Intra-party diversity and ministerial selection in coalition governments

We refrain from discussing the role of intra-party factions for portfolio allocation in coalition governments in more detail. This is not only because we do not have the information on faction membership of possible ministers for all cases under study here. In addition, Ceron (2014a) and Bernauer and Bräuninger (2009) show that computerized scaling models like Wordscores (Laver et al. 2003) or Wordfish (Slapin and Proksch 2008) are doing a very good job in correctly locating faction representatives, so that the main independent variable here—the policy distance to the dominant principal—should cover factional membership of potential ministers.

Ideological proximity is an expectation that is line with both intuition and powerful theoretical arguments. Yet situations may arise when other criteria that may conflict with proximity temporarily gain in importance (e.g., signalling economic competence and/or independence from partisan positions; see, e.g., Bäck et al. 2009; Hallerberg and Wehner 2012; Alexiadou and Gunaydin 2014). In such situations (e.g., economic or judicial crises) the pool of appointments for which proximity considerations are important might shrink, but the principle should still be valid for the remaining positions.

We here use the database created by Julian Bernauer on the full text of speeches in the 15th German Bundestag (Bernauer 2006; see also Bernauer and Bräuninger 2009).

One further, often discussed problem of the Wordscores approach is that the standardization proposed by Laver et al. (2003), conducted in a way that the mean score and the standard deviation of the reference and the virgin texts are congruent, is not robust. When adding one or more further virgin texts to the set of texts to be scored, the estimated positions of all virgin texts will change. Martin and Vanberg (2008a, b) developed a standardization method for Wordscores estimates. Their method makes it possible to handle the standardization problem that arises when applying the technique by Laver et al. (2003). The Martin and Vanberg (2008a) rescaling approach does not depend on the standard deviation of all texts that are of interest. Martin and Vanberg (2008a, p. 99) propose to select two reference texts as “anchor texts” which should cover, first, a wide range of words and, secondly, should reflect the most extreme positions on the policy dimension under study. These two reference texts are then used to transform the raw scores of the other reference texts. This set of reference texts is then, in a second step, used to estimate the positions of the virgin texts (Benoit and Laver 2008, p. 105). As Martin and Vanberg (2008a) put it, researchers who use their standardization technique face a trade-off: “any increased accuracy in the word dictionary that is gained by adding reference texts must be purchased at the expense of some degree of internal consistency” (Martin and Vanberg 2008a, p. 99). Faced with this choice, we follow the suggestion by Benoit and Laver (2008, p. 110) to use the transformation by Laver et al. (2003) since more than two high-quality reference texts are available. This choice is also necessary to cover the wide array of the ideological spectrum in the parliamentary debates of the three countries under study. Other transformation strategies or estimation techniques like Wordfish (Slapin and Proksch 2008) might produce different scores for the parliamentary speeches. Since Wordscores estimates—based on the transformation technique by Laver et al. (2003)—of left–right positions of parties in several European countries correlate highly with the left–right positions of parties according to the estimates of the Comparative Manifesto Project (Volkens et al. 2013) according to recent studies (Bräuninger et al. 2013), we apply Wordscores here to measure policy positions of individual MPs.

We include a control variable in our multivariate analyses that identifies those MPs who did not give a parliamentary speech related to the respective policy dimensions. In doing so, we try to account for the problem identified by Proksch and Slapin (2012), who suggest that MPs who did not deliver a speech in parliament are probable defectors from the party line.

Given the high party unity scores in parliamentary systems, one might critically argue that the high degree of heterogeneity within parliamentary parties that Figs. 1, 2 and 3 show is not realistic and can change when applying a different estimation procedure. However, MPs can have incentives to deviate strongly from the party line in parliamentary speeches to relate to the specific economic situation or the policy attitudes in the electoral district they represent. Such behaviour is less costly than voting against the party line and suitable for sending out signals to the citizens in the electoral districts the MPs represent (see for empirical evidence, e.g., Debus and Bäck 2014; Baumann et al. 2015; Bäck and Debus 2016, pp. 128–135).