Identifying at-risk students by early warning systems is bogus.
Early warning systems are predicated on deficits. Usually the data used to identify students at risk is limited to attendance, grades, accountability test scores and behavior during a specific time period. Such information is too limited to predict who is at risk. Such systems do little for students that do not reside in the deficit model. More important, valid predictions require more data than that employed in early warning systems. The inherent flaws that exist in early warning systems include at-risk designations based on limited events, failure to account for historical trend lines that defy membership in a category, disproportionate weight being assigned to too few incidents, and/or failure to credit student performance because of behavior or an event.
An example includes the student whose academic performance is minimized because of excessive absenteeism. Another example is a precipitating event that defies historical patterns of growth and achievement — a fight in the cafeteria that results in a suspension that puts the student at risk. Students whose grades do not align with state accountability exams can be deemed at risk inappropriately. Perhaps the most dangerous application of early warning systems is in the early grades where the acquisition of on-grade reading skills is highly variable. Failure to demonstrate on-grade reading skills can result in retention putting the student at risk for the balance of their K-12 experience. The influence of each example could be minimized by analyses of all related data. Here is where the use of predictive analytics exceeds the limitations of early warning systems.
In education, predictive analytics must depend upon clearly identified predictable events and an array of all data that support predictions within a margin of error. Clearly, the more data supporting the prediction, the more credible the prediction is likely to be. Practitioners must be able to view all historical data associated with a student. A review of current data or that from the previous academic year is inadequate because trends and patterns are critical to meaningful, non-punitive predictions. The historical data stored in student information systems and/or archived in data warehouses is essential to predictive analytics.
Trouble is, such data is not readily available to teachers or administrators for appropriate analysis. Meanwhile, students are subject to ill-informed predictions. The time for predictive analytics in education is now.