Cues and heuristics have been studied extensively in the past decades and have been shown to help citizens make up their minds when making political choices (Campbell et al., Reference Campbell, Converse, Miller and Stokes1960; Sniderman et al.,Reference Sniderman, Brody and Tetlock1991; Popkin, Reference Popkin1991; Lupia, Reference Lupia1994; Lupia and McCubbins, Reference Lupia and McCubbins1998). Given that ordinary citizens are poorly informed about politics (Converse, Reference Converse and Apter1964; Delli Carpini and Keeter, Reference Delli Carpini and Keeter1996), one of the main objectives of research on cues and heuristics was to rehabilitate citizens by showing that no matter how uninformed they may be, they were nonetheless able to come to reasonable political decisions. Cues and heuristics aside, adherents of the “aggregationist” argument tried to rescue the entire public by arguing that individual errors caused by information biases cancel out in the aggregate, meaning that democratic decisions of the entire electorate should be considered informed even if individual citizens are not (Erikson et al., Reference Erikson, MacKuen and Stimson2002 ; Miller, Reference Miller, Grofman and Owen1986; Page and Shapiro, Reference Page and Shapiro1992).
While cues and heuristics are certainly useful to citizens, it is now acknowledged that they cannot compensate fully for a lack of political information, and that individual errors do not simply cancel out during the aggregation process (Althaus, Reference Althaus2003; Bartels, Reference Bartels1996, Reference Bartels2008; Blais et al., Reference Blais, Gidengil, Fournier and Nevitte2009; Gidengil et al., Reference Gidengil, Blais, Nevitte and Nadeau2004; Hansen, Reference Hansen2009; Oscarsson, Reference Oscarsson2007; Tóka and Popescu, Reference Tóka and Popescu2007). These various studies came to this conclusion by simulating votes or opinions of uninformed citizens using votes and opinions of their more informed and sociodemographically similar counterparts. Comparing the real opinions or votes to the simulated ones then allowed the researchers to look at the magnitude of the effect of information biases on vote choice and opinions as well as their impact on collective preferences in the aggregate. The main reason that individual errors—opinions or vote choices that are different than they would have been were citizens fully or more informed—do not simply cancel out in the aggregate is that errors are not randomly distributed in the population. Uninformed citizens are more prone to make errors, and since political knowledge is highly related to variables linked to social status (Delli Carpini and Keeter, Reference Delli Carpini and Keeter1996), and demographic variables are also known to be related to vote choice and opinions, individual errors in political decision making are much more likely to occur among the least affluent. Bartels (Reference Bartels2008) has argued that this leads to a vicious circle in which the least affluent citizens are less well represented because they are more likely to make mistakes, which results in policies benefiting the most privileged, rather than themselves. This, in turn, makes it even more unlikely that the least affluent move upward in the social ladder, which pretty much ensures that they remain less informed and continue to make costly errors.
Yet, although information effects are certainly normatively important because we intuitively expect that informed choices and opinions are “better” than uninformed ones, the lack of information may manifest itself in other ways as well. When lacking crucial information such as meaningful partisan cues, citizens may be more prone to be swayed by cognitive biases. In the voting booth, a simple and easily detectable cognitive bias is the impact of the serial position of the candidates on the ballot paper. Similarly, since in most polities candidates are alphabetically ordered on the ballots, serial order effect may also be the product of citizens’ alphabetical preferences. In either case, though, alphabetical preferences or “true” ballot order effect are both completely irrelevant criteria to choose one candidate over another.
As we will see, past studies have investigated ballot order effect, as well as alphabetic preferences bias, but these studies are mostly focused on the US. To our knowledge, no study on the topic has been conducted in Canada. We examine how variations in partisan cues stemming from different types of elections affect the prevalence of ballot order effect in the province of Québec. More specifically, we look at two very different settings: provincial elections, in which party cues are strong, and municipal elections in which party cues are rather weak. Our main expectation is that elections with weak party cues do not provide voters with the required information to overcome simple cognitive biases, hence making it more likely that irrelevant factors such as serial order effect—or alphabetical preferences—will impact their vote choice. On the other hand, when cues are sufficiently strong, such biases should not have a meaningful impact. Should our expectations be sound, this would highlight an important avenue for improving democratic elections in settings where political parties are not coherently structured.
Party Cues and Ballot Order Effect
Shortcuts and heuristics are typically understood as pieces of information that, although somewhat superficial and incomplete, allow one to adopt behaviours that emulate or resemble behaviours that would stem from more complete knowledge. For instance, Lupia (Reference Lupia1994) has shown that in the absence of clear party cues during a referendum on auto insurance policies in California, simple awareness of the positions taken by stakeholders helped otherwise uninformed voters to emulate the votes of more informed citizens. Sniderman and colleagues (Reference Sniderman, Brody and Tetlock1991) proposed the likability heuristic, which posits that uninformed citizens are able to correctly attribute various political positions by relying on their positive or negative affect—likes and dislikes—towards various groups. Obviously, knowing where one's preferred party stands on various issues can also help citizens form opinions without the need for complete information about why a position is better than another. The idea that citizens can rely on party cues is hardly new. Campbell and colleagues (Reference Campbell, Converse, Miller and Stokes1960) demonstrated early on that voters indeed do use partisan cues when making up their mind about various issues. Besides political opinions per se, one instance in which party cues are certainly crucial is vote choice (Downs, Reference Downs1957). Most voters are unlikely to possess meaningful knowledge about the issue positions of most candidates, and in some instance, voters may not really hold any relevant information about candidates besides their affiliated party. In the absence of such party cues, ambivalent voters should be very likely to be influenced by very simple cognitive biases. The serial position effect is a well-known phenomenon in psychology (Cowan et al., Reference Cowan, Scott Saults, Elliott and Moreno2002). When presented with a list of items, an individual is more likely to remember the item at the beginning (primacy effect) or at the end of the list (recency effect). Such effects have namely been observed on persuasion (Haugtvedt and Wegener, Reference Haugtvedt and Wegener1994; Li, Reference Li2010), and impression formation of other people (Forgas, Reference Forgas2011; Webstera et al., Reference Webstera, Richterb and Kruglanskib1996). This bias is also well known by survey professionals as response orders within web surveys are known to affect respondents’ answers (Krosnick and Alwin, Reference Krosnick and Alwin1987; Malhotra, Reference Malhotra2008). Krosnick and Alwin (Reference Krosnick and Alwin1987) used a survey experiment to show that individuals with lower “cognitive sophistication”—those with lower vocabulary scores and education—are more likely to be influenced by response order in surveys. Although voting is arguably very different from answering survey items because the former is much more consequential than the latter, it remains that the action itself is similar. For that reason, some scholars have looked at the possible relationship between ballot order and vote choice.
The impact of primacy effect on ballot order has first been tested and shown in studies conducted during US elections in the 1960s (see Miller and Krosnick, Reference Miller and Krosnick1998). These studies were, however, criticized because the candidates’ position on the ballot was not randomized. In these studies, researchers assessed, when looking at a large number of elections, whether candidates fared better or worse depending on their position on the ballot. Miller and Krosnick (Reference Miller and Krosnick1998) critiqued this design because it may be assessing an alphabetic-based name preference instead of a true serial order effect. Recently, some polities modified their election regulations to introduce randomization of candidates order on their ballots. These new regulations offered an opportunity to examine the ballot order effect using more convincing designs. Studies conducted in New York City (Koppell and Steen, Reference Koppell and Steen2004), Ohio (Miller and Krosnick, Reference Miller and Krosnick1998), California (Pasek et al., Reference Pasek, Schneider, Krosnick, Tahk, Ophir and Milligan2014) and North Dakota (Chen et al., Reference Chen, Simonovits, Krosnick and Pasek2014) all took advantage of some form of randomization, and all confirmed the effect of ballot order. Two studies from Ho and Imai (Reference Ho and Imai2008; Reference Ho and Imai2006) reached similar conclusions, mostly for minor party candidates, confirming the moderating effect of party cues. Another study from Kim and colleagues (Reference Kim, Krosnick and Casasanto2014) also pointed to more moderators, like information availability, voters’ cognitive skills and effort devoted to candidates’ evaluation. Hence, the literature confirms that ballot order has an impact stemming from the biases known as the primacy effect. Nevertheless, most of this research is based on American electorates which, considering the various specificities of the American democracy, may be somewhat different than other electorates. Moreover, ballot order effect has not been studied in Canada, where we also tend to devote much more attention to federal and provincial elections, which, albeit important, are only two of the multiple settings in which elections take place.
Hence, the main goal of this article is to evaluate the prevalence of ballot order effect on a non-American electorate, and to see if it varies depending on the strength of party cues that an electoral setting provides. Our main hypothesis is simple: in the absence of sufficient cues to make a political decision, most voters should remain ambivalent and therefore be more likely to be influenced by cognitive biases such as ballot order effect. In other words, in the absence of an informational shortcut, people who are ambivalent about their vote choice should not be expected to vote randomly among all the possible candidates but to vote randomly while still being subjected to this well-known cognitive bias. In this article, we intend to study two settings. The first one, provincial elections in Québec, is characterized by a very strong party cue. Our second setting, municipal elections in Québec is exactly the opposite. It offers, for the most part, a very weak party cue. This difference leads to these two hypotheses concerning the effect of ballot order:
H1:
The ballot rank of candidates will affect their vote share.
H2:
The relationship stated in H1 will be moderated by the election level.
The next section describes in detail the chosen settings.
The Setting: Québec Provincial and Municipal Elections
Our analysis uses results from provincial and municipal elections in Québec, Canada.Footnote 1 This setting offers two main advantages. First, the two municipal elections (2009 and 2013) and three provincial elections (2008, 2012 and 2014) were held in a close enough time frame to allow us to compare their electorates.
The second advantage of this setting relates to the unique Canadian party system. While in most advanced democracies political parties compete in elections at all levels or office positions, this is not the case in Québec since the party systems at the different levels of governments are distinct. Even parties who share the same name mostly do not have any institutional relations with each other.Footnote 2 Furthermore, the Québec municipal party system is very limited and many municipalities have elections in which all candidates have no party affiliation. While in some cases candidates for municipal office are former federal or provincial MPs, the municipal party for which they are running—when they have one—has no ties whatsoever with the federal or provincial party with which they were originally affiliated.
For instance, the city of Longueuil, on the south shore of Montréal, is the fifth most populous city in the province (following Montréal, Québec City, Laval, and Gatineau). Yet, since 1848, its mayor has had a party affiliation only four times. In the last decades, the city's mayor has been Claude Gladu (1994 to 2001, without party affiliation), who was followed by Jacques Olivier from 2001 to 2005. Olivier, a former Liberal MP from 1972 to 1984, became mayor in the aftermath of large scale municipal mergers in the province; he had no party affiliation and was closely affiliated with Gladu who was on his team. The latter won the 2005 election, this time as the leader of the “Parti municipal Longueuil” (Longueuil Municipal Party), which could hardly be a more generic name for a municipal party. In 2009, Caroline St-Hilaire became the city's mayor after leaving federal politics where she had been a Bloc Québécois MP for many years. The “Action Longueuil, équipe Caroline St-Hilaire” party was created at that time. Hence, her party provided few cues to voters besides its clear affiliation with her.
The only exceptions are Montréal and Québec City, the provincial metropolis and capital, where the party system is structured in a more cohesive fashion. Yet, this is mainly due to the visibility that these mayors enjoy in the media. Hence, in Montréal and Québec City, party cues are somewhat stronger, but these cues are heavily dependent on the visibility of the leader. This is perhaps best exemplified by the fate of municipal parties when their leader leaves municipal politics. Pierre Bourque was Montréal's mayor from 1994 to 2001, he led the “Vision Montréal” party, created in 1994 a few months before the election. He lost to Gérald Tremblay (former Québec liberal MP) in 2001 who led “Union Montréal, équipe Gérald Tremblay.” Both parties have now disappeared from municipal politics in Montréal. Denis Coderre (who also happens to be a former federal liberal MP), has been elected mayor in 2013 with his new party “Equipe Coderre pour Montréal” (Team Coderre for Montréal). Such a party label leaves little doubt as to the fate of the “party” should Coderre decide to leave municipal politics. Obviously, his current team may very well unite under a new banner, but that new party would hardly provide any meaningful cues to voters unless the new leader has strong name recognition.
Hence, even though municipal elections do matter since decisions at this level have direct and visible consequences in citizens’ lives, party cues for most municipalities are weak or almost non-existent. Many parties do exist in cities outside Montréal and Québec City, but they are usually created by mayoral candidates for the purpose of a few elections. In these cases, it is impossible to develop a party identification because the party typically disappears when its leader retires from politics. Accordingly, elections at the provincial and municipal levels vary greatly in terms of party cues, and allow us to compare electoral behaviour within two different settings. To sum up, using provincial and municipal elections in Québec to study ballot order gives the opportunity to look at two types of non-simultaneous elections in which voters have a very limited number of choices to make and in which party cue strength differs.
The names appearing on the ballot in Québec are not randomized as they are listed in alphabetical order. As we said earlier, it was argued that some past observational studies on ballot order have found the effect of alphabetical preferences instead of a real ballot order effects (Miller and Krosnick, Reference Miller and Krosnick1998). We acknowledge that a setting in which names were randomized would be best to identify a causal mechanism, but we nonetheless believe that the benefit of the setting we use outweighs this limitation. First, there is no randomized ballot in Canada. Therefore it is the only possible design to study this phenomenon in the country. Second, past studies using randomized designs came to the same conclusion as those that did not. Therefore, this indicates that the effects found in studies in which candidates’ names were not randomized were attributable to the primacy effect (Chen et al., Reference Chen, Simonovits, Krosnick and Pasek2014; Pasek et al., Reference Pasek, Schneider, Krosnick, Tahk, Ophir and Milligan2014). Third, our main interest is not to identify a causal mechanism, but to look at ballot order effect as the manifestation of cognitive biases, either serial ordering of the candidates or alphabetical preferences that are irrelevant to vote choice, but that may nonetheless affect voters when informational cues are not sufficient.
Normatively, ballot order effect is an interesting and important phenomenon as it arguably hinders the democratic process. Apart from potentially leading to improvements in electoral regulations, the importance of research on this topic comes from the fact that the simple existence of a ballot order effect—regardless of its true cause—is a challenge to democratic politics. From a normative standpoint, alphabetical preferences and ballot order are equally irrelevant to vote choice. Although the randomization of candidates on ballot papers could at first appear to be a solution to ballot order effect, one could as well argue that it would mostly hide rather than solve the problem. If ballot order effects are more likely in settings with weak party cues, worrying about the potential confounding impact of alphabetical preferences seems a little beside the point. Although we acknowledge that our design does not empirically allow to control for alphabetical preferences, we nonetheless think that this specific problem is relatively small when we take into account what ballot order effect means for democracy. Our argument is that this bias is one of the manifestations of noise obfuscating the signal in elections. Whether the effect is truly related to serial ordering of the candidates or to alphabetic preferences does not solve the issue since both explanations are normatively irrelevant to vote choice.
Data and Method
All municipal elections across the province are held on the same day in the Fall. We analyze results from the 2009 and 2013 Quebec's municipal elections. In both of these elections, a rather large percentage of candidates were elected by acclamation, 35.6 per cent in 2009 and 33.3 per cent in 2013. These cases were dropped from the analysis. Among the remaining elections, 6400 candidates ran for the 2250 offices of district councillors, and 1307 candidates ran for the 544 mayoral positions in 2009. In 2013, 6874 ran for the 3438 district councillor contested offices, while 1419 individuals ran for the 580 contested mayors’ positions. Among these candidates, 23.9 per cent had a party affiliation in 2009 and 22.8 per cent in 2013. Again, when candidates have a party affiliation, it is never vertically related to provincial or federal political parties. Hence, this provides very little information to most voters. The mean number of candidates per race was 2.4 in 2009 and 2.3 in 2013.
To compare how the same electorate behaved at the provincial level during the time covered by the municipal elections, we use election results from the 2008, 2012 and 2014 Québec provincial elections during which candidates were competing within 125 ridings, with one seat each in the Québec provincial legislative assembly. As in most modern democracies, provincial elections in Québec are usually held at four- to five-year intervals, but the 2012 elections resulted in a minority government, which ultimately led to another provincial election less than two years later in 2014. Table 1 presents a descriptive statistics of the municipal and provincial data per elections.
Table 1 Number of Candidates per Elections
Source : Ministère des Affaires municipales et de l'Occupation du territoire and Directeur général des élections du Québec
The 2008 provincial election was held on December 8, 11 months before the November 1, 2009, municipal election. The 2012 provincial election was held on September 4, a little more than a year before the 2013 municipal elections held on November 3. Finally, the 2014 provincial elections, held on April 7, took place six months after the last municipal elections in 2013. This context allows us to test ballot order effect on approximately the same electorate in two very different settings. Provincial elections are structured by classic party competition and are well covered in both French and English media in the province. On the other hand, municipal elections are best characterized as competitions between low visibility individuals who may or may not happen to be affiliated with a municipal political party which is largely unknown to most voters. Those campaigns are usually covered in local media and focus on the mayoral races, leaving very little coverage for the city councillor candidates.
In both cases, the dependent variable is the vote share of each candidate in one election. Ballot order, our independent variable of interest, is determined by using the alphabetical order of the candidates’ names. In the municipal models, candidates’ rank is included as a four categories factor variable. The first three ranks are divided into an equal number of categories, and all the remaining positions are combined into a fourth one. In the provincial model, we included a seven-categories rank variable (rank seven onward are combined in a single category).Footnote 3 The first rank serves as the reference category in all our models. The number of candidates included in each category of the rank variable is presented in Table 2 and 3. All models also include a control for the number of candidates appearing on each ballot and incumbency. In addition, provincial and municipal analyses will include specific control variables, as well as a different number of models. The provincial analysis is pooled into a single model. In addition to the already mentioned variables, it includes an interaction term between the rank variable and the election year (2008, 2012, 2014) to account for potential heterogeneity between specific elections. Finally, dummy variables are included to take into account the candidates’ party affiliation (PLQ, PQ, CAQ, QS, Others).
Table 2 Number of Candidates in Each Rank Category - Municipal
Source : Ministère des Affaires municipales et de l'Occupation du territoire
Table 3 Number of Candidates in Each Rank Category–Provincial
Source : Directeur général des élections du Québec
The municipal analysis comprises three models. As previously mentioned, the party systems in most cities are very weak, or absent. The two exceptions are Montréal and Québec City, which both have somewhat more coherent party systems that are mostly the product of the visibility their mayors. For this reason, it is necessary to control for the candidates’ specific party affiliation in these two cities, which makes it difficult to pool the entire municipal data. Therefore, Québec City and Montréal each have their own model, and the “rest of Québec” is pooled together.
The first model for the rest of Québec, includes the basic variables already discussed (vote share, rank, number of candidates and incumbent). Even if the party cues are weak, the simple affiliation of a candidate might affect her vote share by enhancing the candidate's credibility. Having a party affiliation might also moderate the effect of their rank. Therefore a dummy variable for affiliated candidates is included, as well as an interaction with the rank variable.Footnote 4 In the municipal models for Québec City and Montréal, dummy variables for party affiliation are also included. Furthermore, all three municipal models include an interaction for the election year (2009, 2013).
Vote share, our dependent variable, is a continuous variable and one could therefore conclude that OLS regression is the appropriate statistical tool to analyze the data. However, such ratio variables are usually heteroskedastic and asymmetric, leading to biased results when used in combination with an OLS.Footnote 5 We will therefore use a beta regression developed specifically for ratio data bounded between 0 and 1 (Cribari-Neto and Zeileis, Reference Cribari-Neto and Zeileis2010). The results of these beta regressions are presented in the next section.
Results
Figure 1 presents the bivariate linear relationships along with the 95 per cent confidence intervals between ballot order and vote share in all the elections we are analyzing. At first glance, a simple visual inspection of the bivariate relationships clearly indicates that ballot order effect seems to be occurring within both types of municipal elections, whether mayoral or council races. Such an effect does not appear in our three provincial elections, which all exhibit flat relationships between ballot order and vote share. These results are consistent with our expectations regarding the effect of party cues at the provincial level.
Figure 1 Ballot Order and Vote Percentage in Québec Municipal and Provincial Elections
Relevant controls obviously need to be accounted for before coming to tangible conclusions. Hence, Table 4 displays the results of the three municipal beta regression models.Footnote 6 The results confirm that in 2009, candidates without party affiliation in the rest of Québec were negatively affected by a lower rank on the ballot paper. Although there is no systematic difference in vote share between the candidates appearing in the first and the second position on the ballot, this is not the case for candidates appearing in the third and fourth positions. On average these candidates received fewer votes than candidates appearing first on the ballot. Since interactions between the election year and the rank are not significant, this effect also holds for candidates in the 2013 election.Footnote 7 Furthermore, pairwise hypothesis tests with the rank categories show there a significant difference between ranks two and three but not between two and four. Therefore, the third rank seems to be the position with the biggest disadvantage.
Table 4 Effect on Vote Percentage in Québec 2009 and 2013 Municipal Elections
a p < .05
b p < .01
Interestingly, the interaction term between the 4th and subsequent ranks and party affiliation displays a positive and significant effect. This means that among candidates who have a party affiliation, not only does the effect observed for fourth and subsequently ranked candidates cancel out, it also reverses. On average, candidates who have a party affiliation and were placed lower on the ballot received more votes than candidates who appeared in first, second or third place. Yet among candidates who have a party affiliation, those who are third on the ballot still received fewer votes than candidates who appear first. The pairwise hypothesis test shows there is still a difference between ranks two and three.
The model for Montréal also exhibits a weak significant relationship between candidate ballot position and vote share. This effect exists for both elections since the interaction with the election year variable is not significant. More specifically, candidates who appeared in the third position on the ballot received, on average, significantly fewer votes than candidates who appeared first. Pairwise hypothesis tests also show that they received significantly fewer votes than candidates who appeared at the fourth or subsequent positions. This effect, however, is weaker than the one observed in elections outside Montréal, and the effect disappears without the interactions.Footnote 8 However, no difference exists to candidates in the second position. In addition, the interaction term with the election year variable shows that this effect is observed across both elections. Finally, the results from the Québec City model do not display a significant effect of ballot order on vote share. These results might be explained by the limited sample size. It might also be explained by the fact that party affiliation seems to be much stronger in Québec City compared to the other model, especially the effect of being affiliated with “Equipe Labeaume.” More analysis including more elections will be needed to confirm if there is no ballot order in the province's capital.
Beta regressions use a logit link function, and it is therefore impossible to directly interpret the resulting coefficients as they appear in Table 4. Hence, in order to better grasp the size of the relationships, model based predictions of the dependent variable are presented in Table 5. Since party affiliation moderates the effect of rank in the model for the rest of Québec, two series of predictions are presented. Among candidates who have no party affiliation, the predicted vote share of a third-ranked candidate is five points lower than first-ranked candidates, and four points lower than candidates appearing in the second position. The effect of appearing in the fourth position is weaker, about 3.5 points lower than the first rank. As previously mentioned the differences in vote share between the other ranks are not significant.
Table 5 Predicted Vote Share in the 2009 and 2013 Municipal Elections
The predictions among the candidates with a party affiliation are quite different. First, the effect size of the third rank is weaker. The difference in vote share between the first and the second rank is 4 and 2.5 percentage points respectively. However, the effect for the fourth rank is reversed. Among candidates with a party affiliation, candidates appearing on the fourth and subsequent position receive, on average, about four percentage points more votes than candidates appearing in first or second rank, and seven points more than those appearing third. In the Montréal municipal elections, candidates who appeared at the third rank received, on average, four percentage points less overall vote share than candidates who appeared first.
Table 6 displays the results of the models for the last three provincial elections in the province.Footnote 9 Results presented in Table 6 show that in provincial elections, ballot order has almost no effect as the rank variables have no significant impact in each of the three elections under study. This is true even if, as we have seen on Table 3, there are typically more candidates to choose from in provincial elections.
Table 6 Effect on Vote Percentage in Québec 2008, 2012 and 2014 Provincial Elections
ap < .05
b p < .01
Discussion
This paper has looked at the frequency of ballot order effects in two very different kinds of elections in Québec. We argued that ballot order is the manifestation of cognitive biases that are more likely to occur in settings that provide little information to voters. We analyzed election results from municipal and provincial elections in the province of Québec because they allowed us to compare electoral behaviour within two different party cue environments. Québec provincial elections are characterized by strong party cues whereas—apart from Québec City and Montréal—municipal elections in the province provide much weaker party cues to voters because of the nature of the parties themselves. Moreover, the five elections analyzed were held within a narrow time frame, which allowed us to compare the behaviours of roughly similar electorates.
Provincial elections, which are characterized by very prominent party cues and high visibility, do not exhibit signs of serial order bias. In these elections, candidate vote share is not affected by the order in which names appear on the ballot paper. On the other hand, municipal elections are characterized by almost no party cues and much less visibility. Hence, these elections ask for a little more effort from the voters compared to provincial elections. In this more demanding setting, our results show signs of ballot order effect, indicating that this small increase in complexity has an important impact on voting behaviour. The strongest effect was observed among candidates with no party affiliation outside Montréal and Québec City. Yet the effect was present as well for candidates with a party affiliation outside of the two largest cities and in Montréal. No such effect was observed in Québec City. Hence, these results confirm H1 and H2, which stated that ballot order would have an effect on vote share (H1), but that this effect would be moderated by election level (H2).
Comparing provincial elections to municipal ones allows us to tap into the mechanism at play. Since these different elections vary mostly by the extent to which they provide party cues to voters, the fact that ballot order seems to be present only in elections without party cues indicates that they are indeed a crucial piece of information to voters. Yet not all elections provide clear party cues and, hence, our results cast doubt on the ability of electorates to come to meaningful political decisions in the absence of clear and easy-to-use heuristics. The assumption behind a healthy democratic process is that voters know what they are doing. It assumes that they will be able to punish or reward incumbents according to their performance. It also assumes that people are voting for the candidates they prefer, which entails that they actually have meaningful preferences to begin with. Individual errors are inevitable, but there are important democratic implications if they do not cancel out during the aggregation process. In the 2009 municipal election, 30 third-rank candidates lost their race by five percentage points or less. In the 2013 election, this number increased to 69. The outcome of all these races could have been altered because of the ballot order bias. Therefore, our results show that in the absence of sufficiently strong party cues, the above democratic expectations are likely not met.
To some extent, the presence of ballot order effect in elections can also be interpreted as clear evidence that individual errors do not simply cancel each other out in the aggregation process. In the absence of a cue to make a political decision, most voters appear to remain ambivalent about their voting decision, and, given that they nonetheless cast their ballot, this ambivalence seems to leave room for serial position effect, a well-known cognitive bias. Our results indicate that in the absence of an informational shortcut, people who are unsure about how they want to cast their vote should not be expected to vote randomly among all the possible candidates, but to vote randomly while still being subjected to this well-known cognitive bias. Hence, the uninformed votes of those citizens should not cancel each other out in the aggregation process but rather will advantage the candidates lucky enough to be among the first few appearing on the ballot. This is a real instance of noise taking over the signal in collective preferences.
Although our results suggest that party cues appear to be sufficient to prevent such a bias having an effect, they also indicate that their absence has potentially dramatic consequences on the quality of the democratic process. Finally, municipal elections are arguably understudied in the discipline and, considering their concrete importance in citizens’ lives, our results also highlight the need to study further these types of elections. Such an endeavour, we believe, may prove to be important should we wish to improve the democratic process.
Appendix
Table A Effect on Vote Percentage in Québec 2009 and 2013 Municipal Elections–No Interactions
