Skip to contents

Leverage plot

Source: R/plot-mlnmr.R
plot_leverage.Rd

Ported from multinma::plot.nma_dic() with type = "leverage" (Phillippo et al. 2020). Each data point's leverage (its contribution to the effective number of parameters) is plotted against its signed square root residual deviance, with contours of constant DIC contribution. Points outside the DIC = 3 contour are usually the ones spoiling the fit.

Usage

plot_leverage(object, ..., dic_contours = 1:4)

Arguments

object

A cmlnmr() fit.

...

Unused.

dic_contours

Numeric vector of DIC contours to draw. Default 1:4.

Value

A ggplot object.

Details

Leverage is the pointwise pV penalty, half the posterior variance of the point's deviance, matching the penalty used by dic(). Because deviances covary across points, the pointwise leverages need not sum exactly to the model's total pV.

Leverage plots need a saturated model and so are not available for survival outcomes, where the censored contributions have no saturated reference. multinma declines them for the same reason.

See also

dic(), plot.cpaic_dic()

Examples

if (FALSE) {
plot_leverage(fit)
}