THE CANADA ELECTION NOWCASTER
———————————————————————————-
FINAL FORECAST FOR: OCTOBER 2019 (41.23%)
———————————————————————————–
Post-election update: The 2019 election has proven to be quite harsh for the model, producing its second highest error (around eight percentage points) over the 1953-2019 period. Although I will look more deeply into this in the coming weeks, the SNC-Lavalin scandal and the Bloc Québécois’ leader (Yves-François Blanchet) strong performance on the camapign trail appear to be the main explanations for the Liberals’ relatively poor performance. The model shows what could have happen under more “hospitable” conditions. Although a bad forecast is never pleasant for a forecaster, it allows us to underline particular features of an election that might explain why there is a wide margin between the forecast and the actual outcome.
———————————————————————————–
Election forecasting is now a thriving discipline in the United States, where a large number of different models are being used to forecast the outcome of congressional elections or the fate of presidential candidates. Although forecasting models have been developed for France, Germany and the United Kingdom over the past years, Canada, like most other democracies, has received very little attention. Consequently, we developed a theoretically-driven model that can be used to predict the popular vote share of the incumbent party in Canadian federal elections with sufficient lead time (i.e., a 3-month lag).
———————————————————————————-
The model is composed of five variables: (1) the difference between the unemployment rates in Canada and the United States three months before the vote; (2) the natural logarithm of the number of consecutive months the incumbent party has been in office; (3) a dichotomous variable related to the substitution of the Prime Minister near an election; (4) the number of years of political experience gained by the Prime Minister in relation to his/her main opponent; and (5) a factor related to the province of origin of party leaders. The equation of the model is the following:
———————————————————————————-
V = 59.83 + (–3.52 × U) + (–4.28 × M) + (6.44 × L) + (12.81 × Q) + (0.32 × E) + ε
R2 = 0.89; Adj. R2 = 0.86; SEE = 3.00;
DW = 2.72; N = 21 (1953−2015)
———————————————————————————-
where
V = | the incumbent vote share; |
U = | the difference between the unemployment rates in Canada and the United States three months before the vote (benchmark); |
M = | the natural logarithm of the number of consecutive months the incumbent party has been in office; |
L = | a dichotomous variable related to the substitution of the Prime Minister near an election scored 1 if the Prime Minister resigns before the election and 0 otherwise; |
Q = | the provincial origin of the Prime Minister in relation to the other party leaders coded -1, -0.5, 0, +0.5 or +1 depending on the situation; |
E = | the number of years of provincial and federal political experience gained by the Prime Minister in relation to his/her main opponent (i.e., the Liberal leader or the Conservative leader depending on the party affiliation of the Prime Minister); and |
ε = | an error term. |
———————————————————————————-
This blog presents nowcasts derived from our model from June 2017 (the first month following Andrew Scheer’s election as leader of the Conservative Party of Canada) to the next election which should take place in October 2019. Nowcasting consists in updating forecasts on a quarterly, monthly, or even daily basis by using the most recent values for each independent variable. Our own forecasts are also compared to the average of voting intentions.
———————————————————————————-
Assuming Justin Trudeau and Andrew Scheer will face off in the upcoming federal election, the final values of our five independent variables are already known and should remain the same until election day. In October 2019, the Liberal Party of Canada will have spent 48 months in power which means that the value of our time variable will be equal to 3.87 [ln(48)]. Since we are making nowcasts, the time variable will evidently vary from month to month (for example, in June 2017, the value of the time variable was equal to 3.00 because the Liberal Party had by then only spent 20 months in office). Unless Justin Trudeau resigns before the election, the dichotomous variable related to the substitution of the Prime Minister will be scored 0. When it comes to the political experience variable, Justin Trudeau is less experienced by about 4.30 years in comparison to his main opponent. Finally, because the Prime Minister is from Quebec, no other major party leader is from that province, and two minor party leaders are from Quebec (the leader of the Bloc Québécois and the leader of the newly founded People’s Party of Canada), the provincial origin variable takes a value of 0.50 (for more details on how this variable was coded, see Mongrain 2019). The economic benchmark (i.e., the difference between the unemployment rates in Canada and the United States) in July (i.e., the third month before the election month) was equal to two percentage points (5.7% – 3.7%). The value of the economic variable was updated every month according to the unemployement rates published by Statistics Canada and the U.S. Bureau of Labor Statistics.
———————————————————————————-
Model forecasts and voting intentions were also used to estimate the incumbent party’s percentage of seats using a swing ratio. A swing ratio is obtained by regressing the share of seats collected by a party on its share of the popular vote.
———————————————————————————-
Liberal Party’s vote share | Liberal Party’s seat share | ||||
Month, Year | Polls | Model | Polls | Model | |
June 2017 | 38.31 | 44.63 | 44.84 | 57.49 | |
July 2017 | 39.47 | 44.77 | 47.16 | 57.78 | |
August 2017 | 40.27 | 43.87 | 48.76 | 55.97 | |
September 2017 | 39.97 | 43.68 | 48.16 | 55.59 | |
October 2017 | 36.85 | 44.20 | 41.91 | 56.63 | |
November 2017 | 38.38 | 44.73 | 44.97 | 57.70 | |
December 2017 | 40.50 | 43.86 | 49.23 | 55.95 | |
January 2018 | 39.11 | 43.34 | 46.45 | 54.92 | |
February 2018 | 37.90 | 44.24 | 44.02 | 56.72 | |
March 2018 | 35.22 | 44.45 | 38.65 | 57.13 | |
April 2018 | 37.59 | 43.95 | 43.39 | 56.13 | |
May 2018 | 34.69 | 44.16 | 37.59 | 56.56 | |
June 2018 | 34.94 | 44.03 | 38.10 | 56.29 | |
July 2018 | 36.74 | 43.19 | 41.69 | 54.61 | |
August 2018 | 37.05 | 42.71 | 42.32 | 53.65 | |
September 2018 | 38.67 | 42.59 | 45.55 | 53.40 | |
October 2018 | 37.04 | 42.82 | 42.30 | 53.87 | |
November 2018 | 38.58 | 42.00 | 45.38 | 52.22 | |
December 2018 | 35.44 | 41.53 | 39.09 | 51.29 | |
January 2019 | 37.76 | 41.77 | 43.74 | 51.77 | |
February 2019 | 34.31 | 42.37 | 36.83 | 52.96 | |
March 2019 | 32.14 | 42.97 | 32.48 | 54.16 | |
April 2019 | 31.16 | 42.51 | 30.52 | 53.25 | |
May 2019 | 30.69 | 41.70 | 29.58 | 51.64 | |
June 2019 | 30.85 | 41.61 | 29.90 | 51.44 | |
July 2019 | 32.91 | 41.16 | 34.02 | 50.54 | |
August 2019 | – | 42.12 | – | 52.47 | |
September 2019 | – | 42.03 | – | 52.29 | |
OCTOBER 2019 | – | 41.23 | – | 50.70 |
Sources: Mongrain 2019 (for model data); Wikipedia 2019 (for vote intentions).
———————————————————————————-
Figure 1. Liberal Party’s Vote Share
———————————————————————————-
Figure 2. Liberal Party’s Seat Share
———————————————————————————-
This content has been updated on 20 December 2021 at 16 h 16 min.