PCA error: The correlation matrix is singular. What to do?

[Kate Sheahan]

I am running a PCA. It all seems to go smoothly until I run "estat KMO" to see if the data is suitable for PCA. I get an error message "the correlation matrix is singular". I have found a few references on what this means but I cannot find any information on how to address this in stata. I was surprised to see this error because when I initially looked at the correlations between the components and the original variables, I was concerned because they seemed low. Are the results of the PCA invalid if the matrix is singular? Thank you in advance.

[Phil Bromiley]

I don't do PCA, but a singular correlation matrix does not simply demand high bi-variate correlations - the singularity can come from multiple variables .

You're more likely to get a helpful answer if you follow the suggestions in the FAQ - provide Stata code using code delimiters, Stata output, and sample data using dataex.