|
Abstract:
|
For data from cluster randomized trials or observational studies involving cluster sampling, the assumption of independence is likely to fail. Parametric model assumptions may not even be realistic when data is measured in non-metric scale as commonly happens for Quality of Life outcomes. Fully nonparametric methods provide flexible alternative in such situations. In particular, nonparametric methods can accommodate binary, ordered categorical, discrete and continuous data seamlessly. In this talk, nonparametric effect-size measures that allow meaningful and interpretable probabilistic comparison of treatments or intervention programs will be introduced and its properties will be discussed along with procedures for constructing confidence intervals and hypothesis tests. Theoretical as well as empirical properties of the procedures will be discussed. Methods for small sample approximation that retain some of the optimal asymptotic behaviors will be presented. Simulations show favorable performance of the methods for arbitrary settings of complete and incomplete case combinations. Data from Asthma Randomized Trial of Indoor wood Smoke (ARTIS) will be used for application.
|
