Errors associated with simple versus realistic models

This paper addresses the relative errors associated with simple versus realistic (or science-based) models. We take the perspective of trying to predict what the model will predict as we begin to build the model. Any model building process can get the model “wrong” to a greater or lesser extent by making a theoretical mistake in constructing the model. In addition, every model needs data of some sort, whether it be obtained by experiments, surveys or expert judgment, and the data collection process is filled with error sources. This paper suggests a hypothesis that

1.	simple models have a larger variance in their predication of a result than do more realistic models (something most people intuitively agree to), and
2.	more realistic models still have a significant probability of an error because the errors in the model building process will result in a probability distribution that ought to be bimodal, trimodal, or higher multimodal.

The paper provides evidence to support these statements and draws conclusions about what types of models to generate and when.

Dennis BuedeEmail: