Best practice for bimodal type of phenotype

[Diana Carolina Lopez Arias]

Hello,

This is a discussion that follows up this question. I don't have a specific problem with Rqtl but it is more about how to proceed with bimodal phenotypes and discuss a recent paper using 2p for bimodal data. 

So it seems that if the residuals are normally distributed or close to that, no matter if the phenotype itself is bimodal, the normal model can be used, right ? But what they didn't say is : if the residuals are not normal, do we have to use "np" model instead ? 

Because if I understood well, Karl said that bimodal data shouldn't be transformed to maybe not loose the major loci effect, right?

I saw in this paper that they used "2p" model for their bimodal phenotypes (supplementary figure S2). It is very confusing because I thought that "2p" model was developed for spike-like distribution. What are your thoughts ?

Thank you so much for such a wonderful work ! 

Best regards,

Diana

[Karl Broman]

I’m not sure I understand the question, or what I might say beyond what I said in the previous response: “The question of what to do with non-normal phenotype distributions in QTL mapping is the same as the question of what to do with non-normally distributed outcomes in multiple linear regression. Generally linear regression is quite robust, and non-constant residual variance is a bigger problem than non-normality.”

karl