Average change in “gap” since 1936 is 6.6 points
PRINCETON, NJ -- A question of keen interest to election observers is the following: To what degree do presidential elections change between the end of the political conventions and Election Day?
There have been 18 presidential elections since Gallup's election polling began in 1936. In each instance, both parties' conventions had concluded by Labor Day, making it possible to look at the difference in the margin between the Gallup Poll that immediately followed Labor Day and the final popular-vote outcome. (In 2004, the Labor Day poll was conducted during the Labor Day weekend, just after the Republican National Convention had ended.)
The gap change between Labor Day and Election Day over these 18 elections has ranged from 1 point (1940, 1960, 1988, and 2004) to 20 points (1936). (The gap change represents a comparison of the difference in the candidates' vote percentages around Labor Day and the difference in their final Election Day popular-vote percentages.)
Here is another way of looking at the distribution of gap changes between those two milestones over the 18 elections between 1936 and 2004.
The median in the distribution is 6 points -- that is, six is the number that best splits the distribution of gap changes in half (two elections showed a gap change of exactly 6 points).
The mean (mathematical average) gap change is 6.6 points, driven a little higher than the median by the high 20-point gap of the 1936 election.
Based solely on history, that would put the predicted gap change for this year's election (comparing the post-Labor Day poll results with the actual popular-vote percentages on Election Day) in the 6- to 7-point range.
Shrinking or Expanding Gap?
In theory, a gap change, if it does occur, could be in either direction. The difference between the two major-party candidates could grow or decline by 6 points.
The historical analysis suggests that the change is more frequently in the direction of closing a gap, rather than expanding it, between Labor Day and Election Day. In the 18 elections, the gap expanded in 6, while it shrank in 12.
Three of the elections in which the gap expanded involved Franklin D. Roosevelt. Beginning with the 1948 election, the Labor Day gap between the two major-party candidates shrank in all but 3 out of 14 elections. The three post-World War II elections in which the gap expanded were 1956 (Dwight Eisenhower gained steam on Adlai Stevenson as the fall campaign progressed), 1984 (Ronald Reagan expanded his lead over Walter Mondale by Election Day), and 2004 (George W. Bush won over John Kerry by about 2.5 points, just slightly more than the 2-point gap by which he led Kerry in the Labor Day poll).
Thus, if there was a change in the gap in past years' elections, history showed the odds were higher that the leader's lead diminished than they were that the leader moved further ahead.
Winning or Losing?
It is of some interest to look at the relationship between the Gallup Poll numbers for the leader immediately after Labor Day and the winner on Election Day.
Across the 18 elections between 1936 and 2004, a candidate who was behind in the post-Labor Day poll went on to win the popular vote on Election Day in only 3:
In 1980, Gallup's post-Labor Day poll had Jimmy Carter up by 4. Reagan won by 10.
In Gallup's 1960 post-Labor Day poll, Richard Nixon was up by 1. John F. Kennedy won by less than 1 percentage point.
In 1948, Gallup's post-Labor Day poll had Thomas Dewey up by 8 points. Harry Truman won by 5.
In other words, in these previous 18 elections, the candidate who was ahead just after Labor Day has more often than not gone on to win the election, regardless of changes in the margins over the last two months of the campaign.
There were two other recent examples -- 2000 and 2004 -- in which the lead changed hands between Labor Day and Election Day, but the candidate who lost the lead regained it to win the popular vote. Both of these elections were close. Al Gore was ahead in Gallup polling conducted after Labor Day in 2000 and went on to win the popular vote (but of course lost the election after the Supreme Court's decision that put Florida's electoral votes in George W. Bush's column). In 2004, Bush was ahead in Gallup polling conducted after Labor Day and went on to win the popular vote, by about 2.5 points.
The results from the previous elections reviewed here were all based on Gallup's first poll conducted after Labor Day. In all of these instances, both political party conventions were completed before Labor Day, and in some instances, weeks before Labor Day. For example, the last of the two conventions in 2000 was completed on Aug. 17, more than two weeks before Labor Day that year; in 1948, the last of the two conventions was completed in mid-July.
Labor Day this year was Sept. 1, with the Democratic convention taking place immediately before then (Aug. 25-28) and the Republican convention actually taking place on Labor Day and the three days that followed. Thus, it is obviously not appropriate to use Labor Day polling this year to compare to these historical examples. Gallup Poll Daily tracking in fact has shown much movement in the race in the days since Labor Day, with Obama having as much as an 8-point lead in a three-day period that included Labor Day, and McCain as much as a 5-point lead after his convention. As of Sept. 13-15, the race is at a statistical tie: McCain 47%, Obama 46%.
The historical record reviewed here certifies that a change in the gap of up to 6 or 7 points would not be unusual between Labor Day and Election Day. This year, the data have already shown a change in the gap of 13 points from the high-water Obama gap to the high-water McCain gap since Labor Day. The unusual timing of the conventions, both of which fell close to Labor Day, makes it difficult to project probabilities of victory for one candidate or the other based on historical references to the calendar. More generally, it may be useful to refer back to Gallup's previous analysis showing that, historically, competitive campaigns in which one candidate did not pull to a clear post-conventions lead remained close (with the lead switching back and forth) right through to the election.
To provide feedback or suggestions about how to improve Gallup.com, please e-mail firstname.lastname@example.org.