ABSTRACT
This chapter looks at four models: additive models for survival data, matched case-control data, ordinal response data and time series. The first three models are special because they do not fit into the univariate exponential family framework. The fourth model is really a technique for decomposing a time series into a sum of seasonal, trend and remainder components. The chapter describes a new technique for decomposing a time series into three components: seasonal, trend and remainder. The matched case-control and proportional-hazards models for censored survival data are similar in flavour since they both use conditioning arguments for estimation. The linear logistic regression model is often used by epidemiologists and biostatisticians for the analysis of matched case-control data. Brasher studied a number of definitions of residuals, including one derived from the penalized partial-likelihood approach.
