0

I am looking for a way to compare the results from two lmRob() functions statistically.

Here is an explanation what I am trying to do: My professor wants me to compare the results of two ancovas (one for each sex) in another anova with each other. This is because there seems to be a significant interaction between my predictor and the covariate sex when checking for homogeneity of regression slopes. But based on my assumptions for the ancovas, I need to use a robust model -> lmRob().

Would it be an option for me to check if confidence intervalls of the coefficients overlap for a variable between the two lmRob() models?

Emil
  • 21
  • 1
    Why not add an interaction term to the lmRob model? – Peter Flom Jan 20 '24 at 17:02
  • @PeterFlom

    When checking the assumptions for ANCOVA, I already saw that there is a significant interaction between X and sex. Also X is kategorial with5 levels and I am using dummies for it in the lmRob() model. So it would not make sense to add the interaction term to the model, would it?

    Here is my model:

    model <- robust::lmRob(Y ~ X_dummy1 + X_dummy2 + X_dummy3 + X_dummy4 + covariate1 + covariate2 + covariate3, data = df)

     – Emil Jan 20 '24 at 17:30
  • 1
    Why wouldn't it make sense? You can have an interaction with a categorical variable. – Peter Flom Jan 20 '24 at 23:51
  • @PeterFlom Well actually I am getting this error when trying with the interaction:

    Error in lmRob.fit.compute(x, y, x1.idx = x1.idx, nrep = nrep, robust.control = robust.control, : Singular matrix encountered in FORTRAN subroutine rlfastse

     – Emil Jan 22 '24 at 10:07

0 Answers0