ABSTRACT
In biological studies and in several other areas, experimental data are an alyzed using F and Student’s t tests of significance. These tests are based on certain basic assumptions: additivity (the treatments’ effects must be as sumed to be additive (sometimes additivity is referred to as linearity); nor mality (data are obtained randomly from a normal population); homogene ity of variances (equal variances, i.e., the variances are homoscedastic); and independence of errors (errors are noncorrelated). If the assumptions are not met, the basic probabilities may change and lead to faulty conclusions. When we test at P - 0.05 level of significance, we actually will be testing at 8 percent level, which will lead to too many significant results. The normal distribution is one important assumption in the analysis of variance (Cochran, 1947). The experimental data may exhibit violations of the as sumptions. In the ANOVA, if the treatment effects are not really additive, then the sums of squares attributable to such effects do not represent the true effects (Anderson and Bancroft, 1952). Nonadditivity may exist when the true effects are multiplicative in nature and when aberrant observations are present (Ostle, 1988; Zar, 1974). Additivity effects imply a model that does not contain interaction terms. Experimenters commonly use design models, e.g., completely randomized design, randomized complete block design, Latin square design, factorial experiments, etc., which are all based on the assumption of additivity. The use of transformation in order to obtain additivity has received more attention recently than in the past. Use of Tukey’s nonadditivity has been used as a guide for deciding between alter native possible transformation.
