V=.63*A
For example, in 1952, Eisenhower led Stevenson by 55%-38% in August, so A was -17. The predicted margin in the actual vote was .63*(-17)=-10.7, which was almost exactly equal to the actual margin (55.2% to 44.3%). Clinton's +7 leads to a predicted margin of 4.4%. The standard error is about 4.2.
The biggest residual was in 1980, when Carter and Reagan were tied in August, but Reagan went on to win easily (about 51%-41%). I think that even the final polls showed a close race, but I also recall than Carter's campaign seemed to be floundering in the last few months. So probably some of it was survey error but some of it was a real change. The other two large residuals were in 2008, when the polls were pretty much tied in August, and in 1956, when Eisenhower was ahead by 13 in August and increased the margin to 15 in September. The explanation for 2008 is obvious--the financial crisis that started in September 2008. I don't know anything about the details of the 1956 campaign, but my guess is that since it was a rematch of 1952, people made their minds up earlier than usual.
The experience of 1952-2012 suggests that a Trump victory is unlikely but not impossible (maybe 15%). A Clinton landslide (say a margin of 10 or more) is also unlikely.
The raw data:
Aug Final
1952 -17 -11
1956 -13 -15
1960 0 0
1964 41 24
1968 -6 -1
1972 -37 -23
1976 9 2
1980 0 -10
1984 -26 -18
1988 -8 -8
1996 15 9
2000 5 1
2004 -1 -2
2008 0 7
2012 9 4
*I omitted 1992. because Ross Perot had dropped out, but resumed his campaign in September.
[Data from iPOLL, Roper Center for Public Opinion Research]
No comments:
Post a Comment