I'm sure I completely understand your question, but its best for you to use ComBat and include your 4 covariates, and then your downstream analysis (I'd recommend Limma) including all 4 conditions but no need for batch after ComBat. On May 3, 2013, at 11:01 AM, Menezes, Luis (NIH/NIDDK) [F] wrote: > Dear Dr. Johnson, > > I'm using ComBat to merge two illumina datasets of similar sample characteristics, but done at different times. I'm looking 2 experimental conditions in 2 genotypes, so have a total of 4 conditions (80 samples, 2 batches). > > My question is: is it more appropriate for downstream (anova) analysis to perform batch correction using only batch as a covariate, or should I also use the 4 conditions as covariates. Perhaps not surprisingly, the 2 strategies results in very different between sample correlation ranks, and I don't know if there is any statistical basis for choosing between the two strategies. > > Thank you! > Luis.