Characteristics of quantitative data in toxicity rodents - Questioning theusefulness of Bartlett's test for homogeneity of variance to introduce a rank test.

We investigated the changes in sample numbers with time and calculated the F values for various quantitative data between control and each treated gro up from 10 long-term studies using 9,280 rodents by the so-called "decision tree" for computerized statistical analysis. We then examined the detectio n rate of significant differences using Bartlett's test, and compared the i ncidence of significant differences between each dose group and the control group obtained with and without Bartlett's test. As a results, the number of animals decreased in the second half of the long-term studies, and varia nces of body weight, food consumption, hematological tests and organ weight s were higher in the dose groups than in the controls. The detection rate o f significant differences by Barlett's test was 16-35% for the parameters e xamined. We conclude that, it seems practical to test the significant diffe rence among mean values by parametric procedures, even if the data are hete rogeneous in their variances among groups.