Thanks Marc, as you said, this is a priori design. I'm trying to get more events. But for now, I'll have to try to get rid of the collinear genes. Is there any other method than using cor? Any method to state which genes behave in the same manner? Paul
Marc Schwartz escribió: > On Jun 15, 2009, at 5:54 AM, Paul Christoph Schröder wrote: > >> Hi all! >> Maybe someone could help me with the following. I know this hasn't >> directly to do with ecology but I'm also using glm. >> >> I have a list of 16 genes and 10 samples. The samples are of two >> types, 4 Ctrl and 6 Diseased. If, >> >> labelInd<-as.factor(c(rep("0",4),rep("1",6))) >> genes.glm<-glm(labelInd ~ ., family=binomial, data=mat) >> >> >> beeing "mat" the 10x16 matrix (without NAs), I got 17 values, first >> the intercept, 9 numerical values and "NA" for the last 7 genes. Does >> anybody you know why this is happening or how I can model using the >> 16 genes? >> >> I hope anyone could help me with this! >> Many thanks in advance, >> >> Paul > > More than likely, the 7 genes for which you are getting NA's are > collinear to other genes. Thus you get NA's. If you switched the order > of the 7 genes for which you are getting NAs so that they come first > in the formula, you would get NAs for others. > > If you use: > > summary(genes.glm) > > you will likely see a warning message about singularities in the > coefficient table header line. Something like: > > Coefficients: (7 not defined because of singularities) > > I would use cor(mat) to take a look at the correlation matrix for your > data so that you can review this in more detail. > > BTW, with only 10 observations, you are significantly overfitting the > model by using so many covariates. You typically need at least 10 to > 20 'events' for each covariate degree of freedom in a logistic > regression model. With only 6 diseased (events) you really don't even > have enough data to support one covariate. The study, presuming an 'a > priori' design, is way underpowered for what you are attempting to do. > > HTH, > > Marc Schwartz > > [[alternative HTML version deleted]]
______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.