Student Growth Percentiles and Projections
Student Growth Percentiles (SGP) place a student’s performance in the context of academic peers who have similar assessment score paths, regardless of their demographic characteristics or educational program participation. In order to do this, up to five years of test data are used to establish a student’s comparison group. For example, in spring 2024, Simon’s SGP places him among a cohort of sixth grade students who have statistically similar assessment scores to his own.
A statistical procedure called quantile regression is used to identify a student’s academic peer group. For the purposes of SGP analyses, academic peers are all Washington state students who are in the same grade and subject as a given student and who have scored similarly on MCAS in previous years. This process is repeated for up to two additional years in order to produce a set of test results that are representative of the performance of academic peers in each of the identified grades.
In addition to identifying the student’s academic peers, SGP models are also used to produce both Student Growth Percentiles and student growth projections (both cohort and baseline referenced). Both of these calculations require a number of different steps which are conducted simultaneously in operational SGP analyses. In order to simplify the source code associated with running these analyses, the SGP package has a set of wrapper functions (SGP()) which combine and streamline the lower level analysis steps.
The sgp() function takes long formatted data like that contained in sgpData and produces both the SGP percentiles and projections. In addition to a long formatted data source, sgp() requires a list of instructor-student lookup files (sgpData_INSTRUCTOR_NUMBER) and a boolean argument indicating whether or not SGP() will utilize the base case matrices from the specified SGPstateData.
If the boolean argument is not present or is not TRUE, then SGP() will only produce projections when a complete set of coefficient matrices is available for all years being projected. This is a necessary condition to ensure that all confidence intervals are constructed using the same coefficient matrix for each year being projected.
The SGP() function can be run in a variety of ways, but all SGP analyses must be conducted using the R software environment. This is available for Windows, OSX and Linux and, being open source, can be downloaded for free. Before conducting any SGP analyses, it is recommended that you familiarize yourself with the basic features of the R software environment. The following resources can be helpful in doing so.