Trying to predict the next US president is generally like shaking an 8-ball and seeing what pops up. If you took a million people who were all trying to predict who would have won each election, you'd get more than a handful of them who would have gotten each election right from 1984 even if they were all flipping a coin between the two candidates. I can't give too much stock to whoever this pollster is without knowing more about their methodology, and even then, there isn't a great way of breaking down why one method is better than another until maybe after the fact.