The extent of clientelism in Irish politics: Evidence from classifying Dáil questions on a local-national dimension

From AcaWiki
Jump to: navigation, search

Citation: Richard Sinnott, Niall O'Reilly, Sarah Jane Delany (2010) The extent of clientelism in Irish politics: Evidence from classifying Dáil questions on a local-national dimension. Proceedings of 21st Irish Conference on Artificial Intelligence and Cognitive Science (AICS 2010) (RSS)
Internet Archive Scholar (search for fulltext): The extent of clientelism in Irish politics: Evidence from classifying Dáil questions on a local-national dimension
Download: http://arrow.dit.ie/dmccon/31
Tagged: Computer Science (RSS) politics (RSS), automatic text classification (RSS), Dáil (RSS), Ireland (RSS)

Summary

This paper uses supervised machine learning to determine whether Irish elected representatives' questions are more local or national.

It draws on previous work on manual content analysis from A study of parliamentary Questions in Dáil Éireann used as training data and uses a holdout set also classified manually.

Context

Representatives, known as TD's, may question ministers about public affairs. "Question Time" is the name for this process of submitting questions in writing, for either oral or written answer. About 5 hours a week are allocated to responding to these questions in the Dáil Éireann (lower house).

Research Procedures

Parlimentary questions (along with responses and supplementary questions) from September 1922 through December 31, 2008 were collected from the parlimentary debates archive. Thus personal questions, which were anonymized, were assumed to be local. Training data was a random sample from a subset of the years (1957-1973), but a holdout set of 68 questions was from randomly selected from all years.

They used Weka's SMO classifier (an SVM implementation). After experimentation, they chose a bag-of-words representation, removing stopwords and words that appeared in fewer than 500 of the questions in the corpus.

Results

Identifying national questions was easier than identifying local questions, in part because the proportion of local questions in both manually-identified sets was low.

Figure 1 shows that questions volume has grown considerably, with over 20,000 questions asked in some recent years, compared to fewer than 2,000/year until the 1960's. 1984 stands out as an unusual year, with nearly 80% local and personal questions. In most years, these percentages are within the 40-60% range.

The authors comment that the emphasis on local topics "shows substantial correspondence" with periods when the Fáil Fianna Fáil, which typically leads, were in opposition.

Full-text data sources available from the Oireachtas Éireann (national parliment)

Selected References

Theoretical and Practical Relevance

The authors relate this to the Comparative Agendas Project which is developing systematic indicators to compare what issues are at stake in various countries' political systems.

Did removing low-frequency words contribute to the difficulty in identifying local questions?

A second content analysis, Using parliamentary questions to measure constituency orientation: An application to the Irish case, is mentioned but apparently not used, presumably because it was published after work commenced.