Partisan heart and/or rational mind? Party identification, political knowledge and electoral turnout

This paper aims at analysing the relationship between positive and negative party identification and political knowledge, as well as their interplay in regard to electoral turnout. The data from the Serbian 2012 post-election public opinion survey on the nationally representative probability-based sample of voting age citizens were used (N = 1568). The data show that political knowledge is significantly and positively correlated both with positive and negative party identification. Citizens who are more knowledgeable and positively identified are also more likely to vote. However, it is shown that the influence of political knowledge on electoral turnout is moderated only by positive party identification. Political knowledge significantly predicts turnout only in the group of citizens without positive party identification. Being negatively identified with a political party does not modify the relationship between political knowledge and turnout. The results are discussed in terms of the Michigan model of party identification and cognitive mobilisation thesis. The roles of affective and cognitive motivation in electoral participation are additionally stressed and debated.

Party identification is one of the most important concepts in the study of voting behaviour. The very notion of party identification, introduced in the well-known study of the University of Michigan scholars (Campbell, Converse, Miller, & Stokes, 1960), has become a reference point in explaining voting behaviour ever since (e.g. Clark, Sanders, Stewart, & Whiteley, 2004;Holmberg, 2007;Lewis-Beck, Jacoby, Norpoth, & We isberg, 2008).
In the original formulation, party identification was defined as "individual's affective orientation to an important group-object in his environment" (Campbell et al., 1960, p. 121). Party identification is acquired during the socialisation process as a form of social identity and under the prevailing influence of early, family socialisation. This psychological affinity towards a political party is regarded as a stable and long-term attachment, changeable only under some extraordinary personal or social circumstances, and immune to on-going political and economic issues (Campbell et al., 1960;Holmberg, 2007;Lewis-Beck et al., 2008). Once established, it hugely influences voting behaviour (Campbell et al., 1960;Clark et al., 2004;Dalton, Farrell, & McAllister, 2011;Green, Palmquist, & Schickler, 2002;Lewis-Beck et al., 2008).
In the development of party identification, rational considerations and cognitions are overshadowed by non-rational processes, self-understanding, identity process and group-belonging (Campbell et al., 1960;Greene, 2004;Lewis-Beck et al., 2008). Rather than being a product of acquired political information, political awareness and deliberation, the social identity aspects of partisanship are stressed as crucial (Campbell et al., 1960;Greene, 2004). In fact, the concept of party identification offered one explanation of how the average individual, who, as a rule, lacked any deep understanding of politics (Campbell et al., 1960;Converse, 1964;Delli Carpini, & Keeter, 1996), managed the complexities of democratic politics. Becoming well-informed bears costs; instead of closely monitoring the political process, party identification allows voters to rely on this simple rule-of-thumb when making political decisions (Lau, Andersen, & Redlawsk, 2008;Shively, 1979;Sniderman, Brody, & Tetlock, 1991;Sniderman, Glaser, & Griffin, 1990;Sniderman & Stiglitz, 2012).
However, the link between party identification and political cognitions has been intensely debated. One line of theorising argues that party identification is in fact based on political cognitions and evaluations. Party identification has a direct parallel with, for example, Downs' (1957) rational choice theory and the concept of the 'standing vote'. As a way of minimising the costs of collecting political information continuously, voters develop a habit of supporting one party, closest to their ideological and policy positions. Similarly, in his revisionist model of party identification, Fiorina (1981) argues that, rather than being identity-based, party identification is the consequence of rational calculations, the "pros" and "cons", a "running tally" of party's past performances, which is constantly being updated. Quite contrary to the Michigan model, these models imply by far a more important role of political cognition and evaluations in developing party attachments. Party identification is rather the consequence than the cause of political knowledge (Fiorina, 1981). Additionally, the models stressing the importance of cognitive factors and the corresponding changeability of party affective ties can better accompany the well documented data showing the volatility of party identification in a relatively short time-span (Lockerbie, 1989(Lockerbie, , 2002 and under the influence of the election campaign and economic evaluations (Niemi & Weisberg, 1993), which the original model is unfit to fully explain.
On the other hand, political cognitions and evaluations play a substantial role in explaining another well-established phenomenon -a decline in the number of partisans and the growth in the number of independent voters in many Western countries (Dalton & Wattenberg, 2000;Nie, Verba, & Petrocik, 1976). In response to this, several studies have pointed to a related rise in 'consumer voting' (Himmelweit, Humphreys, & Jaeger, 1985). Voters, like percipient consumers, pick and choose between parties on the basis of relatively short-term considerations. Others explain this decline in vote choices based on an affective identification with the rise in issue voting (Carmines & Stimson, 1980), by which a voter chooses the party that is suited best to represent his/her personal preferences. Finally, numerous scholars argue that the registered de-alignment has more general causes, which are most systematically elaborated under the cognitive mobilisation thesis (Dalton, 1984(Dalton, , 2007Inglehart, 1990). The cognitive mobilisation model posits that, during the past decades, the public's ability to process political information has increased as a function of higher levels of education and political sophistication (Dalton, 1984(Dalton, , 2007. The expansion of mass media and other sources of information have lowered the cost of acquiring political information as well (Dalton, 1984(Dalton, , 2007. Citizens nowadays are thus better equipped with the necessary resources and skills which enable them to manage the complexities of politics and reach their own political decisions without relying on affective or other external cues. Increased political knowledge thus makes the development of affective ties superfluous, and enables voting decisions based on issues and short-term factors (Dalton, 1984(Dalton, , 2007Dalton & Wattenberg, 2000). Combining party identification and cognitive mobilisation measures, Dalton (2007) has developed a four-fold typology of voters. It differentiates between apartisans (those high on cognitive measures and low on, i.e. lacking, party identification), cognitive partisans (high on both measures), ritual partisans (high on party identification and low on cognitive mobilisation) and apoliticals (low on both measures). He has focused on the apartisans category, arguing that they represent the today's independent votersrational, informed citizens, able to make their own political decisions, and close to the normative ideal of the democratic voter.
In sum, the revisionist model of party identification and the cognitive mobilisation thesis both stress the importance of political cognitions for party identification. However, their supposed effects are just the opposite -enhancing and enabling party identification in the former case, and eroding it in the latter. The research evidence on the relationships between party identification and political sophistication/knowledge is rather mixed. Independence in voting behaviour has been well documented (Dalton & Wattenberg, 2000), implying the diminishing party dependability among the electorate as a consequence of rising levels of political sophistication on the aggregate level. Similar findings showed that political sophistication increases vote intention switching (Dassonneville, 2014). Others failed to find evidence in support of the cognitive mobilisation thesis in different countries (Albright, 2009;Arzheimer, 2006;Huber, Kernell, & Leoni, 2005;Poletti, 2015), showing that cognitive mobilisation in fact does not decrease, but quite the contrary, increases the probability of being a partisan. Zaller (1992) also argued that party attachments should be the strongest just for the cognitively sophisticated because they are best able to filter out information that conflicts their political predispositions. Studies have also shown that party identifiers exhibit higher levels of political knowledge, but only in some specific areas, such as familiarity with the number of political parties or differentiating between them (Smith, 1989).
On the other hand, a small number of studies suggest that the relationship between party identification and political knowledge, on the one hand, and voting behaviour, on the other, is far more complex. There is no dispute that party identification (Campbell et al., 1960;Clark et al., 2004;Dalton et al., 2011;Green et al., 2002;Lewis-Beck et al., 2008) or political sophistication and knowledge (Delli Carpini & Keeter, 1996;Galston, 2001;Gronlund, 2007;Gronlund & Milner, 2006;Marquis, 2010) make one more prone to voting. Yet, the interaction between party identification and political knowledge in regard to turnout has been insufficiently explored.
Furthermore, all of the aforementioned models stress the link between political knowledge and positive party identification. Although the original study allowed for the possibility of negative identification (Campbell et al., 1960), only recently has the notion of negative party identification gained prominent research interest (e.g. Mayer, 2015;Medeiros & Noël, 2013;Rose & Mishler, 1998;Vlachová, 2001). The notion of negative party identification has been debated and elaborated under the social identity theoretical framework (Mayer, 2015;Vlachová, 2001), found to be an autonomous and significant determinant of electoral participation (Mayer, 2015;Medeiros & Noël, 2013), with different sources and outcomes than the positive one (Medeiros & Noël, 2013).
The role of negative party identification in the post-communist societies is often discussed as especially relevant. The citizens of these countries had to pay close attention to the messages of the communist elites and developed a sort of dualistic political thinking, the outward conformity and inward rejection (Rose & Mishler, 1998). It resulted in the politically aware, knowledgeable or sophisticated public. On the other side, the post-communist societies often suffer from the lack of trust in political institutions (including the political parties), which makes a more intense and widespread negative identification a reality in such countries (Rose & Mishler, 1998). Still, the role of political cognitions in developing negative party identification has been rarely empirically studied. The scarce evidence suggests that one of the very important determinants of negative party identification is ideology (Medeiros & Noël, 2013). Since ideology is a rather a matter of rational dimensions of political thinking, the role of political cognitions in negative party identification could be prominent.
The reasoning becomes even more complex if we have in mind that being "independent" from party attachments almost exclusively means independent from the positive affective tie. Hence, while political sophistication might erode positive party identification, the question of the "effects" of cognitive mobilisation on negative party identification is open. Additionally, the issue of the negative and positive party identification and political knowledge interplay in terms of the turnout has remained unresolved. Independents, in other terms, might not be so independent after all when negative party evaluations are taken into account.

Present study
This paper analyses the interplay of political knowledge and (positive and negative) party identification in regard to voting. The aim of the study is twofold: (1) to analyse the relationship between positive and negative party identification and general political knowledge, and (2) to analyse the possibly moderating role of positive and negative party identification in the relationship between political knowledge and electoral turnout. In doing so, we tend to add to the on-going debates in three ways.
Firstly, it should be stated that the link between political knowledge and party identification has in fact been rarely studied. The measures of cognitive mobilisation used in the analyses of this kind usually comprise an index combining the education level and political interest (e.g. Albright, 2009;Dalton, 2007). We use a "pure" measure of political knowledge that higher education and political interest should supposedly emulate. Secondly, we differentiate between and analyse both poles of party identification, the positive and negative, and their associations with political knowledge. Finally, the main purpose of the present survey is to analyse the role of party identification in moderating the link between political knowledge and turnout. When the relationship between political cognition and party identification is discussed, it is almost exclusively in relation to vote choice -how a voter makes a decision who to vote for. When, for instance, Dalton (2007) argues that "new independents" rely less on party cues and more on the questions of issues and ideology, it is in the context of the vote choice criteria (see also Kuan & Lau, 2002). However, the question of turnout, i.e. the decision to vote at all, as well as the role of negative feelings, especially for the cognitively mobilised group, is under-researched. In other words, apartisans might vote for different reasons than partisans do, but are they equally likely to vote at all? This overview pointed out that there are different views regarding the role of political knowledge in the formation of either positive or negative party identification. Cognitions play a minor role in the Michigan model of party identification, but an important one in the Fiorina's model (Fiorina, 1981); the cognitive mobilisation thesis states that increasing political awareness erodes party affective ties. The Serbian political context provides a fertile ground for the present analysis and adds a new frame of reference in debating these issues. The multiparty political system in Serbia has been introduced relatively recently and is still highly unstable. Growing either positive or negative feeling towards political parties in such a social context should rather be a matter of cognitions and evaluations than of social identity acquired in early socialisation. (It goes without saying that the majority of Serbian citizens today are in fact "older" than all of the relevant political parties). Hence, we propose the following hypothesis: H1: Higher political knowledge is related to greater likelihood of having both positive and negative party identification.
The question of the effects of political knowledge and party identification on turnout is rather a different one. There is no dispute that those more knowledgeable and those party identified are more likely to vote (Campbell et al., 1960;Clark et al., 2004;Dalton et al., 2011;Delli Carpini & Keeter, 1996;Galston, 2001;Green et al., 2002;Gronlund, 2007;Gronlund & Milner, 2006;Lewis-Beck et al., 2008;Marquis, 2010). The same goes for the Serbian political context: party identification and political knowledge are very important factors of turnout and/or vote choice (Mihailović, 1996;Milošević, 1999;Milošević-Đorđević, 2005Milošević-Đorđević, , 2006Pavlović, 2012Pavlović, , 2013. Having in mind that a small number of surveys have shown that negative party identification increases turnout as well, we hypothesise that: H2: More knowledgeable and (positively and negatively) party identified citizens will be more likely to vote.
Finally, we hypothesise that the role of political knowledge in boosting turnout should be dependent on positive and negative affective party ties. One can speculate that although positive party identifiers are biased in favour of one party (Campbell et al., 1960;Green et al., 2002;Lewis-Beck et al., 2008), they otherwise respond to cognitive factors (issues, evaluations etc.) in the same way as the voters who do not identify with any party (Jessee, 2010;Lachat, 2015). Increasing political knowledge could thus make such voters additionally prone to voting in general or voting for the preferred party in particular.
Yet, this rationale, a sort of an "additive" logic concerning the role of cognitive and affective factors is disputable. The majority of voters base their vote choice both on positive party identification and issue proximity (Carmines & Stimson, 1980), but the latter is conditioned by political sophistication (Gerber, Nicolet, & Sciarini, 2015). Similarly, those more knowledgeable are under a weaker influence of positive party identification when making vote choice decisions than the less knowledgeable voters (Kuan & Lau, 2002), and deploy more intensely economic evaluation and ideology (Kuan & Lau, 2002) or issue proximity (Gerber et al., 2015;Lachat, 2015;Marquis, 2010).
In line with this, the present study suggests a rather different role of party loyalties in regard to electoral turnout. Whatever its sources, party identification could be seen as a sort of a pre-election made decision to vote, with possibly varying vote choices. Party identification is an indication of psychological investment in politics (Campbell et al., 1960;Lewis-Beck et al., 2008). Therefore, in case of a positive identification, a person has a party to vote for and does not want to betray those he/she likes. Similarly, he/she also has parties he/she dislikes, a negative identification, and does not want to see them winning. In other terms, being a (positive or negative) partisan low or high on political knowledge could be of special relevance for the decision to vote or not to vote for the preferred party, as already shown (e.g. Kuan & Lau, 2002;Lachat, 2015). Still, the differing levels of knowledge should not further undermine or enhance the already made decision to vote, i.e. participate in the elections.
Those not positively or negatively identified, on the other hand, suffer from low initial investment into political process, and therefore are less likely to vote. What can compensate for the absent identification are increased political interest and the correspondent body of political knowledge, based on various motives: the sense of civic duty (Delli Carpini & Keeter, 1996), or, for instance, simply the entertainment value of elections (Downs, 1957). At the same time, higher knowledge is likely to reflect higher acceptance of civic cuties (Delli Carpini & Keeter, 1996) and better understanding of democratic norms (Galston, 2001;Memoli, 2011). Thus, given the absence of party identification, it can be expected that more knowledge leads to higher turnout. Yet, the impact of political knowledge on electoral turnout decision should be weaker among those with either negative or positive party identity. Studies have already shown that positive party identification moderates the impact of issue preferences on vote choice, it being stronger in a non-partisan group (Lachat, 2015). We similarly hypothesise that the relationship between political knowledge and electoral turnout should be conditional on party identification and propose the following hypothesis: H3: Political knowledge increases turnout, but only in the group of citizens without positive and negative party identification.
This hypothesis, simply put, states that political knowledge will be a far more important electoral resource for those who are independent of positive and negative affective party ties.

Participants
The data from a post-elections survey conducted on the nationally representative sample of voting age citizens of Serbia (N = 1,568) were used. The sample was probabilitybased with multiple stages of selection. Data were collected by face-to-face CAPI interviews, from December 2012 to February 2013. Data were weighted to correct for the sampling procedure and demographic parameters during analysis.

Data and measures
Political knowledge. The level of political knowledge was measured by four multiplechoice questions knowledge test, a part of the Comparative Study of Electoral Systems (CSES) Module 4 questionnaire, covering the core political knowledge themes -factual politics and foreign affairs (Delli Carpini & Keeter, 1996;Gronlund, 2007;Gronlund & Milner, 2006;Pavlović, 2012Pavlović, , 2013. The respondents answered by choosing one of the four offered answers to the questions probing for the familiarity with the name of the Serbian Finance Minister, unemployment rate in Serbia in the first trimester of 2012, a party or coalition that came in second in seats in the National Assembly, and the name of the current Secretary-General of the United Nations. The number of correct answers was treated as a measure of the level of political knowledge. Positive party identification. As already stated, we differentiate between positive and negative party identification. The standard question and common procedure in constructing the measure of party identification was used as an indicator of positive party identification. The respondent was asked whether he/she felt close to any political party (Yes/No). Those who said "Yes" were coded as partisans (1); "No" response was treated as an absence of party identification (0).
Negative party identification. Negative party identification is usually operationalised in two ways. Respondents may be asked whether there is a political party that they would never vote for; having the least preferred political party is treated as an indication of the presence of negative party identification (Medeiros & Noël, 2013;Rose & Mishler, 1998;Vlachová, 2001). The other procedure includes probing for the likeability of political parties, most commonly by a sort of feelings/likeness thermometer question (e.g. Maggiotto & Piereson, 1977;Mayer, 2015). We used the latter approach, since it focuses on a negative affective evaluation of a party rather than on a negative behavioural attitude (Mayer, 2015). The respondents were asked to assess the (dis)likeability on an eleven-point scale (0 -Strongly dislike, 10 -Strongly like) of seven major political parties in Serbia (SNS, DS, SPS, DSS, LDP, URS and SRS). Those who said that they strongly disliked at least one of the main political parties were coded as having negative party identification (1); others were coded as not negatively identified (0). Electoral turnout. Electoral turnout was measured by the following question: "Which of the following statements best describes you: One, I did not vote in the elections on May 6th of this year; Two, I thought about voting this time -but didn't; Three, I usually vote, but didn't this time; or Four, I am sure I voted?". A dichotomous measure of turnout was constructed, differentiating those who voted (1) from those who did not (0). Socio-demographics. Socio-demographic variables, such as age, education, socioeconomic status and educational level are well-known correlates of party identification (e.g. Campbell et al., 1960;Lewis-Beck et al., 2008), political knowledge (e.g. Delli Carpini & Keeter, 1996) and turnout (e.g. Verba, Nie, & Kim, 1979;Verba, Schlozman, & Brady, 1995). Since we are primarily interested in the relationship between political knowledge, party identification and turnout when other relevant factors are controlled for, we used several variables as controls: gender, education, age, socio-economic status (SES) and urbanisation. Age was measured in years since birth. Educational level was measured in 8 degrees (ISCED categorisation). Socio-economic status was measured by questions probing for material possessions of the respondent (ownership of home, business, stocks, car less than 8 years old, bank savings). Scores from individual items (0 -No, 1 -Yes) were summed up and treated as a measure of respondents' SES. The size of settlement was used as a measure of urbanisation.

Results
The analysis of the relationship between political knowledge, party identification and electoral turnout begins with correlation analysis and descriptive measures. The relative importance of political knowledge for electoral turnout was analysed by hierarchical logistic regression models, performed separately in the partisan (having positive party identification) and non-partisan (without positive party identification) subsamples. Political knowledge and socio-demographic variables were entered as predictors in the first step, negative party identification in the second, and the interaction term Political Knowledge × Negative Party Identification in the third step.
The correlations between all variables included in the analysis and their descriptive measures are presented in Table 1. We are mainly interested in the relationship between political knowledge, party identification measures and turnout.
Political knowledge is, as predicted, positively and significantly correlated with positive and negative party identification. Quite contrary to the cognitive mobilisation thesis, both positive and negative party identifiers are more knowledgeable of politics. Similarly, there is a positive correlation between both political knowledge and positive party identification with turnout. The turnout propensity is higher among the highly informed citizens, as well as among those with positive party attachments. On the other hand, negative party identification is not significantly related to turnout. Negative party identification is, as shown in Table 1, more than two times more widespread in the sample, while the two sorts of party identification are significantly correlated. Having a party one is close to at the same time implies that there is a party one also strongly dislikes. Finally, in regard to the role of socio-demographic variables, gender, age and SES are the only variables positively correlated with both party identification measures; gender, educational level and SES are positively correlated with political knowledge, while the turnout is higher among older, more affluent and the respondents from rural areas. Socio-demographic variables are also significantly inter-correlated, which additionally suggests controlling for their variation in analysing the relationship between political variables.
The results of the three hierarchical logistic regression models, performed on the whole sample as well as separately in the subsamples of those with and without positive party identification, are shown in Table 2. Socio-demographic variables and political knowledge are entered as predictors in the first model. Controlling for the socio-demographic variables, political knowledge significantly predicts the electoral turnout in the whole sample as well as in the group of citizens without a positive party identification, but not in the group of those with it. For the positive partisan group, socio-demographic variables, age, education and urbanisation are far more important predictors of turnout than political knowledge. On the other hand, aside from the fact that being older is associated with a higher propensity to vote, all of the other socio-demographic variables prove to be irrelevant determinants of turnout in the non-partisan group.
The negative party identification is then added as a predictor in the second step. Controlling for the socio-demographic variables and political knowledge, negative party identification is not a significant predictor of turnout in the sample as a whole, as well as in either of the two subsamples defined by positive party identification.
Finally, since we hypothesised that the effects of political knowledge on turnout should be dependent on the presence/absence of positive as well as negative party identification, the interaction term Political knowledge × Negative party identification is added in the third step of the analysis. However, it has proved non-significant in the sample as a whole and the partisan and nonpartisan group. To obtain a better insight into the relationship between political knowledge and electoral turnout in regard to party identification measures, the predicted probabilities of voting by the level of political knowledge and positive and negative party identification are graphed, as shown in Figure 1. The probability of voting in the positive partisan group is generally high, irrespective of the political knowledge level or the presence of negative party identification. Increasing political knowledge or the fact that one has or does not have the negative party identity does not significantly change the probability of voting as long as one is positively attached to a political party.
On the other side, political knowledge dramatically increases the probability of voting in the group of those without a positive party identification. The effects of political knowledge on voting are, in other terms, highly dependent on the presence or absence of positive party identification. However, they are not under the influence of negative party identification; the propensity to vote increases with the growth of political knowledge irrespective of the presence or absence of negative party identification. Last, but not least, the least knowledgeable positive partisan is more likely to vote than the most knowledgeable citizen without positive party identification. It seems that political knowledge makes one prone to voting (irrespective of negative party identification), but cannot fully compensate for the lack of positive party identification.

Discussion
The purpose of this paper was twofold. In the framework of the on-going debate regarding the relationship between party identification and political knowledge, the link between these two crucial determinants of electoral turnout was analysed. We firstly hypothesised that political knowledge should increase the likelihood of having positive and negative affective ties. The data have confirmed such expectations. Partisans are more knowledgeable of politics: an increase in political knowledge increases the probability of being a positive party identifier, which is in line with recent research elsewhere (Albright, 2009), and counters the main assumptions of the cognitive mobilisation thesis (Dalton, 2007). The results have additionally revealed a similar relationship between political knowledge and negative party identification. Thus, it seems that we have essentially dealt with different indicators of the general involvement in the political process. One is the affective connection with parties, the other is cognitive involvement. They are correlated because there is, so to speak, mutual affinity between them. Liking a party makes it easier to pay attention and retain some information about politics and vice versa. Paying attention to politics and knowing about it should also help forming negative affective links with some other parties.
This bears significant relevance for the described models of party identification. The multi-party system in Serbia is relatively novel, and the party system itself is marked by high instability. It takes time for citizens to habituate to the idea of a competitive party system and for socio-economic cleavages to relate to political parties. In places like Serbia, party identification is primarily the matter of likeability of a party, rather than a long-term consequence of (early) socialisation (Holmberg, 2007;Pavlović & Todosijević, 2016). In addition, the revised model of party identification (Fiorina, 1981) could be of use in describing the nature of party identification in such instances and is supported by the presented data regarding the positive relationship between political knowledge and party identification measures. More knowledgeable individuals could be more able to evaluate past performances, relate them to their preferences and grow affective ties to a political party in the end. Political knowledge helps citizens translate a cognitive judgment (e.g. evaluations, issue proximity) into an evaluative judgment (i.e. voting preferences) (Marquis, 2010). Failing to perceive differences between parties, for example, due to the lack of relevant political knowledge, may lead to abstention (Downs, 1957;Popkin & Dimock, 1999). Similarly, the more complex the political system, in terms of the number of parties or its novelty (as in the post-communist countries), the more important are cognitive resources for information processing in determining the development of a party identification (Miller & Klobucar, 2000). Finally, being negatively identified is a demanding cognitive task, which is obviously enhanced by political knowledge.
Overall, whether being the cause of party attachments or its effect, political knowledge plays a prominent role in party identification. Although this argument could imply that party identification has different sources than originally proposed, it does not, however, disqualify the basic assumption of the classic model which depicted the independents (i.e. non-partisans) as generally uninvolved in politics, politically uninterested and less sophisticated (Campbell et al., 1960). Similarly, party identification could still be connected to the social identity theory (Greene, 2004;Green et al., 2002) -identification can be quickly formed and requires little information. In our case, the main party had been formed briefly before the elections, but there were plenty of reasons for voters to recognise its position in the ideological space relevant for Serbia and to positively or negatively attach to it.
Our second hypothesis has proposed that more knowledgeable and affectively involved citizens are more likely to be voters. Yet, it has been confirmed only partially, since negative party identification, unlike political knowledge and positive identification, does not boost turnout. Similarly, our third hypothesis, stating that positive and negative party identification will moderate the link between political knowledge and turnout, has been partially confirmed as well. The logistic regression models have shown that political knowledge significantly predicts electoral turnout only in the group of citizens without positive party identification. The influence of political knowledge on voting is thus highly dependent on the presence of positive party identification. Furthermore, positive party identifiers are generally more prone to voting than non-partisans, irrespective of the level of political knowledge. These findings have several important implications.
The fact that for the positive partisans political knowledge does not make a significant difference in increasing the proneness to vote does not mean that political knowledge is completely insignificant for them. This is, actually, supported by the presented data showing the positive relationship between political knowledge and party identification measures. It implies that the partisans need fewer pieces of information or rely less on them when making their decision to vote. To use Dalton's typology once more (Dalton, 2007), it does not make much difference whether a partisan is cognitive (high on cognitive mobilisation) or ritual (low on cognitive mobilisation) as long as he/ she is partisan. He/she is equally prone to vote, because he/she is committed to politics and affectively invested. Either way, the act of voting for those who are partisans seems to be less demanding -adding more information does not make much difference.
Citizens without positive affective ties to political parties need some additional motivation in order to take part in elections. More extensive knowledge enables people to participate in politics (Delli Carpini & Keeter, 1996;Galston, 2001) by making their political attitudes more consistent, and by promoting support for democratic values, trust in the political system (Galston, 2001;Memoli, 2011) or the sense of political efficacy (Memoli, 2011;Reichert, 2016). As others have already shown, a non-partisan (or 'rational') citizen votes for different and more varying reasons than the emotive one, i.e. partisan (Kuan & Lau, 2002). However, our finding is that knowledge-based ("rational") motivation is clearly weaker than positive affection-based motivation when it comes to explaining electoral participation. Extensive political knowledge, in terms of turnout propensity, cannot fully compensate for the lack of positive party identification. After all, at elections, one has to vote for a party, not for democracy, or civic virtues, or handsome election administrators.
This directly relates to the thesis that, due to cognitive mobilisation, the nature of independents has changed. Described as essentially apolitical in the classic study (Campbell et al., 1960), the independents nowadays are rather apartisan -well-informed citizens able to make their own political judgments without relying on party cues (Dalton, 2007). In these terms, being cognitively mobilised indeed has significant payoffs in terms of electoral turnout, but only in a situation when there are no positive party attachments. No matter how highly cognitively mobilised, those citizens are still underachievers in terms of electoral turnout compared to partisans.
We further reasoned that negative party identification could play its role in the relationship between political knowledge and turnout in the absence of positive party identification. Yet, the data showed that 'negative partisanship', i.e. intense dislike of some party (parties), is almost irrelevant. It is neither an equally strong incentive as the positive one nor changes anything regarding the effects of political knowledge on turnout. This contradicts several previous findings showing that negative party identification is an independent source of turnout (Mayer, 2015;Medeiros & Noël, 2013;Vlachová, 2001). Rose and Mishler (1998) developed a typology of partisanship based on the criteria of positive and negative party identification. They differentiated between negative (only negative party identification), closed (both positive and negative), open (only positive) and apathetic (neither negative nor positive) partisanship. The first one was registered with highest frequencies in numerous post-communist countries (Rose & Mishler, 1998). Our data have shown that, in terms of turnout, there are no great differences in the Serbian political context between the open and the closed type, on the one hand, and between negative and apathetic, on the other. Only among those with the negative and apathetic partisanship does political knowledge make a difference, increasing the probability of election participation. The main reason for that is the fact that what really makes a difference is the positive party identification. Based on the data obtained in Serbia, it seems that the role of negative party identification stressed in the literature in regard to post-communist societies is overrated.
Finally, negative sentiments other than the negative identification might exist as well. Being close to a political party does not only mean caring who wins or loses, but the people who identify strongly (and positively) with a party are generally expected to be more supportive of the political system and political parties as institutions that are necessary for the functioning of democratic governance than the non-partisans (Holmberg, 2003;Miller & Listhaug, 1990). On the other hand, being informed, but non-partisan, could lead to being more demanding in regard to government responsiveness (Neuman, 1986). It could, similarly, cause greater dissatisfaction and criticism in regard to, for instance, 46 PSIHOLOGIJA, 2018, Vol. 51(1), 31-49 the output of the political system (Dalton, 2007), due to perceiving an inadequate representation of one's own interest in the parliament. Consequent dissatisfaction could make some informed citizens more passive and lead to absenteeism.

Limitations and recommendations for future research.
Our measure of negative party identification is only one way of its operationalisation. The other, behavioural attitude (the party one would never vote for), could have shown a different pattern of the relationship with political knowledge and turnout. Furthermore, we included all of the available party (dis)like items; they relate to relevant political parties, but it is a restricted area of possible party dislikes. We have also shown that the relationships between political variables are quite robust and unaffected by the standard set of sociodemographic variables. On the other hand, future research could focus more thoroughly on possible national differences in the observed patterns. Comparing the Western type democracies and Eastern, post-communist countries could be especially relevant in that sense.