Data SGP is a longitudinal student assessment dataset that enables educators to identify areas of strength and weakness for individual students as well as classroom instruction. In addition, administrators can use the results to guide district-wide improvement initiatives. SGP analyses compare each student’s historical growth trajectories with projections from a selected base cohort. These comparisons can help teachers determine whether their students are on track to meet proficiency standards, and what their projected scores will be in future years.
The SGP package includes documentation, vignettes and examples that provide thorough explanations of its calculations and processes. These resources should be reviewed by users prior to conducting any analyses. If questions arise during the course of a SGP analysis, users are encouraged to contact support for assistance.
SGP analyses rely on a variety of techniques to estimate latent achievement trait models and compare them against growth standards established via teacher evaluation criteria and student covariates. These estimates are inevitably error-prone, as it is not possible to observe the actual values of student achievement traits directly with any degree of precision. Errors can be minimized by using a baseline-referenced model where the student’s original score is compared to that of an identical baseline cohort. However, the design of a baseline cohort can introduce other error sources.
For example, the initial average score may be influenced by differences in student demographics or by teacher-level characteristics such as class size or training. Furthermore, there is no guarantee that a student will remain with the same instructor throughout their educational experience. These factors can lead to spurious correlations in the estimation process. These issues can also lead to misinterpretation of SGP results, including overestimating student progress toward proficiency and the number of future years required to achieve that progression.
In order to reduce the impact of these errors, SGP analyses use an iterative methodology that recalculates the estimated latent achievement trait models for each student every year. This recalculation reduces the impact of changes in the estimated models caused by variation in the data or by errors in the calculation process. The recalculations are then used to generate new projections for each student, which are compared against the projected scores of the baseline cohort to estimate how close the student is to meeting the proficiency standard.
The SGP package includes exemplar WIDE and LONG format data sets for users to practice their analyses. The sgptData_LONG format includes student assessment records for 8 windows (3 windows annually) of longitudinal (time dependent) data. The sgptData_LONG data set is comparable to the sgpData_LONG data set but is anonymized and contains additional variables for creating aggregate student SGPs. The sgptData_LONG also contains a lookup table called sgpData_INSTRUCTOR_NUMBER that enables the user to associate an instructor number with each student assessment record. This lookup table is required for running SGP analysis with long data sets.