Bulldogtech.org

Bulldogtech.org

How to Improve Accuracy of Data SGP Estimates

How to Improve Accuracy of Data SGP Estimates

Data sgp is a package for organizing longitudinal student assessment data into statistical growth plots. It allows educators to quickly create reports of student percentile ranks, which help families understand how their children are performing in comparison to similar students. These percentile reports can then be shared with teachers to highlight student progress and provide context for future instruction. The sgpData_CONTENTS page of this website provides examples of these percentile reports.

A student’s growth percentile indicates how much they have improved compared to other students with similar prior test scores (their academic peers). The value of this statistic lies in its ability to describe relative performance. For example, if a student’s math growth percentile is near the 40th percentile, this indicates they have made significant improvements in their knowledge of mathematics, but are still significantly behind their peers. This information can be useful in guiding future instructional decisions, for example, by helping educators identify what topics or concepts students need to focus on for further improvement.

We have already established that true SGPs are correlated between math and English/Language Arts, and that they are related to student background characteristics. This article shifts our attention to investigating to what extent these relationships could be exploited to improve the accuracy of SGP estimates. For this purpose, we construct a model for latent achievement attributes and define true SGPs within this framework. We then demonstrate that using only the current and prior math test scores of a student, one can estimate e4,2,i, a student’s math SGP. Moreover, we show that estimating this SGP using varying amounts of data results in varying error variances.

The resulting error variances can be interpreted as the RMSEs for conditional mean estimators of e4,2,i conditioned on different amounts of data. As expected, the RMSE decreases as the amount of data used to condition on the SGP increases. However, a key finding is that the RMSEs are not inversely proportional to the square root of the MSE, as one might expect given the fact that the MSE increases as the number of points added to the estimate grows.

We then extend this basic idea to establish multi-year growth standards based upon official state achievement targets/goals. This is done by estimating what growth, expressed as a per/year growth percentile, is required for a student to reach their achievement target. This is accomplished by establishing what the growth standard is and then quantifying how much the student must grow, both in percentage of their prior test score, to attain it. This approach is unique to SGP methodology and allows for the establishment of both what a student must achieve and how much they have actually grown, both of which are important information to stakeholders.