Comparison of quantile regression models on SGP estimation and the effect on classification consistency and accuracy

There has been a movement in state education to measure student progress not only on their achievement but their academic growth over time. One common way to measure this growth is through student growth percentiles (SGPs), which compare students’ academic growth to other students of similar prior ability. This study investigates how consistently and accurately students can be classified into growth categories like “high/medium/low” growth based on their SGPs.

In addition, this study investigated how certain factors impacted the classification consistency and accuracy of SGPs. One such factor was the model used to estimate student abilities at each time point. Similarly, two different models for estimating SGPs were compared. The number of categories into which student growth was being classified (two, three, or four categories) was also explored. Finally, models that included each student’s classroom membership were compared to similar models that did not include each student’s classroom membership. This was investigated because ignoring effects related to a student’s classroom can lead to wrong conclusions being drawn about student growth.