% \VignetteIndexEntry{Exposure to Dust - Logistic Regression and Search for Outliers}
%\VignetteEngine{knitr::knitr}
%\VignetteEncoding{UTF-8}
\documentclass[a4paper]{article}

\title{Exposure to Dust - Logistic Regression and Search for Outliers}

\begin{document}

\maketitle
First of all, the dust data are loaded:
<<results='hide',eval=FALSE>>=
library(catdata)
data(dust)
attach(dust)
@
First, the subsample of non-smokers is considered. A main effect logit model yields the following results:
<<eval=FALSE>>=
dustlogitnon1=glm(bronch ~ dust+years, family=binomial, data=dust[(dust$smoke==0),])
summary(dustlogitnon1)
@
The same model as above is used without observation 1245 which can be regarded as an outlier:
<<eval=FALSE>>=
dustlogitnon2 <- glm(bronch ~ dust+years, family=binomial, 
                     data=dust[(dust$smoke==0)&(dust$dust<10),])
summary(dustlogitnon2)
@

The following calculations are based on the complete dataset. Therefore, main effect logit models are fitted for all observations and without
observation 1246, respectively:
<<eval=FALSE>>=
dustlogit1 <- glm(bronch ~ dust+years+smoke, family=binomial, data=dust)
summary(dustlogit1)
@

<<eval=FALSE>>=
dustlogit2 <- glm(bronch ~ dust+years+smoke, family=binomial, 
                  data=dust[(dust$dust<20),])
summary(dustlogit2)
@

\end{document}