Dear Bioconductor Community,
as I'm currently writting a report about my gene expression analysis on two affymetrix microarray datasets, regarding differential expression, i found an paper (http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0012336) mentioning the necessary assumptions and "possible requirements" in order to use different categories of statistical tests. In my case, i performed paired limma moderated t-test to check for gene expression alterations between cancer and control samples in each patient comprized the 2 above mentioned datasets, but as im "fresh in R" i have checked the normality of my data(boxplots,histograms,Q-Q plots after normalization), but i havent cheched for equal variance !! As i have read from the above paper limma is a homoscedastic test(thus makes the assumption for equal variances between the groups of interest) could i have violated my results regarding the false positive rate ? Or due to the paired nature of my analysis(and thus not generally two group comparison) this does not affect my study ?
Finally, if this pinpoint a great concern, how could i check in R and in both datasets prior of using limma the homoscedasticity of my data(test for unequal variances) regarding the two groups(cancer and control samples) ?
Any advice or consideration on this matter would be grateful !!
Dear Mr Ryan C.Thompson,
thank you for your suggestion. Besides the above implementation in limma which i will search, is there also another test to check for homo- or heteroscedasticity prior to limma testing ?
Formal tests of homoscedasticity are not a good idea, because such tests make assumptions of their own that are stronger than the original assumptions you are trying to check (see Box's quote about row boats and ocean liners). Instead, just look at the MDS plot.