ABSTRACT
The choice of model for a given process may be governed by knowledge of the physical properties of the process. In the absence of any pointers, some graphical exploratory data analysis is required to investigate such aspects as the behaviour of the empirical reliability or hazard functions, which may then suggest a plausible family of distributions. Whichever way is used to make the initial selection of a model, the fitting of that model must always be complemented by graphical and other means to assess how well the model fits the data. The named theoretical model, Weibull, exponential or some other well-defined form, can be thought of as specifying the model 'family'. This family will involve parameters which are usually unknown and require estimation from the available data. Censoring may arise for a variety of reasons. It is important to be aware of its existence and not to ignore any observation simply because a precise value is not known.
