Ever since Campbell et al. (Reference Campbell, Converse, Miller and Stokes1960) first published The American Voter, scholars of political behaviour have been aware of the influence that parties have on citizens’ political attitudes. Campbell et al. described the political party as “an opinion-forming agency of great importance” (128). Numerous studies have adopted experimental approaches to show that parties influence a wide variety of policy preferences (Cohen, Reference Cohen2003; Druckman et al., Reference Druckman, Peterson and Slothuus2013; Merolla et al., Reference Merolla, Stephenson and Zechmeister2008). Nearly all of these studies rely on the concept of party identification; that is, they assume that citizens who identify with a party adopt the policy positions of that party in order to show support for it. More recently, this phenomenon has been called “partisan motivated reasoning” (Bolsen et al., Reference Bolsen, Druckman and Cook2014; Druckman et al., Reference Druckman, Peterson and Slothuus2013; Leeper and Slothuus, Reference Leeper and Slothuus2014).
To be sure, there is considerable evidence that parties influence preferences in the United States and elsewhere. The most prominent study to consider the influence of party cues in Canada, however, concludes that party cue effects are limited because Canadian parties have long had more ambiguous ideological profiles than parties elsewhere. That study, Merolla et al. (Reference Merolla, Stephenson and Zechmeister2008), along with most other studies, considers party cue effects on partisan groups. In other words, it assesses whether parties influence the attitudes of people who identify with them. These studies all show that parties influence partisan groups by interacting party cue treatments with party identification variables. However, interactions between randomly assigned treatments and non-experimental covariates cannot show that covariates (in this case, party identification) influence the effect in question (here, the influence of parties on policy preferences) (Gerber and Green, Reference Gerber and Green2012; Kam and Trussler, Reference Kam and Trussler2017). All they do is show that, on average, a given partisan category reacts differently from another.
In this article, we cast doubt on the interpretation of these results as reliant solely on party identification in multi-party systems such as Canada, and potentially beyond. While positive identification with a party (Merolla et al., Reference Merolla, Stephenson and Zechmeister2008) may lead citizens to adopt its positions and negative identification with a party (McGregor et al., Reference McGregor, Caruana and Stephenson2015; Meideros and Noël, Reference Medeiros and Noël2014) may lead others to reject that party’s positions, there is nothing in either of the two dominant theories used to account for party cue effects that is necessarily about either positive or negative identification. In addition to the influence that parties have on their positive and negative identifiers, they likely influence many other people who simply have positive or negative attitudes toward them regardless of their identification. This is particularly true in multi-party systems where more general feelings toward parties and leaders have been found to be consequential for political behaviour (Blais et al., Reference Blais, Guntermann and Bodet2017; Guntermann, Reference Guntermann, Oscarsson and Holmberg2020). In other words, partisan identification may not be the only moderator of party cue effects; those who like a party more should adopt policy positions that are closer to that party’s positions, while those who like it less should adopt positions that are more distant from it, irrespective of party identification.
To test this, we first reanalyze data from the Merolla et al. (Reference Merolla, Stephenson and Zechmeister2008) study, adding party feeling thermometers to their analyses. We find significant results on issues where the authors found none. Specifically, some treatments identified by the authors as having no effect in fact had positive effects on participants who liked the party and had negative effects on those who did not, regardless of their party identification. We extend this replication to analyses of original data from two party cue experiments on environmental policy proposals. Here we show that while effects on partisan groups are weak or non-existent, the cues nevertheless had powerful effects on participants who strongly liked or disliked each of the parties or their leaders. We conclude that parties have more influence than scholars previously thought. Focussing on partisan identification alone may lead to null results, whereas taking into consideration party or leader evaluations is more likely to reveal the party cue effects that actually occur.
Why Party Identification May Not Be the Only Moderator of Party Cue Effects
Numerous studies convincingly show that parties influence people’s policy preferences. Beginning with Cohen (Reference Cohen2003), scholars have experimentally manipulated exposure to policy positions, with treated participants seeing positions attributed to parties and/or their leaders and control group participants seeing them attributed to anonymous politicians (Druckman et al., Reference Druckman, Peterson and Slothuus2013; Kam, Reference Kam2005; Merolla et al., Reference Merolla, Stephenson and Zechmeister2008). These scholars find that their treatments are conditioned by partisan identification.
There are two theories that account for the influence of these cues on opinions. According to the first, citizens use parties as heuristics to help them figure out their positions on policy issues (Downs, Reference Downs1957; Kam, Reference Kam2005). Research in this tradition suggests that such informational shortcuts help voters form preferences even though they lack information (Lupia, Reference Lupia1994; Lupia and McCubbins, Reference Lupia and McCubbins1998). More recently, scholars have argued that parties influence citizens’ preferences because citizens identify with a party and seek to support that party by adopting its policy positions. This theory has been called partisan motivated reasoning (Bolsen et al., Reference Bolsen, Druckman and Cook2014; Druckman et al., Reference Druckman, Peterson and Slothuus2013; Leeper and Slothuus, Reference Leeper and Slothuus2014).
Recent studies have found more support for the view that party cue effects are about showing support for the party with which one identifies, rather than about facilitating decision making (Bullock et al., Reference Bullock, Gerber, Hill and Huber2015). Citizens take longer to answer policy questions when exposed to party cues, suggesting that they do not make it easier for people to express their opinions (Petersen et al., Reference Petersen, Skov, Serritzlew and Ramsøy2013). Moreover, only people who experience strong physiological reactions to party cues are influenced by party positions, suggesting that party cues depend on affective reactions (Petersen et al., Reference Petersen, Giessing and Nielsen2015). Finally, people with more political knowledge react more strongly to cues than do people with less knowledge (Slothuus, Reference Slothuus2016). While partisan motivated reasoning has received more support than the heuristic perspective in recent years, the main prediction of both theories is the same: partisans of a party become more supportive of that party’s positions when they see them.
Experimental studies on party cue effects were initially conducted in the United States. However, subsequent studies have been conducted in other contexts. Scholars have notably conducted party cue experiments in Canada (Merolla et al., Reference Merolla, Stephenson and Zechmeister2008), Denmark (Aaroe, Reference Aaroe2012; Slothuus, Reference Slothuus2016), Mexico (Merolla et al., Reference Merolla, Stephenson and Zechmeister2007) and Spain (Guntermann, Reference Guntermann2019). Overall, these studies have found weaker effects than in the United States (Bullock, Reference Bullock2011). Nearly all of these studies have adopted the same approach as earlier American studies, which is to use party identification as a moderator of reactions to party cue effects.
However, such studies rely on the strong assumption that party identification is the only variable that moderates party cue effects. If it is not, and one or more variables that vary within partisan groups moderate the impact of parties, these studies will underestimate party influence. The difficulty is that experiments do not allow researchers to randomize respondents’ partisanship and thus prevent scholars from clearly testing moderation effects. Assessing moderation effects requires the same kind of systematic assessment of alternative explanations that scholars routinely conduct in observational studies (Gerber and Green, Reference Gerber and Green2012; Kam and Trussler, Reference Kam and Trussler2017). Particularly since there is strong evidence that people adapt their policy preferences to their candidate and party preferences rather than simply to fit their party identification (Lenz, Reference Lenz2012), it is important to consider whether party identification adequately moderates party cue effects.
We argue that there is no reason to expect party identification to be the only moderator of party cue effects, at least outside the United States. There is nothing about either of the two theories commonly used to explain party cue effects that is necessarily about party identification. The heuristic perspective (Downs, Reference Downs1957; Kam, Reference Kam2005) is about citizens using parties to help them figure out their own positions on policy issues. However, according to a prominent argument that falls within this perspective, what really matters is whether people perceive the cue-giver as knowledgeable and as sharing their interests (Lupia and McCubbins, Reference Lupia and McCubbins1998). Lupia (Reference Lupia1994), in a widely cited article, shows that low-knowledge people used a cue from car insurance companies, which they presumably perceived as opposing their interests, to express the same positions as those with higher knowledge. Thus, people can compensate for low knowledge by adopting policy positions that are closer to those of cue-givers they are more positive about (or more distant from cue-givers they are more negative about).
The other theory commonly used to account for party cue effects, partisan motivated reasoning (Bolsen et al., Reference Bolsen, Druckman and Cook2014; Druckman et al., Reference Druckman, Peterson and Slothuus2013; Leeper and Slothuus, Reference Leeper and Slothuus2014), is based on the more general theory of directional motivated reasoning whereby citizens seek to support their pre-existing attitudes (Kunda, Reference Kunda1990; Lodge and Taber, Reference Lodge and Taber2013). However, there is nothing about motivated reasoning that is necessarily about party identification (Druckman and McGrath, Reference Druckman and McGrath2019). To take one mechanism as an example: Lodge and Taber (Reference Lodge and Taber2013) have found that people transfer affect from one object to another; why, then, would citizens not transfer the affect they feel toward political parties and leaders to policy issues?
We thus expect that, other things being equal, experimental participants’ attitudes toward the sources of elite messages should influence their reactions to them. They should adopt preferences that are closer to the positions expressed by cue-givers they like, while adopting preferences that are more distant from the positions of cue-givers they dislike. Note that while we focus on feeling thermometers toward parties and leaders, our argument is not that these attitudes are the only moderators of party cue effects. We simply argue that how people feel about parties and their leaders plays a role in moderating how they respond to party cue effects that party identification does not account for.
Partisan Attitudes by Partisan Groups in Canada
Studies of party cue effects focussing on the reactions of partisan groups assume that no other variables condition the reception of cues. We argued above that attitudes toward cue-givers, either parties or leaders, are a likely additional moderator. Here we show that party and leader evaluations are not well summarized by party identification. There is no reason for partisan attitudes to be a perfect reflection of party identification. As Mason (Reference Mason2018) shows, such a reflection is dependent on a convergence between partisan and other political and social identities.
The 2015 Canadian Election Study (Fournier et al., Reference Fournier, Cutler, Soroka and Stolle2015) asked respondents how much they like or dislike each of the main parties and leaders on a scale from 0 to 100, where 0 means they really dislike the party or leader and 100 that they really like them. Party identification adequately summarizes party ratings if most partisans of a particular party prefer that party to the alternatives—if they like that party (that is, rate it above the midpoint) and dislike others (that is, rate it below the midpoint). Figure 1 plots ratings of the Conservative and Liberal parties relative to each other by Liberal and Conservative partisan identifiers, respectively, where ratings of the in-party are on the vertical axis and ratings of the out-party are on the horizontal axis.
Figure 1. Attitudes toward Parties by Party Identification
Note: Figures 1(a) and 1(b) are jittered scatterplots of ratings by Liberal and Conservative partisans, respectively, of their own party (y-axis) and of the other large party (x-axis). Points above the diagonal dashed line represent partisans who prefer their party to the other party. Respondents in the top left quadrant are partisans who rate their own party above the midpoint and dislike the other large party below the midpoint.
For party identification to fully reflect attitudes toward parties, we would expect partisans to be above the 45-degree lines, indicating that they prefer their party to the other party. Moreover, we would expect them to be in the top left quadrant indicating that they evaluate their party above the midpoint (50) and other parties below it. As we can see in Figure 1, partisans of the two main parties in Canada do not have a clear preference for their own party. Only 49 per cent of Liberal identifiers have a consistent preference for their party over the Conservatives, and only 47 per cent of Conservatives have a consistent preference for their party over the Liberals. Party identification even less adequately reflects partisan attitudes if we consider the New Democratic party (NDP) (see online Appendix, section 1). Only 9 per cent of Liberal partisans consistently prefer their party to both the Conservatives and the NDP, and 18 per cent of NDP partisans have a consistent preference for their party over the Liberals and Conservatives. The percentage of Conservatives with a consistent preference over the other two parties is 35 per cent. Party identification thus reflects party ratings particularly badly for Liberal and NDP partisans. In the United States, party attitudes reflect party identification much more clearly. In the 2016 American National Election Study, 72 per cent of Democrats like their party and dislike the Republican Party while 69 per cent of Republicans have a similar preference for their party (ANES, Reference American National Election Studies (ANES)2019).
Does party identification reflect leader evaluations any better? Clearly it does not. Only 13 per cent of Liberal partisans had a clear preference for Justin Trudeau over the Conservative (Stephen Harper) and the NDP (Thomas Mulcair) leaders. The percentages with consistent leader preferences were 33 per cent for Conservatives and 16 per cent for NDP partisans (see section 2 of the online Appendix for leader ratings by partisan group). Thus, leader evaluations are also poorly represented by party identification, particularly for Liberal and NDP partisans, making them possible additional moderators of party cue effects. Thus, to the extent that party cue effects are driven by affect in addition to party identification, relying exclusively on party identification as a moderator of party cues may miss an important part of partisan cue taking in Canada.
Study 1: Reanalysis of Merolla et al. (Reference Merolla, Stephenson and Zechmeister2008)
Merolla et al. (Reference Merolla, Stephenson and Zechmeister2008) recruited a sample of 196 university students in 2004 and exposed them to cues from the three major parties in Canada (Conservative, Liberal and NDP) on four different issue: legalizing same-sex marriage, reducing spending on social services, changing the Employment Insurance Act to establish a status for seasonal workers, and creating an Office of Ombudsman for Older Adult Justice. Respondents were randomly assigned to receive cues attributed to anonymous politicians, to the Liberals, the Conservatives or the NDP. The 4 × 3 issue-party combinations thus provide 12 tests of party cues. Respondents were asked whether they strongly support, support, neither support nor oppose, oppose or strongly oppose each policy. The authors scaled the outcome variable so that higher values correspond to more left-wing responses. We rescaled them from 0 to 1. Following the usual practice in the analysis of experiments, we ran ordinary least squares analyses. To determine whether there was any evidence that the party cue treatments worked, we used the kernel estimator proposed by Hainmueller et al. (Reference Hainmueller, Mummolo and Xu2019).Footnote 1
Following the conventional practice of interacting treatment variables with party identification, the authors found significant effects of six of the cues. When considering both party identification and party ratings, we found significant effects of nine of them. More importantly, we found anomalies that a focus on party identification cannot explain. For instance, we found that on some of the issues, there was no overall effect on a partisan group but there was an effect on partisans who gave their party a high rating. This finding could arguably be explained by stronger identifiers being more attached to their party and thus following it more (see Morin-Chassé and Lachapelle, Reference Morin-Chassé and Lachapelle2020). As such, this is a weak challenge to the notion that party identification fully moderates the effect. On other issues, we found that nonpartisans were influenced by parties with which they did not identify.
Figure 2 shows examples of each anomaly. It shows the average treatment effect (ATE) of two treatments for participants with ratings of the cue-giving party in the low, middle and top terciles.Footnote 2 In Figure 2(a), we show that the Conservative social services cue had a positive effect only on Conservative partisans who gave the Conservatives a rating in the top tercile. Figure 2(b) shows that the NDP social services cue had an effect on respondents who did not even identify with the NDP but rated that party above the bottom tercile. Overall, we found that 3 of the 12 party cue treatments had effects on identifiers that differed depending on their party ratings (either the effect changed sign or was only significant at some values of party ratings). We also found that 7 of the treatments had effects on non-identifiers who gave party ratings above a certain level. The first anomaly can perhaps be explained by stronger partisans being more easily persuaded by their parties than weak partisans. The second cannot, and it is clear evidence that party identification is not the only reason people are influenced by parties. Many people like parties and leaders with which they do not identify, and this affect is likely to influence their policy attitudes when they are exposed to cues from them.
Figure 2. Party Cue Effects by Party Identification and Party Rating
Note: Figures 2(a) and 2(b) show the average treatment effect (ATE) of two of the cues at different values of the feeling thermometer for the party giving the cue in the Merolla et al. (Reference Merolla, Stephenson and Zechmeister2008) study. CPC: Conservative Part of Canada; NDP: New Democratic Party.
Study 2: Cues from the Prime Minister on Environmental Policy
To further test for additional moderators of party cue effects, we analyzed data from a survey experiment administered by telephone in the 2013 wave of the Canadian Surveys on Energy and Environment (Lachapelle et al., Reference Lachapelle, Borick and Rabe2012). This experiment was administered while Stephen Harper was prime minister, and it informed respondents that a possible policy to reduce greenhouse gas emissions was to set a hard cap on emissions from coal-fired electricity (this policy was proposed by the Harper government at the time). A random half of respondents was simply informed that this was a possible policy. The other half was told that Stephen Harper had proposed this policy. Respondents were then asked whether they strongly support, somewhat support, somewhat oppose or strongly oppose the policy. The outcome variable was rescaled from 0 to 1 (0 = strongly oppose, 0.33 = somewhat oppose, 0.67 = somewhat support, 1 = strongly support).
We first assess how partisan groups reacted to the treatment (where party groups are operationalized by vote choice, a reasonable proxy, since no partisan identification variable was available in these data). The average treatment effect among self-identified Conservative voters was 0.04. The strongest effect was found among Bloc Québécois supporters (0.11). However, these results were far from significant at conventional levels (p = .328 and p = .314).
In recent years, a number of scholars have argued that negative partisanship is an important determinant of how people relate to the political world (McGregor et al., Reference McGregor, Caruana and Stephenson2015; Medeiros and Noël, Reference Medeiros and Noël2014). The question used by Medeiros and Noël (Reference Medeiros and Noël2014) to assess negative party identification was included in the 2013 survey. It asked if there was a party that respondents would never vote for. Among the 65 per cent who do not reject the Conservatives, there was a weak effect of 0.07 (p = .003), which contrasts to no effect (0.003, p = .915) among those who say they would never vote Conservative. Negative partisanship thus does not clearly moderate the party cue effect: negative partisans do not react to the cue while other participants do.
However, leader ratings did a much better job moderating the treatment effect. Respondents were asked to rate Harper’s performance on a scale from 0 to 10, where 0 is very poor and 10 is very good. Figure 3 shows the average treatment effect for respondents in the bottom, middle and top terciles of the Harper rating scale. As we can see, there is no effect at Harper ratings in the bottom two terciles. In the top tercile, the effect of the Harper cue is significantly positive. Thus, when assessing the effects of cues from the prime minister, the best moderator was participants’ assessments of the prime minister.
Figure 3. ATE of Harper Cue by Harper Rating
Note: Figure 3 shows the average treatment effect (ATE) at different values of the Stephen Harper rating scale. Confidence intervals that only cover positive values show that the ATE is positive at the corresponding Harper rating.
Study 3: Party Cues on Environmental Policy
The final study is an experiment that was included in the Canadian Climate Politics panel study of 2019. The survey asked respondents to indicate their level of support for four distinct policies that were proposed by the Liberal party, the Conservative party, the NDP and the Green party during the 2019 federal election. A random half of respondents read descriptions of the policies, while the other half read identical descriptions that attributed each to the respective political party.
The policies are: [The Green party’s proposal to] End all imports of foreign oil and supply the Canadian market with Canadian oil and gas; [The Liberal party’s policy of] Apply[ing] a carbon price in every province without one and returning the proceeds to taxpayers; [The Conservative party’s proposal to] Require companies to pay into a clean technology fund if they release more carbon than the government allows; and [The NDP’s proposal to] Eliminate government aid to the fossil-fuel industry and cancel the expansion of the Trans Mountain pipeline. Respondents were asked whether they strongly support, somewhat support, neither support nor oppose, somewhat oppose or strongly oppose each proposal. We recoded the outcome variables from 0 to 1 (0 = strongly oppose, 0.25 = somewhat oppose, 0.5 = neither support nor oppose, 0.75 = strongly support, 1 = strongly support).
The left panel of Figure 4 shows the average treatment effect of the party cues on support for each proposal by party group. As we can see, the only party cue that has an effect in the right direction on the relevant party group at a conventional level of significance is the Conservative cue on clean technology. The effect is only 0.05 though (p = .013). The right panel shows how the average treatment effect varies by party rating. Respondents were asked how they feel about each party on a scale from 0 to 100, where 0 means they really dislike a party and 100 that they really like it. We present treatment effects in the bottom, middle and top terciles of the party rating variables. As we can see, the Conservative Party cues had a positive effect among respondents in the top tercile of the feeling thermometer and a negative effect among those in the bottom and middle terciles. The Green, Liberal and NDP cues had negative effects among respondents with low or moderate ratings of those parties (although the NDP effect just misses significance in the bottom tercile). Once again, ratings of the cue-giver are a better moderator of party cue effects than are people’s partisan group.
Figure 4. Party Cue Effects
Note: Figures 4(a) and 4(b) show the average treatment effect (ATE) at various values of party preferences. Confidence bands that do not overlap 0 tell us that the ATE is positive or negative at given values of vote choice or of the party feeling thermometers. GP: Green Party; LPC: Liberal Party of Canada; NDP: New Democratic Party; BQ: Bloc Québécois; PPC: People's Party of Canada.
Conclusion
We follow up on the conclusion in Merolla et al. (Reference Merolla, Stephenson and Zechmeister2008) that party cue effects are weak in Canada due to the ambiguous ideological positions of Canadian parties. We argue that party cue effects are not necessarily dependent on party identification. A long line of literature leads us to expect citizens’ evaluations of the source of a message to shape how they react to it (Lodge and Taber Reference Lodge and Taber2013; Lupia, Reference Lupia1994; Lupia and McCubbins, Reference Lupia and McCubbins1998). Thus, we argue that responses to party and leader cue experiments depend on people's ratings of those cue-givers.
We reanalyze the experiment run by Merolla et al. (Reference Merolla, Stephenson and Zechmeister2008) using party ratings in addition to party identification and find more significant party cue effects than they originally found. We also find some results that clearly show that partisanship does not fully account for people’s reactions to the cues. In particular, on some issues, partisans’ reactions depend on how much they like or dislike the party with which they identify. We also find that, on other issues, nonpartisans react positively to cues from parties they like.
We then present the results of a prime-ministerial cue experiment and a party cue experiment on environmental policy. We showed that simply focussing on the effects of those cues on partisan groups leads to weak or even non-existent effects. When considering how people feel about the sources of the messages, however, we find much stronger positive effects among participants who like (or negative effects among participants who dislike) the leader (or party) that the cue is coming from.
In sum, focussing exclusively on partisan groups as being affected by elite cues potentially obscures many of their effects. We suggest that scholars should instead consider how evaluations of the source of the cue moderate those effects, perhaps in addition to party identification. Doing so uncovers effects of parties on attitudes that are not picked up when the focus is exclusively on party identification. Echoing the recent literature on negative partisanship (McGregor et al., Reference McGregor, Caruana and Stephenson2015; Medeiros and Noël, Reference Medeiros and Noël2014), we have uncovered additional ways in which citizens’ relation to parties affects their receptivity to cues that are not uniquely attributable to party identification. Therefore, greater consideration of the extent to which attitudes toward leaders and parties condition the reception of party cue effects can help show that parties and leaders have a stronger influence on people’s policy preferences than previously thought.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0008423920000608
