Default behavior for posterior summaries (mean over quantiles)

[nk027]

At the moment, most posterior summaries report quantiles (to convey uncertainty) and not the mean. Especially for methods lie coef or vcov this may not be ideal. Maybe we should change the default behavior?

Perhaps any of

• as is (median + sometimes a range),
• mean instead of median (+ sometimes a range),
• mean in addition to median (if there's a range), otherwise mean?

[nk027]

Currently we have this for:

function	outputs	options
irf, fevd	draws, quantiles	thinning, bands, no mean!
predict	draws, quantiles	thinning, bands, no mean!
plot for both	time series	uses outputs
coef, vcov, companion	summaries	mean, bands, no thinning
fitted, predict, resid	summaries	bands, mean

I'd say that we generally want a location and a scale for the posteriors. At the moment that is mostly the median + quantiles. I guess for the location we would generally want the mean? For the scale, we may want SD, HPDI, etc.

For the functions that means we may want to

• Add the mean or change median to mean?
• Allow more and potentially ~custom summaries, e.g. coda::HPDinterval?
• Allow single draws?

Maybe the default argument would look like type = c("mean+quantiles", "mean", "quantiles", "draw")?

[nk027]

Just to note -- this would change default behavior and would only be a feature for a major version update.