Editor’s note: After the author wrote this column, she attended the Marjory Stoneman Douglas High School Public Safety Commission Meeting and was heartened that members appeared receptive to her technological considerations shared during a public comment period.
Hoping to prevent another mass school shooting, Florida is assembling a huge central database called the Florida Schools Safety Portal. It will use a shockingly large amount of data from state agencies, school records, mental health data and social media monitoring to flag warning signs of a student who might be the next school shooter. Even setting aside questions of privacy, security and civil rights, how do you actually build such a data-driven system?
I’ve researched the Florida initiative in my work as a fellow with the Aspen Tech Policy Hub, and I am troubled by the lack of transparency about how the safety portal actually functions. I have worked with extremely large student datasets, and I know how much time and expertise it takes to create a centralized database of student information. I am also aware of the biases and issues of equity and fairness inherent in technology design. The Statewide Threat Assessment Database Workgroup was established to make recommendations for technical aspects of the safety portal. The group consists of people from several state agencies, including the Department of Law Enforcement, Department of Children and Families, Department of Juvenile Justice and the Agency for Health Care Administration. But it’s not clear if technical experts were included.
Preventing school shootings through data is a herculean task fraught with ethical and technical considerations. When developed in the absence of technical expertise about ethical and fair systems, tools such as the Florida Schools Safety Portal can harm students and their families. Say, for example, a student who happened to be in foster care was incorrectly flagged as a threat in the centralized database. If a developer does not identify and fix this algorithmic error, over time, the system will incorrectly correlate every student who has been in foster care as likely to incite school violence. Bringing technical experts into conversations about the design, development and deployment of school safety systems can fix these problems on the front end. Some of the basic questions we should be asking include:
Is it even possible to use data to predict school shootings? It is almost impossible to create such a system. There just isn’t enough information about the small number of students who actually commit mass acts of violence, compared to data about students who make up the entire school population. There are too few data points to accurately model school shootings because they are statistically rare events. Any conclusions drawn from an analysis of the data will be unreliable at best.
What is the quality of the data being used in the system? Before bringing large amounts of data from separate sources into a single database, developers need to “clean” the data. That involves identifying, removing and correcting inaccurate information in each student record. Otherwise, the system will be rife with errors and inconsistencies, and team members who use the data will make poor decisions based on distorted insights. Once cleaned, data from each record must then be cross-referenced with all other records for each student, a process called data integration. In addition, data errors will slow down the time it takes to access the data. It would most likely take a team of 10 people working full time at least one year to clean all the data proposed for use in the safety portal.
What are the potential biases? The data sets included in the database are problematic from the outset due to their high correlation with race, gender and socioeconomic status. For instance, 47 percent of intakes in the Florida Department of Juvenile Justice for fiscal year 2017–2018 were black youth, while 60 percent of children placed in out-of-home care through the Florida Department of Children and Families are white. Other potential entry points for bias lie in the creation of algorithms used to analyze the data (algorithmic bias) and in the people who will view student data, such as threat assessment team members (observation bias). Schools must ensure that teams are trained and pre-screened for biases. Procedures should be established to prevent conscious and unconscious prejudices against student groups that are disproportionately impacted by digital surveillance technologies and increased policing.
Policymakers should be transparent about the details of how the Florida Schools Safety Portal operates, accountable for evidence that it actually works, and that it will be fair to all students. It is unwise to neglect the insights of technical experts, students, parents, educators, community members and other stakeholders in the development of such a data-driven school safety system.
Ora D. Tanner is assistant director of the Office of Undergraduate Research at the University of South Florida. She previously worked as a nuclear physicist, science educator and more recently as a graduate researcher on NSF-funded grant projects related to digital game-based learning and assessment. She spends her free time working with several nonprofit organizations to help equip underserved students with technology skills. She is the parent of three children who graduated from Hillsborough public schools and is currently a doctoral researcher in educational technology at USF. She spent the summer in San Francisco as a fellow at the Aspen Tech Policy Hub where she completed a technology policy project on data-driven school safety initiatives.