Non-Randomized Version of the PIT Histogram (for Count Data)
pit.Rd
See Czado et al. (2009).
Usage
pit(x, ...)
# Default S3 method
pit(x, pdistr, J = 10, relative = TRUE, ..., plot = list())
Arguments
- x
numeric vector representing the observed counts.
- pdistr
either a list of predictive cumulative distribution functions for the observations
x
, or (the name of) a single predictive CDF used for allx
(with potentially varying arguments...
). It is checked that the predictive CDF returns 0 atx=-1
. The name of its first argument can be different fromx
, e.g.,pdistr="pnbinom"
is possible.
Ifpdistr
is a single function and no additional...
arguments are supplied,pdistr
is assumed to be vectorized, i.e., it is simply called aspdistr(x)
andpdistr(x-1)
. Otherwise, the predictive CDF is called sequentially and does not need to be vectorized.- J
the number of bins of the histogram.
- relative
logical indicating if relative frequency or the density should be plotted. Due to a historical bug,
relative=TRUE
(the default) actually plots a density histogram whilerelative=FALSE
plots relative frequencies.- ...
ignored if
pdistr
is a list. Otherwise, such additional arguments are used in sequential calls ofpdistr
viamapply(pdistr, x, ...)
.- plot
a list of arguments for
plot.histogram
. Otherwise, no plot will be produced.
Value
an object of class "pit"
, which inherits from class
"histogram"
(see hist
).
It is returned invisibly if a plot is produced.
References
Czado, C., Gneiting, T. and Held, L. (2009): Predictive model assessment for count data. Biometrics, 65 (4), 1254-1261. doi:10.1111/j.1541-0420.2009.01191.x
Examples
## Simulation example of Czado et al. (2009, Section 2.4)
set.seed(100)
x <- rnbinom(200, mu = 5, size = 2)
pdistrs <- list("NB(5,0)" = function (x) ppois(x, lambda=5),
"NB(5,1/2)" = function (x) pnbinom(x, mu=5, size=2),
"NB(5,1)" = function (x) pnbinom(x, mu=5, size=1))
## Reproduce Figure 1
op <- par(mfrow = c(1,3))
for (i in seq_along(pdistrs)) {
pit(x, pdistr = pdistrs[[i]], J = 10,
plot = list(ylim = c(0,2.75), main = names(pdistrs)[i]))
box()
}
par(op)
## Alternative call using ... arguments for pdistr (less efficient)
stopifnot(identical(pit(x, "pnbinom", mu = 5, size = 2, plot = FALSE),
pit(x, pdistrs[[2]], plot = FALSE)))