FiveThirtyEight’s Prediction Model Is Failing in the Clinton-Trump Race - That's According to Its Guru, Nate Silver
Telling political junkies to stop checking fivethirtyeight.com obsessively is likely to be met by the same hollow-eyed stare from a lab rat that spends its day clicking a lever for bumps of cocaine.
The internet venerates the website’s founder, Nate Silver, as “America’s Chief Wizard” after his statistical polling model correctly predicted the results of the 2012 presidential contest in every state and the District of Columbia. But Silver’s fans are freaking out as Trump’s chance of winning has tripled in two weeks.
However, the wild swings toward Trump are more a sign that an orange-furred monkey wrench has jammed FiveThirtyEight’s soothsaying machine than a candidate who is detested by nearly 60 percent of voters is suddenly floating to the top of the 2016 shitshow. Trump can win and possible hacking of the electoral process increase unpredictability, but his path is still precarious and not one other professional election handicapper records the seismic shift in Trump’s chances that FiveThirtyEight does.
To Silver’s credit, he’s cast doubts on his poll-based forecasting this year. He mentions the lag time in how polls react to dramatic news such as the FBI’s email bombshell. He highlighted a survey showing little movement in voter preferences since January 2016. FiveThirtyEight also notes, “Trump supporters are more likely than Clinton voters to make it through the likely voter screens, indicating they are more vocal and enthusiastic in their support.” In other words, it’s likely Comey’s ill-advised letter about Clinton’s emails has depressed pro-Clinton respondents in polling more than it will at the voting booth. On the flip side, after Clinton’s average lead stretched to 7.1 percent in mid-October, this was likely affected by Trump supporters shunning pollsters, particularly after Trump’s rants about a rigged media, rigged polls, and a rigged election.
If you’re hell-bent on checking FiveThirtyEight constantly (as I still do), then it’s worth keeping in mind reasons why a Clinton victory is likelier than Silver’s model predicts.
The main issue is FiveThirtyEight works best when electoral coalitions are well-defined and change slowly as they have been since Bill Clinton was in office. Trump has scrambled all that, however, and analyzing polls cannot account for rapid demographic shifts. Trump’s bastion of working-class white men includes significant numbers of Democrats who’ve crossed over since 2012. But he’s alienated typically Republican college-educated whites and spurred Latinos, Asians, and Muslims to mobilize in record numbers to defeat him. And the gender gap is wider than ever, with women favoring Clinton over Trump by an average of 16 points in October polls. The focus on Trump’s working-class support also misses one crucial aspect: a majority of the white working class is female, and they began rejecting Trump after the Access Hollywood tape scandal. This overlaps with a remarkable geographic shift in which Democrats are quickly gaining ground in Sunbelt states that were a redoubt of Reagan conservatism, while traditional Democratic strongholds are crumbling in Midwest regions that are older, white and working class.
These groups are huge at a national scale, but fine-grained detail disappears in polls of a thousand respondents. FiveThirtyEight apparently doesn’t include polls that could clarify the picture by zeroing in on groups such as Latino voters, who favor Clinton by an astonishing 48-point gap. Most public polls are marked by shoddier methodology that amplify small swings as opposed to internal campaign polling that show more stability.
Two other significant reasons why FiveThirtyEight overestimates Trump’s chances is early voting and get out the vote. More than 35 million Americans have already voted, which is 28 percent of the total vote in 2012. In nearly every battleground state early voting is outpacing last election, meaning Clinton banked millions of votes when her numbers were peaking. One poll of early voters indicates Clinton is ahead by 5 points, which is enough to tack on a 1.3 percent advantage in the final election results.
Many of these factors are evident in Nevada. FiveThirtyEight shows Trump neck and neck with Clinton, but Silver says early voting shows a six-point gap favoring Democratic voters. This discrepancy is consistent with FiveThirtyEight’s failure to predict Harry Reid’s 2010 Senate re-election and its underestimate of Obama’s 2012 victory margin. Six years ago Reid edged out Tea Party kook, Sharron Angle, thanks to an extraordinary turnout machine he’s built after more than 30 years in public office. Nevada also has its own election oracle in reporter John Ralston, who correctly called the two races Silver flubbed. Ralston’s dive into 622,000 votes already cast, which already accounts for 61 percent of 2012 totals, bears grim tidings for Trump even in best-case scenarios.
If Trump loses Nevada, Silver says he wins in only 9 percent of scenarios. That’s a far cry from the 35 percent mark Trump just touched. Trump’s campaign looks to be treating Nevada as a lost cause as he eyes richer electoral prizes like Michigan and Wisconsin that are even further out of his grasp. Even if Trump wins Ohio, Florida, Iowa, Arizona, North Carolina, and New Hampshire, he still loses the election. So he needs to wrestle a Midwestern state away from Clinton.
That’s unlikely as Trump is paying the price for outsourcing get-out-the-vote operations. GOTV is warfare through electoral means and requires a command staff, hundreds of offices, tens of thousands of paid staff and volunteers, and coordination both strategic and minute with the campaign. Trump has none of that. A month before the election, FiveThirtyEight counted 2.5 Clinton field offices for every one of Trump’s. His campaign is also plagued by chaotic websites and offices listed in demolished buildings. Field offices can’t be thrown up like lawn signs, and in battleground states a ground game could add as much as three to five points.
Clinton’s GOTV is robust enough to target strongly Democratic but low-propensity voters. Silver points out polls can’t factor this in. For example, one organization in Florida has dispatched 500 paid canvassers to the field for months to activate a pool of 384,000 Latino voters. Trump is also outmuscled by unions spending hundreds of millions of dollars and deploying thousands of full-time canvassers for Democrats. These efforts are yielding dividends among Latinos in Arizona and Asian-Americans in Nevada. And Michigan’s Arab-American and Muslim-American firewall will easily keep Trump at bay there.
Democrats are hand-wringing over declining African-American turnout, and this has put North Carolina and Ohio in peril, but the surge from women, college-educated, and racial minorities from Virginia to Nevada will offset this deficit. Republicans are already embroiled in civil war, which is damaging Trump in Wisconsin and Utah. Democrats are united and feature a stellar array of surrogates on the hustings, such as the Obamas, Bill Clinton, Joe Biden, Tim Kaine, Bernie Sanders, Elizabeth Warren, and an A-list of celebrities. Donald Trump has the Pharma Bro, an underwear model, a disgraced baseball star, and a felonious “Real Housewife.”
There is also no evidence of “shy” Trump voters that polls allegedly fail to register. And Trump’s attempts to activate whites who normally don’t vote looks to be a dud as well. An analysis of newly registered and “missing” voters favoring Clinton. Even if there is a hidden upsurge for Trump, it’s unlikely to overcome all these obstacles.
Silver cautions as well that the history of presidential elections and scientific polling is so scanty in terms of data that it’s dicey to draw any broad-based conclusions. For instance, conventional wisdom says as goes Ohio, so goes the nation. But Trump looks set to win Ohio and lose the general election, a first in 14 straight presidential races.
FiveThirtyEight’s model worked fantastically well when the sailing was smooth, but it is foundering in the violent seas of 2016. The Cook Political Report, the brainchild of the dean of election forecasting, Charles Cook, observes that despite Clinton’s eleventh-hour woes, “The race has tightened to its ‘natural resting place’ with a 2-4 point lead for Clinton,” while Trump’s “path to 270 electoral votes remains decidedly and almost impossibly narrow.”
The best prediction of how the 2016 campaign ends is likelier to be a metaphor than math: Hillary Clinton’s near-collapse at the September 11 memorial where her entire team mobilized to shield her and carry her over the finish line. And no statistical model can predict a moment like that.