The current issue and full text archive of this journal is available at www.emeraldinsight.com/0309-0566.htm Towards the development of a cross-cultural model of voter behavior Comparative analysis of Poland and the US Wojciech Cwalina and Andrzej Falkowski Department of Marketing Psychology, Warsaw School of Social Psychology, Warsaw, Poland, and Development of a model of voter behavior 351 Received November 2007 Revised March 2008 Accepted May 2008 Bruce I. Newman Department of Marketing, DePaul University, Chicago, Illinois, USA Abstract Purpose – The purpose of this paper is to reinterpret and test empirically Newman’s model of voter’s choice behavior, where three elements influencing the choice of a given candidate were included: the media’s role in the election; cognitive reasons to vote for; and emotional feelings toward the candidate. Design/methodology/approach – The data from Polish and US 2000 presidential elections have been analyzed and compared. The research purpose is concerned with the measurement of the model’s seven domains of political object evaluation (issues and policies, emotional feelings, candidate image, current events, social imagery and epistemic issues), and its predictive power for the choice of a presidential candidate. The data are analyzed using the methodology of structural equations modeling, and are interpreted with the terms of mutual causal relationships between these domains. Findings – The key factor in influencing voting behavior is evoking positive emotions towards the candidate and then providing voters with a justification for such affect. However, this is especially applicable in an evolving democracy like Poland. In an established democracy like the USA, voters have already learned to more carefully analyze messages from presidential campaigns and they are more resistant to the unconscious power of affect. Furthermore, in the case of challengers, the media exerts a stronger influence on the valuation of their candidacies, but for incumbents, the role of the media is not so pervasive. Research limitations/implications – The main limitation of the results of this study is the non-representative character of the samples used and its size. Research should be carried out on a representative sample as the next step toward a complete verification of the voter behavior model. Practical implications – The reported results make it possible to elaborate more precisely and effectively strategies to segment voters and position candidates in the minds of voters. Political consultants can use the level of electorate cognitive stability to accurately control the media in political campaigns for their candidate. Originality/value – The paper presents an interdisciplinary and cross-cultural approach to analyzing and understanding voter behavior. It offers a baseline to carry out research in disciplines pertinent to political marketing, communication, and psychology. Keywords Elections, Individual behaviour, Cognition, Cross-cultural management, Poland, United States of America Paper type Research paper European Journal of Marketing Vol. 44 No. 3/4, 2010 pp. 351-368 q Emerald Group Publishing Limited 0309-0566 DOI 10.1108/03090561011020462 EJM 44,3/4 352 Introduction Contemporary marketing tools used to control consumers’ as well as voters’ behavior profit to a large extent due to the findings of psychology. Creating the corporate as well as the political brand image requires advanced knowledge and research methodology in the domain of cognitive and emotional processes. In the course of communication between voter and candidate, the image of political reality is formed on the basis of affect and cognition. Thus, to effectively control this image is not an easy task, and psychologists should find their place in the group of political consultants. The candidate or political party image formed in voters’ minds is an important factor determining voter behavior. However, this behavior also depends on some other factors known as cognitive domains, which were developed in the original Newman and Sheth (1985) model of voter behavior. The main objective of this study is to reinterpret and test empirically the model of voter behavior, where three elements influencing the choice of a given candidate were included: Voter’s beliefs about the role of media in elections, cognitive domains, and emotional feelings. All three of these elements influence one another while creating a voter’s cognitive structure, which defines the attitude towards the candidate (Conover and Feldman, 1984; Falkowski and Cwalina, 1999). However, the question arises, what is the structure of this interaction (i.e. what causal relationships exist between those elements)? Is such a structure different, depending upon the level of society’s democratization? Or, do all voters evaluate political candidates in a similar way? Given the different historical experiences and the difference in mentality between the citizens of highly developed and developing countries, it can be assumed that such a cognitive structure may be considerably different. Currently, there are two prevailing approaches to causal relationships of cognitive and affective elements (and their influence on voting behavior). The first approach, called the constructivist approach, stresses the role of a politician’s cognitive domains on the affective attitude toward him. The second approach, known as the realist approach, establishes the relevance of affects in forming the cognitive domains. (Cwalina and Falkowski, 2003; Cwalina et al., 2000, 2005). The primary criterion to differentiate the two approaches is whether perception of affect is direct or not. According to the constructivist approach, the affect is the result of stimuli processed in cognitive structures. According to the realist approach, the affect is the result of direct perception and is independent of information processing (e.g. Buck, 1988; Forgas, 2000; Lazarus, 1991; Zajonc, 1980). This approach is supported by Zajonc’s (1980) cognitive theory of affects. The author, on the basis of a number of empirical studies, demonstrates that an affective reaction evoked by the perceived object appears first, and is only then followed by cognitive elements. In research on political behavior, Wattenberg (1987) found that, as a matter of fact, one-third of the voters know nothing about particular politicians. Despite this lack of knowledge, voters still have strong feelings toward them. Therefore, it is quite reasonable that the affect plays an important role in forming the cognition of politicians. Cwalina et al. (2004; see also Cwalina et al., 2008) empirically tested the constructivist approach using the causal – effect models of voter behavior developed within the structural equation methodology. The research was carried out on presidential election data from Poland and the USA, and parliamentary election data from Slovenia. The comparative analysis of voting behaviors in two countries, Poland (Central Europe) and the USA, will be presented below using the realistic approach. Subsequently, both the constructive and realistic approaches will be compared on the basis of US and Poland voting behavior empirical data. While Poland represents an evolving democracy, the USA is an example of a traditional, well-established democracy. The research of a voter’s cognitive structure was a concept developed by Falkowski and Cwalina (2002) in a methodology of structural equation comparisons. The concept is a development of the model of voter’s choice behavior proposed by Newman (Newman, 2002, 1999; Newman and Sheth, 1985). Structural models of voter behavior following the realistic approach The analysis of structural models of voter behaviors has been presented in detail in earlier publications. Therefore, only the key elements of these models will be presented in this study (see Falkowski and Cwalina, 2002). According to Newman and Sheth’s (1985) model of voter’s choice behavior, the following seven domains are assumed to guide voter behavior: (1) Issues and Policies (refers to the personal beliefs of the voter about the candidate’s stand on economic, social, and foreign policy issues, which represent the rationale for the candidate’s platform). (2) Social imagery (represents the stereotyping of the candidate to appeal to voters by making associations between the candidate and selected segments in society). (3) Emotional feelings (represent the voters’ emotional attitude toward the candidates). (4) Candidate image (refers to the candidate’s image based on personality traits). (5) Current events (refers to issues and policies which develop during the course of the campaign). (6) Personal events (refers to situations in the personal life of the candidate). (7) Epistemic issues (refers to the change-of-pace value a candidate acquires as a result of novelty, curiosity, boredom, or satiation associated with the election process). This model includes an implicit assumption that all of these domains are independent from each other and, potentially, equally important for determining voting decisions. The model then determines a particular approach to the empirical analysis of data: regression analysis or discriminant analysis[1]. However, it is well documented in psychological literature that cognitive and emotional elements should be treated not as separate vectors, but as interactive vectors (e.g. Cwalina et al., 2000, 2005; Falkowski and Cwalina, 2002; Nadeau et al., 1995). Therefore, one can empirically test the question regarding possible causal relationships among the set of variables previously treated as distinct and independent. In particular, this concerns the proper placement in this causal chain of a voter’s emotional attitudes toward candidates. Newman and Sheth’s model does not include the function of the media and election polls as important factors that mediate and oftentimes directly influence voter behavior (e.g. Ansolabehere et al., 1995; Cwalina and Falkowski, 2000; Negrine, 1994). With that in mind, it seems justifiable to supplement the model with a factor including voters’ beliefs about the media’s and polls’ influence on decision making. There is a common view that the media is the “fourth estate” and the supporters of this view stress the Development of a model of voter behavior 353 EJM 44,3/4 354 Figure 1. Structural model of voter behavior I importance of mass media as an element in the political fight and a way of influencing society through this medium. Thompson (1994), trying to define the mutual relations between social development and mass communication, suggests that the media plays an important role in the mechanisms of power. They use symbolic forms while transferring information in order to influence events, making it a vehicle for those who want to use it to achieve particular ideological, economic, or political benefits. The term “media” is understood here as an additional cognitive domain which reflects voters’ beliefs, perceptions or imaginations concerning the importance of the media for political elections in general. Also, the term takes into account the voters’ feelings about the media’s influence on decision making. The developed and re-interpreted model of voter‘s choice behavior may, hypothetically, reflect six primary ways of structural-causal relationships between individual elements. Newman points out that the general idea is that cognitive domains influence voter behavior. However, one should know that according to the realistic approach, these domains are formed in voters’ minds on the basis of affect toward the candidate. Thus, Figure 1 presents the first hypothetical model assuming causal relationship between affect and cognitive domains. Here, the structure of causal relationships is intuitively obvious. The media influences emotional feelings that, in turn, form voter’s cognitive domains that directly influence voter choice. The assumption for this model (namely, that emotions influence the cognitive domains) is based on the above metioned Zajonc’s cognitive theory of affect and the results of Wattenberg’s research on political behavior. One can also assume different combinations of the structural model. It is quite reasonable to assume that emotional feelings exert a direct influence not only on cognitive domains, but also influence the media. Figure 2 presents this hypothesis. It may also be assumed that there exists a mutual interaction between media and emotional feelings. That is, media influences the affect as well as the affect influences the media (as presented in Figure 3). This model assumes that broadcast messages not only evoke the affect, but are also perceived in the context of voter’s emotional feelings. It is well known in psychological literature that evoked emotions cause different perception of the events (e.g. Forgas, 2000). Voters add their own interpretations of the information they receive, depending on their emotional state. Development of a model of voter behavior 355 Figure 2. Structural model of voter behavior II Figure 3. Structural model of voter behavior III Each one of the three presented models can be further differentiated depending on the causal connections between Media and Cognitive domains. Thus, one should assume that Media 1 and Media 2 influence cognitive domains, or cognitive domains influence Media 1 and Media 2, or there is mutual interaction between these variables. An example of the structural equations that underlie model I is presented in Figure 4. Here, we see that in addition to the causal relations presented in Figure 1, it is assumed that there is a mutual and direct influence between the Media and Cognitive domains. That means that the particular shape of cognitive domains in voter’s minds is the result of both the direct influence of broadcast messages and the indirect influence of emotions. On the other hand, the way the broadcast messages are perceived depends on already-formed Cognitive domains. That means that Media and Cognitive domains reinforce one another. Overview of the study In order to test these four hypothetised reinterpretations of Newman and Sheth’s model of voters’ choice behavior the empirical research was conducted in Poland and the USA in 2000 during those countries’ presidential elections. Because the presented research EJM 44,3/4 356 Figure 4. Structural model of voter behavior IV results are an introductory step to determine the causal structure of factors influencing voter behavior, the research samples are non-representative, although based on random sampling. As a consequence the interpretation of the obtained results should be very careful as far as the generalization of the samples to the whole populations of Polish and American voters is concerned. During the presidential elections in Poland, the main candidates running for the office were Aleksander Kwasniewski (the incumbent; supported by leftist parties) and Andrzej Olechowski (the challenger; not connected with any political party, representing center-liberal options). The 2000 US presidential election featured two candidates who faced very different challenges. Al Gore, the Vice-President in the Clinton administration for the previous eight years, was working to separate himself from President Clinton in an effort to create his own identity. His opponent was the candidate of the Republican Party, Texas Governor George W. Bush. Subjects Poland. The empirical research was conducted all over Poland in November 2000, a month after the Polish presidential election. A total of 240 respondents were chosen randomly from different parts of the country. Men represented 46 percent of the sample, while women represented 54 percent. Regarding age, 45 percent of the sample consisted of respondents between 18 and 29 years of age, 30 percent of the sample were between 30 and 44 years, and 25 percent of the sample were above the age of 44. Thus, the research sample over-represents young voters, which limits the generalization of the obtained results to the whole population of Polish voters. Respondents with primary and vocational education represented 13 percent of the sample, those with secondary education represented 59 percent, and those with higher education represented 28 percent. The majority of the respondents came from big cities (53 percent), whereas 32 percent came from small towns, and 15 percent hailed from rural areas. There were only two candidates who were seriously considered in the Polish presidential election in 2000: Aleksander Kwasniewski – the incumbent, and Andrzej Olechowski – the challenger. In the research sample Kwasniewski got 52.9 percent of votes while Olechowski got 34.5 percent. The official results of the Polish presidential election show that Kwasniewski obtained 54.5 percent, and Olechowski got 17.4 percent. Thus, the voters who supported the second candidate were over-represented in the research sample. USA. The study was carried out approximately two weeks before the 2000 US presidential election in Chicago, Illinois. The sample (n ¼ 151) included a slightly younger, better-educated voters than one would find in the population as a whole. Approximately two-thirds of voters in both the Republican and Democratic parties were loyal to their party. While close to 45 percent of voters were very interested in the results of the election, only 2 percent were concerned about the outcome. This last statistic indicates that this was an election that was not very exciting for voters. Respondents were asked to fill out a survey before they voted, and to indicate their intention on the questionnaire. Women constituted 68.5 percent of the sample, and men represented the remaining 31.5 percent. Of the subjects, 39 percent were between 18 and 30 years of age, 22 percent between 31 and 45, and 39 percent over 45. Of the subjects, 14 percent had either secondary or lower education, 60 percent were attending or graduated from a college, and 26 percent had higher education. In the US research sample, Gore received 53.2 percent of the votes while Bush received 43.5 percent of the votes. The official results of the US presidential election show that Bush obtained 271 electoral votes and 47.89 percent of popular votes, whereas Gore received 266 electoral votes and 48.40 percent of popular votes. Method This same questionnaire was used for voters in the USA and Poland. The only differences between particular versions of the questionnaire consisted of including political and social questions specific for particular countries (e.g. membership in the EU, co-operation with secret communist services, or caring about US interests abroad). All the questionnaires consisted of the nine sections: (1) “Background questions” was concerned with political preferences. The subjects marked the candidate they were voting for during the elections. In addition, they defined their own political affiliation, their interest in politics and the political options they supported (from left-wing to right-wing). The question about the candidate the subject voted for was a question directing answers to the other parts of the questionnaire. That is, in answering the following items the subject always referred them to the given candidate and only to him. (2) “Issues and policies” specified the respondent-supporter views of the candidate’s standing on the economy (nine items), foreign affairs (four items) and domestic social issues (seven items). The subjects had to mark whether their candidate supported a given issue or not. (The scale’s Cronbach alpha coefficients were 0.86 and 0.79 in the Polish and US research studies respectively.) (3) “Social imagery” included questions on what support, according to the respondent, her or his candidate received from various social groups (workers, farmers, entrepreneurs, religious believers, women, men, etc.). In the Polish version of the questionnaire, 19 such groups were mentioned, and in the American version – 15 groups. The respondent marked (“yes” or “no”) whether Development of a model of voter behavior 357 EJM 44,3/4 (4) 358 (5) (6) (7) (8) (9) his or her candidate would get support from representatives of a given group. The more groups the respondent marked, the wider social support the candidate enjoyed. (The scale’s Cronbach’s alpha coefficients were 0.77 and 0.82 in the Polish and US research studies respectively.) “Candidate image and emotional feelings” is related to image of and emotion toward the candidate. This section consists of a standard thermometer of feelings (from 0 to 100). In addition, The subjects were asked to mark (“yes” or “no”) whether the candidate they supported had the enumerated personal traits. The Polish questionnaire included 16 such traits, and the American – 12 traits. (The scale’s Cronbach’s alpha coefficients were 0.80 and 0.59, in the Polish and US research studies respectively.) “Current events” was concerned with the respondent’s attitude to possible events, which could change his or her voting decision. The Polish questionnaire included a list of ten such events; the American version, eight. The subject marked “yes” or “no” for every event, meaning whether he could transfer his vote to another candidate or not. (The scale’s Cronbach’s alpha coefficients were 0.83 and 0.85 in the Polish and US research studies respectively.) “Personal events” included ten possible pieces of information from the candidate’s personal life that could change the respondent’s voting decision. The Polish questionnaire included a list of ten such events, whereas the American version had nine. The subject marked “yes” or “no” for every event, meaning whether he could transfer his vote to another candidate or not. (The scale’s Cronbach’s alpha coefficients were 0.78 and 0.67 in the Polish and US research studies respectively.) “Epistemic issues” refers to reasons that would justify the perceived satisfaction of voter’s needs offered by the candidate. The Polish questionnaire included a list of 12 such reasons, whereas the American questionnaire had ten reasons. The subject marked “yes” or “no” for every reason, indicating whether it was the basis of his choice. (The scale’s Cronbach’s alpha coefficients were 0.75 and 0.55 in the Polish and US research studies respectively.) This section of the questionnaire referred to the media. The respondents expressed their own opinions on several five-point Likert scales (from 1 – “strongly agree” to 5 – “strongly disagree”) on the influence of the media and polls on voting decisions and democratic processes. The Polish version consisted of seven items each, whereas the American one of nine. (The scale’s Cronbach’s alpha coefficients were 0.68 and 0.46 in the Polish and US research studies respectively.) Finally, Section IX covers such Demographic information as gender, age, education, occupation, family status, income, and place of residence. Results In order to verify structural models, a methodology based on the analysis of structural equations was used, also called path analysis (see Loehlin, 1987). It is a statistical method that makes it possible to verify hypotheses concerning the structure of causal dependencies in a defined set of variables. The results presented below refer only to model III, which turned out to be the best fit for the empirical data. The research results are presented for each country, respectively. Poland The detailed analysis of the Polish results of fitting the models to the empirical data has been presented in detail elsewhere (Cwalina et al., 2008; Falkowski and Cwalina, 2002). Thus, only the general findings are presented here. In order to simplify the structure of data with reference to the importance of the media in forming voting preferences, we conducted separate principal component analyses on the items from the media section in the questionnaires for Kwasniewski’s and Olechowski’s electorates. We obtained two different factors called “Media 1”, and “Media 2”, respectively. Among Kwasniewski’s electorate, a two-factor solution accounts for 58.07 percent of the total variance. Factor 1 called “Media 1” refers to the “media and polls in elections” and explains 40.97 percent of the variance, whereas Factor 2 defined as “Media 2” may be defined as “money and media in democracy” and explains 17.09 percent of the variance. Among Olechowski’s electorate, two factors were also distinguished which accounted for 43.11 percent of the total variance. Factor 1 (Media 1) refers to “polls and money” in election (27.39 percent), and Factor 2 (Media 2) refers to the “media in elections and democracy” (15.72 percent). These two factors were used in testing individual structural equation models, depending on a given candidate’s electorate. The results of factor analysis on the items from the media section in the questionnaires are presented separately for Kwasniewski’s and Olechowski’s electorates (Falkowski and Cwalina, 2002). The fit of data for models I, II, and III for Kwasniewski and Olechowski was much worse than in the case of model IV. One should also stress that model IV much better explains the data than the models, including causal connections between Media and Cognitive domains added to the second and third model. The goodness-of-fit indices (x 2) for each of the four models are presented in Table I. Because model IV has the best fit (it differs significantly from the fit of the other models), only the results of path analysis for this model will be discussed in detail. Structural equation model IV for Aleksander Kwasniewski. Figure 5 presents the empirical structural equation model IV for Aleksander Kwasniewski. The model goodness-of-fit parameter (x 2 ¼ 59,89; p , 0,001) in comparison to other models (see Table I) allows us to treat this model as an adequate approximation of the empirical data. The model structure shows that there aren’t any significant paths between variables. Thus, it can be stated that the model structure of Kwasniewski’s electorate exhibits little variation. This means that the voters’ attitudes toward Kwasniewski are very stable and solid. Using Blumler and McQuail’s (1968) Aleksander Kwaśniewski x2 p Structural Structural Structural Structural equation equation equation equation model I model II model III model IV 179.67 179.86 153.95 24.89 , 0.001 , 0.001 , 0.001 , 0.001 Andrzej Olechowski x2 p 88.98 89.94 83.59 8.75 , 0.001 , 0.001 , 0.001 , 0.188 Development of a model of voter behavior 359 Table I. Structural equation models’ goodness-of-fit for Aleksander Kwasniewski and Andrzej Olechowski EJM 44,3/4 360 Figure 5. Structural model of voter behavior IV: Aleksander Kwasniewski terminology, the supporters of this candidate have deep-seated or “impervious” attitudes. It is interesting to note that this same conclusion was derived using the constructivist approach (Cwalina et al., 2004), and multiple regression analysis (Falkowski and Cwalina, 1999; Cwalina et al., 2000). Structural equation model IV for Andrzej Olechowski. The results of fitting model IV to the empirical data of Olechowski’s electorate are presented in Figure 6. Thex 2 coefficient (8,75; p ¼ 0,188) is the lowest of all obtained in analyzed models, and allows us to treat this solution as a very good fit to the empirical data. The arrows represent relevant statistical relations between particular elements of the model. A statistically standardized parameter of the path was marked above each of them. For clarity, the arrows connecting the cognitive domains with voter’s behavior, and representing a hypothetical established connection between these variables (see Figure 4), were omitted. In the case of Olechowski’s supporters, the Media 1 directly influences the perceived social support (Social imagery) for the candidate in a positive way. In other words, the perceived role of polls and money in election is higher than the wider social support Olechowski gets[2]. On the other hand, the Media 2 enhances the Candidate image. One should also note that Media 2 exerts an influence on Emotional feelings, which, in turn, influence the Epistemic issues and Candidate image. This means that there is an indirect influence of broadcast messages on cognitive domains where Emotional feelings distort the perceived messages. Furthermore, three cognitive domains – Social Figure 6. Structural model of voter behavior IV: Andrzej Olechowski imagery, Current events, and Candidate images – influence the manner in which Media 1 are perceived. As much as Social imagery and Current events increase the importance of the Media 1, the Candidate image diminishes it. The mutual interaction between Media and Cognitive domains mediated by Emotional feelings creates a picture of voters’ minds consistent with the assumptions of cognitive psychology where it is stressed that a man “goes beyond” the perceived stimuli, constructing, on the basis of them, his or her own world as a cognitive representation of the surrounding reality (Bruner, 1973). Development of a model of voter behavior 361 USA The principal component analysis used for media section items for George W. Bush and Al Gore separately gave two-factor solutions. For the Bush electorate, the factors account for 43.74 percent of the total variance. Media 1 refers to “media and polls’ informativeness” and explains 30.22 percent of the variance. Media 2, called “money and polls’ influence”, accounts for 13.52 percent of the variance. Also, two factors were identified in the Gore electorate and account for 48.7 percent of the total variance. The first one, Media 1, refers to “media in election and democracy” (33.64 percent of variance), while the second one, Media 2, refers to “polls and money in election” (15.06 percent of variance)[3]. As in the case of the analyses on the Polish sample, the structural solutions for models I, II, and III for Bush and Gore do not meet statistical requirements of goodness-of-fit and were much worse than model IV in this regard. The respective goodness-of-fit indices are presented in Table II. Because of the best fit of model IV (it differs significantly from the fit of the other models), only the results of path analysis for this model will be discussed in detail. Structural equation model IV for George Bush. Figure 7 presents structural equation model IV for George W. Bush. Its fit to empirical data is the best from the tested models (x 2 ¼ 12,66; p ¼ 0,049, see Table II). What is characteristic here is the presence of only one positive connection between Media 2 and Emotional feelings. A lack of connections between the other variables means that as in Kwaśniewski’s situation in Poland (see Figure 5), that the voters’ attitudes toward Bush are very stable and impervious to any influence. The supporters of this candidate are already very confident of their decision about for whom to vote. Structural equation model IV for Al Gore. The structural model IV for Al Gore, presented in Figure 8, is more complex compared to the model for his opponent. The fit of this model to empirical data is very good (x 2 ¼ 2,68; p ¼ 0,848; see Table II). We observe two connections where Emotional feelings exert an influence on the Candidate image and Issues and policies, and two connections where Media 2 influences Current events and Candidate image. The belief of his supporters in the George W. Bush x2 p Structural Structural Structural Structural equation equation equation equation model I model II model III model IV 53.59 52.82 45.43 12.66 0.003 0.003 0.01 0.049 x2 56.81 56.80 54.81 2.68 Al Gore p 0.001 0.001 0.001 0.848 Table II. Structural equation models’ goodness-of-fit for George Bush and Al Gore EJM 44,3/4 362 Figure 7. Structural model of voter behavior IV: George W. Bush Figure 8. Structural model of voter behavior IV: Al Gore importance of the polls and money in elections increases together with the increase of these both cognitive domains. The lack of influence of Media on Emotional feelings means that American voters present a coherent approach to voting behavior and are relatively resistant to affect. The perceived broadcast messages are not distorted by evoked emotions. Thus, the voter’s mind separates the emotional feelings from the messages that directly influence some of the cognitive domains. That means that the base of voter behavior is rather rational than emotional. Comparative analysis of the structure of voter behaviors in Poland and the USA Cognitive stability of the electorates The results of the research on fitting the model of structural equations (according to the realistic approach) to voting behavior in Poland and the USA leads us to the following conclusions. First of all, every candidate, irrespective of the country he comes from, has an individualized pattern of the structure of factors influencing his support among the voters. This inference is similar to that which was derived from the analysis of voter behavior according to the constructive approach, and it demonstrates the difficulty and limitations connected with building universal models of voting behaviors (Cwalina et al., 2004). On the other hand, there appears to be an interesting structural pattern of voter behavior that is similar in Poland and the USA. It was observed that Kwasniewski in Poland and Bush in the USA have considerably less significant paths between variables in the tested models than do their opponents (compare the Figures 5 and 7 and Figures 6 and 8). Having considered this, it is apparent that the electorates of these former candidates are more stable and resistant to any influence from outside broadcast messages and internal cognitive domains and affect. It means that it is more difficult to switch them to another candidate. One should note that Kwasniewski was the incumbent candidate. So, his voters had already established feelings and a well-formed image that enabled them to keep psychological distance from problems covered in such cognitive domains as Issues and policies, Current events, or Personal events. As the challenger Olechowski’s voters are more sensitive to their own candidate than incumbent voters are to their own candidate, one should expect the better fit of the model for the first one than for the second. Figure 9 presents the x 2 fitness parameter for both candidates in Poland and in the USA. It’s clear that the structural equation model significantly better fits the challenger than the incumbent in Poland (x2difference (1) ¼ 16,14, p , 0,001) and better to Bush than Gore in the USA (x2difference (1) ¼ 9,98, p , 0,01). Therefore, it seems useful to propose the so called cognitive stability index of the electorate consisted of both, the number of significant paths between variables of the structural equation model, and the model goodness-of-fit parameter expressed by x 2. Because both Kwaśniewski in Poland and Bush in the USA won their respective elections, one should conclude that the cognitive stability index of the electorate, developed here within the framework of structural equation methodology, might be a useful sign allowing us to predict the outcome of election. The importance of x 2 coefficient in such predicting is presented in Figure 10 which in a slightly different way presents data of Figure 9. The goodness-of-fit model parameter is much better for the winner, the stronger candidate, than for loser, the weaker candidate, in Poland as well as in the USA. Development of a model of voter behavior 363 Comparison between realist and constructivist approach Cwalina et al. (2004) analyzed voter behavior using the structural equation modeling according to the constructivist approach. The difference with realistic paradigm is that Figure 9. Goodness-of-fit parameters ðx 2 Þ of structural equation model IV for candidates in Poland and in the USA EJM 44,3/4 364 Figure 10. Goodness-of-fit parameters ðx 2 Þ of structural equation model IV for winner and loser in Poland and in the USA Figure 11. Structural constructivist model of voter behavior Emotional Feelings do not precede, but follow the cognitive domains, thus directly influencing the voter behavior. An example of the constructivist structural model of voter behavior best fitting the data is presented in Figure 11. This same empirical data used for both realist and constructivist approaches enables us to make a direct comparison of the tested models. Figure 12 shows the fitting parameter x 2 of the best models selected in realist and constructivist paradigms, and Table III shows the models’ significance for this second paradigm. The results show that from the statistical point of view, the goodness-of-fit (x 2) of the constructive model is not satisfactory for Poland, while it is quite acceptable for the USA. However, the realist model is satisfactory for the USA and partly satisfactory for Poland (see Tables I and II). Therefore, it seems obvious that the realistic paradigm better explains vote behavior than the constructivist one, which means that cognition of the candidate is already “colored” by affect. This is especially applicable in Poland, which means that the key factor in influencing voting behavior is evoking positive emotions toward the Development of a model of voter behavior 365 Figure 12. Constructivist and realist structural models fitting parameters ðx 2 Þ USA Bush Gore Poland Kwaśniewski Olechowski x2 df p 7.12 7.73 8 6 0.52 0.26 30.37 19.10 6 6 , 0.001 , 0.005 candidate and then providing voters with a justification for such affect. That means that some of the cognitive domains directly influencing voter behavior are already distorted by emotional feelings. In the USA, however, one would say that both approaches are relevant. This means that in an established democracy like the USA, voters have already learned to more carefully analyze messages from presidential campaigns. These American voters are more resistant to the unconscious power of affect than are voters in evolving democracies, such as Poland. It is interesting to note that this same conclusion was reached by testing the sequential model of the influence of political advertising on voting behavior in Poland and the USA (Cwalina et al., 2005). Authors applied both the realist and constructive approaches to test whether the emotional attitude influences the candidate image or, on the contrary, the emotional attitude controls the candidate image while the voter perceives political ads. The used analyses of structural equation methodology, consistent with present findings, showed that the realist worked better than the constructive paradigm for Poland, and both are applicable for the USA. Such a result allows us to state that Poles are more emotional and less rational than Americans in their voter behavior. Final remarks The presented analysis shows the importance of Newman’s model of voter’s choice behavior (Newman, 1999; Newman and Sheth, 1985; see also Cwalina et al., 2008) Table III. Structural equation constructivist models’ goodness-of-fit for Poland and the USA EJM 44,3/4 366 because the cognitive domains distinguished by Newman allow us to predict very accurately the support for individual candidates. Although the model was developed within the framework of structural equation methodology specifying the causal relations between variables, and not all of the cognitive domains are relevant to all the cases, it, nevertheless, offers a sufficient frame for making such predictions. However, one should approach the obtained results with caution, due to both the non-representative character of the samples, and the time when the data were collected. Research should be carried out on a representative sample as the next step toward a complete verification of the voter behavior model. The important finding was the difference in the role of media and its influence on the election process. In the case of weaker candidates, challenger Olechowski in Poland as well Gore in the USA, the media exerts an important influence on the valuation of them both. A significant mutual influence of Media 1 and Media 2 with some of the cognitive domains (see Figures 6 and 8) was observed. On the other hand, in the case of stronger candidates, Kwasniewski in Poland and Bush in the USA, the media is not active. Voters’ formed cognitive domains are resistant to any broadcast messages (see Figures 5 and 7). Such a result is expressed here by the concept of cognitive stability of the electorate, and the structural equation statistics allows us to give a precise value to this stability. From a marketing point-of-view, political consultants can use the level of electorate cognitive stability to accurately control the media in political campaigns for their candidate. It requires them to make an effort to understand the specificity of voter behaviors toward the candidate and the role of situational factors (media and polls, current events, etc.). The results presented here provide a strong test for the specificity of social cognition of the surrounding political reality. The slight advantage of the realist paradigm over the constructivist one is consistent with the Zajonc’s (1980) cognitive theory of affect explaining why emotions precede cognition, and Wattenberg’s (1987) statement that although most of the voters know nothing about particular politicians, they still have strong feelings toward them. From the perspective of political marketing, the reported results make it possible to elaborate more precisely and effectively strategies to segment voters and position candidates in the minds of voters. Cognitive domains and their interactions in the reinterpreted Newman and Sheth model of voting behavior allow us to focus on those elements which have real importance for the voters in a particular election, and in a particular country. Thus, marketing strategies can be developed more effectively to reach different voter segments. An excellent starting point for such segmentation strategies is the classification of voters proposed by Newman (1999): . Rational voters, who correspond to the social domain of Political issues, referring to the problems and directions of social and political actions; . Emotional voters, who are particularly sensitive to Candidate image and his/her Personal events; . Social voters, for whom Social imagery is very important, and who vote for a particular candidate associated with a particular social group (e.g. the affluent, educated, national minorities or religious); and . Situational voters, who are very sensitive in their choices to anything that has or might have happened recently (Current and Epistemic issues). In the context of the presented research results, it is possible to identify which of these segments is dominant and critical to achieve election success. Notes 1. Discriminant analysis was used for empirical verification of the model of voter’s choice behavior during the US primary elections in 1980 (Newman and Sheth, 1985), presidential election in 1996 (Newman, 1999), and presidential election in 2000 (Newman, 2002). 2. Note that the higher the factor result for the media, the smaller the significance attributed to it. It is connected with the construction of the scale, where 1 – “strongly agree” and 5 – “strongly disagree”. This note is particularly important for interpreting the parameters of path analysis. 3. Detailed results of principal component analysis for the USA are available from the authors. References Ansolabehere, S., Iyengar, S. and Simon, A. (1995), “Evolving perspectives on the effects of campaign communication”, Research in Political Sociology, Vol. 7, pp. 13-31. Blumler, J.G. and McQuail, D. (1968), Television in Politics, Faber and Faber, London. Bruner, J.S. (1973), Beyond the Information Given. Studies in the Psychology of Knowing, Norton and Company, New York, NY. Buck, R. (1988), Human Motivation and Emotion, John Wiley, New York, NY. Conover, P.J. and Feldman, S. (1984), “How people organize the political world: a schematic model”, American Journal of Political Science, Vol. 28, pp. 95-126. Cwalina, W. and Falkowski, A. (2000), “Psychological mechanisms of political persuasion: the influence of political advertising on voting behavior”, Polish Psychological Bulletin, Vol. 31, pp. 203-22. Cwalina, W. and Falkowski, A. (2003), “Advertising and the image of politicians. National elections in Poland, France, and Germany”, in Hansen, F. and Christensen, L.B. (Eds), Branding and Advertising, Copenhagen Business School Press, Copenhagen, pp. 205-31. Cwalina, W., Falkowski, A. and Kaid, L.L. (2000), “Role of advertising in forming the image of politicians: comparative analysis of Poland, France, and Germany”, Media Psychology, Vol. 2, pp. 119-46. Cwalina, W., Falkowski, A. and Kaid, L.L. (2005), “Advertising and the image of politicians in evolving and established democracies: comparative study of the Polish and the US presidential elections in 2000”, Journal of Political Marketing, Vol. 4, pp. 19-44. Cwalina, W., Falkowski, A. and Newman, B.I. (2008), A Cross-cultural Theory of Voter Behavior, Haworth Press/Taylor & Francis Group, Binghamton, NY. Cwalina, W., Falkowski, A., Newman, B.I. and Verčič, D. (2004), “Models of voter behavior in traditional and evolving democracies: comparative analysis of Poland, Slovenia, and US”, Journal of Political Marketing, Vol. 2, pp. 7-30. Falkowski, A. and Cwalina, W. (1999), “Methodology of constructing effective political advertising: an empirical study of the Polish presidential election in 1995”, in Newman, B.I. (Ed.), Handbook of Political Marketing, Sage Publications, Thousand Oaks, CA, pp. 283-304. Falkowski, A. and Cwalina, W. (2002), “Structural models of voter behavior in the 2000 Polish presidential election”, Journal of Political Marketing, Vol. 1, pp. 137-58. Development of a model of voter behavior 367 EJM 44,3/4 368 Forgas, J.P. (ed.) (2000), Feeling and Thinking: The Role of Affect in Social Cognition, Cambridge University Press, Cambridge. Lazarus, R.S. (1991), Emotion and Adaptation, Oxford University Press, New York, NY. Loehlin, J.C. (1987), Latent Variable Models: An Introduction to Factor, Path, and Structural Analysis, Lawrence Erlbaum, Hillsdale, NJ. Nadeau, R., Niemi, R.G. and Amato, T. (1995), “Emotions, issue importance, and political learning”, American Journal of Political Science, Vol. 39, pp. 558-74. Negrine, R. (1994), Politics and the Mass Media in Britain, Routlege, London. Newman, B.I. (1999), “A predictive model of voter behavior: the repositioning of Bill Clinton”, in Newman, B.I. (Ed.), Handbook of Political Marketing, Sage Publications, Thousand Oaks, CA, pp. 259-82. Newman, B.I. (2002), “Testing a predictive model of voter behavior on the 2000 US presidential election”, Journal of Political Marketing, Vol. 1, pp. 159-73. Newman, B.I. and Sheth, J.N. (1985), “A model of primary voter behavior”, Journal of Consumer Research, Vol. 12, pp. 178-87. Thompson, J.B. (1994), “Social theory and the media”, in Crowley, D. and Mitchell, D. (Eds), Communication Theory Today, Polity Press, Cambridge, pp. 27-49. Wattenberg, M.P. (1987), “The hollow realignment: partisan change in a candidate-centered era”, Public Opinion Quarterly, Vol. 51, pp. 58-74. Zajonc, R.B. (1980), “Feeling and thinking: preferences needs no inferences”, American Psychologist, Vol. 35, pp. 151-75. Corresponding author Wojciech Cwalina can be contacted at: wcwalina@swps.edu.pl To purchase reprints of this article please e-mail: reprints@emeraldinsight.com Or visit our web site for further details: www.emeraldinsight.com/reprints