A Multilevel, Multivariate Model for Studying School Climate With Estimation Via the EM Algorithm and Application to U.S. High-School Data

Abstract

In many studies of school climate, researchers ask teachers a series of questions, and the responses to related questions are averaged or summed to create a scale score for each teacher on each dimension of climate under investigation. Researchers have disagreed, however, about the analysis of such data: Some have utilized the teacher as the analytic unit, and some have utilized the school as the unit. In this article, we propose a three-level, multivariate statistical modeling strategy that resolves the unit-of-analysis dilemma and unifies thinking about the analysis in such studies. A reanalysis of U. S. high-school data illustrates how to estimate and interpret: (a) the level of interteacher agreement on each climate dimension; (b) the internal consistency of measurement at the teacher and school levels; and (c) the correlations among “true” climate scores at each level. A linear model analysis utilized teacher control over school and classroom policy and teacher morale as bivariate latent outcomes to be predicted by school-level variables (e.g., sector, size, composition) and by teacher-level variables (e.g., education, race, sex, subject matter). Implications for conceptualization, design, analysis, and interpretation in future studies of school climate are considered.