influence_plot creates an influence plot for a
fitted lm object. The y-axis is either the
studentized (the default) or standardized residuals
versus the leverage values for each observation. The size
of the point associated with each observation is
proportional to the value of the Cook's distance (the
default) or the DFFITS statistic for the observation.
Details about the different types of residuals are
discussed in the get_residuals
function.
influence_plot(
model,
rtype = c("studentized", "standardized"),
criterion = c("cooks", "dffits"),
id_n = 3,
add_reference = TRUE,
alpha = 0.05,
size = c(1, 4.8),
...,
text_arglist = list(),
abline_arglist = list(),
extendrange_f = 0.08
)A fitted model object from the
lm function.
The residual type to plot on the y-axis. The
default is "studentized". The other option is
"standardized".
The criterion that decides the size of
the points. The default is "cooks". The other
option is "dffits".
The number of points to identify with labels
with respect to largest absolute criterion. The default
is 3 labels.
A logical value indicating whether a
reference line should be added. The default is
TRUE.
The default quantile used for the horizontal reference lines. The default is 0.05. See Details.
A numeric vector of length 2 that provides guidelines for the size of the points.
Additional arguments passed to the
plot function.
Additional arguments passed to the
text function, which is used to
display the points that are identified.
A named list specifying additional
arguments passed to the abline
function for the horizontal reference line added to the
plot.
Positive number(s) specifying the
fraction by which the range of the residuals should be
extended using the extendrange
function. If longer than one, f[1] is used on
the left and f[2] on the right.
Joshua French
The range of the criterion statistic is mapped to
cex_pt = size[2]^2 - size[1]^2 and then the size
of the points is sqrt(cex_pt).
If add_reference is TRUE, then horizontal
reference lines are added at the \(\alpha/2\) and
\(1-\alpha/2\) quantiles of a t distribution with
degrees of freedom given by
stats::df.residual(model).
If add_reference is TRUE, then vertical
reference lines are added at \(2p/n\) and \(0.5\)
where \(p=\)length(stats::coef(model)) and
\(n=\)stats::nobs(model).
The vertical position of the reference lines can be
customized by setting the h argument of
abline_arglist. The horizontal position of the
reference lines can be customized by setting the v
argument of abline_arglist.
lmod <- lm(murder ~ hs_grad + urban + poverty + single,
data = crime2009)
# studentized residuals vs leverage
influence_plot(lmod, id_n = 3)
# standardized residuals vs leverage
influence_plot(lmod, rtype = "stan")
# similar plot from plot.lm
plot(lmod, which = 5)
Run the code above in your browser using DataLab