"X = Not a Whole Lot" -- John Allen Paulos in The Guardian, 11/18/04:
Excuse my mathematician's obsession with coin flips, but consider this. There is a large bloc of people who will vote for the Republican candidate no matter what, and a similarly reliable Democratic bloc of roughly the same size. There is also a smaller group of voters who either do not have fixed opinions or are otherwise open to changing their vote.
To an extent, these latter people's votes (and thus elections themselves) are determined by chance (external events, campaign gaffes, etc).
So what conclusion would we draw about a coin that landed heads two or three times out of four flips (or about a sequence of two or three Democratic victories in the last four elections)? The answer, of course, is that we would draw no conclusions at all.
One reason we tend to draw far-reaching conclusions about elections is the charming superstition that significant events must be the consequence of significant events.
This psychological foible is illustrated by an experiment in which a group of subjects is told that a man parked his car on a hill. It then rolled into a fire hydrant. A second group is told that the car rolled into a pedestrian.
The members of the first group generally view the event as an accident; the members of the second generally hold the driver responsible. People are more likely to attribute an event to an agent than to chance if it has momentous or emotional implications. Likewise with elections.