ERIC Number: ED371033
Record Type: RIE
Publication Date: 1994-Apr
Reference Count: N/A
Are Multilevel Techniques Necessary? An Attempt at Demystification.
Kreft, Ita G. G.; Yoon, Bokhee
The merits of the multilevel model for educational research and its uses for school effectiveness research are considered. The main goal of the paper is to establish what intelligent applications of multilevel models can do, helping researchers decide what they must do to make a rational choice between models. Multilevel models are random line models. The data is assumed to be hierarchically nested, with the lower observations nested within the higher levels, resulting in intraclass correlation. Random line models are random coefficient, fixed variable models. The parameters of the models and some alternative estimation procedures are explored. Monte Carlo studies illustrate the use of these models. It is concluded that the random line model is technically an improvement over the traditional multiple regression model because it calculates the correct standard errors. In addition, researchers interested in prediction for separate schools will find the random line model technically superior because the procedure improves the estimation of the parameters for the separate schools. Random line models can be useful, but will often lead to the same conclusions as classical regression models. (Contains 38 references.) (SLD)
Descriptors: Correlation, Data Analysis, Educational Research, Effective Schools Research, Elementary Secondary Education, Error of Measurement, Estimation (Mathematics), Instructional Effectiveness, Models, Monte Carlo Methods, Prediction, Regression (Statistics), Research Problems, Statistical Studies
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: N/A
Authoring Institution: N/A
Identifiers: Multilevel Analysis; Random Line Models
Note: Paper presented at the Annual Meeting of the American Educational Research Association (New Orleans, LA, April 4-8, 1994).