amfp
New Member
Posts: 1
|
Post by amfp on Aug 11, 2014 14:44:42 GMT
I just want to clarify one thing with regards to the --covar function for sex, and also for the --gxe function.
For the the --covar function: I understand that GCTA reads "M" and "F" characters in the input file for the --covar command and constrains one sex to zero and calculates the effect of the other sex in relation to the one it set to zero. How do I know which one it set to zero? Do I assume that it set M to zero? Also, if I wanted to get some kind of output for sex and age, could I?
For the --gxe function: I understand that GCTA includes the main effect in as the fixed effect in the model (so the mean difference between males and females) at the same time that it models the interaction between the covariate and the genotype, so I have taken this to mean that I should NOT also put in this same covariate into the --covar command. Is this correct? Or would there be no conflict in the model if I were to model sex as a covariate with --covar and as an interaction effect with --gxe (knowing that the interaction is a higher-order effect that that one must model the main effect before modeling the interaction).
Just want to double-check my modeling assumptions.
|
|
|
Post by Jian Yang on Aug 16, 2014 12:06:26 GMT
Re 1) This is similar to an ANOVA analysis, where male and female are modelled as a factor with 2 levels. The effects are just the means of the phenotype in males and females (no overall mean fitted here because otherwise it will cause collinearity issue). You can output these effect by adding --reml-est-fix in the REML analysis.
Re 2) You are right that --gxe and --covar do the same thing for the fixed effect components.
|
|