When Lynzi Ziegenhagen, ’93, graduated with a symbolic systems degree, her area of expertise was so new that people working in the field hadn’t yet agreed on what to call it. Some used names like data mining and knowledge discovery, but the one that stuck was data science—to describe the extraction of useful information from often quite messy data. As the field developed, many data scientists focused on enabling media-darling moon shots—self-driving cars or robotics—but Ziegenhagen had her eye on education. “I was trying to figure out how all these cool new things that are available could actually be helpful to regular, hard-working people doing their job to make the world a better place, like teachers and principals and district leaders,” she says.
In 2007, after more than a decade as a corporate technical consultant, Ziegenhagen saw the opportunity to bring together “the thing I love to do and was good at and the thing I deeply cared about.” Aspire Public Schools, a nonprofit charter school system serving low-income communities throughout California, hired her as a data analyst to help teachers and administrators evaluate their schools’ health and efficacy. At the time, Aspire teachers used between 12 and 30 incompatible data systems to search for a student’s grades, standardized test scores, and attendance, among other metrics. “They would literally go page by page in the student information system to look up the number and write it down in a separate Google sheet or Excel file or on paper,” Ziegenhagen says. “And they did that for a hundred students, hypothetically.”
The challenges of America’s underfunded schools have hardly been a secret. A 2018 story in the New York Times, for example, ran with the headline “25-Year-Old Textbooks and Holes in the Ceiling: Inside America’s Public Schools.” That was before COVID-19 brought a slew of new challenges. Pre-pandemic estimates of how many teachers quit—or planned to—within their first five years of teaching ranged from 17 percent to 50 percent. Their reasons included poor pay, lack of respect, and far too much work time spent on documentation: of parent phone calls, student-teacher conferences, departmental meetings, and student behavior and performance, not to mention teacher self-evaluations and growth plans. As for lesson planning and grading, many teachers are left to do it at home. According to a 2022 National Education Association study, 55 percent of educators now say they are “more likely to leave or retire from education sooner than planned.”
The software that Ziegenhagen’s team built, based on interviews with the Aspire schools’ staff members, aggregates information such as student assessments, attendance statistics, and behavioral reports, allowing teachers to quickly access and compare metrics. For example, they can correlate poor grades with spotty attendance, and, when they assess English scores, take into consideration the language students speak at home. By the software’s third year, 90 percent of Aspire teachers reported that they used it regularly to support students. Ziegenhagen says they can “look up information in a minute versus an hour” and use the saved time to make plans for students who need help. “Then you can sleep, and maybe you’ll stay in the profession longer because you’re not working every Sunday night.”
The software has also “catalyze[d] conversations about equity issues,” Ziegenhagen says, because it allows educators to see inequities along racial lines—for instance, the breakdown by racial background of the students who take AP classes, or the ones who face disciplinary measures. “People may know there’s an issue around oversuspension of African American students,” she says, “but it’s not until they see that it’s happening in their school and at a dramatically different rate that the conversation is very different—when it’s not just a problem out there in the world but a problem ‘in my classroom and I am part of it.’ ” Similarly, the software has helped bring to light gender-specific issues. In one school with low attendance numbers, the software identified that ninth-grade girls were frequently absent. This allowed administrators to dig into the cause—which turned out to be anxiety around social dynamics—and create a support system for that specific group.
Ziegenhagen created her own company in 2011 so that other schools could use the software too. That company, dubbed Schoolzilla, sold in 2019 to education technology firm Renaissance Learning, where Ziegenhagen now serves as senior vice president of product platform interoperability. Educators from some 2,100 school systems serving 20 million students use her software.
Drew Sarratore, principal of William G. Paden Elementary in California’s Alameda Unified School District, says Schoolzilla turned out to be particularly valuable during the first year of the COVID-19 pandemic, when his staff used it to identify and support the students most likely to struggle with at-home learning, such as non-native English speakers and those qualifying for free or reduced-price lunch. “We improved their growth metrics from the year prior to COVID,” he says, “in a time where it could have completely fallen apart.”
Sarratore says the service is also broadly useful for managing complex information. “I can compare my chronic absenteeism [data] to my academic growth data and add in my office discipline referral data and have everything in one place, where I can look across and pinpoint specific populations or grade levels or places where we can set really clear goals.”
Ziegenhagen highlights the importance of taking individual data points in context. She references a study that showed that a single F in a freshman-year course is the highest indicator of a student dropping out later in high school. With Schoolzilla, she aims to catch problems long before the F appears on the report card and to recognize which trends lead to which outcomes.
“You don’t fixate on the one number,” she says. “You are thinking about the overall health of the student and of the school system. I think that’s an important maturity step for education.”
Prison’s Big Picture
As an associate product manager at Google, Clementine Jacoby, ’15, focused on criminal justice reform for her “20 percent” project, an initiative allowing employees to divert that much salaried time to their topic of choice. Jacoby worried about mass incarceration and the overcrowding of prisons, in which Black, Indigenous, and Latinx people were overrepresented. “I grew up with a dad who is a political scientist and a mom who does addiction medicine, so it was sort of causally overdetermined that I would come to think that criminal justice reform was the most important public policy issue of our time,” she says. “I also grew up with family members who were in the system.”
Jacoby soon learned that prison administrators and parole officers didn’t have the information they needed to support people in the system, and nobody—neither administrators nor lawmakers and advocates—had the data to determine which programs and services worked best. “I was shocked that this system, which touched every community—and sits at the intersection of mental health challenges, addiction, economic mobility, and poverty—didn’t have modern analytics,” Jacoby says. She realized that providing as much information as possible would help parole officers to determine who needed support, prisons to decide whom to release, and lawmakers to assess how they might reform sentencing recommendations.
Jacoby and two co-workers decided to build an open-source database to clean and aggregate publicly available data about the criminal justice system, including information such as length of sentencing, incarceration rates, and reincarceration rates. “By the time we realized the enormity of the problem, we were in too deep—we had to start a company,” she says. In 2018, Jacoby and her colleagues launched Recidiviz (which now has 56 employees), and the next year she left Google to become the nonprofit’s full-time CEO.
Its name is a portmanteau of recidivism and visibility. Early on, Jacoby’s team focused on recidivism, derived from the Latin cadere—“to fall”—and thus meaning to fall back into criminality. But recidivism, she learned, “was measured differently in almost every context.” From state to state or county to county, a recidivism rate could indicate rearrest, reconviction, or reincarceration; it could be measured over one to 10 years; and it could be based on cohort—say, everyone who was released from prison during a certain period—or on everyone uncategorized. Different definitions made comparing the effectiveness of programs between states—or even between the United States and other countries—extremely difficult. “One of the key heartbeat metrics for the system was fundamentally incomparable,” Jacoby says, “and that was just the tip of the iceberg.”
The iceberg in question is a sprawling criminal justice system consisting of incarcerated people as well as those on probation (supervision in lieu of imprisonment) and parole (early release with monitoring). The problem of recidivism led her team to a much larger problem: “Criminal justice data is so fragmented that we can’t even tell you how many people are in jail today,” she says. The state departments of correction had a tough job, she realized. “Collectively, the 50 directors run an $80 billion system. They have hundreds of thousands of staff,” she says. “You probably have better analytics on your personal website than these folks get on their flagship programs.” Such a warning system is especially important in preventing reincarceration. “Once you go to prison, the likelihood that you will be rearrested within 9 years is 83 percent,” she points out. But people often age out of crime by 30. For social and developmental reasons, they become less inclined to take part in crime or less vulnerable to the circumstances fostering it. However, a criminal record often limits a person’s options for work and education, leading them to be more likely to commit crimes and get rearrested. If they’ve spent time in prison before age 30, Jacoby says, “they’re going to have a really, really hard time getting back on their feet.”
Recidiviz’s software consolidates information from a variety of criminal justice databases. “Instead of going through 13 steps, we just have one,” says Joshua Graham, district director for community supervision in the Tennessee Department of Corrections, where he oversees 70 employees in charge of more than 5,000 people on probation and parole. To decrease the officers’ workloads, Recidiviz’s online platform gathers information about individuals in the system and offers “a snapshot of the person,” Graham says. The software alerts parole officers when a person has met the eligibility requirements for release from supervision. It also sends notifications when people qualify for good-behavior programs. “It increases safety because it leaves less room for error,” Graham says. “It helps us reduce recidivism by giving us an opportunity to actively and efficiently manage the cases that we have.” Officers can then focus time and resources on the people most likely to struggle—those who are homeless, unemployed, or not receiving treatment. “These tools probably sound quite basic,” Jacoby says. “But they didn’t exist and in many places still don’t exist.”
To reduce prison overcrowding, Recidiviz saw an opportunity in COVID-19. The team forecasted potential outbreaks across the country according to prison capacity. By summer 2020, major COVID clusters were appearing in prisons and jails, and Recidiviz ended up collaborating with corrections leaders in 34 states to find people who were either eligible for release or could be released safely due to medical conditions, good behavior, or time served. “The federal prison system released 11,000 people early during COVID-19,” Jacoby says, “and only 17 committed new crimes, only one of which was violent.”
Recidiviz also measures racial disparities at transition points—sentencing, reduced sentences for good behavior, parole, and revocation of parole—and compares disparities among counties and states to give “a more granular picture of what is happening in the system,” Jacoby says. This allows lawmakers, corrections officers, and advocates to see how race correlates with an individual’s experience in the system. Recidiviz has also supported advocates working on racial disparities in sentencing for crack and cocaine. Though nearly chemically identical, the two drugs have different cultural associations: crack with the Black community, and cocaine with the white, especially in more affluent or glamorized contexts. “For a long time, crack cases have been sentenced far more harshly than powder cocaine sentences,” says Molly Gill, vice president of policy for FAMM (Families Against Mandatory Minimums). To build arguments for sentencing reform, FAMM appeals both to fiscally conservative politicians, many of whom favor reducing prison costs, and to progressive politicians intent on reducing mass incarceration. Recidiviz supplies FAMM with data models for both, showing how a reform could decrease expenses while reducing prison populations. “Recidiviz provides these services for free to a lot of nonprofits,” Gill says. “That’s incredibly valuable. Most criminal-justice nonprofit organizations don’t have the staff or expertise or resources to do this kind of analysis themselves.”
Though Recidiviz now offers pro bono policy impact modeling to lawmakers and advocates in 19 states, renders services to 11 state departments of correction, and provides public-facing dashboards for researchers, journalists, and community members, Jacoby doesn’t see tech as the answer to pressing social problems. “Tech is not great at creating political alignment or changing hearts and minds. To do that well, you need to listen to people in the field, on the ground, who have been doing the work for longer than you have.” Rather, she views it as an accelerant: “That’s where tech can be so powerful.”
Deni Ellis Béchard is a former senior writer at Stanford and the author of eight books. Email him at firstname.lastname@example.org.
A Worldwide Web of Women
By Kali Shiloh
In the midst of an exciting period of growth in the field of data science, Margot Gerritsen had a frustrating realization. A mathematician and computer scientist by training, Gerritsen, PhD ’97, had routinely used data science in her work since becoming the first woman hired to the faculty of Stanford’s petroleum engineering department in 2001. She was one of the few women in the room wherever she worked, but bit by bit, she saw signs of progress. Yet in 2015, she attended a string of conferences featuring the “heroes” and “experts” of data science, not one woman among them. “The representation is nothing better than I experienced when I first started,” she remembers thinking.
So Gerritsen, now a professor emerita of environmental resources engineering, decided to build a technical conference at Stanford featuring only female speakers. In November 2015, the inaugural Women in Data Science (WiDS) conference debuted to a sold-out crowd of 400 and thousands of online viewers. It was designed to showcase outstanding women in the field, but it was also intentionally small in order to facilitate career connections and social bonds. The idea struck a chord, and female data scientists from around the world soon signed on as ambassadors to start their own local WiDS conferences, which now number about 200 annually in more than 50 countries.
Among the ambassadors are a passionate group of eight Icelandic undergraduates, the sole female data science professor at Japan’s Yokohama City University, and data scientists in Saudi Arabia who were able for the first time to attend a conference without being accompanied by a man. Colombia native Cindy Orozco Bohorquez, MS ’17, PhD ’21, attended the inaugural conference as a doctoral student. “[WiDS] helped me to discover that we don’t need to be in a specific location (i.e., Bay Area) to do cool data science,” she said in a email interview with WiDS last year. Orozco Bohorquez went on to volunteer with the organization and eventually became a WiDS Worldwide speaker.
Kali Shiloh is a staff writer at Stanford. Email her at email@example.com.