We explore the possibilities for measuring school value-added using published data that are aggregated by school and grade as an option for situations in which student-level data are unavailable. We investigate two kinds of value-added measures from public data: one that uses published average assessment scores by school, grade, subject, and year; and another that uses estimates of average assessment scores produced from published school proficiency data using heteroskedastic ordered probit. We find, using data from a consortium of districts in California, that value-added measures produced from published, aggregated data are highly correlated with value-added measures produced from student-level data. Although we do not believe that school value-added should be measured from aggregated data when student data are available, these results suggest such measures may be useful for some applications, particularly in lower-stakes cases, in the absence of student data.
The appropriate selection of covariates in value-added models (VAMs) remains a subject of considerable debate, particularly regarding its implications for the accuracy of school-effectiveness measures. This systematic review synthesizes evidence from 25 empirical studies conducted across multiple national contexts and published between 1996 and 2023. The findings suggest that incorporating relevant covariates can improve the accuracy of school-effectiveness measurement in VAMs. Specifically, students' prior achievement, particularly prior achievement in the same subject, was consistently identified as a salient covariate. At the student level, demographic characteristics and socioeconomic status were also frequently identified as relevant covariates. In contrast, evidence concerning the relevance of school-level covariates was mixed. Moreover, the contribution of covariates to school-effectiveness measurement varied across subject areas, educational stages, measurement intervals, and student population characteristics. These findings provide methodological guidance for covariate specification in VAMs and offer practical implications for educational policy and school management.
Public laws and professional guidelines in educational testing require that students with disabilities and English learners be provided with accommodations to standard test administration conditions if those conditions reduce their access to the assessment. However, changes to standardized test conditions can alter the construct measured and lead to scores that may not be comparable to those from standard test administrations. Thus, research is needed to evaluate the effects of test accommodations. In this study, we illustrate data analysis strategies and two statistical procedures that are not commonly applied to this problem - person-fit and multidimensional scaling analyses - researchers can use to evaluate the comparability of scores from standard and accommodated test administrations.
