Why do people generalize EU instead of making an entirely new model, or create a model that is neither a special case nor an extension of EU?
To my knowledge, most utility functions under risk and uncertainty are generalization of expected utility (EU) or expected value (EV). Examples are regret theory, prospect theory, and all the models for Ellsberg paradox. The list goes on.
I am guessing the reasons:
EU is historically the classic model so everyone follows it. But by doing research people should try to find the better model, not the model that is closely link to the classic model.
EU performs pretty good experimentally and empirically. By totally abandoning EU some goodies might be missed. By relaxing EU those goodies can be kept.
If a model is a generalization of EU, then the model can be empirically compared with EU easily with likelihood ratio tests. But even if a model is not a generalization of EU, statistical tests are also available.
EU is well-studied and a lot of tools are ready. Relaxing EU is technically easier than creating an entirely new model.
But these still do not explain why no one (to my knowledge) make a model that is not a generalization of EU or EV. There must be something deadly wrong if the model is not a generalization of EU or EV.
Edit: I understand that EU is a good model. I am looking for some conceptual rationale that generalization of EU is good while non-generalization of EU or EV is not good.