Dear Thomas, You could have created the same plot more simply by gof(fit, plot=TRUE) Your plot looks as if the dispersion estimates entered to glmFit were systematically too big. This could arise, for example, if your counts are less dispersed than Poisson, because edgeR can't estimate dispersions to be less than zero. This of course would suggest a problem with the source of your data. With more usual data, it is very hard to imagine how correct use of edgeR could lead to a gof plot like you attach. Best wishes Gordon On Tue, 9 Oct 2012, Thomas Frederick Willems wrote: > Recently, I've run into another issue and I was curious whether this > might have to do with poor dispersion estimates. In particular, I > created a Q-Q plot using the following commands, > > g <- gof(fit) > z <- zscoreGamma(g$gof.statistics,shape=g$df/2,scale=2) > qqnorm(z) > abline(0, 1) > > In this case, fit corresponds to the output of glmFit run using the > tagWise dispersion estimates. In the Q-Q plot (which I've attached), the > sample quantile lies systematically below the theoretical quantile. Is > this possibly due to a poor dispersion estimate, or if not, what else > could it arise from? > > Thanks as always for the help > > Thomas ______________________________________________________________________ The information in this email is confidential and intend...{{dropped:4}}