Elsevier Editorial System(tm) for Electoral Studies Manuscript Draft Manuscript Number: Title: Campaign Civility under Preferential and Plurality Voting Article Type: Original Research Paper Keywords: Preferential voting, campaigns, electoral systems, Alternative Vote, ranked choice voting, negative campaigns Corresponding Author: Prof. Todd Donovan, Corresponding Author's Institution: Western Washington University First Author: Todd Donovan Order of Authors: Todd Donovan; Caroline Tolbert; Kellen Gracey Abstract: Preferential voting systems have been said to facilitate civility in elections, yet there has been little systematic study of this proposition. This paper reports results from surveys of candidates' and voters' perceptions of campaigns, comparing places with plurality elections to those that used preferential voting rules. Candidates from these two sets of jurisdictions reported engaging in different campaign activities, and candidates from plurality places reported that campaigns were less negative. Surveys of voters indicate that people in cities with preferential voting were less likely than people in cities with plurality elections to view campaigns as negative, and people experiencing campaigns under preferential rules were more satisfied with the conduct of campaigns than people in places using plurality elections. Results are consistent across a series of robustness checks.
Manuscript without author identifiers Click here to view linked References Abstract Preferential voting systems have been said to facilitate civility in elections, yet there has been little systematic study of this proposition. This paper reports results from surveys of candidates' and voters' perceptions of campaigns, comparing places with plurality elections to those that used preferential voting rules. Candidates from these two sets of jurisdictions reported engaging in different campaign activities, and candidates from plurality places reported that campaigns were less negative. Surveys of voters indicate that people in cities with preferential voting were less likely than people in cities with plurality elections to view campaigns as negative, and people experiencing campaigns under preferential rules were more satisfied with the conduct of campaigns than people in places using plurality elections. Results are consistent across a series of robustness checks. 0
Campaign Civility under Preferential and Plurality Voting This paper examines if electoral systems affect the tone, or civility, of campaigns and elections. We test if the type of electoral system affects levels of negativity in election campaigns by isolating, at least partially, the effects of plurality versus preferential voting. 1 Elections in nearly all US cities are conducted under some variant of plurality, winner-take-all rules, where each voter has the capacity to express a preference for a single candidate. However, a handful of US cities have adopted preferential voting, where voters may rank their preferences for multiple candidates. Existing literature suggests that the latter system gives rival candidates more incentives to cooperate, or weakens incentives for conflict. The American case, then, provides a unique opportunity for systematic, empirical tests of this intuition. As we discus below, there are reasons to expect that the manner in which votes are solicited and how votes are cast may have a measurable influence on how candidates campaign and on how voters experience and perceive campaigns. This is not to say that other features of these local elections have no effects on how candidates campaign. A candidate seeking office in an election conducted exclusively in a single member district (SMD) has little, if any, incentive to campaign city-wide. Partisan elections affect local campaigns by altering resources available to candidates and altering the cues that voters use when voting (Schaffner et al., 2001). The racial / ethnic composition of districts may also affect campaign mobilization efforts (Barreto et al., 2004; Bowler et al., 2001). Yet we have little reason to expect that these factors have substantial effects on 1 We leave the question of whether or not negativity in campaigns is a good or bad thing to others (e.g. Mayer, 1996; Geer, 2008; Mattes and Redlawsk, 2014). 1
the incentives candidates have to engage in negative campaigns. Furthermore, given America's political geography, any effects of partisan versus non partisan local elections on campaign tone may also be muted. Many cities, particularly larger places, are dominated by a single party, regardless of whether or not the city uses partisan or nonpartisan ballots. 2 However, we suggest there are reasons to expect variation in the local electoral system may affect how candidates and voters experience campaigns, and that local elections in the US provide a unique opportunity for modeling the effects of electoral systems. In the past decade, a number of US cities adopted the Alternative Vote, a form preferential voting that is commonly referred to as Ranked Choice Voting (RCV) in the United States. This study compares candidate and voter perceptions of campaign discourse in these cities to perceptions measured in similar cities that conducted elections with plurality voting. Campaigns Under Preferential versus Plurality Voting We have reason to expect that the electoral system affects incentives candidates have to engage in negative campaigns. Preferential voting systems can be expected to alter the nature of electoral campaigns, in part, by changing incentives candidates have for attacking rival candidates. Under winner-take-all plurality voting, a voter has a single preference to cast and candidates compete for each voter's single vote. Campaigns are conducted in a zero-sum context: a candidate runs in a situation where any vote cast for another candidate may be seen as a loss for the candidate. In such a context, (and possibly 2 All cases where voters were surveyed for this study used nonpartisan elections conducted in November of an "off" (odd numbered) year. Nearly all the cases had a similar dominant (Democratic) partisan composition. 2
contingent on the number of candidates, see Skaperdas and Grofman 1995) the candidate may have strong incentives to criticize and attack opponents, and maximize (or exaggerate) differences between the candidate and rivals in order to attract a voter's single preference. Preferential systems such as the Single Transferable Vote (Bowler and Grofman 2000) and Ranked Choice Voting (RCV), in contrast, may reduce a candidate's incentive to engage in negative campaigns because they allow (if not require) voters to express multiple preferences. Horowitz (1985; 1991) suggests that by allowing voters to cast multiple preferences, preferential voting encourages bargaining, reciprocity and accommodation among rival elites. Reilly (1997; 2001; 2004:263) notes that preferential voting changes elections from a zero sum situation to something more of a positive sum context, and encourages candidates to explore potential areas of commonality on policy, and on political strategy. Guinier (1994) proposes that Cumulative Voting could produce a "mutually beneficial system of cooperation" in the context of racially polarized voting because, by allowing voters to support multiple candidates, it would encourage candidates to make appeals to voters that cut across racial lines. 3 The mechanics of voting, then, are expected to condition how campaigns are conducted. Under plurality elections, politicians depend on single votes cast from their supporters, and on single votes cast from others who may be persuaded that rival candidates are ideologically distant from the voter, or otherwise threatening. Under preferential systems candidates rely not just on support from their core supporters; they 3 Cumulative voting has a mechanism similar to preferential systems. But rather than ranking multiple candidates, voters can support multiple candidates by distributing multiple votes. 3
may also benefit from lower-ordered preferences cast by supporters of their rivals. Given this, we might expect candidates to campaign differently under preferential than plurality rules. With each voter casting second, third, or additional preferences, candidates may have less incentive to attack their rival, for fear of alienating their rival's supporters and losing those voters' lower preferences. Preferential systems have received substantial attention as a mechanism for mitigating ethnic group conflict and reducing tension in divided societies such as Papua New Guinea, Fiji, Northern Ireland, and Estonia (Reilly 2001; 2002), and Australia's experience with the Alternative Vote has been well studied (e.g. Bean, 1997; Farrell and McAllister, 2003; Bowler et al., 1996). Case study evidence suggests that preferential voting may have a moderating effect on electoral politics in Australia (Reilly, 2001; 2002), and research documents that Australian political elites bargain and coordinate preference trades in a sophisticated manner (Sharman et al., 2002). Yet we know little about how (or if) preferential voting may affect the tone and style of campaigning in advanced, established democracies. There are few, if any, systematic, cross jurisdictional studies that compare campaign tone under preferential voting to campaigns conducted under other electoral rules. There is, however, some evidence that candidates campaign differently under preferential rules, and there is some evidence that preferential voting is associated with a form of politics that citizens value. A study comparing US elections conducted under plurality and cumulative voting found candidates in the latter system were more likely to work to mobilize voters (Bowler et al., 2003). A cross-national study of public opinion (Farrell and McAllister, 2006) concluded that, other things equal, voters were more 4
satisfied with how democracy worked in nations where people voted for candidates with preferential voting. 4 From this literature, we develop our core research questions: Does preferential voting (RCV) correspond with greater cooperation and civility among candidates? Do voters and candidates perceive less negative campaigning where elections were conducted under preferential voting rules? Candidate and Citizen Perceptions of Campaigns: A Comparative Method This research builds on a method employed in Bowler et al. (2003), in order to compare campaigns in U.S. cities using preferential voting to campaigns in similar places using plurality elections. Our study takes advantage of natural variation in election rules at the local level by conducting surveys with similar content that measured opinions at the elite (candidate) and mass (voter) levels. We identified multiple control cases (plurality cities) that were demographic matches of cities using RCV to provide additional leverage for assessing potential effects of electoral systems. The research design brings us some way toward isolating the potential effects of RCV elections on campaigns and perceptions of campaigns. Voter and candidate perceptions of campaigns were measured in RCV jurisdictions having competitive local elections, and in plurality jurisdictions with competitive local elections that were identified as (approximate) demographic matches 4 Carey and Shugart (1995:425) could be read as suggesting that incentives for cultivating a personal vote may be lower in RCV elections than in plurality systems. 5
for the RCV communities. 5 This design allows us to compare mass perceptions of campaigns across different electoral systems, and it also provides a rare opportunity to assess how well elite and mass perceptions of campaigns overlap, and to test if the public perceived campaigns as elites did. As much as possible, RCV cities were matched to cities using plurality elections based on population, median income, race/ethnic composition, electoral context, and region. Methods of the voter study are presented in more detail in the next section of this paper. We surveyed candidates for local offices in three cities that conducted RCV elections in 2013 6 (Cambridge, Minneapolis, and St. Paul) and in four cities that conducted RCV elections in 2012 (Berkeley, Oakland, San Leandro, San Francisco). In addition, we also surveyed candidates who ran in similar plurality jurisdictions these same years. The RCV cities in the candidate study, and their matching cases, are listed in Table 1 (top half). 7 The major assumption behind this matching method is that the composite of these plurality jurisdictions form a comparative context that is similar enough to the RCV places, so that difference in attitudes and behavior between candidates from RCV and non-rcv places can be attributed to the electoral system. Table 1 about here We assume that there were no major idiosyncrasies associated with any of these places that would make cross-jurisdictional comparisons problematic. The likelihood of 5 This process was constrained somewhat by the limited number of jurisdictions with competitive mayoral elections in 2013 or other competitive elections. See Appendix A3 for details. 6 Three offices were elected in San Francisco with RCV in 2013, but none were contested by more than one candidate. 7 Given there are far fewer candidates than voters, we surveyed candidates in more jurisdictions than voters in order to increase or pool of candidate respondents. 6
violating this assumption is reduced by the fact that we have a range of places that were using RCV, and by the fact that each RCV city was matched with multiple plurality cities. We suggest this multiple case comparative method has advantages over the standard cross-sectional, cross-national statistical analyses of institutional effects (e.g. Karp and Banducci, 2008; Blais and Dobrzynska, 2003; Anderson and Guillory, 1997) since it allows us to minimize non-electoral system differences across the cases that are being compared. That is, as much as Seattle (using plurality elections) and Minneapolis (using RCV) have features which differ that are unrelated to their electoral systems, such differences are likely much narrower than non-electoral system differences between Canada and Italy, or Great Britain and Norway. Surveys were mailed in late 2013 to all candidates who ran in the most recent election in these places (approximately 670 candidates). Follow up mailings, and emails, produced a response rate of 36%, with similar response rates from RCV and non-rcv jurisdictions. 8 Responses were slightly higher from places having elections in 2013. Given there were more plurality cities surveyed by design, this produced 154 responses from candidates who ran under plurality rules, and 80 responses from candidates who ran under preferential rules. Thirty-five percent of respondents had won their recent election. Overall, 70% were male, 75% identified as white, the mean age of respondents was 52 years, and just over half had post graduate education. 8 This rate is near the mean for response rates for several surveys of parliamentary candidates. The Comparative Candidate Survey reports response rates such as 47% for Australian candidates, 23% for Austrian candidates, 55% in Canada, 32% in Estonia, 44% in Germany, 34% in Greece, 32% in Hungary, 35% in The Netherlands, 17% in Portugal, and 43% in Sweden. 7
Candidate Assessments of their Campaign Activities The candidate survey included a battery of questions asking about campaign activities, and a battery of questions asking candidates about their perception of how their most recent campaign was conducted (the candidate survey included some questions very similar to those asked on the voter surveys). We begin here with a comparison of candidate campaign activities across the two election systems, and then compare candidate perceptions of campaigns. Given the literature on preferential voting, and given that RCV presents candidates with opportunities to appeal for voters' first and lower-ordered preferences, we expected that candidates (or their allies) in RCV cities may have been more likely than candidates (or their allies) in plurality places to ask voters to support them and to also support another candidate. We asked "when campaigning, did you ask voters to 1) vote for you only, or 2) vote for you and support other candidates seeking the same office?" We also included questions asking (separately) if there were groups working on the candidate's behalf, or working on behalf of the candidate's opponent, that instructed people to "support me and other candidates in my race." With these items we found limited evidence that there was more mutual cooperation among rivals for the same office under RCV than plurality elections. Twenty nine percent of candidates for a plurality office said they asked voters to "support me and support other candidates in my race," while 36% of candidates for an RCV office reported this. Similarly, 37% of candidates in RCV cities said that there were groups working on behalf of their opponent who were instructing people to support multiple candidates. Twenty seven percent of candidates in plurality jurisdictions reported this. 8
Candidates were also asked about the types of campaign activities they engaged in, and about what they spent campaign funds on. In terms of total campaign spending reported by these candidates, there was no significant difference between candidates running in RCV versus plurality elections (this remains so even when we control for higher levels of spending reported by candidates from the larger cities). As for spending on specific campaign activities, candidates from RCV places were much less likely to report spending on mail (56%) than candidates who ran under plurality rules (74%, Chi Sq. p. <.02), but RCV candidates were more likely to hire paid staff (44 % vs. 29%, Chi Square p. <.03). RCV candidates were slightly more likely to spend money on Internet ads (30% vs. 25%) and RCV candidates were significantly less likely to report spending campaign funds on radio and TV ads. 9 Table 2 about here Table 2 presents the mean of candidate responses to questions asking which campaign activities were seen as most important to the candidates. We had expected that candidates in RCV places, having an eye toward attracting preferences from supporters of rival candidates, would have been more likely to report that "building alliances with rival candidates" was an important part of their most recent campaign. Yet, as indicated in Table 3, there is no difference between RCV and plurality candidates in the mean levels of importance assigned to building alliances with rivals. Nonetheless, we do observe significant differences between the campaign activities that were seen as important to RCV candidates versus plurality candidates. Among all respondents, RCV candidates 9 18% of plurality candidates, and 5% of RCV candidates spent on radio (Chi Sq. p. <.00). 10% of plurality candidates and 2.5% of RCV candidate spent on TV ads (Chi Sq. p =.04). 9
were significantly less likely than plurality candidates to say that money for advertising was very important, and they were more likely to say a campaign web site was important. Among winning candidates, RCV candidates were significantly more likely to say it was important to have volunteers on their campaign, were more likely to say it was important to maintain a web site, and were less likely to say that having money for advertising was important compared to candidates in plurality cities. These findings suggest that, at the local level in the US, campaigns conducted under preferential elections had a different style or tactics than campaigns conducted under winner-take-all plurality rules. Under preferential voting there may have been less importance of conventional print and broadcast advertising, and a greater importance of paid staff, volunteers, door-to-door contacts, 10 and web advertising. At this point, we cannot speculate as to how, or if, these potential differences in campaigning translated into campaign content that was less negative or less adversarial. We can start at this by examining what the candidates thought about how the campaigns were actually conducted. Candidate Perceptions of Campaign Tone Candidates' perceptions of campaigns were measured with several different survey items. Our primary goal with these was to assess how candidates perceived the quality of the campaign they participated in, in terms of the tone of the campaign (negative or not), and in terms of information that the campaigns disseminated. Candidates were asked how satisfied they were with the way other candidates conducted 10 Experience with local campaigns suggests paid staff are often field coordinators who direct volunteers in canvasing efforts. 10
their campaigns, and if they thought that other candidates provided people with useful information. On these two items, we found no notable differences between candidates from places using plurality or RCV elections. Figure 1 about here Responses to a number of additional questions do suggest that there were differences in how candidates perceived the tone of these campaigns. Candidates were asked how they perceived how "your main rival's campaign described you" They were also asked, "how did your own campaign describe your main rival in the race. 11 The upper panel of Figure 1 illustrates the percent who indicated that they were portrayed somewhat or mostly negatively by their rival during the campaign. Nearly 40% of candidates who ran for office under plurality rules reported their rival described them in negative terms, whereas 28% of candidates running under preferential rules reported this. The second panel of Figure 1 displays how candidates described the manner in which their own campaign portrayed the candidates' main rival. Thirty six percent of RCV candidates reported that their own campaign described a rival positively; this compares to 20% of candidates who campaigned under winner-take-all plurality elections. Ordered logit models estimating responses to both questions illustrated in Figure 1 demonstrate the effect of RCV is statistically significant. 12 Figure 2 about here 11 On each of these items, response options were: mostly positive, somewhat positive, not really positive or negative, somewhat negative, or mostly negative. Candidates with no opponent are omitted from the analysis here. 12 Anderson et al. (2005) demonstrate that attitudes about politics are conditioned by whether or not someone was on the winning or losing side of an election. Ordered logit models of responses to these questions in Appendix A1 account for this, for candidate age, gender, and if the jurisdiction used district vs. at large elections. 11
Figure 2 provides further illustration that campaigns conducted under RCV may have been less negative than plurality contests. The upper panel in Figure 2 reports the results of a question that asked candidates if they recalled candidates praising or endorsing any of their opponents. Fifty-three percent of RCV candidates replied that this occurred frequently or occasionally, compared to 41% of plurality candidates. Again, the coefficient for RCV is significant in ordered logit estimates of this item. 13 The lower panel of Figure 2 displays responses to a question that asked if the candidate's election was more or less negative compared to other recent political contests. 14 Response options ranged from "a lot more negative" to "a lot less negative." Forty-nine percent of candidates seeking office under RCV said their contests were less negative than before, while 37% of candidates in plurality contests gave this response. Conversely, just 8% of RCV candidates thought the most recent campaign was "a lot more negative" than before (compared to 15% of plurality candidates). The independent effect of RCV on responses to this question does not remain significant in multi-variate ordered logit models. Thus, although RCV and plurality candidates did not differ in their retrospective evaluations of campaign tone, there were significant, notable differences in how they perceived the tone of the campaigns they had just participated in. Citizens' Perceptions of Campaign Activity 13 See Appendix A1. 14 Candidates and voters were asked about negativity in the context of "other political contests" rather than being asked to compare things to previous contests in their specific city, or to contests for state or federal office. Our question wording is identical to an item used by Magleby et al 2004. This wording gives respondents in all places a common reference for assessing negativity, and it allows us to compare their responses to those collected from national surveys conducted during federal contests. 12
Our candidate survey suggests that compared to campaigns conducted under plurality elections, campaigns conducted under preferential voting relied less on mail, TV and radio, while relying more on paid staff, volunteers, and Internet. In addition, candidates from RCV places perceived less negativity and more cooperation (in terms of candidates praising and endorsing rivals) in their contests than candidates from plurality places. The magnitude of the effects are modest, but we do find a pattern here: RCV candidates were more likely than plurality candidates to report that rivals were praising each other, and RCV candidates were less likely to report that their campaign or their opponent's campaign was negative. An important question that follows from this is, did the voters notice? Given the potentially limited (Bowler and Donovan, 2013; but see Anderson and Guillory, 1997), perverse (Berinsky, 2005) and unanticipated effects (Burden et al., 2014) that electoral reforms can have on political attitudes and behavior, and given that the differences we observed in campaign tone across election systems were rather subtle, we might expect that mass perceptions about negativity in campaigns would be unrelated to the local electoral system. However, there is evidence (from federal elections) that exposure to negative campaigns may increase the likelihood that people perceive a campaign as negative (Ridout and Fowler, 2012). We employed a method for matching RCV and plurality cities very similar to that used in the candidate survey to isolate the effects of local election system on public attitudes about local campaigns. Plurality cities were matched to the RCV cities based on criteria used in the candidate survey (but limited to places with competitive elections in 2013). We designed a telephone survey to measure voters' perceptions, and a random 13
sample of 2432 respondents were contacted by the Eagleton Poll (Rutgers University) immediately after the November 2013 elections. 15 The lower half of Table 1 lists the distribution of voter interviews across these RCV and plurality cities. 16 Our voter sample included approximately 1200 likely voters from three jurisdictions where elections were just conducted under RCV (Cambridge, Minneapolis, and St. Paul) and about 1200 likely voters from seven similar jurisdictions that had just experienced plurality elections (Boston, Seattle, Des Moines, Cedar Rapids, Tulsa, Lowell, MA and Worcester, MA). This design enables us to estimate models of public perceptions that hold citywide and individual level factors constant, so we may test if respondents from RCV cities perceived campaigns differently (or as less negative) than did people in places using plurality elections. Bivariate results from the likely voter survey are reported below, and the Electronic Supplementary Material Appendix contains several multivariate models estimating responses to these survey items as well. Control variables in the multivariate models include standard demographic factors associated with participation in elections, as well as political interest, satisfaction with local government, partisanship, and whether a person supported a candidate who was elected mayor and / or elected to the city council. Our survey of likely voters included several items asking people what they recalled about local campaign activities, and asking questions about perceptions of campaign tone, using items very similar to those employed in the candidates survey. 15 The random digit dial sample included cell and landline phone numbers and was conducted in English and Spanish. 16 RCV cities were matched with plurality cities based on electoral system traits (off-year, open seat, mayoral or city council race) as well as city size, demographics, socioeconomic conditions, region and political attributes. See Appendix A3. 14
People were asked what modes of campaign contacts they experienced, and what types of advertising they could recall. As with the candidates, citizens were also asked if they thought campaigns were more or less negative than other recent contests; how often they found that candidates were praising each other; how frequently they perceived candidates to be criticizing each other; and how satisfied they were with how candidates conducted their campaigns. Table 3 about here Table 3 illustrates the distribution of voter responses to several of the questions about campaign activities. These results suggest there were clear differences in perceptions of how campaigns were conducted across plurality and RCV cities, and that voters' perceptions of campaign activities mesh, at least in part, with what we found with the candidate survey. Most notably, voters in RCV cities were more likely to report being contacted by a campaign. Although there were no differences across jurisdictions in total spending reported by candidates, candidates from plurality jurisdictions were more likely to report that money for advertising was important to their campaign (see Table 2). For their part, voters in plurality cities were much more likely to recall television or radio ads than voters in RCV cities. Conversely, voters in RCV cities were far more likely to report finding a pamphlet on their door, and were more likely to report they were contacted in person at their home. This suggests that campaigns in these RCV cities directed more resources organizing face-to-face contacts with voters, rather than 15
broadcast advertising. These effects of election system in Table 3 remain significant when multivariate logit models of responses to these questions are estimate. 17 Citizens' Perceptions of Campaign Civility At this point, we can not determine if the differences in campaign activities corresponded with more or less negativity in the content of campaign discourse or campaign advertising. In future research, we hope to assess the content of advertising and media coverage of subsequent campaigns in order to have additional measures of the civility and tone of the campaigns. However, if we assume broadcast advertising is more likely to contain negative and \ or combative content than face-to-face contacts with voters, 18 results in Table 3 might be read as showing that campaigns under plurality elections had greater capacity to disseminate negative advertising. Our survey of voters, however, does provide a direct means of measuring how people perceived the tone of the campaigns. Voters' responses to several questions about campaign civility are presented in Table 4. Although residents of RCV cities did not share the RCV candidates' perspective about candidates praising each other, responses to other questions demonstrate substantial differences in voters' perceptions of campaigns across plurality versus preferential elections, differences that, again, are consistent with what candidates reported. When asked "do you believe the campaigns this year in [CITY NAME] were more negative, 17 Respondents were also asked if they were if they were contacted via social network sites such as Facebook or Twitter, and if they recalled seeing web ads. There were no differences across elections systems on these items. 18 The assumption is plausible if we consider that campaigns often seek to distance themselves from negative ads produced on their behalf. Candidates and their supporters may thus be less likely to use negative information in face-to-face contacts with voters. 16
less negative or the same compared to other recent political contests?", 14.6% more of the respondents in RCV cities replied the campaigns were a lot less negative (Chi Sq. 131.8 p. <.000). Although most people in plurality cities reported that candidates criticized their opponents a great deal of the time or some of the time, only half as many people in RCV cities reported this (Chi Sq. 257.4 p. <.000). 19 Respondents in the RCV cities were also significantly more likely to report being satisfied with how candidates conducted their campaigns (Chi Sq. 41.7 p. <.00). Table 4 about here As demonstrated in a series of tables reported in Appendix A2, these effects illustrated in Table 4 remain significant in multivariate models that control for a range of demographic and attitudinal factors, including race, age, education, partisanship, and supporting a winning candidate, and results are robust when estimated with ordered logit or dichotomous logit models. The primary results are reported in Appendix A2 Table 1, showing four ordered logistic regression models estimating perceptions of positive electoral campaigns. Controlling for host of demographic, partisan and other factors, respondents living in RCV cities were significantly more likely to perceive the election campaigns as more positive. Appendix A2 Table 2 replicates this model using logistic regression where perceptions of campaigns are coded as a binary variable (1=little less negative, lot less negative; 0 = other). Again, the coefficient for RCV places remains significant in logit models. On additional survey items, we found no differences associated with a city's electoral system. Regardless of whether they experienced a plurality or RCV election, 19 Respondents were asked, "Thinking about the [CITY] election, how much time would you say the candidates spent criticizing their opponent?" 17
most voters were fairly or very satisfied with their choices of local candidates, and most thought the campaigns provided at least some useful information. Figures 3, 4 and 5 about here Nonetheless, on our measures of perceptions of campaign tone, the differences we observe are rather striking. Voters in places where RCV elections were conducted in 2013 (Minneapolis, St. Paul and Cambridge) were substantially more likely to view campaigns as being less negative than voters in similar places that used plurality elections. Figures 3, 4 and 5 illustrate the relationships between electoral system and voter perceptions of campaigns with post-estimation simulations of predicted probabilities generated from multivariate models that control for demographic and attitudinal characteristics of individual respondents (see Appendix A2). These figures illustrate the independent effects of electoral system when the effects of the control variables are accounted for. Other things held constant, people in RCV cities were nearly twice as likely to say local campaigns were a lot less negative than other recent contests (Figure 3), people in plurality cities were twice as likely to say candidates criticized each other some or most of the time (Figure 4), and people in RCV cities were significantly more likely to report being very satisfied with how local campaigns were conducted (Figure 5). Robustness Checks for Control Cases using Jack-knifed Samples and Matching Methods Given that the quality of the matched jurisdictions is important for establishing confidence in these results, we conducted four separate robustness tests to determine whether our control cities are a good match for our treatment cases, or RCV cities. First, cities were selected based on demographic traits listed in Appendix A3. All cities where 18
voters were surveyed used non-partisan elections, so we did not have local results that revealed city-level partisanship. We examined this with a redistricting tool that provided precinct-level data on Democratic vote share in cities where we surveyed voters. 20 Appendix A3 lists each city's demographic traits, and city-level partisanship constructed from precinct data. Other than Tulsa, our plurality matches for Minneapolis (Boston and Seattle) were nearly identically liberal, as measured by the percent vote for Obama in the 2008 presidential election and Democratic vote in the 2010 US Senate contests. St. Paul (RCV) and its matched plurality places (Des Moines and Cedar Rapids) were very similar in partisan composition, and Lowell and Worcester (plurality) were only slightly less liberal than Cambridge, MA (RCV). Second, it is possible some plurality cities were poor matches for the RCV cities, or were outliers in some other unmeasured dimension that might be affecting our results. To check for this we jack-knifed our voter sample by omitting, sequentially, the respondents from each of the control cities and re-estimating our multivariate regression models. The results of the jack-knife robustness tests are reported in Appendix A4, Table 1. Column 1 replicates our primary model predicting perceptions of campaign negativity from Appendix A2 (Table 1) but omitting respondents from Lowell, MA. Column 2 includes the full sample but omits respondents from Worchester, MA. Column 3 omits the respondents from Boston, etc. Across the seven models reported in Appendix A4, our primary finding holds: respondents residing in cities with RCV had more positive perceptions of electoral campaigns. The size of the coefficient for RCV, furthermore, is 20 This was done using an on-line application that supplies precinct-level data on partisan vote share from the 2008 presidential and 2010 US Senate elections. See Dave s Redistricting 2.2 http://gardow.com/davebradlee/redistricting/davesredistricting2.0.aspx 19
very similar across the model specifications. This is a strong indication that the choice of control cities is not driving the results we report; omitting the cases from any one particular control city does not change the primary results. As a third robustness test we use a statistical matching process, coarsened exact matching (CEM), in R statistical software to preprocess the data (Iacus et al., 2012). We matched on individual level demographic, economic and partisan variables using the balance function. This produced only small differences between the RCV and non-rcv city samples, and no systematic differences. Specifically, the multivariate imbalance measure, or L1 statistic (L1=.473; the percentage of local common support equaled 41%) revealed that our full voter samples from RCV and non-rcv cities were relatively balanced. RCV and non-rcv cities were compared, but they were also clustered/matched by the city pairs (Minneapolis vs. Boston, Seattle and Tulsa, for example). Only the variable age showed a small imbalance between the RCV and non- RCV cities, but the mean differences were small (.11 on the scale of age measured in years). Appendix A5 reports these results of the univariate imbalance measures for each explanatory variable. Finally, the CEM matching method is used as a robustness check for our survey matching design to ensure that the samples from the RCV cities have roughly the same demographic composition as samples from the control cities. The purpose of CEM is to approximate a treatment and control design for data that are not experimental. After preprocessing our data with this statistical matching, respondents who cannot be matched were omitted from the analysis. Models were estimated with only the cases that were matched to identify a SATT (sample average treatment effect) statistic. This pre- 20
processed survey data was then used to generate point estimates to see if there was still a statistically significant effect of residing in an RCV city on perception of positive campaigns. The CEM procedure revealed a statistically significant SATT statistic of.81. In short, even when using a reduced sample of survey respondents who can be statistically matched, we find similar results as reported in the paper. These robustness checks increase our confidence in the results reported here. Conclusion The academic literature suggests campaigns conducted under preferential electoral systems such as the Alternative Vote (known as Ranked Choice Voting in the US) may have more cooperation, or less conflict, among political rivals. We have tested that intuition here and found some support for it. Our results are based on comparisons of campaigns held under plurality and preferential voting in just one election cycle in the US (2013), and are based on the only three cities with competitive RCV elections that year. There is, then, a potential problem with edogeneity here. Our three RCV cities could have adopted their election system for reasons that are related to what we observe and, and thus related to what we suggest might be caused by the election system. Although we have no reason to expect this to be the case, it may be that cities with a more civil and less adversarial political context are somehow more likely to adopt RCV. Related to this, it might be that the 2013 elections in these RCV places were outliers for those jurisdictions, in that they were atypically civil for unknown reasons. We propose our comparative matching method brings us some way toward accounting for this potential endogeneity. 21
As we noted at the outset, we will leave judgments about the pros or cons of negativity in campaigns to others. As much as American citizens, pundits, and elites bemoan negativity in politics, social science paints a much more nuanced picture about the ills, or potential benefits of civil versus negative campaigns (e.g. Brooks and Geer, 2007; Brooks, 2008; Mattes and Redlawsk, 2014). Voters may also be able to distinguish between information that they see as negative or critical, and campaign information that is useful. Put differently, a negative campaign, or campaigns with candidates criticizing each other, may not mean that the campaign is uninformative (Sides et al., 2009). Nonetheless, from the perspective of students of electoral reform our results are somewhat striking. As we mentioned earlier, there are many reasons to expect that electoral reforms have minimal, if any, effects, on how people view politics. As examples, term limits did not restore trust in state legislatures, campaign finance reforms had no effect on public attitudes about politics, changes in electoral system can have only weak effects on mass attitudes and behavior (Bowler and Donovan, 2013), and 'early voting' may correspond with lower voter turnout (Burden et al., 2014). Yet in 2013 (in these places) it appears preferential voting corresponded with campaigns being perceived by candidates and voters as less negative, with voters saying that candidates criticized each other less, and with voters having greater satisfaction with the conduct of campaigns. Although additional research will be needed to assess if these cross-sectional observations are supported with cross-time observations from places that change from plurality to preferential systems, the intuition about how preferential voting affects campaigning, and our study of candidates and voters, lends strong support to the idea that the electoral system is the causal mechanism here. 22
Does this mean that the adoption of RCV could alter campaigns for higher level elections in the US or in other established democracies, and affect how voters experience those campaigns? We are unable to make inferences about that. We can, however, put some of our survey results in a broader perspective with the US case. In the 10 US cities where we surveyed voters in 2013, we found few people who thought campaigns for city offices were 'more negative' than usual, and found high levels of satisfaction with local campaigns. Overall, 77% of our respondents were satisfied "with the way most candidates have conducted their campaigns in the recent local election" (with even higher satisfaction in RCV places). Most (71%) were happy with the quality of information they received from local campaigns, and only 10% thought their local campaigns were more negative than "other recent political contests" (again, with less negativity perceived in RCV places). In contrast, there is likely to be far less satisfaction with federal campaigns in the US, and much more widespread perceptions of negativity. A 2002 survey found 33% of Americans thought campaigns for the US Senate / House of Representatives were more negative "compared to other recent political contests." In states with heated federal races, most people viewed federal campaigns as more negative (Magleby et al., 2004). A 2010 Gallup Poll using another question similar to one we employed found that just 48% of Americans were satisfied with the way "the congressional candidates in your area have conducted their campaigns this year." It may be then, that given general satisfaction with local elections in these cities, there was a bit of a ceiling effect for RCV on campaign tone and perceptions of campaigns, and greater room for effects where there is less satisfaction with campaigning. 23
Or, given the polarized nature of federal contests in the US, it may be that incivility and negativity are so deeply entrenched in federal elections that adoption of preferential voting may have no effect on such elections. Whatever the case, the diffusion of RCV in the US presents scholars with important new opportunities for studying the consequences of electoral systems. 24
Table 1: Jurisdictions included in the Candidate Study and Voter Survey Candidate Survey RCV cities Berkeley, CA (2012) Cambridge, MA (2013) Oakland, CA (2012) Minneapolis, MN (2013) San Leandro, CA (2012) San Francisco, CA (2012) Matched plurality cities Alameda, CA (2012); Pasadena, CA (2012); Santa Monica, CA (2012 Ann Arbor, MI (2013); Lowell, MA (2013); Stamford, CT (2013); Worcester, MA (2013) Richmond, CA (2012); Sacramento, CA (2012); Vallejo, CA (2012) Boston, MA (2013); Cincinnati, OH (2013); Seattle, WA (2013); Tulsa, OK (2013) Daly City, C (2012); Fairfield, CA (2012); Hayward, CA (2012); Union City, CA (2012) Portland, OR (2012); San Diego, CA (2012); San Jose, CA (2012) St. Paul, MN (2013) Cedar Rapids, IA (2013); Des Monies, IA (2013); Madison, WI (2013); Spokane, WA (2013) Voter Survey RCV cities Cambridge, MA (n=202) Minneapolis, MN* (n=812) St. Paul, MN (n=203) Matched plurality cities Lowell, MA (n=100); Worcester, MA (n=100). Boston, MA* (n=268); Seattle, WA* (n=270); Tulsa, OK* (n=269). Cedar Rapids, IA (n=108); Des Monies, IA (n=100). *City had a competitive mayoral election Note: Number of respondents per jurisdiction in parentheses. 25
Table 2: How important were the following aspects of campaigns (1=did not do / not important; 5= extremely important), reported by all candidates. Plurality RCV Volunteers 4.12 4.09 Money for advertising 3.85* 3.32* Speaking at forums 3.70 3.65 Campaign website 3.32* 3.88* Local news coverage 3.31 3.53 Endorsements from groups 3.30 2.93 Social networking sites 3.15 3.32 Support from party 2.66 2.64 Building alliances with rivals 1.98 2.15 n=146 n=74 How important were the following (1=did not do / not important; 5= extremely important), reported by winning candidates. Plurality RCV Money for advertising 3.92+ 3.41+ Volunteers 3.88* 4.54* Endorsements from groups 3.35 3.27 Speaking at forums 3.33 3.50 Social network sites 2.96 3.04 Campaign website 2.94* 3.61* Local news coverage 2.90 2.95 Support from a party 2.79 2.60 Building alliances with rivals 1.68 1.72 n=54 n=24 Note: Mean response scores, per item. Source: Authors' survey of candidates, December 2013 - January 2014. * Significant difference between means (t test at p. <.05 two-tail). + Significant difference at p. <.08 one tail 26
Table 3: Voter recall of campaign activities. Contacted by a campaign: Plurality RCV Difference (col 2 minus col 1) No 34.4 (414) 24.9 (300) Yes 65.6 (791) 75.1 (907) +9.5 Total: 100 100 X 2 =26.1** If contacted, by direct mail to your house : No 8.3 (65) 5.0 (45) Yes 91.7 (720) 95.0 (858) + 3.3 Total: 100 100 X 2 = 7.5* If contacted, was it via email : No 77.4 (603) 68.5 (609) Yes 22.6 (176) 31.5 (280) + 8.9 Total: 100 100 X 2 =16.4** If contacted, in person at home or in public : No 58.5 (461) 43.4 (392) Yes 41.5 (327) 56.6 (511) + 15.1 Total: 100 100 X 2 =38.3** Recall seeing TV or radio ads: No 20.2 (244) 37.9 (452) Yes 79.8 (964) 62.1 (740) - 17.7 Total: 100 100 X 2 =91.5** Recall a pamphlet on your door: No 50.1 (599) 24.5 (296) Yes 49.9 (597) 75.5 (913) +25.6 Total: 100 100 X 2 =168.6** Source: Authors' survey of likely voters conducted November 2013 by the Eagleton Poll (Rutgers University). N=2432. Note: ** = significant at p. <.001; * = at p. <.01 27
Table 4: Voter perceptions of local election campaigns. Negative Campaigning: Non-RCV RCV Difference (col 2 minus col 1) A lot more 10.02 (115) 1.84 (20) - 8.18 A little more 4.97 (57) 2.30 (28) - 2.67 About the same 55.57 (638) 48.39 (527) - 7.18 A little less 13.68 (157) 16.80 (183) + 3.12 A lot less 15.77 (181) 30.39 (331) + 14.6 Total: 100 100 X 2 = 131.8** Candidates Praising Each Other: Frequently 6.26 (71) 4.50 (51) - 1.76 Occasionally 18.68 (212) 19.15 (217) + 0.47 Rarely 14.10 (160) 16.33 (185) - 2.23 Never 60.97 (692) 60.02 (680) - 0.95 Total: 100 100 X 2 = 5.3 Candidates Criticizing Each Other: A great deal of time 25.25 (277) 5.24 (55) - 20.01 Some of the time 34.73 (381) 23.74 (249) - 10.99 Not too much 27.62 (303) 45.28 (475) + 17.66 Not at all 12.40 (136) 35.74 (270) + 23.34 Total: 100 100 X 2 = 257.4** Overall Conduct of Candidate Campaigns: Very satisfied 23.31 (266) 33.13 (374) + 9.82 Fairly satisfied 53.11 (606) 51.37 (580) - 1.74 Not very satisfied 14.90 (170) 11.07 (125) - 3.83 Not at all satisfied 8.68 (99) 4.43 (50) - 4.25 Total: 100 100 X2=41.7** Source: Authors' survey of likely voters conducted November 2013 by the Eagleton Poll (Rutgers University). N=2432. Note: ** = significant at p <.001 28
Figure 1: Candidate perceptions of how they and their rivals were described during the campaign 40 Percent described negatively by rival during campaign (n=213) 35 30 25 20 15 10 5 RCV Plurality 40 35 30 25 20 15 10 5 0 Percent who described their rival in positive terms (n=213) RCV Plurality 29
Figure 2: Candidate perceptions of tone of campaigns 55 50 45 40 35 30 25 20 15 Candidates praised or endorsed others during the campaing? (% yes, n=224) RCV Plurality 55 50 45 40 35 30 25 20 15 Was campaign less negative compared to others? (% saying less negative, n=223) RCV Plurality 30
Figure 3: Probability of Voter Perceiving Campaign as Less Negative 0.30 0.25 0.20 0.15 0.10 RCV Plurality 0.05 0.00 A little less negative A lot less negative Note: Predicted probability of voter perception of campaign tone. Error bars represent 95% confidence intervals for estimates. Source: Post-estimation simulations from multivariate estimates in Appendix 2 Table 1 (Race, age, gender, education, employment status, marital status, party, and interest held constant at mean values). 31
Figure 4: Probability of Voter Perceiving Candidates Criticize Each Other Some or a Great Deal of the Time. 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 Candidates criticize Candidates do not criticize RCV Plurality Note: Predicted probability of voter perception of campaign tone. Error bars represent 95% confidence intervals for estimates. Source: Post-estimation simulations from multivariate estimates in Appendix 2 Table 3 (Race, age, gender, education, employment status, marital status, party, and interest held constant at mean values). 32
Figure 5: Probability of Voter Being Satisfied with Overall Conduct Local Campaigns 0.35 0.30 0.25 0.20 0.15 0.10 RCV Plurality 0.05 0.00 Not at all satisfied Very satisfied Note: Predicted probability of voter perception of campaign tone. Error bars represent 95% confidence intervals for estimates. Source: Post-estimation simulations from regression coefficients in Appendix 2 Table 4 (Race, age, gender, education, employment status, marital status, party, and interest held constant at mean values). 33