Covariance prior via lkj_corr_cholesky

[jsocolar]

Welcome to the forums! The default construction of covariance matrices in Stan can lead with high probability to numerical issues at initialization when the matrices are large. Fortunately, there are good ways to construct/initialize these matrices that avoid problems. This post and the links therein are a good place to start:

Using the onion method in occupancy model hierarchical model

I am trying to build a hierarchical, two-level occupancy model. The model estimates the correlation between species detections using a \Sigma matrix, similar to the model in SUG 1.13. The model works works until I scale up to greater than ~5 to 10 covariates. This post ( -nan results with lkj_corr_cholesky - Developers - The Stan Forums (mc-stan.org)) describes why. Looking through the answers, I see an example onion method file, in -nan results with lkj_corr_cholesky - #3 by bgoodri. Howeve…