Thsting for Heterogeneity of Experimental Error for Combined Analysis

Abstract

For experiments conducted at several sites or repeated several seasons, the purpose in to obtain information of the interaction of environment and treatment to increase the scope of inference. Before using a combined analysis over sites or seasons the homogeneity of error variance must be examined. In practice, the error variances can be considered homogeneous if the highest error mean square is not three-larger than the smallest error mean square where it is refrred to as the three-fold procedure. However, the result that obtained by this procedure often differ from that obtained by the Bartlett's test. Therefore, a Cochran's procedure, the results of testing the homogeneity of error variance that obtained between the two procedure are compared considering the Bartlett's test as the standard procedure. The 102 sets of data with 3 more than 3 sites from the rice research project of rice variety trials for irrigated area, 1986-93 were used. The results indicated that the Cochran's test gave 87% of the time for getting the same results as Bartett's test, while the three-fold procedure gave 51% of the time. It was threfore recommended that the Cochran's test was more efficient in testing the homegeneity of error variance than the three-fold procedure.