Canada - On the Lucky Effect
In some election, one side wins a disproportionate amount of the close seats. I refer to this as the "lucky effect". It was first described here: http://electionpolls.blogspot.com/2006/01/general-comments.html.
The lucky effect is quantified as the difference being the actual seats won and the seats predicted using probability using the actual results. If a seat was won by one vote, the probabilities would be about 50-50. Based on my current models, a 10 point win in a two party race is around 87 %, a 20 point win is around 96 %, and a 30 point spread is getting close to 100 %. The regional results have higher probabilities for these spreads, since the standard deviation in the projection is lower. Anyway, here are the numbers:
2006:
Actual: Con: 124 Lib: 103 Ndp: 29 Blc: 51
Probability: Con: 122.4 Lib: 103.2 Ndp: 30.0 Blc: 51.5
2004:
Actual: Con: 99 Lib: 135 Ndp: 19 Blc: 54
Probability: Con: 93.7 Lib: 134.9 Ndp: 23.0 Blc: 55.4
In 2004 the Conservatives benifitted by the lucky effect a lot, and the NDP was hurt. In this election, the effect was much smaller. In both cases, the liberal totals were about even with probability, while the Conservatives were higher and NDP and BQ lower.
Micky