Disparate Impact: Who's Afraid of It?

Hi, I’m Aaron Sankin, and I’m a reporter here at The Markup. Over the past year, I’ve focused on investigating why the internet connection at your house is slow and what you can do about it (other than shoot dirty looks at your modem, which doesn’t really care about your opinion).

Tucked deep in the Infrastructure Investment and Jobs Act, signed by President Joe Biden in 2021, is a short provision giving the Federal Communications Commission (FCC) broad leeway to fundamentally reshape how America connects to the internet.

Under a section titled “Digital Discrimination,” the legislation tasks the FCC with adopting “rules to facilitate equal access to broadband internet access service, taking into account the issues of technical and economic feasibility presented by that objective, including … preventing digital discrimination of access based on income level, race, ethnicity, color, religion, or national origin.”

The section itself is short—under 400 words—and relatively light on detail, leaving it largely up to the FCC to decide how much power it wants to grab for itself to close the digital divide. Various actors have rushed into that void of legislative ambiguity, forcefully advocating for their own agendas.

On one side are advocacy groups and some local governments who hope the FCC will take the opportunity to aggressively hold internet service providers (ISPs) to account for what they see as generations of inequitable broadband infrastructure deployment across the country.

Some propose levying fines on ISPs guilty of discrimination or, as the agency did when it approved a merger between Charter Communications and Time Warner Cable in 2016, mandating companies build out high-speed infrastructure in new areas.

On the other side are ISPs, the industry groups representing them, and ideologically aligned pro-business think tanks.

They want nothing more than for the FCC to merely supplement its historic light-touch regulation of the broadband industry with generous subsidies—that make investing in high-speed internet infrastructure in marginalized communities more beneficial to the companies’ bottom lines.

The crux of this fight is over a concept called “disparate impact” and whether the FCC should use it to identify whether an ISP has engaged in discrimination. The idea behind disparate impact is that discrimination isn’t always explicit.

So, if an institution’s policies have created a situation where certain protected groups experience markedly worse outcomes than others, this impact matters just as much as any intentionally biased practices.

For example: A company’s CEO decides that people should always physically look up to their managers, so only people over 6 feet, 3 inches get promoted to managerial roles. As a result, the management staff is entirely male, since men tend to be taller on average.

Did the company engage in gender discrimination? Under a disparate impact standard, yes. But, under a standard where intent prevails, this discrimination is a lot harder to prove, because the rules that resulted in a management team where everyone can dunk said nary a peep about gender.

“From my point of view, disparate impact is really just discrimination through the back door,” said John Yinger, a professor of economics and public administration at Syracuse University, who has studied disparate impact and wrote a book about discrimination in mortgage lending.

“If you have a system that doesn’t recognize and address disparate impact, you allow people to, either on purpose [or] by accident, engage in a lot of discrimination.”

The problem, Yinger noted, is that to assess disparate impact, the first step is determining which allocation of a certain scarce set of resources—be it broadband access or corporate promotions—is fair, and then determine if different groups’ receipt of that resource violates that standard of fairness.

Intentions Matter, or Do They?

Consider an investigation we published last year showing how a quartet of ISPs consistently gave the worst deals for internet service to households in low-income, less-White, and historically redlined neighborhoods in major cities across the country.

When we were reporting that story, we certainly were curious why these disparities existed, but that why was less important than the fact that we could prove they existed in the first place.

Last year, when we reached out to CenturyLink, one of the companies we found consistently charging high prices for sluggish service in marginalized areas, spokesperson Mark Molzen insisted at the time that the company was committed to anti-racism.

“While we can’t comment on behalf of other providers, we can say we do not enable services based on any consideration of race or ethnicity, and the methodology used for the report on our network is deeply flawed.”

Molzen never pointed to any specific inaccuracies in our data collection or analysis. The “flaw” he identified was in an inference the company read into the story about its motives.

Our October 2022 investigation showed absolutely zero evidence that CenturyLink had partitioned cities into geographic areas based on race and used that to determine where to deploy high-speed internet infrastructure. And, honestly, it would be pretty wild if it actually had.

Nevertheless, we found that CenturyLink offered addresses in the least-White parts of Des Moines, Iowa, slow-speed internet service 19 percentage points more frequently than the Whitest parts of the city—and all for the same price. In Las Vegas, Nev., that gap was 31 percentage points. In Omaha, Neb., 33 percentage points.

A comment within a June 2022 FCC digital discrimination proceeding with Lumen, CenturyLink’s parent company, mentioned the company’s expectations “to deploy fiber to approximately one million new locations in 2022 and around 1.5 to 2 million new locations in 2023.”

However, there wasn’t any specificity regarding how many of these new fiber deployments would go to low-income or predominantly non-White areas.

When it comes to the FCC’s responsibility, the question is whether the FCC will look at the levels of access that different racial, ethnic, and socioeconomic groups have to affordable, blazing-fast internet, find substantial inequality, and then do something about it.

Or, as the telecom industry would prefer, will the agency only intervene after finding direct evidence of an ISP making inequitable decisions about where to deploy infrastructure explicitly based on characteristics like race or ethnicity?

“In Some Circumstances, It Can Be Profitable”

The legal theory behind disparate impact goes back to Griggs v. Duke Power Co., a 1971 Supreme Court case about how a company required all employees to pass a pair of intelligence tests if they wanted to move into higher-paying roles. Black people were far less likely to pass those tests, which were unrelated to the jobs’ responsibilities.

The court ruled that these tests—which the company just so happened to implement on the very day the 1964 Civil Rights Act—prohibited race-based employment discrimination, and were illegal due to the effect they had on Black employees’ career advancement.

The use of disparate impact to identify discrimination grew in the ensuing decades and was finally codified into law in 1991. “Because individual motives may be difficult to prove directly, Congress has frequently permitted proof of only discriminatory impact as a means of overcoming discriminatory practices,” reads a U.S. Department of Justice manual on the subject.

“The Supreme Court has, therefore, recognized that disparate impact liability under various civil rights laws ‘permits plaintiffs to counteract unconscious prejudices and disguised animus that escape easy classification as disparate treatment.’ ”

During the administration of President Donald Trump, who famously settled a lawsuit in the 1970s over his family’s real estate company’s practice of not renting properties to Black tenants, the Department of Housing and Urban Development dramatically weakened the government’s ability to use disparate impact, but the rules were reinstated shortly after President Biden took office.

While disparate impact has been used by the government to regulate employment and housing, it hasn’t yet been used to assess the role ISPs are playing in the digital divide. As a result, when the FCC went about implementing the digital discrimination rules mandated by Congress, the agency specifically requested that the public weigh in on whether the standard should be used and, if so, how.

The responses the agency received on the issue largely fell into two categories. The telecom industry argued in favor of using disparate treatment as a metric, which holds that discrimination occurs when people are explicitly divided up by a protected category (like race) and then treated worse as a result. Digital inclusion advocacy groups argued that disparate impact is the only feasible metric.

“Rational, profit-seeking actors will discriminate because, in some circumstances, it can be profitable to do so. Whether this is intentional or not is beside the point,” argued the National Digital Inclusion Alliance and Common Sense Media, two nonprofit advocacy groups, in one comment.

“The point … is that discrimination occurs and will continue to occur unless providers’ incentives are changed.”

The groups’ comment urges the FCC to look for disparate impact in a plethora of ISP practices ranging from the prices customers are charged for a service, and under what contract terms, to the actual speeds different areas receive and how companies prioritize network maintenance.

“For decades, the disability community has noted that discrimination occurs unintentionally and often results from seemingly neutral policies,” the American Association of People with Disabilities concurred in another comment.

“Too often, disabled people experience discrimination not because of malicious intent or explicit exclusion within programs or policies but because the disabled people were simply not considered in the first place.”

The Greenlining Institute, a California-based nonprofit that advocates for digital inclusion efforts, insisted in its comment that applying disparate impact to broadband would be consistent with how it has been used across the government for decades.

“Adoption of the disparate impact standard in the wake of the Civil Rights Act of 1964 allowed for a whole-of-government approach that targeted discrimination in public goods such as education and housing,” Greenlining asserted. “It would be consistent to apply this similar treatment to digital public infrastructure and services such as broadband internet.”

The tenor of comments from folks who might theoretically be held accountable if their actions were found to have treated some communities worse than others was, unsurprisingly, markedly different.

A primary industry worry is that requiring ISPs to constantly weigh equity concerns may dissuade them from deploying new infrastructure widely in the first place.

“Adopting a disparate impact test would cause carriers to choose between prioritizing deployment based on closing the digital divide, as required by Congress, and paralyzing deployment for fear that a regular, common business decision may disproportionately affect a minority community,” asserted the Lincoln Network, a free market think tank, in its comment.

Another industry argument is that Congress didn’t really mean for the FCC to use disparate impact to identify “discrimination of access based on income level, race, ethnicity, color, religion, or national origin.”

Verizon, for example, asserted that the phrase “based on” in the legislation specifically means that discrimination has to be a direct result of company decisions made because of protected demographic characteristics of a particular area.

For its part, AT&T asserted in its comment that, even if a disparate impact standard were applied, the broadband industry wouldn’t be found to have delivered less connectivity to historically marginalized areas.

While there have been localized studies showing inequitable deployment in places like Cleveland, Dallas, Oakland, and Los Angeles, the company said that the only way to accurately assess deployment would be to look at a comprehensive, national picture of everywhere every single ISP offers service.

AT&T pointed to an analysis from former FCC chief economist Glenn Woroch that showed that “broadband availability rates for census-based ‘non-white’ households are higher than for white households and that availability rates for households above and below the poverty line are nearly identical.”

Woroch’s research relies on data from a broadband availability map produced by the FCC that is supposed to show what service ISPs offer across the country.

While a footnote in Woroch’s comment insists his data source doesn’t skew the results, that map has been widely criticized for inaccuracies, and a 2020 Princeton study found that “the FCC’s data disproportionately overstates coverage in rural and minority communities.”

The FCC released a new map last year designed to rectify those issues, but that map has its own set of profound accuracy problems and is still effectively in its beta testing phase. As a result, obtaining data accurate enough to meet the standards pushed for by AT&T remains a tall order.

The battle over the FCC’s potential use of disparate impact hasn’t been limited to the public comment process.

In a recent op-ed, The Wall Street Journal’s historically laissez-faire editorial board used the specter of disparate impact to lobby against Public Knowledge’s co-founder Gigi Sohn, whom Biden unsuccessfully nominated to lead the agency, but withdrew following telecom industry opposition.

“Public Knowledge, the outfit that Ms. Sohn co-founded and led from 2001 to 2013, has urged the FCC to interpret its regulatory mandate broadly to micromanage broadband competition and development,” the editorial charged.

Now would be the time to mention that The Markup does, in fact, have a pack of dogs in this fight. Our newsroom is based on the idea that if you evaluate the inputs and outputs of a big, complex technical system, you can identify serious problems in how it treats different groups without having to know the precise intentions of the people who developed it.

The technology does not yet exist to do a regression controlling for the vagaries of the human heart.

In March, Markup reporter Malena Carollo and freelance data journalist Ben Tanen used data analysis to determine that an algorithm for deciding which patients got life-saving liver transplants systematically disadvantaged people in a handful of states, almost all of which were in the South and Midwest. The day after that story was published, the Department of Health and Human Services announced major reforms to that very system.

One thing I’m working on—along with my colleague, investigative data reporter Leon Yin—is a step-by-step guide showing everyday people how they can evaluate whether ISPs have left a digital divide in their towns. It should be dropping sometime in the next few weeks.

We’re hoping that if people use the guide and find something interesting, they can tell the FCC about it. And, who knows, depending on how this disparate impact proceeding goes, maybe the agency will even do something about it.

Sincerely,

Aaron Sankin

Investigative Reporter

The Markup

By Aaron Sankin

Also published here

Photo by Annie Spratt on Unsplash