On Monday, the Libertarian candidate for Senate in Iowa, Doug Butzier, died in a plane crash. Butzier, an emergency room physician, was polling at only 2 percent. But Republican candidate Joni Ernst leads the Democratic candidate, Bruce Braley, by an even narrower margin. How Butzier's supporters eventually vote is a question of national significance.

Half a dozen other Senate races are equally likely to come down to the wire. That's why control of the chamber is still uncertain, even though the election is less than three weeks away. Such uncertainty is apparently belied by the fact that the major aggregators of Senate polls are pointing in the same direction, with probabilities ranging from 60 percent and up for Republican control. This includes my own site, the Princeton Election Consortium, which recently switched away from a Democratic advantage in early October.

Indeed, Republicans hold the lead in key states, but this unanimous agreement among election forecasters conceals an Achilles heel: We all rely on the same poll data. What if that data is off?

**A Few Front-Runners in Close Elections Will Lose on Election Day**

Josh Katz at *The New York Times'* The Upshot has analyzed the performance of Senate polls since 2004. He found that the predictive accuracy of polls depends on how soon the election is and the size of the front-runner's lead. For instance, if the election is three weeks away and the front-runner leads by 3 percent or less, that candidate will still lose 38 percent of the time—nearly two times out of five.

In the period Katz analyzed, only three or four Senate races in each election were decided by 3 percentage points or less. But as of Tuesday, such narrow margins existed in six races:

Four more races—Alaska, Arkansas, Kentucky, and New Hampshire—are within 4 or 5 percentage points. So the potential exists for many close outcomes. For election nerds like me, this is Christmas.

If every front-runner today were to win, the Senate outcome would be 52 Republicans and 48 Democrats and independents. But history tells us to expect two or three of the current leaders to lose. Democratic wins in the key states of Iowa and Colorado would give a 50-50 split, with Vice President Joe Biden breaking Senate ties in his party's favor. (This assumes that Kansas independent Greg Orman would caucus with Democrats and the other independents, which is not a sure bet.) Of course, if the opposite were to happen North Carolina—with Republican Thom Tillis defeating Democratic Senator Kay Hagan, who leads narrowly—the GOP would end up with a convincing 53-47 majority.

Some races are less likely to reverse than polls would indicate. In polls spanning the last two weeks, Georgia Republican David Perdue's median lead over Democrat Michelle Nunn dropped to 1 percent after he proclaimed last week to be "proud" about outsourcing jobs as an executive at now-defunct Pillowtex Corp. To avoid a runoff, Nunn must exceed 50 percent of the vote. Turnout tends to decrease in Georgia runoff elections, so current polls are of little help in predicting the outcome of a runoff.

Voters in these six closely-fought states are disproportionately powerful, which accounts for the concentration of advertising and get-out-the-vote activity in the battleground states. Measured in terms of their ability to influence Senate control, a vote cast in Iowa is about 110 times as powerful as that of a vote cast in Minnesota. Similarly, a Colorado voter is over 30 times as powerful as that of an Oregon voter. And my vote in New Jersey, a populous state where Senator Cory Booker has a 12-percentage-point lead, is worth less than even an Oregon or Minnesota vote.

**Democrats Tend To Perform Better Than Polls Predict**** **

Even in the week before the election, polls are not perfect. For example, in two out of ten close Senate races in 2010 and 2012. In those two cases, both in 2010, the eventual winner was the Democrat (Harry Reid in Nevada and Michael Bennet in Colorado). This reflects a broader pattern: When errors occur, the outcome tends to be more favorable to the Democrat.

Let's examine the percentage difference between polls and election results:

The above table, calculated for state-level presidential and Senate contests, shows the difference between Election Eve polls with actual election results, using the median across all races decided by less than a 10-percent margin.

Overall, these numbers set a range for how wrong we would expect a poll-based view to be. Pollsters as a group underestimate Democratic performance by an average of 1.2 percent. This bias is asymmetric: When Republicans outperformed, they did so by 1.2 percentage points or less. But in four out of eight cases, Democrats surpassed polls by 2.4 to 3.7 percentage points. This bias was even larger in Colorado, where Democrats outperformed polls by at least 4 percentage points in the 2010 and 2012 elections.

**Instead of Probability, Estimate How Much Swing Is Needed**

What if this year's polls are off by 2 percentage points in one direction or the other? A 2-point advantage for Democrats would make the most likely outcome a split of 50 Democrats/independents to 50 Republicans. And a 2-point advantage for Republicans would propel them to a 53-47 majority. These outcomes match the "what if two races flip" scenarios I outlined above.

Neither extreme is guaranteed. The introduction of new survey methods may have changed the accuracy of polls—for better or for worse—and partisan polls are more frequent this year, creating another source of uncertainty. In fact, the probabilities offered by poll aggregators might not even be the best way to think about this year's Senate forecasts, in part because we tend to mentally round probabilities up to 100 percent. The Princeton Election Consortium, for instance, gives Republicans a 69 percent chance of taking the Senate; if that fails to happen, many will claim that PEC was wrong. But a loaded coin that's biased to come up heads 69 percent of the time would still come up tails three times out of ten. Or think of it like a weather forecast: If there's a 69 percent chance of rain, you'd be well-advised to bring an umbrella, but you might end up not using it.

Instead of probability, I prefer a statistic I developed in 2004, the Meta-Margin. The Senate Meta-Margin is defined as how much the final vote would have to differ from polls in key states to create an exact toss-up for control of the chamber. Given the polling errors I have analyzed in this article, neither side can breathe easier until the Meta-Margin gets to at least 3 percent in their direction. Only at that point can we predict a winner with high confidence. As of Wednesday, the Senate Meta-Margin shows a Republican lead of 1.3 percent—too close to call. So although Republicans have the advantage in polls, Democrats' track record of outperforming polls works in the other direction. For the moment, there's a decent probability that polling nerds will be surprised on November 4.

This article has been updated.