More Poll Follies

The latest New York Times/CBS News poll came out this morning, and, like several others, it shows President Bush and John Kerry locked in a 47% to 47% tie, compared to an eight point lead for President Bush after the Republican convention in September.
The poll’s internals are easily accessible–which is praiseworthy–so it takes only a moment to determine that the October poll sampled 34% Democrats and 29% Republicans, while the September poll sampled 33% Republicans and 31% Democrats. So it’s hardly a surprise that Kerry did better in the October survey. If the pollsters sampled only Democrats, they could show that Kerry was sweeping toward an unprecedented victory.
We aren’t going to be able to untangle the pros and cons of “correcting” samples between now and November; suffice it to say, however, that 4% of the population didn’t abandon the Republican party for the Democrats over the last 30 days.
I also note that the proportion of liberals sampled in today’s Times/CBS poll is the highest they have recorded in any Presidential poll since 1995. Maybe Kerry should be worried that the best he could manage was a tie.
It’s noteworthy that all of the polls that over-sampled Republicans in September are now over-sampling Democrats in October. Is this a coincidence, or a deliberate effort to manufacture a Kerry “comeback” to generate momentum for the Democrats? One possible explanation, as least as to the Times/CBS poll, is that their September poll was taken on a Monday through Wednesday, while the poll released today was entirely done on the weekend, when pollsters know they will tend to find more Democrats at home. So was the choice of polling dates deliberate, or coincidental?
For what it’s worth, those polls that weight samples to produce a consistent blend of Republicans, Democrats and independents have found little or no change since the first Presidential debate.