Models should be mechanical

Unless there is a very strong reason not to do it, models should be "mechanical".

That is, the connections of states should follow a causal structure. This provides two benefits:

The purpose of the model is obvious from the structure The use of the model is flexible to perturbations Context: There are often disagreements with how breast cancer should be modelled given there are disagreements about the "natural history" approach. I think, regardless of this, assuming a natural history can be useful, because it allows people to understand how interventions would affect populations in a much more clear way than other models.

Another reason that models should be mechanical in nature, is because otherwise, it's too easy to accumulate evidence about their effects without thought to causal mechanisms. For instance, sugar sweetened beverage taxation may reduce BMI, Overweight, Diabetes, CVD Risk, Total Mortality, but these are not independent. These are obviously all related to weight. Having a mechanical model forces the developer to identify the causal pathway, and reduce the risk of double counting.