The empirical analyses outlined in the following were based on a hierarchically structured data set with two levels. The second level of the data set is related to the 25 countries in which the 2004 election to the European Parliament was held. If one of these countries had only been composed of a single national constituency in the European election 2004, this country was represented in the second data set level as one case. If, however, a country had been divided into several constituencies, then each of these subnational constituencies represented its own level-2 unit. The latter applied to Belgium (3 constituencies), France (8 constituencies), Italy (5 constituencies), Ireland (4 constituencies), Poland (13 constituencies) and the United Kingdom (12 constituencies). In total, the level-2 data set comprised 64 constituencies, of which 19 corresponded to an entire country.6 For these constituencies there were several level-2 variables available: the number of parties, combined lists and independent candidates running in each constituency, the information about sub-national constituencies (if applicable) as well as data about the respective electoral laws of the constituency (electoral duty, barring clause, flexible candidate lists).

In order to prevent countries with a large number of constituencies from disproportionately influencing the results of the empirical analyses, a weighting variable was also created that weighted the 64 constituencies in such a way that all 25 countries were represented with equal weights.7 The equal weighting of all countries seemed appropriate, since we were particularly interested in the context dependence of the attractiveness effect. Against this background, it may be plausible that the various institutional arrangements should have equal impact on the results of the empirical analysis.

The cases at the first level of the data set represented all parties and combined lists that were running at the European election in the 64 constituencies and that were able to gain at least one per cent of the valid votes. If a party was competing in several sub-national constituencies in a country, it was recorded in the data set multiple times according to its number of candidateships (provided that the party gained at least one per cent of votes in each case). This resulted in a total of 574 cases in the level-1 data set. For each of these cases, the proportion of votes obtained at the European election was recorded for each constituency and added to the data set as a level-1 variable. In addition, the result at the latest general national parliamentary election was included in the data set in each case (where '0' represented the absence of a candidateship at this election).8 We used this variable in our empirical analysis as a kind of omnibus-control variable, since it covers in some way all conceivable influences on the electoral success of the various parties and combined lists in the different countries. Furthermore, it was also recorded whether a party at the time of the European election was represented in the respective national government9, if it had an anti-European program10 and if it was already a member of the European Parliament before the election11 (in each case as dummy variables).

As this article focuses on the influence of front-runners' physical attractiveness on electoral success, each party's or combined list's constituency front-runner was identified and the corresponding details of their sex, age and attractiveness were included in the level-1 data set. While sex and age could be determined with relative ease, the identification of candidates' attractiveness proved much more difficult. As is usual in attractiveness research, portrait photographs of the candidates were used as a basis for measuring attractiveness. These photographs were researched on the internet, and in the case of 47 candidates it was not possible to find a suitable photograph. The number of level-1 units available for the empirical analyses was thereby reduced to 527 (= 91.8 %).

The actual measurement of attractiveness was carried out using the Truth of Consensus Method (Patzer 1985, p. 17). In this, a group of participants (with each participant acting independently) rated a person's attractiveness, and by averaging the individual ratings this provided a measured value of the person's attractiveness. The basis for this process was the aforementioned attractiveness consensus, the observation that a person's attractiveness is a feature which various observers perceive in a very similar way (Grammer et al. 2003; Henss 1987, 1992; Iliffe 1960). Variations in attractiveness rating were therefore essentially attributable to secondary differences in taste between the observers. At the same time this meant that even with a very small, non-representative group of participants a comparable, stable and dependable attractiveness measurement could be achieved. In the literature it is generally seen as sufficient if the attractiveness classification is based on the ratings of two dozen people. The average attractiveness value is then said to be so stable that one is unlikely to achieve a different result even with a large number of participants (Henss 1992, p. 308).

The measurements used in this study of front-runners' attractiveness were based on assessments supplied by 12 male and 12 female sociology students from the University of Cologne, aged between 20 and 26. These students were paid for their participation in the study and were able to rate the candidates' photographs from their home PCs through an online questionnaire. In the survey, each photograph was presented on an individual page and was scaled to the same height.12 The participants received no information at all about the people they had to evaluate. Assessment was carried out via a seven-point scale with the extremes 'unattractive' (coded as '0' in the data set) and 'attractive' (coded as '6' in the data set). An analysis of the reliability of individual ratings confirmed empirically the expected high consensus in the evaluation of attractiveness. According to common practice in attractiveness research, the participants were considered as variables and the photographs as cases. Cronbach's alpha here was 0.95. In addition, each assessor was presented with nine of the pictures twice for the purposes of ascertaining individual response stability. In this case the average Cronbach's alpha for the 24 assessors was 0.83 (with a standard deviation of 0.13).

The ratings of the 24 assessors were calculated according to the Truth of Consensus Method by averaging the attractiveness scores and including the result in the data set as another level-1 variable. The average attractiveness score was 2.05. The lowest value was 0.08 and the highest 5.33. Table 1 represents the attractiveness score of the candidates according to sex and age. There it is shown that attractiveness varied considerably with sex and age, to the effect that young people appeared more attractive than old people, and women more attractive than men. These patterns corresponded to the findings frequently replicated in sociopsychological attractiveness research (for an overview: Henss 1992).

< Zurück   INHALT   Weiter >