A theory that predicts the results of US presidential elections... ?
November 2004, corrected June 2006
Occam's razor says we should prefer the simpler of two explanations. I begin by reminding readers of this principle because I'm about to propose a theory that will offend both liberals and conservatives. But Occam's razor means, in effect, that if you want to disagree with it, you have a hell of a coincidence to explain.
Theory: In US presidential elections, the more charismatic candidate wins.
People who write about politics, whether on the left or the right, have a consistent bias: they take politics seriously. When one candidate beats another they look for political explanations. The country is shifting to the left, or the right. And that sort of shift can certainly be the result of a presidential election, which makes it easy to believe it was the cause.
But when I think about why I voted for Clinton over the first George Bush, it wasn't because I was shifting to the left. Clinton just seemed more dynamic. He seemed to want the job more. Bush seemed old and tired. I suspect it was the same for a lot of voters.
Clinton didn't represent any national shift leftward.  He was just more charismatic than George Bush or (God help us) Bob Dole. In 2000 we practically got a controlled experiment to prove it: Gore had Clinton's policies, but not his charisma, and he suffered proportionally.  Same story in 2004. Kerry was smarter and more articulate than Bush, but rather a stiff. And Kerry lost.
As I looked further back, I kept finding the same pattern. Pundits said Carter beat Ford because the country distrusted the Republicans after Watergate. And yet it also happened that Carter was famous for his big grin and folksy ways, and Ford for being a boring klutz. Four years later, pundits said the country had lurched to the right. But Reagan, a former actor, also happened to be even more charismatic than Carter (whose grin was somewhat less cheery after four stressful years in office). In 1984 the charisma gap between Reagan and Mondale was like that between Clinton and Dole, with similar results. The first George Bush managed to win in 1988, though he would later be vanquished by one of the most charismatic presidents ever, because in 1988 he was up against the notoriously uncharismatic Michael Dukakis.
These are the elections I remember personally, but apparently the same pattern played out in 1964 and 1972. The most recent counterexample appears to be 1968, when Nixon beat the more charismatic Hubert Humphrey. But when you examine that election, it tends to support the charisma theory more than contradict it. As Joe McGinnis recounts in his famous book The Selling of the President 1968, Nixon knew he had less charisma than Humphrey, and thus simply refused to debate him on TV. He knew he couldn't afford to let the two of them be seen side by side.
Now a candidate probably couldn't get away with refusing to debate. But in 1968 the custom of televised debates was still evolving. In effect, Nixon won in 1968 because voters were never allowed to see the real Nixon. All they saw were carefully scripted campaign spots.
Oddly enough, the most recent true counterexample is probably 1960. Though this election is usually given as an example of the power of TV, Kennedy apparently would not have won without fraud by party machines in Illinois and Texas. But TV was still young in 1960; only 87% of households had it.  Undoubtedly TV helped Kennedy, so historians are correct in regarding this election as a watershed. TV required a new kind of candidate. There would be no more Calvin Coolidges.
The charisma theory may also explain why Democrats tend to lose presidential elections. The core of the Democrats' ideology seems to be a belief in government. Perhaps this tends to attract people who are earnest, but dull. Dukakis, Gore, and Kerry were so similar in that respect that they might have been brothers. Good thing for the Democrats that their screen lets through an occasional Clinton, even if some scandal results. 
One would like to believe elections are won and lost on issues, if only fake ones like Willie Horton. And yet, if they are, we have a remarkable coincidence to explain. In every presidential election since TV became widespread, the apparently more charismatic candidate has won. Surprising, isn't it, that voters' opinions on the issues have lined up with charisma for 11 elections in a row?
The political commentators who come up with shifts to the left or right in their morning-after analyses are like the financial reporters stuck writing stories day after day about the random fluctuations of the stock market. Day ends, market closes up or down, reporter looks for good or bad news respectively, and writes that the market was up on news of Intel's earnings, or down on fears of instability in the Middle East. Suppose we could somehow feed these reporters false information about market closes, but give them all the other news intact. Does anyone believe they would notice the anomaly, and not simply write that stocks were up (or down) on whatever good (or bad) news there was that day? That they would say, hey, wait a minute, how can stocks be up with all this unrest in the Middle East?
I'm not saying that issues don't matter to voters. Of course they do. But the major parties know so well which issues matter how much to how many voters, and adjust their message so precisely in response, that they tend to split the difference on the issues, leaving the election to be decided by the one factor they can't control: charisma.
If the Democrats had been running a candidate as charismatic as Clinton in the 2004 election, he'd have won. And we'd be reading that the election was a referendum on the war in Iraq, instead of that the Democrats are out of touch with evangelical Christians in middle America.
During the 1992 election, the Clinton campaign staff had a big sign in their office saying "It's the economy, stupid." Perhaps it was even simpler than they thought.
Opinions seem to be divided about the charisma theory. Some say it's impossible, others say it's obvious. This seems a good sign. Perhaps it's in the sweet spot midway between.
As for it being impossible, I reply: here's the data; here's the theory; theory explains data 100%. To a scientist, at least, that means it deserves attention, however implausible it seems.
You can't believe voters are so superficial that they just choose the most charismatic guy? My theory doesn't require that. I'm not proposing that charisma is the only factor, just that it's the only one left after the efforts of the two parties cancel one another out.
As for the theory being obvious, as far as I know, no one has proposed it before. Election forecasters are proud when they can achieve the same results with much more complicated models.
Finally, to the people who say that the theory is probably true, but rather depressing: it's not so bad as it seems. The phenomenon is like a pricing anomaly; once people realize it's there, it will disappear. Once both parties realize it's a waste of time to nominate uncharismatic candidates, they'll tend to nominate only the most charismatic ones. And if the candidates are equally charismatic, charisma will cancel out, and elections will be decided on issues, as political commentators like to think they are now.
 As Clinton himself discovered to his surprise when, in one of his first acts as president, he tried to shift the military leftward. After a bruising fight he escaped with a face-saving compromise.
 True, Gore won the popular vote. But politicians know the electoral vote decides the election, so that's what they campaign for. If Bush had been campaigning for the popular vote he would presumably have got more of it. (Thanks to judgmentalist for this point.)
 Source: Nielsen Media Research. Of the remaining 13%, 11 didn't have TV because they couldn't afford it. I'd argue that the missing 11% were probably also the 11% most susceptible to charisma.
 One implication of this theory is that parties shouldn't be too quick to reject candidates with skeletons in their closets. Charismatic candidates will tend to have more skeletons than squeaky clean dullards, but in practice that doesn't seem to lose elections. The current Bush, for example, probably did more drugs in his twenties than any preceding president, and yet managed to get elected with a base of evangelical Christians. All you have to do is say you've reformed, and stonewall about the details.