> sometimes you even want to use a more-wrong model because it fits your problem better.

Or they are equal in fit, but the more wrong is easier to use.