I've read a few studies on merging microarray datasets (corresponding to specific series) from GEO. Some of them RMA normalize the series data (es. GSExxx) separately and then, once merged, renormalize the whole dataset. Some of them normalize the whole dataset they intend to build basically RMA analyzing all the CEL files from multiple series (es. GSExx1, GSExx2, GSExx3, etc..).
What's the correct approach according to your opinion and, if you can, please cite some useful papers, it would be a great help,
thanks in advance!