Dear members, I have the same design as in section 8.2 Paired samples in Limma user guide (http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.182.22 71&rep=rep1&type=pdf). Treatment or control is apply on the same cell. The recommendation is to perform an paired t-test (equivalent to a fixed model) but in my mind, I will choose random effect and use duplicatecorrelation in lmfit. Results are totaly different. What are the main aspect to choose between fixed or random. thanks for any comments [[alternative HTML version deleted]]