Gore Wins! (At least that's what the model says)

By Douglas A. Samuelson

At this writing (mid-September), the polls and the pundits say the presidential election is too close to call, a real horse race. To at least one analyst, however, the result was already pretty much a sure thing months ago: afor Vice President Gore. To another analyst, the big question is not how the election will turn out, but being the first again — to project state-by-state results correctly from early actual returns.

OR/MS Today readers may remember Allan J. Lichtman, professor of history at The American University in Washington, D.C. He was a consultant to the 1992 Gore campaign (among others), is a successful early predictor of the last four presidential elections, and authored the popular book, "The Keys to the White House" (1996), now updated and reissued to include the analysis of the 1996 election and predictions for 2000. His forecast of a Gore victory is on record at least as early as September 1999, when he stated, "There are five keys against the incumbent party at this time. If the Democrats have a serious contest for the nomination, they will lose; otherwise, they will win."

Lichtman is one of the analysts highlighted in the 1996 INFORMS-produced videotape on careers in operations research. Ironically, he had never identified himself as an OR analyst before he was invited to participate in making the tape; he describes himself as a quantitative historian. His Ph.D. is in history, from Harvard. The invitation resulted from an appearance at the Washington, D. C., chapter's banquet in June 1996. In October 1996, OR/MS Today profiled him, reporting on his prediction of the 1996 election at the banquet.

Jack Moshman, on the other hand, is not well known in the OR profession. He is a Fellow of the American Statistical Association and a seasoned consultant in the Washington, D.C., area. In 1960 he invented the method NBC used to make early projections of state-by-state results. For NBC, and later for ABC, he has been calling state-by-state results in presidential elections, and in some Senate and governors' races, ever since. He will be at it again on Election Night this year.

What both these analysts have in common is their pathbreaking uses of OR techniques to solve a complicated problem — and little appreciation, within the profession and outside it, of just what they accomplished.

Lichtman's predictions are based on 13 questions (see box), each with a "yes" or "no" answer. "Yes" answers favor the incumbent party. If five or fewer answers are "no," the incumbent party retains the presidency; if six or more are "no," the challenger wins.

The Thirteen Keys to the Presidency

- The incumbent party holds more seats in the U. S. House of Representatives after the midterm election than after the preceding midterm election.
- There is no serious contest for the incumbent-party nomination.
- The incumbent-party candidate is the current president.
- There is no significant third-party or independent candidacy.
- The economy is not in recession during the campaign.
- Real (constant-dollar) per capita economic growth during the term equals or exceeds mean growth for the preceding two terms.
- The administration has effected major policy changes during the term.
- There has been no major social unrest during the term.
- The incumbent administration is untainted by major scandal.
- There has been no major military or foreign policy failure during the term.
- There has been a major military or foreign policy success during the term.
- The incumbent-party candidate is charismatic or is a national hero.
- The challenger is not charismatic and is not a national hero. If six or more of these statements are false, the incumbent party loses.

For this election, Lichtman says the Democrats have lost Key 3 (the incumbent-party candidate is not the current president), Key 7 (there were no significant policy changes during this term), Key 9 (there was a scandal), Key 11 (there were no major military or foreign-policy successes), and Key 12 (the incumbent-party candidate is not very charismatic or a national hero.) Key 4, the significant third-party presence, does not fall because neither Pat Buchanan nor Ralph Nader seems likely to get five percent of the vote, the criterion for turning this key. Even if Buchanan did get five percent, the key would not fall, according to Lichtman's criteria, because Buchanan is a breakaway candidate from one of the major parties (Republican) rather than a true independent. Key 2, the contest for the incumbent-party nomination, falls if no candidate comes to the convention with two-thirds of the delegates — so Bill Bradley's challenge did not turn this key.

Not Just Reading Tea Leaves

Lichtman is quick to point out that his method is based on a solid statistical model that incorporates a test of competing theories of politics, and the prediction results validate some of the theories and contradict others. His method is based on a statistical pattern recognition algorithm for predicting earthquakes, implemented by Russian seismologist Volodia Keilis-Borok. In English-language terminology, the technique most closely resembles kernel discriminant function analysis. The highest plurality of the popular vote, not the electoral vote which actually decides the presidency, is the criterion — which means that in two elections, 1876 and 1888, the "winner" as defined in this method did not end up as president. Out of nearly 200 questions, which were all binary ("yes" or "no") variables, the algorithm picked those which displayed the greatest difference between the proportion of the time the variable was "yes" for years when the incumbent party won and the corresponding proportion for years when the challenging party won, using all U. S. elections from 1860 through 1976 as the training set. Interestingly, the single most powerful variable was the key that turned out to be critical in 2000 — the presence or absence of a significant contest for the incumbent-party nomination.

"Naturally, we thought the intensity of the fight for the challenging-party nomination would also be important, but it turned out not to be," Lichtman adds. "There's a big fight for the challenging-party nomination either when they're pretty sure they'll win, or when they're in horrible shape and they're all blaming each other. So that one didn't have much of an association with who won."

Licthman also points out that a number of other variables didn't have much effect: adverse reports on candidates' health, running mates and endorsements, among others. "The point," he asserts, "is that elections are less about campaigning than people like to believe, and more about governance. That doesn't mean one party could just stay home, do no campaigning at all, and stillif the keys were in its favor. It does mean, though, that the little ups and downs in the campaigns don't have all that much effect, no matter what the pundits claim. The people are sensible, and they decide based on how well the party in power has governed.

"There's some irony in the impeachment," he adds. "If Clinton had been removed, Gore would be running as the incumbent president, and the Republicans would have lost another key. It seems counter-intuitive, but what the theory says is that Gore would have gone into the campaign with more public support because they would already see him as a successful president, a known quantity."

Election Night Projections

Whatever factors end up determining the election, there is still another challenge on Election Night: being first to project correctly, as the actual votes come in, who will carry each state. Enter Jack Moshman, statistician and OR analyst, inventor of the method that has been most successful in this highly competitive endeavor. He ought to be good at it by now: he's been doing in since 1960.

"There was some earlier work," he explains. "Max Woodbury, at NYU, working for CBS, developed a nice probability model, based on a generating function, that gave a probability of a win. In 1956, very early in the evening, he got a probability of .997 that Eisenhower would win. He and the other analysts and the network people talked it over and decided that seemed too high, and concluded they must have entered the trend factor wrong. They Œcorrected' it and went out with a probability of .6. Ten minutes later, even with the Œcorrection,' the model was back at .997. I think that was the highest probability it could yield, as it was formulated. And, of course, Eisenhower won in a landslide. So people started thinking these models were worth something.

"In 1960," he continues, "NBC hired CEIR, the company I was with at the time, to do something with computers for Election Night. I worked with Max and pulled in John Tukey and Richard Scammon to develop a method that would be easier to implement. This was back when computers were large and expensive and slow, and much of Max's method had to be done by hand. So we focused on what John Tukey called Œswing-o-metric' precincts, the ones that most often reflected the way the state went. The absolute percentage was unimportant — the precinct could be one that went Republican every year. What mattered was how closely the swings from year to year in that precinct reflected the swing in the state total.

"The other critical thing was that the precincts we used had to have machine ballots, so we could get their results quickly. We would pick another election that seemed to have similar issues and circumstances, and use that as a starting basis for comparison. Then we would have a big meeting the Sunday before the election and put in the latest poll numbers as our Time Zero estimate, then update those numbers and track the differences as actual votes came in. That's essentially still what we do now, except that now we get county and sub-county totals electronically via the media pretty quickly. When we started, we had only statewide totals reported, so we had to rely on a nationwide collection of volunteers, recruited mostly by the League of Women Voters, to phone results in from the precincts we used." (This reporter was one of those volunteers in 1970.)

"In 1960 and 1964," he relates, "we predicted only nationwide electoral vote totals, not state by state. That helped us, because we called both California and Illinois wrong, but the errors more or less canceled each other. Usually big-city votes come in first and run Democratic, then late returns from rural areas run Republican. We didn't know that Cook County's chairman held his vote to make sure Kennedy got enough to win. And in California, we neglected the absentee ballots, which — as they usually do — went heavily Republican. As it turned out, we were the only network that called it for Kennedy early, and we stayed with him all night, so we skunked the competition. But we bit our nails as the returns from the West came in."

In 1964, Moshman moved to ABC, where he has been ever since — and will be again this year. The networks have economized over the years, first by sharing the collecting and transmission of returns from the precincts, and later running much the same analysis on a shared computer. The analysts' teams are still separate and competing, however, "massaging the data" somewhat differently. "We've had the best record of correct calls early," he asserts, "with maybe five states wrong in presidential, gubernatorial and Senate elections in the 32 years since we started doing state-by-state projections. And the networks have been moving more and more away from calling states before all their polling places close, so many times we sit for 45 minutes after we have a projection we're ready to release."

CBS, lagging in the race, introduced exit polls in 1980. Moshman is not coy about them: "I was in Europe on vacation when the Israeli networks called the Prime Minister election in 1996 for Peres," he recounts. "Aside from any personal political preferences, I was tickled when Netanyahu won and proved the exit polls wrong."

Ironically, as technology has made the job easier in many ways, it has made one critical step much slower. "Many voting machines now generate punch cards or magnetic disks that get trucked to one central location and counted there," he explains. "So it's becoming harder to get those precinct results early.

"But now the networks are starting their coverage later, too, keeping commercial programming on until 9 p.m. or even 10 p.m. instead of starting continuous coverage at 7 p.m.. Every year, the competition is different, but it's always there."

And so is Jack Moshman. Will the "Thirteen Keys" prediction be right again? Moshman will be among the first to know and tell the rest of us.


- Lichtman, Allan J., 2000, "The Keys to the White House, 2000," Madison Books, Lanham, Md.

- Lichtman, Allan J., 1996, "The Keys to the White House, 1996," Madison Books, Lanham, Md.

- Lichtman, Allan J., and DeCell, Kenneth, 1990, "The Thirteen Keys to the Presidency," Madison Books, Lanham, Md.

- Lichtman, A. J., and Keilis-Borok, V. I., 1981, "Pattern Recognition Applied to Presidential Elections in the United States, 1860-1980: Role of Integral Social, Economic and Political Traits," Proceedings of the National Academy of Science, Vol. 78, No. 11, pp. 7230-7234.

- Samuelson, Doug, 1996, "Unlocking the Door to the White House," OR/MS Today, pp. 28-30.

Doug Samuelson, a frequent contributor to OR/MS Today, is president of InfoLogix, Inc., a consulting company in Annandale, Va. He worked as a campaign staffer in a U. S. Senate campaign in Nevada in 1970 and as a county coordinator in a gubernatorial campaign in California in 1974.