Plot a model-neutral influence diagnostic

Usage

# S3 method for class 'influ_diag'
plot(
  x,
  type = c("influence", "index", "cdi", "components"),
  term = NULL,
  component = NULL,
  scale = NULL,
  coefficient_reference = "centred",
  coefficient_scale = "auto",
  coefficient_order = "original",
  ...
)

# S3 method for class 'influ_diag'
autoplot(object, ...)

Arguments

x: An influ_diag object.
type: One of "influence", "index", "cdi", or "components".
term: Optional term selection.
component: Optional component selection.
scale: Optional influence scale. By default the natural response contrast is plotted rather than the link-scale contrast.
coefficient_reference: For CDI plots, "centred" (default) subtracts the term's mean over the same weighted reference distribution used for influence. "model" displays the original model-coded contribution on the link scale, including its reference factor level.
coefficient_scale: For CDI plots, "auto" (default) displays centred log-response effects as ratios on a logarithmic axis. Other links remain in their labelled link units. "link" displays centred link effects for every model. The "model" reference always uses link units.
coefficient_order: For CDI plots, "original" (default) retains the diagnostic's level order. "ascending" and "descending" order categories by their displayed coefficient or relative-effect point estimate. Ties retain their original order. Both coefficient and distribution panels use the same ordering; annual influence values and the input object are unchanged. Use this for nominal categories such as vessels. The default is preferable when horizontal order has meaning, such as months or a continuous covariate.
...: Reserved for future plotting options.
object: An influ_diag object passed to autoplot().

Value

A ggplot or patchwork object.

Details

CDI intervals use the probabilities supplied to influ() (95% by default). Centring propagates the joint coefficient covariance or is performed within each posterior/simulation draw. Ratio summaries are calculated after transforming those draws. A CDI plot displays one component at a time; select component when a term occurs in several model components. Zero-probability components retain their fitted link orientation and are explicitly labelled as such. Ratio-scale CDI panels use the y-axis label "Relative Effect"; the term name remains on the horizontal axis rather than lengthening the y label. Logit-scale panels use "Effect (log-odds)" (or "Effect (log-odds of zero)" for zero-probability components). This shorter label does not change the selected centring or component orientation. Short term labels (including months) are horizontal on the upper fitted- effect axis and the lower composition axis, for fixed and random effects. At drawing time, the paired axes measure the available width and their label text. All labels are shown when they fit; otherwise every second, third, or subsequent level is labelled, starting at the first level. The last level is not forced onto an off-stride position. The same regular spacing is used above and below, retaining every coefficient, composition column, and tick. Wider output devices can show more labels. Longer labels are angled and justified for their respective top or bottom axis. The influence panel's focus labels are on the right, with the same level ordering as the composition. The proportion legend has one column and at most four reference bubbles.

Plot a model-neutral influence diagnostic

Usage

Arguments

Value

Details

See also