Trying to forecast a presidential election - by the numbers
Boston — You might call it a presidential election ciphering mechanism. ... Or just another of those mathematical formulas into which you plug a set of statistical variables and - voila - out comes the probable next president of the United States, complete with percent of popular vote.
Whatever you call it, Michael Lewis-Beck and Tom Rice - two among a handful of political scientists in the United States who are investigating models to predict the outcome of presidential and congressional elections - have developed one such formula.
They say their ''multiple regression model'' has correctly predicted the winner in seven of the nine presidential campaigns since World War II, using data available six months before the elections.
''We realized that the other work that had been done on forecasting elections faced a main problem in that they didn't forecast far enough in advance to be a real forecast,'' says Dr. Lewis-Beck, a political science professor at the University of Iowa.
He and Mr. Rice, who teaches political science at the University of Vermont, combined a measure of economic performance (second-quarter election-year growth rate in real per capita GNP) and a measure of presidential popularity (the May Gallup poll's presidential approval rating) into a single model. Barring unforeseen events, it should be able to predict the outcome of presidential elections to within a 2.5 percent margin of error.
When tested against past postwar campaigns, the only two the model hasn't accurately called were very close races - Dewey-Truman in 1948 and Nixon-Kennedy in 1960.
''If the models are right it looks like a good year for the Republicans,'' Lewis-Beck says. His model predicts Reagan winning reelection with 54 percent of the popular vote. A similar model of the congressional races predicts the Republicans will pick up 10 seats in the House of Representatives.
''People are still somewhat leery of these models, they want to see them work ,'' says Rice, a former student of Lewis-Beck. ''I think it is only a matter of time before we see models becoming part of campaigns and being publicized widely before elections.''
Some in the political science community question the utility and significance of such predictive models in a world of proliferating polls, polling techniques, and political analysis. Everett Ladd, executive director of the Roper Center for Public Opinion Research sees the development of such models as part of ''the reckless interest of quantifying things that at this stage can't be quantified.''
He notes that a presidential election is a ''complex interaction of many things'' including national attitudes, economic factors, party identification, and party loyalty.
He adds, ''I don't know why one would want to restrict the range of applicable material.''
Michael Traugott of the University of Michigan's Center for Political Studies , says he is generally in favor of the use of models. ''It is important for us to search out patterns and to discern irregularities in the world around us,'' he says.
But he adds that beyond predicting a winner and a vote percentage, the Lewis-Beck and Rice model ''has no ability at all to offer any explanation for individual or group behavior in the context of a single election.'' This is a point the two men readily acknowledge, saying their primary aim in designing the model was to forecast, not to explain.
Stanley Kelley, a political scientist at Princeton University, says the model is ''interesting'' because ''it suggests that economic conditions are very important to the presidential vote.'' He says that such models can help political scientists not just predict the outcome of an election but to determine why people vote the way they do.
But there is more than academic interest in forecasting the outcome of elections. The key word is planning: Foreign countries and businesses, legislators and government bureaucrats in the US, and American businesses would all be in a position to use accurate election forecasts to plan for the future.
Lewis-Beck sees a parallel between his work and that of weather forecasters: ''There are parallels in that election forecasts and weather forecasts both are scientific enterprises. Weathermen look at meteorological data and try to tell what will happen tomorrow. We are trying to tell a little bit farther down the road. They are both scientific attempts in terms of looking at relevant past behavior and asking what can this tell us about the future.''
Techniques for forecastering elections have probably been around for as long as there have been elections.
One piece of lore has it that the taller presidential candidate always wins. But in 1976, a 5-foot 10-inch Jimmy Carter defeated a 6-foot 1-inch Gerald Ford. The theory - modified from always to usually - got back on track in 1980, when Carter was lost to a 6-foot 1-inch Ronald Reagan. (For the record Walter Mondale is 5-feet 11-inches).
Another time-tested presidential election forecasting method is the World Series test: If an American League team wins the series in an election year the Republicans will be in the White House. If the National League wins, the Democrats win. Since 1952, this method has proved accurate in seven of the last eight elections. (Perhaps Ring Lardner was right when he called it the ''World Serious.'') Lewis-Beck says it is merely a ''statistical coincidence.''
Another method is the Dow Jones industrial average test: If the Dow is higher on the first Monday of November than it was in January 1984, Reagan will win. If it is lower, Walter Mondale will be president. Since 1900, this method has proved accurate in all but five cases.