Data SGP – How to Estimate Student Growth Percentiles (SGP) and Percentile Growth Projections (Projects)
The data sgp package provides classes, functions and data for estimating student growth percentiles (SGP) and percentile growth projections/trajectories using large scale, longitudinal education assessment data. The SGP model uses a number of advanced methods to perform this estimation, including sparse graphical models and variational inference. These approaches reduce computational complexity and memory usage (typically O(NM2)) while preserving predictive accuracy.
A SGP is a percentage that shows how much a student has improved over time in comparison to their academic peers with similar prior test scores. SGPs allow us to fairly compare students who enter school at different levels and provide a more complete picture of student performance than simply whether or not a student is passing state assessments.
In order to generate an SGP, a student must have a valid and comparable test score for at least one year prior to the current year of testing. The test used for this calculation must be taken in a testing window during the autumn, winter, or spring. A held back student will need to take two tests during two separate windows in their holding period before having an SGP generated for them.
Each year, a student’s SGP is calculated for each subject and grade by comparing them to their academic peers. The academic peers are all Washington students who have valid and consecutive test scores in the same subject and grade. A student’s academic peers are chosen based solely on their assessment history, not any other factors such as demographics or program participation.
The SGP data set sgpData_INSTRUCTOR_NUMBER contains an anonymized, student-instructor lookup table that associates each student’s test record with the teacher who taught them during the test window for that year. Each teacher can have multiple students associated with their test records in a single content area, but each student can only be associated with one teacher for a given year.
The lower level SGP analysis functions (studentGrowthPercentiles and studentGrowthProjections) use WIDE formatted data while the higher level wrapper functions, studentGrowthMetrics and studentGrowthTrajectories, require LONG formatted data. For most operational analyses that will be run year after year, the higher level SGP functions are recommended as they have many preparation and storage benefits over the lower level analysis functions when using the LONG data format. These higher level functions also often assume that a specific state’s embedded SGPstateData meta-data has been loaded for each year of analysis, but this is optional. See the SGP data specification for more detailed information on how to manage and access the SGP state specific metadata. If you have questions about using the SGP data, please consult our documentation or contact an SGP Data Team member. Almost all errors that occur when running SGP analyses revert back to data preparation issues so we recommend taking the time to prepare your data properly. This will ensure that your analyses are accurate and will help you make more informed decisions about instructional practices based on the data.