Hi, Thanks for the quick reply, yes the replicate spots would add extra error strata, I was trying to simplfy, and should have clarified. I was referring to time as the repeated scannings. Any suggestions about what I might use, before I dive into lme4 and write something home grown? Thanks for any suggestions, the question of interest is just paired comparisons between treatments, so I would be happy just doing moderated t statistics if the efficiency is ok and the bias not too great. Nicholas On Fri, 20 Aug 2004 18:57:57 +1000, "Gordon Smyth" <smyth@wehi.edu.au> said: > At 06:10 PM 20/08/2004, Nicholas Lewin-Koh wrote: > >Hi, > >first the design: > >I have 12 hand spotted arrays in 3 blocks with 4 treatments in each > >block. > >Each array is scanned at 2days, 3days and 4days exposure (this is > >phospholuminescince). > >Each probe is replicated twice on the array in neighboring spots, so the > >array > >rows look like > > > > ** ** ** ...... ** > > ** ** ** ...... > > : > > : > > : > > > >If this were a univariate response eg, one gene, I would probably just > >use a split plot, something like > > > >Block > >treatment (Block) > >Block * Treatment (Error1) > >Time > >Time*Treatment > >Time*Block + Time*Block*Treatment (Error 2) > > > >and just concentrate on the treatment effect before > >modelling the covariance and mixed effects, and hope it > >is a reasonable approximation. > > > >Is it possible to do something like this in limma? How do I force > >it to get the correct error, or is this a bad idea? > > Don't the duplicate probes and the repeated scannings of the same array > also introduce error strata? Anyway, this is too complicated for limma. > > Gordon > > >Thanks > >Nicholas >