I think I asked this question on the list before but regarding one-way ANOVA and pairwise comparison. And I am no expert in multiple comparison either. In the following paper, there are two main effect group and time. If I remember correctly the authors argue that interaction term is most important (otherwise one-way ANOVA would suffice) followed by groups effect. Statistical tests for identifying differentially expressed genes in time-course microarray experiments. Park T., Yi S.G., Lee S., Lee S.Y., Yoo D.H., Ahn J.I., Lee Y.S. Bioinformatics 2003; 19(6):694-703 12691981 (Don't you just hate p-values ?) On Tue, 2004-07-27 at 21:38, Eric wrote: > Hi Andy, > > Thanks for the reply. My reasoning here is a little Byzantine so bear with me. > > If the significant results are relatively evenly distributed across the > main effects and the interaction (about the same number of genes found in > each), then using the omnibus test will not make much of a difference. > However, say one of the two main effects is much stronger than the other, > then I have a case where the overall test will pick up all of those changes > from the 'powerful' treatment (or most of them). Because of that, multiple > testing correction at the overall level will allow genes with larger > p-values from the second main effect through the filter compared to the > list of genes that would make it through a multiple testing correction > applied at the level of the second main effect. > > Contrast this with the case where multiple testing is applied separately to > each of the three outputs. Here the first main effect is relatively > unaffected, but the second main effect is nuked (if the second main effect > has no more genes than would be expected by chance). IMHO it doesn't matter > what the original question was, the two multiple testing corrections change > the list of genes and the experimental question does not address which of > these procedures should be used. It would be disingenuous to say "Well, > we're mainly interested in main effect 2 (the weak one), so we'll use the > overall correction and at least see a list of genes" or "We wanted to > disagree with previous work about main effect two's importance to research > so we used individual correction to show the world that main effect two is > not doing anything". Perhaps the proportion of genes assigned an > interaction significance could be used to gauge the dependence of the two > main effects; the more dependent they are, the more applicable the overall > testing correction. While the smaller the proportion of genes showing an > interaction term, the more appropriate independent correction for each main > effect would be. > > > At 03:05 PM 7/27/2004, you wrote: > >I am absolutely no expert in multiple comparison / multiple testing / gene > >expression data analysis, so take the following with appropriate dose of > >salt: > > > >It really depends on what you are looking to get out of the data. Just > >because you have multi-factor data with > 2 levels and thousands of > >responses, it doesn't automatically mean that the usual multiple comparison > >procedures are appropriate. You design the experiment to answer some > >specific questions (hopefully). How you analyze the data depends greatly on > >what those questions are, and (hopefully, therefore) how the experiment is > >designed. > > > >Best, > >Andy > > > > > From: Eric > > > > > > Hi, > > > > > > I apologize for this being off-topic- it's really a > > > statistical question > > > but I'd be interested in the community's input. If I run a > > > 'per gene' 2-way > > > ANOVA on single channel microarray data (i.e., each gene is tested > > > separately by 2-Way ANOVA), should I run multiple testing > > > correction for > > > each factor and interaction separately? Alternatively, should > > > I use an > > > overall (omnibus) F-test, correct that for multiple testing, > > > and treat the > > > main effects and interaction results as post-hoc to the overall test? > > > > > > Thanks, > > > -E > > > > > > Eric Blalock, PhD > > > Dept Pharmacology, UKMC > > > 859 323-8033 > > > > > > STATEMENT OF CONFIDENTIALITY\ \ The contents of this e-mail > > > ...{{dropped}} > > > > > > _______________________________________________ > > > Bioconductor mailing list > > > Bioconductor@stat.math.ethz.ch > > > https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor > > > > > > > > > > > >------------------------------------------------------------------- ----------- > >Notice: This e-mail message, together with any attachments, contains > >information of Merck & Co., Inc. (One Merck Drive, Whitehouse Station, New > >Jersey, USA 08889), and/or its affiliates (which may be known outside the > >United States as Merck Frosst, Merck Sharp & Dohme or MSD and in Japan, as > >Banyu) that may be confidential, proprietary copyrighted and/or legally > >privileged. It is intended solely for the use of the individual or entity > >named on this message. If you are not the intended recipient, and have > >received this message in error, please notify us immediately by reply > >e-mail and then delete it from your system. > >------------------------------------------------------------------- ----------- > > Eric Blalock, PhD > Dept Pharmacology, UKMC > 859 323-8033 > > STATEMENT OF CONFIDENTIALITY\ \ The contents of this e-mail ...{{dropped}} > > _______________________________________________ > Bioconductor mailing list > Bioconductor@stat.math.ethz.ch > https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor >