Labelled High Risk in Ninth Grade

What happened

Since 2012, Wisconsin’s Department of Public Instruction (DPI) has run the Dropout Early Warning System, known as DEWS. It scores students in grades 6 to 9 on how likely they are to graduate on time, and its inputs included race, gender, disability, and family income. In 2020–21 it labelled more than 32,000 students, about 15 percent, as high risk.

In April 2023 The Markup and Chalkbeat reported what DPI’s own 2021 analyses had found. When DEWS predicted that a student would not graduate on time, it was wrong nearly three quarters of the time. Its false alarms were 42 percentage points more frequent for Black students and 18 points more frequent for Hispanic students than for white students. In Kenosha in 2018, 43 percent of Black ninth graders were labelled high risk, against 11 percent of white ninth graders.

Who spoke up

Students. Reporters interviewed students who had carried the label, and in December 2023 two Black students at Cudahy High School, Mia Townsend and Maurice Newton, wrote a first-person piece about it. They wrote that they were surprised a state tool used race, that labels like this can discourage students, and that their own lives did not match what the system predicted. DPI at first defended the system; a spokesperson told reporters, “It’s math that reflects our systems.”

What changed

In September 2023 DPI said it was studying its early warning systems and considering changes. In October 2023 it removed DEWS data from the dashboards schools use.

What to keep straight

• The 42 and 18 figures are gaps in percentage points. DPI did not publish the underlying false-alarm rates.
• A study by researchers at UC Berkeley found that DEWS ranked students by risk fairly accurately. The reporting’s concern is mainly false alarms. Both can be true at once.
• No documented case shows a specific student harmed by a label. The concern is what labels do: to expectations, to support, and to how students see themselves.

Talk about it

• Is a warning that is usually wrong still useful if it gets help to some students who need it?
• Should race be an input to a prediction about a student? What would change if it were removed?
• Would you want to know if a system had labelled you? Who should be able to see the label?