GENERALIZED LINEAR MODEL APPROACH TO ANALYZING PROPORTIONAL DATA

Data such as proportions and counts often defy the standard statistical procedures that assume normality of residuals and homogeneity of variances. Such data are often transformed, analyzed using nonparametric tests or rely on the robustness of classical ANOVA. This study therefore presents a simpler and direct approach to analyzing proportional data (data obtained from count only) without subjecting to rigorous transformation. The methods used were Generalized Linear Model (GLMs) procedure using Genstat 15. Based on the results of this study, we recommend that the first step always be using the original, untransformed data to examine the residuals. Only if the residuals do not meet the assumptions, would one consider an arcsine transformation. If the arcsine transformation improved the residuals, then it would be a suitable method. However, based on the results of this study, that is not likely to be the case. More recently, it is being suggested that the logistic regression (GLMs) be used as an alternative to the arcsine transformation.