ABSTRACT
Fitting a model to a set of data involves searching for parameter estimates that optimize the relationship between the observed data and the predictions of the model. The parameter estimates are taken to represent some aspect of the world in which we are interested. While it should be possible to žnd optimum parameter values in each situation, it remains the case that the data used are only a sample. A different sample of the same kind of data would very likely lead to at least slightly different parameter estimates; so the exact value of the estimated parameters is not really the issue. Rather, we want to know how repeatable such estimates are likely to be if we had the luxury of having multiple samples. That is, the parameters are just estimates, and we want to know how conždent we can be in those estimates. Some aspects of uncertainty, such as highly variable data leading to parameter estimates with wide conždence intervals, are relatively familiar, and more details will be given here. However, in modelling there are many potential sources of uncertainty, and we need to be aware of them all to avoid drawing stronger conclusions than our data and models should allow.
