Stuffed shirts vs. 'skins: an econometric critic tells how to predict the presidential race.

For nearly two hundred years, journalists and other pundits have sought a rule of thumb (or two) that might accurately predict the outcome of U.S. presidential elections. Since 2000, this search must be informed by two important events: First, the possible breakdown of the venerable and highly accurate Washington Redskins standard. For fifteen straight elections, the Redskins accurately predicted the outcome of U.S. presidential elections: A win in their last home game of the regular season prior to the election indicated that the incumbent party would retain the White House. Alas, this failed to be the case in 2000; in their last home game, the Redskins crushed the hapless St. Louis Cardinals, and had the predictive power of the Redskins standard stood, Al Gore should have won the election.

But, now of course we have a complication: There are many who still say that Al Gore actually did win this election, or at least did so in the normal sense of "win." Thus, even though it is George W. Bush who sits in the White House, perhaps the predictive power of the Redskins standard was never meant to handle election irregularities in the state of Florida and weird Supreme Court decisions whereby a loser emerges as the winner. So, maybe, the Redskins standard indeed is a perfect role of thumb. On this, we might note that in the 2004 season, in their last home game before the election, the hapless Red skins will be hosting the mighty Green Bay Packers. Is the fate of George Bush in the hands of returning Redskins Coach Joe Gibbs? And even so, does this likely defeat for the Washington football club foretell the election of John Kerry?

The second event for election watchers was a move toward economic determinism, with Clinton's campaign motto "It's the economy, stupid" as emblem. The academic experts picked up on this trend. Yale University economist Ray C. Fair published a book in 2002 that uses econometric techniques to investigate all sorts of things, including but not limited to outcomes of U.S. presidential elections. (1) Fair's methodologies are claimed to have very general applicability; indeed, he also investigates whether people are likely to have extramarital affairs. (On this, he concludes "yes".) On presidential election outcomes, he concludes that, statistically speaking, only two factors really matter: (1) which party is in power: the incumbent is likely to be returned to off]cc unless his party was in charge of the Oval Office for "too long" (i.e., three or more terms); and (2) whether the economy is improving or deteriorating just prior to the election. The latter factor would seem to be more heavily weighted than the former, such that, if one indeed is looking for a rule of thumb, Clinton's aphorism is a good guide.

Fair indeed has predicted since publication of his book that, given the U.S. economy is likely to continue to be on an upward tick for the remainder of this year and that Bush the Junior has been in office for only one term, the odds are overwhelming that Bush will be returned to the White House. Could it be that we need not wait to see the likely outcome of the 'Skins game? We should instead bet on an econometric model?

Let's start by noting that there are some problems associated with hypothesis testing via econometric methods. Ideally, to do statistical testing, one needs for everything to be held constant except for the causal variables being tested. For a physicist, this is relatively easy: one devises an experiment where everything but that which is to be tested is controlled so that all else indeed is, as best as one can do, held equal. Even then, to make sure that random errors of control or measurement do not contaminate his or her results, the physicist runs the experiment multiple times. Alas, even so, the outcome can be inconclusive although more often than not, the physicist does return results that yield something like the truth.

In the social sciences, however, life is much harder because it is impossible to test most hypotheses via controlled experiments. Hence one must look toward so-called "natural experiments." One must test hypotheses by analyzing data for events that have actually happened while trying to "control" for factors outside the hypothesis that might affect the outcome, using statistical methods. Econometricians have developed some quite potent tools for doing so; nonetheless, there is still a lot still left to be desired. For example, econometric results often differ from one study to another of the same hypothesis. Think of the active disputes in the globalization literature about the benefits or costs of financial liberalization or acceptance of foreign direct investment in emerging markets--the issues are far from settled. but not due to lack of empirical research.

Such inherent difficulties of empirical analysis of course are no deterrent to those who have great faith in the veracity of a hypothesis when claiming that events prove the hypothesis right. Such persons of great faith usually simply assert that one observed "'success'" of the hypothesis proves it correct. This is particularly tree of supply-side enthusiasts, the editorialists of the Wall Street Journal and Vice President Dick Cheney prominent among them. They claim that the "Reagan boom" of the late 1980s demonstrates that reductions in marginal income tax rates stimulate economic growth. There indeed was a long rise in U.S. economic activity that began toward the end of 1982 and ended in 1990, and this was preceded by cuts in marginal income tax rates. But there was another long rise in economic activity that began early in 1991 and ended in 2001, where this second upturn was preceded by increases in marginal taxes. Moreover, growth was more robust and longer lasting during the second upturn than during the first. Which is it, then, that tax increases or tax cuts stimulate economic growth'? The simple fact is that it is hard to find robust econometric support for either contention--and certainly no dependable rule of thumb for this issue.

Given this, let's take our own look at the data, using the very robust technique of "ocular regression analysis." Ocular regression analysis is done by placing the relevant data in a format that the eye can absorb and then using the eye to see what is to be seen. No one should laugh too hard about this: a noted professor of statistics at Harvard University told me, when I was a graduate student there, that few hypothesized relationships will prove to be robust if they fail the ocular test.

First, I assume that what moves voters the most is what happens to the economy in the third quarter of year, the most recent quarter prior to the election. The reasoning is that Americans do not have particularly long memories, such that their behavior is influenced mostly by what is going on right now rather than by what might happen in the future, or what has happened in the past. Then, all else being equal, if the third quarter economics results are on an up-tick, the incumbent, or the candidate of the incumbent's party if the incumbent is not up for reelection, will likely win; if not, the incumbent or his party's candidate will likely be thrown out. Second, I assume that direction of economic activity is what counts, rather than the level. Level-based economic predictors of elections--meaning things like "misery indices" that combine the level of unemployment and inflation--have proven to be lousy.

Table 1 presents the election results, and the third quarter economic results, lot each election since 1948. In the "election result" column, a plus (+) indicates that the incumbent or the incumbent party's candidate indeed was returned to office, while a minus (-) indicates that the incumbent lost. The "third quarter GDP" column indicates whether real GDP was rising (+) or falling (-) in the third quarter of the relevant year relative to GDP in the second quarter.

If the "it's the economy, stupid" hypothesis were correct, one would expect that for each election year where a plus (minus) appears in the "Third Quarter" column, a plus (minus) would appear in the "Election Result" column as well. This is tree for a majority (eight out of fourteen) of election years, but hardly the kind of a large majority that conveys a mandate for this role of thumb, of course. In many election years, rising third quarter GDP would not have accurately predicted the election outcome.

But those false positive and negative results do not by themselves render the hypothesis invalid, especially if we can argue that other factors weighed heavily enough hl those years so as to dominate the economy as the decisive factor. As for those other factors, in 1952, 1960, 1968, 1976, and 1992 (five of the six years in which counter-expected results obtained), the incumbent party had been in the White House for two or more consecutive terms. So we didn't need any fancy econometrics to get a two-part rule that predicts the election outcome in thirteen out of fourteen cases!

Some might argue that "it's the economy, stupid" all right, but that expansion or contraction of GDP is not the right measure of how the electorate views the economy. Rather, it is the expansion or contraction of employment. Table 2 indicates election results and changes in unemployment, where a minus (plus) indicates that unemployment fell (rose) during the year previous to the election. A zero (0) in the column indicates no change. Again, the "economy" hypothesis leads to the expectation that, for each election year, the entries in the "Election Result" should be of the same sign as those in the "Third Quarter Unemployment" column. These signs match, that is predict election outcomes properly, in eight out of the fourteen years again, but, interestingly, the specific years where they differ are not the same as in Table 1. The years in which Tables 1 and 2 differ are 1948, 1956, 1992, and 2000. By Table 2, the years in which change in unemployment does not predict the election are not necessarily those where the incumbent party has held the White House for more than two terms. This makes a certain amount of sense because we know that the unemployed do not vote in large numbers, and people vote their own not others' pocketbooks in the United States.

A levels-based variant of the "it's the economy, stupid" is that a particularly bad economy results in the incumbent being tossed from office after just one term. Does this hold up? In fact, three presidents since 1948 have served but one term (Gerald Ford, 1973-76; Jimmy Carter, 1977-81: George H. W. Bush, 1989-92). Bush the Senior followed a president of his own party, such that Bush's term was the third term of the party incumbent in the White House, which fits our secondary pale. In any event, in 1992, GDP growth was positive although the public might not yet have known this. In 1976, when Ford was forced out, unemployment was steady and GDP was growing, albeit slowly, so that does not portend well for the levels argument either. In 1980, there was both a rise in unemployment and a fall in GDP, and so poor Jimmy Carter might have been the main case where "it's especially the economy, stupid!" But, of course, Carter was also negatively affected by the Iran hostage crisis, and that's one case.

So where to come out on all of this? The simple ocular analysis does support the economic changes predictor (and the results of Fair's heavy duty econometric analysis) to a point. But the number of observations is not very large and I am not convinced that adding more data points in order to improve statistical significance by going back and including earlier elections would particularly helpful to resolve the difficulties posed by this small number (which is that when the number of natural experiments is small, accurate statistical inference is difficult). After all, time does change society, and if you go back too far in time when doing econometric analysis, the "all else being equal" assumption will be substantially weakened. Moreover, as a purely predictive horse race, it is still a dead heat since the Washington Redskins standard (assuming Coach Gibbs does not work a miracle) is definitely predicting a different outcome for this presidential contest than the economic rule of thumb. Maybe our political pundits should be trying to get their Sunday morning TV appearances on "The NFL Today" rather than on the Washington-based talk shows.

Table 1 U.S. Election Results and
Third Quarter GDP Growth

Year    Election Result    Third Quarter GDP

1948           +                   +
1952           -                   +
1956           +                   -
1960           -                   +
1964           +                   +
1968           -                   +
1972           +                   +
1976           -                   +
1980           -                   -
1984           +                   +
1988           +                   +
1992           -                   +
1996           +                   +
2000           -                   -

Table 2 U.S. Election Results and
Movement of Unemployment

                             Unemployment
                           (- indicates rise
Year    Election Result     in unemployment

1948           +                   -
1952           -                   +
1956           +                   +
1960           -                   0
1964           +                   +
1968           -                   +
1972           +                   +
1976           -                   +
1980           -                   -
1984           +                   +
1988           +                   +
1992           -                   -
1996           +                   +
2000           -                   +

NOTE

(1.) Ray C. Fair, Predicting U.S. Presidential Elections and Other Things (Stanford, CA: Stanford University Press, 2002).

RELATED ARTICLE: The Redskins standard.

For fifteen straight elections, the Redskins accurately predicted the outcome of U.S. presidential elections: A win in their last home game of the regular season prior to the election indicated that the incumbent party would retain the White House.

In the 2004 season, in their last home game before the election, the hapless Redskins will be hosting the mighty Green Bay Packers. Is the fate of George Bush in the hands of returning Redskins Coach Joe Gibbs? And even so, does this likely defeat for the Washington football club foretell the election of John Kerry?

--E. Graham

Edward M. Graham is a Senior Fellow at the Institute for International Economics.