Rasmussen weights his samples using a 39% Dem 35% Rep 26% Ind breakdown. He believes that this will be the composition of the actual voter population on November 2 of this year. A large Harris party ID survey (6000 respondents) earlier this year found a simliar partisan breakdown among voters.
Newsweek, LA Times, and others that show wild fluctuations from poll to poll do not weight their samples by party ID at all. Generally speaking, the large spikes and dips in these polls can be attributed to their samples being skewed in favor of one party in one poll, and the other party in the next. In fact, if you apply a more realstic party weighting to yesterday's Newsweek poll, you will get results similar to Rasmussen.
Rasmussen attributes his embarrassing showing in 2000 to his failure to weight his samples by party. He claims that had he done so, his poll would have shown Gore leading by 1, instead of Bush leading by 8 or 9 in his final 2000 poll.
I am skeptical that this weighting is representative of party affiliation, especially among likely voters and especially if done earlier in the year (when Bush and Republicans generally were in a weaker position than they are today). More generally, I distrust polls that weight by party affiliation because the weighting destroys the randomness of the sample and thereby renders the poll unscientific.
You know all these polls assume a lot and are so much BS. They assume that they are all getting a clear picture of the voting public. I question if that on a holiday weekend an accurate read is even possible. I also question the weighting of the voting public. Turnout is everything and that is something the pollsters cannot even come close to getting right. I think President Bush is 7 to 9 points ahead based on a post convention bump. How long this bump lasts and if it can be increased is up to the president and how he handles his campaign between not and Nov. 2.
If this is the case - then Rass polls can be trusted a little more - And you are exactly right with regard to the Newsweek poll - if you weight it properly it shows GWB with a lead ......but more around 5pts....not the 11pts -
We are up right now - but not by double Digits - Just as the LA Times poll that was so skewed for Kerry a month ago....the same thing happened in our direction with the latest NewsWeek poll -
Isn't that a little arbitrary if used as a blanket nationally? I mean, different regions, and states, have different weights. Calling as many Republicans in New York as in Texas is a bit foolish.