The Real (Lack of) Difference between Republicans and Democrats: A Computer Word Score Analysis of Party Platforms, 1996–2004

Abstract

For years voters and political pundits have grumbled about the lack of real choice between Republicans and Democrats. Scholars have examined party behavior and suggested reasons for concern. Determining whether there is a real ideological and policy difference between U.S. political parties, and the nature of that difference, is important for political science and for democratic politics generally. Ultimately, democracy is about choices, and where choices are few, democracy is degraded. One way to examine the choices that political parties offer voters is by assessing their political platforms. Even in an era of candidate-centric politics, political party platforms spell out the general programs offered by the parties, and the platforms are heavily influenced by the policy positions of the candidates themselves. In addition, the political platform is the one document that spells out the entire program of the party. Individual candidate speeches capture only snippets (at best) of the policy choices offered to voters. Recent innovations in computerized content analysis make it possible to analyze large bodies of text such as party platforms in a systematic way by treating words as data, then analyzing them statistically.

Detecting policy preferences and dynamics in the UN general debate with neural word embeddings

“…As for supervised learning, early applications include the Wordscore model (Laver, Benoit and Garry, 2003), a Naive Bayes classifier that categorizes text across a one-dimensional spectrum, where the ends of the spectrum are learnt by training on the bag-of-word features of a set of anchor corpora. The model has been used to track the ideological differences between US Republicans and Democrats (Kidd, 2008), detect conflict in political preferences (Monroe, Colaresi and Quinn, 2008), and interest group influence (Klüver, 2009). Other supervised methods, such as random forests and support vector machines, have been applied to classify political text into one or more ideological dimensions.…”

The Promise and Pitfalls of Automated Text-Scaling Techniques for the Analysis of Judicial Opinions

“…Among the techniques that have emerged over the last decade, two latent trait scaling algorithms, Wordscores (Laver et al, 2003) and Wordfish (Slapin and Proksch, 2008) have proved remarkably popular with political scientists. Scholars have used Wordscores to estimate party positions from Irish, British, French, US and Danish election manifestos (Laver et al, 2003, Klemmensen et al, 2007, Benoit and Laver, 2003, Laver et al, 2006, Kidd, 2008 and to measure the preferences of delegates at the Convention on the Future of Europe (Benoit et al, 2005). Wordfish, meanwhile, has variously served to measure the content of electoral pledges in German and Japanese elections (Slapin and Proksch, 2008, Proksch and Slapin, 2009, Proksch et al, 2011; to map the policy stances of interest groups at EU level (Klüver, 2009); and to analyse thousands of EU Parliament speeches (Proksch and Slapin, 2010).…”

Left, Right, and Meeting in the Middle: Addressing Addiction is Something we can Agree About

“…Thus, comparing and contrasting party platforms should provide voters with useful information as they decide who they will vote for come Election Day. Investigators have developed methods for systematically comparing party platforms and other forms of political text, [5][6][7][8] although such reviews have never been carried out with regards to addiction issues to our knowledge.…”