The specification of my vote equation has not been changed since the update in November 1994, following the 1992 election. In this sense no data mining has taken place in the last 12 years. The equation has simply been reestimated each four-year interval using the latest economic data and the results of the previous election. This paper presents the estimates through the 2004 election and gives a prediction for 2008.
The theory behind the equation is presented in Fair (1978), which is the original paper. The previous updates are in Fair ( 1982, 1988, 1990, 1996, 1998, 2002). The easiest paper to read regarding the changes that were made to the equation between the original version and the current version (i.e., the 1994 specification) is Econometrics and Presidential Elections, The Journal of Economic Perspectives, Summer 1996, 89-102. A very non technical discussion is in Predicting Presidential Elections and Other Things . The present paper is not self contained. It assumes that at least one of the non technical discussions has been read.
The data are discussed in Section 2; the estimates are presented in Section 3; and predictions for 2008 are presented in Section 4. Appendix C contains a complete description of how the data were collected and a listing of all the data. It allows one to duplicate the results in this paper if desired. Appendix A will likely not be of interest to most readers. It contains 12 different estimates of the vote equation, each corresponding to a small specification change. These results show how sensitive the equation is to small changes---in other words, how robust the estimates are. If data mining is a problem, there is likely to be lack of robustness. It will be seen that with a few exceptions the estimates are quite robust. Appendix B discusses the interpretation of the coefficient of the WAR variable.
Since the last update the Department of Commerce has revised some of the national income and product data and has changed the base year to 2000. The Department of Commerce data available as of October 27, 2006, have been used. Data prior to 1929 have been obtained, as before, from Balke and Gordon (1986). Appendix C discusses the splicing of the Balke and Gordon data to the Department of Commerce data.
The notation for the variables is as follows:
The equation has VOTE on the left hand side and the other variables plus a constant term on the right hand side. It is linear in coefficients. The estimation period begins with the 1916 election. The equation is estimated by ordinary least squares.
Column 1 of Table 1 below presents the estimates using the entire 1916-2004 estimation period (23 observations). This is the equation that is used for the predictions in the next section. It is the "updated" or "current" equation. Column 2 presents the estimates using the 1916-2000 period. The estimates in columns 1 and 2 are based on the revised economic data. Column 3 presents for comparison purposes the estimates using the 1916-2000 period that are based on the economic data available at the time of the November 1, 2002, update. The estimates in column 3 are the estimates used for the previous update---the previous "current" equation. Comparing columns 2 and 3 shows how much difference the data revisions made to the coefficient estimates. Finally, column 4 presents the estimates using the revised data but only the 1916-1960 estimation period (12 observations).
The predicted values and errors are presented at the end of the table for each of the four sets of estimates. All the predicted values are based on the actual (i.e., latest revised) values of the explanatory variables. This means for column 3 that the coefficient estimates have been obtained from a slightly different data set (the pre revised data) than is used for the predictions for this equation. For columns 2 and 3 the predicted value for 2004 is outside the estimation period. For column 4 the predicted values for 1964 and beyond are outside the estimation period, where the predicted value for 2004 is outside by 44 years!
__________________________________________________________________________
Table 1
Four Sets of Estimates of the Vote Equation
1 2 3 4
Current Previous
Sample: 1916-2004 1916-2000 1916-2000 1916-1960
GROWTH 0.680 0.680 0.691 0.801
( 6.14) ( 6.57) ( 6.72) ( 8.18)
INFLATION -0.657 -0.800 -0.775 -0.486
(-2.26) (-2.82) (-2.71) (-1.42)
GOODNEWS 1.075 0.857 0.837 0.698
( 4.31) ( 3.26) ( 3.12) ( 3.06)
PERSON 3.30 3.29 3.25 5.16
( 2.34) ( 2.50) ( 2.50) ( 4.34)
DURATION -3.33 -3.89 -3.63 -2.33
(-2.75) (-3.31) (-3.04) (-2.75)
PARTY -2.68 -2.82 -2.71 -3.67
(-4.27) (-4.77) (-4.65) (-6.65)
WAR 5.61 4.15 3.85 4.01
( 2.09) ( 1.57) ( 1.46) ( 1.82)
INTERCEPT 47.26 49.58 49.61 47.52
(18.62) (18.31) (18.08) (23.09)
SE 2.54 2.38 2.37 1.41
2
R 0.905 0.923 0.923 0.988
No. obs. 23 22 22 12
V V V V V
Year act. pred. error pred. error pred. error pred. error
1916 51.7 | 49.8 -1.9 | 50.7 -0.9 | 50.9 -0.8 | 50.8 -0.9 |
1920 36.1 | 39.1 3.0 | 39.2 3.1 | 39.2 3.1 | 36.3 0.2 |
1924 58.2 | 58.0 -0.3 | 57.5 -0.7 | 57.3 -1.0 | 57.7 -0.5 |
1928 58.8 | 57.2 -1.7 | 57.5 -1.3 | 57.6 -1.2 | 57.4 -1.5 |
1932 40.8 | 38.8 -2.1 | 38.6 -2.2 | 38.8 -2.1 | 41.1 0.3 |
1936 62.5 | 63.9 1.5 | 63.8 1.3 | 63.9 1.4 | 63.5 1.0 |
1940 55.0 | 55.8 0.8 | 55.6 0.6 | 55.9 0.9 | 55.3 0.3 |
1944 53.8 | 52.2 -1.5 | 52.2 -1.5 | 52.4 -1.4 | 53.5 -0.2 |
1948 52.4 | 50.9 -1.4 | 50.8 -1.6 | 51.0 -1.3 | 52.4 0.0 |
1952 44.6 | 45.2 0.6 | 44.5 -0.1 | 45.1 0.5 | 44.1 -0.5 |
1956 57.8 | 56.4 -1.4 | 57.4 -0.3 | 57.3 -0.5 | 57.7 0.0 |
1960 49.9 | 50.9 1.0 | 51.5 1.6 | 51.6 1.7 | 51.7 1.8 |
1964 61.3 | 61.3 -0.1 | 61.1 -0.3 | 61.1 -0.3 | 59.5 -1.9 |
1968 49.6 | 50.1 0.5 | 49.8 0.2 | 50.2 0.6 | 48.9 -0.7 |
1972 61.8 | 58.4 -3.4 | 59.3 -2.5 | 59.3 -2.5 | 61.5 -0.3 |
1976 48.9 | 49.5 0.6 | 49.2 0.3 | 49.6 0.6 | 51.6 2.7 |
1980 44.7 | 45.7 1.0 | 45.6 0.9 | 45.8 1.1 | 45.8 1.1 |
1984 59.2 | 62.1 2.9 | 62.0 2.9 | 62.0 2.8 | 63.7 4.6 |
1988 53.9 | 50.5 -3.4 | 51.1 -2.8 | 51.3 -2.6 | 52.0 -1.9 |
1992 46.5 | 50.9 4.3 | 51.7 5.2 | 52.0 5.5 | 55.4 8.8 |
1996 54.7 | 53.0 -1.8 | 53.9 -0.8 | 54.1 -0.7 | 53.3 -1.4 |
2000 50.3 | 49.6 -0.6 | 49.3 -1.0 | 49.6 -0.7 | 47.3 -3.0 |
2004 51.2 | 54.6 3.4 | 56.5 5.3 | 56.5 5.2 | 58.1 6.8 |
2008 ? | 46.5 | 47.7 | 48.0 | 49.2 |
_____________________________________________________________________________
Notes: error may differ from pred. minus act. because of rounding.
The predictions for 2008 use the forecasts of the economic
variables from the US model discussed in Section 4.
Columns 2 and 3 show that the data revisions had only small effects on the coefficient estimates. The (outside sample) prediction for 2004 is 56.5 percent in both cases, for an error of 5.3 percentage points. This error is a little over twice the estimated standard error, and it is about the same size as the error for the 1992 election.
Columns 1 and 2 show that adding the observation for 2004 had only modest effects on the coefficient estimates. The INFLATION coefficient estimate decreased in absolute value from -0.800 to -0.657, and the GOODNEWS coefficient estimate increased from 0.857 to 1.075. These two changes are as expected. The INFLATION variable was low historically for the 2004 election, which is positive for Bush, and yet Bush did not do as well as the equation predicted he should have. So it is not surprising that adding the 2004 observation leads to a smaller coefficient estimate in absolute value on INFLATION. Regarding the GOODNEWS variable, there was only one good news quarter for the 2004 election, which is negative for Bush. So it is not surprising in this case that adding the 2004 observation leads to a larger coefficient estimate on GOODNEWS. The (within sample) error for 2004 is 3.4 percentage points, compared to the outside sample error of 5.3 percentage points. The error for 1992 fell from 5.2 percentage points to 4.3 percentage points with the addition of the 2004 observation. The larger GOODNEWS coefficient estimate helps the prediction for 1992, since there were only 2 good news quarters for the 1992 election. The equation says that the elder Bush should have done better than he did, but the GOODNEWS variable is helping keep the error modest.
A remarkable feature of the results in Table 1 is how well the equation estimated only through 1960 does (column 4). For this equation the 8 coefficient estimates are based on only 12 observations. There are 11 outside sample predictions for this equation, beginning with 1964, and the mean absolute error is only 3.0 percentage points for these 11 predictions. As noted above, the 2004 prediction is outside sample by 44 years. It is particularly encouraging that the GOODNEWS variable is significant for this much shorter estimation period. This significance suggests that the variable is not just serving as a dummy variable for the 1992 and 2004 elections. Its coefficient estimate is smaller for the shorter period (0.698 versus 1.075), which is part of the explanation for the larger prediction errors for 1992 and 2004 (8.8 versus 4.3 for 1992 and 6.8 versus 3.4 for 2004).
The current coefficient estimate for GROWTH is .680, which says that a one percentage point increase in GROWTH leads to a .680 percentage point increase in the vote share. The coefficient estimate for INFLATION is -.657, which says that a one percentage point increase in INFLATION leads to a .657 percentage point decrease in the vote share. The coefficient estimate for GOODNEWS is 1.075, which says that each good news quarter adds 1.075 percentage points to the vote share.
For the 2008 election PARTY is -1, PERSON is 0, DURATION is 1, and WAR is 0. Multiplying these values by their respective coefficients and adding the intercept gives a value of 46.61. The vote equation for 2008 is then:
The October 27, 2006, economic forecast from the US model on this website has no more predicted good news quarters, so GOODNEWS has a predicted value of 1. GROWTH is predicted to be 1.8 percent (assuming population growth of 0.9 percent). INFLATION is predicted to be 3.6 percent, which is based on the past seven inflation rates and the eight predicted future rates. Using these economic forecasts in the above equation gives a predicted value of VOTE of 46.5 percent. The Republicans are thus predicted to lose the election by 3.5 percentage points given these economic forecasts. It is the case, however, that these economic forecasts from the US model are more pessimistic that the current consensus view of the future course of the U.S. economy, especially regarding inflation. It is possible on this website to enter your own values for the three economic variables and produce a prediction for VOTE. (Compute your own prediction for 2008 using the 2004 update.) If, for example, you use 2.5 for GROWTH, 2.5 for INFLATION, and 3 for GOODNEWS, you get a predicted value for VOTE of 49.9 percent, a dead heat.
So the bottom line using the vote equation is that the 2008 election looks very close. A somewhat pessimistic economic forecast like the current forecast from the US model on this website leads to a modest Democratic victory. A more neutral economic forecast, which is probably currently the consensus forecast, leads to a dead heat---clearly too close to call. Strong growth with modest inflation would lead to a modest Republican victory. You can play around with the numbers on the website to see what you think. Remember that the estimated standard error is 2.54 percentage points and that added to the uncertainty of any prediction of VOTE is the uncertainty of the economic forecasts themselves.
Table 2 examines the sensitivity to alternative measures of GROWTH and INFLATION. Column 1 is the basic equation, repeated from column 1 in Table 1. Column 2 uses the growth rate in the second and third quarters of the election year only, and column 3 uses the growth rate in the four quarters preceding the election. The fit is obviously best using the growth rate in the first three quarters of the election year (the basic equation), and so the data support this specification. Column 4 uses the inflation rate in the 11 quarters prior to the election, and column 5 uses the inflation rate in the 7 quarters prior to the election. These compare to 15 quarters for the basic specification. Again, the fit is best for the basic specification, and in this case the alternative inflation variables are not significant. The bottom of Table 2 shows that neither the error for 2004 nor the prediction for 2008 is sensitive to the alternative specifications.
__________________________________________________________________________
Table 2
Testing GROWTH and INFLATION Measures
1 2 3 4 5
GROWTH INFLATION INFLATION
GROWTH 4 qtrs 11 qtrs 7 qtrs
2nd & 3rd before before before
Current quarters election election election
GROWTH 0.680 0.516 0.600 0.700 0.715
( 6.14) ( 5.02) ( 4.71) ( 5.73) ( 5.24)
INFLATION -0.657 -0.478 -0.735 -0.429 -0.238
(-2.26) (-1.35) (-2.13) (-1.46) (-0.82)
GOODNEWS 1.075 1.088 1.093 0.978 1.012
( 4.31) ( 3.81) ( 3.68) ( 3.40) ( 3.22)
PERSON 3.30 2.39 3.04 3.21 3.17
( 2.34) ( 1.46) ( 1.81) ( 2.10) ( 1.96)
DURATION -3.33 -4.12 -3.28 -2.82 -2.95
(-2.75) (-2.99) (-2.27) (-2.15) (-2.16)
PARTY -2.68 -2.82 -3.14 -2.34 -2.31
(-4.27) (-3.91) (-4.12) (-3.46) (-3.18)
WAR 5.61 7.16 5.12 5.13 6.09
( 2.09) ( 2.26) ( 1.61) ( 1.54) ( 1.67)
INTERCEPT 47.26 48.03 47.45 46.87 46.08
(18.62) (16.64) (15.69) (16.30) (14.94)
SE 2.54 2.91 3.02 2.75 2.88
2
R 0.905 0.876 0.866 0.889 0.878
No. obs. 23 23 23 23 23
V V V V V V
Year act. pred. error pred. error pred. error pred. error pred. error
1916 51.7 | 49.8 -1.9 | 48.2 -3.5 | 51.2 -0.5 | 49.7 -2.0 | 49.7 -2.0 |
1920 36.1 | 39.1 3.0 | 40.8 4.7 | 40.6 4.5 | 38.8 2.7 | 38.7 2.6 |
1924 58.2 | 58.0 -0.3 | 56.6 -1.6 | 58.8 0.6 | 59.1 0.8 | 58.8 0.6 |
1928 58.8 | 57.2 -1.7 | 57.4 -1.4 | 55.5 -3.3 | 56.1 -2.7 | 55.8 -3.0 |
1932 40.8 | 38.8 -2.1 | 40.6 -0.2 | 39.1 -1.7 | 38.5 -2.4 | 38.8 -2.1 |
1936 62.5 | 63.9 1.5 | 65.9 3.4 | 64.1 1.6 | 64.0 1.6 | 64.1 1.7 |
1940 55.0 | 55.8 0.8 | 57.5 2.5 | 58.4 3.4 | 55.2 0.2 | 54.8 -0.1 |
1944 53.8 | 52.2 -1.5 | 51.3 -2.4 | 51.8 -1.9 | 52.3 -1.4 | 52.4 -1.4 |
1948 52.4 | 50.9 -1.4 | 50.2 -2.2 | 49.8 -2.6 | 51.1 -1.2 | 51.2 -1.2 |
1952 44.6 | 45.2 0.6 | 44.4 -0.2 | 44.6 0.0 | 45.2 0.6 | 45.3 0.7 |
1956 57.8 | 56.4 -1.4 | 57.5 -0.2 | 57.1 -0.7 | 55.3 -2.5 | 54.8 -2.9 |
1960 49.9 | 50.9 1.0 | 50.1 0.2 | 51.5 1.5 | 50.8 0.9 | 50.5 0.6 |
1964 61.3 | 61.3 -0.1 | 59.8 -1.5 | 59.9 -1.5 | 60.5 -0.8 | 60.4 -1.0 |
1968 49.6 | 50.1 0.5 | 49.2 -0.4 | 48.9 -0.7 | 50.5 0.9 | 50.6 1.1 |
1972 61.8 | 58.4 -3.4 | 58.2 -3.5 | 57.1 -4.7 | 58.5 -3.3 | 58.8 -3.0 |
1976 48.9 | 49.5 0.6 | 49.3 0.4 | 49.5 0.6 | 50.5 1.6 | 51.7 2.7 |
1980 44.7 | 45.7 1.0 | 46.5 1.8 | 45.4 0.7 | 46.6 1.9 | 47.3 2.6 |
1984 59.2 | 62.1 2.9 | 61.8 2.6 | 62.1 2.9 | 62.3 3.1 | 62.7 3.5 |
1988 53.9 | 50.5 -3.4 | 51.1 -2.8 | 51.4 -2.5 | 50.6 -3.3 | 50.3 -3.6 |
1992 46.5 | 50.9 4.3 | 50.0 3.4 | 50.6 4.0 | 51.3 4.8 | 51.2 4.6 |
1996 54.7 | 53.0 -1.8 | 52.9 -1.8 | 51.9 -2.9 | 53.0 -1.7 | 52.8 -2.0 |
2000 50.3 | 49.6 -0.6 | 50.0 -0.3 | 50.0 -0.2 | 49.7 -0.6 | 49.3 -0.9 |
2004 51.2 | 54.6 3.4 | 54.5 3.3 | 54.5 3.2 | 54.3 3.1 | 53.9 2.6 |
2008 ? | 46.5 | 47.0 | 46.8 | 47.0 | 46.8 |
_____________________________________________________________________________
Notes: error may differ from pred. minus act. because of rounding.
The estimation period is 1916-2004.
The predictions for 2008 use the forecasts of the economic
variables from the US model discussed in Section 4.
Table 3 examines the sensitivity to alternative measures of GOODNEWS and DURATION. Column 1 is again the basic equation. Column 2 uses 2.7 percent rather than 3.2 percent as the cutoff for the GOODNEWS variable, and column 3 uses 3.7 percent. The results are not very sensitive to these changes. Column 4 uses for the measure of the DURATION variable 1.00 if the incumbent party has been in power for two or more consecutive terms rather than 1.00, 1.25, 1.50, etc. for the basic specification. Column 5 uses 1.00, 1.50, 2.00, etc. These changes affect somewhat the coefficient estimates of DURATION and PERSON. There is clearly some collinearity between PERSON and the measure of DURATION.
__________________________________________________________________________
Table 3
Testing GOODNEWs and DURATION Measures
1 2 3 4 5
GOODNEWS GOODNEWS DURATION DURATION
2.7% 3.7% 0.,1., 0.,1.,
Current cutoff cutoff 1.,1.,1.,.. 1.5,2.0,2.5,..
GROWTH 0.680 0.643 0.664 0.677 0.685
( 6.14) ( 5.37) ( 5.35) ( 5.93) ( 6.13)
INFLATION -0.657 -0.715 -0.818 -0.685 -0.631
(-2.26) (-2.30) (-2.52) (-2.27) (-2.14)
GOODNEWS 1.075 1.104 0.918 1.085 1.079
( 4.31) ( 3.78) ( 3.45) ( 4.23) ( 4.27)
PERSON 3.30 3.57 2.80 2.84 3.84
( 2.34) ( 2.34) ( 1.79) ( 1.77) ( 2.89)
DURATION -3.33 -3.20 -3.32 -4.13 -2.45
(-2.75) (-2.45) (-2.45) (-2.50) (-2.64)
PARTY -2.68 -2.48 -2.41 -2.79 -2.61
(-4.27) (-3.76) (-3.51) (-4.32) (-4.11)
WAR 5.61 5.88 3.32 5.49 5.51
( 2.09) ( 1.97) ( 1.20) ( 1.98) ( 2.03)
INTERCEPT 47.26 46.41 49.60 47.78 46.45
(18.62) (15.63) (19.48) (17.01) (19.18)
SE 2.54 2.72 2.84 2.62 2.57
2
R 0.905 0.891 0.882 0.899 0.903
No. obs. 23 23 23 23 23
V V V V V V
Year act. pred. error pred. error pred. error pred. error pred. error
1916 51.7 | 49.8 -1.9 | 49.2 -2.5 | 50.7 -0.9 | 49.7 -2.0 | 49.8 -1.9 |
1920 36.1 | 39.1 3.0 | 39.2 3.1 | 39.6 3.5 | 38.6 2.5 | 39.1 2.9 |
1924 58.2 | 58.0 -0.3 | 57.3 -0.9 | 57.2 -1.1 | 58.1 -0.1 | 57.8 -0.5 |
1928 58.8 | 57.2 -1.7 | 56.3 -2.6 | 57.1 -1.7 | 57.0 -1.8 | 57.2 -1.6 |
1932 40.8 | 38.8 -2.1 | 38.4 -2.4 | 38.8 -2.0 | 38.9 -2.0 | 39.1 -1.8 |
1936 62.5 | 63.9 1.5 | 63.2 0.8 | 64.0 1.6 | 63.9 1.4 | 63.9 1.4 |
1940 55.0 | 55.8 0.8 | 55.6 0.6 | 56.5 1.5 | 55.0 0.0 | 56.5 1.5 |
1944 53.8 | 52.2 -1.5 | 52.1 -1.6 | 52.0 -1.8 | 52.1 -1.7 | 52.5 -1.3 |
1948 52.4 | 50.9 -1.4 | 50.9 -1.5 | 50.7 -1.7 | 51.6 -0.8 | 50.8 -1.6 |
1952 44.6 | 45.2 0.6 | 45.9 1.3 | 45.4 0.8 | 47.3 2.7 | 44.3 -0.3 |
1956 57.8 | 56.4 -1.4 | 55.7 -2.1 | 55.9 -1.8 | 56.5 -1.2 | 56.1 -1.7 |
1960 49.9 | 50.9 1.0 | 50.0 0.1 | 51.9 2.0 | 50.8 0.9 | 51.0 1.1 |
1964 61.3 | 61.3 -0.1 | 62.0 0.7 | 60.6 -0.7 | 61.3 -0.1 | 61.2 -0.2 |
1968 49.6 | 50.1 0.5 | 49.5 -0.1 | 51.1 1.5 | 49.7 0.1 | 50.4 0.8 |
1972 61.8 | 58.4 -3.4 | 58.3 -3.4 | 58.5 -3.3 | 58.5 -3.3 | 58.2 -3.6 |
1976 48.9 | 49.5 0.6 | 49.3 0.3 | 49.5 0.6 | 49.2 0.2 | 49.8 0.8 |
1980 44.7 | 45.7 1.0 | 46.2 1.5 | 45.8 1.1 | 45.4 0.8 | 45.7 1.0 |
1984 59.2 | 62.1 2.9 | 62.1 3.0 | 61.5 2.3 | 62.2 3.0 | 61.9 2.8 |
1988 53.9 | 50.5 -3.4 | 52.7 -1.2 | 50.5 -3.4 | 50.3 -3.6 | 50.6 -3.3 |
1992 46.5 | 50.9 4.3 | 51.1 4.6 | 50.7 4.1 | 51.0 4.5 | 51.1 4.6 |
1996 54.7 | 53.0 -1.8 | 53.6 -1.2 | 53.1 -1.6 | 52.9 -1.9 | 52.8 -1.9 |
2000 50.3 | 49.6 -0.6 | 49.2 -1.1 | 47.0 -3.2 | 49.3 -1.0 | 49.8 -0.4 |
2004 51.2 | 54.6 3.4 | 55.8 4.6 | 55.6 4.4 | 54.7 3.5 | 54.4 3.1 |
2008 ? | 46.5 | 46.5 | 47.9 | 46.3 | 46.6 |
_____________________________________________________________________________
Notes: error may differ from pred. minus act. because of rounding.
The estimation period is 1916-2004.
The predictions for 2008 use the forecasts of the economic
variables from the US model discussed in Section 4.
Table 4 examines the sensitivity to the treatment of Ford, to the treatment of the third party candidates LaFollette and Perot, and to the treatment of World Wars I and II. Column 1 is the basic equation. Column 2 uses a value of 1 for PERSON for Ford (1976 election) rather than 0. This change increased in absolute value the INFLATION coefficient estimate and increased the error for 1976 slightly, but otherwise had little effect. For column 3 all of the LaFollette vote is given to the Democrats rather than just 76.5 percent of it (1924 election). This change had only small effects. For column 4 the Perot vote is given to the Republicans in the 1992 election rather than just considering the two-party vote, as in the basic equation. This change had some noticeable effects. It decreased the GOODNEWS coefficient estimate from 1.075 to 0.836; it made the DURATION variable insignificant; it lowered the error for 1992; and it raised the error for 2004. It lowered the standard error from 2.54 to 2.22. I have argued elsewhere---see, for example, Econometrics and Presidential Elections---that it is probably not sensible to assume that Perot took all his votes from Bush. Columns 1 and 4 show that it does make some difference which assumption about the Perot votes is made. Finally, column 5 drops the special treatment of World Wars I and II regarding INFLATION and GOODNEWS. The actual values of these variables were used for 1920, 1944, and 1948 rather than zero, and WAR was dropped. The values of INFLATION in these three elections are 16.535, 5.644, and 8.482, respectively, and the values of GOODNEWS are 5, 14, and 5. The large outliers are INFLATION in 1920 and GOODNEWS in 1944. This change cut the INFLATION coefficient estimate roughly in half and made it insignificant, and similarly for the PERSON coefficient estimate. The GOODNEWS coefficient estimate decreased from 1.075 to 0.816. The standard error increased from 2.54 to 2.90. Some of the coefficient estimates are thus sensitive to the treatment of the two world wars.
__________________________________________________________________________
Table 4
Testing Ford, LaFollette, Perot, and World Wars
1 2 3 4 5
LaFollette Perot INFLATION &
PERSON all for all for GOODNEWS
Ford = 1 Democrats Republicans No adjustment
Current (1976) (1924) (1992) for world wars
GROWTH 0.680 0.639 0.719 0.726 0.602
( 6.14) ( 5.31) ( 6.30) ( 7.50) ( 4.25)
INFLATION -0.657 -0.826 -0.686 -0.724 -0.336
(-2.26) (-2.49) (-2.28) (-2.84) (-1.56)
GOODNEWS 1.075 1.054 0.876 0.836 0.816
( 4.31) ( 4.03) ( 3.41) ( 3.83) ( 3.54)
PERSON 3.30 3.02 3.05 5.21 1.81
( 2.34) ( 1.88) ( 2.10) ( 4.23) ( 1.20)
DURATION -3.33 -3.79 -3.11 -1.40 -3.77
(-2.75) (-3.12) (-2.49) (-1.32) (-3.03)
PARTY -2.68 -2.57 -2.38 -2.99 -1.96
(-4.27) (-3.91) (-3.68) (-5.46) (-2.91)
WAR 5.61 5.15 4.20 2.63 -
( 2.09) ( 1.78) ( 1.52) ( 1.12)
INTERCEPT 47.26 48.33 48.31 46.94 49.29
(18.62) (19.21) (18.45) (21.13) (20.63)
SE 2.54 2.67 2.62 2.22 2.90
2
R 0.905 0.895 0.896 0.926 0.868
No. obs. 23 23 23 23 23
V V V V V V
Year act. pred. error pred. error pred. error pred. error pred. error
1916 51.7 | 49.8 -1.9 | 49.9 -1.8 | 50.3 -1.4 | 50.2 -1.5 | 51.5 -0.2 |
1920 36.1 | 39.1 3.0 | 39.8 3.7 | 38.8 2.7 | 36.9 0.7 | 35.2 -0.9 |
1924 58.2 | 58.0 -0.3 | 57.7 -0.5 | 56.2 1.9 | 57.0 -1.3 | 57.2 -1.1 |
1928 58.8 | 57.2 -1.7 | 57.3 -1.5 | 56.9 -1.9 | 57.6 -1.2 | 55.9 -2.9 |
1932 40.8 | 38.8 -2.1 | 38.2 -2.6 | 38.0 -2.9 | 41.0 0.2 | 40.5 -0.4 |
1936 62.5 | 63.9 1.5 | 63.7 1.3 | 63.6 1.2 | 63.4 1.0 | 62.7 0.3 |
1940 55.0 | 55.8 0.8 | 55.8 0.8 | 55.6 0.6 | 57.2 2.2 | 54.2 -0.8 |
1944 53.8 | 52.2 -1.5 | 51.9 -1.8 | 52.4 -1.4 | 53.1 -0.6 | 56.5 2.8 |
1948 52.4 | 50.9 -1.4 | 50.5 -1.8 | 51.1 -1.3 | 52.3 -0.1 | 46.9 -5.5 |
1952 44.6 | 45.2 0.6 | 45.0 0.4 | 45.5 0.9 | 46.1 1.5 | 46.1 1.5 |
1956 57.8 | 56.4 -1.4 | 56.7 -1.1 | 55.8 -2.0 | 56.9 -0.9 | 55.6 -2.1 |
1960 49.9 | 50.9 1.0 | 51.0 1.1 | 50.9 1.0 | 51.6 1.6 | 51.1 1.2 |
1964 61.3 | 61.3 -0.1 | 61.5 0.2 | 60.6 -0.8 | 60.3 -1.0 | 60.0 -1.4 |
1968 49.6 | 50.1 0.5 | 50.0 0.4 | 50.4 0.8 | 49.8 0.2 | 51.3 1.7 |
1972 61.8 | 58.4 -3.4 | 57.9 -3.8 | 58.2 -3.6 | 59.3 -2.5 | 58.3 -3.5 |
1976 48.9 | 49.5 0.6 | 51.5 2.5 | 49.4 0.5 | 49.9 1.0 | 51.3 2.3 |
1980 44.7 | 45.7 1.0 | 45.3 0.6 | 45.4 0.7 | 45.1 0.4 | 48.4 3.7 |
1984 59.2 | 62.1 2.9 | 61.5 2.3 | 61.1 1.9 | 62.0 2.8 | 61.1 1.9 |
1988 53.9 | 50.5 -3.4 | 50.3 -3.6 | 50.7 -3.2 | 51.3 -2.6 | 51.1 -2.8 |
1992 46.5 | 50.9 4.3 | 50.3 3.7 | 51.3 4.7 | 54.6 -2.1 | 50.5 3.9 |
1996 54.7 | 53.0 -1.8 | 53.3 -1.4 | 53.3 -1.4 | 53.3 -1.5 | 53.6 -1.1 |
2000 50.3 | 49.6 -0.6 | 49.9 -0.4 | 49.6 -0.7 | 49.0 -1.3 | 50.3 0.0 |
2004 51.2 | 54.6 3.4 | 54.8 3.5 | 55.0 3.7 | 56.2 5.0 | 54.7 3.5 |
2008 ? | 46.5 | 46.3 | 47.3 | 48.1 | 48.2 |
_____________________________________________________________________________
Notes: error may differ from pred. minus act. because of rounding.
The estimation period is 1916-2004.
The predictions for 2008 use the forecasts of the economic
variables from the US model discussed in Section 4.
For column 3 the actual value of V is 54.3 in 1924.
For column 4 the actual value of V is 56.7 in 1992.
Looking across all 12 specifications, the GROWTH and GOODNEWS coefficient estimates are fairly stable and always retain their significance. In general the rest are also fairly stable, although there are a few exceptions. The estimates are most sensitive to the treatment of the two world wars. Regarding the last election and the upcoming one, the error for 2004 varies only moderately across the 12 estimates, as does the predicted value for 2008.
(1) VOTEt = a1 + a2(GROWTHt - GROWTH*) + a3(INFLATIONt - INFLATION*)(1 - WARt) + a4(GOODNEWSt - GOODNEWS*)(1 - WARt) + a5PERSONt + a6DURATIONt + a7PARTYt , t=1,...,23
GROWTH* is the "normal" rate, normal in the sense that growth rates above this value are a plus for the incumbent party and growth rates below it are a minus. The same is true for INFLATION* with the sign reversed, and the same is true for GOODNEWS*. These normal values are taken to be constant over time. Remember that WARt is 1 in the three war elections and 0 otherwise. The above equation can be written:
(2) VOTEt = a1 - a2GROWTH* - a3INFLATION* - a4GOODNEWS* + [a3INFLATION* + a4GOODNEWS*]WARt + a2GROWTHt + a3INFLATIONt(1 - WARt) +a4GOODNEWSt(1 - WARt) +a5PERSONt + a6DURATIONt + a7PARTYt, t=1,...,23
or
(3) VOTEt = b1 + b2WARt + a2GROWTHt + a3INFLATIONt(1 - WARt) +a4GOODNEWSt(1 - WARt) +a5PERSONt + a6DURATIONt + a7PARTYt, t=1,...,23
where
b1 = a1 - a2GROWTH* - a3INFLATION* - a4GOODNEWS*
and
b2 = a3INFLATION* + a4GOODNEWS*
You can thus see that the coefficient on WARt, which is b2, is nonzero because INFLATION* and GOODNEWS* are nonzero. Equation (3) is the form of the equation that is estimated. There are not separate estimates of GROWTH*, INFLATION*, and GOODNEWS*.
The data that were used to estimate the vote equation are presented in Table A-1 below. Data are presented back to 1880 even though only data back to 1916 were used in the estimation work. Quarterly data on nominal GDP, real GDP, and population are needed to construct GROWTH, INFLATION, and GOODNEWS, and these data are presented in Table A-2 at the end of this paper. In Table A-2 Y denotes nominal GDP, X denotes real GDP, and POP denotes population. Let a subscript k denote the kth quarter of the sixteen-quarter period of an administration. Also, let Z = X/POP, per capita real GDP, and let P = Y/X, the GDP deflator. Then GROWTH and INFLATION are constructed as:
VOTE is the incumbent vote divided by the Democratic plus Republican vote except for the 1912 and 1924 elections. For 1912, VOTE is the incumbent vote divided by the Democratic plus Republican plus Roosevelt vote. For 1924, VOTE is the incumbent vote plus .765 times the LaFollette vote divided by the Democratic plus Republican plus LaFollette vote. The voting data for 1880-1916 were obtained from U.S. Department of Commerce (1975), pp. 1078-1079. For 1920-1932 the data were obtained from U.S. Department of Commerce (1988), p. 232, for 1936-1992 the data were obtained from U.S. Department of Commerce (1997), p. 271, and for 1996-2000 the data were obtained from U.S. Department of Commerce (2001), p. 233. The voting data for the 2004 election were obtained from the U.S. Department of Commerce website.
PARTY, PERSON, DURATION, and WAR are defined in the text. In the construction of PERSON Ford is not counted as an incumbent running again, since he was not an elected vice president, whereas the other vice presidents who became president while in office are counted.
The data on nominal GDP were obtained as follows. Annual data for 1929-1945 and quarterly data for 1947:1-2006:3 were obtained from the Bureau of Economic Analysis (BEA) website on October 27, 2005. Quarterly data for 1946:1-1946:4 were obtained from the BEA website on October 30, 2002. Quarterly data for 1877:1-1945:4 are available from Balke and Gordon (1986), pp. 789-795. The Balke and Gordon values for 1877:1-1928:4 were used exactly, but the values for 1929:1-1945:4 were adjusted to take account of the new BEA annual data. For 1929:1-1945:4 each quarterly value for a given year was multiplied by a splicing factor for that year. The splicing factor is the ratio of the BEA value for that year to the respective yearly value in Balke and Gordon (1976), pp. 782-783.
The data on real GDP were obtained in a similar way. Annual data for 1929-1946 and quarterly data for 1947:1-2006:3 were obtained from the BEA website on October 27, 2006. Quarterly data for 1877:1-1946:4 are available from Balke and Gordon (1986), pp. 789-795. The Balke and Gordon values were spliced to the BEA values. All the Balke and Gordon quarterly values for 1877:1-1929:4 were multiplied by the same number. This number is the ratio of the BEA value for 1929 to the 1929 value in Balke and Gordon (1976), p. 782. For 1930:1-1946:4 each Balke and Gordon quarterly value for a given year was multiplied by a splicing factor for that year. The splicing factor is the ratio of the BEA value for that year to the respective yearly value in Balke and Gordon (1976), pp. 782-783.
The data on population were obtained as follows. For 1877-1928 annual data were obtained from U.S. Department of Commerce (1973), pp. 200-201, A114 series. Each of these observations was multiplied by 1.000887, a splicing factor. The splicing factor is the ratio of the A114 value for 1929 in U.S. Department of Commerce (1973) to the value for 1929 in Table 8.2 in U.S. Department of Commerce (1992). For 1929-1945 annual data were obtained from U.S. Department of Commerce (1992), Table 8.2. Quarterly observations for 1877:1-1945:4 were obtained by interpolating the annual observations using the method presented in Fair (1994), Table B.6. For 1946:1-2006:3 quarterly data were obtained from the BEA website on October 27, 2006.
____________________________________________________________________
Table A-1
Data Used in the Estimation
VOTE PARTY PERSON DURATION WAR GROWTH INFLATION GOODNEWS
1880 50.220 -1 0 1.75 0 3.879 1.974 9
1884 49.846 -1 0 2.00 0 1.589 1.055 2
1888 50.414 1 1 0.00 0 -5.553 0.604 3
1892 48.268 -1 1 0.00 0 2.763 2.274 7
1896 47.760 1 0 0.00 0 -10.024 3.410 6
1900 53.171 -1 1 0.00 0 -1.425 2.548 7
1904 60.006 -1 1 1.00 0 -2.421 1.442 5
1908 54.483 -1 0 1.25 0 -6.281 1.879 8
1912 54.708 -1 1 1.50 0 4.164 2.172 8
1916 51.682 1 1 0.00 0 2.229 4.252 3
1920 36.119 1 0 1.00 1 -11.463 0.000 0
1924 58.244 -1 1 0.00 0 -3.872 5.161 10
1928 58.820 -1 0 1.00 0 4.623 0.183 7
1932 40.841 -1 1 1.25 0 -14.499 7.200 4
1936 62.458 1 1 0.00 0 11.765 2.497 9
1940 54.999 1 1 1.00 0 3.902 0.081 8
1944 53.774 1 1 1.25 1 4.279 0.000 0
1948 52.370 1 1 1.50 1 3.579 0.000 0
1952 44.595 1 0 1.75 0 0.691 2.362 7
1956 57.764 -1 1 0.00 0 -1.451 1.935 5
1960 49.913 -1 0 1.00 0 0.377 1.967 5
1964 61.344 1 1 0.00 0 5.109 1.260 10
1968 49.596 1 0 1.00 0 5.043 3.139 7
1972 61.789 -1 1 0.00 0 5.914 4.815 4
1976 48.948 -1 0 1.00 0 3.751 7.630 5
1980 44.697 1 1 0.00 0 -3.597 7.831 5
1984 59.170 -1 1 0.00 0 5.440 5.259 8
1988 53.902 -1 0 1.00 0 2.178 2.906 4
1992 46.545 -1 1 1.25 0 2.662 3.280 2
1996 54.736 1 1 0.00 0 3.121 2.062 4
2000 50.265 1 0 1.00 0 1.219 1.605 8
2004 51.233 -1 1 0.00 0 2.690 2.325 1
___________________________________________________________________