On a Short Leash? Sub-National Party Positions between Regional Context and National Party Unity

There are other aspects of party politics that can be used as indicators of sub-national parties' autonomy. Deschouwer (Citation2006: 294–295) provides examples such as the party's strategy (under which he subsumes campaign tactics, coalition building and the formulation of election manifestos); the selection of candidates for elections; the representation of sub-national parties in higher boards of the party; and their financial independence (see also Laffin et al., Citation2007: 88–108).

In combining both perspectives, this article also extends recent analyses of party positioning that implicitly model local (Bækgaard & Jensen, Citation2012) or regional parties (Bräuninger & Debus, Citation2012: 187–201) as autonomous actors, thereby neglecting the interdependence between different levels of policy-making and possibly existing intra-party constraints.

Although the relevance of single goals may vary between parties, it seems reasonable to assume that most established parties take all of these goals into account when making important decisions.

The interdependence of parties' actions also introduces methodological problems, as it makes it hard to identify the actual reason behind a particular decision.

One reason for the uncertain political environment is the complexity of voting behaviour, meaning that “no party knows how its moves will really affect voting” (Budge Citation1994: 450).

I would like to thank an anonymous reviewer for drawing my attention to this point.

Moreover, Laver Citation(2005) claims that it is possible to comprehend the preferences of Irish voters regarding two parties if their positional change is modelled according to the heuristics stated above.

Budge et al. Citation(2010) make a similar argument with regard to party factions.

Empirical evidence regarding leapfrogging is somewhat inconclusive (see Budge, Citation1994; McDonald & Budge, Citation2005: 64–73). However, differences appear to exist between states and party systems.

The CSU is an independent party that only runs for elections in the state of Bavaria. However, in this article, the CSU is analytically perceived as a sub-national party that is closely linked to the regional party branches of the CDU. It should be noted here that disregarding the CSU does not significantly change the results.

For a discussion of the promises and pitfalls of the Wordscores method, see Benoit and Laver (Citation2007a, Citation2008); Budge and Pennings (Citation2007a, Citation2007b); Klemmensen et al. Citation(2007); Lowe Citation(2008); Martin and Vanberg (Citation2008a, Citation2008b).

For a comprehensive discussion of the data used in this article, see Bräuninger and Debus (Citation2012: 45–52).

The use of the selected source of reference scores – and expert survey data on parties' positions in general – is not without difficulties (see, for example, Budge, Citation2000; Curini, Citation2010; McDonald et al., Citation2007). However, alternative reference scores either refer to the beginning of the time under study (e.g. Huber & Inglehart, 1995) or give rise to other specific problems (e.g. differences regarding the meaning of left and right between countries and over time; uncertainty regarding the quality of (left–right) scores because of the unreliable and possibly wrong classification of text by human coders), as is the case with measures that are based on CMP data (see, for example, Benoit & Laver, Citation2007b; Franzmann & Kaiser Citation2006; Mikhaylov et al., Citation2012). Furthermore, given the focus on a particular time in a single country some of the problems that accompany the use of expert survey data should be less relevant in this context.

Empirically the choice of reference scores seems to be less crucial here anyway. The positions of the virgin texts were re-estimated with four additional specifications: I varied the source of the reference scores (Huber & Inglehart and CMP) as well as the “reference elections” (1990, 2002 and both at the same time). If we correlate the new estimates with the positions used in this article we get the following coefficients: 0.636, 0.823, 0.812 and 0.88. Interestingly the values are often higher than those obtained in a cross-validation based on (German) national parties left–right positions (see Jahn, Citation2011: 757; Keman, Citation2007: 82).

This approach is essentially taken because the manifestos of smaller parties do not meet the requirements of Wordscores, meaning that they focus only on a few issues and/or are extremely short.

These are the PDS's regional branches of Baden-Württemberg and Rhineland-Palatinate.

The West German regional branches of the PDS and its successor parties are the main exception: for these sub-national parties, only one or two election manifestos were available. The only other exception is the Brandenburgian CDU, which did not publish a manifesto in the run-up to the 1990 state election.

In some cases a 12-point scale was used. In these cases, the self-placements were transformed.

Other studies on policy shifts (e.g. Adams & Merrill, Citation2006; Adams & Somer-Topcu, Citation2009) use the Kim-Fording measure of voter ideology (see Kim & Fording Citation1998, Citation2003) to capture the public opinion. The advantage of this measure is that it is very easy to apply: the calculation of the “median voter position” is based on the distribution of party positions and the parties' vote shares (Kim & Fording, Citation1998: 79–80). Leaving aside technical problems, I do not use this measure because it does not seem appropriate to assume that party positions and voters' preferences are that strongly connected in a situation in which it is at least possible that other factors which could have the opposite effect – in this case intra-party factors – outweigh the impact of the electorate.

Summary statistics of all variables can be found in Table 3 (Appendix).

In 1990, East Germans were not asked which party they leaned towards. In this case, the vote intention was used to calculate the party supporters' preferences.

The unemployment rate used is the annual mean share of the unemployed in the working population (“Anteil der Arbeitslosen an den zivilen Erwerbspersonen”).

Most studies that analyse party position changes tackle serial correlation by introducing a lagged dependent variable.

The tables were produced using the estout package for Stata (Jann, Citation2005, Citation2007).

It could be argued that that the relationship between the state of the economy and parties' positions is in fact conditional on the parties' ideology (see Adams et al., Citation2009). However, further tests show that the effect of unemployment on policy shifts does not differ between parties.

The effect remains positive if alternative operationalizations – the non-weighted mean position shift of all other parties or the mean position shift of the adjacent parties – are employed.

This relationship is extremely robust and statistically significant across different sets of party positions that are estimated using different specifications (other reference texts and/or reference scores). A brief presentation of the sensitivity analysis is given in Appendix B in the supplementary material.