Collusion to Conflict: A Quantitative Analysis of Japanese Labor Politics

• In Event: New Methods in Comparative Politics and International Relations

Abstract

Summary:
I make use of advances in quantitative text-as-data tools to analyze the evolution of labor politics in Japan from the 1990s to today. Using a variety of data sources---including hundreds of deliberative council records, parliamentary transcripts, and newspaper articles---I show that labor policy became both more contentious and politically salient as the locus of policymaking migrated from the labor ministry to the prime minister's office. In addition, I demonstrate important changes in labor unions' and political parties' policy positions that came about as a result of changes in the policy process. Finally, I highlight how these quantitative analysis tools might be usefully employed by qualitative scholars of comparative politics and public policy.

Background:
Quantitative text-as-data tools have proliferated in recent years. This latest wave of quantitative techniques also requires deep qualitative and/or area studies knowledge to meaningfully implement the methods and interpret the results. Be it tweets or parliamentary debates, sentiment analysis or content analysis (topic models), scholars need a nuanced understanding of the language and its use-in-context. Moreover, they need access to the text itself, a task often facilitated by a Rolodex of in-country contacts.

Using labor market regulation as a case study, I show how these tools can be usefully applied to the study of Japanese politics and policy making. Postwar Japanese labor policy was traditionally decided within tripartite deliberative councils sponsored by the labor ministry, but in the 1990s the policy initiative shifted to parliament and the prime minister's office. In the former system, labor unions and employers bargained directly over policy. In the latter, political parties set the policy agenda. How did labor unions, employers, and political parties respond to these changes in the labor policy process? In particular:

1) Did deliberative council meetings and parliamentary debate become more contentious over time?
2) How often and in what context did policy actors talk about labor market "insiders" (lifetime employees) versus "outsiders" (dispatch, part-time, and contract workers)?

Data and methods:
I collected thousands of records from deliberative council meetings, parliamentary debates, and periodicals put out by employers and labor unions stretching back to the 1980s. To assess contentiousness in debates over time I used several varieties of sentiment analysis (dictionary methods, document scaling, and Google's NLP API). To assess the salience of labor market insiders / outsiders in debates, I relied on topic models (STM) and correspondence analysis.

Results:
I found that after as labor policy initiative shifted out of the ministerial deliberative councils debate became more contentious. Unions in particular became markedly negative in the deliberative councils. At the same time, labor policy began to receive more attention in parliament, and debate became more contentious. Moreover, I found that parties and unions on the far left devoted much more attention to (non-unionized) labor market outsiders than their centrist counterparts. However, the attention gap began to close in the mid-2000s. Furthermore, as the opposition Democratic Party of Japan became a credible electoral threat, the long-ruling center-right Liberal Democratic Party pivoted sharply away from deregulation, attenuating the contentiousness of debate.

This research contributes to a growing body of scholarship in comparative politics blending deep area studies knowledge with novel text-as-data techniques. In addition, to my knowledge, this is the first analysis to use deliberative council meeting transcripts, and the first to systematically describe the evolution of expressed labor union policy positions vis-a-vis labor market outsiders since the 1990s. It is also the first to make use of Google's natural language processing (NLP) tools to validate results from classical machine learning techniques as applied to Japanese.

Author

• Nathan B. Cisneros, University of California, Irvine