Wednesday, October 27, 2004

Break for the Challenger? Not Quite.

Guy Molyneux along with Mark Blumenthal and Charlie Cook, all make the assertion that undecided voters will break for the challenger in presidential contest, citing the final polls before the election and results of the past few elections. Let’s look at this assertion.

Molyneux claims 1) in the past four elections involving an incumbent only once has the incumbent gained from the final poll to the election and 2) all challengers gained substantial from poll to election. Both are misleading.

In 1) Molyneux rounds, so than an increase of a fraction of a percent will appear to be 0, thus in reality (using Gallup poll numbers) two of the past four incumbents have drawn a higher percentage in the election than in the final Gallup poll. He is correct in pointing out that the final poll number and the actual vote percentage are very close, and no incumbent has improved substantial from final poll to election since 1980.

In assertion 2) Molyneux includes Ross Perot in both 1992 and 1996 (in other words the increase of the challenger implies adding Clinton and Perot’s change in 1992 and adding Dole and Perot’s change in 1996). This does not help us understand the 2004 election in any way. Perot likely did higher in actual votes than in poll number because of lower turnout for the major party candidates, or dissatisfied voters voicing a “protest votes”. If the vote for Nader is higher than his final poll total, it certainly will not help swing the election for John Kerry (or hurt Bush). Rather, we should examine only the change in poll numbers for the major party challenger. Doing this (and not rounding) we find that the percentage for three of the past four challengers facing an incumbent declined between the final poll and the election (Mondale and Dole by less than a percent, Clinton by almost 6% in 1992).

If we were to leave it at this, our guess is that Bush’s final poll number will be very close to his election percentage, with the odds being 50:50 of improving or declining, but only marginally. Whereas Kerry has a 75% of his poll number dropping, though most likely by a very small percent. Nader, on the other hand, should much better than his final poll numbers.

I’m not going to leave it at that, however, as I would like to delve deeper into the numbers. In the past 12 elections featuring an incumbent president seeking reelection (going back to 1936), the average swing from the final Gallup poll to the election has been a gain of .5% for the incumbent; and a decline of 1% for the challenger. Further, as (currently) Bush leads in the polls and the election is expected to be close, I want to examine the effect in close races where the incumbent leads. What the numbers bear is: when the incumbent leads in the final poll, the average change is +.44% for the incumbent, and -.67% challenger. When the final poll is close (within 6% of each other) the average change is (I/C) +1.3/-.76C. When the actual result is close the change average to +1.6/-2.7. Finally, where both the incumbent leads and the final poll is close (as it appears to be in 2004), the incumbent gains and average of 1.5 points while the challenger loses an average of 1.1 points. If these trends (and the current Gallup poll) hold up, Bush should expect to gain over a point and Kerry lose ground between the final poll and the election numbers. However, even this theory is suspect.

Delving slightly deeper into the numbers, it seems that include some of the earlier elections in our analysis (particularly the Roosevelt and Truman elections). Looking at only the past 6 elections (since 1972), the incumbent does decline (as the pundits insist) from final poll to election by an average of 1%. However, the percentage of the primary challenger also declines, albeit by an average of only 3/100ths of a percent. When the incumbent leads in the final poll, the average changes come to (I/C) -.93/+.43; when the final poll is close: -1.9/+2.9 (only 2 occurrences); when the final result is close, the numbers switch:-.1/ -1.8. Finally, to look at cases where the poll is close and the incumbent leads, only 1976 applies (see chart).

The final criticism of this assertion is that it is statistically unsound. The average change in neither the incumbent nor the challenger’s numbers is not statistically significant (meaning that we cannot reasonably expect that the variation will be positive or negative), whether we look at the past 6 or the past 12 elections. Furthermore, our confidence intervals using the past 12 elections, go from –1.5% to 2.6% for the incumbent and from –3.2% to 1.1% for the incumbent. In other words, we can say that it is 95% likely for Bush to be between a point and a half below and 2.6 percent above his final poll number, and 95% likely for Kerry to be between 3.2% below and a 1.2% of his poll. Using only the past six elections, our confidence intervals just as bad or worse (for the challenger): from –2.6 to .64 for Bush and from –3.4 to 3.3 for Kerry. Basically, using past elections and polls has almost no predictive value, as Bush could do slightly better or worse, as could Kerry, and there is no reason to believe that the final poll number will be above or below the actual results.

Election Chart

No comments: