Column: Clinton vs. Trump by the numbers, or how to really understand those election forecasts
Political junkies love numbers. Forget discussions of abortion or LGBT rights, or what to do about Aleppo, or the future of Social Security; what they obsess over are poll results and electoral vote counts.
Among the most popular statistics of all are probability forecasts, which combine poll results and a secret sauce of “fundamentals”— demographics, economic conditions and the political environment, among other factors — to determine each major presidential candidate’s chances of winning on election day, Nov. 8.
As we write, the election forecast at Nate Silver’s fivethirtyeight.com gives Hillary Clinton a 71.2% chance of beating Donald Trump (who accordingly has a 28.8% chance of winning); the New York Times Upshot blog has her at 80%; Daily Kos Elections has her at 72%; and the Princeton Election Consortium at 85% or 91%, depending on how you crunch the raw numbers. Although the percentages vary, these sources all tend to move together—if one is rising or falling, usually they all are. But they differ in the weight they give to individual polls and to fundamentals—and in how thoroughly they disclose their formulas to the public. (For an example, Fivethirtyeight’s description of its methodology is here.)
Things with 30% probability happen all the time. That doesn’t mean you should bet on them, but you shouldn’t rule them out either.
— David Byler, RealClearPolitics
Yet it’s likely that few laypersons actually understand what these numbers really signify. They just use them to validate their mood (grim if their candidate’s score is low or falling; euphoric if big or rising). That’s especially true since none of the forecasts at this stage of the election points to a decisive outcome; even at 28.8%, Donald Trump could still win, just as a weatherman can be judged accurate if he forecasts a 30% chance of rain and it, indeed, rains.
The forecasting phenomenon combines Americans’ love for hard numbers with their well-documented inability to comprehend statistics and probability. No one should take this as an insult, since even math professionals sometimes find statistics and probabilities hard to master. The forecasts also underscore our obsession with horse-race campaign reporting, in which the candidates’ positions take a back seat to snapshots of who’s ahead or behind, by how much, and how some election-related event will affect the trend.
But that does leave open the question of what these forecasts mean, and how we should read them.
It’s important to understand that the forecasts are inherently uncertain, especially early in a campaign. Forecasters don’t often make that very clear. Rather, “most researchers overstate the confidence in their fundamentals-based forecasts, exaggerating the probability that either candidate should be expected to win,” wrote Drew Linzer, a former pollster who oversees the poll analysis for Daily Kos Politics, in an analysis he co-authored in 2013, when he was teaching political science at Emory University.
Uncertainties also result from the sheer complexity of the American presidential election process, which takes individual results from 50 states and filters them through the Electoral College. Some state results correlate with each other—the result in Minnesota may resemble that in Wisconsin, though not with Texas—some may have favorite sons or daughters at the top of the ticket, and the electorate in some may be more sensitive to economic conditions than others.
Deciding how to weight fundamentals in a forecasting model is hampered by the dearth of historical evidence. There have been only 17 presidential elections since World War II, and political and demographic conditions have changed dramatically even over that period.
Forecasters try to communicate these uncertainties in different ways. Some will point out that even in a series of random coin-flips, the chance of coming up heads or tails two or three times in a row can be 25% or so, suggesting that a candidate with a low probability number can’t be counted out entirely. The Upshot compares the probability of a Clinton loss despite her 80% figure to the chance that an NFL kicker will miss a 42-yard field goal.
Yet are these comparisons relevant? A coin flip is the product or nearly total random chance, and a place-kicker’s success the product of countless factors including field conditions, physiology, and the quality of the offense and defense.
More to the point, people normally judge probabilities “in the context of events that can happen over and over,” says Drew Linzer, a former political science professor and pollster who oversees the poll analysis for Daily Kos Politics. “Intuitively, we understand that we don’t know what’s going to happen on the next coin flip, but that over the long run, half of them will come up as heads.” By the same token, about 1,000 field goal attempts are made in the NFL during an average season.
“But in this case, there’s only one presidential election,” Linzer says. “It’s not as if we’re going to run it over and over again and in 30% of those elections Donald Trump is going to win. In a one-event world, we talk about our degree of belief and our level of certainty.”
Some in the polling community believe people are better than they’re given credit for at actually processing the uncertainties. “A person on the street who sees a 30% chance will think, ‘That really could happen,’ and that’s the right conclusion,” says David Byler, elections analyst at RealClearPolitics. “Things with 30% probability happen all the time. That doesn’t mean you should bet on them, but you shouldn’t rule them out either.” (RCP produces an average of election polls, but not a forecast.)
Linzer contends that most people can grasp gross probabilities, though not fine distinctions. “You can look at a weather report that says there’s a 70% chance of rain and know it’s probably a good idea to bring an umbrella,” he says. “But it takes a lot of training to understand the difference between something that happens 60% of the time and 70%.”
In part for that reason, “I actually have a lot of hesitation about the focus that goes on that top-line probability number,” he says. “It’s something we produce because there’s a lot of demand for it.” He believes that a more useful number is “the range of electoral votes that all of us believe are a plausible outcome.” This is represented on Daily Kos Politics and other forecasting sites by a diagram showing the curve of possible vote totals in graphical form.
Underlying this discussion is the question of whether horse-race campaign coverage serves the public. By its nature, says Brendan Nyhan, a media critic and professor of government at Dartmouth, “it leads to too much emphasis on minor swings in the polls that are likely to prove temporary or illusory.”
He also thinks it reflects the characteristics of the more voracious consumers of political news. “People who follow politics almost always have a dog in the fight,” he says. “They’re looking to see who’s up and who’s down because they’re not looking for information about who to vote for.” Nyhan advocates “smarter” horse-race journalism, which at least avoids overemphasizing small changes in the polls and recognizes that most events assiduously covered in the 24-hour news cycle “do not move the needle” in support or opposition to an individual candidate.
The truth is that for nonprofessionals, forecasts of win or lose probabilities tend to make the race look more dynamic than it really is.
The forecasts do lend an aura of drama to a campaign in which polling averages have been fairly stable for many weeks, most showing that Hillary Clinton has been leading nationally since this summer’s conventions. The shifts in the race have been small and transitory.
Only now are the forecasts beginning to provide a more reliable picture of the state of the race, in part because they’re increasingly based on polls rather than fundamentals as election day approaches. Yet they’ve been providing thrills and chills for months. If that’s a reflection of Americans’ view of politics as sport and their lack of interest in what’s at stake in this election or any other, that may not be a good sign for participatory democracy.
Keep up to date with Michael Hiltzik. Follow @hiltzikm on Twitter, see his Facebook page, or email [email protected].
