#### Abstract

Covariance matrix estimation arises in multivariate problems including multivariate normal sampling models and regression models where random effects are jointly modeled, e.g. random-intercept, random-slope models. A Bayesian analysis of these problems requires a prior on the covariance matrix. Here we compare an inverse Wishart, scaled inverse Wishart, hierarchical inverse Wishart, and a separation strategy as possible priors for the covariance matrix. We evaluate these priors through a simulation study and application to a real data set. Generally all priors work well with the exception of the inverse Wishart when the true variance is small relative to prior mean. In this case, the posterior for the variance is biased toward larger values and the correlation is biased toward zero. This bias persists even for large sample sizes and therefore caution should be used when using the inverse Wishart prior.

#### Creative Commons License

This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

#### Recommended Citation

Alvarez, Ignacio; Niemi, Jarad; and Simpson, Matt
(2014).
"BAYESIAN INFERENCE FOR A COVARIANCE MATRIX,"
*Conference on Applied Statistics in Agriculture*.
https://doi.org/10.4148/2475-7772.1004

BAYESIAN INFERENCE FOR A COVARIANCE MATRIX

Covariance matrix estimation arises in multivariate problems including multivariate normal sampling models and regression models where random effects are jointly modeled, e.g. random-intercept, random-slope models. A Bayesian analysis of these problems requires a prior on the covariance matrix. Here we compare an inverse Wishart, scaled inverse Wishart, hierarchical inverse Wishart, and a separation strategy as possible priors for the covariance matrix. We evaluate these priors through a simulation study and application to a real data set. Generally all priors work well with the exception of the inverse Wishart when the true variance is small relative to prior mean. In this case, the posterior for the variance is biased toward larger values and the correlation is biased toward zero. This bias persists even for large sample sizes and therefore caution should be used when using the inverse Wishart prior.