Estimating policy positions from political texts

TLDR: This article proposed a new hand-coding scheme for policy positions, together with a new English language computer coding scheme that is compatible with this, and applied both schemes to party manifestos from Britain and Ireland in 1992 and 1997 and cross validated the resulting estimates with those derived from quite independent expert surveys and with previous manifesto analyses.

Abstract: The analysis of policy-based party competition will not make serious progress beyond the constraints of (a) the unitary actor assumption and (b) a static approach to analyzing party competition between elections until a method is available for deriving reliable and valid time-series estimates of the policy positions of large numbers of political actors. Retrospective estimation of these positions in past party systems will require a method for estimating policy positions from political texts. Previous hand-coding content analysis schemes deal with policy emphasis rather than policy positions. We propose a new hand-coding scheme for policy positions, together with a new English language computer-coding scheme that is compatible with this. We apply both schemes to party manifestos from Britain and Ireland in 1992 and 1997 and cross validate the resulting estimates with those derived from quite independent expert surveys and with previous manifesto analyses. There is a high degree of cross validation between coding methods, including computer coding. This implies that it is indeed possible to use computer-coded content analysis to derive reliable and valid estimates of policy positions from political texts. This will allow vast volumes of text to be coded, including texts generated by individuals and other internal party actors, allowing the empirical elaboration of dynamic rather than static models of party competition that move beyond the unitary actor assumption. eriving reliable and valid estimates of the policy positions of key actors is fundamental to the analysis of political competition. Various systematic methods have been used to do this, including surveys of voters, politicians, and political scientists, and the content analysis of policy documents. Each method has advantages and disadvantages but, for both theoretical and pragmatic reasons, policy documents represent a core source of information about the policy positions of political actors. We explore various ways to extract information about policy positions from political texts. We are particularly interested in using computer-coding techniques to derive reliable and valid estimates of the policy positions of political actors. This is not mere laziness on our part, a lack of stomach for the hard graft of expert coding. If analyses of party competition are to move beyond both static models and a view of political parties as unitary actors, this requires information on the policy positions of actors inside political parties and on the development of these over time and between elections. The laborious expert "hand-coding" of text is simply not a viable method for estimating the policy positions of huge numbers of political actors, for example, all members of a legislature. Any serious attempt to operationalize a model of internal party policy competition, or of dynamic policy-based party competition or coalition government between elections, implies using computer-coding for estimating the policy positions of key political actors. We first review existing methods for estimating policy positions from political texts. These have for the most part concentrated on the expert coding of party manifestos. We then suggest ways to improve these, dealing with both expertand computer-coded content analysis. We then explore the impact of our suggestions upon estimates of party policy positions derived from British and Irish manifestos issued during the 1992 and 1997 general elections in each country, positions for which a range of