Negotiating the analog mainstream with digital methods in hand, by Pertti Ahonen, University of Helsinki, Finland
1 like449 views
This document discusses negotiating between digital and analog research methods in political science. It outlines experiences examining Finnish political party platforms and government programs digitally from 1880-2012 and 1917-2015 respectively. Key findings include: 1) Wordfish and structural topic modeling reveal changing emphases over time from "traditional" to "modern" issues in party platforms. 2) Government programs covered nation-building, economic development, and later welfare policy. 3) Digital methods provide new insights but must be carefully integrated with theory and interpretation to be persuasive to traditional political science. Hybrid research combining methods is most promising.
Negotiating the analog mainstream with digital methods in hand, by Pertti Ahonen, University of Helsinki, Finland
- 1. NEGOTIATING THE ANALOG MAINSTREAM WITH DIGITAL METHODS IN HAND VISIONS FROM THE EXAMINATION OF PARTY PROGRAMS AND GOVERNMENT POLITICAL PROGRAMS Dr. Pertti Ahonen, Professor, Faculty of Social Sciences, University of Helsinki, Finland, pertti.ahonen@helsinki.fi ESS Digital Sociology Mini-Conference, Boston, March 17–20, 2016 Session 43, March 17, Thursday, 3:30 – 5:00 p.m.
- 2. Outline 1 Conditions for digital research to negotiate with the analog mainstream 2 Experiences from examining party programs and governmental political master programs 3 Conclusions Febuary - March 2016 (c) Pertti Ahonen 2016 2
- 3. Personal research background A long-time interest in traditional and newer qualitative and quantitative methods for examining texts Article in Big Data & Society (open access, 2015) Pertti Ahonen, “Institutionalizing Big Data methods in social and political research”, at http://bds.sagepub.com/content/spbds/ 2/2/2053951715591224.full.pdf Febuary - March 2016 (c) Pertti Ahonen 2016 3
- 4. 1 Conditions for digital research to negotiate with the analog mainstream 1.1 Mainstream background See https://manifestoproject.wzb.eu • A 35-year international project, MANIFESTO/ MARPOR • Covering about 35 countries worldwide • Data on political party programs since World War II • Acquired by means of manual coding by 1 to 2 coders, according to a 54-item codebook • Managed at WZB, Berlin Social Science Research Center • Important emphasis on placing political parties in the right-left dimension • Has generated a rich body of research (for a bibliography, see the above website) Febuary - March 2016 (c) Pertti Ahonen 2016 4
- 5. 1, cont’d • The MANIFESTO/MARPOR manual coding has been contested by digital scholars • First, Michael Laver, now with NY University, developed supervised latent trait scaling (R/Wordscore program) • Later, Proksch and Slapin developed unsupervised latent trait scaling (R/Wordfish program, see the R/ Austin library of Will Lowe; http://conjugateprior.org/software/austin/) 1.2 Conclusions • The mainstream (e.g., the MANIFESTO/MARPOR veteran Ian Budge of Essex) has found it hard to accept the digital challenges • The mainstream has often denied that the digital methods bring improvements • The digital alternatives possibly accepted as a source of technical correctives only Febuary - March 2016 (c) Pertti Ahonen 2016 5
- 6. 2 Experiences from examining political party programs and governmental political master programs 2.1 A 2015-2017 project • Funded by the Kone Foundation of Finland • Directed by Pertti Ahonen, 2-3 doctoral research associates, research aide • Examining by means of digital (computational) methods: 1. Modernization of vocabulary, topics (and evaluative sentiments) in: • political party programs in Finland (1880-2012) • government master political programs in Finland (1917-2015; announced by each government at its start) 2. The influence of the party programs upon the government programs (a demanding extension; not in this presentation) 3. Discourse quality by means of DQIs, Discourse Quality Indexes (not in this presentation) 4. Special analyses with smaller datasets (not in this presentation) Febuary - March 2016 (c) Pertti Ahonen 2016 6
- 7. 2, cont’d 2.2 Achievements thus far and challenges ahead 2.2.1 Data • 734 party programs (1880-2012), acquired from the Finnish National Social Science Data Archive (see http://www.fsd.uta.fi/en/; logo to the right), preliminarily examined by March 2016 • Prime Minister’s Office, 74 government political master programs (1917-2015), preliminarily examined by March 2016 2.2.2 Methods • The R/Wordfish unsupervised latent trait scaling program: http:// www.wordfish.org/uploads/1/2/9/8/12985397/wordfish_manual.pdf • The newly evolved R/Structural Topic Modeling (STM) program: https://cran.r-project.org/web/packages/stm/vignettes/ stmVignette.pdf • An R/sentiment analysis program, to use later (not in this presentation) Febuary - March 2016 (c) Pertti Ahonen 2016 7
- 8. 2, cont’d 2.2.3 Experiences and preliminary results • Specificity of the structural and temporal context • Finland applies since 1906/1907 a proportional electoral system that catalyzes a multi-party political system • 5-6 so-called effective parties • The left-right dimension has changed many times over, and • has more lately become hard to distinguish • Note the long time perspective (party programs, 1880-2012, government political programs, 1917-2015) • Crucially reduces the rationale of trying to examine the left-right dimension • Rather suggests the examination of temporal change (”modernization”) in the programs examined Febuary - March 2016 (c) Pertti Ahonen 2016 8 2006 Commemo- rative coin: Universal suffrage in Finland (1906/7)
- 9. 2, cont’d • Picks from the results • Examination of political party programs in Finland • Wordfish • Beta values for each different word (about 25,000), distinguishing ”traditional” and ”modern” vocabulary elements (not in this presentation) • Omega values for each document (party program), indicating the position of each party in a ”traditional-modern” dimension at each point of time during 1880-2012 (provided that the party has existed at each such point) • Figure 1: the position of selected political parties and their programs in the ”traditional-modern” dimension over time Febuary - March 2016 (c) Pertti Ahonen 2016 9
- 10. 2, cont’d (Figure 1) Febuary - March 2016 (c) Pertti Ahonen 2016 10
- 11. 2, cont’d • Party programs, structural topic modeling (STM) • STM cannot (unfortunately according to some) be put in the driver’s seat, as this would give a very high number of topics hard to intepret • Equally as in mainstream research, the researcher must decide on a tractable number of topics on the basis of theoretical interest • Compare, e.g., with certain entrenched classification methods • Hypothesis: The topics reflect big issues in Finnish politics since early nation building to later welfare state elaboration and onwards • For an example, see Figure 2: five topics, 1880-2012, each identified on the basis of their foremost vocabulary elements: • 1: Nation building: human rights, law, and state action in general • 2: Basic economic development: Land reform and the elaboration of taxes and taxation • 3: Public services: Organizing the municipalities to provide for education and other services for children, the youth, and others • 4: Advanced economic development: Concern with the environment, energy supply, and balanced regional development • 5: Europeanization: Integration with Europe but preserving gender equality and the country’s inherited nature • Many refinements of the analysis have been possible and necessary (not in this presentation) Febuary - March 2016 (c) Pertti Ahonen 2016 11
- 12. 2, cont’d (Figure 2) Febuary - March 2016 (c) Pertti Ahonen 2016 12
- 13. 2, cont’d • Examination of government political master programs in Finland • The results are harder to intepret than the results on the political party programs • Wordfish results (not in this presentation) • Structural Topic Modeling (STM) results • The data were first examined in their entirety (1917-2015); for an example with three topics, see a solution with three topics, each identified on the basis of their foremost vocabulary elements, Figure 3: • 1: Nation building: stabilizing the independence won in 1917, preserving democracy, managing the war economy (1939-1948/52) • 2: Basic economic development: employment and economic policy- making • 3: Economic and social policy: welfare legislation, pensions, institutional reform in the public sector • A possible refinement to the above: from linear to curved estimates (not in this presentation) • The data were alsoexamined in subsets according to cutoff points in Finland’s political history (not in this presentation): • 1917-1944 (Finland’s exit from World War II) • 1944 -1986 (start of the dissolution of Soviet influence in Europe) • 1987-2015 Febuary - March 2016 (c) Pertti Ahonen 2016 13
- 14. 2, cont’d (Figure 3) Febuary - March 2016 (c) Pertti Ahonen 2016 14
- 15. 3 Conclusions Challenges from the mainstream to digital researchers • Generally,”more is better” does not suffice, i.e., that the digital methods enable the examination of ever bigger data • It is undesirable that the digital methods researchers become mere technical methdologists supporting the research mainstream • The argument that digital exploration replaces explanation and interpretation is unlikely to convince many mainstream representatives • Quantitative and qualitative mainstream insists on a sophisticated formulation of the research problem and the research questions, and giving explicit answers to these questions • Posing and testing hypotheses remains influential in mainstream research • The mainstream is unlikely to buy the argument of certain digital researchers that explanation and controlled interpretation be rejected in favor of mere exploration Febuary - March 2016 (c) Pertti Ahonen 2016 15
- 16. 3, cont’d What the digital scholars could and should do? • Keep one foot in the analog mainstream, the other in the digital stream • Either become an analog and a digital scholar at the same time, • Or participate in projects in which both aspects are present, either with an analog or digital role, or with both roles • Promote hybrid research, both with digital and other, including traditional or newer qualitative and quantitative methods • Learn enough of good statistical theory and research, both classical and newer • E.g., Bayesian as opposed to ”frequentist” statistics • E.g., the common history of traditional classification methods (such as principal components analysis and factor analysis) and newer digital methods of classification (such as topic modeling) Febuary - March 2016 (c) Pertti Ahonen 2016 16