| FAIRMODEL | Home | Presidential Vote Equation |
With 98 percent of the precincts reporting, President Clinton has received 54.6 percent of the two-party vote (V = .546). Based on the latest economic data, released last week, V was predicted to be .510 by the vote equation. (The latest data has g3 = 2.0, p15 = 2.3, and n = 3.) The equation thus made an error of .036, which is slightly less than twice the estimated standard error of the equation, .019. The equation was right in predicting that the election would be much closer than the polls were suggesting even less than a week ago, but the election was not as close as the equation predicted it would be.
It is important to note that the above vote prediction is based on preliminary economic data. As the economic data are revised, the predicted vote changes. One key revision will be the growth rate in the third quarter of 1995. Using the current estimates, the per capita growth rate in this quarter is 2.8 percent (annual rate), which makes the quarter not quite a good news quarter, since a good news quarter requires 2.9 percent. If future revisions increase this growth rate to 2.9 percent or above, then n will be 4 instead of 3 and the predicted value of V will rise, other things being equal, by .0099 to .520. The error then would be only .026. So the final judgment on how the equation did for 1996 will not be in until more economic data are in. Note that by using the latest data to evaluate the equation, it is implicitly assumed that the new data better approximate the economic conditions known to the voters at election time than do the old data. If voters look at the economic conditions around them and not at the numbers themselves, which is assumed here, then one should always attempt to use the most accurate data to evaluate the equation.
The following was put on the site after the August 2, 1996, economic forecast from the US model, and it has been retained for reference purposes.
The vote equation originally presented in Ray C. Fair, "The Effect of Economic Events on Votes for President," The Review of Economics and Statistics, 1978, pdf file (1,276KB), has been updated through the 1992 election. The update is discussed in two papers. The first, which is in the June 1996 issue of Political Behavior, is Ray C. Fair, "The Effects of Economic Events on Votes for President: 1992 Update," pdf file (982KB), pdf file (1,674KB). The second, which is in the summer 1996 issue of the Journal of Economic Perspectives, is Ray C. Fair, "Econometrics and Presidential Elections," pdf file (714KB). Both papers cover the same issues. The first is more complete, and the second is somewhat easier to read.
Given that President Clinton is running for reelection, the equation for
1996 can
be written as:
V = .4859 + .0065*g3 - .0083*p15 + .0099*n
where V is the Democratic share of the two-party vote,
g3 is the annualized growth rate of real, per capita GDP
in the first
three quarters of 1996, p15 is the annualized growth rate of the GDP
price index in the first 15 quarters of the Clinton Administration, and
n is the number of quarters out of the first 15 quarters of the Clinton
Administration in which the annualized growth rate of real, per capita
GDP exceeds 2.9 percent. n is called the "good news" variable.
Given values for g3, p15, and n, the equation can be used to predict V. A prediction of V was made in the Journal of Economic Perspectives paper mentioned above. The economic values that were used in the paper for this prediction were 2.1 for g3, 3.0 for p15, and 2 for n. This gives a prediction of V of .495, which is a narrow Republican victory (assuming the electoral college vote is consistent with the popular vote). It was stated in the paper that the economic values were from the US model forecast made May 3, 1996. This is true for the forecasts for g3 and n, but not for p15. The May 3, 1996, forecast from the US model for p15 was 2.3, not 3.0. When 2.3 is used in place of 3.0 for p15, the predicted value of V changes from .495 to .5003, which is a very narrow Democratic victory (assuming electoral college consistency).
Between May and now the actual data for 1996:2 have become available. The main change in economic assumptions in light of the new data is that 1996:2 is a good news quarter, whereas it was not predicted to be so in the May 3, 1996, forecast. The growth rate for 1996:2 was forecast to be 3.7 percent (2.7 percent per capita), which is not quite a good news value, whereas the actual growth rate was 4.2 percent (3.2 percent per capita), which is a good news value. So n is now at least 3. The current US model forecasts (August 2, 1996) imply values of 2.3 for g3, 2.3 for p15, and 3 for n. (The predicted growth rate for 1996:3 is 2.9 percent--1.9 percent per capita--so 1996:3 is not predicted to be a good news quarter.) Using these values, the predicted value for V is .5115, which is a narrow Democratic victory (again, assuming electoral college consistency).
Although the predicted value of V between May and now has increased by .0115, the main message from the equation is the same, namely that the election is predicted to be very close. If 1996:3 turns out to be roughly as forecast (an OK but not great growth-rate quarter) and if the election is close, the vote equation will have done well regardless of who wins. If, on the other hand, 1996:3 turns out to be roughly as forecast and the election is not close, the vote equation will have done poorly regardless of who wins. In particular, if President Clinton wins with nearly 60 percent of the two-party vote, as many of the polls are currently showing, the equation will have done poorly even though it got the winner right.
To summarize, given the current economic forecasts, the vote equation says that the election is too close to call with any confidence. If, of course, 1996:3 turns out to be a much stronger quarter than currently forecast, the predicted victory margin for the Democrats increases.
John S. Irons at M.I.T. has put together a neat Web page that allows you to enter your own economic assumptions and predict V. You can also do historical analyses. You can go to the site by clicking the following: http://www.mit.edu:8001/people/irons/myjava/ecalc.html