I'm using R's ppcor package and from there spcor function to calculate the pairwise semi-partial (part) correlations. In my data there are 11 variables and 95 observations.

When calculating partial correlations I get a warning that "The inverse of variance-covariance matrix is calculated using Moore-Penrose generalized matrix inverse due to its determinant of zero" and some of the returned estimates are -Inf.

How should (or could) I interpret these values?

  • $\begingroup$ Btw, Gavin edited the title and the question. In the future, it might not be the best idea to fix typos from quoted error messages as those are something that people copy and paste to google when experiencing the same problem $\endgroup$ – Petri Nov 17 '17 at 15:35

If you have zero determinant on your covariance matrix, then I guess your data is unconditionally independent of at least one variable, or a linear combination thereof. The semi-partial correlations over that variable will naturally be zero. So you can probably get away with setting the returned estimates to zero. If you want a more principled approach, do a singular-value decomposition on the covariance matrix to identify that variable and eliminate it from your dataset.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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