Data SGP
The data sgp package provides an efficient means of organizing longitudinal (time dependent) student assessment data into statistical growth plots. The package does this for state wide data sets using either WIDE or LONG formatted data. Most lower level SGP functions such as studentGrowthPercentiles and studentGrowthProjections require WIDE formatted data, but higher level functions (wrappers for these lower level functions) can work with either the wide or long data formats.
The basic model used by the sgp functions relates latent achievement attributes to a student’s actual performance on assessments. These latent traits are assumed to have a normal distribution with a standard deviation of 1 (which is the standard error in linear regression). The model then uses this information to calculate a student’s current SGP. This is then compared to the student’s prior SGP. The difference between these two values gives the estimate of growth that a student has made relative to their academic peers.
SGPs are a powerful and useful tool for describing student achievement in terms that teachers and parents can understand. They also provide a measure of teacher effectiveness that can be used in educator evaluations. However, the calculation and interpretation of SGPs requires significant computing resources. This is a major reason why MDE has recommended that the use of SGPs for high stakes purposes should be delayed until 2018/19.
One of the challenges in interpreting SGPs is excessive measurement error. While it is impossible to eliminate all measurement error at the individual level, this error can be reduced by aggregating estimated SGPs to the teacher or school levels. However, this can create an additional source of error due to individual-level relationships between true SGPs and student background characteristics and differences in the types of students assigned to teachers.
This article explores the sources of this extraneous variation and illustrates how these can be detected by examining the distributional properties of SGPs. It also discusses how these can be addressed by using additional covariates.
To reduce the computational costs required to run sgp analyses on large state wide datasets, MDE has developed the data sgp package. This software allows users to create and analyze SGPs from a set of predefined covariates. The package also provides a number of other features for customizing and displaying SGPs. The package can be downloaded from http://www.mde.gov/data/sgp/.