With a bunch of new polls released over the weekend, speculation about how a general election race between Hillary Clinton and Donald Trump ends is beginning to matter. Just not in the way you would think, based on all the hype.
We have three ways to predict what will happen, each with strengths and weaknesses. First, there are prediction methods based on the "fundamentals" of the election. Then we have betting markets and polls.
Political scientists and others have developed models to predict elections based only on what we can know before the candidates are chosen and their campaigns take place.
It turns out that roughly four factors matter. The incumbent party's candidate does better when the economy is strong, when presidential approval is high and when the nation is at peace.
But the candidate's chances appear to suffer if the incumbent party has already had two or more terms in the White House.
All this may seem obvious, but the trick is in measuring these variables and figuring out the proper weights to give each. Scholars disagree about the extent to which these "fundamentals" determine outcomes, compared with the effects of individual candidates and campaigns.
With those caveats, most models predict a close race, with perhaps a small advantage for the Republicans.
The economy is growing, but not thriving. Barack Obama is popular, but not excessively so: The HuffPollster estimate of his approval rating stands at 50 percent right now. And third terms for the same party are rare.
A different way of forecasting election results is to follow the wisdom of crowds, found in election betting markets (such as those aggregated by PredictWise). Some critics of these tools believe they only quantify conventional wisdom, which is as likely to be wrong as right.
But prediction markets have the advantage that the participants, and therefore the results, can take into consideration any relevant information -- as opposed to the "fundamentals" analysis, which excludes anything specific to this election cycle, and polling, which only looks at current public opinion and therefore ignores predictable changes.
So far this year, Clinton has been the solid favorite. Predictwise currently gives the Democrats a 67 percent chance of winning in November.
Trump has been closing the gap in recently released polls; HuffPollster's current estimate has Clinton at 43.6 percent of the vote, Trump at 42.0, with the remainder undecided or other.
As several commentators have noted, the relevant context for understanding current polls is that Trump has won the Republican nomination (or at least all other candidates have dropped out), while Clinton still faces an active challenger. Right now, Trump is ahead of her in consolidating his vote.
But at least some Bernie Sanders supporters who currently have an unfavorable view of Clinton will wind up supporting her in November.
The best way to treat these indicators is as objective starting points.
Combined, they tell us that Trump's stronger polling in the last few weeks doesn't really change the basics: He's still the underdog.
But from there, each of these methods requires us to make subjective judgments about several unknowns.
Will elite Republican unhappiness about Trump have any effect on Republican voters?
Will Democrats have difficulty unifying after Sanders eventually concedes defeat?
Which will eventually turn out to predict swing voters' behavior better: Trump's historically high unfavorable ratings, Clinton's slightly lower but still very high unfavorables, or some combination of both?
Will early indicators of high anti-Trump turnout by groups he has insulted prove accurate, or will antipathy for him fade, especially if he tones down his act?
We have to wait until the conventions are over before the polls start to be good predictors of November results.
Write Jonathan Bernstein, a Bloomberg View columnist on U.S. politics, at firstname.lastname@example.org.