A Short Diversion on Models in Science
- In the previous exercise, we made a generative model, that allowed us to make mock data and compare it with real data until the datasets matched.
- A model is good if it is able to reproduce the data: we learn by adjusting our models in the light of new information.
- The model then provides form to our understanding of the physical object. All of your understanding of everything you know about is in terms of a model.
- Models are of central importance in science: without data we have no information, and without models we have no understanding.
- Sometimes results are described as being "model-independent"; the only model-independent quantities are the data, handed to us as constants, or summaries of those data. Any physical interpretation of those data will necessarily be model-dependent.
- What people sometimes mean by "model-independent" is that their model is not simply-parameterized, and/or does not involve strong assumptions. But beware: aspiring to model-independence can lead you to overlook the assumptions that are inevitably involved.
- In general, it's best to own your model - and test your assumptions.