Let's walk through this. My post states that a random voter is "more likely to have voted for her than for Trump." This is a fact. Non-voters do not affect whether the P(H)>P(T). You see. If two people voted for Hillary, one person voted for Trump, and 997 people were non-voters, it is still two times more likely that a random person of that 1000 voted for Hillary. If you think that is otherwise, then I do not know what to tell you.
As for carefully selecting states, well then that would not be a random selection.