ruling. Viewed individually or on a combined basis, these studies suggest that digital redlining exists in California. Furthermore, in addition to the three studies

22 Affordability Metrics Framework Staff Proposal, R.18-07-006, January 24, 2020, p. 22. 23 Greenlining Institute, On the Wrong Side of the Digital Divide (June 2020) https://greenlining.org/publications/online-resources/2020/on-the-wrong-side-of-the-digital-divide/ (“Greenlining 2020 Study”); ALJ Ruling at 2 (providing a brief description). 24 Communications Workers of America and National Digital Inclusion Alliance, AT&T’s Digital Redlining: Leaving Communities Behind for Profit (October 2020) https://www.digitalinclusion.org/wp- content/uploads/dlm_uploads/2020/10/ATTs-Digital-Redlining-Leaving-Communities-Behind-for- Profit.pdf (“CWA/NDIA 2020 Study”); ALJ Ruling at 2-3 (providing a brief description). 25 Hernan Galperin, et al. USC Annenberg Research Network for International Communication and the USC Price Spatial Analysis Lab, Who Gets Access to Fast Broadband? Evidence from Los Angeles County 2014-17 (September 2019) http://arnicusc.org/wp-content/uploads/2019/10/Policy-Brief-4- final.pdf (“Original Annenberg Study”); ALJ Ruling at 3 (providing a brief description of the 2019 study). Since this writing, Dr. Galperin has since published an updated version of the 2019 publication. See Hernan Galperin, et al., “Who gets access to fast broadband? Evidence from Los Angeles County, Government Information Quarterly, (July 2021) https://www.sciencedirect.com/science/article/abs/pii/S0740624X21000307 (“Updated Annenberg Study”). A copy of the study, in press form, is provided in the appendix for this comment. TURN provides a summary of the Updated Annenberg study later in this comment.

11 / 67 from the ALJ Ruling, TURN provides other accounts that suggest evidence that digital redlining exists in California.

A. The Greenlining Study

The ALJ Ruling requested comment on the Greenlining Institute’s study, “On the Wrong

Side of the Digital Divide.”26 The Greenling study is based on data collected from a statewide telephone survey conducted in January 2019 by the Berkeley IGS Poll on behalf of the California

Emerging Technology Fund (CETF).27 The Greenlining study also relies on the 2019 Original

Annenberg Study. Since the Greenlining study provides qualitative analysis through testimonials and is based on quantitative analysis from these two other sources. We will first review the quantitative analysis, the 2019 CETF Survey and later in this comment, the 2019 Original

Annenberg Study.

The 2019 CETF Survey appears to have been conducted using sound methodology. The survey collected data from both cellular and landline telephone customers, based on random- digit dialing, during the period between January 21, 2019, to February 21, 2019.28 The survey used a sample size of 1,625 and was also conducted in six different languages.29

26 ALJ Ruling at 2. 27 Greenlining 2020 Study; California Emerging Technology Fund and the Berkeley IGS Poll, Institute of Government Studies, University of California Berkeley, Internet Connectivity and the “Digital Divide” in California – 2019 (2019) https://www.cetfund.org/wp- content/uploads/2019/08/005_003_002_CETF_2019_002_IGS_Poll_CA_Digital_Divide_ppt.pdf (“CETF 2019 Survey”). 28 CETF 2019 Survey at 2. 29 The CETF 2019 Survey was conducted in English, Spanish, Mandarin, Cantonese, Vietnamese, and Korean. CETF 2019 Survey at 2.

12 / 67 The 2019 CETF Survey is focused on broadband adoption. Results are presented in terms of “connectivity” which generally suggests whether the user has subscribed to the service.30 As a result, the 2019 CETF Survey addresses the redlining issue indirectly. Issue of affordability may be influencing decisions of whether or not to adopt broadband, even if the service is available from multiple broadband ISPs. At the time of the 2019 CETF survey nearly

22% of Californians either had no internet access or used only their smartphone to connect to internet services.31 The Greenlining study states that Latino households “are 21% less likely to have access to home internet than White ones.”32 This statement is based on data from the 2019

CETF survey that shows 89% of White Non-Hispanic households are connected through a home computing device (i.e., home desktop, laptop, or tablet computer) in 2019, compared to only

68% of Latino households.33 The 2019 CETF survey also reports that among all Latino households, those households that are Spanish-language dominant fare worse than any other ethnic group, with only 57% connected through a home computing device.

The Greenlining study also points to the disparity in Internet access in the home based on income. As reported in the 2019 CETF survey, the most frequently reported reason given for not having internet access in the home is “[t]oo expensive/no computer or smartphone at home.”34

In addition, the 2019 CETF survey indicates striking disparities in home internet based on

30 CETF 2019 Survey at 3-12 (presenting various findings of Internet connectivity layered with other information). 31 CETF 2019 Survey at 6; see also Greenlining 2020 Study. 32 Greenlining 2020 Study. 33 CETF 2019 Survey at 6. The CETF 2019 Survey distinguishes the households, “underconnected” meaning connected through a smartphone versus connected through a computing device. CETF 2019 Survey at 6. 34 CETF 2019 Survey at 12.

13 / 67 income. While 97% of households with incomes of $100,000 or more had home internet, 82% of households with incomes between $20,000 and $39,000, and only 52% of households with incomes less than $20,000, had home internet access.35

Next, TURN examines the qualitative analysis in the Greenlining Study. The

Greenlining Study provides the results of seven interviews with individuals with various ethnic and racial backgrounds. The interviews completed and presented in the study are about the participants’ home internet access. The Greenlining Study presents interview summaries for five individuals who resided in Fresno, California and two individuals who resided in Oakland,

California. These individuals tell compelling stories about the hard choices faced by low-income individuals in their efforts to obtain and maintain broadband service.36 Therefore, through this methodology, the Greenlining study attempts to provide both quantitative and qualitative analysis that can be helpful.

The Greenlining study has some areas where it could be improved, such as the study’s discussion of competition and fiber-based service deployment.37 The Greenlining study provides a “heat map” comparing a map showing redlining associated with bank practices in real estate markets with broadband deployment in the City of Oakland and surrounding areas. However, a more compelling presentation of this information would explain how the “heat map” was developed and the data sources used in the development of the heat map.38

35 CETF 2019 Survey at 10. 36 Greenlining 2020 Study. 37 Greenlining 2020 Study. 38 The Greenlining study contains six endnotes that identify sources. However, the sources appearing in endnotes 3, 4, 5, and 6 do not appear to relate to the corresponding discussion in the report.

14 / 67 TURN finds that the policy recommendations associated with the Greenlining Study are reasonable. Greenlining recommends that affordable broadband plans, defined as 50 Mbps for

$10 per month should be offered to all households earning up to 200 percent of the federal poverty line.39 Greenlining also recommends that fiber internet should be available to all, and that more competition is needed.40 TURN supports fiber deployment for most new broadband buildouts because fiber is a “future-proof” technology that will serve best the needs of

Californians in the long run.

B. The CWA & NDIA Study

The ALJ Ruling summarizes some of the key findings of the CWA and NDIA Study.41

TURN finds that the CWA & NDIA Study is based on geospatial analysis using FCC Form 477 data, as well as income data from the U.S. Census. Because FCC Form 477 data tracks the availability of broadband (not the adoption of broadband) the focus of this study is on broadband availability. The CWA & NDIA Study is based on FCC Form 477 data and poverty rates from the Census’s American Community Survey. Dr. Whitacre of Oklahoma State University replicated and confirmed NDIA’s analysis of digital redlining in Cleveland. Dr. Whitacre’s study of redlining in Cleveland was filed to support a formal complaint regarding AT&T practices at the Federal Communications Commission. Dr. Whitacre notes in the Cleveland study:

[T]he study offers clear evidence that AT&T has withheld the standard product offering for most suburbs- its fiber-enhanced “Fiber To the Node” VDSL infrastructure (“FTTN”)– from the overwhelming majority of census blocks with individual poverty

39 Greenlining 2020 Study. 40 Greenlining 2020 Study. 41 ALJ Ruling at 2-3; CWA/NDIA 2020 Study.

15 / 67 rates above 35%. As a consequence, residents of these neighborhoods: suffer uneven, often severely limited Internet access, in many cases 3 mbps downstream or less, and also lack access to AT&T’s competitive fiber-enabled video service and the benefits such competition and service would bring.42

TURN’s review of Dr. Whitacre’s studies (Cleveland and Dallas43) indicates that they are based on publicly available data sets and on sound methodology. The studies provide links to their source materials and access to data files. TURN’s review of all of these materials leads TURN to conclude that Dr. Whitacre’s results are reliable, and that the general conclusions associated with the CWA & NDIA study are reasonably supported. TURN notes that the CWA & NDIA study does not focus extensively on AT&T operations in California, however, given the results of the Commission-ordered Network Exam Study44, discussed below, TURN expects that a

California-centric analysis like that presented in the CWA & NDIA Study would generate comparable results.

TURN notes that the CWA & NDIA Study contains the following recommendations, indicating that “AT&T should:”

• Invest in next-generation networks -- AT&T should commit to capital investment in fiber deployment that would double the number of households passed by fiber in two years. If AT&T invests one quarter of its annual free cash flow (projected to be more than $25 billion) into rapid fiber deployment, it could deploy to more than 6 million locations per year. • Stop leaving rural communities behind -- AT&T must upgrade its network in rural communities to meet the FCC’s broadband definition, at least, and renew its efforts to deploy next-generation fiber.

42 Id., Whitacre Declaration, ¶27. 43 https://www.digitalinclusion.org/blog/2019/08/06/atts-digital-redlining-of-dallas-new-research-by-dr- brian-whitacre/ 44 See, Examination of the Local Telecommunications Networks and Related Policies and Practices of AT&T California and Frontier California. Study conducted pursuant to the CPUC Service Quality Rulemaking 11-12-001, Decision 13-02-023, and Decision 15-08-041, April 2019, p. 3. Hereinafter “Network Exam Study.”

16 / 67 • Stop leaving parts of urban communities behind -- AT&T must renew its efforts to equitably deploy next-generation fiber in urban locations. • Make its low-income product more accessible -- AT&T should invest more in advertising these products and agree to bulk sales of its wireline service to school districts and other public entities which redistributes plans to households. • Invest in good jobs -- AT&T must stop laying off its skilled, unionized workers and stop outsourcing work to subcontractors in order to pay lower wages and avoid being held legally responsible for the subcontractors’ conduct. An experienced workforce is a prerequisite to reliable, high-quality internet service, particularly in areas where AT&T’s network is outdated or deteriorated. Rather than layoffs, AT&T should invest in a workforce that will connect customers, rural and urban, to next-generation fiber networks.45 TURN believes that these recommendations would mitigate the digital divide in California, though AT&T would need to agree to implement these proposals. These recommendations would lead to a direct improvement in broadband availability and adoption and would also spur a competitive response from AT&T’s cable rivals in the areas where cable has decided to serve.

TURN also believes that application of this study’s methodology should be considered by the

Commission for a California-specific study.

C. The Annenberg Study

The ALJ requested comment on the 2019 Annenberg Study (hereinafter, “Original

Annenberg Study”), which TURN provides below. In addition, TURN describes and comments on the Updated Annenberg Study that was released in June of 2021.

1. The Original Annenberg Study

The ALJ ruling also request comment on “Who Gets Access to Fast Broadband?

Evidence from Los Angeles County 2014-17,” a study published as a Connected Cities and

Inclusive Growth Policy Brief from the University of Southern California’s Annenberg Research

45 CWA/NDIA 2020 Study at 8.

17 / 67 Network on International Communication.46 This study uses data from the CPUC on broadband availability, thus like the CWA & NDIA, study focuses on broadband availability.

The Original Annenberg Study is based on CPUC data regarding broadband service availability and demographic data from the American Community Survey.47 The Original

Annenberg Study focuses on broadband investment and competition in Los Angeles County, using measures of the number of broadband ISPs and availability of fiber services for the years

2014-2017.48 The study finds that broadband competition is more likely in affluent communities. The study also finds that the chances of fiber availability fall significantly in lower income areas.

The Original Annenberg Study examines both competition and fiber deployment in areas with various racial makeups. The Study finds that for low-income African American communities, the chances of facing broadband competition are significantly lower than for non- low-income African American communities. According to data presented in the Original

Annenberg Study, a higher the percentage of low-income African Americans leads the odds of broadband competition to decline from over 65% for communities with close to zero African

American residents to about 45% for communities with close to 100% African American residents.

46 Hernan Galperin, et al. USC Annenberg Research Network for International Communication and the USC Price Spatial Analysis Lab, Who Gets Access to Fast Broadband? Evidence from Los Angeles County 2014-17 (September 2019) http://arnicusc.org/wp-content/uploads/2019/10/Policy-Brief-4- final.pdf. 47 Original Annenberg Study at 6. 48 Original Annenberg Study at 1

18 / 67 Similarly, with regard to fiber deployment, the higher the percentage of African

American residents, the lower the probability of finding fiber connections available. In areas with close to zero African Americans, the odds of fiber availability are about 35%. In areas with close to 100% African American, the odds of fiber availability are just above 5%.

2. The Updated Annenberg Study

TURN notes that the Original Annenberg Study was updated in 2021 to include data from

2018, thus covering the years 2014-2018.49 The “Updated Annenberg Study” shows results that are similar to the earlier study.50 Below TURN provides an analysis of the Updated Annenberg

Study.

The Updated Annenberg Study is also based on CPUC data regarding broadband service availability and demographic data from the American Community Survey. The Updated

Annenberg Study focuses on broadband availability in Los Angeles County, using measures of the number of broadband ISPs and availability of fiber services for the years 2014-2018.51 The study finds that broadband competition is more likely in affluent communities,52 with the chances of fiber availability falling significantly in lower income areas.53

The Updated Annenberg Study examines both competition and fiber deployment in areas with various racial populations. The Study finds that for low-income African American

49 Galperin, Hernan, Thai V. Le, Kurt Daum, “Who gets access to fast broadband? Evidence from Los Angeles County, Government Information Quarterly, in press, https://annenberg.usc.edu/sites/default/files/2021/06/21/GIQ%202021.pdf . (Hereinafter, “Updated Annenberg Study.”) 50 TURN has attached a copy of the Updated Annenberg Study to these Comments as an appendix. 51 Updated Annenberg Study at 1 52 Updated Annenberg Study at 5. 53 Updated Annenberg Study at 6.

19 / 67 communities, the chances of facing broadband competition are significantly lower than for non- low-income African American communities. According to data presented in the Updated

Annenberg Study, a higher the percentage of low-income African Americans leads the odds of broadband competition to decline. The Updated Annenberg Study reports that broadband competition is at 77% for communities with close to zero African American residents but declines to about 68% for communities with close to 100% African American residents. The

Updated Annenberg Study also examines the combined impact of income and race on broadband competition. The study finds that low-income Black communities are much less likely than low- income white communities to have broadband competition. For low-income communities with few black residents, the probability of broadband competition is above 70 percent. However, in low-income communities that are mostly Black, the probability of broadband competition falls below 50 percent.

Similarly, with regard to fiber deployment, the Updated Annenberg Study shows that the higher the percentage of African American residents, the lower the probability of finding fiber connections available. In areas with close to zero African Americans, the odds of fiber availability are about 35%. In areas with close to 100% African American, the odds of fiber availability are just above 5%.54

The Updated Annenberg Study also conducts regression analysis to further evaluate fiber availability and broadband competition.55 The results of the regressions, similar to the results of the regressions conducted by TURN regarding the availability of broadband at 100 Mbps,

54 Updated Annenberg Study at 7. 55 This regression analysis is a new addition to the Updated Annenberg Study and did not appear in the original study.

20 / 67 indicate that median household income has a direct impact on broadband availability, both in terms of broadband competition and fiber availability. The Updated Annenberg Study’s regression analysis further supports the proposition that broadband “competition is particularly lacking in areas that combine poverty and a large concentration of Black residents.”56

In summary, both the original and updated Annenberg Studies providing convincing evidence of digital redlining. The original Annenberg Study, with reference to the nondiscriminatory provisions of California Public Utilities Code §5810, indicates that that study’s findings:

. . . suggest that broadband investments in LA County during 2014-17 did not adhere to these non-discriminatory standards. Regardless of intent, the practical effect has been the reproduction of the inequalities that have characterized the region for generations, with particular adverse effects for low-income Black communities in South LA.57

TURN believes that the results from both the original Annenberg Study and the Updated

Annenberg Study provide strong support for this conclusion.

D. The 2021 Update of the CETF Survey

As noted above, the Greenlining Study was based in part on the 2019 CETF survey. The results of the 2021 CETF survey were released in March 2021 and are based on collaboration with the University of Southern California.58 This telephone survey had a sample size of 1,650 households and was conducted in four languages. The 2021 CETF survey was conducted during

56 Updated Annenberg Study at 5-6. 57 Annenberg Study at 5. 58 “Statewide Survey on Broadband Adoption 2021: Internet Adoption and the ‘Digital Divide’ in California,” Results from a survey conducted for the California Emerging Technology Fund (CETF), University of Southern California, Principal Investigator: Dr. Hernan Galperin, March 2021 at 2. https://www.cetfund.org/wp- content/uploads/2021/03/Annual_Survey_2021_CETF_USC_Final_Summary_Report_CETF_A.pdf.

21 / 67 the period between February 10, 2021, to March 22, 2021, and has a margin of error of 2% for a

95% confidence level.59 The results from the new survey are generally consistent with the trends summarized for the 2019 CETF Survey, however, the survey results demonstrate that the impact of the pandemic has had an impact on broadband adoption. The necessity of using broadband connections for work, education, and other daily activities appears to have spurred broadband adoption, and it is also likely that the increased availability of emergency broadband subsidies and rate relief, have also a positive impact on broadband adoption. Table A summarizes the changes in areas highlighted in the Greenlining Study.

Table A: Comparison of Key Metrics from the CETF Survey 2019 CETF 2021 CETF Survey Survey No internet/Smartphone Only 22% 15% White Non-Hispanics with Home Internet 89% 91% Hispanic Households with Home Internet 68% 76% Spanish Language Hispanic Household with Home 57% 65% Internet Households with Income of $100,000 or More with 97% 98% Home Internet Households with Income Between $20,000 and 82% 81% $39,000 with Home Internet Households with Income Below $20,000 with Home 52% 77% Internet

While the most dramatic change in subscription to home broadband is in households with income below $20,000, it is likely that those gains face the greatest risk. Because emergency rate relief

59 “Statewide Survey on Broadband Adoption 2021: Internet Adoption and the ‘Digital Divide’ in California,” Results from a survey conducted for the California Emerging Technology Fund (CETF), University of Southern California, Principal Investigator: Dr. Hernan Galperin, March 2021 at 2. https://www.cetfund.org/wp- content/uploads/2021/03/Annual_Survey_2021_CETF_USC_Final_Summary_Report_CETF_A.pdf

22 / 67 offered and subsidies60 contributed to those gains, and the expiration of those programs are likely to lead to backsliding among those households. As is noted in the 2021 CETF survey, by a wide margin, the “main reason” given for households that either had no internet or smartphone-only connections is that home internet is “too expensive.”61 It is also notable that the 2021 CETF survey found that 62% of households who either had no internet or who were smartphone-only were unaware of low-cost home broadband plans.62 This lack of awareness suggests that the marketing of those low-cost plans is failing to reach a significant number of households.

E. The Studies Reviewed Demonstrate that the Commission Must Expand Universal Service Programs to Support Broadband Adoption

Some of the studies discussed above demonstrate that there is a clear correlation between broadband adoption and whether the services are affordable to customers. Thus, the

Commission must also work to address the consequences of high broadband service prices charged by broadband service providers in California. There is no question that this Commission has the ability to exercise its authority over broadband to promote universal service requirements contained in state law.63

60 Federal Communications Commission, Emergency Broadband Benefit Program, Report and Order, WC Docket No. 20-445, FCC 21-29 (February 2021) https://docs.fcc.gov/public/attachments/FCC-21- 29A1.pdf (establishing the Emergency Broadband Benefit Program to provide eligible consumers a discount on their broadband service plan under specific conditions). 61 “Statewide Survey on Broadband Adoption 2021: Internet Adoption and the ‘Digital Divide’ in California,” Results from a survey conducted for the California Emerging Technology Fund (CETF), University of Southern California, Principal Investigator: Dr. Hernan Galperin, March 2021, p. 20. https://www.cetfund.org/wp- content/uploads/2021/03/Annual_Survey_2021_CETF_USC_Final_Summary_Report_CETF_A.pdf 62 Id., p. 21. 63 Section 701 grants the Commission broad regulatory powers over public utilities; Section 709 advances state telecommunications policies to support universal service, ensure affordable telecommunications, encourage deployment of new technology, and bridge the digital divide. Pub. Util. Code §701, §709(a), (c), (d).

23 / 67 TURN believes that the Commission must also consider broadband affordability when addressing digital redlining. Thus, as part of the digital redlining solution the Commission must expand support for the broadband LifeLine program and work with broadband ISPs to promote the availability of affordable broadband service offers. The Commission can help address the problems of redlining by enhancing universal service support. For example, programs such as

LifeLine should be expanded to enable participation by individuals who have incomes above current program benchmarks and program rules should be adjusted to more clearly support wireless and wireline broadband services. The Teleconnect Fund could also be expanded to more easily allow nonprofits and smaller institutions to qualify for discounted broadband services and to encourage carriers to participate. Because universal service policies are critical to tackling problems of broadband deployment and adoption, the Commission should also expand the contribution base to better support both the deployment and adoption of broadband services.

IV. Table 1 - City and Census Designated Place Analysis

The ALJ Ruling requests comment on the analysis shown in its Table 1, which summarizes a Staff analysis of broadband availability at 100 Mbps download.64 TURN provides its comments about Table 1 below and presents TURN’s additional findings based on the data set used to generate Table 1.

The information shown in the ALJ Ruling’s Table 1 shows that households located in lower income areas are less likely to be served at download speeds of 100 Mbps. Specifically, the ALJ Ruling’s Table 1 shows that in California cities and Census Designated Places with

64 ALJ Ruling at 4.

24 / 67 median household income above $84,000, less than 1 percent of households are unserved.

However, for areas where median household income was below $53,221, that 75% or more households are unserved.

In addition to the review of Table 1, TURN used the detailed data65 that was used by the

Communications Division Staff to generate Table 1 to conduct additional statistical analysis.

TURN’s findings show further support to the trend shown in Table 1. Like Table 1, the detailed data provided with the ALJ Ruling contains information on the percentage of households served at 100 Mbps in California cities and Census Designated Places and the associated weighted average median household income. However, the detailed data also provides additional information on (1) the number of community anchor institutions in each geographic area, (2) the number of households in each geographic area, and (3) the size of each geographic area, stated in square miles. This additional data can be used to evaluate the results shown in Table 1 with more statistical rigor. TURN conducted a multivariate linear regression using the data underlying Table 1. Regression analysis can add depth to the evaluation of data and generates results that can be stated with statistical confidence.

The data set provided with the ALJ Ruling contains a total of 1,511 observations, which were associated with various cities and Census Designated Places.66 The regression conducted by TURN uses the percentage of each geographic area that is served as the dependent variable and uses the median household income, number of anchor institutions, and density of the geographic area (households per square mile) as explanatory variables.67 By including variables

65 The ALJ Ruling provided a link to the data at 4. 66 ALJ Ruling at 4. 67 I.e., 푃푒푟푐푒푛푡 표푓 퐻표푢푠푒ℎ표푙푑푠 푆푒푟푣푒푑 = 훼 + 훽1푀푒푑𝑖푎푛 퐻퐻 퐼푛푐표푚푒 + 훽2퐴푛푐ℎ표푟 + 훽3퐷푒푛푠𝑖푡푦

25 / 67 other than median household income as explanatory variables, the regression allows influences other than income to be analyzed. This approach also better tests the hypothesis that household incomes in a geographic area influence the availability of broadband at the 100 Mbps benchmark. The regression results are shown in Table B.

Table B: Regression Results—Dependent Variable: Percent of Households Served at 100 Mbps Variable Regression Std. Error of t Statistic Significance Coefficient the Regression Level Coefficient Constant 0.40437 0.0201 20.11 1% Median HH Income 0.0000023379 0.00000002425 9.64 1% Anchor Institutions 0.00084 0.00037 2.26 5% Density 0.00013 0.000008336 15.13 1% F of the regression: 140.66. Significant at 1%. R2 of the regression: 0.21876

The results in Table B lend further support to the information contained in the ALJ Ruling’s

Table 1. Table B shows that each of the explanatory variables (Median Household Income, the

Number of Anchor Institutions, and the Density of the geographic area) have a direct relationship with the number of households in each geographic area that are served at 100 Mbps.

The Median HH Income variable shows a positive sign and is highly statistically significant.

This result supports the proposition that the higher the household income in a geographic area, the higher the likelihood that the area will be served at 100 Mbps. The income coefficient in

Table A suggests that an increase in median household income of $10,000 in an area leads to a

2.3 percent increase in households served at 100 Mbps.

Similarly, for the number of anchor institutions in a geographic area the coefficient for anchor institutions suggests that the higher the number of anchor institutions in a geographic area, the higher the likelihood that the area will be served at 100 Mbps. Likewise, the coefficient

26 / 67 for customer density indicates that the higher the customer density in a geographic area, the higher the likelihood that the area will be served at 100 Mbps. These last two factors are consistent with the observed digital divide between urban and rural areas, as rural areas typically have fewer anchor institutions and have lower population density.68

The addition of other explanatory variables to this data set (e.g., variables reflecting other characteristics of the populations in the cities and Census Designated Places such as race, ethnicity, age, and education), could identify other factors that influence the percentage of households that are served at 100 Mbps. Expanding the types of variables available in the data set is something that the Staff should consider.

In conclusion, the ALJ Ruling’s Table 1, and the data underlying that data support the proposition that digital redlining is a reality in California.

68 The significance of the overall regression leads to the rejection of the null hypothesis that the explanatory variables have no impact on the percentage of households served at 100 Mbps in a geographic area. The F of the regression, with three degrees of freedom, is 140.66, which is significant at the 1 percent level. Regarding the explanatory power of this regression model, the R2 value of 0.219 suggests that the regression explains about 22 percent of the variation in the dependent variable of percentage of households unserved at 100 Mbps. While the R2 does not have a high value, that does not diminish the statistical significance of the regression results.

27 / 67 V. Response to ALJ Ruling Questions

TURN provides responses to the eight questions posed in the ALJ Ruling below.

A. ALJ Ruling Question 1

Are the inputs and assumptions of the studies discussed above accurate? How could one improve these studies?

As noted above in the detailed discussion, TURN generally believes that the three studies—the Greenlining Institute;69 the Communications Workers of America (CWA) and the

National Digital Inclusion Alliance (NDIA);70 and the USC Annenberg Research Network for

International Communication (ARNIC) and the USC Price Spatial Analysis Lab71—use sound approaches, use different methodologies, and focus on different elements.

The Greenlining Study relies on the 2019 CETF 2019 Survey and the Original

Annenberg Study. The 2019 CETF survey and USC Annenberg study appear to also use sound methodologies even though they respectively, have a different focus. For example, the 2019

69 Greenlining Institute, On the Wrong Side of the Digital Divide (June 2020) https://greenlining.org/publications/online-resources/2020/on-the-wrong-side-of-the-digital-divide/ (“Greenlining 2020 Study”); ALJ Ruling at 2 (providing a brief description). 70 Communications Workers of America and National Digital Inclusion Alliance, AT&T’s Digital Redlining: Leaving Communities Behind for Profit (October 2020) https://www.digitalinclusion.org/wp- content/uploads/dlm_uploads/2020/10/ATTs-Digital-Redlining-Leaving-Communities-Behind-for- Profit.pdf (“CWA/NDIA 2020 Study”); ALJ Ruling at 2-3 (providing a brief description). 71 Hernan Galperin, et al. USC Annenberg Research Network for International Communication and the USC Price Spatial Analysis Lab, Who Gets Access to Fast Broadband? Evidence from Los Angeles County 2014-17 (September 2019) http://arnicusc.org/wp-content/uploads/2019/10/Policy-Brief-4- final.pdf (“Original Annenberg Study”); ALJ Ruling at 3 (providing a brief description of the 2019 study). Since this writing, Dr. Galperin has since published an updated version of the 2019 publication. See Hernan Galperin, et al., “Who gets access to fast broadband? Evidence from Los Angeles County, Government Information Quarterly, (July 2021) https://www.sciencedirect.com/science/article/abs/pii/S0740624X21000307 (“Updated Annenberg Study”). A copy of the study, in press form, is provided in the appendix for this comment. TURN provides a summary of the Updated Annenberg study later in this comment.

28 / 67 CETF Survey asks questions about broadband adoption. On the other hand, both the original and updated Annenberg studies are focused on broadband availability. The CWA & NDIA study, also focused on broadband availability, appears to employ reasonable methodology. As discussed earlier, studies that focus on adoption address digital redlining indirectly.

Affordability of broadband services has a significant impact on adoption, so the CETF surveys do not isolate the impact of the lack of broadband availability.

In addition, the Original and Updated Annenberg studies rely on reasonable data sources, including CPUC data regarding broadband availability and Census demographic data. As discussed earlier, the CWA & NDIA Study relies on FCC Form 477 data and Census data. The results of CWA & NDIA studies have been replicated by Professor Whitacre.72 However, Form

477 data is recognized to have weaknesses and to overstate broadband availability. This is because Form 477 data identifies broadband as “available” for the entire population residing in a

Census block if at least one household in the Census block is served.73 The potential for overstatement is especially pronounced when considering the identification of broadband availability in rural and suburban areas where large and irregularly shaped Census blocks74 may lead to over-reporting of broadband availability. Given that the focus of the CWA & NDIA study is urban redlining, that weakness of the Form 477 data is less likely to be a concern.

TURN believes that the studies may be improved to help the Commission review the impact of digital redlining in the state. Both the Original and the Updated USC Annenberg

72 CWA & NDIA Study at 5. 73 See, for example, Inquiry Concerning Deployment of Advanced Telecommunications Capability to All Americans in a Reasonable and Timely Fashion, GN Docket No. 20-269, Fourteenth Broadband Deployment Report, FCC 21-18, at 13, para. 22 (January 19, 2021). 74 https://www.census.gov/newsroom/blogs/random-samplings/2011/07/what-are-census-blocks.html

29 / 67 studies limit their scope to Los Angeles County. Thus, expanding the scope of the USC

Annenberg studies would be a reasonable improvement. While the CWA & NDIA study contains a brief profile of fiber deployment in the California Central Valley,75 the CWA & NDIA study is based primarily on studies of AT&T that have been conducted outside of California. A focus on AT&T California, as well as other broadband ISPs in California would improve the study. The USC researchers updated the 2019 CETF Survey with recent data for the 2021 CETF

Survey. Moreover, the 2021 CETF Survey also incorporates the impact of the pandemic on broadband demand and the impact of emergency broadband programs on consumers’ ability to afford broadband. These are important and relevant considerations for the Commission to address surface level and underlying disparate impacts in different communities across the state.

B. ALJ Ruling Question 2

Do the findings of these studies provide evidence of a systemic problem in California?

Yes, while the three studies identified in the ALJ ruling approach the digital divide issue from alternative perspectives, they all lend support to the existence of a systemic problem in

California. The Original and Updated Annenberg studies demonstrate that broadband deployment has lagged in low-income areas in Los Angeles County, especially in those areas with high African American populations.76 The 2019 and 2021 CETF surveys demonstrate that several elements correspond to low broadband adoption levels, such as low household incomes,

75 CWA & NDIA study at 5. 76 Original Annenberg Study at 3; Updated Annenberg Study at 7.

30 / 67 non-white race, and Hispanic origin.77 Likewise, the CWA studies of areas outside of California also show evidence of digital redlining.78

Furthermore, as described in detail below (in response to Question 7), the Commission’s own Network Exam Study reflects evidence of a systemic problem.79 As noted in the Network

Exam Study, ILECs’ failure to invest in California contributes to the state’s broadband availability and competition problems. In fact, TURN notes that the CWA & NDIA study demonstrates that this problem is not unique to California; AT&T has engaged in digital redlining practices in its other operating states.80

C. ALJ Ruling Question 3

Do these studies indicate discrimination based on race, socioeconomic status or otherwise, and, if yes, what are the societal implications?

The three studies indicate that several factors, including race, socioeconomic status, and others are present with digital redlining practices as described below (in response to Question 5).

Moreover, the Commission’s own Network Exam Study, as discussed below in response to

Question 7, demonstrates underinvestment and with additional demographic data layered on top, could lead to additional startling results. In addition, the three studies provide significant evidence that standard business practices that reject investing in areas that unlikely to meet the broadband ISP's target investment returns, result in market failure. These decisions also reflect a

77 2019 CETF Survey at 6; 2021 CETF Survey at 6. 78 CWA & NDIA Study at 5-6. 79 Examination of the Local Telecommunications Networks and Related Policies and Practices of AT&T California and Frontier California, (Network Exam Study), conducted pursuant to R.11-12-001, D. 13- 02-023 and D. 15-08-041 at 2. 80 CWA & NDIA Study at 5-6.

31 / 67 callous disregard for the consequences of underinvestment or withholding investment decisions by broadband ISPs, an indication to the Commission that the market is not correcting itself.

TURN notes the high degree of correlation between race and income, with African Americans tending to have lower incomes.81 Thus, the discrimination associated with broadband ISPs withholding investment has led to racial disparities associated with inequitable broadband investment.

The market failure observed through digital redlining practices lead to significant societal implications. As the COVID-19 pandemic is clearly illustrating, access to affordable and high- quality broadband is a basic necessity which the Commission’s concluded previously.82 Full participation in economic, educational, and social activities depend on access to broadband services. The investment decisions that impact low-income and rural communities systematically deny economic opportunity for the residents of those communities. Unaffordable prices may also reduce broadband subscriptions if household income fluctuations result in households dropping home broadband when economic conditions deteriorate.83 TURN believes that the societal implications from digital redlining leaves communities unserved or underserved, and harms large numbers of Californians.

For the California communities that already have historic underpinnings of housing redlining, the layers of underinvestment or no investment for broadband infrastructure

81 See, for example, “Median Income in the Past 12 Months (in 2019 Inflation-Adjusted Dollars), U.S. Census Bureau, Table ID S1903, which shows median income for the White race of $69,823 and African American median income of $43,862. 82 D.20-07-032 (R.18-07-006) at 5, 25. 83 See, for example, “Determinants of broadband access and affordability: An analysis of a community survey on the digital divide.” Christopher G. Reddick, Roger Enriquez, Richard J. Harris, and Bonita Sharma, Cities: the International Journal of Urban Policy and Planning, 2020. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7480260/pdf/main.pdf

32 / 67 exacerbates the negative impacts that have plagued these communities for decades. In these areas, the Commission should commit to making rigorous and strident efforts to investigate the digital redlining practices to avoid missing nuanced problems in these communities.

D. ALJ Ruling Question 4

If the Commission were to undertake an investigation into whether ISPs are not serving certain communities or neighborhoods within their service or franchise areas, a practice generally referred to as redlining, how should the Commission conduct that investigation? What data should the Commission rely on for its investigation?

TURN emphatically supports the Commission’s efforts to investigate broadband ISPs service offerings and delivery to communities within their service or franchise areas to identify the presence of digital redlining practices. As explained throughout these comments, digital redlining practices and their impacts do not exist in isolation and have some roots in past inequitable practices. By identifying the factors contributing to digital redlining, i.e., resulting in areas where residents do not have two providers of wireline broadband services that offer downstream services of at least 100 Mbps, the Commission can have a robust, and data-driven understanding of the digital redlining practices occurring within a given community and its impacts. This will equip the Commission to then quickly identify the solution needed for similarly situated communities and target the impediments to broadband deployment and competition.

As noted above, the Commission has a comprehensive collection of studies that demonstrate that broadband service providers fail to invest in certain communities, resulting in digital redlining. Furthermore, the Commission’s own study of AT&T and Frontier’s operations

(the Network Exam), the three studies referenced in the ALJ Ruling, and widely available data

33 / 67 on broadband availability and adoption levels, as well as statements made by AT&T’s CEO

(discussed in response to Question 7), all illustrate disparate investment decisions. As discussed above, these decisions tend to adversely affect communities of color. Furthermore, the

Commission’s universal service programs alone are not sufficient to address the digital redlining problem. Universal service support does little to help those who reside in communities that do not have broadband options, or which have broadband options that are low-quality or are only affordable on a short-term basis because of temporary subsidies available.

To investigate all the ways in which digital redlining is present, the Commission should seek better data. The Commission should identify per Census block, the number of broadband

ISPs that serve the households within, the broadband service data speeds advertised, as well as the speeds that are actually delivered, the prices of service offerings available to each household, and other characteristics of broadband offerings, such as data caps or bundling practices. This data can be linked to Census data on demographics by geographic region, such as the Census

Designated Place data relied on by the Commission’s Staff, as shown in Table 1 of the ALJ ruling.84

84 For example, the 2021 Annenberg study utilized Census block data on broadband availability, then aggregating that data to the Census block group level to match the demographic data availability from the American Community Survey. 2021 Annenberg Study, at 3.

34 / 67 E. ALJ Ruling Question 5

Historically, redlining has meant that some neighborhoods, generally with affluent, white residents, have access to a particular service while poorer residents do not. How should the Commission define redlining? In the context of broadband Internet service, should Internet speeds offered to residents be taken into consideration?

As discussed earlier, TURN believes that digital redlining should be identified as occurring in areas where residents do not have two providers of wireline broadband services that offer downstream services of at least 100 Mbps. Thus, the speeds offered to residents, as well as the number of service providers, should be taken into consideration. TURN recommends that the

Commission should also use metrics based on reliable, high-speed broadband service deployment and service offerings with data speeds of at least 100 Mbps,85 also keeping in mind service quality and service affordability. As the Commission approaches the digital redlining problem, TURN believes that it should establish priorities for remedies. Table C provides a general summary of priorities. Additional factors that may affect priorities are described below

Table C.

Table C: Priorities for Correcting Digital Redlining 1 Areas without broadband service at any speed. 2 Areas with a single broadband provider offering speeds less than 100 Mbps. 3 Areas with a single broadband provider offering speeds greater than 100 Mbps. 4 Areas with two broadband providers, one offering speeds above 100 Mbps; the other offering speeds below 100 Mbps.

In addition, when setting priorities, TURN believes that the Commission should consider other factors, such as the persistent lack of investment by broadband ISPs that affect historically disadvantaged areas, leading to inequitable outcomes not unlike the real estate redlining practices discussed earlier in these comments and in the Appendix. TURN believes that remedying this

85 TURN continues to encourage the Commission to address upload speeds as it considers issues of broadband deployment, including digital redlining.

35 / 67 lack of investment is a matter of social justice. As noted in the Commission’s Environmental and Social Justice Action Plan:

The CPUC will work to facilitate improved access to high-quality water, communications, and transportation services in communities with less reliable access to those services, so that the CPUC can achieve its goal of providing high quality service to all.86

TURN believes that digital redlining is consistent with the “less reliable access” envisioned in the Commission’s Environmental and Social Justice Action Plan. As a result, within the context of the priorities shown in Table C, additional weight should be given to correcting the identified deficiencies associated with ESJ communities.

The Commission must also consider cable company decisions to avoid rural areas to remedy the lack of opportunity associated with the lack of cable investment in those areas.

Another set of priorities are associated with the characteristics of services that are available in terms of reliability and service quality, the prices charged by service providers, and the existence of data caps. A complete and data-driven assessment will generate a robust investigation that would lead to workable and effective policy solutions for California communities negatively impacted by digital redlining practices.

1. Digital Redlining Should be Identified Based on the Availability of Wireline Broadband Service

For purposes of digital redlining, the availability of wireline broadband service should be identified, and not wireless broadband service, due to continuing limitations associated with

86 Goal 3 of the CPUC’s Environment and Social Justice Plan.

36 / 67 fixed and mobile wireless broadband services.87 While TURN does not believe the Commission should identify digital redlining based on any part of wireless service availability, should the

Commission choose to use this approach, any wireless alternative must be equivalent to wireline broadband service of at least 100 Mbps. Specifically, wireless broadband service quality and reliability must be the same as wireline broadband. In addition, any potential wireless broadband service must not throttle during peak periods or be subject to data caps that are more restrictive than those associated with wireline broadband services. Likewise, any wireless alternative must have prices that are similar to those associated with wireline broadband prices at similar speeds.

In other words, TURN does not believe that the Commission should allow inferior and more expensive wireless services to be counted when identifying digital redlining as such an approach would simply institutionalize, with Commission approval, digital redlining practices. The

Commission must not create a government-sanctioned two-tier solution that would all but guarantee that low-income communities continue to receive inferior broadband services.

2. Digital Redlining Should Consider Affordable and Reliable Service

TURN also urges the Commission to think broadly regarding the availability of broadband in California. Even in the areas of the state where high-quality broadband is deployed and service offerings are provided by multiple providers, high prices arising from the lack of competition make that service unaffordable for many California households. Broadband ISP pricing practices, which reflect significant market power resulting from the lack of competition

87 See e.g., Columbia Telecommunications Corporation, Mobile Broadband Service is Not an Adequate Substitute for Wireline (October 2017) https://www.ctcnet.us/wp-content/uploads/2017/10/CTC-Mobile- Broadband-White-Paper-final-20171004.pdf.

37 / 67 in broadband markets,88 are also disadvantaging Californians. According to FCC data for 2019, only 51.3 percent of California households adopted broadband at the 100 Mbps speed, and only

6.5 percent had adopted at the 250 Mbps download speed.89

Relatedly, TURN encourages the Commission to view whether a service is truly offered and available if the technical service quality of the broadband service is poor. A consistently unusable service reflects badly on whether the broadband service is truly available for paying customers. In addition, the Commission should be concerned about the communities that live through extended broadband service outages or significant degradation that impacts video quality or other dynamic two-way applications compared to other communities whose service is restored.

F. ALJ Ruling Question 6

Does the table in Section 3 of this ruling indicate redlining or some other form of systemic issue? It appears to indicate that poorer communities are more likely to be unserved, and wealthier communities are more likely to be served. Is this analysis accurate? Please explain why it is or is not accurate.

TURN believes that the analysis shown in the ALJ Ruling’s Table 1 is accurate. As discussed above, the data presented in Table 1 of the ALJ Ruling is supportive of the proposition that broadband ISPs make broadband investment decisions based in large part on the income characteristics of geographic areas. As also noted in TURN’s analysis of the data underlying

88 See, for example, “Profiles of Monopoly: Big Cable and Telecom,” institute for Local Self-Reliance, August 2020. https://ilsr.org/wp-content/uploads/2020/08/2020_08_Profiles-of-Monopoly.pdf . 89 Inquiry Concerning Deployment of Advanced Telecommunications Capability to All Americans in a Reasonable and Timely Fashion, GN Docket No. 20-269, Fourteenth Broadband Deployment Report, FCC 21-18, Appendix G at 284 (January 19, 2021).

38 / 67 Table 1, shown in Table B, above, the Staff data set also supports the proposition that household income, customer density, and the presence of anchor institutions influence broadband speeds in

Census Designated Places. Table B shows that each of the variables contained in the data set provided with the ALJ Ruling (Median Household Income, the Number of Anchor Institutions, and the Density of the geographic area) have a direct relationship with the number of households in each geographic area that are served at 100 Mbps.

TURN’s regression analysis shows that the Median HH Income variable shows a positive sign and is highly statistically significant. This result supports the proposition that the higher is household income in a geographic area, the higher the likelihood that the area will be served at

100 Mbps. The income coefficient in Table A suggests that an increase in median household income of $10,000 in an area leads to a 2.3 percent increase in households served at 100 Mbps.

Similarly, for the number of anchor institutions in a geographic area, the coefficient for anchor institutions suggests that the higher the number of anchor institutions in a geographic area, the higher the likelihood that the area will be served at 100 Mbps. Likewise, the coefficient for customer density indicates that the higher the customer density in a geographic area, the higher the likelihood that the area will be served at 100 Mbps. These last two factors are consistent with the observed digital divide between urban and rural areas, as rural areas typically have fewer anchor institutions and have lower population density.

39 / 67 G. ALJ Ruling Question 7

Are there other studies or analysis that parties wish to submit for the record in this proceeding?

Yes, as TURN suggests above, the Network Exam Study, the 2021 CETF Survey, and the

2021 Annenberg study are relevant and contain updated information that is helpful for this proceeding to understand the digital redlining mutations that persist in the state.

1. The Commission Should Include the 2019 Network Exam Study of AT&T as a Data Source in this Proceeding, and Update the Study

In addition, the three studies the ALJ ruling provided for comment, TURN suggests that the CPUC also consider its own Network Exam Study. The Commission has evidence that

California’s largest telephone company, AT&T, fails to invest in lower income communities.

According to the Examination of the Local Telecommunications Networks and Related Policies and Practices of AT&T California and Frontier California (Network Exam Study) ordered by the Commission and released in 2019, “AT&T wire centers that have been upgraded with fiber optic facilities and other broadband-related investments disproportionately serve higher income communities.”90 As noted in the Network Exam Study, not only do lower income communities suffer from a lack of investment, but the lack of investment in these communities also results in poor service quality performance.91

The methodology associated with the Network Exam Study, if updated with current information, would provide the Commission a roadmap to prioritize ILEC investment agendas.

90 Examination of the Local Telecommunications Networks and Related Policies and Practices of AT&T California and Frontier California, (Network Exam Study), conducted pursuant to R.11-12-001, D. 13- 02-023 and D. 15-08-041 at 2. 91 Id.

40 / 67 By identifying areas where investment has not been pursued, and where broadband performance does not match the Governor’s 100 Mbps mandate, the Commission can ensure that digital redlining is eliminated in ILEC service areas. The Network Exam Study offers conclusions that are important to the Commission’s current assessment of digital redlining.

The Network Exam Study found that:

• AT&T and Frontier suffered from persistent disinvestment during the period 2010 and 2017.92 While data from this period may seem dated, it is important to consider as the state’s two largest telephone companies, by failing to make capital investments in their local networks during that period, reduced the potential for expanded broadband deployment and competition.

• AT&T lagged in the deployment of fiber facilities in its network, resulting in an outside- plant network that “is still largely copper based.”93 This lack of fiber deployment reduces the potential for expanded broadband availability and competition.

• AT&T has deployed fiber optic facilities in “roughly half of its California wire centers,” and these facilities are “primarily in the feeder plant supporting Fiber-to-the-Node (“FTTN”) architecture.”94 This limited fiber deployment reduces the potential for expanded broadband availability and competition.

• The Network Exam Study also found that “There is an inverse relationship between household income and wire center service quality performance. AT&T wire centers that have been upgraded with fiber optic facilities and other broadband-related investments disproportionately serve higher income communities.”95 This focus on high-income communities has disadvantaged lower-income communities by reducing broadband availability and competition.

92 Network Exam Study at 1. 93 Network Exam Study at 9. 94 Network Exam Study at 11, emphasis added. 95 Network Exam Study at 2.

41 / 67 • The lack of investment in rural areas served by AT&T and Frontier was also documented in the Network Exam Study. The study found that “the distance from the Central Office to many users is well beyond 18,000 feet resulting in long loops.”96 These long loops reflect a lack of investment that will result in low-quality broadband, if any broadband at all.97

• The Network Exam Study also found that “In rural areas, customers have fewer (if any) competitive options.”98 Thus, the lack of investment by AT&T and Frontier in rural areas leaves rural communities with either no broadband at all, or low-quality broadband.

These findings from the Network Exam Study, based on detailed geographic analysis using data from the service providers, offers significant evidence of digital redlining in the service areas of

California’s two largest telephone companies. Given that these companies are also broadband

ISPs, and that their ability to offer high-quality broadband services depends on investments in fiber optic broadband facilities, TURN believes that the Network Exam Study provides evidence of one the root causes of digital redlining and the digital divide in California. These telephone companies have failed to invest in their facilities in California in a manner that serves the public interest.

This outcome is not surprising. The inequitable investment decisions of broadband ISPs, such as AT&T, are not driven by social objectives, such as goals established by the California

Legislature regarding bridging the digital divide.99 As a result, the Commission should expect that the redlining identified in the Network Exam Study will continue in California, unless countervailing actions are taken by the Commission or the legislature.

96 Network Exam Study at 36 97 The service distance limit on DSL is 18,000 feet. 98 Network Exam Study at 37. 99 See, for example, Cal. Pub. Util. Code, §871.7(c)(4).

42 / 67 It is unlikely that AT&T will voluntarily remedy its inequitable investments in broadband infrastructure in California. Rather than investing in broadband infrastructure in California,

AT&T has instead made other business decisions and investments by pouring billions into failed business deals.100 AT&T’s debt burden, generated by the failure of AT&T’s entertainment strategy, will also make it less likely the AT&T will voluntarily invest in broadband infrastructure.

In addition, AT&T has recently conceded that there is a digital divide in their service area and made a pledge to remedy the problem, “. . .we’re also addressing the digital divide, by investing $2 billion over the next 3 years.”101 TURN notes that this $2 billion pledge, while certainly welcome, is for AT&T’s entire 21-state service area. If fiber-to-the-home (FTTH) connections each require $1,000 per household in investment,102 AT&T’s commitment amounts

100 As recently noted in the New York Times: Offloading WarnerMedia would cap a reversal of deals worth hundreds of billions of dollars. Under Randall Stephenson, AT&T bought a slew of companies, including Time Warner and DirecTV, in the hopes of marrying an enormous content-creation machine with a broadband and satellite TV powerhouse. Those bets haven’t worked out, which led AT&T to sell a stake in DirecTV to TPG and its Crunchyroll streaming service to Sony earlier this year. . .

One of the biggest numbers to keep in mind: $170 billion. That’s how much debt AT&T amassed as part of Stephenson’s deal spree. It’s a burden that has made it harder to compete against streaming giants on one hand — Netflix plans to spend $17 billion on content this year, with Disney spending a similar amount — and traditional telecom rivals like Verizon and T-Mobile on the other. “AT&T Unwinds Billions’ Worth of Deal- Making,” New York Times, May 17, 2021. https://www.nytimes.com/2021/05/17/business/dealbook/att-discovery-deal.html.

101 Connecting America: AT&T’s Broadband Commitment, AT&T Press Release, April 26, 2021. https://about.att.com/inside_connections_blog/2021/connecting_america.html?source=En0JAN0a400000 00L&wtExtndSource=digital_divide-att-article2. 102 Based on statements made by AT&T CEO John Stankey in a January 27, 2021, earnings call, because AT&T, has already deployed fiber to the node, $1,000 appears to be a conservative estimate. According to Stankey: “We had the conversation last quarter. I don’t want to repeat myself, but we shared with you that this build is a little bit different than when we initially started because we have wire centers that are already fiber capable. The infrastructure is in place. We’re going back in and picking up the next adjacent neighborhood or the next successive area. And as a result of that, the speed to get up and moving is not

43 / 67 to two million newly connected homes—across AT&T’s entire 21 state service area. TURN also notes that the magnitude of AT&T’s pledge pales in comparison to the approximately $180 billion in investments in the now failed and unwound DirecTV and Time Warner mergers. It is also notable that compared to AT&T’s scale, the much smaller Frontier Communications of

California has recently agreed to invest $1.75 billion in its California network over the next four years to expand fiber broadband availability (with fiber to be deployed to an additional 350,000 customer locations) and to improve service quality.103 That AT&T pledges to invest $2 billion in

21 states while Frontier commits to invest $1.75 billion in California alone demonstrates the lack of commitment on AT&T’s part to close the digital divide.

Furthermore, AT&T indicates that it is in no hurry to invest, which points to the potential for the redlining problem to persist. AT&T’s CEO John Stankey conceded earlier this year that

AT&T’s rural areas will not be seeing investment from AT&T any time soon:

There is always the first third of the customer base that’s kind of the low-hanging fruit slam dunk that is the part of the customer base that you cut your teeth on, get your processes squared away on, get the vendor community up the first part of the cost curve. And it’s like the no-brainer economically, and you kind of make your decision that that’s where you’re going to start. And I would say, we went through that largely through that first third in the investment we’ve made over the last several years.

the lift that it was the first time we started ramping up on this. So, we’ve got a little bit smoother dynamic around it than what you might think, because of the increase in the fiber dynamic.” Emphasis added. https://seekingalpha.com/article/4401349-t-inc-t-ceo-john-stankey-on-q4-2020-results-earnings-call- transcript . See also, for example: “CenturyLink: FTTP deployment costs range from $500-800 per home,” Fierce Telecom, August 17, 2016. https://www.fiercetelecom.com/telecom/centurylink-fttp- deployment-costs-range-from-500-800-per-home ; 103 Application of Frontier Communications Corporation, Frontier California Inc. (U 1002 C), Citizens Telecommunications Company of California Inc. (U 1024 C), Frontier Communications of the Southwest Inc. (U 1026 C), Frontier Communications Online and Long Distance Inc. (U 7167 C), Frontier Communications of America, Inc. (U 5429 C) For Determination That Corporate Restructuring Is Exempt From or Compliant With Public Utilities Code Section 854, A. 20-05-010, Joint Motion for Adoption of Settlement Agreement, December 24, 2020, p. 9.

44 / 67 Then what happens is as you get your -- up the learning curve, and the vendors start to scale and you start to get confidence in your economics and your business processes start to improve, and you start to see the marginal economics improve on the subscriber, which is what you’re going to start to see happen on some of those broadband numbers on the fiber base that we’ve broken out for you as we move forward here over time, the next third becomes the opportunity. And we’re into that segment -- the middle innings of the game, if you want. It’s the next 30% to 60% of the customer base that we can work our way through, and you look at the economics around it that makes sense.

Then you get into that last third, and it’s always the hardest part. And I think, in many instances, you’re still sitting with the last third that in parts of the rural America that still really don’t have effective broadband options.104

AT&T’s lack of investment in broadband infrastructure in the state denies some

Californians any access to high-quality broadband services, resulting in highly inequitable outcomes. This same problem also diminishes competition and investment incentives for cable companies in the areas of AT&T’s service territory where cable has chosen to deploy broadband services.

In conclusion on the Network Exam Study, this study lends additional support to the findings of the other three studies identified by the ALJ ruling. Combined, these four studies provide the Commission with significant evidence of digital redlining in California. The

Network Exam Study provides ample evidence of ILECs’ unwillingness to invest in broadband in California’s low-income communities.

104 AT&T CEO John Stankey in a January 27, 2021, earnings call, emphasis added. https://seekingalpha.com/article/4401349-t-inc-t-ceo-john-stankey-on-q4-2020-results-earnings-call- transcript.

45 / 67 2. Cable Business Models Contribute to the Redlining Problem

TURN also urges the Commission to review the analysis about the cable business models. Cable broadband providers select their service territories based on the calculated objectives of their business model, not on social goals. Many rural areas of California do not have cable service available.105 Figure 1 shows the DOCSIS 3.1 service areas for the three largest cable providers in the state.106

Figure 1: DOCSIS 3.1 Service Area of California's Three Largest Cable Providers

Figure 1 clearly shows that these large cable companies gravitate toward serving urban areas in southern California, the Bay Area, and larger population centers in the Central Valley, leaving large numbers of California households without the ability to choose cable broadband internet access services. These maps also show that these cable companies, even with the grant of a

105 See, for example, California Interactive Broadband Map. 106 Figure 1 was created using the California Interactive Broadband Map.

46 / 67 “statewide” franchises through legislative mandates, continue to maintain their own “turf” and fail to compete against one another. The lack of customer choice and competitive pressure leads to higher prices and lower service quality.107

H. ALJ Ruling Question 8

The Commission’s Environmental and Social Justice Action Plan has as a stated goal (Goal 3) to increase access to high quality communications services for Environmental Justice and Social Justice communities. If it is found that ISPs have engaged in redlining practices, what actions should this Commission take to ensure high quality Internet service becomes available to previously redlined communities?

TURN is encouraged that the Commission is cognizant of its Environmental and Social

Justice Action Plan, specifically Goal 3 to increase access to high quality communications services for the identified communities. As explained in these comments, the historical underpinnings of “redlining” cannot be bifurcated from the investment decisions’ disparate impacts felt by the “environmental justice and social justice communities.”

However, the focus on broadband “availability” that underlies the question is insufficient for the needs of these communities the focus of the ESJ Action Plan. This Commission has a statutory obligation to ensure that affordable and high-quality voice and broadband services are available to all Californians. This is more than generating an environment in which a wireline broadband service provider offers service at any cost, or at any speed, which is what the word

“available” alone implies.

107 As noted in the study prepared for the Commission by ETI in 2019, areas with lower levels of competition experience greater degradation of service quality. See, Examination of the Local Telecommunications Networks and Related Policies and Practices of AT&T California and Frontier California. Study conducted pursuant to the California PUC Service Quality Rulemaking 11-12-001, Decision 13-02-023, and Decision 15-08-041, April 2019, at 3. Hereinafter “Network Exam Study.”

47 / 67 AT&T’s lack of investment in California continues to be problem for California’s digital divide. Likewise, the fact that cable broadband providers largely self-define their service areas also undermines the ability of all Californians to even have access to high quality broadband, never mind affordability. Given the overarching market failures, the Commission should reconsider its decision to forbear from exercising its authority over AT&T California and other large LECs. TURN believes that the Commission’s authority over ILECs could successfully address digital redlining by promoting investment in un- and underserved areas, by promoting affordable prices, and by expanding low-income programs that support broadband adoption.

VI. Conclusion

As evident from the analysis set forth in the studies cited in the ALJ Ruling, and the

Commission’s Network Exam, the impact of broadband ISP business decisions has resulted in a digital divide in California based on several elements, including race, ethnicity, and geography.

These inequitable investment decisions have denied opportunities and benefits to residents of the affected communities. Eliminating digital redlining practices will help close the digital divide and should be the highest priority of the Commission. The digital divide will be “closed” only when all Californians have access to high-quality and affordable broadband services.

TURN continues to encourage the Commission to think broadly about its authority and responsibility to ensure that all Californians have access to high quality broadband telecommunications services, as the Legislature has directed. The COVID-19 crisis has emphasized the fundamental need for ubiquitous, affordable broadband nationwide and in

California. TURN has previously outlined steps that the Commission must take to address

48 / 67 California’s digital divide.108 However, for the digital redlining problem, TURN encourages to build a comprehensive picture to quickly address the areas with the greatest need.

Dated: July 2, 2021 Respectfully submitted,

//S//

Brenda D. Villanueva The Utility Reform Network 785 Market Street, Suite 1400 San Francisco, CA 94103 (415) 929-8876 [email protected]

108 See, R. 20-09-001, Comments of The Utility Reform Network and the Center for Accessible Technology on the Commission’s Order Instituting Rulemaking, October 12, 2020.

49 / 67 Appendix A

50 / 67

Appendix A:

Redlining is Rooted in the Expansion of Homeownership

Gabriela Sandoval, Ph.D. Director of Strategic Initiatives The Utility Reform Network July 2, 2021

The history of redlining cannot be divorced from the historical factors that made homeownership possible for white working- and middle-class Americans. Prior to the Great

Depression, the federal government began encouraging homeownership among its citizens. The objective was to discourage communism by instilling in the US populace a commitment to capitalism through the acquisition of private property—in this case, homeownership. However, prior to the policies enacted as part of the New Deal, homeownership was not attainable for most families. If a borrower could access a home loan, it required a 50 percent down payment, was usually a short term (5 to 7 years), interest-only loan with a balloon payment at the end of the life of the loan. If, at that point, the family was unable to make the balloon payment they would be required to take out another short term loan.1

The Home Owners’ Loan Corporation (HOLC) was created by the federal government in

1933, during the Great Recession and as part of the New Deal, to help homeowners avoid foreclosure. The HOLC provided low-interest, amortized loans that made it possible for the first time in US history for working- and middle-class residents to pay for a home over 15 to 25 years and to build equity.

1 Richard Rothstein, The Color of Law, Liveright Publishing, 2017, pp. 63.

51 / 67 The agency hired real estate appraisers to create a risk assessment system in order to evaluate the level of risk that a borrower might default on their loan. Neighborhoods were designated as “Best”; “Desirable”; “Declining”; and “Hazardous.” With those neighborhoods identified as “best” outlined on the maps in green and those deemed “hazardous” outlined in red—thus, the term “redlining”.

The Home Owners’ Loan Corporation (HOLC), a now-defunct federal agency, drew maps for over 200 cities in order to document the relative riskiness of lending across neighborhoods. Neighborhoods were classified based on detailed information about housing age, quality, occupancy, prices, and other related risk-based characteristics. However, non-housing characteristics such as race, ethnicity, and immigration status were influential factors as well. Since the lowest rated neighborhoods were drawn in red and often had the vast majority of African American residents, these maps have been associated with the so-called practice of “redlining” in which borrowers are denied access to credit due to the racial composition of their neighborhood.2

Black people were designated high-risk borrowers simply because of their race and irrespective of their credit rating and integrated. Black neighborhoods were regarded as “high risk” simply by virtue of their racial heterogeneity and because they were not homogenously

White. In contrast, only all-White communities were designated as “Best,” meaning low-risk for loans. This, in combination with the 1934 National Housing Act that created the Federal Housing

Administration, which insured home loans with a federally backed guarantee, spurred White suburbanization throughout the country while simultaneously concentrating Black residents into

2 Aaronson, Daniel, Hartley, Daniel, Mazumder, Bhaskar, “The Effects of the 1930s HOLC ‘Redlining’ Maps”, Federal Reserve Bank of Chicago Working Paper 2017-12, Revised August 2018. https://fraser.stlouisfed.org/files/docs/historical/frbchi/workingpapers/frbchi_workingpaper_2017-12.pdf See also: Richard Rothstein, The Color of Law, Liveright Publishing, 2017, pp. 63-64. Bakelmun, Ashley and Shoenfeld, Sarah, “Open Data and Racial Segregation: Mapping the Historic Imprint of Racial Covenants and Redlining on American Cities,” in Open Cities | Open Data—Collaborative Cities in the Information Era. Palgrave, 2020. Kushner, James, “Apartheid in America: An Historical and Legal Analysis of Contemporary Racial Residential Segregation in the United States,” 22 Howard Law journal 547, 1979.

52 / 67 the least desirable housing and sorting other racial groups into their other homogenous neighborhoods as well.

White home buyers were almost exclusively granted loans to live in newly constructed suburbs. Black homebuyers were not only locked out of these new suburbs and other

“greenlined” neighborhoods, they were systematically denied FHA-backed loans. Black families were unable to buy a home unless they were willing to accept high-interest loans that were not guaranteed by the FHA. These loans did not allow them to build equity and their homes could be repossessed if they missed just one loan payment.3 In the first 28 years of its existence, the FHA backed more than $120 billion in new housing. Ninety-eight percent of these loans went to White home buyers and less than two percent went to Black and other homebuyers of color.4

There are several aspects of redlining the Commission must keep in mind with respect to redlining:

• Redlining was an explicitly racially discriminatory project with the intent of

creating and maintaining racially homogenous communities.

Redlining was a very specific practice rooted in blatant discrimination that began with federal government policy and was perpetuated by private markets. The objective of this experiment in increasing white homeownership with federally-backed home loans was clear: to spatially separate racial and ethnic groups—particularly Black and White populations.

Although the HOLC stopped operating in 1951, the maps created by the real estate appraisers for HOLC continued to be used by the Federal Housing Authority as it insured

3 Richard Rothstein, The Color of Law, Liveright Publishing, 2017, pp. 97. 4 George Lipsitz, The Possessive Investment in Whiteness: How White People Profit from Identity Politics, Temple University Press, 1998.

53 / 67 mortgage loans almost exclusively for White borrowers in newly built (greenlined) suburbs and explicitly and intentionally segregated, not only neighborhoods but whole regions of the United

States. These “federal mortgage guarantees . . .were promulgated locally and spread through the private market.”5

• Redlining was not used throughout the United States.

The HOLC maps and their color-ranked risk assessment, later used by the FHA and subsequently by private real estate interests did not include the whole country. The original maps focused on ranking neighborhoods in 200 metropolitan areas.

• Redlining was not restricted to existing housing stock.

Because redlining was used to encourage White homeownership in what would become new all-

White suburbs, many communities that would eventually be “greenlined” were not yet in existence in the 1930s. Rather they would be built—indeed, mass-produced—as part of the push to increase homeownership and create homogenous White suburbs free of “racial strife”— especially after World War II when Veteran Affairs home loans were provided for White veterans but not Black veterans.

• Although we have made some strides toward integration, the legacy of redlining

remains largely relevant—and intact—today.

5 Stephen Menendian, Arthur Gailes, and Samir Gambhir, The Roots of Structural Racism: Twenty-First Century Racial Residential Segregation in the United States (Berkeley, CA: Othering & Belonging Institute, 2021). https://belonging.berkeley.edu/roots-structural-racism.

54 / 67 The Fair Housing Act of 1968 finally prohibited racial discrimination in housing and began to remove barriers to residential integration, but as Menendian, et al. note:

following the passage of the federal Fair Housing Act in 1968, residential integration increased significantly between 1970 and 1980,11 to such an extent that many reasonable observers felt that the residential patterns established in the early and middle decades of the twentieth century might actually fade away in time. Previously all-white neighborhoods changed complexion as non-white neighbors arrived, and vice versa.12 Although progress incrementally slowed each subsequent decade, the in-migration of people of color into the suburbs—especially between 1990 and 2000—seemed to suggest a different and more hopeful racial trajectory, such that two economists declared the “End of the Segregated Century.”13 The downward trend of residential segregation, at least as popularly measured, seemed to portend eventual widespread residential integration. But . . . these encouraging observations turned out not to reflect the actual dynamics of what was occurring. In most regions, segregation was in fact increasing.6

Ultimately, their report finds that 83 percent of areas redlined in the 1930s remained highly segregated as of 2010. 7

6 Stephen Menendian, Arthur Gailes, and Samir Gambhir, The Roots of Structural Racism: Twenty-First Century Racial Residential Segregation in the United States (Berkeley, CA: Othering & Belonging Institute, 2021). https://belonging.berkeley.edu/roots-structural-racism, p. 3. (citations omitted)

7 Id., p. 19.

55 / 67 Additional notes: the CPUC again has an opportunity to be a national leader by weighing in on the legality of this type of discrimination (see Galperin p. 2)

56 / 67 Appendix B

57 / 67 Government Information Quarterly xxx (xxxx) xxx

Contents lists available at ScienceDirect

Government Information Quarterly

journal homepage: www.elsevier.com/locate/govinf

Who gets access to fast broadband? Evidence from Los Angeles County

Hernan Galperin a,*, Thai V. Le b, Kurt Daum b a Annenberg School for Communication, University of Southern California, 3502 Watt Way, Los Angeles, CA 90089-0281, United States of America b Sol Price School of Public Policy, University of Southern California, 650 Childs Way, Los Angeles, CA 90089, United States of America

ARTICLE INFO ABSTRACT

Keywords: Regulatory and market changes in residential (fixed) broadband have raised concerns about Internet Service Broadband deployment Providers (ISPs) prioritizing investments in the most profitable areas, thus relegating low-income and minority Telecommunication policy communities to fewer broadband options and legacy networks. This study examines these concerns for Los Urban inequality Angeles (LA) County during the 2014–18 period. The analysis uses rollout data collected by the California Public Concentration effects Utilities Commission (CPUC) in combination with demographic information from the American Community Survey (ACS). Because the spatial distribution of broadband investments cannot be directly observed, compe tition and the availability of FTTH services are used as proxies. The findingsindicate that competition and fiber- based services are less likely in low-income areas and minority communities, with the most severe deficits observed in census block groups that combine poverty and a large share of Black residents. We outline alternative policy tools to address intracity inequalities in broadband investments in the conclusion.

1. Introduction significantly less likely to have Internet service at broadband speeds (Holmes & Wieder, 2016). Despite the media attention generated by High-quality, affordable broadband is as critical to the social and these and other reports, the issue has received surprisingly little atten economic vitality of communities as transportation and electricity were tion in the academic literature. in the 20th century. However, the private sector is responsible for most In order to explore the intracity patterns of broadband investments of broadband investments in the US. This raises fundamental questions for the 2014–18 period we create a longitudinal dataset that combines about potential underinvestment in areas of low expected returns. information on residential (fixed)Internet service availability for every Further, market consolidation and the relaxation of rules governing in census block in LA county with demographic information from the dustry organization have intensifiedconcerns that network upgrades to American Community Survey (ACS). Because the spatial distribution of fast broadband services are not reaching distressed urban communities broadband investments cannot be directly observed, we use competition (Blevins, 2019; Crawford, 2013). Underinvestment in broadband intensity and the availability of residential fiber services (FTTH) as therefore threatens to amplify urban inequality by depriving commu proxies. With this dataset, we use two empirical strategies to estimate nities of the basic infrastructure for commerce, education and civic how income and racial composition affect broadband service rollout: engagement (Mossberger, Tolbert, & Franko, 2012). first,a pooled logistic regression with error clustering, which allows for This study probes for evidence that broadband infrastructure in isolating the effect of income and race from other factors shown in vestments in Los Angeles (LA) County during the 2014–18 period have previous studies to affect broadband deployment; second, a fixedeffects favored affluent areas, thus relegating low-income and minority com estimation that further controls for unobserved heterogeneity across our munities to fewer broadband options and legacy networks. Similar units of observation. patterns have been documented by advocacy groups in several other US The findingsfrom these two strategies are consistent and support the cities. A study by the National Digital Inclusion Alliance (NDIA) suggests hypothesis that network upgrades in the 2014–18 period are associated that AT&T has failed to upgrade Internet and video services in low- with income and racial factors. Perhaps the most remarkable finding is income communities in Cleveland, OH, and Dallas, TX (NDIA, 2019). that the largest deficits are observed in areas that combine low income Similar findings have been reported for Goochland County, VA, where and a high share of Black residents. This finding is consistent with pre the Center for Public Integrity found that low-income neighborhoods are vious studies of urban segregation in Los Angeles, which emphasize the

* Corresponding author. E-mail addresses: [email protected] (H. Galperin), [email protected] (T.V. Le), [email protected] (K. Daum). https://doi.org/10.1016/j.giq.2021.101594 Received 28 February 2020; Received in revised form 29 June 2020; Accepted 7 May 2021 0740-624X/© 2021 Elsevier Inc. All rights reserved.

Please cite this article as: Hernan Galperin, Government Information Quarterly, https://doi.org/10.1016/j.giq.2021.101594

58 / 67 H. Galperin et al. Government Information Quarterly xxx (xxxx) xxx clustering of multidimensional poverty in formerly redlined neighbor In one of these early studies, Gillett and Lehr (1999) use county-level hoods, particularly in the South Los Angeles area (Matsunaga, 2008). data to analyze the factors driving investments in cable modem services. These communities have historically been bypassed by investments in Their findings suggest that these services concentrate in high-income health facilities, transportation, education and other public goods. and high-density urban areas, which the authors attribute to the ex Whereas Internet adoption could theoretically compensate for such pected diffusion pattern of new information technologies. In a similar deficits(for example by facilitating remote work, telehealth services and study, Prieger (2003) finds evidence of a rural gap in broadband avail remote learning), lagging investments in next-generation broadband ability. However, after controlling for cost factors, competition in threaten to aggravate community distress and inhibit socioeconomic tensity, and demographic variables that affect broadband demand, the development. author does not find evidence that income or racial/ethnicity factors These findings raise multiple policy questions. Whether and how influencebroadband investments. Grubesic and Murray (2005) similarly federal law requires deployment of communication facilities on a find that broadband competition is significantly weaker in rural areas. nondiscriminatory basis is the subject of much debate among legal Other studies have used more restricted datasets to address similar scholars (Baynes, 2004). Interestingly, antidiscrimination provisions in questions. For example, Prieger and Hu (2008) use a unique dataset of federal and state law have rarely been tested in the courts - possibly due DSL subscribers at the ZIP+4 level for two incumbent operators across to the lack, until recently, of appropriately disaggregated information fivestates in the Midwest region.4 Their findingssuggest that household about service availability and adoption. This strengthens the case for income is a significant driver of investments in DSL infrastructure. improving the collection and reporting of broadband data at the federal However, after controlling for income and other demand factors, race and state levels.1 Further, the findings also bear on the debate over the and ethnicity are not found to affect DSL rollout. Similarly, Kolko (2010) classification of broadband as an information service (and therefore combines ZIP code level availability data with geolocated household- more lightly regulated under Title I of the Communications Act) or as an level information from a proprietary dataset. Using households as the essential communication service (and thus subject to the more stringent unit of observation, the regression estimates suggest that income is a obligations of Title II). We elaborate on the implications of our findings strong predictor of broadband supply. for these debates and the policy tools available to redress intracity gaps With funding from the American Recovery and Reinvestment Act of in broadband investments in the conclusion. 2009, federal and state governments began collecting broadband Our findings also inform scholarship about smart cities and digital deployment data at the census block level. This change in data granu inequality. Weak competition and lack of high-quality services may larity significantly improved researchers’ ability to tease out the de partly account for the racial/ethnicity gaps in residential broadband terminants of investments in broadband infrastructure. The existing data adoption (after controlling for income and other characteristics) that collection procedure is far from perfect, and both researchers and prior studies have identified but struggled to explain (e.g., Campos- oversight agencies have exposed numerous flaws (e.g., GAO, 2018; Castillo, 2015; Fairlie, 2004; Flamm & Chaudhuri, 2007). Further, and Grubesic, 2012; Turner, 2016). In the case of fixed residential services, in the context of the current COVID-19 pandemic, our findings call for the most significant shortcoming is that an entire census block is increased attention to the unintended amplification of inequalities in considered served if a provider is able to serve a single household in that education, health and job opportunities that results from differences in block. Further, ISPs are allowed to report availability in blocks where connectivity opportunities available to residents. they could potentially offer services “within a service interval that is typical for that kind of connection—that is, without an extraordinary 2. What drives broadband infrastructure deployment and why it commitment of resources.”5 As reported by the Government Account matters? ability Office(2018), this leads to significantavailability overstatement, which affect in particular areas with large census blocks. 2.1. The determinants of broadband investments Despite these shortcomings, the availability of more granular data has significantly contributed to scholarship about the relationship be From the early days of the transition from dial-up to broadband tween broadband investments, competition intensity, community de Internet in the mid-1990s, questions have been raised about the spatial mographics and residential adoption. For example, Whitacre, Strover, distribution of investments in network upgrades to high-speed services, and Gallardo (2015) use a regression decomposition strategy to examine and about investment lags in low-income communities and rural areas. the extent to which differences in broadband availability explain the Congress addressed this question in the Telecommunication Act of 1996, rural-urban adoption gap. Their findings suggest that supply variations establishing a broad mandate that communication services be made account for over a third of the observed rural gap in residential broad available “to all people of the United States, without discrimination on band adoption. Using a matching estimator strategy, Prieger (2015) the basis of race, color, religion, national origin, or sex”.2 This broad finds that, despite widespread availability of fixed residential services policy mandate resulted in several federal initiatives to expand broad nationwide, Black and Hispanic households tend to have fewer broad band services across the US, including the Connect America Fund and band choices.6 the e-rate program. The availability of deployment data at the census block level has also The Act also mandated the FCC to collect broadband deployment shifted the research focus from the federal or state to the local level. As a data to establish “whether advanced telecommunications capability is result, digital inequality scholarship has moved beyond the traditional being deployed to all Americans in a reasonable and timely fashion”.3 demographic factors, exploring how idiosyncratic community factors Until 2010, however, the FCC collected such data at the ZIP-code level, and municipal policies affect broadband investment and adoption pat thus limiting precise measurement of the sociodemographic patterns terns. For example, Rhinesmith and Reisdorf (2017) combine FCC and market parameters affecting investment decisions made by ISPs. deployment data with information from the ACS to analyze the spatial Despite this limitation, several early studies addressed the nexus be tween broadband availability, socioeconomic factors and service adoption. 4 The ZIP+4 level is significantlymore geographically disaggregated than the five-digit ZIP code level. 5 FCC Report and Order and Second Further Notice of Proposed Rulemaking, 1 See FCC (2019). Establishing the Digital Opportunity Data Collection, WC WC Docket Nos. 19–195, 11–10, p. 3. Docket No. 19–195, released August 6, 2019. 6 Interestingly, the reverse is true for mobile broadband, though it must be 2 47 U.S.C. § 151. noted that current data collection procedures for wireless broadband are 3 47 U.S.C. § 706. notoriously imprecise (see GAO, 2018).

59 / 67 H. Galperin et al. Government Information Quarterly xxx (xxxx) xxx distribution of broadband competition and service quality differences LA county for two main reasons. First, as (Perry, 2020) argues, the across the Kansas City metro area. Their findingshighlight the value of legacy of urban segregation and redlining in Los Angeles has largely local policy efforts to address deficitsin broadband investments in low- affected Black residents. Second, Black residents are significantly more income communities. Similarly, Grubesic, Helderop, and Alizadeh spatially segregated than other racial/ethnic minorities. Consider an (2019) examine the deployment of Google Fiber in Provo, UT, and analysis based on the dissimilarity index proposed by Massey and Austin, TX, using a novel empirical strategy that contrasts results ob Denton (1988), which represents the percentage of one demographic tained with FCC data with results from a query-based data collection group who would need to relocate to another neighborhood to achieve effort at the residential address level. Interestingly, their findings indi equal distribution across an urban area. In 2018, the Black-non Black cate that fiber deployment in Provo and Austin has favored areas with dissimilarity index for Los Angeles was 0.7, compared to 0.5 for His fewer minority residents but lower median incomes, which the authors panics and Asians. Therefore, though representing a relatively small partly attribute to the presence of a large population of college students minority in LA County (relative to Asians and in particular Hispanics), in both cities. and despite significant changes in the racial composition of LA neigh ¨ borhoods in the past decade (Clark, Anderson, Osth, & Malmberg, 2.2. Connecting digital inequality and urban segregation 2015), Blacks remain spatially concentrated and overrepresented in areas affected by the legacy of racial discrimination in the postwar era. The ability to quantitatively explore how disparities in broadband provision manifest at the local level has opened multiple opportunities 3. Data and methods to connect digital inequality scholarship with the broader literature on urban segregation and broader debates about fairness in the provision of This study draws from two data sources. The first is the California essential public goods. Two particularly useful concepts from this Public Utilities Commission (CPUC), which annually collects informa literature are “cumulative adversity” and “concentration effects,” both tion from ISPs about service availability, speed and transmission tech of which refer to the intergenerational accumulation of socioeconomic nology at the census block level. This data collection initiative is disadvantages in urban neighborhoods that combine poverty and a large separate from the data collected by the FCC through Form 477. Further, concentration of underrepresented minorities (Quillian, 2012; Sampson the CPUC performs several additional validity checks, the most impor & Wilson, 1995; Wilson, 1987). Drawing from studies in various urban tant being that service availability is validated through a provider- settings, these scholars point to the compounding of disadvantages in supplied list of customers showing their address and subscribed speeds specific urban clusters, which is both caused by and contributes to un (CPUC, 2016). This validation procedure significantly reduces the derinvestment in critical infrastructure such as transportation, health overstatement of availability found in FCC data. facilities and sanitation. It is important to note that our analysis is limited to the availability of For example, Massey (1990) argues that the multiplicative effects of residential (wired) broadband services, and thus excludes mobile poverty and racial segregation in the housing market during the postwar broadband and fixed terrestrial wireless alternatives. A thorough dis era has resulted in inner-city communities that are more prone to cussion about whether fixed and wireless broadband are truly separate deteriorating infrastructure, violence, fractured social networks, and markets is beyond the scope of this study. Our premise, following the more limited access to quality public goods. This cycle has had a FCC’s latest broadband deployment report (FCC, 2020), is that con particularly negative impact on Black residents, who regardless of actual sumers continue to use both services concurrently and in distinct ways, income are significantly more likely to live in less affluent neighbor thus suggesting complementarity rather than substitution. hoods, and therefore with fewer access opportunities to quality educa Using the CPUC residential broadband deployment data, we create a tion, transportation and other critical resources. series of new variables, including the total number of unique ISPs of Several studies have examined these clustering effects in the Los fering broadband speeds and the number of ISPs offering fiber services Angeles area. Matsunaga (2008) finds that South Los Angeles and the (e.g., FTTH) in each census block.7 We use these variables as proxies for adjacent downtown neighborhoods represent areas of concentrated investments in residential broadband, which cannot be directly poverty with disproportionately large shares of Hispanic and Black observed. The dependent variables to be estimated are thus: residents. In these areas, deficits in transportation, housing, education, and other public goods combine with poverty to perpetuate community 1) whether broadband competition exists in a particular census block distress. Kneebone and Holmes (2016) show that these trends have (indicated by the presence of two or more ISPs offering broadband exacerbated since the Great Recession. Others examining concentration speeds); effects in Los Angeles focus on the resource environment as a source of 2) whether residential fiberservices are available in a particular census environmental and health injustice. For example, Wolch, Wilson, and block (indicated by the technology and speed reported by ISPs Fehrenbach (2005) findthat families in low-income, majority Black and servicing that block). Hispanic areas such as South LA have significantly more limited access to parks while other studies report similar deficits with respect to We combine this data with sociodemographic information from the healthy food options (Lewis et al., 2005). American Community Survey (ACS). With a current sample size of about Drawing on this body of literature, we examine whether the same 3.5 million households, the ACS allows for reliable estimates at the patterns of infrastructure underinvestment exist in the case of broad census block group level, which on average contains about 40 census band. In other words, we probe for evidence of concentration effects blocks (or about 1500 residents). To join these datasets at the same whereby formerly redlined areas of Los Angeles that combine low in geographical scale, we aggregate the block-level CPUC data to the block come and a disproportionate share of underrepresented minorities are group level. Despite the fact that in a highly urbanized metro area such less likely to receive investments and network upgrades to fast broad as Los Angeles the typical census block is relatively small (about 30,000 band services. A key component to these investments is fiber.As several square feet), this aggregation inevitably results in overestimation of scholars argue, there is no viable path to a faster and more robust service availability. Our results should thus be interpreted as upper Internet for households and businesses without significant investments bound estimations. Further, as is the case for most spatial statistical that bring fiber infrastructure closer to the end-user (Crawford, 2018; Grubesic et al., 2019). Even the new generations of ultrafast mobile technologies (e.g., 5G) depend on fiberbackhaul deployment to connect 7 Following the FCC definition we consider broadband an Internet access base stations to the public Internet. service with advertised speeds of at least 25Mbps for data download and 3Mbps Our analysis pays particular attention to historically Black areas in for data upload (FCC, 2015).

60 / 67 H. Galperin et al. Government Information Quarterly xxx (xxxx) xxx inquiries involving aggregate data, we acknowledge that our results are strategies. The first is a fixed-effects panel data specification that esti vulnerable to the ModifiableAreal Unit Problem (MAUP), definedas the mates the effect of race and income on the two outcomes of interest sensitivity of estimates to zonation (i.e., the arbitrary nature of block (broadband competition and fiber availability), conditional on de group boundaries) and scale (i.e., the size of block groups) (see mographic factors. The regressors are lagged one period in order to Fotheringham & Wong, 1991; Green & Flowerdew, 1996; Oberwittler & account for the prolonged investment cycle involved in broadband Wikstrom,¨ 2009). network deployment.Formally, the model is: Our full dataset contains 32,135 observations, which correspond to ) ′ 1 6427 block groups observed over fiveyears from 2014 to 2018. For each Pr(Yit = 1|Xit 1, β, αi = 1 + e αi xit 1β observation, we estimate the probability of observing our two outcomes of interest (the presence of broadband competition and the availability [ ′ ] Yit = 1 X β + αi + εit > 0 of fiber services), conditional on a series of demographic factors that it 1 previous studies have shown to affect broadband investments (Flamm & where Y is the binary outcome of interest for census block group i in Chaudhuri, 2007; Hauge & Prieger, 2010; Whitacre et al., 2015). These it year t, X’ is the vector of (lagged) block group characteristics, α is the include population density, racial composition, median household in it-1 i time-invariant error term for block group i, ε is the logistically come (logged), housing value (logged), median age, education (per it distributed, time-varying error term and β is the vector of coefficients centage of population with bachelor’s degree or higher), the percentage that are estimated through maximum likelihood. of households with children under 18 years old, and the percentage of This strategy best approximates the effect of income and racial fac English-only households. tors on the spatial distribution of broadband investments by controlling In order to test for the concentration effects discussed in the previous for time-invariant unobserved differences across blocks groups (such as section, we create an interaction between the share of Black residents in topography and local rules for civil works) known to affect broadband a block group and whether the block group is low income, using the rollout (Greene, 2003). These unobserved differences between block bottom quartile of median household income as a proxy for low income. groups are particularly relevant in Los Angeles County, which contains Broadly speaking, this term tests the hypothesis that, above and beyond 88 incorporated cities over a 4083 square-mile area with considerable the separate effect of income and race factors, broadband investments variation in topography. Results from a Hausman test confirm that the lag in areas that combine poverty and a disproportionately large share of fixed-effects specification is preferred over a random-effects Black residents. Table 1 contains additional details about the model specification. variables. At the same time, this empirical strategy results in considerable in The empirical analysis is based on two reduced-form modelling formation loss, primarily because the fixed-effects estimator only uses information from block groups for which changes in the outcome vari Table 1 ables are observed during the study period. For example, there are 4834 List of variables and sources. block groups (from a total of 6427) for which no change in fiber avail Type Variable name Variable explanation Source ability status is observed between 2014 and 2018 (in order words, fiber Dependent Broadband Whether block group has CPUC service was either available or unavailable in these blocks throughout Variable Competition more than one ISP offering the entire period). These observations are thus dropped from the fiber broadband speeds (YES = model estimations, thus resulting in significant information loss. 1) We therefore use a second estimation strategy that, while not con Availability of Fiber- Whether FTTH services are CPUC to-the-Home Services available in block group trolling for unobserved heterogeneity across block groups, uses infor (YES = 1) mation from all observations. These second set of estimates are based on Independent Concentration Effects Interaction between Black ACS a pooled logit specification with robust standard errors clustered at the Variable of of % Black and Low- residents (%) and low- block group level. While this strategy essentially ignores the panel Interest Income income block group (YES = 1) structure in the data, the clustered errors account for correlation be Covariates Population Density Population per square mile ACS tween errors from repeated observations across periods (Wooldridge, in block group 2001). Results for both empirical strategies are reported and discussed in % Black Percent of block group ACS the following sections. population that is Black % Non-white Percent of block group ACS Hispanic population that is non- 4. Results white Hispanic % Asian Percent of block group ACS 4.1. Descriptives and trends in broadband infrastructure rollout population that is Asian Median Age Median age of block group ACS – population During the 2014 18 period there is evidence of a significant % English-only Percent of block group ACS expansion in fixed residential services in LA County. The share of resi Households households that report dents served by at least two ISPs offering broadband speeds increased speaking English only from 65.5% in 2014 to 90.4% in 2018. This represents an additional 2.5 Presence of Children Percent of block group ACS M residents who can choose from high-speed residential plans offered by households with at least one child (individuals less competing ISPs. Similarly, the share of residents served by FTTH ser than 18 years old) vices increased from 26.4% in 2014 to 49.7% in 2018, thus suggesting % Bachelor or Higher Percent of block group ACS robust investments in gigabit-level services (Fig. 1 left panel). population with a At the same time, other results raise concerns about weakening bachelor’s degree or higher competition and investments. First, the rate of expansion of broadband Median Income (log) Block group median ACS household income (logged) competition appears to be trending down (Fig. 1 right panel). This is Housing Value (log) Block group housing value somewhat to be expected as competition approaches full population (logged) coverage, and likely reflects(at least in part) topography and regulatory Low Income Area Whether the block group ACS challenges in deploying service to the less than 10% of residents who median household income falls in the bottom quartile lack broadband choice. By contrast, the slowdown in residential fiber (YES = 1) rollout raises questions as coverage remained below 50% in 2018.

61 / 67 H. Galperin et al. Government Information Quarterly xxx (xxxx) xxx

Fig. 1. Trends in broadband competition and residential fiber availability, 2014–2018.

Whether this reflects a temporary slowdown or a long-term trend re We observe that most control variables take the expected sign. mains to be seen. The most recent FCC Broadband Deployment report Perhaps surprising is that education level (bachelor’s degree or higher) (FCC, 2020) notes that homes passed by fibergrew nationally by 16% in and population density are not significant predictors of broadband 2019. At the same time, industry analysts note that large operators have competition, while the share of Asian residents has a positive effect. We scaled-down fiberefforts to favor 5G deployment (Falcon, 2019), though return to these results in the next section. other reports suggest that new entrants are filling in the void (FCC, Turning to our main variables of interest, we findthat the probability 2020). of observing competition decreases with the share of Black residents Second, there is evidence that the residential internet access market while increasing with household income and housing value. To quantify is increasingly characterized by duopoly competition. Between 2014 the magnitude of these effects, Fig. 2 offers a visual representation of the and 2018 the share of residents able to choose between three or more results based on conditional predictions across the range of values for competing ISPs dropped by about half from 10% to 5.7%. Further, these two variables. As shown (left panel), the probability of competi considering the share of residents who lived in areas where the number tion between two or more ISPs in a census block group is about 77% in of ISPs increased, decreased or remained unchanged between 2014 and areas with a small share of Black residents, dropping to about 68% in 18, the data shows that 11.5% of residents (about 1.2 M) saw a reduction traditional Black neighborhoods. Similarly (right panel), in low-income in the number of local competitors. Several studies have noted similar block groups the probability of broadband competition falls below 70%, trends nationally, pointing to lack of regulatory incentives that promote climbing above 80% in the more affluent areas (notice the effect is effective broadband competition in the US in comparison to other nonlinear by construction of the logged income variable). Although advanced countries (Flamm & Varas, 2018; Frieden, 2009). competition levels are relatively high overall, the results suggest that As a prologue to the presentation of results in the next section, a table low-income and Black residents have fewer broadband options, which is of summary statistics is presented below (Table 2:). typically associated with lower quality services and higher prices. In order to test for the concentration effects discussed in section 2.2., 4.2. Broadband competition Model 2 introduces a term that captures the interaction between the share of Black residents and whether the block group is low income, 4.2.1. Pooled logit results using first (bottom) median household income quartile as a proxy. The We begin by examining results from the pooled logit models for results reveal several interesting patterns. First, the bottom income broadband competition, our firstproxy variable. Model 1 estimates the quartile variable absorbs the effect of median household income, thus likelihood of competition in a block group controlling for population suggesting that underinvestment is clustered in high-poverty areas. density and household characteristics. As noted, while this empirical Further, the interaction term also absorbs the effect of Black, which in strategy ignores the panel structure of the data, standard errors are Model 2 is not significant. At the same time, the interaction term is clustered at the block group level to account for correlation between highly significant, thus validating the hypothesis that competition is observations across periods. The results are presented in Table 3. particularly lacking in areas that combine poverty and a large

Fig. 2. Predicted probability of broadband competition by HH income and share of Black residents (95% confidence intervals).

62 / 67 H. Galperin et al. Government Information Quarterly xxx (xxxx) xxx concentration of Black residents. above, remains negative and significant at p < 0.01. This validates the To illustrate these interaction effects, Fig. 3 replicates the left panel hypothesis that new broadband investments are not reaching histori plot in Fig. 2 (probability of broadband competition along the share of cally Black, low-income communities. Black residents), but divides block groups into low income (bottom quartile) and not low income (that is, the remaining three quartiles). In 4.3. Fiber availability other words, the figure compares the likelihood of broadband compe tition between low-income and more affluent areas along the share of 4.3.1. Pooled logit results Black residents in each block group. As shown, while the probability of Results for fiber availability, our second proxy variable, generally observing competition is higher and relatively similar in affluent areas point in a similar direction. As expected, the probability of observing regardless of the share of Black residents, the same probability falls fiberavailability decreases with the share of Black residents. The results rapidly in poor communities as the share of Black residents increases. also indicate that fiberis more likely to be available in the more affluent Notably, it falls below 50% in low-income block groups that concentrate areas. However, contrary to the results for broadband competition dis large Black populations. cussed above, the interaction term between the share of Black residents and bottom income quartile is not significant(Model 6), suggesting that 4.2.2. Fixed-effects logit results race and income independently (but not jointly) affect the probability of The results above suggest that intracity variations in investments in observing fiber availability. fast broadband are associated with race and income factors. These re Fig. 4 quantifies the effect magnitude of racial and income factors sults however do not account for unobserved differences across block based on the predicted probability of observing fiberalong the range of groups that may affect broadband rollout. This include important cost values for the two variables of interest (the share of Black residents and factors such as topography and permitting laws as well as other demand median household income). As shown, the probability of fiber avail factors not captured in the pooled logit models. By contrast, Table 4 ability is about three times lower in majority Black block groups relative presents results from the fixed effects estimations, which as discussed to comparable areas with fewer Black residents (left panel). Similarly, take advantage of the panel structure of the data to control for unob while in the more affluent areas the likelihood of observing fiber are served, time invariant heterogeneity across units of observation. approaching 1 in 2, in the less affluent areas they stand at about 1 in 5 The fixedeffects estimations generally validate the findingsfrom the (right panel). pooled logit models, with minor differences in coefficients for the con trol variables. While the predictor variables in Models 3 and 4 are 4.3.2. Fixed-effects logit results identical to those in Models 1 and 2, it is worth noting that results from The fixed effects estimations for fiber availability are presented in Models 3 and 4 are based on a smaller sample of 8464 observations Table 6. The results are qualitatively similar to those in Table 5, with (2206 block groups observed over 4 periods), which corresponds to some variations in effect size and statistical significance. Interestingly, block group where competition status changed (in either direction) over we observe that while the positive effect of the Asian and Hispanic the study period. Note for example that education and the presence of variables disappears, the share of Black residents remains a strong children in the household are now, as expected, positively correlated negative predictor of fiber services. In turn, household income has a with broadband competition, while the share of Asian residents is no positive independent effect in Model 7, which is nonetheless absorbed longer significant. by the introduction of the interaction term in Model 8. Overall, the fixed Turning to our main variables of interest, we observe that household effects estimates validate the main hypothesis about broadband in income no longer has an independent effect on broadband competition, vestments being associated with income and racial factors, with although housing value remains a strong predictor, which suggests a particularly adverse effects found for low-income Black communities. correlation between broadband investments and areas that experienced faster gentrification(and thus stronger growth in housing values) during 5. Discussion and limitations the study period. The dummy variable for low income areas (bottom income quartile) remains significant, thus suggesting an independent This study examines the correlation between demographic factors effect on the likelihood of broadband competition (model 4). Notably, and broadband service rollout in LA County over the 2014–18 period, the interaction term between bottom income quartile and the share of using competition and fiber availability as proxy variables. Using two Black residents, which captures the concentration effects discussed different empirical strategies, we findconsistent evidence that racial and income factors are associated with the spatial distribution of broadband investments. In particular, we find that competition between two or more ISPs offering services at broadband speeds is less likely in low- income areas as well as in areas with a large share of Black residents. Similarly, these areas are also less likely to be served by residential fiber services capable of delivering gigabit-level speeds. In addition, the study explores concentration effects that generate particularly adverse outcomes for areas that combine these demographic attributes. As noted, many scholars have linked the legacy of racial discrimination in the postwar era to the spatial concentration of eco nomic distress in particular neighborhoods (Massey & Fischer, 1999; Sampson, 2017). A key barrier to social mobility in these communities is the low quality of the public goods provisioned to residents, including public transportation, public safety, and public education. If—as many have argued—information is the foundation of the modern economy, deficitsin the provisioning of high-quality broadband can be expected to have a similarly adverse impact on the opportunities afforded to small businesses and residents in these areas. Overall, our findingssuggest that these concentration effects exist and adversely affect broadband infra Fig. 3. Predicted probability of broadband competition by HH income and structure in predominantly Black, low-income neighborhoods. share of Black residents (95% confidence intervals). Unfortunately, the lack of reliable data about broadband prices and

63 / 67 H. Galperin et al. Government Information Quarterly xxx (xxxx) xxx

Fig. 4. Predicted probability of fiber availability by HH income and share of Black residents (95% confidence intervals). actual service speeds (as opposed to advertised speeds) limits our ability from the dissimilarity index as well as the variable descriptives in to quantify the direct impact of these broadband infrastructure gaps. Table 2). This makes the potential effect of the Hispanic variable harder However, both basic economic principles and evidence from other to detect. Further, because the fixedeffects estimations capture changes studies suggest that the implications of weak competition and un in predictors over time, differences in the rate of neighborhood gentri derinvestments in next-generation broadband infrastructure are poten ficationacross racial lines may also account for these results.Ultimately, tially far-reaching (see Kongaut & Bohlin, 2017). For example, studies further studies are needed to untangle this question. suggest that deficits in network infrastructure are likely to limit the ability of firms in distressed communities to use broadband to 6. Policy implications and conclusions compensate for deficits in other inputs, such as transportation and financing. Forman, Goldfarb, and Greenstein (2005) show that Addressing deficitsin the local availability of critical infrastructure is geographical isolation promotes broadband adoption by allowing firms a key government mandate. This is reflected in several provisions in to substitute for higher transportation costs. Further, Prieger (2019) federal and state law that set forth policies to ensure the deployment of finds evidence that broadband promotes small-business formation in low-income and minority areas by compensating for constraints in local banking availability. These and other studies indicate that disparities in Table 3 the provision of broadband are likely to deepen urban inequality. Pooled logit estimation for broadband competition. At the same time, there are several limitations to this study that Broadband competition (yes = 1) warrant caution in the interpretation of results. The most relevant is the VARIABLE Model 1 Model 2 potential for omitted variable bias in the estimation models. While our fixed effects modelling strategy effectively accounts for unobserved Population density 4.45e-06 5.20e-06* variations in time invariant factors (such as local terrain) across block (2.89e-06) (2.92e-06) groups, this strategy is not immune to bias from time-varying factors Black (% pop) 0.430** 0.139 that affect broadband investment incentives. An example of such a factor (0.179) (0.194) is business demand for broadband, which correlates with incentives to Hispanic (% pop) 0.168 0.148 (0.213) (0.213) deploy fiber in a particular community. Because the presence of Asian (% pop) 1.297*** 1.279*** technology-intensive firms also correlates with community de (0.223) (0.223) mographics, this omission potentially biases our estimation results. Bachelor or higher (% pop) 0.144 0.151 Another question that deserves further attention is why similar def (0.251) (0.252) Median age 0.0148*** 0.0142*** icits in broadband buildout are not found for areas with large Hispanic (0.00438) (0.00438) populations. One possible answer relates to the spatial distribution of HH with children <18 (%) 0.312* 0.233 Hispanics across block groups in LA county, which is significantlymore (0.176) (0.176) homogeneous than the distribution of Black residents (recall the results English-only HH (%) 0.997*** 1.040*** (0.194) (0.195) Median HH income (log) 0.195*** 0.00832 (0.0466) (0.0616) Table 2 Housing value (log) 0.161** 0.175** Summary statistics. (0.0740) (0.0742) = Variable Mean Std. Dev. Min Max First income quartile (yes 1) 0.182*** (0.0670) Broadband Competition 0.751 0.433 0 1 First income quartile X Black (% pop) 0.895*** Fiber Availability 0.348 0.476 0 1 (0.240) Population density 13,740.6 11,455.1 0.205 148,004.1 Constant 3.563*** 1.540 Black (% pop) 0.085 0.148 0 1 (1.111) (1.194) Hispanic (% pop) 0.466 0.303 0 1 Observations 28,273 28,273 Asian (% pop) 0.137 0.166 0 1 Bachelor or higher (% pop) 0.303 0.22 0 1 Robust standard errors in parentheses. *** Median age 37.5 7.85 13.6 80.8 p < 0.01. HH with children <18 (%) 0.397 0.161 0 1 English-only HH (%) 0.429 0.244 0 1 ** p < 0.05. Median HH income 68,140 36,456 4987 250,000 Housing value 508,189 271,340 10,400 2,000,000 * p < 0.1 First income quartile 0.244 0.43 0 1

64 / 67 H. Galperin et al. Government Information Quarterly xxx (xxxx) xxx

Table 4 Table 5 Fixed effects logit estimation for broadband competition. Pooled logit estimation for fiber availability.

Broadband competition (yes = 1) Fiber availability (yes = 1)

VARIABLE Model 3 Model 4 VARIABLE Model 5 Model 6

Population density 2.57e-05* 2.56e-05 Population density 5.81e-06* 4.62e-06 (1.77e-05) (1.77e-05) (3.19e-06) (3.21e-06) Black (% pop) 1.271 0.795 Black (% pop) 1.414*** 1.372*** (0.982) (0.996) (0.180) (0.187) Hispanic (% pop) 0.838 0.807 Hispanic (% pop) 2.525*** 2.512*** (0.745) (0.746) (0.215) (0.214) Asian (% pop) 0.0567 0.0742 Asian (% pop) 2.410*** 2.360*** (0.929) (0.930) (0.218) (0.218) Bachelor or higher (% pop) 2.462*** 2.468*** Bachelor or higher (% pop) 0.389 0.345 (0.740) (0.740) (0.243) (0.243) Median age 0.0636*** 0.0642*** Median age 0.00779* 0.00619 (0.0103) (0.0104) (0.00453) (0.00454) HH with children <18 (%) 2.450*** 2.453*** HH with children <18 (%) 0.243 0.373** (0.329) (0.330) (0.167) (0.167) English-only HH (%) 0.231 0.222 English-only HH (%) 3.396*** 3.407*** (0.620) (0.620) (0.213) (0.213) Median HH income (log) 0.0468 0.100 Median HH income (log) 0.695*** 0.431*** (0.0826) (0.106) (0.0502) (0.0620) Housing value (log) 7.389*** 7.416*** Housing value (log) 0.0521 0.0705 (0.312) (0.313) (0.0769) (0.0781) First income quartile (yes = 1) 0.239** First income quartile (yes = 1) 0.413*** (0.113) (0.0682) First income quartile X Black (% pop) 1.228*** First income quartile X Black (% pop) 0.148 (0.392) (0.282) Observations 8464 8464 Constant 11.81*** 9.190*** Number of block groups 2206 2206 (1.116) (1.177) Observations 28,273 28,273

Robust standard errors in parentheses Standard errors in parentheses. *** p < 0.01. *** p < 0.01. ** p < 0.05 ** p < 0.05. * p < 0.1 * p < 0.1

most technologically advanced cable and video services to all California communication facilities regardless of race, income and other de communities in a nondiscriminatory manner, regardless of their socio mographic factors.A key takeaway from our findings is that broadband economic status”.9 However, the FCC has explicitly ruled that local cable buildout in LA County during the 2014–18 period did not adhere to franchising authorities (such as the CPUC) may not exercise this au these policy standards. The steady pace of fiber deployment in affluent thority to regulate broadband, despite the fact that broadband services areas stands in contrast to the slow rollout in less affluentand minority are provided over the same infrastructure, and that consumers are communities. Similarly, while affluent residents can choose from rapidly abandoning traditional video and cable services in favor of broadband offerings by competing providers, in many low-income Internet-based video services.10 communities broadband speeds are offered by a single provider (typi Some local jurisdictions (including the City of Los Angeles) have cally the incumbent cable TV provider). attempted to use their control over utility poles, public rights of way, At the federal level, several programs exist to promote broadband municipal buildings, and other assets that are critical to the rollout of 5G rollout, most notably the Connect America Fund which since 2011 has wireless services as levers to negotiate network build-out commitments provided about $4.5B in annual funding for underserved areas.However, with broadband operators. However, these attempts have been met with the program guidelines favor high-cost areas, which result in subsidies resistance from federal authorities, for example in the FCC’s 2018 pre being directed almost exclusively to rural and sparsely populated areas emption of local authority over pole attachment rules.11 And earlier the at the expense of poorly served urban communities. Interestingly, the same year, the FCC reversed course by reclassifying broadband as an same is true for California’s own broadband infrastructure program, the information service under Title I of the Communications Act, a shift that California Advanced Services Fund (CASF), created in 2007 to promote signaled its reluctance to extend nondiscrimination obligations (which network deployment in “unserved and underserved areas in the state”. apply to common carriers under Title II) to the rollout of broadband As of April 2020, about 90% of the roughly $310 M in awarded funds services. 8 have gone to broadband infrastructure projects in rural areas. In response to these policy changes, many local governments are The challenge of promoting network buildout in distressed urban exploring alternatives based on the public-utility model. While there are areas is compounded by the contested nature of policy oversight over many variations of this model (Menon, 2016), it is typically structured broadband deployment. Such oversight spans multiple layers of policy around a municipally-owned wholesale fiber network that serves authority, and is characterized by legacy legislation that does not reflect current market trends. Consider for example California’s Digital Infra structure and Video Competition Act of 2006 (DIVCA), which gives 9 DIVCA (2006) Section 5810 (a). broad authority for the CPUC to “promote the widespread access to the 10 FCC Third Report and Order, FCC 19–80, adopted on August 1, 2019. 11 Accelerating Wireline Broadband Deployment by Removing Barriers to Infrastructure Investment, FCC Docket No. 17–84, Third Report and Order and 8 Own calculations based on CPUC data available at www.cpuc.ca.gov. Declaratory Ruling, 33 FCC Rcd 7705.

65 / 67 H. Galperin et al. Government Information Quarterly xxx (xxxx) xxx

Table 6 Funding Sources Fixed effects logit estimation for fiber availability.

Fiber availability (yes = 1) This study did not receive any specificgrant from funding agencies in the public, commercial, or not-for-profit sectors. VARIABLE Model 7 Model 8

Population density 2.69e-06* 1.42e-06 (2.29e-05) (2.30e-05) Submission declaration and verification Black (% pop) 3.669*** 3.136** (1.239) (1.258) This study has not been published previously, is not under consid Hispanic (% pop) 0.439 0.441 eration for publication elsewhere, and its publication is approved by all (0.952) (0.952) authors and tacitly or explicitly by the responsible authorities where the Asian (% pop) 0.571 0.620 (1.166) (1.163) work was carried out. If accepted, it will not be published elsewhere in Bachelor or higher (% pop) 2.515*** 2.576*** the same form, in English or in any other language, including elec (0.927) (0.928) tronically without the written consent of the copyright-holder. Median age 0.0824*** 0.0826*** (0.0137) (0.0137) < HH with children 18 (%) 4.117*** 4.117*** Declaration of Competing Interest (0.413) (0.413) English-only HH (%) 1.140 1.090 (0.803) (0.804) None. Median HH income (log) 0.314*** 0.146 (0.108) (0.137) Acknowledgements Housing value (log) 11.81*** 11.81*** (0.465) (0.465) First income quartile (yes = 1) 0.0704 We thank Dr. Lisa Schweitzer and three anonymous reviewers for (0.142) their thoughtful comments that greatly improved the quality of the First income quartile X Black (% pop) 1.256*** manuscript. (0.480) Observations 6227 6227 Number of block groups 1593 1593 References Standard errors in parentheses *** < Baynes, L. M. (2004). Deregulatory injustice and electronic redlining: The color of access p 0.01. to telecommunications. Administrative Law Review, 56. Blevins, J. (2019). The use and abuse of “light-touch” internet regulation. Boston ** p < 0.05. University Law Review, 99(1), 177–232. https://doi.org/10.2139/ssrn.3168055. Campos-Castillo, C. (2015). Revisiting the first-level digital divide in the United States: – * p < 0.1 Gender and race/ethnicity patterns, 2007 2012. Social Science Computer Review, 33 (4), 423–439. https://doi.org/10.1177/0894439314547617. ¨ Clark, W., Anderson, E., Osth, J., & Malmberg, B. (2015). A multiscalar analysis of neighborhood composition in Los Angeles, 2000–2010: A location-based approach to government needs while also leasing capacity to private operators who segregation and diversity. Annals of the Association of American Geographers, 105(6), 1260–1284. https://doi.org/10.1080/00045608.2015.1072790. operate the retail access network. Until recently, this model was limited CPUC. (2016). California Broadband Data Processing and Validation Data as of December. to relatively small cities, or in some cases consortia of cities (such as in 31 p. 2016). Utah’s UTOPIA network). However, in January 2020 New York City Crawford, S. (2013). Captive audience: The telecom industry and monopoly power in the new gilded age. Yale University Press. unveiled an ambitious plan to build an open access fibernetwork across Crawford, S. (2018). Fiber: The coming tech revolution–and why America might miss it. Yale the entire city, leveraging multiple city assets. This fibernetwork will be University Press. overlaid with a wireless access network capable of providing fast and Fairlie, R. W. (2004). Race and the digital divide. Contributions to Economic Analysis and Policy, 3(1), 155–194. https://doi.org/10.2202/1538-0645.1263. affordable broadband service in every neighborhood, but the plan Falcon, E. (2019). Thank Laws Supported By AT&T and Comcast for California’s explicitly prioritizes low-income areas where competition and FTTH Broadband Monopoly Problem. https://www.eff.org/deeplinks/2019/07/thank-la rollout have lagged in comparison to the more affluentneighborhoods. 12 ws-att-and-comcast-support-californias-broadband-monopoly-problem. Federal Communications Commission. (2020). Broadband Deployment Report. It remains to be seen whether the municipal network model can be Flamm, K., & Chaudhuri, A. (2007). An analysis of the determinants of broadband access. successfully replicated at scale in large metro areas. Regardless, ambi Telecommunications Policy, 31(6–7), 312–326. https://doi.org/10.1016/j. tious projects such as New York City’s point to growing frustration telpol.2007.05.006. Flamm, K., & Varas, P. (2018). The evolution of broadband competition in local US among local policymakers with federal and state policies that have failed markets: A distributional analysis. In TPRC 46: The 46th Research Conference on to catalyze broadband investments in distressed urban communities. In Communication, Information and Internet Policy. SSRN Electronic Journal. https://doi. 2018, California passed AB 1999 which eliminated the last set of re org/10.2139/ssrn.3142329. strictions on municipal broadband.13 Municipalities seeking to redress Forman, C., Goldfarb, A., & Greenstein, S. (2005). How did location affect adoption of the commercial internet? Global village vs. urban leadership. Journal of Urban disparities in the provisioning of broadband now have a broader toolset Economics, 58(3), 389–420. https://doi.org/10.1016/j.jue.2005.05.004. at their disposal, and the findings of this study suggest that a more Fotheringham, A. S., & Wong, D. W. S. (1991). The modifiable areal unit problem in – proactive role for local governments may indeed be warranted. multivariate statistical analysis. Environment and Planning A, 23(7), 1025 1044. https://doi.org/10.1068/a231025. Frieden, R. (2009). Lies, damn lies and statistics: Developing a clearer assessment of Author statement market penetration and broadband competition in the United States. Virginia Journal of Law and Technology, 14(100), 101–125. GAO. (2018). Broadband Internet: FCC’s Data Overstate Access on Tribal Lands. Hernan Galperin: Conceptualization, methodology, writing, original Gillett, S. E., & Lehr, W. (1999). Availability of broadband internet access: empirical draft preparation. Thai Le: Dataset preparation, data analysis, writing. evidence. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3215923. Kurt Wyatt: Dataset preparation, data analysis, writing. Green, M., & Flowerdew, R. (1996). New evidence on the modifiableareal unit problem. In P. A. Longley, & M. Batty (Eds.), Spatial Analysis: Modelling in a GIS Environment (pp. 41–54). John Wiley & Sons. Greene, W. (2003). Econometric analysis (5th ed.). Prentice Hall. Grubesic, T. H. (2012). The U.S. National Broadband map: Data limitations and 12 implications. Telecommunications Policy, 36(2), 113–126. https://doi.org/10.1016/j. The New York City Internet Mater Plan. Available at www.nyc.gov/tech. telpol.2011.12.006. 13 AB 1999 lifted restrictions to publicly-owned broadband networks in un Grubesic, T. H., Helderop, E., & Alizadeh, T. (2019). Closing information asymmetries: A incorporated areas in the state. scale agnostic approach for exploring equity implications of broadband provision.

66 / 67 H. Galperin et al. Government Information Quarterly xxx (xxxx) xxx

Telecommunications Policy, 43(1), 50–66. https://doi.org/10.1016/j. Prieger, J. E. (2015). The broadband digital divide and the benefitsof mobile broadband telpol.2018.04.002. for minorities. Journal of Economic Inequality, 13(3), 373–400. https://doi.org/ Grubesic, T. H., & Murray, A. T. (2005). Geographies of imperfection in 10.1007/s10888-015-9296-0. telecommunication analysis. Telecommunications Policy, 29(1), 69–94. https://doi. Prieger, J. E. (2019). The importance of broadband for entrepreneurship in org/10.1016/j.telpol.2004.08.001. disadvantaged areas. In Conference paper presented at the 41st Annual Fall Research Hauge, J. A., & Prieger, J. E. (2010). Demand-side programs to stimulate adoption of Conference for the Association of Public Policy Analysis & Management. Denver, CO. broadband: What works? Review of Network Economics, 9(3). https://doi.org/ Prieger, J. E., & Hu, W. M. (2008). The broadband digital divide and the nexus of race, 10.2202/1446-9022.1234. competition, and quality. Information Economics and Policy, 20(2), 150–167. https:// Holmes, A., & Wieder, B. (2016). DSL providers save faster internet for wealthier doi.org/10.1016/j.infoecopol.2008.01.001. communities. Quillian, L. (2012). Segregation and poverty concentration: The role of three Kneebone, E., & Holmes, N. (2016). U.S. concentrated poverty in the wake of the great segregations. American Sociological Review, 77(3), 354–379. https://doi.org/ recession. American Journal of Public Health, 104. https://doi.org/10.2105/ 10.1177/0003122412447793. AJPH.2013.301420. American Public Health Association Inc. Rhinesmith, C., & Reisdorf, B. C. (2017). Race and digital inequalities: policy Kolko, J. (2010). A new measure of US residential broadband availability. implications. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.2944205. Telecommunications Policy, 34(3), 132–143. https://doi.org/10.1016/j. Sampson, R. J. (2017, August 22). Urban sustainability in an age of enduring telpol.2009.11.015. inequalities: Advancing theory and ecometrics for the 21st-century city. Proceedings Kongaut, C., & Bohlin, E. (2017). Impact of broadband speed on economic outputs: An of the National Academy of Sciences of the United States of America. National Academy empirical study of OECD countries. Economics and Business Review, 3(2), 12–32. of Sciences.. https://doi.org/10.1073/pnas.1614433114. https://doi.org/10.18559/ebr.2017.2.2. Sampson, R. J., & Wilson, W. J. (1995). Toward a theory of race, crime, and urban Lewis, L. B., Sloane, D. C., Nascimento, L. M., Diamant, A. L., Guinyard, J. J., inequality. In D. Karp (Ed.), Community justice: An emerging field. Rowman & Yancey, A. K., & Flynn, G. (2005). African Americans’ access to healthy food options Littlefield Publishers. in South Los Angeles restaurants. American Journal of Public Health, 95(4), 668–673. Turner, S. D. (2016). Digital denied: The impact of systemic racial discrimination on home- https://doi.org/10.2105/AJPH.2004.050260. internet adoption. Massey, D. S. (1990). American apartheid: Segregation and the making of the underclass. Whitacre, B., Strover, S., & Gallardo, R. (2015). How much does broadband American Journal of Sociology, 96(2), 329–357. https://doi.org/10.1086/229532. infrastructure matter? Decomposing the metro-non-metro adoption gap with the Massey, D. S., & Denton, N. A. (1988). The dimensions of residential segregation. Social help of the National Broadband map. Government Information Quarterly, 32(3), Forces, 67(2), 281–315. 261–269. https://doi.org/10.1016/j.giq.2015.03.002. Massey, D. S., & Fischer, M. J. (1999). Does rising income bring integration? New results Wilson, W. J. (1987). The truly disadvantaged: The Inner City, the underclass, and public for blacks, Hispanics, and Asians in 1990. Social Science Research, 28(3), 316–326. policy. The University of Chicago Press. https://doi.org/10.1006/ssre.1999.0660. Wolch, J., Wilson, J. P., & Fehrenbach, J. (2005). Parks and park funding in los Angeles: Matsunaga, M. (2008). Concentrated poverty neighborhoods in Los Angeles. An equity-mapping analysis. Urban Geography, 26(1), 4–35. https://doi.org/ Menon, S. (2016). Access to and adoption of a municipal broadband middle-mile 10.2747/0272-3638.26.1.4. network: The case of the community access network in Washington, D.C. Government Wooldridge, Jeffrey, M (2001). Econometric Analysis of Cross Section and Panel Data. MIT Information Quarterly, 33(4), 757–768. https://doi.org/10.1016/j.giq.2016.08.010. Press. Mossberger, K., Tolbert, C., & Franko, W. (2012). Digital cities: The internet and the geography of opportunity. Oxford University Press. Hernan Galperin is Associate Professor and Assistant Dean at the Annenberg School for National Digital Inclusion Alliance. (2019). AT&T’s Digital Redlining of Dallas. Available Communication at the University of Southern California. at www.digitalinclusion.org/blog/2019/08/06/atts-digital-redlining-of-dallas-new- research-by-dr-brian-whitacre. Oberwittler, D., & Wikstrom,¨ P. O. H. (2009). Why small is better: Advancing the study of Thai Le is a Ph.D. candidate in Public Policy and Management at the Sol Price School of the role of behavioral contexts in crime causation. In Putting Crime in its Place: Units Public Policy at the University of Southern California. of Analysis in Geographic Criminology (pp. 35–59). Springer New York. https://doi. org/10.1007/978-0-387-09688-9_2. Kurt Daum is a Ph.D. student in Urban Planning and Development at the Sol Price School Perry, Andre, M (2020). Know Your Price: Valuing Black Lives and Property in America’s of Public Policy at the University of Southern California. Black Cities. Brookings Institution Press. Prieger, J. E. (2003). The supply side of the digital divide: Is there equal availability in the broadband internet access market? Economic Inquiry, 41(2), 346–363. https:// doi.org/10.1093/EI/CBG013.