Skip to contents

A normal density on the log of the coefficient. Under fit_control(bias_adjust_proc = TRUE), the default, it is centred at meanlog - sdlog^2/2, so exp(meanlog) is the prior mean; with FALSE it matches stats::dlnorm() and exp(meanlog) is the prior median.

Usage

prior_lognormal(meanlog, sdlog)

Arguments

meanlog

log of the prior mean (of the median when bias_adjust_proc = FALSE).

sdlog

prior standard deviation of the log (must be positive).

Value

An Rceattle_prior of family "lognormal".