It depends on the experimental design, like which array type you are using, but say for microarray translating pixel intensities into transcript expression requires a series of computations, generically known as preprocessing and normalization steps. Correcting for batch effects is when you have large differences in error estimation than expected between two or more identical samples, or replicate microarrays in this case, so you need to add this correction categorically in your phenotypic data when re-normalizing the data; unless you know ahead of time that you do have or will have a batch effect. I’m sure someone else on here can explain it better than I can, but you don’t divulge much in your question/post to give a more clear answer.