Looking at the data I can see via princomp that pc1 is mostly noise, pc2 and pc3 are much more relevant. What is the best way to remove the effects of pc1 from the dataset. Chris

Hi Chris, On Fri, Jan 10, 2014 at 12:59 PM, Fenton Christopher Graham <christopher.fenton at="" uit.no=""> wrote: > Looking at the data I can see via princomp that pc1 is mostly noise, pc2 and pc3 are much more relevant. > What is the best way to remove the effects of pc1 from the dataset. > If it's really, really crap, ;.-) : just subtract it: mod <- prcomp(data) new.data <- data - mod$x[,1] %*% t(mod$rotation[,1]) If there may be some interesting variance in PC1, e.g it may have some variance related to your targets/reponse values (if you have some), then look into the surrogate variable package (sva) and related material. Arnar