nrPIT: Non-randomized Probability Integral Transform

In thomas-fung/mpcmp: Mean-Parametrized Conway-Maxwell Poisson (COM-Poisson) Regression

Non-randomized Probability Integral Transform

Description

Functions to produce the non-randomized probability integral transform (PIT) to check the adequacy of the distributional assumption of the COM-Poisson model. The majority of the code and descriptions are taken from Dunsmuir and Scott (2015).

Usage

compPredProb(object)

compPIT(object, bins = 10)

Arguments

object	an object class "cmp", obtained from a call to glm.cmp.
bins	numeric; the number of bins shown in the PIT histogram or the PIT Q-Q plot.

Details

These functions are used to obtain the predictive probabilities and the probability integral transform for a fitted COM-Poisson model. The majority of the code and descriptions are taken from Dunsmuir and Scott (2015).

Value

compPredprob returns a list with values:

upper	the predictive cumulative probabilities used as the upper bound for computing the non-randomized PIT.
lower	the predictive cumulative probabilities used as the upper bound for computing the non-randomized PIT.

compPIT returns a list with values:

conditionalPIT	the conditional probability integral transformation given the observed counts.
PIT	the probability integral transformation.

References

Czado, C., Gneiting, T. and Held, L. (2009). Predictive model assessment for count data. Biometrics, 65, 1254–1261.

Dunsmuir, W.T.M. and Scott, D.J. (2015). The glarma Package for Observation-Driven Time Series Regression of Counts. Journal of Statistical Software, 67, 1–36.

Examples

data(takeoverbids)
M.bids <- glm.cmp(numbids ~ leglrest + rearest + finrest + whtknght
  + bidprem + insthold + size + sizesq + regulatn, data = takeoverbids)
compPredProb(M.bids)
compPIT(M.bids)

thomas-fung/mpcmp documentation built on June 13, 2022, 6:20 p.m.