I am trying to analize my experiment results according to the multgee reference pdf, but I have some warning message.

My dependent variable is a 5 level categorical variable. I have 12 eyes of 6 rat, examined 4 time daily for 3 days. My ID is a numerical variable (1 -> 12), different for each rat and each eyes between rats. The grouping variable is a 2 level categorical variable (Drug). I want to know if there are any significant differences in terms of the 5 levels of the dependent variable between subject and within subject (repeated measure) among drugs.

here a str() report of my data:

 Classes ‘tbl_df’, ‘tbl’ and 'data.frame':  264 obs. of  19 variables:
 $ Eye                            : chr  "R.E" "L.E" "R.E" "L.E" ...
 $ Drug                           : chr  "Study D" "None" "Study D" "None" 
 $ Timepoint                      : num  0 0 1 1 2 2 3 3 4 4 ...
 $ Corneal_Defect                 : Factor w/ 5 levels "No staining",..: 1 1 3 2 3 2 3 2 3 1 ...
 $ newid                          : num  1 2 1 2 1 2 1 2 1 2 ...

My code, according to reference manual is:

intrinsic.pars( y = Controlli$Corneal_Defect, data = Controlli, id = newid, repeated = Timepoint,rscale = "ordinal")

and the results (that require more than 30 minutes), with the first warning are:

 [1] -1.485344e+00 -6.996468e-01 -1.179613e+01 -1.179301e+01
  [5] -6.428373e+00 -1.365166e+01 -4.638385e-01 -2.289868e-01
  [9] -9.975028e-02 -7.761781e-01 -6.750473e-01 -1.107365e+00
[229] -2.537813e-01 -2.537813e-01  2.197225e+00
Warning message:
glm.fit: fitted rates numerically 0 occurred 

When i try to fit the model fit <- ordLORgee(formula = Corneal_Defect ~ factor(Timepoint) + factor(Drug),link = "logit", id = newid, repeated = Timepoint, data = Controlli,LORstr = "uniform")

I have this error:

Error in fitLORgee(Y, X_mat, coeffs, ncategories, id, repeated, offset,  : 
  Robust covariance matrix is not positive definite
In addition: Warning message:
In eval(slot(family, "initialize")) :
  response should be ordinal---see ordered()

Can someone please help me?? Moreover, if there are other packages of statistical tools to perform the analysis I need please, can you suggest me?


1 Answer 1


Regarding the second problem: try 'ordered(Corneal_Defect)' in the model formula. If the covariance structure is still no positive definite then, try a different (easier) structure.

Regarding the first problem: save the results of intrinsic.pars and have a look at them via the summary() function.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.