tau_prior_scale: Global scale prior scale choice

In jrnold/bayz: Miscellaneous Bayesian functions

Description

Piironen and Vehtari (2017) suggest the following choice of the scale for the global hyperparameter in a regression shrinkage problem with a global-local scale mixture of normal prior distribution,

τ_0 = \frac{p}{D - p} \frac{σ}{√{n}}

where p is the expected number of non-zero parameters, sigma is the scale of the observation disturbances, D is the number of parameters, and n is the number of observations.

Usage

tau_prior_scale(p, D = 1, n = D, sigma = 1)

Arguments

p	Expected number of non-zero parameters
D	Number of parameters
n	Number of observations
sigma	Scale of the observation disturbances

Value

A numeric vector

References

Piironen, J., Vehtari, A. 2017. https://arxiv.org/abs/1610.05559.

jrnold/bayz documentation built on May 5, 2019, 5:52 p.m.