Related Experiment Videos
Methods for comparing event rates in intervention studies when the unit of allocation is a cluster
1Department of Epidemiology and Biostatistics, University of Western Ontario, London, Canada.
American Journal of Epidemiology
|August 1, 1994
Summary
This study addresses statistical methods for clustered data in group-randomized trials. Choosing the right analysis depends on whether interventions were randomly assigned to clusters.
Area of Science:
- Biostatistics
- Public Health Research
- Epidemiology
Background:
- Many studies allocate intact social units (e.g., schools, communities) to intervention groups.
- Standard statistical methods fail with clustered data due to dependencies within units.
- This can lead to inaccurate study findings.
Purpose of the Study:
- To review statistical approaches for analyzing clustered data in group-randomized trials.
- To evaluate the strengths and weaknesses of different methods.
- To provide guidance on selecting appropriate statistical techniques.
Main Methods:
- Review of statistical methods for clustered data analysis.
- Application of methods to data from a school-based smoking cessation trial.
- Comparison of approaches based on study design features, particularly random allocation.
Main Results:
- Standard methods are inadequate for clustered designs.
- Different statistical approaches have varying strengths and weaknesses.
- The appropriateness of a method is influenced by the use of random allocation.
Conclusions:
- The selection of statistical methods for clustered data should consider the presence or absence of random allocation.
- Appropriate statistical analysis is crucial for valid interpretation of group-randomized trials.
- Guidance is provided for researchers working with clustered intervention designs.