Fully Nonparametric Methods for Partially Complete Data in Repeated Measures Design
No Thumbnail Available
Date
2018-07-04
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
In this dissertation, two related but distinct problems are studied. The first one is a fully nonparametric
rank-based method for comparing samples with partially paired data. Partially-paired
(correlated) data naturally arise, for example, as a result of missing values, in incomplete block
designs or meta analysis. In the nonparametric setup, treatment effects are characterized in terms
of functionals of distribution functions and the only assumption needed is that the marginal distributions
to be non-degenerate. The setup accommodates binary, ordered categorical, discrete
and continuous data in a seamless fashion. The use of nonparametric effects also addresses the
Behrens-Fisher problem from the nonparametric point of view and allows construction of confidence
intervals. Although, the nonparametric methods are mainly asymptotic, methods for small
sample approximations are also proposed. The second problem studied is also a fully nonparametric
rank-based method but for partially repeated measures data. Here a vector of nonparametric
relative effect measures are defined and linear hypotheses on these effects are considered.
A multitude of tests are available for hypothesis related to a vector of relative effects. We focus
on asymptotic results and finite sample performance for Wald-type statistic (WTS), ANOVA-type
statistic (ATS) and Multiple Comparison Test Procedure (MCTP). Notwithstanding the limitation
that the theory is thoroughly investigated for the three time point case, the results can formally be
extended to the more general set up but the involved expressions will be much more complicated.
The finite sample behavior of the tests are investigated via simulation studies. The results provide
numerical evidence of favorable performance of the nonparametric method. The new methods
vi
have overwhelming power advantage when treatment effects are reflected in the shape of the distribution
while they perform comparably better with parametric methods for location-type alternatives.
Data from a therapeutic-drug clinical trial and a randomized controlled epidemiological
study are used to illustrate the application of the methods.
Description
Keywords
Nonparametric Methods, Partially Complete Data, Repeated Measures Design