The remaining Republican presidential candidates have spent a lot of time attacking one another in debates, but they are also starting to talk more urgently about who can win the real fight in November.
"If you're our nominee," Marco Rubio told Donald Trump during Thursday's debate, "we will lose."
"Wrong," Trump countered. "I beat Hillary Clinton. I beat Hillary Clinton in many polls."
"Since we're talking about polls," John Kasich chimed in, "I beat Hillary Clinton by more than 11 points."
It's hard to know what to make of polls fielded months before the general election, though.
For example, a CNN poll last week found that voters preferred Rubio and Ted Cruz to Clinton, but that Clinton would beat Trump, and Bernie Sanders would beat any of those three Republican candidates. Yet many Americans still aren't following the race closely, and neither party has spent much money on negative advertising to try to change how voters see the other party's candidates. Americans' opinions about the candidates could change substantially before November.
There's another measure of a candidate's likely success that economists have found tends to be more accurate than polls: bets that people place on the candidates. It's illegal in much of the United States, but internationally, gamblers can bet on who they think will win.
And contrary to the CNN poll, a new analysis of data from two major betting exchanges suggests that Clinton would be more electable in November than Sanders - but that both candidates are likely to beat the eventual Republican nominee. And Chris Christie, possibly the Republican with the best chance of overcoming the Democrats, is already out of the race.
When bettors choose candidates, they receive odds from bookmakers on candidates who are unlikely to succeed, just as in horse racing. The bettors win substantially more money than they gambled if their candidate of choice comes out ahead.
The bookies' goal is to make money, and if a lot of their customers are betting on a particular candidate, they will shorten the odds to limit the firm's losses if that candidate wins. If customers aren't betting on a candidate, the bookies will lengthen the odds, offering more generous payments to get more people to place bets.
As a result, the odds on a candidate reflect how likely bettors around the world think that candidate is to win, a numerical representation of a candidate's chances of victory.
Typically, bookmakers will allow customers to bet that a candidate will win the party's nomination or that a candidate wins the general election. The difference between the odds that a candidate wins the primary and the odds that a candidate wins the general election is one measure of how electable that candidate is - how much he or she appeals to a general electorate.
Researchers Jonathan Hartley of the University of Pennsylvania and Kevin Hassett of the American Enterprise Institute used this fact to produce estimates of the likelihood that each candidate wins the general election, assuming they win the primary.
The figures reported here are current through Feb. 21. While more recent data is available, Cruz's and Sanders's chances of winning their parties' respective nominations have declined to nearly zero since that date as more voters have cast ballots. Bookmakers are placing such small probabilities on Cruz and Sanders defeating the other contenders in their own parties - and even smaller probabilities on a win for either candidate in November - that it's impossible to come up with an exact measure of Sanders beating the Republicans or of Cruz beating the Democrats.
Rubio is the most electable Republican candidate still in the field. Even if he wins the nomination, his chances of winning in November are only about 4 in 10.
If either Trump or Cruz is nominated, the chances that a Republican becomes president would be about 3 in 10. Trump's chances have been declining as he continues to make controversial statements about women, Latinos, Muslims and the Ku Klux Klan.
Christie, the governor of New Jersey, appeared more electable than Rubio, but he is no longer in the race and is instead campaigning for Trump.
The estimates of Christie's electability are a few months old, based on data from when he still looked like a contender in the primary. Then, his chances of winning the general election if nominated were about 56 percent on average. As his odds of winning the primary declined, the researchers' estimates of his chances against the Democrats lost precision.
Kasich has also looked electable in recent weeks - as to whether he would win if the GOP nominated him, his odds are no better or worse than tossing a coin. Again, though, because the chances that he wins the nomination are so slim, it's hard to know what to make of Kasich's figures.
On the Democratic side, Clinton had a 64 percent chance of winning in a general election if she won the primary, while Sanders' odds of winning in November were 53 percent as of Feb. 21 if he won the primary - still more likely than not, but a closer call than for Clinton.
Clinton's figure has remained stable since the campaign began, while views on Sanders have shifted over the past several weeks. Bettors have improved their opinion of Sanders' chances in November if he somehow beats Clinton for the nomination.
The recent shift in Sanders' numbers are probably due to a couple of factors, including the ever-increasing likelihood that Republicans nominate Trump or Cruz, said Hassett, a former economic adviser to the Bush and Clinton administrations.
"People are looking at the potential that a really unelectable Republican gets nominated, and thinking that even Bernie Sanders could beat these guys," Hassett said.
Meanwhile, he added, Sanders has benefited from having more time on air as the media pays more attention to the Democratic contest.
"As people have gotten to see Bernie Sanders on television, he comes across as a very charismatic and sympathetic guy," Hassett said.
The numbers are positive for both Democrats, reflecting the widespread view that Democrats will have an advantage in November's election no matter whom they nominate. The economy is improving, favoring the party that currently controls the Oval Office, and the country has slightly more Democrats than Republicans in any case. The estimates of electability show that the differences between the candidates in policy and personality aren't all that important relative to these larger, impersonal factors.
Betting markets yield only probabilities, not certainties - saying that Clinton has a 64 percent chance of winning means she also has a 36 percent chance of losing. And the markets can be wrong if candidates have crucial strengths or weaknesses that, for some reason, no one can see. These estimates aren't perfect, but they're one of the best measures available at this early date of how well the candidates would perform in November.
• 03/07/16: The key difference between how Trump and Romney made their money
• 02/15/16: What happens when a Supreme Court justice dies in an election year?
• 12/07/15: Why you should stop calling Donald Trump a fascist
• 12/04/15: Contrary to conventional wisdom, the U.S. has actually had a major decline in gun violence. But here's why