Very informative. I have an idea that would take the weight factor away and be a bit more fair.
Take 1500 people divide 52% that voted for Bush and the 48% that voted for Kerry and ask 3 questions with the choice of one, 1) More often than not approve of Bush, 2) about the same, 3) more often than not disapprove of Bush. Take the "about the same" and divide equally amongst the other two questions. This method would eliminate all party affiliation and take a percentage based only on an elected result.
Your suggestion would seem to make sense. In any event, there are a couple lifetimes until voting time so any poll means little right now....