Student Growth Percentiles (SGPs) compare a student’s assessment score on any test to that of his/her academic peers who have scored similarly on prior assessments. SGPs serve as an invaluable indicator of progress even when not meeting grade level standards, enabling teachers to celebrate important strides made while also identifying those in need of intervention before they lag behind.
OSPI has created the Student Growth Report as a tool to assist educators with understanding SGPs – it can be found in our Tools for Teachers section of our website and is intended as an overview. However, its primary use will be as an initial starting point for further investigation and training on this subject matter.
To conduct SGP analyses with this tool, educators must first format their data correctly. Since most of the time is spent preparing and running SGP analyses rather than calculations themselves, this tool was specifically created to be as user-friendly and straight forward as possible.
Data preparation for SGP analyses involves two steps: 1. Formatting data to either WIDE or LONG format and 2. Calculating SGPs using this data. Lower level functions like studentGrowthPercentiles and studentGrowthProjections rely on WIDE data, while wrapper functions such as SGPs prefer LONG data, with LONG having numerous preparation and storage advantages over WIDE format. In most instances, LONG is best when conducting SGP analyses on an ongoing basis since it offers greater preparation benefits over WIDE format in terms of preparation and storage advantages over WIDE format.
sgpData is an anonymized panel data set containing five years of annual, vertically scaled assessment scores from each of five years for five student identifiers; assessment scores per year in columns 2-5 are followed by five columns showing growth percentiles as students progress through their state’s window system; finally the last five columns contain student growth percentiles from year five onward.
If your data exceeds memory capacity, the sgpSimulator tool provides an alternative. By using a stratified random sample of test results from previously established longitudinal datasets as well as coefficient matrices from those datasets, sgpSimulator creates synthetic growth curves (SGPs). Furthermore, confidence intervals based on those SGPs will also be created and returned in PDF form for review.
In addition to sgpSimulator, this tool also supports BASELINE referenced projections and lagged SGPs as well as being capable of including test CSEMs into its calculations. Because sgpSimulator and SGPs produced with it can be computationally intensive, it is advised to only run them on quad core machines with at least 4GB memory per core; this amount should provide enough space to run all three programs at the same time on most laptop or desktop computers without reaching memory errors or out of memory errors and inability to complete analyses.
