Analyzing Multilevel Data in the Presence of Heterogeneous Within-Class Regressions
Leigh Burstein, Robert L. Linn, Frank J. Capell
Abstract
Leigh Burstein, Robert L. Linn, Frank J. Capell
Abstract
The concerns of this investigation are multiple sources of complications in the analysis of multilevel educational data. The focus is on problems that arise when within-group regressions of outcome on input are related to teacher/ school characteristics. Single-level and multilevel analytical approaches are applied to hypothetical data for which the relationship between teacher/class quality and the heterogeneity of within-class slopes is varied systematically. It is shown that single-level analyses and the proposed multilevel approaches can all yield mislèading estimates of teacher/class effects on mean class outcomes. However, selected multilevel methods provide some indication of misspecification and can identify the direction of the bias in estimating teacher/class effects on mean class outcomes.
OpenAlex reports 95 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
The concerns of this investigation are multiple sources of complications in the analysis of multilevel educational data. The focus is on problems that arise when within-group regressions of outcome on input are related to teacher/ school characteristics. Single-level and multilevel analytical approaches are applied to hypothetical data for which the relationship between teacher/class quality and the heterogeneity of within-class slopes is varied systematically. It is shown that single-level analyses and the proposed multilevel approaches can all yield mislèading estimates of teacher/class effects on mean class outcomes. However, selected multilevel methods provide some indication of misspecification and can identify the direction of the bias in estimating teacher/class effects on mean class outcomes.
Key concepts: Multilevel model, Class (philosophy), Statistics, Outcome (game theory), Multilevel modelling, Class size, Econometrics, Regression analysis