|
Session Title: Individual Growth Curve Analysis and Structural Equation Modeling of Longitudinal Data: Alternatives to Hierarchical Linear Modeling (HLM)
|
|
Panel Session 295 to be held in Panzacola Section H2 on Thursday, Nov 12, 1:40 PM to 3:10 PM
|
|
Sponsored by the Quantitative Methods: Theory and Design TIG
|
| Chair(s): |
| Lee Sechrest, University of Arizona, sechrest@u.arizona.edu
|
| Abstract:
Hierarchical Linear Modeling seems to have become the default standard for analysis of longitudinal data. Alternatives do, however, exist, and they afford at least some advantages over HLM. One alternative is the analysis of individual growth curve parameters, primarily intercept and slope, by application of conventional statistics, including multiple regression. Another alternative is to include a chronological variable in a structural equation model. Data will be presented showing that the three methods of analysis of longitudinal data yield comparable, if not identical results, and the two methods alternative to HLM will be illustrated by analyses of actual data. The presentations will highlight the strengths of each approach as well as compare them to HLM.
|
|
An Empirical Comparison of Three Methods of Analysis of Longitudinal Data
|
| Katherine McKnight, Pearson Corporation, kathy.mcknight@gmail.com
|
|
A longitudinal data set on heroin use by veterans was available for analysis and was subjected to HLM, structural equation modeling, and individual growth curve analyses. Comparisons of the three methods showed only rather small, probably inconsequential, differences in the estimates of parameters of interest, indicating that the choice of method of analysis is largely a matter of preference. Each method did, however, result in unique information that could be useful depending on circumstances.
|
|
|
Individual Growth Curve Analysis of Longitudinal Data on Intervention for Back Pain
|
| Patricia Herman, University of Arizona, pherman@u.arizona.edu
|
|
Persons suffering from lower back pain were assigned randomly to an experimental intervention and treatment as usual. Effects of treatment were assessed by widely used measures of distress from back pain. Data were converted into growth curves and intercept and slope parameters were estimated. In addition, trajectories of change were identified on the basis of theoretical considerations regarding the likely courses of change, and fits to those models were estimated. Results indicated that the experimental intervention could be considered likely to be effective, with an asymptotic trajectory being the best fit to the data. Nonetheless, individual trajectories were varied, and the advantages of preserving change at the level of the individual case were obvious..
| |
|
Structural Equation Modeling of Longitudinal Data
|
| Lee Sechrest, University of Arizona, sechrest@u.arizona.edu
|
|
Longitudinal processes can be represented in structural equation models by employing one or more chronological variables to represent times of measurement. Depending on the number of measurement occasions, various theoretically interesting trajectories of change can be modeled, as can interaction s between time of measurement and other variables of interest. Illustrated by results from studies of naval recruits in training and participants in substance abuse treatment, it is shown that structural equation modeling of longitudinal data facilitates examination of specific pathways and their associated path coefficients and the identification of differential trajectories of change.
| |