% \VignetteIndexEntry{Insolvency - (Quasi-)Poisson Model and Negative Binomial Model}
%\VignetteEngine{knitr::knitr}
%\VignetteEncoding{UTF-8}

\documentclass[a4paper]{article}

\title{Insolvency - (Quasi-)Poisson Model and Negative Binomial Model}

\begin{document}

\maketitle
First the insolvency data are loaded:
<<results='hide',eval=TRUE>>=
library(catdata)
data(insolvency)
attach(insolvency)
@
%
For the number of insolvent firms between 1994 and 1996 a Poisson model is fitted with time as predictor.
Time is considered as a number from 1 to 36, denoting the month from January 1994 to December 1996.
<<eval=TRUE>>=
ins1 <- glm(insolv ~ case + I(case^2), family=poisson(link=log), data=insolvency)
summary(ins1)
# plot(ins1)
@
Scatter-Plot of number of insolvent firms dependent of the month (1-36). With estimated curve of the log-linear model.
<<eval=TRUE>>=
plot(case, insolv)
points(ins1$fitted.values, type="l")
@
In many real-world datasets the variance of count-data is higher than predicted by the Poisson distribution.
So next a Poisson model with disperison parameter is fitted (Quasi-Poisson model).
<<eval=TRUE>>=
ins2 <- glm(insolv ~ case + I(case^2), family=quasipoisson, data=insolvency)
summary(ins2)
# plot(ins2)
@
An alternative to a quasi-poisson model is to use the negative binomial distribution.
<<eval=TRUE>>=
library(MASS)
ins3 <- glm.nb(insolv ~ case + I(case^2),data=insolvency)
summary(ins3)
@
Since counts are rather large in addition a normal distribution model is fitted.
<<eval=TRUE>>=
ins4 <- glm(insolv ~ case + I(case^2), family=gaussian(link=log), data=insolvency)
summary(ins4)
@
\end{document}