Introduction
The idea that incumbents hold an advantage in subsequent elections simply by virtue of their incumbency is a significant focus in political science research. Extensive research has consistently demonstrated that incumbents seeking re-election generally experience a substantial electoral advantage (Ade et al., Reference Ade, Freier and Odendahl2014; De la Cuesta and Imai, Reference De la Cuesta and Imai2016; Eggers and Spirling, Reference Eggers and Spirling2017; Hainmueller and Kern, Reference Hainmueller and Kern2008; Kendall and Rekkas, Reference Kendall and Rekkas2015; Lee, Reference Lee2008; Sevi, Reference Sevi2023).
Understanding how this incumbency advantage translates into greater vote shares and higher probabilities of winning elections is essential for shaping public policy. The existing literature suggests a number of possible explanations for the existence of an incumbency advantage including factors such as name recognition, established support networks, access to campaign funds and the ability to showcase legislative records (Fouirnaies and Hall, Reference Fouirnaies and Hall2014; Green and Sevi, Reference Green and Sevi2023; Kam and Zechmeister, Reference Kam and Zechmeister2013; Tolley et al., Reference Tolley, Besco and Sevi2022). However, the exact causes of incumbency advantage remain a puzzle. One important possible explanation is the availability of political cues on the ballot.
In this article, I leverage an amendment to the Canadian Election Act in 1970 that mandated the inclusion of candidates’ party affiliations alongside their names on the ballot from the 1972 election onwards, thereby potentially amplifying the role of party labels in the electoral process. Before 1972, the only information available to voters on the ballot were candidate names and their occupation (Courtney, Reference Courtney1974). Therefore, party labels were not available as a heuristic cue to voters unless they sought out that information prior to casting their ballot. Given that most voters tend not to seek out factual information about candidates and rely on heuristics instead (Sniderman et al., Reference Sniderman, Brody and Tetlock1991; Schaffner and Streb, Reference Schaffner and Streb2002), prior to this change, it is likely that party labels were not available to inform most voters’ decisions. Furthermore, historically, individual Members of Parliament enjoyed greater influence and visibility than in recent times, where parties have consolidated their power and discipline (Godbout, Reference Godbout2020; Loat and MacMillan, Reference Loat and MacMillan2015; Marland, Reference Marland2020; Savoie, Reference Savoie1999; Sevi et al., Reference Sevi, Yoshinaka and Blais2018). Consequently, it is possible that the incumbency advantage before 1972 relied on voters’ familiarity with individual candidates and their personal track records rather than party affiliations. However, since party labels are an extremely salient heuristic (Campbell et al., Reference Campbell, Philip and Warren1964; Downs, Reference Downs1957), their introduction on the ballot from 1972 onwards may have affected this dynamic by reducing the salience of the individual candidate, thereby reducing the incumbency advantage.
Thus, the 1972 change provides a clean test case to analyze the stability and evolution of incumbency advantages in a changing political context. In the Canadian context, understanding these distinctions is paramount because partisan affiliation wields substantial influence in elections and parliamentary dynamics (Merkley, Reference Merkley2021; Sevi et al., Reference Sevi, Yoshinaka and Blais2018). Using a comprehensive dataset of Canadian parliamentary elections from 1867 to 2021, I begin by analyzing the overall effects of incumbency over this period. I then explore differences in electoral dynamics before and after 1972, as well as between the two major political parties: the Liberal and Conservative parties. This approach contributes to our understanding of how political parties have evolved as influential gatekeepers in shaping parliamentary representation.
To identify the circumstances under which incumbency has an impact, I use a regression discontinuity (RD) design, with two distinct outcome variables from Canadian general elections: the probability of winning in the subsequent election and the percentage of votes received. RD design, pioneered by Lee (Reference Lee2008), confines the analysis to candidates who narrowly win or lose an election, thereby isolating the impact of incumbency. This approach is particularly advantageous as it approximates experimental designs, enabling causal inferences to be drawn from observational data (Lee, Reference Lee2008).
I find that, on average, candidates narrowly winning elections have a 10-percentage-point-higher probability of winning in the next election compared to those narrowly losing. Before 1972, this advantage is 15 percentage points, but decreases to 2 percentage points post-1972, becoming statistically insignificant. Before 1972, Liberal incumbents experience a significant advantage of 16 percentage points in their probability of winning, while Conservative incumbents show a non-significant advantage of 8 percentage points. After 1972, Liberal incumbents exhibit a marginal advantage of 1 percentage point in their probability of winning, whereas Conservatives face a penalty of 3 percentage points, both of which are not statistically significant at the conventional levels. The analysis of the incumbency effect on vote share yields similar results.
Data and Research Design
The data for this paper originates from Sevi’s (Reference Sevi2021) dataset on Canadian federal electoral outcomes and candidates’ sociodemographic characteristics since 1867. Key variables of interest in this research include candidate unique identifiers, party affiliations, election dates, riding, vote shares, and electoral success. Notably, the dataset uniquely assigns identifiers to each candidate, facilitating precise tracking of candidates across multiple elections.Footnote 1
I use an RD design to analyze the impact of incumbency status. RD designs are quasi-experimental methodologies suitable when treatment assignment is determined by a threshold on a continuous running variable. These designs concentrate on observations near the threshold (in this case, narrow electoral victories or losses) to simulate the conditions of a randomized experiment (Cattaneo et al., Reference Cattaneo, Idrobo and Titiunik2019).
I examine two distinct outcome variables: the probability of winning and the vote share, both of which range from 0 to 100. Therefore, the RD estimates can be interpreted as percentages. In the analysis, I use the margin of victory from the previous election as the running variable,Footnote 2 with incumbency status as the treatment variable. The unit of analysis is the individual candidate. The margin of victory for the winning candidate is calculated as the difference between the percentage of votes received by the winning candidate and the percentage received by the runner-up. Conversely, for the runner-up and lower-ranked candidates, the margin of victory is computed as the difference between the percentage of votes received by the candidate and the percentage received by the winning candidate. Consequently, winning candidates have a positive margin of victory, while losing candidates have negative margins of victory. In this context, a positive margin of victory denotes the treatment group (winners), while a negative margin of victory denotes the control group (losers).
Canada is a Westminster democratic system where Members of Parliament are elected using a single-member plurality voting electoral system. Under this system, voters select one candidate in their electoral district (riding), and the candidate with the highest number of votes wins the seat. Each electoral district corresponds to one seat in Parliament. From 1867 to 2021, a total of 44 federal elections have been conducted. Over this period, the average number of seats in the House of Commons was 246, starting at 121 seats in 1867, and increasing to 338 seats by 2021.
Between 1867 and 1920, Canada’s political landscape was dominated by a two-party system, with the Liberal and Conservative parties alternating in government and opposition roles. The Liberal Party tends to occupy an ideological position ranging between the left and center, whereas the Conservative Party tends to lean ideologically towards the right. After 1920, the political scene diversified with the emergence of additional parties, often characterized as insurgent, geographically concentrated or regional. Since Confederation in 1867, the Liberal Party has been the dominant force in Canadian federal politics, winning 26 out of 44 elections. In contrast, the Conservative Party has won 18 elections during this period. In Canada’s multiparty system, an incumbency advantage in vote share does not always translate into a higher probability of winning. This occurs because incumbents may gain votes from smaller parties rather than from their closest competitors.
I focus exclusively on general federal elections, excluding observations from by-elections in the dependent variables. However, I include by-elections in determining incumbency status. This results in 43,515 candidate-level observations and 10,840 elected members of Parliament (MPs). While party switching is infrequent in Canada (Sevi et al., Reference Sevi, Yoshinaka and Blais2018), I exclude all candidates who switch parties and subsequently contest another election. For all of the RD analyses, I compute cluster-robust standard errors using province and riding as clustering variables and include year and province fixed effects. I adopt established methodologies to estimate the individual incumbency advantage of candidates conditional on rerunning (Anagol and Fujiwara, Reference Anagol and Fujiwara2016; Kendall and Rekkas, Reference Kendall and Rekkas2015; Song, Reference Song2018; Uppal, Reference Uppal2009; Sevi, Reference Sevi2023).Footnote 3
For estimation, I use the standard framework for RD analysis, utilizing a local linear regression with a triangular kernel function. The local linear approach, characterized by a polynomial degree of 1, offers enhanced robustness compared to higher-order polynomials.Footnote 4 In kernel-based estimation, a weighted linear regression is fitted around the cut-off point, assigning greater weight to observations closer to the threshold while disregarding observations significantly distant from it (Cattaneo et al., Reference Cattaneo, Idrobo and Titiunik2019). The weighting scheme is selected through a data-driven process that optimizes the mean squared error of the local polynomial estimator, ensuring the bandwidth size is chosen objectively rather than arbitrarily (Cattaneo et al., Reference Cattaneo, Idrobo and Titiunik2019). To determine the precision of estimates, I use robust confidence intervals developed by Calonico et al. (Reference Calonico, Cattaneo and Titiunik2014).
Validity of Research Design
It has been argued by scholars that winners and losers in close elections may not be directly comparable, as incumbents typically benefit from greater resources and hold a systematic advantage (Caughey and Sekhon, Reference Caughey and Sekhon2011; Grimmer et al., Reference Grimmer, Hirsh, Feinstein and Carpenter2012; Snyder, Reference Snyder2005). However, Eggers et al. (Reference Eggers, Anthony Fowler, Hall and Snyder2015) conduct a comprehensive assessment of electoral RDs across approximately 40,000 closely contested elections in various electoral contexts across Canada and nine other nations. They conclude that there is no systematic evidence of strategic sorting or imbalance around the electoral cut-off.
Nonetheless, there are several standard tests that can be used to assess the validity of RD designs. In accordance with established practices, I first investigate whether individuals in the treatment and control groups are comparable in terms of observable characteristics in Sevi’s (Reference Sevi2021) dataset, particularly near the cut-off. The characteristics are the number of female candidates, the number of candidates and electoral experience, as indicated by the number of previous elections contested.Footnote 5 The balance tests across treatment and control groups reveal no evidence of discontinuity in any measured dimension. Specifically, candidates who narrowly won or lost elections exhibit remarkable similarity in these characteristics for elections situated close to the cut-off, suggesting that there are no imbalances in observable traits. The results of these robustness checks are presented in A4.
Another crucial validity check for RD designs involves examining the density of the running variable at the cut-off. Figure 1 illustrates the distribution of the running variable for candidates on either side of the cut-off. The density of treated and control observations at this point is found to be highly comparable. Thus, there is no indication of discontinuity or sorting effects around the cut-off.
Figure 1. Density of Elections with Narrow Winners and Losers.
Note: Density plots of the running variable for the control group (grey) and treatment group (black) based on the procedure developed by Catteno et al. (Reference Cattaneo, Idrobo and Titiunik2019). As can be seen from the plots, there is no discontinuity in the density around the cut-off (at margin of victory = 0)
Results
In this section, I begin by presenting general statistics for the entire sample to assess the impact of incumbency on election outcomes. Then, following Kendall and Rekkas (Reference Kendall and Rekkas2015), I examine whether there is evidence in the data supporting the presence of strategic exits among incumbents. This is an important validity check as strategic exits may bias the results. After reporting on these preliminary observations, I present the main results, starting with the RD analyses with probability of winning as the dependent variable and following with the analyses for vote share.
In the Canadian context, the probability that an incumbent will seek re-election in the subsequent election is 78 per cent since 1867, and 84 per cent since 1972. Among incumbents who seek re-election, the proportion winning the next election is 76 per cent since 1867 and 78 per cent since 1972. These statistics indicate that incumbent candidates are more likely than not to both run for re-election and to win.
While nearly 80 per cent of incumbents from 1867 seek re-election, strategic exits may present a potential challenge for the analysis. Specifically, if individual incumbents who opt not to run again are doing so because they are more likely to lose in the subsequent election, this could introduce bias into the findings. Following Kendall and Rekkas (Reference Kendall and Rekkas2015), I ran logistic regressions to test whether strategic exits could be a problem in the data. The rationale for this approach is that margin of victory from the previous election may be viewed as a proxy for an incumbent’s expectation of success in the next election. Therefore, strategic exit should manifest itself as a positive relationship between an incumbent’s margin of victory and their probability of running again. While this approach has its limitations (see Kendall and Rekkas (Reference Kendall and Rekkas2015) for a discussion), it is nevertheless a way to check for signs of strategic exit with available data. Table 1 presents the results of the logistic regressions. The coefficient for margin of victory is negative and not significant, suggesting that an individual’s decision to run again may not depend on past electoral performance.
Table 1. Effect of Margin of Victory at Time t– 1 on Decision to Run at Rime t for Incumbents
Note: Clustered standard errors in parentheses; clustering is by riding and province. Year and province fixed effects are included.
*p < 0.05, **p < 0.01, ***p < 0.001
Moving on to the RD analyses, I first present the results of the probability of winning estimation, followed by an examination of vote share outcomes. The analysis first shows the entire sample with all the parties followed by focusing on the two predominant political parties: the Liberals and the Conservatives. This selection is warranted as these two parties collectively secure 85 per cent of the seats over the entire study period, with the Liberals winning 46 per cent of the seats and the Conservatives 39 per cent.
Figure 2 plots the incumbency effects across the entire sample period, as well as pre-1972 and post-1972.Footnote 6 In the leftmost panel, we can see that candidates who narrowly win elections have a 10-percentage-point advantage in probability of winning the subsequent election compared to those who narrowly lose. This finding is consistent with the results of Kendall and Rekkas (Reference Kendall and Rekkas2015) and Sevi (Reference Sevi2023). The subsequent panels show that this advantage is 15 percentage points before 1972 and reduced to 2 percentage points after 1972. Table 2 presents the specific numerical values of the incumbency effects on the probability of winning, confirming that the discontinuities observed in panels 1 and 2 of Figure 2 are statistically significant at conventional levels.Footnote 7 Additionally, I analyze results separately for the Liberal and Conservative parties across the entire sample and both time periods. Before 1972, the Liberal party demonstrated a significant incumbency advantage of 16 percentage points, whereas the Conservatives held a nonsignificant advantage of 8 percentage points. After 1972, the Liberal Party maintained a marginal advantage of less than 1 percentage point, while the Conservative Party faced a penalty of 3 percentage points. Neither post-1972 statistic reached conventional levels of statistical significance.
Figure 2. Overall Incumbency Effect by Time Period.
Note: The dots show the quantile-spaced binned averages within the optimal bandwidth. The bins are calculated automatically using the rdplot software designed by Catteno et al. (2019). The lines plot the curve of the local linear regression using a polynomial fit of degree 1. Treated observations (where an individual is an incumbent) are located to the right of the zero and control observations (where an individual is not an incumbent) are located to the left of the zero. The cut-off is at zero margin of victory. The size of the discontinuity is the difference between the right-hand side and the left-hand side at the cut-off. Year and province fixed effects are included.
Table 2. Incumbency Effects on Probability of Winning by Time Period
Next, I investigate the association between vote share and the probability of winning, using vote share as the dependent variable.Footnote 8 Table 3 presents the results of incumbency effects on vote share across the entire sample period, as well as pre-1972 and post-1972. I also conduct separate analyses for the Liberal and Conservative parties across both the entire sample and these two time periods. For the entire sample, incumbents achieve an additional 4 percentage points in vote share. Before 1972, the overall incumbency effect amounts to 5 percentage points and is statistically significant at the 1 per cent level. However, following the changes from 1972 onwards, this advantage decreases to 2 percentage points and becomes nonsignificant. Examining the results for Liberals and Conservatives separately, before 1972, the Liberal Party gains a significant 5 percentage points in their vote share, while the Conservative Party achieves a substantial 7 percentage points increase, both significant at the 1 per cent level. Post-1972, the incumbency effect for the Liberal Party diminishes to 3 percentage points, which remains statistically significant at the 5 per cent level. Conversely, the Conservative Party experiences a 1-percentage-point increase in vote share post-1972, which is not statistically significant.
Table 3. Incumbency Effects on Vote Share by Time Period
These results highlight a clear trend in the evolution of incumbency effects in Canadian elections, particularly when considering the findings on both probability of winning and vote share. Before 1972, incumbents consistently enjoyed substantial advantages in both metrics, with significant increases in probability of winning and vote share observed across the entire sample and the Liberal and Conservative parties. However, after 1972, these advantages diminish notably. While incumbents still maintain some edge in vote share, especially for the Liberal Party, the effects are smaller and often not statistically significant for the Conservative Party. This trend suggests a shifting dynamic where the influence of individual incumbency has waned over time, potentially reflecting broader changes in voter behavior and party strategies in the Canadian political landscape.
Discussion
In this study, I use an RD approach to examine the incumbency advantage within the Canadian Parliament, focusing on both the enhanced vote shares and the increased probability of winning for incumbents. The analysis indicates that the incumbency effect was more pronounced in the pre-1972 period compared to the post-1972 period, which aligns with amendments to the Canadian Election Act. This finding underscores the need for further research into the magnitude and underlying reasons for this effect.
Prior to 1972, incumbents, particularly from the Liberal and Conservative parties, enjoyed significant electoral advantages. However, these advantages diminished considerably after 1972, suggesting a shift towards greater influence of party dynamics in election outcomes. The findings suggest that the decline in individual incumbency advantage may be linked to the increasing visibility and prominence of political parties.
Moreover, the results highlight a changing electoral landscape characterized by a decline in the influence of individual incumbency over time. This trend may reflect broader transformations in voter behavior, party dynamics and institutional structures, including increased party polarization, the centralization of Canadian political parties and the growing authority of party leaders. The diminishing significance of incumbency following the 1972 reforms highlights the rising importance of party affiliations and other electoral considerations in shaping election outcomes.
Future research could consider these alternative explanations and their potential impact on the observed trends. Disaggregating party and individual incumbency advantages (for example, as done in Kendall and Rekkas (Reference Kendall and Rekkas2015)) could provide a more nuanced understanding of these dynamics and clarify the factors influencing the changing nature of electoral advantages. Further investigating the reasons behind the decline in incumbency effects post-1972, such as shifts in voter preferences or party strategies, could yield deeper insights into evolving electoral dynamics. Comparative studies across different electoral systems may reveal whether similar patterns are present in other democratic contexts. Additionally, examining variations in incumbency effects across diverse regions or demographic groups within Canada could uncover important nuances in electoral behavior.
In conclusion, this study contributes to a nuanced understanding of the historical and contemporary roles of incumbency in Canadian politics. While it is commonly assumed that there exists a universal incumbency effect, the findings in this article suggest this assumption should be reconsidered. Specifically, the results indicate that the incumbency advantage may depend on institutional factors, such as the presence of party labels on ballots.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0008423925000058
Open access
