| Abstract: A study of behaviour investigates the relationship between experimentally controlled environmental variables and the likelihood of response (Skinner, 1950). Skinner proposed that response rate should be the canonical datum for indexing the likelihood of response. However, the validity of response rate does not dominate all other types of dependent variable, especially for experimental designs for which rates cannot be calculated or when derived variables are required. Regardless of the type of dependent variable, theories of behaviour attempt to link environmental variables to some behavioural metric. This constrains those theories to particular dimensions of behaviour. An alternative approach is to construct a theory that links environmental variables to a constellation of behavioural metrics, thereby explaining "behaviour" as opposed to a particular behavioural metric. Computational models of behaviour comprise a set of models that have this multivariate property. Because computational models of behaviour output punctuate responses, almost any arbitrary behavioural metric can be calculated. Hence, one may fit a computational model of behaviour to any vector of dependent variables. The present paper outlines the philosophy of the Computational Analysis of Behaviour and illustrates the efficacy of the philosophy with a demonstration of multi-objective optimisation with Catania's Operant Reserve (Catania, 2005). |