Loss functions are one of the most interesting parts of statistics. They directly connect inference and the domain the problem is in. One thing not mentioned is that the loss function is another degree of freedom in your overall model. This is a good thing, as we saw in this chapter; loss functions can be used very effectively, but can be a bad thing, too. An extreme case is that a practitioner can change his or her loss function if the results do not fit the desired result. For this reason, it’s best to set the loss function as soon as possible in the analysis, and have its derivation open and logical.