Reasons besides prediction to build models? Joshua Epstein wrote a paper titled "Why Model?" available at http://www.santafe.edu/media/workingpapers/08-09-040.pdf in which gives 16 reasons:


*

*Explain (very distinct from predict)

*Guide data collection

*Illuminate core dynamics

*Suggest dynamical analogies

*Discover new questions

*Promote a scientific habit of mind

*Bound (bracket) outcomes to plausible ranges

*Illuminate core uncertainties.

*Offer crisis options in near-real time

*Demonstrate tradeoffs / suggest efficiencies

*Challenge the robustness of prevailing theory through perturbations

*Expose prevailing wisdom as incompatible with available data

*Train practitioners

*Discipline the policy dialogue

*Educate the general public

*Reveal the apparently simple (complex) to be complex (simple)


(Epstein elaborates on many of the reasons in more detail in his paper.)
I would like to ask the community:


*

*are there are additional reasons that Epstein did not list?

*is there a more elegant way to conceptualize (a different grouping perhaps) these reasons?

*are any of Epstein's reasons flawed or incomplete?

*are their clearer elaborations of these reasons?

 A: 
Reason 17.  Write a paper.

Sort-of just kidding but not really.  There seems to be a bit of overlap between some of his points (eg 1, 5, 6, 12, 14).
A: 
Save money

I build mathematical/statistical of cellular mechanisms. For example, how a particular protein affects cellular ageing. The role of the model is mainly prediction, but also to save money. It's far cheaper to employ a single modeller than (say) a few wet-lab biologists with the associated equipment costs. Of course modelling doesn't fully replace the experiment, it just aids the process.
A: 
For fun!

I'm sure most statisticians/modellers do their job because they enjoy it. Getting paid to do something you enjoy is quite nice!
A: 
dimension reduction

Sometimes there can be too much data, so forming an initial model allows for further analysis. 
A: 
regulation

Government agencies require firms to provide reports using certain models. This provides for a degree of standardization in oversight. An example is the use of Value-at-Risk in the financial sector. 
A: 
Control

A major aspect of the dynamic modelling literature is associated with control. This kind of work spans a lot of disciplines from politics/economics (see, e.g. Stafford Beer), biology (see e.g. N Weiner's 1948 work on Cybernetics) through to contemporary state space control theory (see for an intro Ljung 1999). 
Control is kind of related to Epstein's 9 and 10, and Shane's answers about human judgement / regulation, but I thought it made sense to be explicit. Indeed, at the end of my engineering undergraduate career I would have given you a very concise response to the uses of modelling: control, inference and prediction. I guess inference, by which I mean filtering/smoothing/dimension-reduction etc, is maybe similar to Epstein's points 3 and 8. 
Of course in my later years I wouldn't be so bold as to limit the purposes of modelling to control, inference and prediction. Maybe a fourth, covering many of Epsteins's points, should be "coercion" - the only way you should "educate the public" is to encourage us to make our own models... 
A: This is closely related to some of the others, but:

Eliminate human judgement

Human decision making is subject to many different forces and biases. That means that you not only get different answers to the same question, but you can also end up with really suboptimal outcomes. Examples would be the over-confidence bias or anchoring.  
A: 
To take (useful) action.

I'm paraphrasing someone else here, but suppose we built a system of public health around the model that infectious diseases are due to malevolent spirits that spread through contact.  The science of microbes may be an infinitely better model, but you could prevent a good number of contagions nonetheless.  (I think this was on reading a history of cybernetics, but I can't remember who made the point.)
The point is that, along the lines of "all models bad, some useful", we need to formulate models and refine them in order to undertake any useful actions with lasting consequences.  Otherwise, we might as well flip coins.
A: 
Repetitive problems that involve some form of benefit / cost

In my field, we model the same set of variables in different locations, time frame, and magnitudes
A: In my opinion 16 are too many reasons, too fine of a specification and sort of overlap at times. Instead I would personally streamline into  broad groups. We can classify study objectives in 3 main categories: single hypothesis testing, exploratory study and to predict.
