Related Experiment Videos
Partial likelihood analysis of within-unit variances in repeated measurement experiments
V M Chinchilli1, J D Esinhart, W G Miller
1Center for Biostatistics & Epidemiology, College of Medicine, Pennsylvania State University, Hershey 17033, USA.
Biometrics
|March 1, 1995
Summary
This study introduces a new statistical method for analyzing repeated measurements, focusing on within-unit variances. The approach uses a flexible random effects model for improved accuracy in comparing treatments or techniques.
Area of Science:
- Statistics
- Biostatistics
- Experimental Design
Background:
- Comparing variances of multiple treatments, products, or techniques is a common experimental objective.
- Repeated measurement designs are crucial when within-unit variances are of primary interest over between-unit variances.
Purpose of the Study:
- To propose a statistical methodology for analyzing repeated measurement designs with heterogeneous within-unit variances.
- To provide methods for both population-based and individual-based inference.
Main Methods:
- A random effects model is employed, allowing for heterogeneous within-unit variances.
- No distributional assumptions are made for random effects.
- Random errors are assumed to follow either a normal or multivariate t distribution.
- A partial likelihood analysis is proposed for inference.
Main Results:
- The methodology is demonstrated using a practical example involving serum cholesterol measurements.
- The approach accommodates complex variance structures in repeated measures data.
- The partial likelihood analysis provides robust inference for comparing treatments.
Conclusions:
- The proposed random effects model and partial likelihood analysis offer a flexible and powerful tool for analyzing repeated measurement data, particularly when within-unit variances are heterogeneous.
- This method enhances the ability to compare treatments or techniques by accurately accounting for within-unit variability.
- The study provides a valuable statistical framework for researchers in various fields utilizing repeated measures designs.