Politics & Analytics: The election that confounded everybody

How the ‘13 Keys’ model outperformed its creator.

In a divided nation, the unprecedented 2016 presidential election was difficult to predict. Image © delcreations | 123rf.com

In a divided nation, the unprecedented 2016 presidential election was difficult to predict. Image © delcreations | 123rf.com

By Douglas A. Samuelson

What an election! Only two prominent forecasters got it right, and one of those only with some after-the-fact spin. Actually, three, if you count film director Michael Moore. Maybe. Perhaps a fourth, if we stretch a bit. We’ll get back to that.

Professor Allan Lichtman’s “13 Keys” model [1] was spot-on once again, in one of the most hard-to-forecast presidential elections in living memory. Well, sort of. Actually, Lichtman got one key wrong and missed the popular vote result, which is what his model is supposed to forecast, but.... Oh, OK, it’s complicated. Patience, please.

As reported in Analytics magazine last July [2], at the time the Keys model looked too close to call. It uses 13 yes-no variables (see box) that reflect satisfaction with the incumbent party. It’s a statistical pattern recognition, a kernel discriminant function analysis [3]. As of July, three keys were still undetermined: incumbent-party contest (Key 2), third party (Key 4), and foreign policy success (Key 11). By mid-September, however, Lichtman felt sure enough to make his prediction. The persistence of the Bernie Sanders effort through the convention, garnering more than one-third of the delegate votes, “probably should have” tipped Key 2 against Clinton, but Sanders’ endorsement of Clinton before the convention left Lichtman doubtful about it. “I had six other keys,” Lichtman explained, “so I didn’t think much more about it.”

Lichtman, a quantitative historian and a professor of history at American University in Washington, D.C., decided that the Iran nuclear deal, the Paris climate change accord and the shrinking territory held by ISIS were not enough to hold Key 11 – foreign policy or military success – in Clinton’s favor, as these achievements did not seem to have generated widespread popular acclaim. This meant that Gary Johnson’s third-party candidacy, consistently polling over 5 percent – actually, over 10 percent in September – was enough to turn the sixth key negative. (The other negative keys were Key 1, House seats; Key 3, incumbency; Key 7, policy change; and Key 12, incumbent party candidate’s charisma.) Therefore,  Lichtman forecast a Trump win in September, when Clinton was leading comfortably in the polls.

Success, right? Ah, but wait! The Keys model forecasts the popular vote, not the electoral vote. Clinton won the popular vote – narrowly, but she won, as much of her lead evaporated in the last two weeks before the election. But also during that time, Gary Johnson’s vote slipped below five percent – he wound up with about 3.7 percent. So the model was right, after all, but not in the way Lichtman interpreted it.

However, as Lichtman is quick to point out, predictive models have to be correct ahead of time, not in retrospect. He called Key 4 against Clinton when Johnson was polling above 10 percent, consistent with how he has interpreted the third-party key in past years.

“You’ll notice that this time I didn’t distinguish between the popular and electoral vote,” he added. “I just predicted that Trump would win, because that’s what the keys indicated. In close elections, the Democrats always have an edge in the popular vote because of the huge margins they can run up in a few big states, notably California and New York. But the electoral vote doesn’t always follow the popular vote that closely any more. So far, nobody has commented on how that relationship has changed.”

At least Lichtman deserves credit for consistently insisting the election would be close. More important, this election indicates more strongly than ever that general satisfaction or dissatisfaction with how things are going eventually overwhelms specific events in the campaign.

It also marks, in Lichtman’s opinion, a major change in the election dynamics. “Truth, your record, demeaning large groups within the populace, even clear illegal acts don’t seem to matter,” he adds. “Herman Cain had to withdraw in 2012 because three women accused him of sexual harassment, which he denied. Trump had more than a dozen women accusing him of sexual harassment, plus a video of him bragging about how he’d done it, and he just kept right on going. All the precedents are out the window. If a candidate like Trump can win, anyone can.”

Why the Polls Were Wrong

On the other hand, one thing this election clearly does is cast serious doubt on the polls. “Polls are snapshots,” Lichtman says dismissively. “They’re poor predictors.” But this year was worse than most others in recent memory, with changes of 4 percent or 5 percent between the actual vote and polls taken less than a week before the election. The most likely explanation for the disparity between the polls and the actual result, according to a number of pollsters and political scientists including Lichtman, is that the pollsters’ estimates of respondents’ likelihood of voting were seriously wrong.

John Zogby, a senior partner at John Zogby Strategies and one of the most accurate pollsters for the past 30 years, saw the problem coming. Three weeks before the election [4], despite Hillary Clinton’s double-digit leads in most polls at the time, he declared, “I can’t tell you who’s really going to win. Tell me who will vote, and I’ll tell you who will win. If we get around 132 million votes, as we did in 2008 and 2012, Hillary wins. If we get 121 million, as we did in 2004, Trump wins.”

We actually got around 126 million, giving Clinton about a 2 million vote lead. Shortly after Election Day, with 122 million votes counted, her lead was closer to 200,000. Several million non-voters, particularly Democratic-leaning people in the north central states, helped tip those states to Trump. So Zogby’s assessment looks quite accurate.

Zogby also asserted, a week before the election, that the memo from FBI Director James Comey was a non-factor. The week of Oct. 23-28, an ABC poll showed Clinton leading Trump by 14 points, but on Monday, Oct. 31, it turned into a 1-point lead for Clinton, and on Tuesday, a 1-point lead for Trump just a week before voters went to the polls (except, of course, those who had already voted early). “Very simply [Clinton has] been in a downward spiral for the last nine days, so this was not an overnight thing or a Comey thing. This has been a slow and steady decline by Hillary,” Zogby said Nov. 1, a week before the election, to Steve Malzberg on “America Talks Live.”

Another person who called it right, surprisingly, was film director Michael Moore, who said in July that anger and frustration among white male skilled workers, mostly union members, in the north central industrial states could tip those states – Wisconsin, Michigan, Ohio and Pennsylvania – away from Hillary Clinton. He was far from alone – according to some unconfirmed reports, the state Democratic chairs and Bill Clinton were trying to convey the same message to Hillary and urging her to push in those states with a strong economic message. She had never appeared in Wisconsin during the general election campaign, and she had skipped the traditional Labor Day Democratic campaign rally in Cadillac Square in Detroit.

Lichtman is skeptical that it made a difference in the outcome: “Strategic campaign decisions, like whether to campaign hard in Wisconsin, are only right or wrong in retrospect. If she’d carried Wisconsin, Michigan and Arizona you’d be calling her a genius.” Again, those campaign decisions occur within a context of general satisfaction or dissatisfaction with how things are going in the country, and those metrics of general satisfaction or dissatisfaction, according to the Keys model, drive the result.

The New Tribalism

There are a couple of other factors that have apparently escaped notice so far. Zogby, who now focuses more on business targeting, sees a marked tendency toward tribalization in America: that is, Americans identify more and more with virtual tribes of like-minded people, and less and less with region, party or other traditional organizational structures [5]. Interestingly, he did not apply his tribalism analysis to the election, but he does see the relevance. One of the reasons he got the turnout effect more accurately than most pollsters is that he used, among other sampling frames, weekly Walmart shoppers.

Certainly many commentators have noted that the populace currently is increasingly divided into “bubbles,” wherein most people rely on news outlets that share a strong point of view, and they associate mostly with other people who share that point of view. This tendency goes along with a rapidly declining number of competitive congressional districts, as reapportionments tend to create mostly safe districts for one party or the other. This, in turn, means that representatives run more toward the “true believer” base voters, to make sure they win their primaries, and have little incentive to move toward the center as one would in a swing district. This also means that strong partisan views by voters get reinforced rather than broadened or moderated by their representatives.

This change in the structure of electorate is closely related to change in communications. Many social media sites have algorithms that direct to the user more content similar to what they have been viewing, reinforcing insularity and strong partisan views. Possibly most polls, still relying heavily on telephones and email, missed the shift in some parts of the populace to a reliance on social media, especially Twitter. There may also have been a decline in some voters’ willingness to respond to polls at all. Trump, a master salesman, used social media more effectively than Clinton, both for messaging and for analytics.

Previous articles in Analytics and OR/MS Today [6, 7] recounted President Obama’s success in using social media. Although Hillary Clinton and the Democrats learned much from Obama and others, the question of how well they used social media and related analytics will no doubt feed some lively discussions, both among political staffs and scholars, for the next few years.

Marshall McLuhan Called It

Looking again at the increasing insularity of large blocs of voters, the reinforcing effects various media and the decreasing importance of policy issues, it seems that this election also provided dramatic confirmation for one more analyst: Marshall McLuhan. As early as 1964, in his penetrating analysis of the 1960 election, he noted the ascendance of style over content because of the new visual medium (TV) and foretold the still-accelerating changes in politics [8]. People who listened to the Kennedy-Nixon debates on the radio thought Nixon won. People who watched on TV thought Kennedy won. McLuhan explained that Kennedy was visually appealing and harder to classify – and that advantage overwhelmed any disadvantage on specific issues.

This phenomenon helps to explain the voters’ indifference to issues and misstatements. McLuhan said that TV is “an extension of the sense of touch,” steering attention away from “linear” content. McLuhan asserted, “Anyone whose appearance strongly declares his role and status in life is wrong for TV.” Applying this insight to the 2016 campaign, we can see that Hillary Clinton’s appearance all but screamed, “corporate lawyer turned politician.” As McLuhan said of Nixon, the ease of classifying Hillary left the viewer vaguely uneasy about her without being able to articulate why. Donald Trump, at first glance seemingly too “hot” and harsh for TV, did keep the viewer guessing about who he was and what he might say next – and that promotes visual engagement, the key to success. He was not a reality TV star by coincidence, nor was that experience irrelevant to the campaign. So once again the American people awoke the morning after the election wondering who the president-elect really was – a predictable consequence of McLuhan’s theories [9].

McLuhan also predicted that the emergence of new media would change the structure of society (“The Medium Is the Message”) and produce a form of tribalism on a global scale – the “Global Village,” as he called it. Now we see it. The next question is what to do with it. Perhaps a new coalition-building approach, building on the emerging tribal structure, will prove successful. Or perhaps personal visual appeal will prevail over content, and our future leaders will be TV personalities first, with experience counting less, if at all. Or there may be a reaction against the current situation, producing a renewed emphasis on proven political experience. The 13 Keys model still implies that successful governance determines who wins the presidency, notwithstanding all the other phenomena. Who knows? It seems highly likely that those political leaders, opinion leaders and analysts who first figure it out will be the ones who will succeed in the future.

Doug Samuelson (samuelsondoug@yahoo.com), a frequent contributor to OR/MS Today, is president of InfoLogix, Inc., a consulting company in Annandale, Va. Samuelson worked as a paid campaign staffer in a U.S. Senate campaign in Nevada in 1970, as a county coordinator in a gubernatorial campaign and targeting analyst for a local campaign in California in 1974, and as a Federal Civil Service policy analyst from 1975 to 1982. More recently, Samuelson worked for the Election Science Institute on the analysis of discrepancies between the exit polls and the actual vote in Ohio in 2004 and on the trials of new voting machines in 2006.

References

  1. Allan J. Lichtman, 1988, “The 13 Keys to the Presidency”; 2016 ed., “Predicting the Next President: The Keys to the White House,” Rowman and Littlefield, Lanham, Md.,
  2. Douglas A. Samuelson, 2016, “Politics & Analytics: Who Holds the Keys to the White House?” Analytics, July-August.
  3. Douglas A. Samuelson, 1996, “Unlocking the Door to the White House,” OR/MS Today, October.
  4. John Zogby, 2016, “Election 2016: How Did We Get to This?” webcast, Center for Strategic and International Studies, www.csis.org, Oct. 19.
  5. John Zogby, 2016, “We Are Many, We Are One: Neo-Tribes and Tribal Analytics in 21st Century America,” John Zogby, New Hartford, N.Y.
  6. Douglas A. Samuelson, 2013, “Analytics: Key to Obama’s Victory,” OR/MS Today, February.
  7. Douglas A. Samuelson, 2014, “Analytics Penetrates Deeper into Politics,” OR/MS Today, October.
  8. Marshall McLuhan,1964; revised, critical edition, 2003, “Understanding Media: The Extensions of Man,” Gingko Press, Corte Madera, Calif.
  9. Douglas A. Samuelson, 2008, “Marshall McLuhan’s Parable,” OR/MS Today, December.