The data sgp package provides the tools and functions to run student growth percentile analyses using large scale longitudinal education assessment data. The analysis uses a combination of sparse generative models and variational inference to estimate conditional density distributions associated with each student’s achievement history. These distributions can then be used to construct student growth projections/trajectories which show the expected percentile gains needed for students to reach future performance targets.
The package is written for the statistical programming language, R, which is available on Windows, OSX and Linux and is free. Running SGP analyses requires a computer with sufficient memory to process the large state assessment data sets. A good rule of thumb is to use 4GB of memory per core utilized during the analysis.
Student Growth Percentiles (SGP) compare a student’s growth to that of their academic peers – students in the same grade with similar score histories on Renaissance Star assessments. This measure helps teachers highlight the significant growth their students are making, even if those students are not yet meeting grade level standards.
While the calculation of SGPs is complex, it is straightforward to implement. This is because the calculations rely on well established and proven methods, including the use of student averages to calculate relative performance and percentiles and the use of variational inference to model the posterior Gaussian distribution.
The SGP package has been designed to make it easy to use these established and proven methods for calculating student growth percentiles and to construct student growth projections/trajectories. The majority of the work required to implement these analyses is spent on preparing and processing the assessment data. Once this work is completed, the calculations can be executed quickly and accurately on a desktop or laptop computer.
SGPs can be calculated for individual students or entire cohorts of students. Currently, the SGP package supports two common formats for representing longitudinal data: WIDE and LONG format. The package installs exemplar WIDE and LONG data sets, sgpData and sgpData_LONG, to assist in setting these up.
When calculating SGPs, the most important step is to ensure that the underlying data is correct. This includes ensuring that the number of assessments taken by each student is correct and that the appropriate assessment scales are applied to the data. Detailed information on the requirements for valid data can be found in the sgpData vignette and in the SGP data analysis document.
The sgpData vignette also explains how to create a sgpstateData object, which contains all of the state assessment data for a student in one year. This can then be used to calculate sGPs and mSGPs. The mSGP calculations are only available for teachers of courses that have valid sgpData and which contain students with the necessary score history to generate an mSGP. The course roster submission process through NJ SMART determines which courses are eligible to receive mSGP scores and which teachers will receive them. Teachers of the eligible courses are notified when their mSGP results are available in fall 2024.
