The 131st Annual Meeting (November 15-19, 2003) of APHA

Abstract #61308

Elizabeth Baumler, PhD and Kathryn M Cardarelli, MPH. Center for Society and Population Health, University of Texas School of Public Health, 1200 Herman Pressler, Suite E 935, Houston, TX 77030, 713/500-9439, ebaumler@sph.uth.tmc.edu

To elucidate the pathways by which education influences health, we examined data from the U.S. Department of Education’s National Assessment of Educational Progress (NAEP). We analyzed state-level data collected on 4th and 8th grade children in schools in 1998 and 2000 and found a striking gap in academic performance between White and African-American students. Because the NAEP datasets are drawn using a multi-stage clustered sample, with data taken from students within the same cluster, application of traditional regression techniques to these correlated data would yield inaccurate estimates because of violation to the assumption of independence between observations. Inclusion of a random intercept into the model allows for the estimation of both the variance between schools as well as the variance between students. We used a model where yij denotes the NAEP achievement score measurement on the ith student in the jth school. Let x1j, …..,xpj , p =1, 2, … s where s is the number of independent variables such as parent involvement and home and family environment, which vary at the student level. Independent variables that vary among schools, remaining constant between students, may also be entered into the model. Our analyses illustrate the necessity of applying appropriate analytical techniques when intra-class correlation is present by comparing our results with those obtained using traditional techniques. Our research suggests that multilevel modeling methods may serve as a useful tool to investigators wishing to incorporate individual and contextual variables into a model.

Learning Objectives:

• At the conclusion of this session, the participant will be able to:
• 1. Articulate the contribution of multilevel modeling in public health research.
• 2. Build a multilevel model incorporating salient predictors of academic achievement to disseminate to policy makers.
• 3. Demonstrate the impact of failure to apply the correct analytic procedures to correlated health data.

Keywords: Statistics, Education

Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.