The Electoral Choices of Voters with Coalition‐Straddling Consideration Sets

ObjectivesPreelectoral party coalitions are common in multiparty systems. We examine the proposition that individuals who choose between parties from competing coalitions think and behave different ...

can stand to gain. Political campaigns can benefit from knowing which groups of voters are more likely to be persuadable, and the political parties and the media can improve their campaign information, which potentially could lead to higher-quality voter decisions.
Second, we assess if they are more likely to engage in postelection attitude change in favor of the selected coalition parties. After all, when they ultimately cast their votes, they presumably have to resolve a larger internal conflict. Liking the coalition parties is one solution. Here, we study the attitudinal consequences of behavior, a chain of causality that is fascinating, but sometimes overlooked. Taken together, these sets of outcomes can illustrate the concept's importance both before and after the election. 1 Since preelectoral coalitions are formed in many democratic countries (Golder, 2006), the potential for this concept in improving the understanding of political behavior is substantial. Furthermore, it is straightforward to measure coalitionstraddling, which allows for an easy adoption of the concept in many different multiparty systems.

Preelectoral Coalitions and Political Behavior
Preelectoral coalitions can have important consequences for political behavior. This is particularly evident in studies on strategic voting. For example, if a smaller party in a favored coalition is at risk of falling below an electoral threshold, supporters of bigger coalition parties sometimes vote for the threatened party because losing its seats would undermine the strength of the coalition (Cox, 1997;Gschwend, 2007). This type of research alludes to the usefulness of accounting for how coalitions influence voter behavior. Next, we turn to another, previously undertheorized area of how coalitions affect citizens.
There are primarily three reasons to believe that individuals who choose between parties from competing coalitions think and act differently during the election campaign compared to individuals who only consider voting for parties of the same coalition. First, individuals who choose between parties from opposing coalitions affect the relative size of the government coalitions whereas those who only consider parties from one coalition do not. The traditional median-voter model posits that voters in the political center are the most pivotal since they decide the outcome of elections (Downs, 1957:117). Here, we suggest that a crucial group of pivotal individuals is voters with coalition-straddling consideration sets because they may swing the election outcome from one to the other. It is possible that they are mindful of their important electoral role. While voters in general elections have a very small chance of influencing its outcome, the p in their voting calculus should be slightly higher, which, in turn, may influence the importance of election-related activities.
Second, coalition-straddling individuals face larger differences in overall policy positions. Figure 1 illustrates a hypothetical election where two preelectoral coalitions of three parties each are competing. Parties are placed on a left-right dimension where Party A is the most leftleaning and Party F is the most rightleaning. For an individual who chooses between Party A and Party B, the choice may not be easy, but at least this person can take comfort in knowing that the parties are from the same coalition (Coalition I) and that the overall policy outcomes of this coalition will not be too far from the individual's preferred outcome. In contrast, the situation for an individual who is choosing between Party C and Party D is considerably different. Regardless of whether the person ultimately selects Party C or Party D, the policy outcomes of the coalition will be further from what the Multiparty System with Two Coalitions on a Left-Right Dimension individual prefers. With a negative outcome looming larger, there is more at stake for an individual who is considering parties from opposing coalitions.
The illustration above is obviously a simplification. It might appear as though the parties are at the same level of conflict. They are generally not. Instead, parties within the same coalition coordinate with each other before the election against the parties of the opposing coalition. It should also be noted that the model simplifies the ideological space. Research shows that voters in Western democracies differentiate between at least two relevant dimensions (Caughey, O'Grady, and Warshaw, 2019;Feldman and Johnston, 2014).
Third, individuals with coalition-straddling party preferences may anticipate that their vote choice is likely to have larger implications for their political attitudes after the election compared to individuals who only consider parties from one coalition. Once coalition-straddling citizens cast their votes, they may feel internally conflicted because they voted for a coalition that they did not wholeheartedly embrace. Since cognitive dissonance is uncomfortable and a powerful determinant of change (Festinger, 1957), this group of individuals should decrease it by liking all of the coalition parties more. Conversely, feelings toward the parties of the rejected coalition are likely to become more negative in order to make it easier to live with the decision.

Considering a Vote for Some but Not All Parties
The choice facing voters in multiparty or multicandidate elections can be overwhelming because of the many options. In order to simplify the choice, citizens tend to form what is known as a consideration set, that is, a set of parties or candidates that they are actively choosing from (Karlsen and Aardal, 2016;Wilson, 2007). For example, in an election where there are eight different parties represented in parliament, one voter may only actively decide between three centrist parties whereas another voter only considers the two parties with the most generous welfare state policies.
The consideration set approach is inspired by consumer research, which shows that consumers are not actively choosing between all the available options. For instance, consumers often have some acceptable brands of cereal that they are deciding between and some that are not even considered (Roberts and Lattin, 1991). In the realm of politics, evidence shows that factors such as policy positions and strategic considerations affect what parties or candidates end up in the consideration set, and that other factors are more influential at the choice stage, that is, when voters make their final decision (Oskarson, Oscarsson, and Boije, 2016).
While the burgeoning literature on consideration sets is providing a more complete understanding of voter behavior in multiparty and multicandidate elections (Oscarsson and Rosema, 2019), more work is needed on the effects of the qualitative differences between types of sets. As discussed above, one plausible key distinction is between having a set that only includes parties from one coalition versus having a set that includes parties from opposing coalitions.
Having parties from different coalitions in a consideration set may theoretically seem close to ambivalence, but our concept is different in several respects. For example, ambivalence is generally characterized as having conflicting positive and negative thoughts about parties or candidates (Lavine, Johnston, and Steenbergen, 2012), whereas a coalitionstraddling consideration set is about voting considerations. Moreover, it is more related to the communication from the parties and their intentions to cooperate with others. Another difference is that research on the impact of ambivalence generally focuses on two-party systems such as the United States, where it helps in explaining political behavior. However, it is less straightforward how it should be applied in multiparty contexts, since vote consideration may concern parties within the same coalitions or over different coalitions.
Being an independent is a related label, but like ambivalence, it is primarily focused on the U.S. system, and is tainted with social desirability problems (Klar and Krupnikov, 2016). The idea of swing voters is another similar concept (Gelman et al., 2016), yet it makes little difference between voters who waver between parties of the same coalition and those who are undecided between parties of different coalitions, which is a key feature of coalitionstraddling.

A Difficult Decision
The literature on determinants of difficulties in deciding between parties is sparse, but more research has been conducted on factors that influence the timing of the vote choice, a related concept. In general, individuals who decide early are party identifiers (Chaffee and Choe, 1980) and they tend to be somewhat older, less politically aware, and pay less attention to campaign media coverage (Fournier et al., 2004). Besides individual-level factors, the information environment matters as well. The less politically engaged are deciding much earlier if they live in battleground states (Henderson and Hillygus, 2016). Moreover, if the information environment is more balanced in how it describes the candidates, a larger share of voters take longer to decide whereas there is no such effect if the media provides a one-sided message (Nir and Druckman, 2008).
Voting advice applications aim at helping voters in finding the party or candidate that is closest to their overall policy preferences. The applications can improve electoral decision making because they have up-to-date records of party positions and can remove unnecessary biases that individuals may have about the various parties. The applications are not simply political diversions for voters, as research shows that their usage increases turnout (Gemenis and Rosema, 2014) and that individuals take the results seriously. In particular, they affect the vote choice of citizens who have not made up their minds about a specific party (Wall, Krouwel, and Vitiello, 2012).
Concerning difficulties related to casting a vote, we expect that (H1) coalition-straddling individuals are more likely to think that deciding on a party is difficult and that (H2) they will rely more on voting advice applications.

Postelection Dissonance Reduction
Attitudes often explain behavior (e.g., Ajzen and Fishbein, 1980), yet there is also a rich literature on effects from behavior to attitudes, as work on cognitive dissonance theory demonstrates. Dissonance arises when an individual has two psychologically inconsistent cognitions (Festinger, 1957) and there are multiple situations behind dissonance-driven change, as experimental studies demonstrate. For example, in a study on HIV prevention, the group of subjects who became the most favorable toward condom use were those who gave a speech on the benefits of condoms and who also were reminded about their own unsafe sexual behavior, presumably because they felt hypocritical (Stone et al., 1994). Dissonance happens when attitudes or behavior are misaligned, as McCrea, Leviston, and Walker (2015) also show. In fact, they find that voting for a climate sceptic party is a stronger predictor of attitudes on climate change than climate change attitudes on voting behavior. That is, behavior influences attitudes more than attitudes affect behavior.
In the same vein, we expect that coalition-straddling individuals will be the most affected after casting their vote because they have, in effect, voted for a coalition that they did not fully embrace. This should lead to a more pronounced attitudinal change toward parties in the coalition that was finally chosen in order to create a better alignment with behavior. Consequently, we predict that (H3) vote choice will lead to more aligned postelection party preferences among coalition-straddling individuals.

The Swedish Case
In the 2014 election, the incumbent center-right government (the Alliance for Sweden) stated clearly that it intended to govern together. In contrast, the statements from the left were more ambiguous; the Social Democrats and the Greens suggested to voters that they intended to govern together, yet this coalition did not formally include the Left Party or the Feminist Initiative (Aylott and Bolin, 2015). Nevertheless, these parties and their potential voters clearly position themselves to the left. The Left Party has been a former government support-party to the Social Democrats and the leader of the Feminist Party publicly supported the left-wing government (Jakobson, 2014). Therefore, we choose to include them in the left-green potential coalition alternative. Since the populist, anti-immigrant party Sweden Democrats is positioned outside the traditional coalitions, individuals who had it in their consideration set have been excluded from the analysis.

Participants
We rely on panel data collected between March 2014 and October 2014. Panelists were interviewed seven times from March to October 2014. Between these time points, two major elections took place in Sweden: the European Parliamentary election on May 25 and the general national election on September 14. Of the 19,432 individuals who completed wave 1, 9,270 individuals finished all seven surveys. Participants were recruited via physical postcards and online ads. All data collections were conducted online.

Measures
Consideration sets were measured at multiple occasions, but we rely on the measure from the first wave because it was conducted before the election campaigns had started in earnest. Our key independent variable, coalition-straddling consideration set, was coded as 1 if respondents weighed voting for parties from both preelectoral coalitions and coded as 0 if respondents only considered parties of the same coalition. In our data at wave 1, 17 percent (n = 2,485) are coded as coalition-straddling individuals and 83 percent (n = 12,423) are not. The consideration set was measured with the following question: "What party or parties do you consider voting for in the parliamentary election of 2014?" Ten major parties were labeled as options, and respondents could select none and other. We operationalize the two coalitions as the red-greens (Social Democrats, Green Party, Left Party, and Feminist Initiative) and the Alliance for Sweden (Moderate Party, Liberal People's Party, Centre Party, and Christian Democrats). For the number of considered parties, we count each party as one (wave 1: M = 2.08, SD = 0.98). That is, respondents considered voting for about two parties, on average.
Left-right ideology was measured on an 11-point scale and recoded to range from 0 = most leftleaning to l = most rightleaning (wave 1: M = 0.44, SD = 0.25). Ideological strength is an indicator that folds the left-right ideology measure, recoded to vary from 0 to 1, with higher values associated with more extremeness on either left or right (wave 1: M = 0.43, SD = 0.28). The latter indicator functions, in effect, as a squared term of left-right ideology. Party identification operationalizes the degree to which respondents identify with parties on the left or right coalition, with higher scores signifying great identification with any of the parties of the right coalition. The six-point measure has been recoded to vary from 0 to 1 (wave 2: M = 0.40, SD = 0.31). Partisan strength measures how much respondents identify with any party, with 0 = not partisan, 0.5 = somewhat partisan, and 1 = strong partisan (wave 2: M = 0.48, SD = 0.38). In addition, we control for political interest, a four-point measure recoded to range from 0 to 1 (wave 1: M = 0.82, SD = 0.20) and political awareness, a 19-item scale created from knowledge questions posed at multiple waves and coded to vary from 0 to 1 (␣ = 0.77, M = 0.64, SD = 0.27). As for background variables often associated with political attitudes and behavior, we control for female sex (1 = female (39.3 percent), 0 = male (60.7 percent)), age in years (M = 44.7, SD = 15.3), education, an eight-point measure recoded to range from 0 to 1 (M = 0.75, SD = 0.19) and income, a 12-point measure also recoded to vary between 0 and 1 (M = 0.48, SD = 0.28). With the exception of income, which was measured at wave 2, the other demographic variables were measured at wave 1.
We are relying on two outcome variables that tap difficulties associated with voting decisions. The extent to which respondents thought it was hard to decide on which party to vote for was measured on a five-point scale and recoded to range from 0 to 1, where 0 = "very easy" and 1 = "very hard." Voting advice application usage was assessed with a question on whether respondents had relied on independent tests, often available at online news sites, where their policy preferences were matched with the party most in line with those preferences. Responses to the question have been recoded as 0 = "None," 0.5 = "Yes, one or few," and 1 = "Yes, several." In terms of voting preferences, a right-of-center vote (i.e., a vote for any of the four parties of the right-leaning coalition) has been coded as 1 while a vote for any of the four left-of-center parties has been coded as 0. The variables were measured at wave 6, which took place immediately after the general election on September 14. Postelection change in party approval (M = 0.04, SD = 0.08) was measured by subtracting approval ratings of the eight parties that form either coalition at wave 3 (␣ = 0.90, M = −0.07, SD = 0.23) from approval ratings of the same eight parties at wave 9 (␣ = 0.91, M = −0.02, SD = 0.24). The measure has been recoded to range from −0.5 to 0.5 where higher ratings indicate higher approval for right-leaning parties.

The Difficult Vote Choice
Two dependent variables are used to test our predictions that individuals with parties from different coalitions in their consideration set face more challenging electoral decisions. Both dependent variables are treated as continuous and we estimated the models with ordinary least squares. The results are presented in Table 1. First, we expect that a coalition-straddling consideration set makes it harder to decide which party to vote for. In line with this prediction (H1), our results in the first column of Table 1 show that effect is positive and statistically significant (b = 0.072, SE = 0.010, p < 0.01). That is, a coalition-straddling set is associated with an around 7 percentage points higher value, all else equal. Second, if the vote choice is challenging, we expect that this group of voters will be more likely to turn to voting advice applications, which are available at many online news outlets. As predicted (H2), we find that the conflicted respondents are more likely to rely on voting advice applications (b = 0.040, SE = 0.012, p < 0.01). 2 In substantive terms, a 4percentagepoint higher value for coalitionstraddling is a small effect, yet it is still an indication that they may be behaviorally influenced by the importance of their choice.

Change in Postelection Party Approval Ratings
We posit that individuals with coalition-straddling consideration sets should experience larger changes in their postelection political perceptions. That is, in order to make peace with their ultimate party choice, they should increase their approval of all parties of the coalition, and not just the one they voted for. We test this proposition by interacting vote choice with coalition-straddling consideration set. Our dependent variable is the gain score of pre and post approval ratings for the eight political parties. Since we expect a stronger effect of the vote among individuals who straddle two coalitions, the interaction effect should be positive and statistically significant in order to support our hypothesis.
In line with our prediction (H3), we find that the interaction is indeed in the right direction and significant (b = 0.020, SE = 0.007, p < 0.01). Figure 2 illustrates the predicted values from the interaction in Table 2. Individuals who vote for a right-of-center party increase their approval of all parties on the right. However, there is no statistical difference between consideration set type for individuals who chose a left-of-center party. One reason for the lack of a statistically significant effect could be that the parties on the left were less united in their preelectoral coalition. Moreover, one of the parties (the Feminist Initiative) did not reach parliamentary representation, and only two of the left-wing parties (the Social Democrats and the Greens) created a formal government, whereas the Left Party played the role of a support party. This could have made reactions from left-of-center voters more heterogeneous.
generalized ordered logit or OLS. Since it easier to interpret OLS coefficients, we chose this model for presentational purposes. The generalized ordered logit coefficient is 0.21 (SE = 0.06, p < 0.01), which can be exponentiated to the proportional odds ratio of 1.24.

Discussion and Conclusion
We introduced the concept of coalitionstraddling, which refers to individuals who consider voting for parties from opposing coalitions, and outlined theoretical reasons as to why it is consequential for political behavior, including the fact that these individuals are crucial to election outcomes. In our conception, individuals with coalition-straddling consideration sets are aware, implicitly or explicitly, of their central role during election campaigns. That makes them think and act differently compared to individuals who only consider voting for parties of the same coalition.
We found that voters with coalition-straddling consideration sets were more likely to think that the vote choice was hard and to use voting advice applications. We also found support, albeit not complete, for the notion that once the coalition-straddling individuals cast their vote, they were more affected by their choice and shifted their party preferences more toward the favored coalition. Overall, the findings suggest that coalitions are important throughout the decision-making process, and that voters who support parties from two competing government alternatives make their decisions differently than other voters.
A limitation with our study is that we do not have direct evidence that voters with coalition-straddling consideration sets are aware of why their electoral choice is more important than that of individuals who only consider voting for parties of the same coalition. We suggested three possible reasons (i.e., they decide which coalition wins the election, their ideal policy platform is generally far from coalition policy platforms, and they will experience stronger cognitive dissonance after the election), but we do not know the relative importance of the individual reasons. Uncovering such evidence in a future study would be valuable.
Another limitation is that we have only conducted our study in one country, Sweden. Whether effects of coalitionstraddling are stronger or weaker should be tested with comparative data. While in the U.S. context there seem to be few real swing voters or independents (Gelman et al., 2016;Klar and Krupnikov, 2016), two related concepts, our results suggest that a sizable share of the electorate is torn between different coalitions. The exact size of this group would depend on the electoral context and electoral system, but it seems premature to call it "mythical" from a comparative perspective.

Supporting Information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Table S1. Conditional effects of party identification and left-right ideology on retrospective governmental performance Table S2. Effect of coalition straddling on ideological strength (fixed effects) Figure S1. Effect of coalition-straddling consideration set by left-right ideology on retrospective governmental performance Figure S2. Cross-lagged model of ideological strength and coalition-straddling consideration set