Student growth percentiles (SGP) describe a student’s growth compared to students with similar prior test scores (“their academic peers”). Although the calculations for SGPs are complex, information can be shared in percentageile terms that are familiar to teachers and parents. This makes SGPs an ideal candidate for future accountability systems that emphasize student growth.
SGPs can be used to identify trends in student performance over time, as well as by student groups such as gender, race and socioeconomic status. These insights could then be incorporated into curriculum and instructional practices that increase student achievement. In addition, SGPs are more accurate than value-added models (VAMs) at predicting student achievement. However, while SGPs are a good candidate for future accountability systems, they do not capture the full complexity of learning.
To run SGP analyses, you need access to a dataset of student assessments that contain both raw scores and percentiles. sgpData is an anonymized, panel data set that contains 5 years of annual, vertically scaled, assessment data in WIDE format. This exemplar data set models the format for data used with the lower level studentGrowthPercentiles and studentGrowthProjections functions. However, for operationally operational SGP analyses year after year it is better to use the higher level wrapper functions with LONG data.
LONG data provides several preparation and storage benefits over WIDE data. However, you must be willing to invest in the additional resources required to store and process long data sets. For example, you will need a machine with enough memory to hold the entire LONG dataset, a fast network connection for processing large files, and access to R. These additional resources are not always feasible or cost-effective, particularly for schools with limited financial support.
The sgpData package is available on CRAN and can be installed on any computer that runs the open source software R. A basic knowledge of the R programming language is required to run SGP analyses. A more in-depth knowledge of the underlying statistical methods is also necessary, though not strictly a prerequisite.
SGP analyses can be executed on a single machine or distributed across a network of machines. The sgpData package is designed to support this type of distributed analysis. To execute SGP analyses on a network, you will need to install the sgpData software on each machine that will participate in the analysis and be able to communicate with the server running the SGP analyses. For more details, see sgpdata_installation_requirements.