Student Growth Percentiles and SGP Data Sets
Student Growth Percentiles (SGP) provide a snapshot of a students’ performance on state assessments. They are based on trends in statewide data from the years preceding the current year. A number of factors can influence overall student growth, such as the flu pandemic or changing proficiency levels for a subject/grade. In such cases, a large percentage of students may experience more or less growth than expected.
To help teachers understand what their students’ SGP scores mean, the state provides technical resources on student growth and student assessment. These resources include background information on SGP models, and detailed explanations of how to interpret the various SGP scores.
SGP scores are reported on a 1-99 scale, with higher numbers indicating greater relative growth. The SGP score tells a teacher or administrator whether a student grew more than, less than, or about the same as their academic peers with similar MCAS score histories in the subject area being measured.
The SGPdata package contains 4 examplar data sets for use with SGP analyses. The first, sgpData, specifies the data in the WIDE format used by lower level SGP functions studentGrowthPercentiles and studentGrowthProjections. The second, sgpData_LONG, specifies the data in the LONG format used by higher level SGP functions abcSGP, prepareSGP, and analyzeSGP. The third, sgptData_INSTRUCTOR_NUMBER, specifies a teacher-student lookup table utilized to produce teacher level aggregates.
While the ability of SGP models to describe a range of test score relationships is significant, they have certain limitations that limit their usefulness. First, their computational complexity (O(N3) time and O(N2) memory) prevents them from being able to handle very large datasets. This complexity stems from the inversion of the covariance matrix K.
A common approach to addressing this issue is through the use of sparse GP approximation techniques, which reduce computational costs by reducing the size of the covariance matrix while maintaining model accuracy. While these methods are not without their drawbacks, they are a reasonable alternative to GPR models when the number of training points is relatively limited.
Lastly, the fact that the SGP score is based on comparing a students’ current MCAS scores to their academic peers’ previous MCAS score histories means that two students with very different MCAS score histories can have the same SGP. This is why the SGP scores are reported on a 1-99 relative scale rather than on a 5-point scale like other percentiles.
The state also reports a Median Student Growth Percentile (mSGP). mSGP measures the average rate of change of a student’s performance on statewide assessments over time, in comparison to the median student’s score history in the subject/grade being tested. The mSGP data is available through NJ SMART following the submission of district course rosters to the system. This data is not provided for Science, Writing, EOC Biology or EOC Math.