Measuring Student and School Progress with the California API

This paper focuses on interpreting the major conceptual features of California’s Academic Performance Index (API) as a coherent set of statistical procedures. To facilitate a characterization of its statistical properties, we first cast the index as a simple weighted average of the subjective worth of students’ normative performance and present its estimation in the form of a linear model. In the process, we illustrate with an example several problems with this index for the study of a school’s year-to-year progress. In its current usage the API lacks realistic estimates of precision and, on closer examination, further misrepresents conceptually student and school performance. We present an alternative analysis of the API index, based on a Bayesian meta-analysis of results from school-specific multilevel models of longitudinal student test scores. We introduce a display for the precision of estimated relative gains of each school in the form of a profile that represents the probability that a gain estimate exceeds set fractions of the distance the pretest is from the statewide target of 800. Along with estimates of their reliabilities, we also produce rank estimates of school API gains rather than simply ranking schools. We illustrate our approach with an elementary school student cohort who took the Stanford 9 at the Long Beach Unified School District in the Spring of 2000.