I have got a data matrix generated on some plant species as proportions from certain quantitative measurements. Since such data (proportions, percentages, probabilities) are generally known to be skewed with unequal variances, I intend to transform mine before ANOVA. I understand that the arcsine or logit transform can be the best for such data. However I am constrained by the fact that the proportions obtained by me do not only lie between 0 and 1. Many of them include values above 1 (i.e. percent increase, giving values greater than 100%). How best (with reasons) can the data be transformed prior to the proposed analysis?