Using Data to Inform Decision-Making in Maryland Public Schools: Race to the Top Data Dashboard

Introduction: Dropout is a major concern for youth and school districts, as youth who drop out are at increased risk for a number of negative social, economic, and mental health outcomes. Significant predictors of school dropout can be incorporated into an early warning system in order to inform schools which students may be at risk for dropping out. Early warning systems utilize real-time data on the students within the school system and can be highly efficient, by focusing attention and resources to those students who are at the greatest risk. The current study utilized data from the Maryland State Department of Education (MSDE) to identify variables that predict high school graduation and dropout and determine practical risk cutpoints which could be used by schools to indicate when a student is in need of additional intervention.

Method: Data regarding all 8^th grade students in the state of Maryland during 2007-2008 school year were utilized for this study (N=67,148). Data from 5 years (i.e., 2007-8 through 2011-12) were used to examine predictors of not graduating. Multilevel modeling was conducted in HLM to predict the graduation outcome, using behavioral (i.e., suspension), grade retention, academic performance (i.e., on standardized assessments), and attendance (i.e., absences and mobility) data. Demographic (e.g., gender, race) data were also examined. Using the HLM results, beta coefficients were applied as “risk points” and cumulative points were analyzed to determine whether specific thresholds of risk for not graduating could be identified.

Results: Variables assessing behavior, retention, achievement, and attendance were all significant in predicting graduation, predicting the complexity promoting graduation. Grade retention stood out as the strongest predictor of not graduating among all included variables. The predictive accuracy using risk cutpoints for not graduating was very high and provided the State clear levels of risk for each grade level. These included 0-2 risk points for low level risk; 3-5 indicating elevated risk and the need for selective intervention; and 6+ risk points indicating nearly perfect prediction of not graduating and therefore implying intensive interventions are needed.

Discussion: The results of this study demonstrate that specific and clear risk thresholds can be utilized within data dashboards to inform data-based decision making in schools. These thresholds have predictive validity and, within the state of Maryland, identify expected percentages of students (i.e., 20%) outlined by a public health model. Therefore, it is expected that these data would be an efficient way to allocate resources and interventions to students. Further research is needed to test this hypothesis.