March is Women’s History Month, which means I’m often asked to name the most pressing issue facing women in America. Answers spring to mind, sometimes faster than I can form the words. The fall of Roe and the Black maternal mortality crisis. The persistence of the gender wage gap and on-the-job sexual harassment, more than five years after #MeToo. Barriers to safe, affordable housing. Policing of Black and Brown mothers, leading to needless family separation. The lack of universal paid family leave coupled with the skyrocketing cost of childcare. The list goes on.

None of these ills, however, is the subject of so-called “Women’s Bill of Rights” laws being introduced in a growing list of states including Kansas, Arizona, South Carolina, Oklahoma, and Montana. Instead, this legislation would create a legal definition of womanhood based on the capacity to produce ova, or human eggs. This definition of “woman,” which is gerrymandered to exclude trans women and girls, would then apply throughout state law — and could make it impossible for trans people to live openly at work, at school, or anywhere in the states they call home.

That should set off alarm bells for all of us, not just those engaged directly in the struggle for LGBTQ rights. The “Women’s Bill of Rights” is only a sliver of the cruel campaign to deny basic rights to trans people currently underway across the country. And despite its misleading label, it shares a through-line with a long and ugly history of gender-based subjugation in the name of “biology.” For centuries, laws and policies premised on women’s biological capacities and “delicate” nature were used to shut women out of educational, economic, and civic opportunities. On these grounds, the U.S. Supreme Court upheld laws barring women from becoming attorneys — or bartenders. Similar “biological” arguments were used to exclude Black women from “the fairer sex” in order to justify extraction of Black women’s labor under the institution of slavery and beyond.

As feminists, we reject efforts to appropriate the rhetoric of “women’s rights” to inflict life-threatening harm on trans people, men or women. Attacking trans people does nothing to address the real problems women face. To the contrary, limiting freedom for trans people worsens conditions for all women by re-entrenching the very gender stereotypes that have underpinned centuries of women’s oppression and that the ACLU Women’s Rights Project has worked for more than half a century to dismantle. After all, the very notion that a person should identify with the sex they were assigned at birth for their entire life is a stereotype, as the more than 1.5 million trans people living in the United States attest to every day.

Formed in 1972, the Women’s Rights Project’s earliest cases focused on establishing rigorous judicial review of laws that classified people on gender lines, often based on long-held stereotypes about men’s and women’s capacities and without regard to individual abilities, needs, and wants. That work, led by Ruth Bader Ginsburg for the ACLU, included challenging a probate rule that preferred men to women based on the stereotype that any man is more capable of settling an estate than every woman; a housing allowance offered to servicemen, but not servicewomen, based on the stereotype that men should be primary breadwinners; and an income tax deduction available to women, but not men, based on the stereotype that only women should be caregivers.

The plaintiffs in these cases included men as well as women. What they had in common was that each defied gender stereotypes, out of desire or necessity. And all fought to live fully and authentically, without laws and policies that constrained them based on gender or their ability to bear children. To live openly as transgender is to seek that same freedom.

Not only is there no conflict between demanding rights for women and for all transgender people, advances in trans rights hold a specific promise for women’s liberation. By tearing down laws and policies based on gender stereotypes, we can create the opportunity for each of us to determine our own life story. That’s why the Women’s Rights Project strives to represent people of all genders, transgender, nonbinary, and cisgender, who face barriers based on their sex.

Today’s avalanche of attacks on trans people, with over 400 anti-LGBTQ bills introduced in state legislatures in 2023 so far, makes plain that the gender discrimination of the past is all too present today. Defending trans people is not only a moral duty for the feminist movement; it is central to it.

Legal intervention as an avenue for social change has been central to social, economic, and political progress for women, but is difficult to access for the people who most need it. Until recently, the mainstream women’s rights movement in the United States was dominated by white and wealthy women — such as Elizabeth Cady Stanton, Susan B. Anthony, and Betty Friedan — fighting for the rights of others who look like them. As a result, the movement has too often excluded and erased the experiences of non-white and working-class women, limiting the movement’s progress in advancing equality for Black women, Indigenous women, and other women of color in this country.

As the women’s movement shifts to an intersectional approach — a term coined by Kimberlé Crenshaw to describe that identities such as race and gender cannot be isolated in understanding the oppression of people who hold multiple marginalized identities — elevating the voices and experiences of women who have been historically excluded is critical. Lawyers must ally with community organizers, who work on the ground with those most impacted, to make the women’s movement more accessible and inclusive — an approach the ACLU Women’s Rights Project strives to model.

Though I provide legal support to attorneys now, I co-founded a tenants’ union in Ithaca, New York in 2019, and I know firsthand that there is heavy skepticism between lawyers and organizers. Lawyers can be out of touch with the communities they are supposed to be helping, yet often deem themselves experts in tackling a problem over people who are closest to the issue. Organizers, on the other hand, sometimes utilize bold strategies that test the limits of the law, which some lawyers are uncomfortable with. Bridging this gap and finding more ways to work together will fill needs on both ends and lead to more progress.

In 2018, the ACLU and several state affiliates, along with private counsel, partnered with Fight for 15 (FF15) organizers to confront rampant sexual harassment at McDonald’s restaurants nationwide. Senior Staff Attorney Gillian Thomas and FF15 organizer Allynn Umel explained to me that people in politically and economically vulnerable positions (such as undocumented people, teenagers, and low-wage workers) frequently avoid complaining about workplace abuse, let alone taking legal action, out of fear of retaliation. Because lawyers can be removed from their clients’ everyday realities, collaborating with organizers helps the attorneys stay informed and build the trust that is essential for time-consuming and emotionally draining cases. Recognizing the risks and pressures of challenging their employer, FF15 organizers met with workers to speak about the importance of legal intervention and make sure they knew they were not alone.

“Particularly for the most marginalized and vulnerable, it is pretty critical to have organizers familiar with what workers are putting on the line every single day as part of a larger fight entails as the workers engage in the legal process, as an organizer’s main job is to help push people past their fear and into a place of power,” explained Umel.

There is a marked difference between attorneys interested in supporting further organizing and attorneys looking to limit organizing to protect organizational or institutional concerns, Umel noted. The central question, she said, is “Are we on the same team? Do the attorneys here fully appreciate and value the role of organizing and overall campaign and making sure the workers and community members have some greater degree of ownership of what happened beyond just trying to enforce the law?” In speaking to her experience working with the ACLU, Umel emphasized that the respect for the organizers, coupled with an understanding of the support that survivors of harassment needed to come forward, was critical for the campaign and fight.

In 2020, during the height of the pandemic, the ACLU Women’s Rights Project and ACLU of Missouri worked with KC Tenants to file a federal lawsuit challenging the Jackson County Circuit Court for permitting eviction cases despite the nationwide eviction moratorium. On navigating the relationships and tensions between organizers and lawyers, Senior Staff Attorney Sandra Park explained that KC Tenants was the client, and ultimately, they determined if and how the litigation advanced their goals.

“In the early days of the pandemic, it was crucial for us to be in communication with folks on the ground, organizing, to determine what the greatest needs were and whether litigation could be used to help stop mass evictions,” said Park, reflecting that she does not see it as her role to determine organizers’ goals and strategies.

These are examples of integrated advocacy at its most effective. In both instances, the lawyers and organizers were able to respect each other’s expertise and collaborate to advocate their causes. These frameworks provide a roadmap that must be replicated in order for the women’s movement to be as inclusive and impactful as possible.

You hear a knock on your door. Expecting a neighbor or perhaps a delivery, you open it, only to find a child welfare worker demanding entry. It doesn’t seem like you can refuse so you let them in and watch as they search every room, rummaging through closets, drawers, cabinets, the fridge — all without a warrant. They ask questions making it sound like you’re a bad parent and, finally, say they need to do a visual inspection of your kids, undressed, without you in the room, and take pictures.

The agency receives many reports of ordinary neglect, which are distinct from physical abuse or severe neglect allegations, but it doesn’t investigate all of them. Instead, the agency had started using an algorithm to help decide who gets the knock on the door and who doesn’t. But you can’t get any information about what it said about you or your child, or how it played a role in the decision to investigate you.

Though an algorithm may sound neutral, predictive tools are designed by people. And the choices people make when creating the tool aren’t just decisions about what statistical method is better or what data is necessary to make its calculations. The same people can be flagged as more or less in need of investigation based on how a tool was designed. One recurring concern is that the use of these tools in systems marked by discriminatory treatment and outcomes will result in those outcomes being replicated. But this time, if that history repeats itself, the disparate results will be deemed unquestionable truths supported by science and math, and not the result of residual or ongoing discrimination, let alone the policy decisions that resulted when tool designers decided to choose model A instead of model B.

To better understand whether this concern is warranted, two years ago, the ACLU requested data and documents from Allegheny County, Pennsylvania related to the Allegheny Family Screening Tool (AFST) so we, together with researchers from the Human Rights Data Analysis Group, could independently evaluate its design and practical impact. We found, among other things, that the AFST could result in inequities in screen-in rates — the percentage of reports (i.e., neglect allegations received by the county child welfare agency) that are forwarded for investigation (“screened in”) out of the total number of reports received. We found that the tool could result in screen-in rate disparities between Black and non-Black families (i.e., the percentage of Black families flagged for investigation out of all allegations about Black families received could be greater than the same percentage for non-Black families). We also found that households where people with disabilities live could be labeled as higher risk than households without a disabled resident. What really stood out though was that we found the AFST’s algorithm, or the way its conclusions about a family were conveyed to a screener, could have been built in different ways that may have had a less discriminatory impact. And this alternative method didn’t change the algorithm’s “accuracy” in any meaningful way, even if we accept the tool’s developer’s definition of that term. We asked the county and tool designers for feedback on a paper describing our analysis, but never received a response. We share our findings for the first time today. But first, a quick overview of how the AFST works.

Allegheny County has been using the AFST to help screening workers decide whether to investigate or dismiss neglect allegations. (The AFST is not used to make screening decisions about physical abuse or severe neglect allegations because state law requires that those be investigated.) The tool calculates a “risk score” from 0 to 20 based on the underlying algorithm’s estimation of the likelihood that the county will, within two years, remove a child from the family involved in the report. In other words, the tool generates a prediction of the “risk” that the agency will place the child in foster care. The county and tool designers treat removal as a sign that the child may be harmed, so that the higher the likelihood of removal, the higher the score, and the greater the presumed need for child welfare intervention. Call-in screeners are instructed to consider the AFST’s output as one factor among many in deciding whether to forward the report for agency action.

However, in a child welfare system already plagued by inequities based on race, gender, income, and disability, using historical data to predict future action by the agency only serves to reinforce those disparities. And when reborn through an algorithm, people are liable to interpret the disparities as hard truths because, well, a mathematical equation told us so.

In this way, the AFST creators are doing more than math when building a tool. They also have the ability to become shadow policymakers — because unless the practical impact of their design decisions is evaluated and made public, this power can be wielded with little transparency or accountability, even though these are two of the reasons why the county adopted the tool.

Here is a summary of the design decisions and resulting policies and value judgments that we shared with the county as cause for concern:

When Allegheny County receives a report alleging child neglect, the AFST generates an individualized “risk score” for every child in the household. However, call screeners don’t see individual-level scores. Instead, the AFST shows an output based on only the highest score of all the children in the household. For referrals where the maximum score falls between 11 and 18, the AFST displays the score’s numeric value. For maximum scores of 18 and up, the AFST displays a “High Risk Protocol” label as long as at least one child in the household is under 17. This subset of referrals is subject to mandatory investigation, which only a supervisor can override. For referrals with a maximum score less than 11 and no children under 7, screening workers see a “Low Risk Protocol” label.

We found that the decision to communicate only the AFST’s predictions of the highest-scoring child could have created inequitable outcomes. We say “could have” because we could not run our analysis on the actual numbers of Black and non-Black families so instead, as is common practice including by the county and its tool developers, we looked at Allegheny County data collected before the AFST was deployed to model what the risk scores would have been.

Compared to other ways of conveying the AFST’s scores, the method in use could have resulted in the AFST classifying Black families as having a greater need for agency scrutiny than non-Black families. Through our analysis of data from 2010-2014, we found that the AFST’s method of showing just one score would have resulted in roughly 33% of Black households being labeled “high risk,” thereby triggering the mandatory screen-in protocol, but only 20% of non-Black households would have been so labeled.

Fig. 1. Distribution of AFST Scores by Race Under Different Scoring Policies, using testing data from 2010-2014. Under policies that assign a single score or screening recommendation to the entire household, AFST scores generally increase for all families, and Black households receive the highest scores more often than non-Black households. Under the current “Single Household Score” policy, nearly 35% of Black households are labeled as “high risk” for future separation while only 20% of non-Black households are labeled as “high risk.”

To build the tool’s algorithm, its designers needed to look at historical records to identify circumstances and individual characteristics most associated with child removals, since the tool bases its risk scores essentially on whether and how those factors are present in the incoming report. Thus to further one of the county’s stated goals in adopting the AFST — to “make decisions [about whether to screen in a report] based on as much information as possible” — the county gave the AFST designers access to government databases beyond the county’s child welfare records, such as juvenile probation and behavioral health records. The problem is that these databases do not reflect a random sample or cross-section of the county’s population. Rather, they reflect the lives of people who have more contact with government agencies than others. As a result, using such a database to identify the characteristics of households more likely to have a child removed means selecting from a pool of factors that over-represents some groups of people and underrepresent others, making it more likely that the tool will classify the same overrepresented populations as higher risk, not because they are more likely to be harmed or to cause harm, but because the government has access to data about them but little or no access to data about others.

Take for instance the county’s juvenile probation database, which was used to construct the AFST. A recent study found that Black girls in the county were 10 times more likely and Black boys were seven times more likely than their white counterparts to end up in the juvenile justice system. As a result, in using the related juvenile probation database to build the tool, the tool developers are mining records that overrepresent Black children as compared to white children.

The behavioral health databases the county used to create the AFST are similarly problematic. Because they expressly include information about people seeking disability-related care, these databases will inevitably contain information about people with disabilities, but not necessarily others. These databases are also skewed along another axis: Because the county doesn’t record information about privately accessed health care, data about individuals with higher incomes is far less likely to be reflected.

By partly basing the AFST’s removal prediction on factors that families can never change, such as whether someone has been held in the Allegheny County Jail at any time or for any reason, the AFST effectively offers families no way to escape their pasts, compounding the impacts of systemic bias in the criminal legal system. We found that households with more children are more likely to include somebody with a record in the county jail system or with HealthChoices, Allegheny’s managed care program for behavioral health services. We found that by including information that tracks whether someone has ever been associated with these systems, the AFST could have produced greater disparities in Black-white family screen-in rates than an alternate model design that did not take these factors into consideration.

Fig. 3. As household sizes increase, the likelihood that at least one member of the household will have some history with the Allegheny County Jail, the HealthChoices program, or juvenile probation increases as well. Black households are disproportionately likely to have involvement with these systems.

Previous reporting has shown that many families do not even know the county is using the AFST, much less how it functions or how to raise concerns about it. Furthermore, government databases, from public benefits databases to criminal justice databases, are rife with errors. And what happens if the tool itself “glitches,” as has already happened to the AFST?

These challenges demonstrate the urgent need for transparency, independent oversight, and meaningful recourse when algorithms are deployed in high-stakes decision-making contexts like child welfare. Families have no knowledge of the policies embedded in the tools imposed upon them, no ability to know how the tool was used to make life-altering decisions, and are ultimately limited in their ability to fight for their civil liberties, cementing long-standing traditions of how family regulation agencies operate.

Read the full report, The Devil is in the Details: Interrogating Values Embedded in the Allegheny Family Screening Tool below:

American Civil Liberties Union

The Devil is in the Details: Interrogating Values Embedded in the Allegheny Family Screening Tool | American Civil Liberties Union

We're examining how algorithmic design choices function as policy decisions through an audit of the Allegheny Family Screening Tool (AFST.)