LIST OF PANELS

Computerized text analysis is a growing field in the social sciences and humanities (Gentzkow et al. 2019; Grimmer and Stewart 2013; Wilkerson and Casas 2017). Interestingly, while there is an evolving interest in ‘big data’ for policy analysis in general (Giest 2017; Giest and Samuels 2020; Pencheva et al. 2018), there are few empirical applications that actually focus on ‘policy’ as such, let alone ‘policy design’ more specifically (Clinton 2017). As highlighted by Brady (2019) in his review of data science approaches in political science, existing research foci include parliamentary speeches (e.g., Gurciullo and Mikhaylov 2018; Proksch, Lowe, et al. 2019; Proksch, Wratil, et al. 2019), party agendas (e.g., Loftis and Mortensen 2020), or stakeholder statements (e.g., Senninger and Blom-Hansen 2020) that are analysed to detect actors’ policy preferences. Data science approaches have also been employed for assessing the amount of government spending as indicators of (changing) policy preferences (Jones et al. 2003) and for identifying whether policies focus on particularistic issues in policy-makers’ voting districts (Gamm and Kousser 2010). Despite this lack of a clear focus of computerized text analysis to assess legal or policy texts themselves in order to achieve a better understanding of ‘policy’ or ‘policy design’, there is an emerging literature in the subfields of public administration (e.g., Anastasopoulos and Whitford 2019) and comparative legal studies (e.g., Alschner 2020) that aims at leveraging machine-learning approaches for categorizing policies in a general manner: For example, Hurka and Haag (2020) have used computational approaches to quantify the ‘complexity’ of policy proposals. There are also first approaches to tackle the question of ‘complexity’ of political texts, although first applications have focused on political speeches rather than policy or legal texts as such (Benoit et al. 2019).

However, these existing approaches, aiming at broad characterisations of policy or legal texts, are far from being able to engage with or contribute to the research questions that the ‘new’ policy design literature (Howlett et al. 2015) is interested in, namely which design characteristics contribute to the effectiveness of public policies (Knill et al. 2012; Schaffrin et al. 2015). Empirical approaches that rely on hand-coding of policy design characteristics to analyse policy effectiveness (e.g., Schmidt and Sewerin 2019) are not scalable, leading to calls for developing automated or semi-automated approaches for coding such policy design characteristics (e.g., Sewerin 2020; Siddiki et al. 2019). Only very recently have first computational applications been proposed that aim at identifying one specific design feature of legal texts, namely the degree of ‘discretion’ and ‘delegation’ (Vannoni et al. 2020).

This panel seeks to bring together scholars from different disciplines and (sub-)fields that are interested in utilizing computerized text analysis and machine-learning approaches for developing new tools to assess policy design characteristics in a scalable way. We also invite contributions that critically discuss how the production of policy-related data can be scaled up by collaborative or crowd-sourcing approaches (Benoit et al. 2016).

This panel will discuss how computerized text analysis and machine-learning approaches can contribute to a better understanding of policy design. We welcome contributions that demonstrate how such computerized text analysis approaches can be leveraged to analyze policy and legislative texts. Contributions from any discipline are welcome.

We are interested in submissions that

(1)         propose an approach for (semi-)automating the analysis of policy or legal texts, or refine already existing applications or tools

(2)         use such tools to analyze policies or policy design at scale and showcase the potential for using such analysis in new kinds of research projects

(3)         engage with the broader question of how the analysis of policies and policy design can be supported by computerized text analysis and machine-learning.