There should be something like Heisenbergs Uncertainty Principle for politics (and movie reviews). By observing, analysing and publishing, pundits will (and are probably trying to) affect the outcome.

There is actually something to that. Statisticians look for correlations in voting trends to help predict future results. For instance, some analyst finds that one suburb in Michigan strongly represents the results of every election for the past 50 years. So, the news media picks up on the story and uses that to talk about how the candidates stack up in that suburb. This makes the candidates focus heavily on that suburb, so that the news will say their campaign is strong, which makes people think they are a good candidate, so they get more votes. But candidates could be hyperfocused on the one suburb and don't bother polling the surrounding counties where they have fallen far behind. All it takes is for some people who haven't traditionally voted in large numbers to show up at the polls, and the campaign is caught sleeping.