The role of candidate evaluations in the 2014 European Parliament elections: Towards the personalization of voting behaviour?

Abstract

We study the personalization of voting behaviour in European Parliament elections. We argue that information from the media is crucial for providing linkages between candidates and voters. Moreover, we contend that candidates can serve as information short-cuts given the complexity of European Union politics. We use a four-wave Dutch panel survey and a media study that enable us to link evaluations of lead candidates, party preferences, and vote choice to exposure to news about these candidates. We show, firstly, that exposure to candidate news is a strong explanatory factor for candidate recognition. Secondly, we find that candidate evaluations positively affect party choice, albeit mainly for those voters who tend to be politically aware. Our research has implications for debates about the European Union’s accountability deficit.

Keywords

Content analysis European Parliament elections media effects personalization of politics voter survey

Introduction

The European Union (EU) suffers from an accountability deficit (Hobolt and Tilley, 2014). One reason for this is that elections to the European Parliament (EP) do not provide voters with the opportunity to hold the EU’s executive to account. Contrary to national parliamentary elections, the executive does not emerge from within the majority (party) in the new parliament. Consequently, scholars have concluded that there would be little at stake in EP elections in terms of political and personal alternatives (e.g. van der Brug and De Vreese, 2016). The introduction of the Spitzenkandidaten procedure in 2014 represents a first counter-development with the lead candidate of the winning party, Jean-Claude Juncker, having been elected by parliamentary majority as the new Commission President.

Amplifying the accountability deficit, the media are accused of providing insufficient information about the responsibilities in EU politics during election campaigns (De Vreese et al., 2006; Hobolt and Tilley, 2014). Voters are hence unable to reward or sanction their representatives when they go to the polls. The aim of the Spitzenkandidaten procedure, by means of the campaigns of pan-European lead candidates, was to motivate European voters to take part in the elections.¹ First assessments of this kind of personalization of the campaigns at the supranational level remain inconclusive about the effects on voter interest and participation (e.g. Hobolt, 2014; Schmitt et al., 2015).

Against this background, we seek to contribute to the broader question of whether the personalization of EU politics may have the potential to contribute to alleviating the alleged accountability deficit (Gattermann, 2017). The personalization of politics is understood as a process in which the focus shifts towards the individual at the expense of parties and institutions. We study whether and to what extent individual politicians play a role in EU elections and embed our research in the wider literature about the personalization of voter behaviour (e.g. Kaase, 1994; Karvonen, 2010; Lobo and Curtice 2015; Renwick and Pilet, 2016; Wattenberg, 1991). Our focus lies on national lead candidates, i.e. candidates at the top of the party lists on the EU electoral ballot, which have thus far received relatively little attention in the extant literature (Giebler and Wagner, 2015). Given the particularities of the EU electoral system most citizens vote for national parties, which then form European party groups inside the EP. This means that – contrary to the Spitzenkandidaten, which voters are unable to elect directly – national politicians offer the primary means of personalizing EU politics. Put differently, it is national politicians who are able to provide an electoral connection and who are therewith directly accountable to EU citizens (see also Hix and Hagemann, 2009).

Following Gschwend and Zittel (2015: 341), we distinguish between two types of personalization at the voter level: ‘cognitive personalization’, addressing the extent to which voters are aware of individual candidates, and ‘behavioural personalization’, that is the role individual candidates play for vote choice. Previous research has argued that candidates’ campaign activities and the extent to which voters are exposed to their campaigns play a central role for candidate recognition (e.g. Giebler and Weßels, 2017; Wolak, 2009) and for the personal vote (e.g. Giebler et al., 2014; Gschwend and Zittel, 2015). In this article, we add another explanatory factor that also relates to both candidates and voters, namely exposure to news about the lead candidates. We argue that information from the media is crucial for providing a linkage between candidates and voters. Moreover, we contribute to the literature on voting behaviour in EU elections by arguing that, alongside party preferences, evaluations of lead candidates are important for party choice given the complexity of EU politics.

To test our hypotheses, we rely on the 2014 EP election campaign study conducted in the Netherlands (De Vreese et al., 2014), which consists of a four-wave panel survey and an accompanying media content analysis. The Netherlands represent an excellent context of our study given the preferential voting system with an ordered ballot structure employed for both EP and national elections. As a consequence, voters tend to be alert to individual candidates, although political parties remain the central actors in Dutch politics (e.g. Van Holsteyn and Andeweg, 2010). In EP elections, list leaders are not party leaders, yet the most high-profile EU politicians Dutch citizens can directly vote for. One important feature of the electoral system is that it ‘does not result in any form of geographical representation’ (Andeweg, 2008: 494) which renders local campaigning less important and thus allows us to examine the role of exposure to candidate news for both cognitive and behavioural personalization.

As regards our first research question – what taps whether voters recognize national lead candidates (cognitive personalization) – our findings show that, while few voters are aware of the lead candidates, alongside party preferences, exposure to news about the lead candidates ceteris paribus has positive effects on candidate recognition. The results suggest further that exposure to candidate news plays a more decisive role than party preferences. Secondly, as regards behavioural personalization, we find that candidate evaluations have a positive effect on deciding which party to vote for, although party preferences have a stronger impact on vote choice. However, candidate evaluations do neither moderate the effect of party preference nor that of exposure to candidate news. Nonetheless, our results indicate that candidates in EU elections may serve as information-short cuts, although this concerns only those voters who tend to be politically aware. Overall, our findings have implications for the debates about the accountability deficit.

Cognitive personalization in European Parliament elections

EU elections do not function in the same way as national general elections: the former are considered second-order national elections which are characterized by lower turnout, and voter and party behaviour being determined by national political considerations (e.g. Reif and Schmitt, 1980). However, the EP has gained considerable policy-making and veto powers over the last few decades, which goes hand in hand with increased media visibility of EP affairs (Gattermann, 2013). Similarly, we find a growing number of genuine career politicians inside the EP at the expense of emeriti national politicians (e.g. Scarrow, 1997). Moreover, there is some evidence for the increasing professionalization of EP election campaigns (see Maier et al., 2011) and even for some increase in media attention at times of EU elections (Boomgaarden and De Vreese, 2016).

These developments suggest that voters have a high chance to be aware of lead candidates standing in EP elections. But we do not know whether and for which voters this is indeed the case. We argue that information is crucial in this respect. In the absence of complete information, voters use cues to make sense of politics and politician behaviour and ultimately to formulate their political preferences and inform their political decisions (e.g. see Lupia, 1994). Specifically, Rahn (1993: 473) argues that partisan cues can be ‘consequential in shaping individuals’ perceptions and evaluations of political candidates’. This also implies that the extent to which voters favour a certain party has an influence on the likelihood that they are aware of individual (lead) candidates. Research has indeed shown that partisan preferences or partisanship have positive effects on candidate recognition (e.g. Giebler and Weßels, 2017; Gschwend and Zittel, 2015; Wolak, 2009). Voters are likely to be familiar with the parties of the individual candidates from domestic politics because voters de facto choose between national parties – and not European parties. And although sometimes new parties form on the occasion of EP elections, we assume that voters are aware of the established parties. Hence, we would expect that if a voter more strongly prefers a certain party over another, she is also more likely to recognize its lead candidate and hence able or willing to provide an evaluation.

H1: The stronger the preference for a certain party, the higher the likelihood that voters recognize the lead candidate of that party.

However, despite the same parties running in EU elections, we argue that voters have less information about the main candidates than they would have in national elections. Voters receive little information about EU politicians from the media during EP election campaigns, although EU elections are nowadays more prominent on the news agenda (e.g. Boomgaarden and De Vreese, 2016). Furthermore, research finds that media attention paid to Members of the European Parliament (MEPs) during the legislative term can be explained by their prominence in domestic politics rather than by their legislative behaviour inside the EP (Gattermann and Vasilopoulou, 2015). In short, citizens hardly know what their European representatives do on their behalf, let alone who they are.

Nevertheless, citizens need information in order to make meaningful choices in elections. We know from existing research that news exposure can have positive effects on the likelihood to participate in EU elections (e.g. De Vreese and Boomgaarden, 2006; Schuck et al., 2016a) and on voter awareness of the Spitzenkandidaten in the 2014 EP election campaigns (Gattermann et al., 2016). However, one pre-condition is that the competing lead candidates also receive attention by the news media; and some are likely to receive more news coverage than others (Gattermann and Vasilopoulou, 2015). The second pre-condition is that voters are exposed to that specific media content in order to learn about the candidates. Importantly, both pre-conditions only work together. In a hypothetical scenario with extensive exposure, but in the absence of any relevant information, cognitive personalization is likely to be hampered.

H2: Exposure to news about the lead candidates increases the likelihood that voters recognize the respective lead candidates.

Exposure to candidate news is also relevant when voters are unable to directly rely on partisan cues. Since information is important for voters to apply cues meaningfully, we also expect that the degree to which party preferences matter for candidate recognition largely depends on exposure to news content about these candidates. If a voter never heard of a particular lead candidate beforehand, the degree to which she prefers the candidate’s party is unlikely to fully explain why or why not she is aware the candidate. Put differently, exposure to candidate media coverage can essentially be understood as a pre-condition for the linkage between party preferences and cognitive personalization.

H3: The effect of party preferences on the likelihood that voters recognize the respective lead candidate (H1) is positively moderated by exposure to candidate news.

European Parliament elections and personalized voting behaviour

Turning to the extent and conditions under which vote choice is influenced by behavioural personalization, extant scholarship is divided over the occurrence and strength of an effect of individual candidate evaluations on party choice. Prominent proponents of the personalization thesis argue that individual politicians have become the main focus of the political space at the expense of political parties (Wattenberg, 1991); and that evidence for such trends would be related to partisan dealignment (e.g. Dalton et al., 2000), to changes in election campaigning (e.g. Reinemann and Wilke, 2007), or to changes in the political media coverage (e.g. Langer, 2007; Rahat and Sheafer, 2007). In line with the personalization thesis, Lodge et al. (1995) argue that candidates provide short-term cues of information, and hence candidate evaluations would be used by voters to decide which party to vote for. This argument has been supported by empirical evidence of voting behaviour in national elections across Europe (Aarts and Blais, 2011; Giebler et al., 2014). The EU political system is characterized by the complexity of EU politics, the length of the decision-making processes and the difficulty to assign responsibility to EU-level politicians (Føllesdal and Hix, 2006; Hobolt and Tilley, 2014). This provides particular reason to expect personalization effects in voter behaviour because ‘faces’ in the form of national lead candidates could actually help citizens to comprehend EU politics and ultimately take an informed decision (see also Adam and Maier, 2010: 239).

H4a: Positive evaluations of lead candidates – ceteris paribus – have a positive effect on vote choice.

The personalization thesis, however, is highly challenged in the literature. Some researchers indeed suggest that leadership evaluations do not play a role for party choice in parliamentary systems (e.g. Kaase, 1994; Schulz et al., 2005), while others assert that party leader evaluations matter for party choice, albeit not increasingly over time (Holmberg and Oscarsson, 2011; Karvonen, 2010). Yet others state that party ratings have stronger effects on the vote than candidate ratings in European countries (Holmberg and Oscarsson, 2011: 39–43). We will hence control for party preferences in the following analysis.

Here, we are interested in the conditions under which the evaluations of the lead candidates can potentially have an effect on vote choice. One would expect that because of partisan dealignment (e.g. Dalton et al., 2000), candidate evaluations only matter for those who do not hold strong party preferences (e.g. see Brettschneider and Gabriel, 2002; Schulz et al. 2005). Aarts and Blais (2011: 155), however, find little evidence to back this assumption and argue that ‘leader evaluations only come into play if these voters see little else to distinguish the parties in terms of policy or performance’. We would hence expect that because EU elections do not serve as a platform for parties to offer policy alternatives (e.g. van der Brug and De Vreese, 2016) alongside some polarization over more or less EU integration such as in the Netherlands (e.g. Meijers and Rauh, 2016), their lead candidates may actually make a difference for voters.

H4b: Evaluations of lead candidates (H4a) are likely to moderate the positive effect of party preferences on vote choice.

As we argued above, the day-to-day political news coverage may further help voters in gathering necessary information to formulate electoral choices. Yet, the literature remains inconclusive about media effects on personalized voting behaviour. Schulz et al. (2005), for instance, find no overtime trend for the impact of general media exposure on the extent to which candidate preferences matter for party choice. Elmelund-Præstekær and Hopmann (2012) find that news exposure is somewhat important for personalized voting behaviour in the form of preferential voting, but that candidate traits and characteristics of the electoral system are more decisive in this respect. However, in their study of Dutch election campaigns, Takens et al. (2015) detect a priming effect and argue that voters ‘weigh in’ their candidate evaluations more into voting behaviour if they have been exposed to personalized news coverage. We would thus expect – in line with the above elaboration about the complexity and abstractedness of EU politics – that being exposed to individual candidate news coverage in EU elections moderates the effect of candidate evaluations on vote choice.

H4c: Exposure to candidate news is likely to moderate the positive effect of the evaluations of lead candidates (H4a) on vote choice.

Research design and methods

Data

We rely on the 2014 EP election campaign study conducted in the Netherlands (De Vreese et al., 2014), which consists of a four-wave panel survey and an accompanying media study. The panel survey was conducted using Computer Assisted Web Interviewing (CAWI); and the fieldwork was carried out by TNS NIPO Netherlands.² The data are representative in terms of age, gender, and education compared to census data. Wave 1 was fielded between 13 and 26 December 2013 and provides major control variables. Our main variable of interest was asked in the second (20–30 March 2014), third (17–28 April 2014) and fourth, post-election wave (26 May–2 June 2014). N comprises 1819 in the second wave (re-contact rate: 83.1%), 1537 in the third wave (re-contact rate: 84.5%), and 1379 in the fourth wave (re-contact rate: 89.7%).

In particular, we rely on the question ‘How do you evaluate the performance of the following lead candidates for the European Parliament, if you know them?’. In the literature, candidate evaluations have been measured in various ways. More general survey questions include ‘feeling thermometers’ (e.g. Aarts and Blais, 2011; Marsh, 2007) or ‘like-dislike’ scales (e.g. Giebler and Wagner, 2015; Holmberg and Oscarsson, 2011). Others have accounted for several different attributes or personality traits, such as competence, leadership, reliability or empathy (e.g. Brettschneider and Gabriel, 2002; Takens et al., 2015). In fact, Funk (1999) demonstrated that individually measured traits have a varying effect on feeling thermometers and recommends distinguishing between traits. Here, we chose to ask respondents to provide performance evaluations because they allow for a broad assessment of less well-known candidates. Indeed, our measure is not without flaws, as respondents could have interpreted the question with respect to past or future performance as well as in terms of general political performance or MEP performance. However, the respective candidates in our study have varying degrees of political experience in general and with respect to the EP. It is therefore an advantage that our question is broad enough to account for different types of political performance.

In wave 2, six lead candidates were listed in random order for the performance evaluation: Hans van Baalen (who was the lead candidate of the People’s Party), Paul Tang (Labour), Dennis de Jong (Socialists), Esther de Lange (Christian Democrats), Sophie in ‘t Veld (Democrats ‘66), and Bas Eickhout (Green-Left). Marcel de Graaff of the populist Freedom Party was added to the third and fourth wave, making it seven candidates overall. Respondents were not provided with the candidates’ party affiliations, i.e. they received no partisan cues. All questions also had a ‘don’t know/cannot say’ response category. The Online appendix provides an overview of the number of evaluations per candidate and wave. Over time, the percentages of those who provide an evaluation is increasing; particularly, there are marked differences between wave 3 and 4. Hans van Baalen appears to be the most prominent lead candidate as fewer respondents picked the ‘don’t know/cannot say’ option (declining from 63.2% to 57.4%). On the other hand, Dennis de Jong (Socialists) and Bas Eickhout of Green-Left were least often evaluated by the respondents (ranging from 10.7% to 18.9%).³

As regards the accompanying media study, news content was sampled and coded for the period between 2 December 2013 and 21 May 2014. Up until 16 April 2014, the first five EU stories in six Dutch news sources, including two television evening newscasts, were coded every third day. For newspapers, these concerned only front pages. EU stories are defined as an article or a news bulletin that mention EU affairs at least twice. From 17 April 2014 until Election Day on 22 May 2014 newspapers’ front pages and one randomly selected page were coded every day; in all media, all EU stories were coded every day (for further details also see De Vreese et al., 2016). This resulted in a total of 4258 news items considered for our analysis (see the Online appendix). The media outlets comprise news from one public and one private television broadcaster (NOS Journaal: 674 news items; RTL Nieuws: 829 news items), two quality newspapers (NRC Handelsblad: 587 articles; De Volkskrant: 615 articles), a tabloid (De Telegraaf: 923 articles), and a major Online news provider (nu.nl: 630 articles).

In each news item up to six political actors were coded. To examine intercoder reliability 16 news items were coded by eight coders. We compared the extent to which coders coded the same individual politicians; Krippendorff’s alpha as a measure of intercoder reliability produced satisfactory results.⁴ For our purposes, we identified the visibility of each candidate in the media prior to the end of each wave, which resulted in a total of 44 references. None of the TV programmes mentioned any of the lead candidates at any time. Generally, candidates were most visible prior to wave 4; Hans van Baalen is the most visible candidate while Bas Eickhout was not visible at all (see the Online appendix).

Dependent variables and models

To examine cognitive personalization, we recoded the candidate evaluation variables into binary variables, where 1 is having provided an evaluation for a lead candidate and 0 is not having done so (category ‘don’t know/cannot say’). We then stacked the data to align the dependent variables with party preferences and to determine explanatory factors within voters (see also Van der Eijk et al., 2006). The unit of analysis is the respondent*candidate*wave. Candidates and waves are not hierarchically interdependent and respondents are cross-nested in candidates and waves. We hence chose cross-classified multilevel logistic regression to examine candidate recognition (see Rasbash and Browne, 2008). This modelling approach allows taking variation both across candidates and waves into account when assessing individual-level explanatory factors.

In order to answer our second research question related to behavioural personalization, we use vote intention as dependent variable in waves 2 and 3 (‘Which party would you vote for if European parliamentary elections were held tomorrow?’) and reported party vote in wave 4 (‘For which party did you vote in the European Parliamentary elections?’). Vote intention and reported behaviour are indeed not the same, but we control for the different waves in our models to address this problem. We identified the parties of the lead candidates and recoded the respective response categories into dummy variables. As above, we aligned the dependent and independent variables in a stacked data matrix. Here, we only consider respondents who took part in all waves. We apply conditional logit models to examine the determinants of vote choice and control for variation between candidates and across waves by dummy variables. This methodological choice appears to be most appropriate given the structure of our data and our research question. However, Van der Brug and Mughan (2007: 36), who also study voting behaviour in the Netherlands, warn that discrete-choice models are ‘not feasible because the small number of votes for some parties makes estimates of their effect parameters unreliable’ (see also Garzia and De Angelis, 2016: 612–613). On the other hand, conditional logit models are also common in electoral research (e.g. Giebler and Wagner, 2015; Gschwend and Zittel, 2015). Furthermore, we only include six and seven parties, respectively. All of them have been elected to the 8th EP as a total of eight parties. We do not have observations for the candidate of the eighth party (Christian Union/Reformed Political Party). The smallest parties in our sample are the Christian Democrats and Green-Left which received a minimum vote share of 9.6 and 11.2% in our data, respectively (wave 4, see the Online appendix).

Independent variables

In order to test H1 and H3, we operationalize party preferences as the respondents’ probability to ever vote for the respective party on a scale from 1 (‘very unlikely’) to 10 (‘very likely’). Van der Brug et al. (2008: 593) state that vote propensities ‘can be regarded for ease of exposition as preferences’. The Online appendix shows that mean values per party fluctuate to some extent between waves. This variable also allows controlling for party preferences in the vote choice model as well as testing H4b. Although there are potential endogeneity problems as determinants of vote propensities can be considered ‘the same as the determinants of party choice’ (Van der Brug et al., 2008: 593), we seek to distinguish between preferences, i.e. attitudes, and (intended) behaviour in our second model (see also Garzia and De Angelis, 2016).

In order to test H2, we linked the number of mentions of each lead candidate in the media to the news exposure of each respondent in a similar manner as Schuck et al. (2016a: 185) who employ weighted measures of news exposure taking into account news content (see also Banducci et al., 2017). Using a weighted measure instead of simple exposure to news content is preferred because – as Schuck et al. (2106b: 209) contend – ‘[h]igh media exposure can mean, de facto, high exposure to celebrity news, sports and weather, so bringing in relevant content features forms a necessary correction for an otherwise potentially inflated or deflated “raw exposure” effect’. In our case, we believe that a voter who actually comes across a lead candidate in her news consumption routine – however this routine looks like – increases the likelihood that she evaluates the candidate. Put differently, a voter who watches television news every day during the campaign period might still not be able to evaluate any of the candidates because she has supposedly never heard of that candidate.

To calculate the weighted news exposure we multiplied an individual respondent’s actual news exposure to each outlet by the number of times a certain lead candidate appeared in the respective outlet prior to each wave. We have taken three important methodological decisions. Firstly, in order to avoid multiplications with zero, we recoded the number of days of exposure, ranging from 0 to 7 days per week, by adding 0.0001 so that the variable ranges from 0.0001 to 7.0001. Secondly, in order not to overstate media visibility of the lead candidates, we calculated the log of media references to each lead candidate ‘[t]o capture decreasing returns of additional information on knowledge’ (Banducci et al., 2017). Thirdly, given the generally low visibility of the lead candidates we also added 1 to the number of candidate mentions per news outlet in the log transformation and in the multiplication with the days of news exposure. Our weighted exposure variables eventually form an additive scale of all six news outlets prior to each wave (see Schuck et al., 2016a: 185):

\sum_{a \in outlets} days * (In (hits + 1) + 1)

The Online appendix shows that there is variation across candidates; and that the weighted exposure values are highest for wave 4. Essentially, if a lead candidate receives no media coverage prior to a certain wave, such as Bas Eickhout, the weighted news exposure scores equal the unweighted values (days of exposure).

This weighted exposure variable also allows testing H3 and H4c by interacting it with party preferences and lead candidate evaluations, respectively. As a robustness check we also estimate our main models with a) the unweighted news exposure to the six news outlets, that is the additive scale of exposure regardless of the candidates’ visibility, and b) comprehensive exposure to news about a lead candidate as well as the respective party, the party leader and any other party member which has been calculated in a similar manner as our main measure (see also recommendations by Schuck et al., 2016b). This helps us determining the extent to which news exposure to the respective lead candidates can indeed explain variation in cognitive and behavioural personalization. We compare the measures in our analyses below.

To test H4a–H4c in our vote choice model, we disregarded all those voters who did not provide a candidate evaluation.⁵ The respondent’s evaluation of a lead candidate is measured by an 11-point scale, ranging from 0 (‘very bad’) to 10 (‘very good’). Lastly, our control variables for both analyses comprise age, the dummy variables higher and lower education (reference category: medium-level education), and gender (dummy female) which we obtained from the first wave of the panel survey. Importantly, we also control for political interest in all models, which becomes particularly relevant for cognitive personalization. Previous research on the determinants of candidate recognition found that political interest has a positive effect (Giebler and Weßels, 2017; Gschwend and Zittel, 2015; Wolak, 2009).

Our modelling approaches require transformation of those independent variables that do not vary within voters during a respective wave. Following the advice of Van der Eijk et al. (2006), we regressed our two dependent variables on all variables – bar probability to vote for a certain party, news exposure to the respective candidates, and comprehensive party exposure – in each wave and computed the y-hats, i.e. the predicted values for each of these independent variables and included these in the stacked matrices. After that, we rescaled all independent variables dividing each by two standard deviations; wave and candidate dummies in the conditional logit models have been mean centred (Gelman, 2008). This allows us to compare effects within models (see also Giebler and Wagner, 2015: 54). The descriptive statistics of all raw variables can be found in the Online appendix.

Findings

We begin by answering our first research question and examine the extent of cognitive personalization among Dutch voters during the 2014 EP election campaigns. We report our results in Table 1 which contains the cross-classified multilevel logistic models. The likelihood-ratio tests are significant for all models. Hence, the chosen models are preferred over ordinary logistic regression.

Table 1.

Cross-classified multilevel logistic regression, predicting candidate recognition.

	Model 1		Model 2
	b	SE	b	SE
Respondent level
PTV for respective party	0.285^***	0.031	0.278^***	0.032
Weighted news exposure to respective candidate	0.741^***	0.032	0.740^***	0.032
PTV for respective party * weighted news exposure to ind. candidates			0.049	0.061
Political interest	1.395^***	0.044	1.394^***	0.044
Age	0.759^***	0.050	0.760^***	0.050
Lower education	−0.093	0.134	−0.090	0.134
Higher education	0.443^***	0.067	0.444^***	0.067
Female	0.200^**	0.096	0.199^*	0.096
Intercept	−1.549^***	0.241	−1.549^***	0.242
Random effects
Candidate level
Variance component	0.195^**	0.123	0.196^**	0.124
Wave level
Variance component	0.090^**	0.081	0.091^**	0.082
N	29,502		29,502
Log likelihood	−12887.022		−12886.696
Wald χ²	2540.35		2542.80
Probability > χ²	<0.001		<0.001
LR test vs. logistic (χ²)	31.13		31.38
Probability > χ² (LR test)	<0.001		<0.001

Note: Independent variables are standardized. *p < .1. **p < .05. ***p < .01.

We expected that the stronger the preference for a certain party, the more likely that voters recognize the lead candidate of that party (H1). Our results lend support to these assumptions. Model 1 shows that party preference has significant positive effects on candidate recognition (b = 0.285). These results are in line with the extant literature that identifies a causal relationship between party preference and candidate recognition (e.g. Gschwend and Zittel, 2015; Wolak, 2009). Turning to H2, which stipulated that exposure to news about the lead candidates increases the likelihood of candidate recognition, our results lend support to the importance of information. Model 1 shows that the effect of exposure to news about a respective candidate is positive holding everything else constant. Furthermore, the effect is considerably larger (b = 0.741) than that of party preferences, which suggests that exposure to candidate news plays a more decisive role than party preferences for explaining why certain voters evaluate lead candidates in EP elections.⁶ Moreover, Figure 1(a) and (b) show that while the effect of party preferences differs across candidates and waves, varying preferences are hardly responsible for these differences. By contrast, Figure 1(c) and (d) demonstrates that we find more variation across candidates and between waves for higher levels compared to lower levels of weighted news exposure. In particular, Figure 1(c) shows that the effects are somewhat stronger for the less well-known candidates Dennis de Jong and Bas Eickhout – although the latter was not visible in the news – compared to the more prominent candidates in our sample, Hans van Baalen and Sophie in ‘t Veld. Put differently, information is crucial for candidate recognition. Similarly, the effects of exposure to news about the respective candidates are also stronger in waves 2 and 3 compared to the post-election wave. This suggests an over-time crystallization effect as voters might have a better idea about candidates and issues the closer the election day and are hence less dependent on information from the media (Figure 1d).

Figure 1.

Fitted values for the probability to vote for respective party (a and b) and weighted news exposure to respective candidate (c and d), by candidate and wave, respectively.

We turn to Model 2 in Table 1 to test H3, which hypothesized a moderating effect of exposure to candidate news on the effect of party preferences. It shows that the effect is not significant. For comparison, when substituting our weighted news exposure measure with either the unweighted or the comprehensive party exposure measures, the results do not yield significant interaction effects either (see the Online appendix). The results thus do not lend support to H3.

The effects of the control variables show that older voters and women are more likely to recognize lead candidates although the latter effect is small compared to the rest. Higher education has positive effects compared to medium-level education, but there are no differences between low and medium levels of education. Lastly, political interest has a positive effect as the literature would expect (e.g. Giebler and Weßels, 2017; Wolak, 2009), and is strongest compared to everything else. Additional tests show further that the positive effect of exposure to news about the respective candidates on candidate recognition decreases with higher levels of political interest and is lower for those who are highly educated (see the Online appendix). This suggests that exposure to candidate news is less important for those who are more politically aware.

Next, we are interested in the extent to which behavioural personalization applies. Table 2 reports the results from the conditional logit models. We expected that the more favourable a voter is towards the lead candidate of a party, the more likely that she will vote for that party (H4a). The results show that the effect of candidate evaluations on vote choice is positive and statistically significant (b = 1.053, Model 1). Similar to voter behaviour in national politics (see Lodge et al., 1995), the results underline that voters in EU politics may use lead candidates as short-cuts to assessing political parties competing for votes in EP elections and thus provide support for H4a. However, party preferences have a stronger effect on vote choice (b = 5.476, Model 1) which resonates with the extant literature on voting behaviour (e.g. Holmberg and Oscarsson, 2011).

Table 2.

Conditional logit models, predicting vote choice.

	Model 1		Model 2		Model 3
	b	SE	b	SE	b	SE
PTV for respective party	5.476^***	0.363	5.406^***	0.385	5.477^***	0.363
Candidate evaluation	1.053^***	0.235	0.761	0.609	1.030^***	0.273
Weighted news exposure to respective candidate	−0.191	0.254	−0.194	0.257	−0.213	0.280
Candidate evaluation * PTV for respective party			0.396	0.737
Candidate evaluation * Weighted news exposure					0.0679	0.334
Political interest	0.119	0.395	0.121	0.394	0.118	0.395
Age	0.549^*	0.302	0.544^*	0.302	0.553^*	0.304
Lower education	0.388	0.303	0.389	0.299	0.386	0.304
Higher education	0.0897	0.237	0.0887	0.237	0.0909	0.238
Female	1.080^***	0.390	1.082^***	0.390	1.081^***	0.390
Wave 3	0.0119	0.153	0.0135	0.154	0.0102	0.153
Wave 4 (ref.: Wave 2)	−0.949^***	0.249	−0.950^***	0.250	−0.949^***	0.248
Tang	1.190^***	0.416	1.184^***	0.417	1.192^***	0.419
de Jong	0.354	0.388	0.365	0.386	0.352	0.386
de Lange	1.119^***	0.340	1.118^***	0.340	1.117^***	0.340
in ‘t Veld	−0.659^**	0.296	−0.659^**	0.297	−0.660^**	0.296
Eickhout	2.127^***	0.608	2.120^***	0.611	2.126^***	0.607
de Graaff (ref.: van Baalen)	1.349^***	0.371	1.347^***	0.367	1.350^***	0.372
N respondents	370		370		370
N	3,545		3,545		3,545
Pseudo R²	0.658		0.658		0.658
AIC	814.371		815.870		816.337
BIC	913.143		920.816		921.283
Log Pseudolikelihood	−391.2		−390.9		−391.2
Chi-Square	327.5		342.6		327.4
Probability > Chi-Square	<0.001		<0.001		<0.001

Note: Robust standard errors; independent variables are standardized or mean centred (dummies); N respondents only includes those who participated in all waves; 308 respondents (1179 observations) dropped because of all positive or all negative outcomes. *p < .1. **p < .05. ***p < .01.

Furthermore, we are interested whether candidate evaluations moderate the positive effect of party preference on vote choice (H4b). The interaction effect in Model 2 does not comply with conventional levels of statistical significance. We thus do not find support for H4b. Likewise, the interaction effect between candidate evaluation and exposure to candidate news (H4c) is also not statistically significant (Model 3).⁷ One potential explanation for these findings could be that we disregarded those respondents who were either unwilling or unable to evaluate a certain candidate. Hence, our sample essentially includes those voters who are more engaged and/or politically aware. This constrains variation between voters.⁸ As regards variation within voters, supposedly the differences in media visibility between the candidates are not large enough to impact on vote decision. Table 2 shows that the direct effect of exposure to candidate news does not meet conventional levels of statistical significance. The same holds for comprehensive party exposure, but the unweighted news exposure has a positive effect (although not in interaction with candidate evaluations, see the Online appendix).⁹ This suggests that information is decisive for voting behaviour, but as a factor that distinguishes between rather than within voters.

Conclusions

EU politics is characterized by complex and lengthy decision-making processes in which individual responsibilities are difficult to identify for citizens; and EP elections hardly provide voters with the opportunity to hold their representatives to account (Føllesdal and Hix, 2006; Hobolt and Tilley, 2014). The personalization of EU elections could potentially make the EU more accessible to its citizens (see Gattermann, 2017). This can ultimately lead to higher levels of citizen participation, such as higher turnout in EU elections (Schmitt et al., 2015). The aim of this article was to shed light onto the phenomenon by examining the role that individual lead candidates play for voting behaviour in EP elections.

Relying on the 2014 EP election campaign study conducted in the Netherlands (De Vreese et al., 2014), we first sought to understand the extent to which cognitive personalization takes place. We have shown that exposure to candidate news plays a more decisive role than party preference for explaining why voters are more likely to recognize lead candidates in EP elections than not. Our aim was to show that information from the news media is highly important during EP election campaigns. This information may eventually help voters to hold their representatives accountable in EU elections (see Hobolt and Tilley, 2014). The wider literature on candidate recognition argues that there are additional explanatory factors related to candidates themselves, including their campaign activities; to voters, such as campaign interest and campaign exposure; or to the local election context (e.g. Giebler and Weßels, 2017; Gschwend and Zittel, 2015; Wolak, 2009). We argued that the campaign context in Dutch EP elections is less important given the particularities of the electoral system. However, we recommend that future research compares the effects of information acquired directly from the media vs information from campaign activities on the recognition of MEP candidates.

Our measure of weighted exposure to candidate news is sophisticated because it takes into account both actual news content and the extent to which an individual is routinely exposed to that type of news content (see Banducci et al., 2017; Schuck et al., 2016a). Yet, we find that unweighted news exposure as well as comprehensive exposure to party news also serve as explanatory factors for the likelihood of providing candidate evaluations in the first place, suggesting that other news features might affect this relationship in addition to specific exposure to candidate news. Moreover, we were unable to account for the entire information environment during the 2014 EP election campaigns. Voters might, for instance, be exposed to additional information about the candidates on social media, which might play a supplemental role for their vote choice (e.g. Kruikemeier, 2014).

Understanding how candidate evaluations come about is relevant because they have a direct and positive effect on vote choice in EU elections. The more positively a voter evaluates the lead candidate, the more likely she is to vote for the candidate’s party. This finding resonates with some research on voting in national elections (e.g. Aarts and Blais, 2011; Giebler et al., 2014) and is important in light of the EU’s accountability deficit. It suggests that individual politicians, in this case lead candidates, may serve as information short-cuts with which voters form their party political preferences in EU elections. However, we also expected that candidate evaluations moderate the effect of party preferences, and that personalized voting behaviour is contingent upon information from exposure to candidate news. Our results do not support these assumptions. We proposed that possible reasons relate to the sample, because those who provide an evaluation in the first place are likely to be more politically aware. Moreover, our data exhibit little within-voter variation when it comes to exposure to individual candidates as these are generally hardly visible. It remains to be seen whether some candidates receive more media coverage in certain outlets in future campaigns (e.g. Labour candidates in left-leaning outlets) and whether selective exposure of voters to this kind of content then has an effect on the evaluations. Furthermore, future research, in the event of additional presence of the candidates in the news, should also investigate the extent to which the tone towards lead candidates in the news during EP elections might influence candidate evaluations and ultimately vote choice, since we know that in national elections positive party evaluations can have positive effects on the motivation to vote for a party (Hopmann et al., 2010).¹⁰

We argued that the Netherlands represent an excellent case given the attentiveness to individual candidates provided by the preferential voting system. Yet, Dutch voters can choose any candidate from a party’s list; and voting for the lead candidate may equally represent a party vote (e.g. see Van Holsteyn and Andeweg, 2010). Our data do not allow us to assess where voters put their cross on the ballot, or whether they vote for a party or a candidate (Elmelund-Præstekær and Hopmann, 2012). Furthermore, our findings might differ in other countries. Voters are likely to be more aware of individual (lead) candidates in small electoral districts and/or electoral systems that employ open electoral ballots supposing more intense and comprehensive campaigning (see Hix and Hagemann, 2009). That would also mean that in such countries more voters take into account their preferences for or against certain candidates when they go to the polls. Although one would hence expect that behavioural personalization is less pronounced in countries where closed ballots in both national and EU elections prevail, such as in Germany (see also Kaase, 1994; Karvonen, 2010), research has shown that evaluations of lead MEP candidates also matter for party choice in German EP elections (Giebler and Wagner, 2015). Comparative research is therefore required to examine the extent to which personalized voting behaviour in EU elections evolves across Europe, if at all. Lastly, comparisons to first-order elections are necessary in order to understand the scope of cognitive and behavioural personalization in EU elections (see Giebler and Wagner, 2015).

Future research should also investigate those factors on which voters base their evaluation of (lead) candidates in EU elections. Voters are likely to take into account past political performance of the candidates – either inside the EP or in domestic politics. This would underline once more that information is crucial in order to evaluate the candidates in the first place. Here, we may also find variation across candidates as regards the degree to which campaigning is personalized within parties (Giebler and Weßels, 2013), differences in journalists’ attention towards MEP candidates (Gattermann and Vasilopoulou, 2015) or exposure on social media (Obholzer and Daniel, 2016).

In the interim, individual (lead) candidates make a difference for only few voters, namely those who are politically aware. Increased concentration of power at the supranational level has led to a politicization of EU affairs, activating politicians, parties, journalists, and ultimately citizens (see Hooghe and Marks, 2009). And since the politicization of EU integration is likely to increase in the future (Kriesi, 2016), the personalization of EU politics is likely to affect more voters. Our aim was to provide a first account of personalized voting behaviour in EU elections; our results show that there is certainly a reason to further investigate the phenomenon.

Footnotes

Acknowledgements

The authors would like to thank three EUP reviewers, Roula Nezi, Wouter de Nooy, Virginie Van Ingelgom as well as participants of the 2016 EPSA Conference, the 2016 Politicologenetmaal and our colleagues of the PCJ research lab for their helpful advice on previous versions of this article. The usual disclaimer applies.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The data collection was largely funded by the Amsterdam School of Communication Research as part of the 2014 European Parliament Election Campaign Study. Katjana Gattermann acknowledges personal funding from the Netherlands Organisation for Scientific Research (NWO Veni Grant, project no. 451-15-003).

Notes

References

Aarts

Blais

(2011) Pull or push? The relative impact of positive and negative leader evaluations on vote choice. In: Aarts

Blais

Schmitt

(eds) Political Leaders and Democratic Elections, Oxford: Oxford University Press, pp. 165–186.

Adam

Maier

(2010) Personalization of politics: A critical review and agenda for research. In: Salmond

(ed.) Communication Yearbook 34, London: Routledge, pp. 213–257.

Andeweg

(2008) The Netherlands: The sanctity of proportionality. In: Gallagher

Mitchell

(eds) Politics of Electoral Systems, Oxford: Oxford University Press, pp. 491–509.

Banducci

Giebler

Kritzinger

(2017) Knowing more from less: How the information environment increases knowledge of party positions. British Journal of Political Science 47(3): 571–588.

Boomgaarden

De Vreese

(2016) Do European elections create a European public sphere? In: van der Brug

De Vreese

(eds) (Un)intended Consequences of European Parliamentary Elections, Oxford: Oxford University Press, pp. 19–35.

Brettschneider

Gabriel

(2002) The nonpersonalization of voting behaviour in Germany. In: King

(ed.) Leaders' Personalities and the Outcomes of Democratic Elections, Oxford: Oxford University Press, pp. 127–157.

Dalton

McAllister

Wattenberg

(2000) The consequences of partisan dealignment. In: Dalton

Wattenberg

(eds) Parties Without Partisans: Political Change in Advanced Industrial Democracies, Oxford: Oxford University Press, pp. 37–63.

De Vreese

Azrout

Möller

(2014) European Parliament Election Campaign Study: Data and Documentation, Amsterdam: University of Amsterdam.

De Vreese

Azrout

Möller

(2016) Cross Road Elections: Change in EU performance evaluations during the European Parliament Elections 2014. Politics and Governance 4(1): 69–82.

10.

De Vreese

Banducci

Semetko

et al. (2006) The news coverage of the 2004 European Parliamentary election campaign in 25 countries. European Union Politics 7(4): 477–504.

11.

De Vreese

Boomgaarden

(2006) News, political knowledge and participation: The differential effects of news media exposure on political knowledge and participation. Acta Politica 41(4): 317–341.

12.

Elmelund-Præstekær

Hopmann

(2012) Does television personalise voting behaviour? Studying the effects of media exposure on voting for candidates or parties. Scandinavian Political Studies 35(2): 117–140.

13.

Esser

Engesser

Matthes

et al. (2017) Negativity. In: De Vreese

Esser

Hopmann

(eds) Comparing Political Journalism, Routledge: Milton Park, pp. 71–91.

14.

Føllesdal

Hix

(2006) Why there is a democratic deficit in the EU: A response to Majone and Moravcsik. Journal of Common Market Studies 44(3): 533–562.

15.

Funk

(1999) Bringing the candidate into models of candidate evaluation. Journal of Politics 61(3): 700–720.

16.

Garzia

De Angelis

(2016) Partisanship, leader evaluations and the vote: Disentangling the new iron triangle in electoral research. Comparative European Politics 14(5): 604–625.

17.

Gattermann

(2013) News about the European Parliament: Patterns and external drivers of broadsheet coverage. European Union Politics 14(3): 436–457.

18.

Gattermann K (2017) Mediated personalization of European Union Politics: Explaining patterns in broadsheet coverage, 1992–2016. Paper prepared for the ECREA Journalism Studies Conference, Odense, 23–24 March 2017.

19.

Gattermann

De Vreese

van der Brug

(2016) Evaluations of the Spitzenkandidaten: The role of information and news exposure in citizens’ preference formation. Politics and Governance 4(1): 37–54.

20.

Gattermann

Vasilopoulou

(2015) Absent yet popular? Explaining news visibility of Members of the European Parliament. European Journal of Political Research 54(1): 121–140.

21.

Gelman

(2008) Scaling regression inputs by dividing by two standard deviations. Statistics in Medicine 27(15): 2865–2287.

22.

Giebler

Wagner

(2015) Contrasting first- and second-order electoral behaviour: Determinants of individual party choice in European and German Federal Elections. German Politics 24(1): 46–66.

23.

Giebler

Weßels

(2013) Campaign foci in European Parliamentary Elections: Determinants and consequences. Journal of Political Marketing 12(1): 53–76.

24.

Giebler H and Weßels B (2017) If you don't know me by now: Explaining local candidate recognition. German Politics 26(1): 149–169.

25.

Giebler

Weßels

Wüst

(2014) Does personal campaigning make a difference? In: Weßels

Rattiner

Roßteutcher

et al. (eds) Voters on the Move or on the Run?, Oxford: Oxford University Press, pp. 139–163.

26.

Gschwend

Zittel

(2015) Do constituency candidates matter in German Federal Elections? The personal vote as an interactive process. Electoral Studies 39: 338–349.

27.

Hix

Hagemann

(2009) Could changing the electoral rules fix European Parliament Elections? Politique Europeenne 28(2): 27–41.

28.

Hobolt

(2014) A vote for the President? The role of Spitzenkandidaten in the 2014 European Parliament elections. Journal of European Public Policy 21(10): 1528–1540.

29.

Hobolt

Tilley

(2014) Blaming Europe? Responsibility without Accountability in the European Union, Oxford: Oxford University Press.

30.

Holmberg

Oscarsson

(2011) Party leader effects on the vote. In: Aarts

Blais

Schmitt

(eds) Political Leaders and Democratic Elections, Oxford: Oxford University Press, pp. 36–52.

31.

Hooghe

Marks

(2009) A postfunctionalist theory of European integration: From permissive consensus to constraining dissensus. British Journal of Political Science 39(1): 1–23.

32.

Hopmann

Vliegenthart

De Vreese

et al. (2010) Effects of election news coverage: How visibility and tone influence party choice. Political Communication 27(4): 389–405.

33.

Kaase

(1994) Is there personalization in politics? Candidates and voting behavior in Germany. International Political Science Review 15(3): 211–230.

34.

Karvonen

(2010) The Personalisation of Politics: A Study of Parliamentary Democracies, Colchester: ECPR Press.

35.

Kriesi H (2016) The politicization of European integration. Journal of Common Market Studies 54(S1): 32–47.

36.

Kruikemeier

(2014) How political candidates use Twitter and the impact on votes. Computers in Human Behavior 34: 131–139.

37.

Langer

(2007) A historical exploration of the personalisation of politics in the print media: The British Prime Ministers (1945–1999). Parliamentary Affairs 60(3): 371–387.

38.

Lengauer

Esser

Berganza

(2012) Negativity in political news: A review of concepts, operationalizations and key findings. Journalism 13(2): 179–202.

39.

Lobo MC and Curtice J (2015) Personality politics? The role of leader evaluations in democratic elections. Oxford: Oxford University Press.

40.

Lodge

Steenbergen

Brau

(1995) The responsive voter: Campaign information and the dynamics of candidate evaluation. American Political Science Review 89(2): 309–326.

41.

Lupia

(1994) Shortcuts versus encyclopedias: Information and voting behavior in California insurance reform elections. American Political Science Review 88(1): 63–76.

42.

Maier

Strömbäck

Kaid

(2011) Political Communication in European Parliamentary Elections, Farnham: Ashgate.

43.

Marsh

(2007) Candidates or parties? Objects of electoral choice in Ireland. Party Politics 13(4): 500–527.

44.

Meijers

Rauh

(2016) Has Eurosceptic mobilization become more contagious? Comparing the 2009 and 2014 EP Election campaigns in The Netherlands and France. Politics and Governance 4(1): 83–103.

45.

Obholzer

Daniel

(2016) An online electoral connection? How electoral systems condition representatives’ social media use. European Union Politics 17(3): 387–407.

46.

Raftery

(1995) Bayesian model selection in social research. Sociological Methodology 25: 111–163.

47.

Rahat

Sheafer

(2007) The personalization(s) of politics: Israel, 1949–2003. Political Communication 24(1): 65–80.

48.

Rahn

(1993) The role of partisan stereotypes in information processing about political candidates. American Journal of Political Science 37(2): 472–496.

49.

Rasbash

Browne

(2008) Non-hierarchical multilevel models. In: Deleeuw

Meijer

(eds) Handbook of Multilevel Analysis, New York: Springer, pp. 301–334.

50.

Reif K and Schmitt H (1980) Nine second order national elections: A conceptual framework for the analysis of European election results. European Journal of Political Research 8: 3–44.

51.

Reinemann

Wilke

(2007) It’s the debates, stupid! How the introduction of televised debates changed the portrayal of chancellor candidates in the German Press, 1949–2005. International Journal of Press/Politics 12(4): 92–111.

52.

Renwick

Pilet

(2016) Faces on the Ballot: The Personalization of Electoral Systems in Europe, Oxford: Oxford University Press.

53.

Scarrow

(1997) Political career paths and the European Parliament. Legislative Studies Quarterly 22(2): 253–263.

54.

Schmitt

Hobolt

Popa

(2015) Does personalization increase turnout? Spitzenkandidaten in the 2014 European Parliament elections. European Union Politics 16(3): 347–368.

55.

Schuck

ART

Vliegenthart

De Vreese

(2016a) Who’s afraid of conflict? The mobilizing effect of conflict framing in campaign news. British Journal of Political Science 46(1): 177–194.

56.

Schuck

ART

Vliegenthart

De Vreese

(2016b) Matching theory and data: Why combining media content with survey data matters. British Journal of Political Science 46(1): 205–213.

57.

Schulz

Zeh

Quiring

(2005) Voters in a changing media environment: A data-based retrospective on consequences of media change in Germany. European Journal of Communication 20(1): 55–88.

58.

Takens

Kleinnijenhuis

van Hoof

et al. (2015) Party leaders in the media and voting behaviour: Priming rather than learning or projection. Political Communication 32(2): 249–267.

59.

Van der Brug W and De Vreese CH (eds) (2016) (Un)intended Consequences of European Parliamentary Elections. Oxford: Oxford University Press.

60.

Van der Brug

Franklin

Tóka

(2008) One electorate or many? Differences in party preference formation between new and established European democracies. Electoral Studies 27: 589–600.

61.

Van der Brug

Mughan

(2007) Charisma, leader effects and support for right-wing populist parties. Party Politics 13(1): 29–51.

62.

Van der Eijk

van der Brug

Kroh

et al. (2006) Rethinking the dependent variable in voting behavior: On the measurement and analysis of electoral utilities. Electoral Studies 25(3): 424–447.

63.

Van der Kolk H, Tillie J, van Erkel P, et al. (2013) Dutch Parliamentary Election Study 2012. Data Archiving and Networked Services (DANS), Statistics Netherlands (CBS), Dutch Electoral Research Foundation (SKON), Amsterdam, June 2013.

64.

Van Holsteyn

JJM

Andeweg

(2010) Demoted leaders and exiled candidates disentangling party and person in the voter's mind. Electoral Studies 29(4): 628–635.

65.

Wattenberg

(1991) The Rise of Candidate-Centered Politics: Presidential Elections of the 1980s, Cambridge, MA: Harvard University Press.

66.

Wolak

(2009) The consequences of concurrent campaigns for citizen knowledge of congressional candidates. Political Behavior 31(2): 211–229.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.27 MB

8.42 MB

0.02 MB