If you had two reports on your desk, one says "this will save X million people" and the other says "this will kill 10000" ...which one would you pick?  There was a research paper that of course, I can't find, that studied this -- it was exactly the same report, one was slanted toward how many lives would be lost vs how many lives would be saved based on a new drug.  By a huge margin, the saving of X million was more desirable than killing 10000, obviously, yet the result is the same (assuming it went to market of course).  Similar, a company that has a revenue of 50 billion that loses 50% of their revenue in one year (taking them to 25 billion) looks very bad, but returning to 50 billion in the next year, a gain of 200% looks awesome.  The net result is still 0.  You can't just look at a number, a percentage and draw a conclusion - there's always more to the story.  Now to bring these two ideas together...

I feel political polls are a tie of Heisenberg's uncertainty principle and statistics.  There's a billion varieties of polls but I'll use the opinion polls in general and point out the lack of accuracy and reliability -- and most importantly, why I think this.  So say you ask 100000 random (ok, seemingly random) people two things, "Do you think the country is going in the right direction" and "Do you think the president is doing a good job" - two very common questions I see on the news a million times over and yes they are similar yet not 1:1 questions.  I would argue that 1st, the first question, more often than not, influences the other and 2nd, it doesn't contain a check/balance.  Example: if you asked "do you think people at your job are using the internet irresponsibility while on the clock?" a check/balance would say "do you abuse the internet at work?".  Further, applying Heisenberg's principle, I feel knowing the answer at that point in time, you cannot assume the influence of that report will not influence the direction it was going, regadless of positive or negative.  Even more unapparent is the influence of asking people the question!  They feel important for that instant - "I took part in a poll, my opinion is valuable".  So twice over you are affecting the system in a manner that will surely skew the results!  Once by asking, another by publishing.  It's like having room for a road (say its packed dirt) and inviting some people to try it -- others will follow even if it wasn't their idea to begin with since there was an influence to the system.

Now let me be clear, I don't have an answer on how to fix this sort of thing, nor is it a perfect theory.  There's a ton of aspects of psychology I could bring into this too, but it might be a bit much for this single post and if you care that much, you'll seek them out (or just message me).  That said, there's one, surefire way to find out.  You have an election.  Just like football (you were wondering when that was coming?) -- just because a team is far better on paper does NOT mean they will win -- that's why you play the game, to find out because there are NO guarantees until after it's over ...then it's fact/history.