Why Are Models Important for Testing Hypotheses?
So why go to the trouble of creating realistic models of brain processes? Basically, because it allows one to compare different models or hypotheses about distributed neuronal computations. In DCM one fits or inverts the models by optimising (the distribution of) their parameters (i.e., connection strengths and other biophysical quantities like rate constants) with respect to the model’s evidence. Put simply, one finds the distribution of parameters that renders the data the most likely, under the DCM considered. This optimisation furnishes two things. It provides the most likely parameters for any given model, and the model evidence itself. This evidence is simply the probability of observing the data under a particular model. The model evidence is a very important quantity because it allows one to compare different models and adjudicate among them . In other words, it allows one to explore model space and find the best model that explains the data in a parsimonious way. If one eq