The effect of valence and ideology in campaign conversion: panel evidence from three Spanish general elections

This paper studies changes in voting preferences over election campaigns. Building on the literature on spatial models and valence issues, we study whether (1) ideological distance to political parties, (2) assessments of party competence to handle different policy issues, and (3) voter-updated candidate evaluations are factors that explain shifts in voter choices in the weeks preceding the election. To test our hypotheses, we use data from three survey panels conducted for the 2008, 2011 and 2015 Spanish general elections. Our findings show that valence factors are more influential than ideological indifference to account for campaign conversion.


Introduction
If we consider elections to be "critical democratic instruments" (Powell 2000, p. 2) and parties as mainly vote-seekers (Downs 1957), election campaigns become one of the most important phenomena in the democratic process.It is at this time when parties maximize their efforts to influence voter decision-making in order to win elections.Campaigns are relevant for three main reasons.First, some voters make their electoral choice in the campaign period (Finkel 1993;Finkel and Schrott 1995;Shaw 1999;Whiteley and Seyd 2003;Issenberg 2012).Second, parties provide more information about their platforms during the campaign, so this is the perfect time to shape voting preferences (Popkin 1991;Shaw 1999;Simon 2002;Clinton and Lapinski 2004;Franz and Rideout 2007;Greene 2011).Finally, campaigns are consequential even for candidates who do not expect to win the election, since to remain in politics they have to garner an acceptable vote share in the eyes of party elites (Gosnell 1950).
Early research on election campaigns mostly focused on their minimal effects.According to received wisdom, the main value of campaigns was to reinforce previous party choices (Lazarsfeld et al. 1944;Berelson et al. 1954).The only relevant persuasion effect of campaigns seemed to be to mobilize voters that otherwise would not turn out on Election Day (Gerber and Green 2000;Holbrook and McClurg 2005;Hillygus 2005;McClurg and Holbrook 2009).As a result, the impact of campaigns on changing prior vote choices drew scant academic attention.It could be argued, however, that this is no longer the case.Following the unfreezing of party systems, many established democracies have recorded high levels of electoral volatility (Mair 2005).Either as a consequence of a new party entry or vote switching, party vote shares have become less stable over elections (Dalton and Wattenberg 2000).Campaigns contribute to explain this pattern of electoral outcomes.
This article examines changes in reported vote choice during the election campaign.More specifically, we study the determinants of conversion as a campaign effect.Following the literature, we define conversion as the switch in party choice during the campaign period.Conversion is important for both parties and voters.For parties, because it means not only that one party increases its vote share, but also that its rivals diminish theirs.The relevance for voters is noteworthy in the case of the 2015 Spanish general election, in which the emergence of two new parties provided voters with more alternatives to change their party choice.According to our findings, valence factors such as heterogeneous assessments of party competence across policy issues and changes in candidate evaluations are more influential than voters' ideological indifference over parties to account for the instability of party choice in the campaign period.
The novelty of our paper is twofold.First, we examine the effect of ideology and valence in a dynamic setting (i.e., conversion during a campaign) rather than a static one (i.e., determinants of vote choice).Of course, studying the impact of valence issues and ideological position is far from being new.Previous studies have found valence effects in the context of the decline of ideological voting for Canada (Bélanger and Meguid 2008), the United Kingdom (Clarke et al. 2004) and the United States (Stone and Simas 2010).But if this pattern is clear for first-past-the-post systems where the Downsian model predicts an ideological convergence of parties, its application to multi-party contexts remains an empirical open question.Secondly, we contribute to the existing literature by framing the conversion effect of election campaigns on swing voters during a period of political transformation from a two-party system to a multi-party system.The literature on elections has mainly focused on the determinants of party choice without distinguishing between stable and unstable voters.Yet, in a context of increasing volatility it is interesting to understand whether incentives of citizens to switch their vote are the same as leading them to prefer one party in the first place.
To test our hypotheses we take advantage of panel survey data collected by the Center for Sociological Research (Centro de Investigaciones Sociológicas, CIS) for the three Spanish general elections held between 2008 and 2015.The three panels consist of a preand post-election survey with data about voter characteristics and attitudes on issues and towards candidates.Spain is an interesting case study because of its recent transformation from a stable de facto two-party system to a multi-party system.
The rest of the article proceeds as follows.Section 2 reviews the literature on campaign effects and develops our hypotheses to explain voting dynamics over the campaign.Section 3 presents the Spanish system.Section 4 deals with data and methods.Section 5 discusses the results and Sect.6 concludes.

The effect of election campaigns
To maximize its share of votes, a party needs not only to keep its core voters, but also to attract voters that either lean towards other parties (swing voters) or are less likely to participate in the election (potential non-voters).For this reason, election campaigns are a key instrument in the repertoire that parties employ to gain office.Even more importantly, the temporal proximity to the election turns campaigns into the ideal period to maximize party support among the electorate because the expected return to the mobilization effort is greater.Hence, parties have to heighten attention to their platforms and candidates to influence the election outcome and increase their policy and portfolios rewards because the effects of campaigning are "large enough" to be important (Campbell 2000, p. 188).
The literature has usually distinguished three main effects of campaigns.First, campaigns can lead to the reinforcement of prior vote intentions.Empirical research finds that this effect is generally prevalent (Lazarsfeld et al. 1944;Berelson et al. 1954;Campbell et al. 1960;Finkel 1993;Iyengar and Simon 2000;Martínez i Coma 2008).Finkel and Schrott (1995) call reinforced individuals "stable voters".In this vein, we could consider campaign reinforcement as a "non-effect" or as the "reference outcome", as reinforced voters report a post-election choice that was unchanged by the campaign.
Another effect of campaigns is activation (or mobilization).Campaigns can induce the participation of voters who otherwise would have abstained.There is plenty of research showing that campaigns increase turnout on Election Day (Gerber and Green 2000;Holbrook and McClurg 2005;Hillygus 2005;McClurg and Holbrook 2009).This effect takes place either by increasing the awareness about the stakes of the election or by convincing voters to make a choice consistent with their social background, party identification or personal ideology (Gelman and King 1993;Finkel and Schrott 1995;Jamieson 2001;Fournier 2006;Kam 2006).
Third, campaigns can also lead to voter conversion.The amount of accessible political information is greater in the campaign period than at any other time. 1 We expect some citizens to switch their vote intention over the course of the campaign, as they collect new information that changes their prior party preference.Furthermore, recent studies (Hillygus and Shield 2009;Willmann 2011;Gallego and Rodden 2016) have found that parties use controversial issues as a campaign strategy to lure cross-pressured voters (i.e., voters subject to contradictory political leanings) in elections in which policy preference and partisanship are in conflict.Conversion affects around a 10% of the electorate in established democracies (Lazarsfeld et al. 1944;Finkel and Schrott 1995;Martínez i Coma 2008;García-Viñuela 2013), although its size is much larger in countries in which partisan identities are weak and campaign spending is highly asymmetric (Greene 2011).
More recent research has explored a fourth effect of campaigns: demobilization.Intended or not as a campaign effect, some survey respondents report their willingness to vote in the pre-campaign period, but they end up abstaining on Election Day.This behaviour might be a result of several factors, such as alienation by the negative tone of campaign rhetoric or disappointment either at the personal qualities exhibited by the candidates or the policy proposals included in party platforms (Ansolabehere et al. 1994;Ansolabehere and Iyengar 1995;Martínez i Coma 2008).
In this paper we focus on the effects of campaigns on voter conversion.To contribute to the literature on elections we posit three hypotheses related to the spatial and the valence theories of voting behaviour.According to the former, party policy proposals and citizen preferences over them can be represented in a linear space.In this model, voters support the party/candidate whose policy proposals are closest to them in the ideological axis.While this approach has been particularly influential on voting research in the last 60 years, relatively little is known about the scenario in which the minimal distance to a voter is the same for two parties.Downs (1957) and Enelow and Hinich (1984) argue that individuals are more likely to abstain when the utility differential between two parties decreases (abstention due to indifference).Within this framework, Leighley and Nagler (2012) showed that turnout would rise in the United States if parties offered more distinct choices.Rodon (2016), however, contested their findings in a paper with a large sample of countries and concluded that the effect of party indifference is relatively small compared to other factors postulated in the turnout literature.
Notwithstanding these contributions, prior research has neglected the potential impact of ideological indifference on the likelihood of party switching over the election campaign.Recent studies show, however, that the rising volatility in parties' vote shares over elections held in advanced democracies in the last decades is associated with voters' indifference between parties (Dassonneville and Hooghe).Therefore, we argue that ideological indifference might be a significant factor of party switching during the campaign period.In the spatial literature, voters that are equally distant from two parties would either abstain or decide which party to vote for by flipping a coin.We do not suggest that indifferent voters decide by chance which party to vote for.We argue instead that if they reported a party preference in the pre-campaign period, they are less likely to stick to it on Election Day and more prone to revise it during the campaign period.Hence, our first hypothesis is: H1 Ideologically indifferent voters are more likely to change their party preference over the campaign.
We compare this hypothesis to the valence politics view.The valence perspective fundamentally departs from the Downsian spatial framework.According to Downs (1957), parties compete over policies on which they dissent.In contrast, Stokes (1963) argues that certain policies are characterized by general agreement.Within this valence frame, there are goals that are "positively or negatively valued by the electorate [as a whole]" (Stokes 1963, p. 373).So, parties do not compete by relocating themselves closer to the electorate since, by definition, everyone (parties and voters) agrees on the ends.This is why the logic of political competition is based on claiming credibility and competence in the achievement of the shared goals (De Sio and Weber 2014).Recently, valence models of voting behaviour have been found to outperform spatial accounts of party choice (Sanders et al. 2011).Here we study two ways in which valence concerns can play a role in campaign dynamics: the assessments of party competence and the evaluation of candidates by voters.
First, there is what we call heterogeneity in assessments of party competence, which we define as the situation in which one voter believes that no single party is the most competent in all policy areas.This is an important topic in the voting behaviour literature.For example, Converse (1964, p. 3) coins the term "belief system" to refer to "a configuration of ideas and attitudes in which the elements are bound together by some form of constraint or functional interdependence".Similarly, Zaller (1991, p. 113) argues that "An extensive research literature has documented that people who are liberal (or conservative) on one issue tend to be relatively liberal (or conservative) on a range of other issues.This tendency is most commonly explained by means of the concept of 'attitude constraint', which implies that one sort of attitude (ideological orientation) constraints other attitudes (policy preferences)".The existence of this "belief system" or "attitude constraint" explains why some voters support all the policies advocated by a single party.
Yet, there are voters whose political attitudes are not constrained in the way described by these authors.For instance, voters could support the policies of a party on moral issues (like abortion) and the policies of a different party on economic issues (like taxes).Within the framework of valence politics, we believe that heterogeneous assessments of party competence can explain voter deviations from their choice before the election campaign.In our view, such voters might be more permeable to campaign messages for three main reasons.
First, following Zaller (1992), while the behaviour of voters with homogenous perceptions of a party's ability across issues could be predicted from those predispositions, the behaviour of voters with heterogenous valence assessments could not.As a result, there is more room for change in voting behaviour over the election campaign for the latter.On the contrary, voters with homogeneous evaluations of what is the most competent party would not have conflicting views about which party to support depending on their valence assessments, leading consequently to higher levels of stability over time.
The new literature on campaigns (Carsey and Layman 2006;Killian and Wilcox 2008;Hillygus and Shields 2009;Willmann 2011;Gallego and Rodden 2016) has studied the choice of voters that face a dilemma between party allegiance and party policies over issues that are highly relevant to them.Such voters might not vote for the party they normally support if it offers policies contrary to their firmly held moral beliefs, like abortion, gay marriage or immigration.If such voters represent a significant segment of the electorate, party elites might design a campaign discourse tailored to exploit the inner conflict of those voters over issues of great expressive importance to them.So, the campaign might have a conversion effect by altering the competing incentives of cross-pressured partisans when they have to decide for which party to vote for.
Secondly, campaign information is more crucial for voters with heterogeneous valence.Rational ignorance theory argues that voters have no incentives to collect information about party platforms or a candidate's qualities.Yet, during the campaign, parties announce their policy proposals and emphasize some issues while understating others.This information might sway the vote intention reported by voters before a party's final manifestos were made public because voters can realize the implications of the policies included in the party platforms.Furthermore, by making some issues more salient, campaigns increase the weight that voters attach to specific issues (priming effect).For voters with non-uniform assessments of party competence, it is more likely that the party they thought the most able to handle the dominant issue in the weeks before the election is not the same that the party they planned to support before the campaign started.If this occurs, such voters will be more disposed to change their party choice on Election Day.
Finally, voters update their perception of a party's ability to handle various issues in the campaign period.We claim that the working of this updating process is not the same across different values of our measure of heterogeneity in competence assessments.For voters that have a uniform perception of party valence across issues, the most likely outcome of the updating process is reinforcement.These voters are probably the ones that exhibit higher levels of selective exposure to the media, making campaign change even more unlikely.By contrast, cross-pressured voters might be more receptive to campaign messages.If the information they receive favours a party different from their pre-campaign choice, an opportunity window for party switching opens.
In sum, we expect campaigns to create more volatility in the electoral choice of voters with heterogeneous assessments of party competence than in the decision of voters who think that a single party is the best equipped to handle all policy issues.We hypothesize that such voters will seek more information during the campaign because they are less sure about which party they should vote for on Election Day.Different types of information are made readily available in the campaign period: from issue saliency to party expertise.When this new information finds the fertile ground for change provided by a heterogeneous structure of party ability assessments, it is more likely to observe changes in the pre-campaign vote choice.In contrast, voters with uniform assessments of party competence tend to make up their mind well before the campaign starts and are more immune to the new information released in the campaign period.This is probably because they use other types of heuristics to decide which party to support.Based on these considerations we posit our second hypothesis.

H2
Voters with heterogeneous assessments of party competence are more likely to change their party preference over the campaign.
An additional valence property has to do with party candidates.Recently, two of the traditionally most relevant predictors of party choice such as social class and party identification seem to have lost explanatory power (Franklin et al. 1992;Curtice and Holmberg 2005).Contrarily, candidate valence appears to have gained importance in spite of the growing cynicism of voters towards politicians (Norris 1999;Pharr and Putnam 2000).According to Bosch and Rico (2003), the process of personalization of politics by which leaders have become increasingly influential in elections may be acknowledged in at least four different levels: institutions (Carey and Shugart 1995;McAllister 2007;Barberá 2010), political communication (Butler and Ranney 1992;Swanson and Mancini 1996;Mughan 2000), electorate (Dalton 1996), and political parties (Scarrow et al. 2000).Although such developments lead to the expectation that the quality of candidates shapes voter behaviour, the importance of the process of personalization of politics in a campaign context has been seldom tested.For instance, Costa Lobo and Curtice (2014) claim that there is strong support for the view that leaders matter for voting behavior, but they do not pay attention to possible changes over the campaign period.
From these studies, we take voter appraisal of candidates as proxies for the qualification of party leaders to the Prime Minister job.Campaign advertising can influence the credibility and competence of party leaders as perceived by the electorate (Bean and Mughan 1989;Stewart and Clarke 1992;Lodge et al. 1995;Alvarez and Shankster 2006;Rico 2009).Accordingly, we expect voters who revise their opinion of party candidates to the Prime Minister office over the campaign period to be more likely to move away from their initial vote preference.Thus, our third hypothesis is as follows: H3 Voters who update their evaluation of party candidates in the campaign are more likely to change their pre-campaign party choice.

The Spanish system
To test our hypotheses, we use survey data for the three general elections held in Spain between 2008 and 2015.The Spanish case is particularly relevant for the purpose of this study because it is a closed-list proportional representation system.In addition, election campaigns in Spain are extremely nationalized.Party platforms have several dominating issues that follow a unique national campaign agenda, while district-level issues are very secondary (Martínez i Coma 2008).2Both features provide an appropriate setting to test our hypotheses on campaign conversion.
The three general elections we study are the most recent ones for which panel data are available.We concentrate on a period of Spanish politics, 2008 to 2015, when the Spanish party system was radically transformed.Hence, our analysis covers different scenarios.In 2008, the election took place after a period characterized by steady economic growth and a strong confrontation between the government and the main opposition party, the conservative Partido Popular (PP).The national economy was on the brink of a major crisis, but its effects were still not visible and the economy played a minor role in the election campaign.This was also the moment when the two parties that have dominated Spanish politics since the late 1980s, PP and PSOE, were at their peak of electoral power.The highly competitive 2008 election, returned the incumbent Socialist Party (PSOE) to office with almost 43% of the vote and a plurality of seats.As the two main parties garnered 80% of the vote, Spain could be considered a de facto two-party system.
The 2011 election took place in the course of a severe economic crisis.The discredit of the PSOE for its mismanagement of the economy led to a landslide victory for the opposition PP, but the two dominant parties still gained 73% of the national vote.However, a major change occurred in 2015.After 7 years of a profound economic and political crisis, the Spanish party system exploded (Orriols and Cordero 2016).The 2015 election was characterized by the surprising performance of two new parties, the radical left Podemos and the center-right Ciudadanos, which won 65 seats and 40 seats, respectively, in the 350 seats lower chamber of parliament (Congreso de los Diputados).Although PP and PSOE remained as the two most voted parties, the former lost more than a third of its electoral support and the latter obtained its worst result since the restoration of democracy in the late 1970s.So, in the 8 years covered by our study, Spain transited from a de facto two partysystem to a true multi-party system.The fragmentation of the party system originated by the electoral success of the new parties forced the repetition of the general election in 2016.
The fact that the political context differed so markedly in the three elections contributes to the potential generalizability of our findings.In our analysis of the first two elections, we concentrate on PP and PSOE because they were the major actors in those campaigns and the only ones that could win the election.Media attention during the weeks before each election was strongly focused on them and, consequently, these were the parties more likely to be benefited by campaign effects.Yet, the breakdown of the party system in 2015 and the emergence of Podemos and Ciudadanos changed unexpectedly the electoral landscape.Whereas Podemos became an appealing option for disappointed PSOE voters, Ciudadanos attracted conservative voters unhappy with the performance of PP governments during the crisis.As both emergent parties increased the opportunities for conversion, we incorporate them in the analyses of the 2015 election.

Data and Methods
Our empirical analyses uses panel survey data collected by the CIS 3 for the Spanish general elections of 2008, 2011 and 2015.The three panels consist of a pre-election survey conducted two to three weeks before the beginning of the official campaign, which in Spain lasts for 14 days, and a post-election survey conducted around a month after each general election. 4The CIS surveys measure a wide variety of socio-economic characteristics and political attitudes by running face-to-face interviews on a representative sample of the Spanish electorate. 5The panel design of the surveys allows us to study the dynamics of individual change.
Our main model includes all respondents that state in the pre-election survey that they have the intention of voting for a party. 6That is, we only consider those voters who could potentially be converted.Our dependent variable, conversion, takes value 1 if the respondent reports a different voting behavior in each wave of the panel, and value 0 if she reports intending to vote (in the pre-election survey) and having voted (in the post-election survey) for the same party.In the party models, we take into account the individual parties involved in the campaign dynamics when we code the dependent variable.For example, conversion to PP takes value 1 when the respondent declares having voted for this party in the postelection survey but had the intention to vote for a different party in the pre-election survey, and value 0 if the voter's choice was not PP in either wave of the survey.
We use two types of analysis to assess the validity of our hypotheses.First, we run logistic regressions with clustered standard errors by district (i.e., province) as observations within each district may not be entirely independent.Secondly, we estimate multilevel models in which we include random intercepts by province to account for different propensities to change the reported choice by district.The results of this second type of analysis and several robustness tests are included in the online Appendix.
Our main independent variables are ideological indifference, heterogeneity in assessments of party competence and shifts in the appraisal of party candidates for prime minister.Regarding the former, respondents were asked in the pre-election survey to place both 3 Although formally dependent on the Spanish government, the CIS is an independent agency with its own legal status and funding.Its aim is to conduct scientific studies of Spanish society. 4CIS studies 2750-2757, 2915-2920 and 3117-3126.  Details about survey questions and sampling techniques are available on the Internet at: http://www.cis.es/cis/openc m/ES/1_encue stas. 6Respondents who state in the pre-election survey that they do not know what they are going to do on Election Day or report in the post-election survey that they do not remember what they did on Election Day are coded as missing.The same applies to people that report a blank or a null vote either in the pre-or the post-election survey or that refuse to answer these questions.
1 3 parties and themselves in a 1 to 10 ideological scale, in which 1 means an extreme left view and 10 an extreme right view.Our indifference variable takes value 1 if the closest party to a voter is equally distant from her than at least a second party and value 0 otherwise.We construct the heterogeneous assessments of party competence as follows: respondents were asked in the pre-election survey which party was in their view most able to handle each of several policy issues.This variable is coded 1 if the respondent reported that no single party was the most competent to handle all issues and 0 otherwise.7Hence, this variable takes value 1 if the voter says a different party is the most able in at least one issue.We also run models with alternative cut-off points: reporting that a different party does better in at least two issues or three issues.The results are similar and are reported in Table A3 of the online Appendix.To build the candidate variables for the pooled models, we averaged the absolute value of the difference in all candidate ratings between the post-and the preelection waves of the panel.8Following previous research on voting behavior, we include in the econometric specifications a number of control variables, such as absence of party identification, gender, age, level of formal education, subjective social class, attendance to religious services, national identification, municipality size, left-right ideology, change in the evaluation of the state of the national economy during the campaign and exposure to political information from the media.In the online Appendix we present the description of all the variables in the models and in Table A1 we show their summary statistics.

Results
Campaigns are often assumed to have minimal effects.Yet, they are far from being irrelevant.Figure 1 confirms the validity of this statement using data from the Spanish general elections examined here.Our results show that campaign conversion has become more consequential and this result connects very well with the dealignment process registered in established democracies in recent years.We find that in the pooled sample for the three elections about 15% of voters changed their reported vote preference during the campaign and that campaign conversion increased remarkably over time.The number of converted voters raised from 11% in the 2008 election to 21% in the election of 2015.These results convey a clear sense of the increasing influence of campaigns and that conversion becomes a more likely behavior in a multi-party system.
Table 1 presents the results of the econometric specifications that take conversion as dependent variable when the three elections and all parties are pooled.Our parameters of interest with regard to valence considerations (i.e., heterogeneous assessments of party competence and changes in candidate evaluations) have a significant impact (P < 0.01) in all models. 9Yet, the impact of ideological indifference is always weaker and fails to reach statistical significance when the controls are included in the analysis (see Model 4).Hence, we find strong support for two of our hypotheses since both valence variables are found to be relevant to explain citizen updates of voting choices made in the pre-campaign period.More to the point, the empirical results highlight valence factors rather than position (i.e., ideological proximity) as a better explanation of why voters switch their party choice during the election campaign.The models by election year displayed in Table 2 offer additional evidence on the performance of our key explanatory variables.Whereas shifts in candidate evaluations remain highly significant across elections, there are important changes in the performance of ideological indifference and heterogeneous assessments of party ability over time.In the 2008 election, the last general election held before the economic crisis, our positional variable (i.e., ideological indifference) has a weak albeit statistically significant influence (P < 0.1) on the probability of changing party choice during the campaign period.By contrast, the effect of heterogeneous valence is indistinguishable from zero.However, in the subsequent elections of 2011 and 2015, ideological indifference is no longer relevant while our valence variable appears in turn as a powerful factor of campaign conversion.So, our findings suggest that campaign vote switchers have paid increased attention to valence considerations as the elections in the last years have revolved around policy alternatives to manage an ailing national economy.
Finally, the party models shown in Table 3 support the main results presented above.Both valence variables are good explanatory factors of campaign conversion for the four parties studied, while ideological indifference fails in all instances but one (i.e., the model for PSOE without controls) to reach conventional levels of significance.This is an important result, as we do not appreciate remarkable variation across parties in terms of conversion in spite of their profound differences with regard to ideology and discourse.
On the role played by the control variables, right-wing and older voters are less likely to change their party choice over the campaign period in the pooled models.Similarly, religious attendance and municipality size tend to decrease and increase, respectively, the probability of changing party preferences in the weeks preceding the election.As theorized (Zaller 1991;Shaw 1999;Franz and Rideout 2007), party identification provides a "perceptive filter" that allows partisans to oppose political messages that are at odds with their beliefs.Consequently, we expect non-partisans to be more open to campaign persuasion than partisans.Our results validate this expectation, although not in the 2011 election or in the case of the new parties, Podemos and Ciudadanos.
We also find that voters with higher levels of education are less prone to alter their reported choice in the pre-election survey when we pool the data in the overall models (Table 1), although the estimated effect is significant only for the 2015 election (Table 2) and for conversion to PP and PSOE (Table 3).In addition, changes in the opinion about the performance of the national economy during the campaign period do 9 To account for any potential collinearity between our valence variables (i.e., heterogeneous assessments and candidates evaluations), we run an OLS regression taking as dependent variable the difference in candidate evaluations between the pre-and the post-election survey and as independent variables all the variables included in the models.As shown in Table A2 of the online Appendix, heterogeneous valence is not a statistically significant predictor of change in candidate evaluations.In addition, we also run multicollinearity tests in which we do not detect any risk of multicollinearity in the models in Table 1.The variance inflation factors in model 2 for heterogeneous assessments and candidates evaluations are 1.57and 2.10, respectively, and in model 4, 1.92 and 3.34, respectively.This indicates that there are no reasons to be concerned.not affect significantly the probability of conversion in the overall models.Yet, when we split the sample by parties we find a negative effect on conversion to PP, and a positive effect on conversion to Podemos (Table 3).The former effect might be explained by the extended view in the Spanish electorate that PP does a good job managing the national economy.So, when voters opinions of the state of the economy deteriorate, conversion to PP is more likely.Contrarily, Podemos is generally perceived by the electorate as a party that is not well suited to run the national economy, so conversion to Podemos should be more likely when voters think that the economy is less of an issue.
As for the media exposure variable, voters who pay more attention to political news are supposed to be less willing to alter their party preference in the campaign period (Butler and Stokes 1969).We expect such well-informed voters to reach the campaign season with a more definite plan about what to do on Election Day (Zaller 2004).Our findings confirm this expectation.The coefficient of the media variable has the predicted negative sign in the pooled models (Table 1) and is significant for the 2011 election (Table 2) and for Ciudadanos (Table 3).
The evidence provided by our tables is limited by the fact that in logistic regressions the strength of the effects cannot be assessed directly from the coefficients.Figures 2  and 3 plot the marginal effects (with 90% confidence intervals) of the explanatory factors on campaign conversion and allow us to get a better measurement of the impact of each factor.To be able to compare the impact of the coefficients, all variables have been standardized to a mean of 0 and a standard deviation of 1.Both figures show that the factors with the strongest marginal influence on conversion are those related to valence: heterogeneous valence and changes in candidate evaluations.The probability of  2, and Models 2, 4 and 6 in Table 3 Fig. 3 Marginal effect of ideological indifference, heterogeneous valence and other independent variables (models by party).Note: Marginal effects are calculated according to Models 2, 4, 6 and 8 in Table 4 Fig. 4 Change in candidate evaluations and predicted probability of conversion (general models and models by year).Note: Predicted probabilities are calculated according to Model 4 in Table 2, and Models 2, 4 and 6 in Table 3 changing party choice as a result of an increase in one standard deviation in the pooled model is around 0.25 and 0.35, respectively.Both effects are of a relevant magnitude.
Likewise, we have calculated the predicted probabilities of conversion associated to changes in the evaluation of party candidates.As depicted in Fig. 4, the impact of this variable is strikingly similar in all elections.Passing from not changing the evaluation of any candidate to the highest possible variation with regard to this variable increases the likelihood of campaign conversion by 50%.Yet, substantial differences are observed depending on the type of party.As illustrated in Fig. 5, the attraction of new voters during the campaign period seems to be more driven by variation in the appraisal of the candidates of established parties than of the emergent ones.
Finally, in Table 4 we re-estimate the party models taken as reference the party that losses rather than the party that wins with conversion behavior.In general, we find consistent results that reinforce our previous results.We observe that in all cases improving the evaluation of the candidate of a party decreases the likelihood of abandoning him, which is the flip side of the effect found in previous analyses.This is farther evidence of the importance of valence for campaign conversion.Likewise, we also find that heterogeneous valence also increases the likelihood of conversion.Voters with mixed preferences are more likely to change the vote intention stated before the campaign starts.Interestingly, however, we find now a slightly clearer effect of the impact of ideological indifference during the campaign.In the models without covariates, those ideologically indifferent are more likely to convert both from PP and PSOE, which are the main traditional parties in Spain.The statistical significance of the results, however, does not hold when we include covariates in the model.Still, and although we did not find a clear direction of those vote flows in Table 3, we see a partial effect of ideological indifference in facilitating conversion  from the mainstream parties.The result, in any case, is of smaller magnitude and weaker than the effect of valence that we have consistently shown in the paper.
To improve confidence in our results, we test their robustness by conducting additional analyses, which are displayed in several tables of the online Appendix.First, we change the cut-off point we use to consider that a respondent has heterogeneous valence and the results are similar (Table A3).Secondly, we change the specification strategy and estimate multilevel models with random intercepts per province (Table A4).In a nutshell, the valence variables outperform once again ideological indifference, which is a significant predictor of conversion only in a bivariate regression.Third, we differentiate between conversion that benefits major parties (i.e., PSOE and PP) and conversion that benefits smaller parties (i.e., all the other).Fourth, in Table A5 we present results that corroborate the explanatory advantage of the valence factors.Fifth, we show that the core results remain robust to the inclusion of the party that the respondent intended to vote in the pre-election wave as a fixed effect (Table A6).Sixth, Tables A7 and A8 display regression models that take as dependent variable conversion to and from PSOE, respectively.10These models aim to capture the role played by individual candidates for conversion in the campaigns.The effects are consistent over elections, indicating that results were not driven by specific candidates.Finally, Table A9 replicates the models in Table 1 including measures of left-right salience from the Comparative Manifesto Project (CMP) as explanatory factors in the models.11This allows us to relax the assumption that all issues are equally relevant in the heterogeneous valence variables and control for some parties making some issues more salient than others.This accounts also for the fact that the assessment of competence might relate to the party itself.Voters can attribute to some parties more competence to solve economic issues and to others (such as the left-wing parties) more competence to solve social issues.However, once included the CMP measures, the results remain unchanged.Altogether, these robustness checks reinforce the conclusions that stem from our main analyses.

Conclusions and future research
The end of cleavage politics (Franklin et al. 1992), the declining importance of party identification (Dalton and Wattenberg 2000) and the end of the two-party system that ruled in the past decades in Spain and other European countries, make election outcomes more difficult to predict and campaigns more salient.Given these patterns, it is surprising the scarce research about the determinants of vote switching in the context of election campaigns.On the whole, previous works tend to be focused on the fluctuations that occur over two consecutive elections.It is remarkable that research on vote change that happens in the period close to Election Day is comparatively rare and that empirical work on the microfoundations of this volatility in multi-party systems is virtually non-existent.
This article is a first attempt to overcome the limitations of previous studies by identifying the factors that account for the instability of voting preferences in the campaign period.We find that the heterogeneous assessments of party competence by voters are a substantial and highly significant predictor of the shifts in party choice over the election campaign.According to our evidence, this variable has a strong influence on changes in the pre-campaign party choice in the overall models, the individual party models and in two of the last three Spanish general elections.Likewise, the literature emphasizes the importance of candidates for elections.We offer evidence consistent with such a view since in our analysis voter changes in perceptions of the quality of candidates are better predictors of the observed volatility of vote choice over the campaign than the ideological indifference of voters towards parties.
These findings may have important consequences for a party's political strategies and the way we perceive voter incentives to participate in elections.Our results suggest that political entrepreneurs that aim to influence voting decisions should focus on valence rather than ideological considerations.Future research should develop this result further in several directions.First, it would be interesting to examine whether some characteristics of voters make them more prone to be influenced by valence considerations or by ideological position when reacting to election campaigns.Secondly, it would be also interesting to assess whether the context matters to explain the effects of valence and ideological indifference.In particular, to what extent the increasing importance of candidates and valence issues is at least partially motivated by the context of economic crisis in which the three studied elections took place.One appealing extension would be to complement our analyses with some experimental evidence in which individuals are exposed to two types of campaigns (i.e., one focused on ideological concerns and one revolving around valence issues) to observe whether conversion effects become more or less likely to happen.Another potential venue for future research would be to enquire into voter evaluations of candidates to identify which particular traits of a politician are valued by the citizens that decide to change their vote preference during the election campaign.
Alternatively, this analysis could be extended to a cross-country research design.According to our results, valence concerns seem to be the leading driver of campaign conversion in the context of Spanish elections.Our analysis by election year also shows the growing numbers of converted voters and the increasing influence of valence motivations over time.It would be interesting to shed more light on a topic for which comparative evidence at the individual level is certainly scarce by clarifying not only whether valence concerns are more relevant than ideology in other democracies, but also how this relevance evolves over time beyond the Spanish case.More specifically, future research may analyze how campaign conversion occurs nowadays in the absence of economic crisis and emerging parties.
To conclude, this article provides some nuance to the classical argument by many scholars that the persuasion effects of campaigns are very limited.For instance, according to Simon (2002), it is unlikely that campaigns will accomplish much because party candidates in election campaigns elude dialogue.Each candidate chooses to talk about the issues on which she is advantaged (that is, closer to the median voter) and avoids discussing themes in which she is disadvantaged.So, in campaigns, candidates will "ignore each other's rhetoric and talk past each other" (Simon 2002, p. 150).Our data allow us to track changes in the vote intention during the campaign.Our view, confirmed by the empirical results, is that campaigns produce more consequential changes in voting preferences.Campaigns allow voters to learn about candidates, issues and policies, persuading some individuals to switch their initial vote choice.In this paper, we have shown that there are voters (like those with heterogeneous valence) who, in order to make their final vote choice, rely on the information conveyed by campaigns and the trustworthiness and competence of parties to carry out their policy proposals.Such voters might revise the preference they reported when the campaign information was not available to them.So, to the extent that campaigns help citizens to make better-informed voting choices, they can be defended as welfare enhancing; that is, as instruments to reach an election outcome more representative of the preferences of the electorate.

Fig. 1
Fig. 1 Magnitude and evolution of campaign conversion

Fig. 2
Fig.2Marginal effect of ideological indifference, heterogeneous valence and other independent variables (general models and models by year).Note: Marginal effects are calculated according to Model 4 in Table2, and Models 2, 4 and 6 in Table3

Fig. 5
Fig. 5 Change in candidate evaluations and predicted probability of conversion (models by party).Note: Predicted probabilities are calculated according to Models 2, 4, 6 and 8 in Table 4

Table 1
The determinants of conversion (general models)

Table 2
The determinants of conversion (models by year)

Table 4
The determinants of conversion from different political parties Variables