Abstract:
|
Community-based interventions can reduce the risk, mortality, and morbidity of diseases such as coronary heart disease and cancer. We consider several studies in diverse populations and discuss improved statistical methods for their analysis using quasi-least squares (QLS). QLS is a method based on generalised estimating equations (GEE) that differs from GEE with regard to estimation of the correlation parameters. We demonstrate that this difference can result in improved efficiency that may result in sample-size reductions and increased precision in estimation of an intervention's effect. We also implement some correlation structures not easily applied using GEE and explain how this improves our understanding of our motivational examples. In particular, we discuss our approach that implements QLS for analysis of data with two levels of correlation (Shults and Morrow, Biometrics, under review). We extend this approach for multi-level correlated data, demonstrating that incorrectly ignoring levels of association can result in substantial losses in efficiency.
|