Search
Browse By Day
Browse By Time
Browse By Person
Browse By Mini-Conference
Browse By Division
Browse By Session or Event Type
Browse Sessions by Fields of Interest
Browse Papers by Fields of Interest
Search Tips
Virtual Exhibit Hall
Change Preferences / Time Zone
Sign In
X (Twitter)
Measuring policy positions of political parties is an endeavor partaken by political scientists from various subfields. Overall, most common data sources for position scaling include party manifestos (for instance, the Comparative Manifesto Project), expert surveys (e.g. Chapel Hill Expert Surveys), voter surveys, legislator surveys, roll-call vote data, and lately parliamentary texts as data. Despite the incredibly wide range of options for measuring party positions, a few core problems have not been exhaustively dealt with. First, common methods for measuring party positions often try to reduce the multiple policy issues into one or a few latent dimensions and assign substantive meanings to them (Laver, 2014).There are obviously sound arguments for this reduction technique: for starters, it has been shown in one case that increasing the number of dimensions to more than two does not explain much more variance in the original 20-issue data set (Benoit & Laver, 2012).
That being said, the allocation of policy issues into latent dimensions relies on either a design pre-determined by the researcher (a priori) or on the data available (a posteriori). Both approaches indirectly assume the same thing: that policy issues making up one latent dimension are independent of policy issues making up the other dimensions. This assumption is questionable, knowing inter-party and intra-party negotiations are much more complicated than that. In fact, parties often make gains and losses when negotiating with other parties across policy issues and over time. There is not a generic “economic” issue that a party puts on the table to trade for a generic gain in the “social” domain. Accordingly, it would be logical to expect party positions on one policy issue to be influenced by their positions on other, perhaps more important, policy issues, be they in the same higher-order latent dimension or not.
Secondly, efforts to measure party positions hitherto have been overshadowed by the quest to capture ideal, or, to quote Benoit and Laver (2006), “sincere” positions of parties. Although learning about parties’ ideal positions could prove useful for descriptive purposes, the end goals of measuring party positions are often explanatory. For instance, partisan politics scholars work out party positions to explain policy changes, party competition scholars wish to understand how parties adjust their policy positions to attract more voters, while coalition theorists aim to untangle strategies parties implement to negotiate with other parties and form coalitions. None of said objectives can be wholly fulfilled through analyzing absolute party positions.
This article proposes an alternative approach to measuring party positions, one that shifts the focal point to parties’ bargaining spaces and how these relate to each other in a competitive political arena. By definition, on each policy issue and for every party, a bargaining range is the distribution of positions that the party manifests through it representatives toward that specific issue. The aggregate, or joint probability distribution, of all bargaining ranges (covering the entire policy space) make up the party’s bargaining space. The form, shape, and statistics of each party’s bargaining space informs us about intra-party, cross-faction struggles and settlements. Meanwhile, interactions among bargaining spaces offers a unique outlook over the multiple strategies and compromises parties make to reach an agreement.
To demonstrate this theoretical concept, I analyze parliamentary speeches of legislators from the German Bundestag from 1991-2017 using quantitative text analysis. Parliamentary speeches are preferred over manifesto data or roll call votes in this case because bargaining spaces are collective in nature – meaning they require multiple actors’ positions, and because intra-party variations can be better observed through legislative speeches than roll-call votes (Schwarz et al., 2017). To categorize legislative speeches, I use annotated texts of the Comparative Agendas Project, which breaks policy agendas down into 21 broad topics, as the training sample for a state-of-the-art DistillBERT machine learning model fine-tuned for text classification. Then, using a neural network that combines bags-of-words with word embeddings, legislators’ positions on each of the 21 policy issues will be scaled. Subsequently, a party’s bargaining range on one issue, comprising its legislators’ scaled positions, will be allowed to covary with other bargaining ranges from the same party, as well as other parties’ bargaining ranges in several multilevel structural equation models. Results from these models, especially when the time dimension is also considered, could reveal parties’ internal and external struggles and compromises, things that might eventually explain how policies come to fruition.