Post Mortem 2011

LISPOP has been in the practice of providing a post mortem of its seat projection since it went online for the 2004 election. What follows will be an instant review, not intended to be the final analysis which hopefully will be provided in the coming days for those watching this space. It consists of two tables, the first comparing our final pre-election seat projection of May 1, with the results as reported on the morning of May 3, they might still change slightly. The second table compares the aggregate regional splits in popular vote available by April 30 from the various polling houses, which was the basis for our final projection, with the popular vote figure available on election night. It is customary for various participants in the modelling process to claim success for their method the day after, and we can do that as well, however it is our intention here to explain the discrepancy between the figures in table 1, by an examination of table 2. The greatest variations occur in Ontario where we used a 10% Conservative lead over the NDP with the actual difference of 18%. Likewise we used a 12% NDP lead over the Bloc in Quebec, when the real figure rounded to 20%. As someone who has applied the "regional swing" model in postdictions back to the election of 1963, I am quite confident of its merit. Nonetheless as I frequently observe to the media, it is not a crystal ball to predict the future, but rather a snapshot of the recent past.. I imagine, when we apply the model fully, I suspect we will show results comparable to those of Eric Grenier at threehundredeight.com, a great site I consulted regularly. Still, the discrepancies remind us that for pre-election modelling, a critical part of the science/art we pursue is how to determine criteria for selecting polls and time frames for inclusion. As I also say frequently, the projections are only as good as the polls upon which they are based. Andrea Perrella, the Director of LISPOP has suggested we have a conference on sat projections, perhaps in the fall, around the time of the Ontario election. Those interested in participating might contact Andrea at aperrella@wlu.ca or myself at bkay@wlu.ca .

Table 1- 2011 Final LISPOP seat projection compared with actual election results (in brackets)

  conservative ndp liberal bq
Other
Canada
144(167)
98(102)
52(34)
15(4)
(1)
Atlantic provinces
14(14)
7(6)
11(12)
--
--
Quebec
4(6)
49(58)
7(7)
15(4)
--
Ontario
56(73)
24(22)
26(11)
--
--
Prairies
22(25)
6(3)
3(2)
--
--
Alberta
27(27)
1(1)
--
--
--
British Columbia
21(21)
11(12)
4(2)
--
(1)

Table 2- 2011 Final pre-election % popular vote estimate by actual election popular vote (in brackets)

  conservative ndp liberal bq
Atlantic provinces
33%(37)
33%(29)
30%(29)
--
Quebec
16%(16)
38%(43)
14%(13)
26%(23)
Ontario
38%(44)
28%(26)
28%(24)
--
Prairies
49%(54)
25%(29)
19%(12)
--
Alberta
57%(67)
18%(16)
16%(9)
--
British Columbia
40%(46)
27%(32)
18%(12)
--

Note: The "regional swing model" is more fully explained in a paper originally prepared and presented by Dr. Barry Kay to the 1990 annual meeting of the Canadian Political Science Association, entitled "Improving Upon the Cube Law: A Regional Swing Model for Converting Canadian Popular Vote into Parliamentary Seats". It should be noted that the application of the model above does not make use of the "incumbency effect" described in that paper. In tests for past elections, using late campaign polls to project electoral outcomes, the model has proved to be accurate within an average of four seats per party since 1963. Readers interested in post-dictions for past federal elections dating back to 1963, for projections using pre-election polls dating back to the 1980 federal election and for three Ontario provincial elections, may contact me at bkay@wlu.ca.

Level of Government: