When I deconstruct the available poll data, I find them to be very consistent - contrary to what the MSM seems to be saying. The internals seem very consistent - Bush getting in the low 90% of Republicans, Kerry getting in the high 80% of Democrats, and independents roughly split - with perhaps a slight lean to Bush.
This means it is all about turnout, and a polls accuracy depends on the validity of the turnout assumptions. The internals of the conflicting polls seem to say the following: 1) If voters turnout in the same ratio's as in 2000, we have a statistical dead heat, 2) if the Republicans can match the Democratic turnout (which would be their best turnout in recent history), Bush has a 3% - 4% lead, and 3) if the Republicans can out draw the Democrats by the amount they are typically outdrawn themselves, Bush is ahead by 8%- 10%. The real story the media should be reporting is "what will voter turnout look like"? I am beginning to wonder if anyone in the media actually performs research and analysis, or do they simply parrot the work of others.
If we are to believe the Gallup and CBS polls, we would need to believe their turnout assumptions. Have they provided any empirical evidence that we have had a dramatic shift in registrations over the last 4 years? For the Gallup ratio's to hold, we are going to have to have one of the following two things happen: 1) 4 million democratic voters have changed to Republican this election cycle, or 2) of all the new voters registered in the past 4 years, there are roughly 8 -9 million more new Republicans than new Democrats. Either of these or some combination is highly unlikely. The real story the press should be focusing on from these polls is the variation of voter registration and turnout assumptions, and which are most likely to be accurate.
Personally, I feel that the Republicans will come close to closing the turnout advantage gap traditionally enjoyed by Democrats, and therefore Bush probably holds a 3 - 4 point edge.