The purpose of the study was to compare two regression-based approaches for measuring educational effectiveness in Tennessee high schools: the mean residual approach (MR), and a more general linear models (LM) approach. Data were obtained from a sample of 1,011 students who were enrolled in 48 high schools, and who had taken the Comprehensive Tests of Basic Skills (CTBS) in grade 8 and the Preliminary American College Test Plus (P-ACT+) in grade 10. Regression models were developed for predicting P-ACT+ scores from CTBS scores. In the MR approach, mean residual P-ACT+ scores were calculated for each high school from a pooled model. The LM approach included effect-coded high school dummy variables as predictors. Indicators of educational change, called Achievement Index (AI) values, were dveloped for each school; AI values were also calculated where race, coursework and grades, or entering achievement level was included in the models. The results show that the LM approach is more sensitive in identifying outlier schools and was less hampered by sample size constraints than was the MR approach. Statistical significance and identifying violations of the assumption of parallel slopes were also easier to determine. Five graphs and one table illustrate the discussion, and there are two references. (Author/SLD)