You may see reference to this article saying that the "undecideds break for the challenger" rule is wrong.
Thus, from the data, incumbents won over the "summer undecideds" in 1936, 1940, 1944, 1948, 1956, 1976, It was a mixed bag in 1972, while the challenger picked up the undecideds in 1964 (barely), 1980, and 1996 (barely). The only strong challenger year was 1980.
But maybe the "undecideds break for the challenger" rule comes from polling data closer to the election? This seemed like a plausible possibility, so my next step -- and the most relevant for purposes now -- was to see what the results were from a month out. This was the easiest, as Gerry from Daly Thoughts had already calculated this from Gallup results. Going from these findings, undecideds broke for the incumbent in half of the years -- 1936, 1944, 1948, 1956, 1976 (barely), and 1992. They broke evenly in four -- 1940, 1972, 1984, and 1996 -- and for the challenger in 1964, and 1980. The data was a little stronger, but it still indicated that undecideds would break for the incumbent or evenly, rather than strongly for the challenger.
When I averaged the final polls and compared that average to the actual results, it showed the following:
- Clinton -1, Dole +3 -- Net +4 Challenger
- Bush +0, Clinton -2 -- Net +2 Incumbent
- Reagan +2, Mondale +2 -- Net 0
- Carter +0, Reagan +6 -- Net +6 Challenger
- Ford +2, Carter +3 -- Net +1 Challenger
- Nixon +0, McGovern +2 -- Net +2 Challenger
- Johnson +5, AuH2O +3 -- Net +2 Incumbent
- Ike -1.5, Stevenson +1.5 -- Net +3 Challenger
- Truman +5.5, Dewey -4.4 -- Net +9.9 Incumbent
- Roosevelt +3.5, Dewey -2.5 -- Net +6 Incumbent
- Roosevelt +3, Wilkie -3 -- Net +6 Incumbent
- Roosevelt +5, Landon -7 -- Net +12 Incumbent
Again, the results are mixed at best. Discounting polls from before the 1948 debacle, there finally is some evidence of undecideds breaking for challengers, as they did in five of the eight elections. But even here predictions of a strong break for the challenger were unfounded -- on average the challenger netted only 3.5 points on the incumbent. Leaving aside 1980 creates even less spectacular gains -- an average net gain of 2.
Update [2004-10-19 3:34:44 by KariQ]: Forgot the link. Whither The Undecideds
Don't panic. The writer is wrong. Here's why:
First of all, in 1992 and 1996, he left out Perot. In 1996, it doesn't matter much. But in 1992 from the final polls to the election, Perot picked up all the votes that Clinton lost. So the net gain for the incumbent was actually 0. Second, Gallup was wildly off that year, so much so that it makes more sense to exclude them entirely (A practice recommend in all elections since 1984). if we exclude Gallup's numbers and recalculate all the numbers we get:
1992 Clinton -1, Bush +0, Perot +3 -- Net +2 Challengers
The only other year after 1956 that he shows the challenger not picking up votes is in 1964. In that year, he has the numbers wrong. The correct numbers are
1964 Johnson -3, Goldwater +3 -- Net +6 Challenger
Polling from 1948 should not be compared because Gallup stopped polling in mid-October that year so there is no true comparison. Polling from before that shouldn't be used period since polling methods have changed dramatically since then. So if we correct the chart it looks like this:
- Clinton -1, Dole +3 -- Net +4 Challenger
- Clinton -1, Bush +0, Perot +3 -- Net +2 Challenger
- Reagan +2, Mondale +2 -- Net 0
- Carter +0, Reagan +6 -- Net +6 Challenger
- Ford +2, Carter +3 -- Net +1 Challenger
- Nixon +0, McGovern +2 -- Net +2 Challenger
- Johnson -3, Goldwater +3 -- Net +6 Challenger
- Ike -1.5, Stevenson +1.5 -- Net +3 Challenger
In other words, since 1956 the best the incumbent can hope for is to equal the results of the poll immediately before the election. On average, the challenger has a net gain of 3 points.
Here's the link in case you want to verify the information for yourself:
Final National Presidential Poll Results, 1936-2000