Sorting between and within coalitions: the Italian case (2001–2008)

Abstract How do voters sort within an electoral coalition? Voting literatures on ideology, character valence, and issue ownership provide explanations for inter-coalition or inter-party voting, yet the coalition context remains understudied. Do voters in proportional coalition-based systems use the same ideological and issue-based heuristics ascribed to them in two-party systems that favor single-party government? Voting behavior in Italy in the 2000s is used to explore this question. This paper examines what motivates the voters of the large center-left and center-right coalitions, specifically whether ideology, economic issues, or other considerations lead voters to select their party of choice. Results indicate that, on average, voters select a coalition ideologically-proximal and deemed the more competent on issues, while they select a specific party based upon character and reputation issues. Findings thus suggest that voters sort for both coalition and party-specific reasons.


Introduction
The decision criterion that voters use to select their preferred party has been a fundamental question challenging political scientists for over a half-century. Various explanations have each been able to explain some part of this phenomenon: voters, it has been suggested, vote for a party based on socio-cultural reasons, because they feel particularly close to a party, to attempt to bring the policy in a specific direction, or because they deem it the most competent at addressing the issues of the day.
In proportional multiparty government systems, however, current theories do not have the explanatory power of smaller party systems. Pessimistically, it has been thought that the complexities involved with post-election government formation are too difficult for voters to accurately evaluate; as such, voting in proportional systems is considered fundamentally different (Downs, 1957). Pre-electoral coalitions, however, provide parties the opportunity to signal to voters how the government formation process will proceed. Even though of 292 elections in 20 countries, 44% included such arrangements, how voters behave in such contexts, however, has not yet been extensively researched (Golder, 2005;Gschwend et al., 2017;Plescia, 2017b). Only recently has research questioned if 'Coalitions [are] just a sum of their parties'? (Plescia and Aichholzer, 2017: 255).
This study suggests that within coalition contexts, voters can use the same heuristics of ideology, character, and issue-voting used in simpler systems. This is tested in the three Italian elections in the 2000s using a sequential logistic regression model. This context best typifies a coalition situation analogous to the two-party voting literature from which many theories are derived. During these years, Italian voters faced a clear choice between a coalition representing the left and a coalition representing the right. Within the coalitions, parties stressed different issues, advocated distinct ideologies, and had their own leaders. Findings suggest that indeed voters use valence and character-based assessments to sort within coalitions while being driven by ideology and issue competencies when choosing between them.
This paper begins with a review of theories of voting behavior. Next, findings from voting literature are integrated into the context of coalitions, and a series of potential heuristics are developed. The Italian case and sequential logistic regression are then introduced along with a description of the dependent and independent variables. Findings from the models are presented followed by a discussion that situates these findings in the context of previous theories.
Determinants of voting behavior in multi-party systems Downs (1957) provides the canonical explanation for votingideology. In short, voters perceive the positions of parties on a left-right continuum dominated by economic issues (Green-Pedersen, 2007). A party is then selected based upon 'proximal' voting to that individual's position in policy space, whereby the voter selects the closest party (van der Brug, 2004), or 'directional' voting, whereby the voter selects the party in her preferred left or right direction (Rabinowitz and Macdonald, 1989). An early critique of this model, as advocated by Stokes (1963), argues that voters are motivated by non-ideological concerns as well: party or candidate character valence and issue competence.

Valence considerations: party and leader characteristics
Valence concerns of desirable non-policy-related party and candidate characteristics include honesty and trustworthiness (Stokes, 1963). Politicians can employ valence-related tactics during campaigns to shift the focus away from ideological or positional dimensions of evaluation (Clark, 2009). Parties that appear more united and leaders that act with integrity have valence advantages. Cross-nationally, as parties and their programmatic offerings become more ideologically dispersed, these character-based attributes are found to have a greater impact (Clark and Leiter, 2014). Alternatively, when political supply is too limited via party policy convergence, voters turn toward valence issues to guide their selection (Green and Hobolt, 2008;Hellwig, 2012). Likewise when political demand, that is, the electorate, converges, as in Britain, parties are limited in their available positional strategies, and voters then rely on these non-policy features to direct their vote . Mauerer et al. (2015) consider party and leader evaluations a 'hard test' in exploring other alternative explanations of vote choice. Adams et al. (2005) argue that party and leader favorability is theoretically the most important non-policy-based character-valence concerns in their theory of party competition. Experimental findings support this contention by arguing that political information and opinions are mediated through one's preferred partisan 'perceptual screen' (Walgrave et al., 2014). Leader evaluations have empirically been found to mediate partisan attachments across Europe (Garzia, 2013).
The Italian context has provided evidence that supports this perspective. While previously dominated by political sub-cultures and a low-threshold proportional system, Italian voters, it is argued, are now guided by their assessment of the performance of incumbents. 'Reasoning voters' assess the competence of party leadership and select the coalition best able to deliver favorable policy outcomes, not necessarily related to policy directions (Bellucci, 2007). These effects have been found within elections of both proportional and majoritarian character (Bellucci, 2007(Bellucci, , 2012, as hitherto Catholic and working-class voters now divide themselves between the left and right.

Issue ownership
Another non-ideological facet of party competence concerns how well parties can 'handle' the important issues of the day. In their theory of party competition, Budge and Farlie (1983) argue that parties selectively emphasize issues, specifically those for which they believe they can enjoy a reputation of issue-handling competence (Walgrave et al., 2014). Thus, voters can be modeled as taking into consideration the reputation of a party in its ability to sincerely resolve a problem of concern and deliver on salient issue dimensions (Green and Hobolt, 2008). While indeed voters may have ideologically preferred solutions to problems, they are also pragmatic and instrumental in seeking out parties to fix these problems (Petrocik, 1996;van der Brug, 2004). An example of this direct effect of 'owning' the agenda occurs when a voter is concerned with crime levels and thereafter finds attractive the party perceived to be credibly toughest on crime (van der Brug, 2004). Clarke et al. (2004) and Bélanger (2003) have found that issue ownership forms the foundation of contemporary British and Canadian politics, respectively. Bellucci et al. (2015) find that issue considerations outweigh leadership character effects in the Italian case. Retrospective evaluations of government performance and beliefs that national economic conditions have improved also drive vote choice (Bellucci, 2007). While left-right positions may remain stable, the issue agenda of electoral campaigns can alter so as to activate this issue-based voting mechanism (van der Brug, 1999).
Party-voter issue linkages and ideological shifts affect party behavior and popularity based upon a party's type (Meguid, 2005;Ezrow et al., 2011;Adams et al., 2012aAdams et al., , 2012bAbney et al., 2013). Mainstream conservative, liberal, and labor/social democratic party families are primarily concerned with the economic dimension of competition and have broad catch-all issue agendas. In turn, niche parties, such as nationalist, green, and regional parties, with narrow agendas focused on salient non-economic dimensions, are electorally rewarded when they follow the left-right shifts of their core partisans (Adams et al., 2006), when they maintain extreme positions (Ezrow, 2008), and when the party's issues have national salience and a proximal mainstream competitor accommodates the issue (Meguid, 2005(Meguid, , 2008. In terms of issue ownership, niche parties succeed when they have issue ownership on its side of the political spectrum (Spoon et al., 2014;Han 2015;Abou-Chadi, 2016).
These explanations of issue ownership and niche success assume independently acting niche and mainstream parties engaged in political competition. While they do account for diverse strategies employed by mainstream and niche adversaries, they do not take into account the common occurrence of coalition governance in democratic politics. These two literatures have developed independently, and the case presented here seeks to bridge this gap, as distinct ideological positions become more difficult in coalitional contexts.

Voting in context of coalitions
Voters have been shown to take government coalitions into account when they vote in proportional systems with pre-electoral coalitions (Gschwend and Hooghe, 2008). When voters believe potential coalition members to be too ideologically incompatible or hold a coalition member in low-esteem, they might alter their vote away from a preferred party (Gschwend et al., 2017). On the contrary, a strong candidate or leader preference of a potential coalition member can lead even those initially not amenable to the coalition parties to support it (Gschwend and Hooghe, 2008). Conversely, individual party or leader preferences can harm one's assessment of a coalition beyond an averaging of constituent members, with coalition leaders perhaps having an outsized influence on vote choice (Plescia and Aichholzer, 2017). In this manner, coalition preferences have an independent effect on vote choice above their views of parties (Blais et al., 2006). Vote-choice functions can be thought of including both party and coalition considerations (Gschwend et al., 2017), as a voter's ideological position has been found to be predictive of their preferred coalition (Debus and Muller, 2014), not necessarily their party.
In Israel's 2006 elections, expectations of future government coalition weakened the propensity of selecting an ideologically preferred party unlikely to participate in parliamentary government, even though that party would likely be represented in parliament (Bargsted and Kedar, 2009). On the other hand, 'coalition-directed voting' induces voters to select non-proximal parties in an attempt to make the expected governing coalition more ideologically proximal than a sincere party vote would predict (Duch et al., 2010). This substantiates earlier studies of the Israeli 2003 election (Blais et al., 2006) that found voter's views on coalition potential had a decisive independent effect on up to 10% of voters beyond their own views about parties, leaders, ideology, or level of political information. Similarly, some voters feel incentives to vote strategically in Germany to counteract electoral rules that restrict representation to those parties that meet a 5% electoral threshold or are non-competitive in single-member districts (Gschwend, 2007) and choose parties other than their most preferred that would potentially join the governing coalition. Individuals could even vote against or for a coalition based on the inclusion or exclusion or a single party (Plescia and Aichholzer, 2017).
While it is uncontroversial to suggest that future-oriented voters have coalition preferences in mind while voting 1 (Hobolt and Karp, 2010), little is known about how voters then select within that potential governing coalition. Understanding this intra-coalition voting serves as the motivation for subsequent analysis. In other words, what drives voters to sort between coalition members, as opposed to sorting in or out of that coalition?
Three potential heuristics When theories on positional, valence, issue ownership, and coalition voting are integrated, a variety of outcomes could result. Previous studies of Italian voting patterns have either focused on coalitions (Bellucci, 2012) or individual parties (Barisione, 2014;Plescia, 2017a). As this is one of the first attempts at such a joint analysis, instead of presenting a single directional hypothesis, a variety of potential outcomes are hereafter presented that will then be tested against post-electoral survey data.
Previous studies of issue-based voting have occurred in systems with single-member districts (Bélanger and Meguid, 2008;Green and Hobolt, 2008). Arndt (2014) and van der Brug (2004) test issue ownership on Denmark and the Netherlands, respectively, but focus on parties as independent entities. Likewise, Kedar's (2005) multi-country study concludes that voters take into account a watering down of policies during government formation in proportional systems; however, it too overlooks specific coalition considerations. No assumptions are here made as to whether voters differentiate between coalition members primarily on the basis of ideology, character valence, administrative responsibility, or issue differentiation, as each can have an impact on a coalition-directed vote (Duch et al., 2010). Instead potential voting heuristics are here presented in ascending order of the primacy of the coalition-centered component over the party-centered component of vote choice.

Coalition-ignorance heuristic
Voters ignore the coalition and vote based on specific party ideology, personalities, issue priorities, or historical linkages. Parties have specific characteristics that appeal to voters. The most extreme parties will attract the most extreme votes while centrist parties will attract centrist voters. If such a model of voting dominates, one would not expect that coalition-based factors like government competence to be a significant predictor of vote choice. While a coalition of parties provides guidance to voters about potential majorities (Golder, 2005), such a claim assumes that 1 It is also uncontroversial to suggest a difficulty in assigning blame when holding coalitions electorally accountable. Empirically, the coalition leader faces the greatest blame (Duch et al., 2015;Fisher and Hobolt, 2010). voters are aware of these signals. Instead of the complexity associated with understanding how different coalitions might alter policy or issue distances, voters could heavily rely on the likability of parties or candidates to make their decision (Gschwend and Hooghe, 2008). In this manner, voters may simply follow a preferred party, ignorant of what coalition it may be potentially a part of (Plescia, 2017b).
H CI : Voters use ideology, issue ownership, and valence not to sort between coalitions, but only for specific parties within them.

Party-within-coalition heuristic
Voters select a party knowing they cannot ignore the eventual post-election coalition. As such, voters will be drawn toward the coalition for either ideological or issue-based concerns and then utilize the criteria to decide within that coalition. If these criteria are the same, then a specific party can be viewed as the epitome of that particular aspect of voting, such as the most ideologically extreme on one side of the political spectrum. Voting criteria could also be different for the coalition and party, whereby voters vote for a coalition ideologically and then select a party based on issue-based concerns, or vice versa. Duch et al. (2010) estimate that more than half of voters condition their vote for a party with a coalition-directed calculation in mind. While some notion of coalition-directed voting is present in most crossnational and Italian studies noted above, there is the possibility that when tested with alternative measures and methods that these findings disappear or that coalition-directed voting is solely ideologically driven (Duch et al., 2010). This conforms with the notion that competitors viewed as similar on non-policy groundsthat is, responsible for policy as members of the same coalitionhave an incentive to differentiate themselves in policy space so as to not split the vote of the same group of supporters (Adams, 1999). Evidence has also suggested that Italian voters differentiate between parties in a coalition based on their leadership, size, presence in parliament, and government portfolios associated with it (Plescia, 2017a). If both mainstream and niche-issue parties are present within a coalition, then voters can vote sincerely on their preferred dimension within the coalition without wasting a vote and reward a party who shares their salient issues. One can thus expect higher issue effects to distinguish between more 'polarizing parties' (Mauerer et al., 2015). If such a model of voting dominates, one would expect that coalition or government competence to be a significant predictor of coalition choice and additional effects of ideology, issue competence, and valence for parties within the coalition HP WC : Voters use ideology, issue ownership, and valence to sort between coalitions and within them.

Coalition-as-party heuristic
Voters explicitly select a governing coalition, not a single party. As such, voters that better conform with a coalition ideologically or based upon the issues it owns will be more likely to select that coalition. Due to the collective government responsibility ascribed to the members of a coalition, 'then, substantively, the voters view these parties as being interchangeable, in the sense that the coalition members no longer represent distinct alternatives to the voters' (Adams et al., 2013: 14). From a psychological perspective, this would suggest that the coalition itself is a meaningful political object beyond the constituent parties (Plescia and Aichholzer, 2017).
Joining a coalition signals a joint governing agenda and accountability (Bellucci, 2012), which ascribes ideological character and valence to the coalition as a whole. Experimental evidence suggests that coalition signals, such as the announcement of a pre-electoral coalition, prime coalition considerations at the expense of party considerations (Gschwend et al., 2017). Some have even suggested that center-left voters in Italy are developing a 'coalition identity' (D'Alimonte and Bartolini, 1998). If such a model of voting dominates, one would expect that coalition or government competence to be a significant predictor of coalition choice and additional effects of ideology, issue competence, and valence to be inconsequential within the coalition.
H CAP : Voters use ideology, issue ownership, and valence to sort between coalitions, but not to sort between individual parties within them.
The Italian case Italy in the 2000s provides an excellent case to examine these voting heuristics. As opposed to most multiparty systems that rely on post-election bargaining for parliamentary majorities, the Italian electoral system has been structured to incentivize pre-electoral coalitions. While voters have been shown to vote with potential post-election coalitions in mind, pre-electoral coalitions provide a more direct test of the heuristics this study seeks to explore. Post-election coalitions are just theoretical constructs before an election while pre-electoral coalitions might manifestly be in the minds, televisions, and newspapers of voters as they may even possess common manifestos (Gschwend et al., 2017).
The presence of pre-electoral coalitions in the Italian case makes both party and coalition considerations clear. Voters need not hypothesize which government may form. With pre-electoral coalitions, voters can select any party in a coalition, even if it is not their most preferred party, knowing that their preferred party will be represented in government. Pre-electoral coalitions reduce the uncertainty of government formation inherent in post-election coalition bargaining. As such, the Italian case is crucial should there be any evidence of non-coalitional directed voting. If it can be observed in the Italian case, then it should also be evident in cases with laxer rules on coalition formation.
The Italian Second Republicthat is, the party system after 1994exemplifies an alternational style of governance with a clearly defined cabinet of either the left or right, which stands in contrast to the pivotal centrist nature of Democrazia Cristiana (Christian Democratic) hegemony of the post-war Italian First Republic. Taking advantage of this type of system is ideal when testing joint ideological, issue, and valence effects on voting. Just as previous studies utilized the clarity of choice in single-member districts to discern between ownership and ideological voting, so too can the Italian case provide clarity between two government coalitions in a multiparty context.
While seven elections have occurred in Italy since 1994, only three exemplify the clarity of a left-right choice for subsequent governance. The 1994 election included a centrist coalition of the Partito Popolare Italiano (Italian People's Party) and Patto Segni (the Segni Pact) which garnered 15% of the proportional vote. The 1996 election saw center-right and center-left coalitions splinter as the Lega Nord (Northern League) ran alone obtaining 10% of the vote and Partito della Rifondazione Comunista (the Communist Refoundation Party) receiving 8. 5%. In 20015%. In , 20065%. In , and 2008, the main center-left and center-right coalitions obtained over 85% of the vote. This clear left-right distinction did not last as the 2013 election included an independent center coalition -Con Monti per L'Italia (With Monti For Italy)that won over 10% of the voteand a movement beyond left and right -Movimento 5 Stelle (Five-Star Movement)that won over 25% of the vote. In 2018, the Movimento 5 Stelle became the largest party with 32% of the vote. Thus, in the elections of 2001, 2006, and 2008, not only were there only two significant preelectoral coalitions, but also voters concentrated their votes on these blocs, with a joint popularity of at least 85%. In light of previous research on the role of government accountability and coalition choice in Italy (Bellucci, 2007) and limited cases of 'quasi bipolar format of competition' (Bellucci, 2012: 492), these three elections will be the subject of this study: Table 1 presents the popularity and left-right position of constituent parties as rated by the Chapel Hill Expert Survey experts (Bakker et al., 2012) and their voting performance from official government sources; the left coalition is in the upper half and the right coalition in the lower half.
Within those three elections, Italy altered its electoral system. From 1994 to 2001, Italy utilized a mixed-member system (Giannetti and De Giorgi, 2006). Three-fourths of both chambers were elected in single-member districts, and the remaining quarter was filled on a second proportional ballot with a 10% coalition threshold. Pre-electoral coalitions were also incentivized in the singlemember districts, whereby the coalition would choose a candidate from one of its constituent parties to run under the banner of the whole coalition: in 2001, Casa delle Libertà (House of Freedoms) was the coalition of the right, while L'Ulivo (the Olive Tree) was the coalition of the left. Cooperation is made quite evident by the fact that while the Lega Nord failed to pass the national 4% threshold for proportional seats, the party earned 30 single-member district seats as fronting a district candidate for the Casa delle Libertà coalition.
A 2005 reform altered the system to a proportional-bonus system. In the Chamber of Deputies, the lower house, the coalition with a plurality of nationwide votes, automatically received 55% of the seats to be distributed proportionally within its coalition. Parties within a coalition face a reduced 2% threshold, providing an even stronger incentive for smaller niche parties to link with, or remain linked with a larger mainstream collective. The persistence of coalitions between 2001 and 2006, even though the elimination of single-member districts, demonstrates this coalition incentive remained. Under these rules, a small party could be greatly over-represented by joining a coalition and gaining the majority bonus. In fact, the 2006 election saw all parties join either the center-left L'Unione (The Union) coalition or the center-right Casa delle Libertà coalition. If consistent evidence is found between these elections, then one can more strongly attest the findings to the nature of voting behavior, as opposed to the effect of electoral systems.
The issues of contention play a major role in the heuristics presented above. Italy in the 2000s experienced a remarkable consistency of the primary issues of political contention. In 2008, general economic issues and social policy reform along with the immigration of foreign workers were the primary issues concentrated upon by the national parties (OSCE, 2008;Wilson, 2009). Two years prior, the joint coalition manifestos in 2006 emphasized immigration and fiscal, labor, and economic reform (Giannetti and De Giorgi, 2006;OSCE, 2006), marking a little departure from the importance of fiscal and immigration issues in the 2001 election (Benoit and Laver, 2006;Giannetti and De Giorgi, 2006). While ideology and competence on these issues have been found to influence crossblock volatility (Bellucci, 2012), their effects on intra-block have not been investigated.

Data and method
The Italian left of the second republic has been highly fractionalized. The political supply of parties catering to left voters is quite saturated, and it was not until 2008 with the creation of Partito Democratico (the Democratic Party) that center-left voters had a clearly identifiable, large mainstream party to support. In the 2006 election, for example, four parties gained representation in the right coalition while eight did so in the left coalition. The largest of these were the Democratici di Sinistra (Democrats of the Left), with origins in the First Republic's Partito Comunista Italiano (Italian Communist Party), and La Margherita (the Daisy), with historical origins coming from more left-leaning elements of the First Republic's Democrazia Cristiana and Partito Socialista Italiano (Italian Socialist Party). At times, the Federazione dei Verdi (Federation of the Greens) appearing as part of the Il Girasole (Sunflower list) and the anti-corruption Italia dei Valori (Italy of Values) have also aligned with the center-left coalition.
The center-right of the 2000s, on the other hand, has had a slightly more stable membership of parties, leaders, core supporters, and ideals. Silvio Berlusconi led the coalition though the mainstream conservative Forza Italia and Popolo della Libertà (People of Freedom) parties. Gianfranco Fini led the former neofascist extremists in the Alleanza Nationale (National Alliance). Pier Ferdinando Casini led the more conservative elements stemming from the First Republic's Democrazia Cristiana through his Unione dei Democratici Cristiani e Democratici di Centro (Christian Democrat Centrists) that prioritized Christian values and centrism. Umberto Bossi led the Lega Nord with its milieu of niche concerns, including Northern regional autonomy, with nationalist views on multiculturalism, immigration, and the European Union.
Method: sequential logistic regression Given the electoral system, a sequential logistic regression model will be utilized. In the Italian case, unlike other multiparty proportional systems where a vote for a party does not directly translate into a vote for prime minister, coalitions state beforehand who their premier candidate is (Giannetti and De Giorgi, 2006). One cannot vote for the Alleanza Nationale without also voting for Berlusconi as prime minister. Crucially, ballot structure reinforces this aspect of the system presenting allied lists in proximal locations with coalitions clearly distinguished from others. Previous research has found that Italian voters do base their voting decisions on coalition or leader, more so than effects associated with individual parties (Bartolini et al., 2004).
Sequential logistic regressionsequential logittakes this into account as voters select a party conditional upon its coalition and premier candidate. Sequential logistic regression models assume that one can be 'at risk' of passing a 'transition' only if one has passed, or 'survived,' previous transitions (Fullerton, 2009;Buis 2010). To rephrase the above example, a voter can only be 'at risk' of voting for the Alleanza Nationale once they have passed the 'transition' of selecting a party of the Berlusconi-premier coalition over other coalitions or independent parties. Similar statistical results can be attained by running separate logistic regressions on each choice set on the appropriate sub-sample (Mare, 1981); however, the sequential logistic model has the added benefit that it can estimate varying effects of variables across multiple stages (Albert and Chib, 2001;Fullerton, 2009).
The SEQLOGIT package in STATA (Buis, 2013) allows for such decomposition and recombination of a variable's effect on categorical dependent variables following (Buis, 2017) suggested coding schema. 2 A discussion of marginal effects associated with each variable in each step and in total can provide a more insightful discussion than just analyzing a vote for a single party or coalition alone. Sequential logistic regression can allow for an analysis of both the effect When measuring the effects of a variable on a specific party choice, that party is assigned the value 1, while all other outcomes are assigned a value of 0. of a variable on each transition, inter-coalition and intra-coalition, and the effect on the final outcome: party choice (Buis, 2017).

Data and measurement
The dependent variable is party vote choice for the Chamber of Deputies as reported in ITANES post-election surveys. 3 For the 2001 election, the proportional ballot was used as only coalitions, not parties, were options for the constituency vote. The sequential logit model first estimates the effect of explanatory variables on the choice of the pre-electoral coalition, and then which party within that coalition. For example, in 2008, the center-left coalition consisted of the Partito Democratico and Italia dei Valori, both coded as 1, while the center-right coalition consisted of the PdL, coded as 2, and Lega Nord, coded as 3. Sequential logistic regression first calculates logit coefficients comparing 1 vs. 2/3, then directly compares 2 vs. 3. In the presentation of results, the inter-coalition coefficients, listed first, have the opposing coalition as the base, with independent variables of ideology and issue ownership coded so as to produce positive values that indicate a greater likelihood to vote for Berlusconi's right coalitions. Intra-coalition logistic coefficients 4 follow with Berlusconi's party's (FI or PdL) as the base, whereby positive values indicate a greater likelihood to vote for a non-Berlusconi party. This process is then switched with parties of the center-right serving as the base and individual coding center-left coalition parties as compared to the coalition leader's party. In 2006, the ITANES survey grouped center-left voters into one of three groups: Communist-Green, Olive Tree, or Other. As such, this case is excluded from the analysis as individual party voters are not obtainable.
Left-right ideology self-placement scores (LR) were asked of voters on a 0-10 scale with 10 being the most right. In the 2001, 2006, and 2008 ITANES surveys, respondents were asked which coalition 5 would best be able to manage specific problems facing Italy. This ownership variable is standard in election research (Bélanger and Meguid, 2008;Green and Hobolt, 2008;Arndt, 2014) and has been used in previous studies of Italian voting behavior (Bellucci, 2007(Bellucci, , 2012. In all years of interest, immigration issue ownership (IOimm)a typical niche issueand economy issue ownership (IOecon) 6the stereotypical mainstream issuewere asked. As mentioned earlier, these issues dominated the political debate surrounding the election. Following Arndt's (2014) coding schema, those that believe the center-right best able to handle the issue were coded with 1, those responding that the center-left was best able to handle the issue were coded with −1, and those who believed there was no difference or no solution were coded with 0. Party preference 7 and leadership evaluations 8 have been found to have an acute importance in the Italian context (Garzia and Viotti, 2011)   Ideally, this would be asked at the party-level; however, the ITANES data lack this variable. That said, if this variable is found to be significant at the party-choice level, this would indicate that voters weighed these concerns differently in their party selection than just ascribing this trait equally to all coalition members. Other leaders were not asked to all respondents and so have extensive missing data. The inclusion of these variables thus isolates direct effect of an individual's ideology and belief of ownership variables (Arndt, 2014) from the indirect effect of parties being associated with promoting the issues and interests of their constituents (Petrocik, 1996;Walgrave et al., 2014) and thus the causal arrow being reversed (Evans and Andersen, 2004).
Standard control variables in Italian electoral studies are also included along with errors clustered by constituency in 2001 and region in 2006 and 2008. Given the anti-clerical nature of some leftist parties as well as the presence of specifically religiously influenced parties on the right, religiosity is included as a 1-4 scale of how important religion is to one's life, ranging from 'not important at all' to 'very important.' Previous findings on Italian voting behavior (Corbetta and Cavazza, 2008) and Berlusconi's electoral appeal (Barisione, 2014) have found that political (dis)interest has had a role in his success along with traditional sociological controls of gender, labor market participation 9 via unemployment, age, and education level. A variable has also been generated (econworse) for whether the respondent believes that the economic situation in Italy has worsened, as studies have found that Italian voters hold incumbent coalitions accountable based on subjective beliefs (Bellucci, 2012;Plescia, 2017a).

Coalition and party choice
The full sequential logit models and summary statistics for each election year are presented in the appendix. For ease of interpretation, these results have been presented visually on Figures 1-3. The upper-most coefficients and 95% confidence intervals in each indicate the variables that affected coalition choice. The subsequent rows indicate choices between the major party and junior members. Significant coefficients on the coalition choice rows indicate that factor led a voter to select that coalition. Significant coefficients on the subsequent rows would indicate that voters are discriminating between parties intra-coalition using that variable. Positive values indicate a greater likelihood for the second of the parties/coalitions listed.
Results for the 2001 election are presented in Figure 1. Looking at voting for the center-right coalition, one sees that being more conservative, believing the center-right coalition is best able to address the issues of the economy and immigration, and believing that the economy is worsening (the previous government was of the center-left) are suggestive of a vote for the center-right coalition; on the figure, each of these estimates has a 95% confidence interval that is greater than 0. Similarly, having favorability toward Forza Italia and Berlusconi, the major party leader, is also indicative of a vote for this coalition. Favoring some junior members of the coalition also led individuals to select this coalition over others. This evidence suggests that Italian voters of the center-right are not ignorant of the coalition that the party they vote for is a part of.
Evidence of factors motivating a center-left vote in 2001 also includes ideology and the belief that the left is best at managing the economy (recall that IOecon is coded positive if the voter believes that the center-right is best able to address the issue and negative if the center-left is best able to address the issue). A voter favoring the Democratici di Sinistra and Rutelli, the left's candidate for prime minister, also was indicative of a vote for this coalition. Favoring 9 Housewives, as suggested by Barisione (2014), are a crucial component to the Berlusconi coalition, and thus an interaction between gender and labor market participation is used to capture this effect.  D'Alema, the leader of the largest party, also led one to more likely select the left coalition. Here too provides evidence that voters for parties within a coalition are not systematically ignorant to its full membership.
While the evidence is presented that voters are cognizant of a full coalition, if variables are significant between the coalition partners, this would indicate that voters use these heuristics to decide between coalition members. On the center-right in 2001, voters for the Lega Nord were distinguished from Forza Italia by their greater conservatism, favorability toward their own party and the party leader (Bossi), and displeasure with Forza Italia as a party. A similar pattern can be found with Alleanza Nationale voters. Voters for the UDC, though not ideologically distinct from Forza Italia voters, had lower favorability toward its coalition partners in the Lega Nord, the Forza Italia organization, and its leader in Berlusconi. A voter's belief in the ability of a coalition to address a pressing issue for the nation does not have an impact on their vote choice.
A similar pattern can be found among the four largest parties of the center-left coalition in 2001. Two of the parties are distinguished ideologically with La Margherita attracting more conservative members than Democratici di Sinistra and the Partito dei Comunisti Italiani (Party of Italian Communists) attracting voters more to the left. While individual party leaders discerned voters of the center-right coalition, this had no effect on the center-left. That said, voters for each of the junior coalition partners had a lower party favorability of the dominant party. It can also be noted that more religious individuals also supported La Margherita, which would be expected given that its progenitor parties included former Democrazia Cristiana. Verdi voters were uniquely motivated by the belief that the center-left coalition would be best for the economy, the only set of voters in this election with evidence that a specific issue might have led to one party's choice over another.
Evidence from the 2006 election confirms many of the findings of the 2001 election. Due to data availability mentioned above, only party-vote data for the center-right can be analyzed. Coalition choice is again dominated by ideology, favorability of the lead party (Forza Italia), and belief that the center-right would be best able to manage the economy. Differentiation between the parties within the coalition is distinguished by voters of smaller parties having a less favorable view of Berlusconi and Forza Italia and a favorable view of their own party leaders. One strong difference does stand out here in this election in that the belief that the center-right was best capable of addressing the immigration issue strongly motivated center-right voters to select the Lega Nord as their party of choice. While the electoral system changed between 2001 and 2006, the mechanisms by which voters selected their party seemed to not have a great difference.
So far, the results have suggested that Italian voters follow a model closest to the 'party-within-coalition' heuristic. They have selected coalitions based on ideology and governing competence and selected within them based on the individual leader and party favorability. Ideology and issue priorities also distinguish smaller party voters. The 2008 election saw a consolidation of parties, though if the evidence is found that voters still are motivated to select coalitions for governing reasons and then sort within them for other reasons, then the 'party-within-coalition' heuristic would seem the most plausible.
Looking at the center-right coalition, this again seems to be the case. Figure 3 consolidates the voting onto a smaller display, still with the upper portion depicting coalition choice and lower figure indicating support for a junior coalition partner. The solid lines centered on a circular point still depict the center-left vote, while a dashed line centered on a triangle depicts the centerright vote. Voters who believed that the economy has worsened under the center-left incumbency selected the more conservative center-right coalition with the belief that this would be best to manage the economy. Favorability for the leading party again drove voters toward this party. Sorting within this coalition, we again see Lega Nord voters having a lower favorability of Berlusconi and his PdL party and more strongly favoring their own organization. Again, the belief that the center-right is best able to deal with immigration distinguishes Lega Nord supports from its coalition partner.
A similar pattern can be found on the center-left. Voting for these incumbents was not motivated by the belief that the economy is doing poorly, and support for the incumbents was bolstered by those who believed the center-left is best able to manage the economy (recall that negative coding means the belief that the left is better able to manage the issue). Center-left coalition voters were indeed more left ideologically. Voters did favor both the parties and party leaders more so than those who voted for the center-right: note how each of these variables is significant for the center-left (circle and solid line) but not for center-right (triangle and dash). However, the strength of these differences is what differentiated a vote within the coalition, as junior Italia Dei Valori members had a lower favorability of the Partito Democratico and a stronger opinion of their own and their leader: note how Veltroni and PD favorability are negative in the lower figure while coefficients for Valori and DiPietro are positive. Valori voters were also more likely to believe the economy had worsened than voters for Partito Democratico. 10

Discussion
This analysis has focused on the determinants of voting within coalitions in Italy through the first decade of the 2000s, specifically interested in whether voters, if at all, distinguished between parties for ideological, governance-based, or character-based valence explanations. Cross-national findings have found that Italians often have the greatest amount of proximity, or sincere, voting (Bargsted and Kedar, 2009;Hobolt and Karp, 2010) whereby voters select the party they evaluate most positively. This analysis suggests this effect is driven by party favorability measures, as opposed to ideological proximity. The Italian pre-electoral coalition system allows voters to both select the group of parties that are directionally closer to them and that they best trust 10 As shown in the Appendix, Valori voters were also more likely to be of working age and to have a higher education.
with the issues facing the country. Therein, voters can reward or punish specific party leaders and party organizations by voting for one of their allies.
Party and leader favorability and the messages sent to voters thus mediate the direct connection between ideology and vote choice (Adams et al., 2005). While favorability toward parties and leaders has been the subject of prior research, the estimated models here allowed for party disfavorability to have a particularly notable effect. Holding constant one's favorability for the party a voter ultimately selected, as a voter held a less favorable view of the Prime Minster's party, so too was she more inclined to select any alternative party. A similar effect can be found in the favorability of party leaders. The measurement of these variables did not preclude a respondent to score all parties high or low on this measure; that is, scoring one's party higher does not automatically reduce the score of other parties or leaders. In this manner, the original methodology of this research design demonstrates that while the 'hard test' suggestion of favorability of one's own party does have an impact on vote choice, so too is vote choice a function of the favorability of other parties in the system. Voters sorting within coalitions face similar character valence 'cross-pressures' (Gschwend and Hooghe, 2008) independent from ideology as those deciding between coalitions.
The role of ideology in sorting between parties is mixed. Ideology certainly played a role in the sorting between coalitions. When there was a large number of parties in a coalition, some voters turned to ideology to sort between allies. Yet results are not consistent. Thus, after including issue-based, party-based, and leader-based factors in the analysis, van der Brug's (2004) conclusion that ideological proximity is the most important determinant for voting behavior in multiparty systems may better apply to coalitions (Duch et al., 2010) than to specific parties.
With the exception of Verdi and Lega Nord supporters, Italian voters did not sort withincoalitions based on their opinion of which coalition could most competently handle pressing issues. In order to gain electorally from the issue-ownership perspective, parties differentiate themselves on the issues they emphasize and voters subsequently sort (Petrocik, 1996) with niche parties specifically needing to differentiate themselves from other mainstream competitors on key issue dimensions (Meguid, 2005;Schofield and Sened, 2005;. In the Italian case, voters for various parties of the right and left coalition did not vary in the ascription of their coalition's economic issue ownership in any systematic ways. While on the surface this seems reasonable, it does slightly challenge the notion of a niche-mainstream distinction. It has been argued that niche parties are less concerned with economic issues (Meguid, 2008;Meyer and Miller, 2013;Bischof, 2017); however, this is not the case with Italian voters. While theory suggests that extremist niche voters should downplay economic concerns and focus on the non-economic issues put forth by their parties, Italians and their parties appear to ascribe a 'collective responsibility'  for the economy to coalitions at large. That said, voters of the anti-immigration Lega Nord party were consistently more concerned about their coalitions ability to handle that issue than voters of other parties.
In toto, it would appear that the party-within-coalition heuristic finds the most support. There is scant evidence that voters have divergent beliefs on the ability of their party's coalition to best manage the economy, nor are voters unmotivated by the allied party and leader favorability influence decisions. These both suggest that the coalition-ignorance heuristic lacks validity. These same notions of favorability ratings along with ideology also allow voters to sort within coalitions. Evidence of sorting within coalitions suggests that voters do not view each coalition-as-party, but instead use measurable facets to differentiate within them. In the Italian context, voters use ideology, issue-based considerations, and character-based valence concerns to select both their preferred coalition and their party within it.

Conclusion
This analysis set out to explore voting mechanisms in the context of coalition voting in multiparty systems. Bridging literature on ideological, issue, niche, and coalition voting, heuristics were presented that suggest a variety of ways that issues and coalitions could interact in the minds of voters. Sequential logistic regression was then used to isolate variables that drove voters in three Italian elections toward one coalition or another, and then toward one party or another. While previous evidence suggests that coalition-directed voting is commonplace, it was not clear what role individual parties had in a voter's ultimate choice. The evidence here presented expands upon previous research focused on ideology (Duch et al., 2010) and found that voters are also divergently motivated by issue, party-level, and character-based concerns. In this manner, voters not only target coalitions when voting, but target parties-within-coalitions. As previous research suggested a greater reliance on non-ideological measures when the party's ideologies converge, perhaps voters believed that parties in coalition were ideologically more proximal and relied upon these other considerations. Future research should seek to isolate whether individual voters who believe parties to be more ideologically convergent after pre-electoral announcements are more likely to be influenced by character and competence-based opinions.
The Italian case was crucial for the development of such an argument through its frequency of pre-electoral coalitions. Forty-four percent of parliamentary elections in 20 democracies contain a pre-electoral coalition and about a quarter of governments formed after elections contain at least one pre-electoral coalition (Golder, 2005). As previous scholars have assumed that voters are aware of post-election bargaining and thus direct a coalition targeting vote, so too, must voters be aware that which party they select within a coalition matters for future governance and policy, especially in light of findings that they punish the parties of particular ministers (Plescia, 2017a). While the Italian case in the 2000s might be extreme in its use of pre-electoral coalitions, the analysis here suggests some additional variables for other scholars to include in their analysis of elections in parliamentary democracies more generally, including most significantly opinions of the likely largest party and prime minister.
The conditional logit 'work horse' (Alvarez and Nagler, 1998;Adams et al., 2005;Mauerer et al., 2015) does not provide for the explicit distinction of favorability of one party or leader and disfavorability of others. The inclusion of ratings for multiple parties and leaders here produced results that indicate voters to be both motivated by the 'hard test' variables of party and leader favorability as well as party and leader displeasure. Future studies should consider the trade-off of including only coalition favorability or leader favorability, as they might be unable to detect differences within coalitions and motivations of direct party voting. Additionally, studies of party voting often use a dichotomous coding for the party of interest, and thus the findings are compared to the mean voter. Using a categorical choice model, however, allowed for a more relevant question to be asked of what distinguishes niche conservative voters from mainstream conservative voters. While indeed ascribing greater ownership to immigration, Lega Nord voters are no more ideologically conservative nor do they ascribe economic stewardship to their chosen party less than mainstream conservative voters. The use of simple logistic models would be unable to discern such variables' effects on an inter-coalition vs. intra-coalition vote; thus, extreme ideologies are often ascribed to niche voters, as done by Passarelli (2013), when in fact this ideology belongs to the coalition at large.
Future studies of party voting in coalition contexts should take into account greater contextual and electoral system factors. As found in cases with electoral thresholds, are voters in larger or smaller districts more or less likely to consider different factors when making their coalition or party choice? While this paper focused on cases with the clearest bi-polar coalition choice, research on other Italian elections could certainly elucidate the models of voting that occur in more complex environments. Does ideology trump all else when the number of coalition choices becomes more complex? Answering such a question would bridge the literatures between Italian two-coalition voter sorting and Northern European multi-party sorting.
Funding. Part of this research was funded by a fellowship from the Italian Club of San Diego.