Creates a horseshoe prior (see Carvalho, Polson, and Scott (2010)). The
horseshoe is usually parametrized as
\(\theta_i \sim N(0, s^2 \tau^2 \lambda_i^2)\),
\(\lambda_i \sim \mathrm{Cauchy}^+(0, 1)\),
with \(s^2\) the variance of the error distribution. We use a single
parameter scale, which corresponds to \(s\tau\) and thus does
not depend on the error distribution.
horseshoe(scale)The scale parameter (must be a scalar).
An object of class horseshoe (a list with a single element
scale, described above).