|Ann A. O'Connell (USA)||email@example.com|
|The analysis of data from longitudinal
studies is one of the more challenging course content areas for multivariate
methods classes. Students are often unaware of the many variants of longitudinal
designs that exist, and as a result may experience trouble moving beyond
the analysis of common pretest-posttest or pretest-post-posttest designs.
Indeed, as the study design becomes more complex, the possible patterns
of change that a researcher might be interested in modeling also become
In this presentation, I discuss the fitting and interpretation of hierarchical linear models (HLM) for data from longitudinal studies of HIV/AIDS prevention programs. Two types of data will be considered; longitudinal (or panel) data for the study of individual behavior change over time, and repeated cross-sectional data for the investigation of group-level (community) change over time. These examples illustrate two different units of analysis, individual-level change versus community-level change, yet both involve data collected repeatedly over time. Through these examples, several issues that can affect the development and/or interpretation of models for longitudinal designs will be highlighted, including model assumptions, data hierarchy, centering of independent variables, non-response across data collection periods, and estimation of model parameters. One of the advantages of using hierarchical linear models for longitudinal studies is that they offer an extremely flexible framework for assessing change over time, particularly with regard to model assumptions. Another benefit is the ability to characterize change over time using familiar regression models, and investigate the extent to which contextual variables, such as gender, intervention program, or community, may contribute to variation in these models.
Much of the work presented here grows out of my own teaching, research, and consulting activities in evaluating the efficacy of HIV/AIDS prevention programs. The examples I have chosen to present will illustrate many of the strengths and weaknesses of longitudinal and repeated cross-sectional data. The longitudinal study (O'Connell, 2001; Fisher et al., 1998) investigated the effect of an eight-week and a weekend intervention for seropositive gay males. The project was designed to increase risk reduction behaviors of participants over time, relative to individuals in a control condition. For each of the three groups, behavior and risk measures were collected on three occasions (pretest-post-posttest). Individual growth curves were fit to the data and analyzed within and across groups. The repeated cross-sectional study evaluated an HIV prevention intervention for women in high-risk communities (Lauby et al., 1999). This community-level study used four annual waves of data collection across eight communities (four intervention and four comparison). Random coefficients models were fit to the data to assess community change over time.
Hierarchical linear modeling provides a relatively new and mathematically sophisticated approach to longitudinal data analysis. Application of HLM will be stressed in this presentation. For optimal understanding of these models, a good working knowledge of regression concepts and methodology is suggested, as expected of students typically enrolled in multivariate methods courses. Extensive references will be provided.
|Download in Adobe Acrobat format (144 Kb).|
Back to top
ICOTS-6, The Sixth International Conference on Teaching Statistics - International Program Committee (IPC) Website.
Copyright © 2001 by the IASE. All rights reserved. This information is subject to change without notice. This page was last modified on July 2, 2002.
For questions or comments, contact the Webmaster, Dagan Ben-Zvi.