Student Growth Percentiles (SGP) are an innovative measurement that provide a more accurate picture of student achievement than traditional percentile scores do. They are calculated by estimating latent achievement trait models and comparing those estimates to growth standards established through teacher evaluation criteria and student covariates.
This allows teachers to understand how their students performed relative to their academic peers. It also helps them to identify areas for improvement in their classrooms. However, the model is complex and requires access to longitudinal data that is able to accurately track student progress over time. This has been a challenge for many school districts.
A growing number of districts are implementing SGP measures as part of their educator evaluation systems, but there are challenges to using this approach. For example, SGP models need to be able to capture a minimum of three years of stable assessment data in order to produce reliable results. Additionally, correlations between prior year assessments and the current year’s model are unlikely to be exactly zero, which can introduce error into the estimated SGPs.
In addition, SGP analyses are computationally intensive and require a significant amount of memory in order to perform properly. These factors have led some schools to opt to use the SGP results from the previous year only instead of the most recent assessment. This practice can lead to misleading reports because SGP results are often based on the median of a five-year trend of test score data.
Thankfully, the SGP analysis tools included in data sgp are designed to overcome these limitations and make the most of available longitudinal data. These tools are based on the statistical software application R, which is open source and is freely available for Window, Linux, Mac OSX, and other operating systems. Running SGP analyses typically requires a relatively powerful computer and a good understanding of how to navigate and manage large datasets.
The data sgp package includes an example WIDE format data set (sgpData) to simulate the time dependent data used by lower level SGP functions like studentGrowthPercentiles and studentGrowthProjections, as well as a LONG format data set to assist with converting that data into SGPdata format. Please refer to the SGP data analysis vignette for more comprehensive documentation of how to use these data sets and the higher level functions.
To run operational SGP analyses, district administrators must have a working copy of the data set sgpData_INSTRUCTOR_NUMBER, which is an anonymized teacher-student lookup table associated with each student’s test record in STAR. This table is updated regularly with new test records as they are submitted to STAR and provides the means for districts to connect instructors with students through unique identifiers attached to each student’s test record. Without this table, the ability of higher level SGP functions to generate student projections would be limited.