Parties and legislators in Canada

Modern parties in Canada play a much larger role in winning elections than they used to. Strong party discipline and patronage appointments are often sufficient to deter Members of Parliament (MPs) from criticizing and voting against their own party. Indeed, Godbout and Høyland (Reference Godbout and Højland2011) report that MPs now almost always vote along party lines. Those who do not toe the party line do so to the detriment of their political careers (Kam, Reference Kam2009). The party leader, in particular, has a lot of leverage and any MP with political ambition (for positions such as cabinet ministries, parliamentary secretariats or Senate appointments) or in need of constituency resources would do well to remain loyal to their party (Rossignol, Reference Rossignol1987).

Parties discipline disloyal MPs and isolate them from their colleagues. Common practices of this in the Canadian context vary from refusal to endorse candidates who have won their constituency nomination to expulsion from caucus (Kam, Reference Kam2009). Many of these dynamics help explain the very high levels of party unity on legislative votes in the Canadian House of Commons (Godbout and Høyland, Reference Godbout and Højland2011; Kam, Reference Kam2009).

This was not always the case. Party unity had not fully matured in the parliaments following Confederation (Carty, Reference Carty and Perlin1988). Many MPs were not tied to political parties. Those who frequently voted against their party on bills were nevertheless able to receive cabinet posts (Chhibber et al., Reference Chhibber and Kollman2009; Underhill, Reference Underhill1935). Godbout and Høyland's analysis (Reference Godbout and Højland2013) of the first ten Parliaments shows that party unity was weak initially and consolidated over time.

On the electoral side, the Elections Act was amended in 1974 to allow a candidate's party affiliation to appear with his or her name on the ballot, thereby increasing the importance of party labels in the electoral arena (Morton, Reference Morton2006). Canada's electoral system went from a candidate-centred system to a party-centred one, where the party leader and the label play a huge role in deciding the fortune of the individual candidate. Parties came to offer more coherent platforms as the era of brokerage politics was replaced by a pan-Canadian partisan context whereby parties developed national agendas and mobilized voters (Carty et al., Reference Carty, Cross and Young2000).

This dual transformation—in the legislature and in the electoral arena—contributed to the institutionalization of parties as exhibited by an increase in party voting in both the legislature and the electorate (Cox, Reference Cox1987). It is this institutionalization, we argue, that provides a mechanism through which we can make sense of the change in the extent to which legislators are willing to switch parties and the extent to which they are punished for doing so.

Legislative Party Switching in Canada and Around the World

Floor crossing in the contemporary era is the ultimate risky proposition. There are costs to switching parties, such as appearing to be driven by opportunism rather than principle (see Evans et al., Reference Evans, Peterson and Hadley2012), which may ultimately lead to electoral defeat (see Yoshinaka, Reference Yoshinaka2016). Nevertheless, party switching does occur in the Canadian context, and it can occur for a number of reasons, including political ambition (Snagovsky, Reference Snagovsky2015).

The recent defection of Eve Adams from the Conservative party to the Liberal party months before the 2015 federal election raised a number of questions about party switchers. Regarding her defection, Adams, who supported the Conservatives since she was 14, said she could no longer back the “mean-spirited leadership” of the Conservative party (O'Malley, Reference O'Malley2015). She announced her defection about a week after the Conservative party rejected her request for renomination because of a series of previous controversial nomination races. Despite the conversion, she was unable to win the Liberal local riding nomination.

There are other examples of political ambition-induced switching in the Canadian context. David Emerson, a former cabinet minister, was re-elected in 2006 as a Liberal in the Vancouver-Kingsway riding. He switched to the Conservatives just weeks later because he did not want to sit in the opposition backbenches (Docherty, Reference Docherty2011). The 2005 defection of Belinda Stronach from the Conservatives to the governing Liberal party two days before a crucial budget vote ensured the government's survival and she was promoted to cabinet right away. This led several male MPs to accuse her of “whoring” or “prostituting” herself for political ambition (Goodyear-Grant, Reference Goodyear-Grant2013).

But not all party switching is done in the service of political ambition. Switchers have been common in Canada during the formation of new parties. Scott Brison's decision to leave the Progressive Conservatives (PC) to join the Liberals was due to the merger of the PC and Canadian Alliance to form the Conservative party in 2004. Brent Rathgeber left the Conservative party in 2013 and lost the 2015 general election, as an independent candidate, citing lack of transparency and accountability (Harper, Reference Harper2013). Winning re-election as an independent is no easy task, but successful independents can hold the government's feet to the fire. In 2005, both the governing Liberal party and the opposition Conservatives rigorously lobbied Chuck Cadman, an independent MP who was in his final stages of terminal cancer, for his vote on the 2005 budget (Zytaruk, Reference Zytaruk2008).

Legislative party switching is common in many advanced democracies and even more so in consolidating democracies. In New Zealand, switching occurs so often that a law was enacted to mandate switchers to resign their seat (Geddis, Reference Geddis2002). Italy, Spain and Brazil each have politicians switching from one political party to another with relative ease (Desposato, Reference Desposato2006; Heller and Mershon, Reference Heller and Mershon2005; Mershon and Heller, Reference Mershon and Heller2003). Compared to these countries, the number of party switchers in the national legislatures of the United States (31 between 1961–2010) and of Canada (295Footnote ¹ between 1867–2015) is low. Their importance, however, both for politics (such as “pivotal” switchers like US Sen. Jim Jeffords in Reference Glasser, Kuzenski, Morelan and Steed2001, whose defection gave Democrats a majority in the Senate) and for our theoretical understanding of parties and political ambition, is no less evident. Each instance of cross-floor defection allows us to examine the extent to which this phenomenon has changed over time in both its incidence as well as its consequences for the re-election fortunes of those who make the jump. By going back to the early days of Confederation, our analysis sheds light on this phenomenon by juxtaposing all instances of party switching against changes in the party system.

Theory and Hypotheses

The extent to which changes in the nature of parties and the party system in Canada are associated with differences in party affiliation decisions of incumbent legislators is an open question. Studies of party switching both in Canada (Snagovsky, Reference Snagovsky2015; Snagovsky and Kerby, Reference Snagovsky and Kerbyforthcoming) and elsewhere (Desposato, Reference Desposato2006; Heller and Mershon, Reference Heller and Mershon2005; Kerevel, Reference Kerevel2017; Mershon and Heller, Reference Mershon and Heller2003; Yoshinaka, Reference Yoshinaka2016; Zielinski et al., Reference Zielinski, Slomczynski and Shabad2005) focus on legislative party switching during a relatively short time period, which precludes an examination of how longer-term changes in party systems affect the incidence and electoral consequences of party switching.

This paper fills a lacuna in the literature on party switching in the Canadian context by contributing to existing research investigating the evolution of parties and party systems. How do the costs of party switching change over time? How do floor-crossing MPs fare in Canada's evolving party system? We argue that an examination of the latter can illuminate our understanding of the former.

Canadian parties have changed since the early days of Confederation as they have become more institutionalized over time. By that we mean that politics both in Parliament and the electorate has been increasingly organized around party discipline and party-based voting (see Cox 1987 for an analogous treatment in the case of the British House of Commons). For one, party unity in the House of Commons is much higher now than it was in the nineteenth century and early twentieth century (Godbout and Høyland, Reference Godbout and Højland2017), a factor that is endemic to the concept of party institutionalization in world parliaments (see Thames, Reference Thames2007). Second, parties have differentiated themselves from each other in terms of platforms and group support in the electorate. According to Carty et al. (Reference Carty, Cross and Young2000), the eras of Confederation and the breakup of the historical Liberal/Conservative stranglehold on federal politics (roughly 1867 to 1920) and brokerage politics and the CCF (1920s-1960s) have been replaced by eras of pan-Canadian politics (1960s-1990s) and the rise of regional parties (post-1993). What the latter two eras share in common—and in contrast to earlier eras—is that parties develop “consistent and coherent messages” (Carty et al., Reference Carty, Cross and Young2000: 21) to appeal to voters with the use of polling and television. The emergence of the NDP as the party of labour “[formalized] the relationship between unions and the party” (Carty et al., Reference Carty, Cross and Young2000: 87), which provided yet another way in which parties distinguish each other.Footnote ²

The pattern through which parties gained prominence in Canadian elections (and elsewhere) is well described by Chhibber and Kollman.

Given Godbout and Høyland's findings (Reference Godbout and Højland2011) about party consolidation in the House of Commons, we hypothesize that the likelihood of party switching declines over time. This is what we would expect theoretically as the party system goes from a less consolidated one to a more institutionalized setting in which parties are paramount for re-election. Scholars who have examined party switching elsewhere find similar patterns. Nokken and Poole (Reference Nokken and Poole2004) report that party switching in the US Congress was much more prevalent prior to the emergence of the sixth party system with two relatively stable coalitions of Democrats and Republicans. Zielinski and colleagues (Reference Zielinski, Slomczynski and Shabad2005) also find that when the party system exhibits less rigidity it is accompanied by more legislators crossing the floor. It is therefore not a surprise that polities with very large numbers of switchers (such as Italy, Brazil, Japan) are also examples of less institutionalized party systems. Thus, in the Canadian case, we expect the incidence of party switching to diminish over time as parties become more institutionalized.

As parties became more prominent, we expect the decision to affiliate with one party—and to later shed that affiliation for another—to be more important at the ballot box. With the growing centrality of parties specifically when it comes to electoral politics, we hypothesize that the cost of defection increases over time. Conversely, when parties are less institutionalized, voters are less inclined to hold floor defectors accountable for their decision to defect. The outgoing party may also not be in a position to retaliate against a defector; in a more institutionalized setting, on the other hand, the “jilted” party will likely attempt to recapture the seat by expending more resources in the riding represented by the switcher. This is what Yoshinaka (Reference Yoshinaka2016) finds with respect to legislative switchers in the United States and this is what we hypothesize has happened in Canada in more recent decades.

Our hypotheses necessitate a longer examination of party switching than what has been done to date (see Snagovsky and Kerby, Reference Snagovsky and Kerbyforthcoming). In order to address these questions, we investigate the fortunes of switchers since Confederation by examining all electoral returns from 1867 to 2015. We separate the data into several periods to assess whether the costs of switching change over different time periods. We then look at whether the direction of the switch has any effect on switchers’ vote shares and whether that pattern is consistent with the evolution of the party system that we hypothesize is central to our understanding of the phenomenon.Footnote ³

Sitting politicians may switch for various reasons such as policy, office, and re-election (Snagovsky and Kerby, Reference Snagovsky and Kerbyforthcoming; see also Müller and Strøm, Reference Müller and Strom1999). When it comes to policy, an MP who is at odds with the party leadership's position on any given issue may decide that switching parties is a better option than staying put. Pertaining to ambition for high office (such as a cabinet position), an MP may view her path in her current party blocked for a number of reasons (Kam, Reference Kam2009); switching may allow her to fulfill this ambition (see Yoshinaka, Reference Yoshinaka2016, for an analogous argument with respect to US party switchers). Finally, an incumbent MP may decide that the path to re-election is made easier by shedding the current party label and attempting to win reelection under the banner of another party.

Before proceeding further, we note that there are other reasons that may lead to party switching, such as expulsion from the party, promotion to the Senate or diplomatic services, and party mergers. In this paper, we are interested in the consequences of an incumbent's decision to switch parties from one existing party to another. That is, we do not consider the decision to/from independent status or the “switch” in labels forced by the merger of existing parties. We do so for theoretical (in the case of party mergers) and empirical (in the case of independents) reasons.

Let us start with party mergers. In our model, we assume that MPs make an individual decision whether to switch or stay with the party. That is the only way that voters can “react” by withholding their support for that MP relative to other MPs who run for re-election under the same party label. In other words, if the MP “must” change labels due to a party merger, then that dynamic does not provide any information for voters to consider when deciding whether to support this or that MP under the new label; they all had to adopt the new label. Certainly, on an aggregate level party mergers may be met with positive or negative reactions from voters. However, individual MPs will not be punished or rewarded relative to other MPs of the same party since they all were affected by the merger. We are modeling the individual decision to switch and voters’ reaction to that decision.Footnote ⁴

We are excluding cases of switching to/from independent status (for example, due to expulsion) because of methodological considerations. Empirically, we establish a baseline of support for MPs of each party, which we use to measure the extent to which party-switching MPs’ support differs from non-switchers. When an MP switches to independent status, the concept of a party baseline no longer makes much sense. As we describe later, we control for a regional “normal” vote for each party by aggregating the votes of all candidates from that party in that region. For an independent MP, this would mean computing a regional “normal independent vote,” a quantity that would not be particularly useful given that the concept of such a normal vote for independents is dubious at best. We note, however, that our results hold when including independents in our analyses (see “Models with Independents” in the online-only appendix).

Data

We collected the data for this paper from the Library of Parliament of Canada. It covers all 42 Canadian federal parliamentary elections spanning 148 years. Our dataset includes observations for every incumbent MP who ran for re-election. This provides us with a total of 7674 candidate observations (3285 unique individuals).Footnote ⁵

Members of the Canadian House of Commons are elected through a first-past-the-post, winner takes all system. Apart from 60 historical constituencies where there were two seats in a single riding, each constituency in Canadian General Elections holds one seat in parliament. Additionally, candidates may also enter contests in multiple constituencies in any given election year, and may win and hold both seats in the corresponding parliament, though this has been very rare and it no longer occurs with any frequency. For simplicity, we removed the few cases where a single candidate ran in multiple ridings at the same time.

Variables

To investigate the questions above, we run a series of linear regression models where each case corresponds to a general election that an incumbent MP contests.Footnote ⁶ Our sample includes a total of 7,504 valid observations of incumbent MPs with a party affiliation running for re-election.Footnote ⁷ The dependent variable is the difference in the incumbent's vote share between the current and lagged election. Our main independent variable is whether the incumbent switched parties between the previous election and the current one. We also conduct a series of placebo tests in which the dependent variable is the difference between the lagged vote share and the double lagged vote share. These placebo tests allow us to rule out the possibility that MPs who switch parties had been returned to Parliament with fewer vote shares than those who remain loyal, which helps alleviate concerns regarding endogeneity. In addition, we estimate a dichotomous model where the dependent variable is a dummy variable indicating whether the candidate won in the current election. Table 1 describes our variables, and we discuss them below.

Table 1 Description of Variables

Incumbents’ Change in Vote Share

To identify incumbency, we first identify unique candidates and parties who run in consecutive general elections. Our main dependent variable is the difference between the incumbent vote share in the current and the previous election for the given incumbent. The mean change in vote share is –2.8 percentage points, with a minimum of –90.4 percentage points, a maximum of 55.4 percentage points, and a standard deviation of 13.7 percentage points.

Party Switching

We define a party switcher as an MP who changes party affiliations while in office. The analysis omits cases where an MP switches parties but switches back before the next election. Our data do not include politicians who take a break from politics and later on return as a non-incumbent candidate with a different party. As described earlier, we also exclude MPs who become independents or the reverse, and we treat those who change party affiliation due to party mergers as non-switchers.Footnote ⁸

Party System Periods

We divide our data into four periods: 1867–1917, 1921–1958, 1962–1988 and 1993–2015 to coincide with transformations in the Canadian party system (see Carty el al., 2000). We look at these four periods to place switching in context and to examine if the costs of switching vary over different time periods. Switching mostly occurs when parties undergo periods of transition, so situating them under the aforementioned periods allows us to better understand how the dynamics specific to each party system may have affected the extent to which switching parties is costly for an incumbent MP.

The first period is marked as the period of loose fish and patronage appointments. The end of the first system broke off the two-party competition among Confederation's founding historic parties, the Liberals and the Conservatives (Carty, Reference Carty1992). The second period was an era of regional brokerage parties which were unable to represent all areas of the country at the same time (Cairns, Reference Cairns1968). The electoral system exacerbated regionalism because it led to a disproportionate allocation of seats to the strongest party in each region. The third period saw the merger of the CCF and the NDP and changes to the electoral ballots that made elections significantly more party centred; and the fourth saw an era of minority governments, instability and entry of the Bloc and merger of the Reform Party and Canadian Alliance, two right-wing parties, to create the Conservative Party of Canada. Each of these periods saw significant changes in the Canadian party system and in the way elections are conducted. These transformations include the introduction of secret ballots, same-day election, redistribution of seats, expansion of suffrage, fixed election dates and party financing laws.

Party Affiliation

Until the 1990s, the Liberals and Conservatives were the two dominant parties in the Canadian House of Commons. The Liberal party has generally been a centrist party seeking to maximize its support by appealing to voters of all stripes and is often characterized as Canada's brokerage party, whereas the Conservatives are ideologically to the right (Carty, Reference Carty, Bittner and Koop2013). Both parties have existed since Confederation, but sometimes they have existed under different names. For the purposes of this study, we have treated the following parties as all being the same Conservative party: Conservative, Conservative Party of Canada, Government, Liberal Conservative Coalition, Liberal-Conservative, National Government, National Government party, PC party, Progressive Conservative, Progressive Conservative party, Unionist. Likewise, Liberal, Liberal Party of Canada, Liberal Progressive, Opposition, Opposition/ Laurier Liberals are coded as the Liberal party. The New Democratic party (NDP) is a left-leaning party created in 1961. Its predecessor, the Cooperative Commonwealth Federation (CCF) existed from 1931–1961. For the purposes of our analysis, they are treated as belonging to one party category. The Bloc Québécois (a nationalist party that runs candidates only in Quebec), and Reform party (a Western-based regional party) have both presented significant challenges to the two major parties and as such they are included in our analysis separately. All other parties are combined in a “third party” category. Parties that run fewer than ten candidates are coded as belonging to the independent category, and are not included in our analysis.

Control Variables

In order to account for the electoral environment in each riding, we include each incumbent's vote share in the preceding election (which also allows us to account for ceiling effects), as well as two variables measuring the regional support for the incumbent's party in both the preceding election and the current election. This accomplishes several aims. First, by controlling for party support in the preceding election, we allow for the possibility that elections follow a cycle whereby a party that does exceptionally well (or poorly) in one election typically will do worse (or better) in the following one. Whether this is due to a regression to the mean or some other ebb-and-flow dynamic is not important for our purposes. The important point is that it allows us to control for this inter-election dynamic. Second, by controlling for the regional party support in the current election, we are essentially removing from the dependent variable that part of the variance that is common across all ridings in the same region. We are therefore controlling for each incumbent's underlying party support. For the purposes of our models, we compute party support in each of the following regions: Atlantic, West, Prairie, Ontario, Quebec, and North. Each riding's party support is therefore the average party vote in the region in which the riding is located.Footnote ⁹

We also include party dummy variables and a time variable equal to the number of years between the current election and 1872. This will allow us to test for time-based dynamics across our entire sample with the inclusion of a multiplicative interaction term between time and switching. We also test for these dynamics by separating the dataset into four distinct periods detailed above, which does not assume that the changes in the electoral fortunes of party switchers follow a linear pattern.

Findings

Descriptive statistics

Our analysis contains 72 unique MPs who switched parties from one party to another and ran for re-election. Of these MPs, 67 switched once and five crossed the floor twice for a total of 77 switches. Incumbent switchers who run in the subsequent election win their bid more often than not, though a nontrivial number (27) were not re-elected in the election immediately following their switch.

Table 2 shows the performance of switchers in each period. As seen in the table, the proportion of winning switchers decreases over time. Whereas in the first two periods the vast majority of switchers are re-elected, in the latter two periods the pattern is reversed, with a relatively small share of switchers re-elected. Overall, 30 switchers switch to government, and 47 switch to opposition (see Table 3).

Table 2 Party Switcher Performance by Party System Period

Table 3 Frequency of Direction of Switch over Time

We also looked at party switching among female MPs specifically. The number of incumbent women in our analyses steadily increases over the party system periods. Nevertheless, the overwhelming majority of incumbents are men. Despite the increasing number of women in Parliament, the sample of female switchers remains extremely small. There are only six female switchers included in our data and they are all in the last period, between 1993 and 2015 (see Table 4). We wanted to investigate whether female party switchers are treated differently than their male counterparts but due to the fact that women switchers are only present in the last period and because we only have six, once we remove independents, our sample is too small to draw any real significant inference and hence we invite more women party switchers for a further study. We also note that the proportion of women who leave their party to become independents (40%) is quite high. Since we exclude them from our study, these cases are not accounted for. However, this may suggest that the party affiliation dynamics differ between women and men, with parties perhaps less likely to accommodate a woman MP who seeks another party label. We leave this investigation for further research.

Table 4 Female Switchers

Four additional incumbent women, who all left a party to become independent, are not included in our analysis. They are: Suzanne Blais-Grenier who left the Conservative party; Bev Desjarlais who quit the CCF/NDP; Louise Thibault who was initially with the Bloc and Helena Guergis who left the Conservatives. All their post-election vote shares were, unsurprisingly, quite small and all lost reelection.

General effect of switching

The first six models summarized in Table 5 measure the effects of switching parties on electoral performance. The first model is a general model of the effect of party switching during the entire era. The second model adds an interaction with time and the next four models break the data up into each of the four party system periods. The general model (first column) shows a 5-percentage point penalty (p < 0.05) for incumbent switchers between 1872 and 2015, even after controlling for each incumbent's previous election vote share and the party's regional support in both the preceding and current elections. This result clearly supports our hypothesis that party switchers pay an electoral price at the ballot box. The average effect over the entire time period is substantively large as a 5-percentage point shift could result in up to a 10-point swing between the incumbent and the closest opponent.

Table 5 The effects of party switching on changes in incumbent vote share: OLS estimations

***p < 0.001, **p < 0.01, *p < 0.05, ^ap < 0.1. The numbers are regression coefficients with clustered standard errors in parentheses.

How does this penalty change over time? To answer this question, we first include a multiplicative interaction term between time and switching on the right-hand side (Model 2). The results of this specification show that switchers did not fare significantly worse than non-switchers in the early years of Confederation (with even a slight positive effect very early on). A few decades later, however, a clear party-switching cost starts to emerge and, by the end of the time series, party switchers are expected to receive close to 20 percentage points less than non-switchers. Figure 1 shows how the costs of party switching evolved over time and how switchers today face a much steeper road to reelection compared to switchers in the nineteenth century.

Figure 1 The estimated impact of switching over time. Estimations based on Model 2 in Table 5.

While the interaction with time clearly shows an increase in the costs of switching, our previous specification assumes that these changes are linear. We next break the data up into four discrete time periods to examine the extent to which switchers do worse (or better) than non-switchers in each of the four eras we identified earlier. Results from Models 3 through 6 show that in the first two eras (covering almost the first century of Confederation) party switchers did no worse than non-switchers. In the last two periods, however, we see a clear difference with party switchers faring significantly worse than non-switchers. These results are consistent with our hypothesis about the effects of party system dynamics on the electoral fortunes of party switchers. The electoral costs of switching emerge as parties gained prominence in elections and in voters’ decision-making.

Two other sets of results deserve to be mentioned. First, we see clearly that the lagged incumbent vote has a large negative effect on change in the incumbent's vote share. This reflects the fact that those incumbents who had the largest vote shares in the previous election were likely to lose more votes in the subsequent election, as it becomes almost impossible for a candidate with many votes in the previous election to do better in the next election.Footnote ¹⁰ Second, we also see that the party-based regional votes have a much larger impact in more recent periods. This is consistent with the notion that parties are more institutionalized today and that candidates running under a party label depend much more on that party's reputation and apparatus than before. In an era in which elections have become more party-centred, we find that the party variables are much more predictive of each incumbent's vote shares. It is in this context that we find that party switchers pay especially high costs when asking voters to return them to Parliament under a new party banner.

Effects of the direction of switch

While the preceding results provide support for our hypotheses, they do not tell us whether switchers who switch to a government party, for instance, are punished less (or more) than those who join an opposition party, and more importantly whether this difference changes over time. This is an important consideration to examine further because if the electoral effect changes over time differently for each type of switch, it would tell us that our finding that the cost has increased is driven primarily by one type of switch or the other.

Theoretically, the punishment could be larger or smaller depending on the direction. On the one hand, joining the government may be beneficial as MPs who join the government may be better able to represent their ridings either because they are now in cabinet or at least belong to the party that is in the majority (see Grose, Reference Grose2004, for a similar argument with respect to US state legislators who join the majority). On the other hand, it is possible that joining the government will be seen as an act of naked ambition viewed in a negative light by voters (see Evans et al., Reference Evans, Peterson and Hadley2012). Or perhaps joining the government will lead to electoral retribution at the individual level in a similar way that Narud and Valen (Reference Narud, Valen, Strøm, Müller and Bergman2008) show that parties that join governing coalitions tend to do worse on aggregate in the next election (see also Müller and Strøm, Reference Müller and Strøm2000; Rose and Mackie, Reference Rose, Mackie, Daalder and Mairs1983). We do not necessarily anticipate one type of switch to be punished at a higher rate, and more importantly, we have no reason to believe that the change over time in the electoral effect will differ based on the direction of the switch.

To investigate the effects of the direction of switch, we re-estimated the general model from 1867–2015, but we substituted the switching dummy variable with two dummy variables indicating the government/opposition status of the party that the switcher joined and we interacted each dummy with time. Each switch is defined as a switch to a government party or to an opposition party based on the status of the new party with which the switcher affiliates. The reference category as always corresponds to non-switchers. Model 7 in Table 5 shows the results of our analyses. Both types of switch display the same general pattern we found earlier, with negative interaction terms that indicate that switchers regardless of government/opposition status fare less well in more recent years than they did in the past. Figure 2 shows this; the slopes of both lines do not significantly differ from each other (the two interaction coefficients are not statistically different from each other with a p-value of the difference of 0.24). Our results also indicate that both types of switching may have been more beneficial in early years, but over time an electoral cost clearly emerges. Overall, the patterns are similar for both types of switchers. These results alleviate any concern that our earlier findings were largely due to one type of switch. Regardless of the government status of the receiving party, the results show that party switchers pay a significant electoral price for the defection, and that this penalty has increased over time.

Figure 2 The estimated impact of two types of switch over time. Estimations based on Model 7 in Table 5. The blue line represents switch to opposition and the pink line switch to government.

Placebo and robustness tests

With these observational data, the question still remains whether party switchers may have been less safe than non-switchers to begin with, which might explain their lower vote shares as they stand for re-election. This does not seem to be the case since switchers’ mean vote share in the election that preceded their switching is 51.2 per cent, which is basically the same mean vote share (52.7%) among non-switchers. Yet, we explore this possibility further by conducting a placebo test in which the vote share in the difference in result between the preceding and the penultimate elections is regressed on the same party switching and control variables (with every variable including the incumbent's previous vote shares lagged by an additional time period). If switchers are in a precarious position to being with, it would very likely show up in this analysis with a negative coefficient on the party-switching variable. Table 6 shows that this is not the case. In all the models save one, there is no evidence that switchers had performed more poorly in the previous election. The only exception is for the second period, but this does not represent a problem for our earlier results since that was not an era during which we had found a party-switching effect. What matters is that the eras in which we had found an effect as well as the overall time period do not show any negative placebo effects. This placebo test serves to bolster the confidence in our result that switchers were not significantly more vulnerable than non-switchers prior to changing their party affiliation.

Table 6 The Placebo Test: Changes in Incumbent Vote Share between the previous and the penultimate elections: OLS estimations.

***p < 0.001, **p < 0.01, *p < 0.05, ^ap < 0.1. The numbers are regression coefficients with clustered standard errors in parentheses.

We also performed a robustness test by estimating logistic regressions to check if the results are similar when the dependent variable is whether the incumbent did or did not win the election, with the same variables on the right-hand side. In all the analyses, the results hold even when the dependent variable is changed to a dichotomous won/lost variable. The pattern in Table 7 is very similar to that of Table 5, further strengthening our findings.

Table 7 A Robustness Test: Winning as the Dependent Variable (logistic regressions)

***p < 0.001, **p < 0.01, *p < 0.05, ^ap < 0.1. The numbers are logit coefficients with clustered standard errors in parentheses.

Finally, we conduct one other series of robustness checks by estimating random-effects models that account for the fact that many incumbents appear more than once in the dataset. By allowing for an MP-specific random coefficient, our models no longer assume that each individual observation is independent. Rather, it takes into account of the possibility that the error term for each MP is not independent across all elections featuring that MP. The results of our random-effects estimations are shown in Table 8, and they are quite consistent with our earlier results: party switchers pay a higher cost in more recent decades. Our earlier results are not an artifact of the panel structure of the data.

Table 8 Changes in Incumbent Vote Share: Random Effects Models (for each individual MP)

***p < 0.001, **p < 0.01, *p < 0.05, ^ap < 0.1. The numbers are regression coefficients with clustered standard errors in parentheses.