Ah, how some of us clung to this alluring historical precedent. You likely heard it touted during the campaign- undecideds traditionally break 2 to 1 for the challenger, any incumbent not polling more than 50% come election is in serious trouble, blah de blah blah. Charles Cook was just one guy noting the importance of the pattern.
That’s why it is a mistake for people to focus on the spread between the two candidates, the far more relevant figure is the actual vote percentage of the incumbent in a poll (or better, average of polls). If you assume that Nader/others get about two percent of the vote (down from combined 3.1 percent last time), if President Bush is at 46, 47 or maybe 48 percent of the vote going into election day, he probably loses, 49 percent, on the cusp, 50 percent wins.
The majority of polls going into Election Day had Bush in the neighborhood of 47–49 percent. Is the Incumbent Rule discredited for future elections?
I’m certainly not putting much faith in it come 2012.
The problem may not be the incumbent rule is wrong, but the polling done was somehow seriously flawed in terms of sampling. Note the exit polls predicted a Kerry win, and Bush won.
There a a LOT of possible problems with sampling in any poll. Some off the top of my head:
#1) Polls only indicate what those who responded said. What if those who hang up as soon as they know a pollster is at the other end are non-representative of the average voter?
#2) Another twist on the cell phone factor: What if those with both a landline and cell phone tend to accept calls directly only on the cell phone, and the landline just goes to an answering machine? I am such a person. And, this is independent of the notion of cell phones being used heavily by the young. I am 43. On average Republicans tend to be wealthier on average. I’d expect that there are a huge number of wealthier Americans 30 and over with both a landline and a cell phone. If so, the pre-election poll samples underrepresented likely Republican voters.
I’m sure others reading can come up with other possible sampling biases of polls.
Location of poll samples can seriously skew the results. The average of the averages is not always the overall average. Population isn’t distributed evenly over a given state and the vote is often inconsistent region to region often independant of population density.