Dennis Meadows, in an email discussing computer models, suggested that beyond their obvious functions, computer models often have one or more of the following purposes:

#1: Provide useful information about the future behavior or the future coefficient values of some system. #2: Attract money that is mainly going to be used for purposes other than building a model - overhead, salaries, proposal writing. #3: Cause the model builder to become respected as an expert by others, so they will ask his or her advice. This often involves publishing the model in some respected journal. #4: Provide a disciplined learning environment within which the model builder actually does become an expert. #5: Generate results and computer output that can be used to justify or illustrate ideas and recommendations that were already existing. #6: Help a group of people learn to work together more effectively - trust, respect, and related variables. #7: Help a group of people learn a vocabulary or paradigm they can use in dealing with some shared problem. ... Note that #3 and #4 are quite independent of each other. People can gain expertise without respect and respect without expertise.

I found it worth pondering for a while.