A Technical Note on Our Final Seat Forecast

AND A BONUS BATTERY OF 2019 TRACKING ON VOTE INTENTION BY REGION AND GENDER

[Ottawa – October 23, 2019] Our final seat forecast was extremely accurate; however, our final four-day roll was somewhat off. In this brief addendum, we wish to explain that contradiction.

	LPC	CPC	NDP	BQ	GP	PPC	OTHER
Final projection:	158	117	31	28	3	0	1
Oct. 21 results	157	121	24	32	3	0	1

Recognizing the inherent instability of a four-day roll, and the need for more reliable regional data, our final forecast was not based on the four-day final roll, but a two-week roll-up of our polling. The overall distribution of that roll is presented below.

As one can see, this two-week roll was a more accurate depiction of the final outcome. The overall error on this roll was nine points as opposed the 14-point error that we had in our final roll.

We presented the four-day roll as our final estimate of vote share and we accept full responsibility for that. We are not trying to revise our estimate after the fact; that would be unethical. Our goal here is rather to explain the paradox of an extremely accurate seat forecast and a final roll that was off in some key regions. We were aware of those discrepancies and that is why we used the two-week roll. It appears in hindsight that there were not that many shifts in the final two weeks of the campaign.

We will be evaluating this more carefully with our exit polling.

We also include the tracking of all of the regions that we completed over the past several months to give a clearer picture of the rhythms of the campaign and pre-campaign period. While we would have preferred to have top levels of accuracy in both the final vote share and the final seat outcomes, we have always placed more emphasis on correctly predicting the overall outcome than the vote shares. It is the former which reveals the configuration of the next parliament and is inherently and more difficult challenge.