Evaluating Random and Systematic Error in Student Growth Percentiles

Abstract

ABSTRACT Student growth percentiles (SGPs) are currently used by several states and school districts to provide information about individual students as well as to evaluate teachers, schools, and school districts. For SGPs to be defensible for these purposes, they should be reliable. In this study, we examine the amount of systematic and random error in SGPs by simulating test scores for four grades and estimating SGPs using one, two, or three conditioning years. The results indicated that, although the amount of systematic error was small to moderate, the amount of random error was substantial, regardless of the number of conditioning years. For example, the standard error of the SGP estimates associated with an SGP value of 56 was 22.2 resulting in a 68% confidence interval that would range from 33.8 to 78.2 when using three conditioning years. The results are consistent with previous research and suggest SGP estimates are too imprecise to be reported for the purpose of understanding students’ progress over time.