Consequences of Misspecifying Levels of Variance in Cross-Classified Longitudinal Data Structures

Frontiers in Psychology

Chris eSchatschneider¹  Donald L Compton¹  Yaacov ePetscher¹  Jennifer eGilbert² 

[1] Florida State University;Vanderbilt University;

关键词: reading;    longitudinal analysis;    multilevel modeling;    Oral reading fluency;    cross-classified model;   

DOI  :  10.3389/fpsyg.2016.00695

来源: DOAJ

【 摘 要 】

The purpose of this study was to determine if modeling school and classroom effects was necessary in estimating passage reading growth across elementary grades. Longitudinal data from 8,367 students in 2,989 classrooms in 202 Reading First schools were used in this study and were obtained from the Progress Monitoring and Reporting Network maintained by the Florida Center for Reading Research. Oral reading fluency (ORF) was assessed four times per school year. Five growth models with varying levels of data (student, classroom, and school) were estimated in order to determine which structures were necessary to correctly partition variance and accurately estimate standard errors for growth parameters. Results illustrate the necessity of modeling classroom cross-classification and school nesting when predicting student growth in ORF across grades. Because ignoring higher-level clustering inflated lower-level variance estimates, the authors recommend including both classroom and school effects so that predicting variance in cross-year growth parameters will not lead to biased results.

【 授权许可】

Unknown