the fixed-effects model matrix is column rank deficient when adding a variable

Question

[Will]

I'm computing LIMMA on a RNAseq Data. When I use this formula:

~feature + age  + (1 | sex) + (1 | group) + (1 | id)

inside voomWithDreamWeights() I have no problems. When I add a variable to this formula I have the following error:

Error in .fitVarPartModel(exprObj, formula, data, REML = REML, useWeights = useWeights,  : 
  the fixed-effects model matrix is column rank deficient (rank(X) = 339 < 340 = p);
the fixed effects will be jointly unidentifiable 

Suggestion: rescale fixed effect variables.
This will not change the variance fractions or p-values

age and the variable that gives me this problem are continuous variable.

Answer 1

[Shab86]

So, adding another variable causes the error? If so, what's that variable?

[Will]

it is just a numeric continuous variable, i.e. the Intracranial volume

[Shab86]

Is it that the intracranial volume is in such units that they are inordinately larger compared to all the rest of your predictors?

[Will]

Yes the value is larger compared to the other variables

[Shab86]

I think that could be why you are indeed getting this error. Maybe you could scale the intracranial volume and then use it in limma-voom.