Abstract:
This article considers the analysis of experiment of crossover design with repeated measures that is used frequently in medical and agricultural research. We investigate a common problem that is the Newton-Raphson algorithm, used to optimize restricted maximum likelihood (REML) function, frequently failing to converge when the two possible covariance structures are fitted with MIXED procedure of the SAS System in the analysis of the crossover design experiments with repeated measures. When REML method fails to converge, the analysis of the crossover design experiments with repeated measures stop from progressing. We propose alternative multivariate method of covariance estimation with two approaches, which can be used to complete the analysis of the crossover design experiments with repeated measure, when REML method fails to converge. The simulation results show that overall, the Kenward-Roger method provided the best control of the nominal Type I error rate for all the approaches and was not inferior in terms of power. The main result of our article is that the two proposed approaches are competitive in term of power when all approaches yield satisfactory Type I error control with the Kenward-Roger method.
Page(s):
47-63
DOI:
DOI not available
Published:
Journal: Pakistan Journal of Statistics, Volume: 23, Issue: 1, Year: 2007