NDP Rebounds as Polls Vary Widely

In conjunction with the following LISPOP seat projection, it must be observed that the current crop of polls utilized varies well beyond appropriate estimates that should be based upon the assumptions of probability samples. Although the entire cluster is drawn from a four-week period published between Sept. 19 to Oct. 17, in the key province of Ontario polls published 10 days apart range from a 20 percentage-point Liberal lead over the Conservatives, to a two-point Conservative lead. Likewise, also over a 10 day period, polls of Quebec vary from a 14-point Liberal lead over the NDP to a 10-point NDP lead. This phenomenon of poll results "all over the place" is occurring with increasing frequency, and is why LISPOP is choosing to incorporate more polls in projections even if they range over a longer time period. It is intended to mitigate the impact of outliers. The following seat projection is drawn from surveys conducted by Angus Reid, Ekos, Forum Research and Abacus among an aggregate sample of 6800 drawn from a mixture of different methodologies. The actual vote preference averages used are a 10-percentage point Liberal lead in Ontario, and a one-point Liberal lead in Quebec. 

Projected distribution of seats by party and region compared with actual election results (in brackets), released October 24, 2014

  liberal conservative ndp bq
Other
Canada
136(34)
114(166)
85(103)
2(4)
1(1)
Atlantic provinces
21(12)
7(14)
4(6)
--
--
Quebec
30(7)
7(5)
39(59)
2(4)
--
Ontario
63(11)
40(73)
18(22)
--
--
Prairies & North
8(2)
14(26)
9(3)
--
--
Alberta
2(0)
31(27)
1(1)
--
--
British Columbia
12(2)
15(21)
14(12)
--
1(1)

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: