One of the things we based rates on was the breeds of dog people owned. Pitbulls and certain other dog breeds do not just have a bad reputation because people irrationally fear/hate them, they actually do pose a greater risk.

This is a classic example of someone observing a statistical correlation between specific factors and using that to assert a direct causal relationship between them. It implies that an insurance agency is able to 1) accurately identify every single breed of dog in every single insurance related incident (which is definitely not the case, because I doubt every insurance company is doing genetic testing on every single dog it comes across) and 2) tie a causal relationship between dog breed and incident. If I were going by typical insurance metrics, and to borrow from your analogy of "teenagers as unsafe drivers," you would also assume that red Camarros, something more expensive to insure than your more conservative sedan, were statistically more dangerous than, say, a white Civic, as if they were what caused their drivers to get into car accidents, as opposed to young, reckless people interested in a fast sports car to simply go out and buy one. These are people who would be reckless behind the wheel of any car, but who are statistically correlated with a particular type of one. But you still mark the red Camaro as more expensive to insure regardless of who buys it because it's statistically correlated with a higher degree of accidents.