I don't mean to dodge the question, but I will answer it in a way that relates to what the information commissioner in the U.K. has ruled in relation to her investigation. Inferred data is considered personal data when it is attached to an identity. It's this idea of creating predictions about people and that when you attach them to their voter file, it constitutes personal data.
I think it gets to your question about campaigns that are trying to predict the behaviour of potential voters, and it's based on predictions rather than verifiable, deterministic facts. That could be one boundary that needs to be further negotiated. For example, if I have a gun license, then you have a verifiable fact that I support gun rights. Whereas, if you're using my social media chatter to infer my feelings about gun rights, that's a different threshold that needs to be defined.