Prior versus posterior

Ported from multinma::plot_prior_posterior() (Phillippo et al. 2020). Posteriors are drawn as histograms, priors as lines. Where a posterior simply reproduces its prior, the data carry no information about that parameter, and any quantity that leans on it is prior-driven rather than estimated. This is the visual counterpart of prior_sensitivity().

Usage

plot_prior_posterior(x, ..., prior = NULL, bins = 40)

Arguments

x

A cmlnmr() fit.

...

Unused.

prior

Which priors to show. Any of "intercept" (mu), "beta" (component effects), "regression" (breg), "gamma" (component by effect-modifier interactions), and "tau" (heterogeneity, random effects only). Defaults to all that the model used.

bins

Number of histogram bins for the posterior. Default 40.

Value

A ggplot object.

Details

It matters most for the component by effect-modifier interactions gamma: interactions informed only by aggregate arms are weakly identified, and prior_gamma_scale then does real regularization.

See also

prior_sensitivity(), prior_predictive_check()

Examples

if (FALSE) {
plot_prior_posterior(fit, prior = "gamma")
}