Author(s): Gilbert Biney, Gabriel Asare Okyere, Abukari Alhassan and William Ofosu Agyapong

Abstract: This paper seeks to find a Robust Adaptive Scheme for models with dependent error structure. The design considered for the paper is Repeated Measures Design. The defining characteristics of repeated measures data are the dependency and covariance structure. The objectives of repeated measures data analysis are to examine and compare response trends over time. The nine winsorised scores proposed by Hettmansperger are used because they are considered the most appropriate set of rank scores for hypothesis testing and accommodate a broad class of continuous distributions which are either symmetric or asymmetric with varying tailweights. The Adaptive Scheme which this paper seeks to find for models with dependent error structure is a two-step procedure in which a selector statistic is first used to examine and classify a given data based on measures of skewness and tailweight. Afterwards, a test statistic, independent of the selector statistic is chosen and a test conducted. A simulation study was conducted to compare the performance of the adaptive test and the traditional parametric test from different continuous distributions. Analysis of real data sets were as well performed to compare efficiency of the two tests. The findings favoured the adaptive test especially for data generated from nonnormal distributions. Our adaptive scheme proved robust and efficient over the parametric test when a data contains outliers. The paper considered four covariance structures namely; Compound Symmetry (CS), Unstructured (UN), First Order Autoregressive (AR (1)) and Autoregressive with Heterogeneous Variance (AHR (1)). On the basis of the values of Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), the best covariance structure is selected.

Pages: 42-53 | Views: 458 | Downloads: 15

Download Full Article: Click Here