| With increasing emphasis being placed on improving student achievement on large-scale assessments, accountability systems across the country are demanding that the impact of teachers and schools on student achievement be evaluated, and many researchers have developed value-added models to address this issue. The purpose of this study was to compare hierarchical linear models to multiple regression models used in value-added approaches. Three key research questions were addressed: (1) Does the relative significance of each variable change as a function of the model being used? (2) Are the different statistical approaches equally sensitive to the inclusion or exclusion of certain key variables? Does variable selection impact the quality rankings? (3) Are the different statistical approaches comparably accurate? Do the quality rankings remain stable across populations or with different sample sizes?; The first statistical approach applied was a standard ordinary least squares (OLS) multiple regression model. The second statistical approach used was a two-level hierarchical linear model (HLM) and a three-level HLM. The predictors in these models included student specific variables, teacher specific variables, and school specific variables from two large school systems in Maryland. Schools and teachers were ranked using a residual analysis to determine the difference between students' actual mean performance and students' expected mean performance. |
