ALTERNATIVE ANALYSES OF CROSSOVER DESIGNS WITH MORE THAN TWO PERIODS

Carla L. Goad, Dallas E. Johnson
1997 Conference on Applied Statistics in Agriculture  
A crossover experiment is a special form of a repeated measures experiment. An appropriate analysis of a repeated measures experiment depends on the form of the variancecovariance matrix of the repeated measures. Certain forms of this matrix yield valid analysis of variance F -tests while other forms invalidate these tests. In a crossover experiment where analysis of variance tests are invalid, two alternative tests of a linear contrast of the parameters are proposed. In addition to these
more » ... tion to these approximate t-tests, three alternative methods for testing for equal treatment effects and equal carryover effects are proposed. A simulation study is conducted to evaluate these proposed alternative test procedures for power. Confidence levels and confidence interval lengths are also examined for those procedures from which an estimate of the linear contrast can be made. each of these approximate t -statistics has a different measure of degrees of freedom associated with it, then what measure of denominator degrees of freedom should be associated with the proposed F-statistic above? One possible choice is to use the value that yields the most conservative test. That is, use U = min {Uj}. Another choice is to use U = max {Uj} J J A Mixed Models Approach A third approach that can be considered is a mixed models approach or a method of using estimated generalized least squares estimates to construct test statistics. To demonstrate, consider once again the sufficient statistics for a crossover experiment. One has
doi:10.4148/2475-7772.1303 fatcat:7ytr52tx3rhw5csoohptdf5mh4