January 15, 2001: Bush's Chances in 2004
Now that George W. Bush has been sworn in and the 2004 presidential election has not yet started, we can pause and ask how well the 2000 election was predicted. The results are mixed at best. Gore got 50.3 percent of the two party vote. The polls even in the last few days before the election had Gore getting about 48 percent of the two party vote. It was a rare poll that had Gore ahead of Bush. On the other hand, the predictions of political science modelers had Gore way ahead. The predictions in September 2000 at the American Political Science Association meetings in Washington, D.C. ranged from 53 to 60 percent of the two party vote for Gore.

With many people making predictions, there is likely to be at least one who does well. This time it is my turn. My presidential vote equation predicted that Gore would get 50.8 percent of the two party vote, which is an error of only half a percentage point. The equation is based heavily on the use of economic variables to explain votes for president, and many have asked me why with the economy so good was not the equation predicting a larger vote share for Gore? The answer is interesting and also says something about President Bush's chances in 2004.

The answer is that there are incumbency effects aside from economic effects that matter. First, there is a historical bias in favor of the Republicans, at least over the sample period upon which the equation is based: 1916-1996. Other things being equal (the economy and the other incumbency variables), the Republicans are predicted to get more votes than the Democrats. Second, there is a duration effect. If a party has been in power at least twice in a row, it has a harder time getting reelected. Other things being equal, voters want a change. Gore had the duration effect working against him, and this along with the fact that he was a Democrat meant that he needed a strong economy to give him a vote share greater than 50 percent. According to the equation, the economy was just strong enough to have this happen (a predicted vote share of 50.8 percent).

The other incumbency effect that is important is if the President is running for reelection. This has a positive effect on votes for the President. Gore, of course, did not have this advantage.

The table below shows the elections over which the equation is estimated, and you can get some insight into Gore's situation by looking at the election of 1968. Humphrey, the Democrat, had the duration effect working against him. Despite the strong economy, he got only 49.6 percent of the vote. The other case in which the Democratic candidate had the duration effect working against him and was not the President was the election of 1920. In this case the economy was terrible and Cox, the Democratic, lost in a landslide to Harding, receiving only 36.1 percent of the vote.

The economic variables that are used in the equation are listed in the table. The growth rate is the growth rate of real per capita GDP in the first three quarters of the election year; the inflation rate is the inflation rate over all but the last quarter of the presidential term; and the good news variable is the number of quarters of the presidential term (except the last) in which the per capita growth rate exceeded 3.2 percent. (All growth rates are at annual rates. Deflation, which occurred in 1924, 1928, and 1932, is listed as a positive value in the table.) The weights that the equation estimates for these variables are .7 for the growth rate, -.7 for the inflation rate, and .9 for the good news variable. Given knowledge of the economy, one can use these weights along with the incumbency weights to predict the vote share. These predictions are presented in the last column of the table. They are based on the weights estimated prior to the 2000 election.

The predicted values are on average quite close to the actual values. The largest error was in 1992, where President G. Bush got 46.5 percent and was predicted to get 50.9 percent. One other election was predicted wrong as to the winner, 1960, where Nixon got 49.9 percent and was predicted to get 51.1 percent. The prediction error was, however, quite small; the prediction just happened to be on the wrong side of 50 percent. (In this business, equations are judged by the size of their errors, not on whether the winners are predicted correctly.)

It is important to realize that in order to use the vote equation for prediction purposes, one first needs a prediction of the economy. The vote predictions in the table are based on actual economic values, not some predicted values. If one is interested in judging the equation as an explanation of voting behavior, which is my main interest, the actual economic values should be used. It so happened for the 2000 election that my economic forecast at the end of April 2000 was close to what actually happened. Using this economic forecast and the vote equation, I was predicting by the end of April 2000 a Democratic vote share of 50.8 percent. This prediction six months before the election was thus quite accurate---and it was made even before the candidates were known!

It is also important to realize that the predictions in the table are based on weights estimated using observations through the 1996 election. Except for the 2000 prediction, the predictions are thus not predictions that could have been made in practice even if the economy were known. One test that has been made of the equation is to estimate it using only observations through 1960 and then see how well this version predicts the elections from 1964 on. Except for the 1992 election, the equation does quite well in this test. (The details are on the above mentioned website.)

Let us return now to the incumbency effects. The best possible incumbency situation is for a Republican President to run for reelection when the Republicans have only been in power for one term (no duration effect). There are 4 cases in the table in which this has been true. In 1984 President Reagan beat Mondale with 59.2 percent of the vote; in 1972 President Nixon beat McGovern with 61.8 percent of the vote; in 1956 President Eisenhower beat Stevenson with 57.8 percent of the vote; and in 1924 President Coolidge beat Davis with 58.2 percent of the vote. These are very large victories, larger than would be predicted solely on the basis of the economic variables.

If President Bush runs for reelection in 2004, he is in this best possible incumbency situation. The bottom of the table shows that if the economy is so-so (a growth rate of 2.5 percent, an inflation rate of 2.5 percent, and no good news quarters), President Bush is predicted to get 55.2 percent of the vote. The economy has to be quite bad before President Bush is predicted to lose. The table shows one possible case, where the growth rate is -2.0 percent, the inflation rate is 6.0 percent, and there are no good news quarters. For these values the predicted vote for President Bush is 49.6 percent. Note that both of these predictions are based on no good news quarters. Since the good news variable has a weight of .9, any good news quarter that President Bush picks up increases his vote share by .9, thus making the predictions even stronger for him.

One must always be cautious using explanations of past behavior to predict the future. In the present case it may be that even though the equation looks like it has done well in explaining the elections since 1916, it may in fact be a poor approximation and only look good because I have tried thousands of versions to come up with one that explains well. It may also be that there will be a sea change in voting behavior beginning with the 2004 election. In this case even though the equation may have been good in explaining voting behavior through 2000, it may no longer be any good. It is never easy predicting human behavior. These caveats aside, the bottom line here is that if there is anything to the vote equation, President Bush will be hard to beat in 2004.


                Presidential Elections Since 1916

    Party                                 Actual                  Predicted
     in                                    Vote  Growth Infl. Good  Vote
    Power                                  Share  Rate  Rate  News  Share
1916  D  Pres. Wilson beat Hughes           51.7   2.2   4.3   3    50.7
1920  D  Cox lost to Harding                36.1 -11.5  16.5   5    38.9
1924  R  Pres. Coolidge beat Davis          58.2  -3.9   5.2  10    57.9
1928  R  Hoover beat Smith                  58.8   4.6   0.2   7    57.3
1932  R  Pres. Hoover lost to F. Roosevelt  40.8 -14.9   7.1   4    39.2
1936  D  Pres. F. Roosevelt beat Landon     62.5  11.9   2.4   9    64.3
1940  D  Pres. F. Roosevelt beat Willkie    55.0   3.7   0.0   8    56.0
1944  D  Pres. F. Roosevelt beat Dewey      53.8   4.1   5.7  14    52.9
1948  D  Pres. Truman beat Dewey            52.4   1.8   8.7   5    50.5
1952  D  Stevenson lost to Eisenhower       44.6   0.6   2.3   6    43.9
1956  R  Pres. Eisenhower beat Stevenson    57.8  -1.5   1.9   5    57.3
1960  R  Nixon lost to Kennedy              49.9   0.1   1.9   5    51.1
1964  D  Pres. Johnson beat Goldwater       61.3   5.1   1.2  10    61.2
1968  D  Humphrey lost to Nixon             49.6   4.8   3.2   7    49.6
1972  R  Pres. Nixon beat McGovern          61.8   6.3   4.8   4    59.8
1976  R  Ford lost to Carter                48.9   3.7   7.7   4    48.6
1980  D  Pres. Carter lost to Reagan        44.7  -3.8   8.1   5    45.6
1984  R  Pres. Reagan beat Mondale          59.2   5.4   5.4   7    61.4
1988  R  G. Bush beat Dukakis               53.9   2.1   3.3   6    52.4
1992  R  Pres. G. Bush lost to Clinton      46.5   2.3   3.7   1    50.9
1996  D  Pres. Clinton beat Dole            54.7   2.9   2.3   3    52.7
2000  D  Gore lost to G.W. Bush             50.3   3.5   1.7   8    50.8
2004  R  Pres G.W. Bush beat ?                 ?   2.5   2.5   0    55.2
2004  R  Pres G.W. Bush lost to ?              ?  -2.0   6.0   0    49.6

Note: Vote share is share of two party vote.