Below is a chart of all reasonably current polls in Pennsylvania. On one axis, I’ve plotted Clinton’s lead in each poll, and on the other, the number of undecided voters in that poll.
As you can see, there is a strong, if not overwhelming relationship between these two things: Obama tends to do better where there are more undecideds, and Clinton tends to do better where there are fewer. The relationship is statistically significant at the 95 percent level.
If we trace the regression line such that it crosses the y-axis — meaning, there are zero undecideds — we’d project a Clinton win by 16 points.
Is this a valid way to look at the polls? I have no idea. But we can run a gut-check of sorts. The average of these 14 polls is: Clinton 48.3, Obama 40.9, Undecided 10.0. For Clinton to win by 16 points on election day, that would imply results of Clinton 57.5, Obama 41.5, assuming that 1 percent of the vote goes to minor candidates.
In order for that to occur, Clinton would need to pick up 9.2 points from undecided voters, to Obama’s 0.6. In other words, she’d have to win nearly every undecided voter. Even when elections break at the end — they don’t break that strongly. Maybe Clinton could win 2 out of 3 undecided voters (which would imply a victory margin of about 10 points), but not 9 out of 10.
So — I wouldn’t take these numbers all that seriously. At the same time, I think there is a case to be made that Clinton has a couple extra points worth of cushion in her Pennsylvania numbers versus what the polling averages currently imply.