Second-Order Saliency Theory - A Theory of Issue Misclassification for Text Scaling

Text scaling is a commonly used and empirically successful method of estimating political positions. It is unclear if the use of words recovers anything politically meaningful beyond document similarity. This approach has little theoretical basis, except saliency theory (ST), which states we can infer position from issue emphasis alone. However, positional statements matter for policy. Positions can be communicated by being pro or contra on political issues. Second Order Agenda Setting aims to integrate agenda setting and framing by adding a second topic to the issue. Following the substantive interpretation of the classification error approach we utilize the uncertainty of machine learning models. We identify the second most likely topic of text as frames. We scale them to estimate relative positions. We test this by predicting the positionality of text as well as scaling positions from positional, non-positional, and second-order issue emphasis. We bridge the Comparative Manifesto Corpus with the Comparative Agendas Project annotations of the same manifestos and demonstrate that this additional information allows position estimation of RILE positions from CAP codings alone. This allows applying position measurement to a wide range of text types and establishes ST as a theoretical foundation for text similarity measures.