Hi, I have some experiments where different nutrients/fertilisers (including but not limited to N) were applied to plants.

Unsurprisingly the data produced is bimodal. The populations are N-applied and N-not-applied.

Is it more robust to separate the populations and do an ANOVA on them separately or to transform the data using something like log/square-root/Boxcox/GMM? If transformation is the answer what kind of transformation is most appropriate?

Thanks very much.