Regression models in standardized test prediction

The objective of this study is to identify the predictors of graduation exam passing rates on standardized tests administered high schools in the United States using weighted linear regression models. Data from high schools in California, Illinois, and Texas were analyzed separately to obtain state-wide predictors, as well as together as one large dataset for overall predictors. Regression models containing only student-related variables, such as the percent of minority students, percent of low- income students, percent of Limited English Proficient students, were better predictors than models that contained only school-related variables, such as total enrollment and number of grade levels. When compared to the best overall model for the combined dataset, all of the states used subsets of the model. Within Texas, the predictors for the percentage of Black students passing were different from those for White and Hispanic students.