First off, you’d need to also factor in the percentage of large dogs, as no matter how vicious a toy breed or even medium sized dog is, it isn’t going to have a high fatal bite count. So out of the 36% of dog households that have a large dog, pitt ownership might be more than 6% of the total.

Then again, we need to look at other factors as well.

Maybe 70% of rescue dogs are pitts and 100% of fatal bites were from rescues? (Or vice versa, that 100% of fatal bites were from rescues and 70% of the rescues that went on to bite were pitts, which is a more subtle but still very different picture of events which might reflect fairly narrow causal environmental factors like prior fight training.)

Without the additional layers of data, the best we can do is draw potentially misleading conclusions around causative factors when we barely have correlative ones.

And the ways in which this could be dangerous in terms of social policy is if actions are taken around the mythos of it being a breed specific trait, it not being that, and then unexpected outcomes occurring, such as a popularity shift towards an even more dangerous breed as pitt ownership declines or ignoring or even exacerbating underlying causal relationships to environmental factors.

We’ve seen how bad data science applied to human crime rates can lead to supremely (supremacist?) misleading claims around the contributing factors with an over representation of demographic data that’s simply correlative to underlying causative environmental factors.

So if we both know full well that saying “XYZ demographic is 2-3x more likely to commit violent crimes so we should get rid of XYZ demographic from the population” is an outrageously bad faith argument predicated on poor data analysis, I’m curious what you think is materially different about the data evaluation aspects that you support the analogous claim here?