A Student Growth Percentile (SGP) is a measure of a student’s current achievement relative to students with similar prior achievement. This measure is appealing because it focuses on a familiar percentile rank scale that remains well-defined when test scores are not vertically or intervally scaled (Betebenner, 2009). Additionally, the SGP calculation uses the same formula as the standard deviation of a school mean, which makes it easy to compare growth results across schools.
When SGPs are aggregated to teacher or school levels, they can also be interpreted as indicators of educator effectiveness. However, it is important to remember that SGPs are only one component of a much more complicated picture of student progress. As with any measure, it is necessary to examine multiple sources of variation in order to understand the full story. For example, SGPs can be influenced by contextual effects, teacher sorting, and individual student characteristics.
In addition, the accuracy of SGP estimates depends on the reliability of the prior and current test scores. This article provides an overview of the factors that affect reliability and how they can be incorporated into SGP calculations.
The most common form of SGP data is longitudinal (time dependent) student assessment data. These data sets are typically structured in a WIDE format, where each case/row represents a unique student and columns represent variables associated with the student at different times. The SGPdata package, which is installed as part of the R statistical software suite, includes exemplar WIDE and LONG format data sets (sgpData_WIDE and sgpData_LONG) to assist users with preparing this type of data for SGP analysis.
This article describes how to use the SGPdata package to estimate SGPs using time-dependent student assessment data. It provides details of the model for latent achievement attributes, defines true SGPs under this model, and demonstrates how to evaluate the distributional properties of estimated SGPs.
Finally, the article discusses some of the limitations of SGP analyses and provides recommendations for future research. The authors highlight some areas for future exploration including incorporating contextual effects, teacher sorting, and individual students’ characteristics into SGP calculations, and comparing the stability of SGPs from year to year.
In general, the simplest and most reliable method to calculate SGPs is to apply a statistical modeling framework that incorporates both prior and current data (i.e., a regression-based SGP model). However, for more complex applications such as estimating teacher or school effects, it may be necessary to use a more sophisticated modeling framework such as latent variable models. For instance, hierarchical linear models (HLM) are more flexible than regression-based SGP models and are more likely to be robust to violations of assumptions such as normality. Therefore, they can be used to identify important effects in more complex education systems that cannot be easily identified with traditional regression-based approaches. These more complex models also enable researchers to control for the effect of confounding variables. This flexibility is particularly valuable in education, where many factors may influence the outcome of an intervention, making simple effects identification difficult or impossible.