Do we need to examine the quantitative data obtained from toxicity studies for both normality and homogeneity of variance?

ABSTRACT

Most of the statistical techniques used to evaluate the data obtained from toxicity studies are based on the assumption that the data show a normal distribution and homogeneity of variance. Literature review on toxicity studies on laboratory animals reveals that in most of the cases homogeneity of variance alone is examined for the data obtained from these studies. But the data that show homogeneity of variance need not always show a normal distribution. In fact, most of the data derived from toxicity studies, including hematological and biochemical parameters show a non-normal distribution. On examining normality of data obtained from various toxicity studies using different normality tests, we observed that Shapiro-Wilk test is more appropriate than Kolmogorov-Smimov test, Lilliefors test, the normal probability paper analysis and Chi square test. But there are situations, especially in the long-term toxicity studies, where normality is not shown by one or more than one of the dosage groups. In this situation, we propose that the data maybe analyzed using Dunnett multiple comparison test after excluding the data of the groups that do not show normality However, the biological relevance of the excluded data has to be carefully scrutinized. We also observed that the tendency of the data to show a normal distribution seems to be related to the age of the animals. Present paper describes various tests commonly used to test normality and their power, and also emphasizes the need of subjecting the data obtained from toxicity studies to both normality and homogeneity tests. A flow chart suggesting the statistical techniques that maybe used for both the types of data showing a normal or non-normal distribution is also proposed.