Using nearPD-corrected correlation matrices for EFA/CFA and genomic SEM

[Abhiram D B]

Hi team,

I am currently performing an exploratory factor analysis (EFA) followed by confirmatory factor analysis (CFA). During the exploratory stage, I obtained the following correlation matrix:

[1,] 1.0000000 0.7173847 0.8724855 0.3366344 0.9096827
[2,] 0.7173847 1.0000000 0.4370086 0.2603012 0.3647780
[3,] 0.8724855 0.4370086 1.0000000 0.3266520 0.2879775
[4,] 0.3366344 0.2603012 0.3266520 1.0000000 0.6120234
[5,] 0.9096827 0.3647780 0.2879775 0.6120234 1.0000000

When attempting to perform EFA, the analysis failed due to the correlation matrix not being positive semidefinite. After reviewing the literature, I found that a commonly suggested solution is to apply a near–positive definite correction using the nearPD() function from the Matrix R package.

I wanted to ask whether anyone here has experience using a nearPD-corrected correlation matrix for EFA/CFA and subsequently continuing with downstream analyses such as GenomicSEM (multivariate GWAS). Specifically, I am interested in understanding whether applying this correction is methodologically acceptable in this context and whether it is reasonable to proceed with factor modeling and genomicSEM based on the corrected matrix.

Any insights, experiences, or references would be greatly appreciated.

Best regards,

Abhiram