Shifting Constitutions and Democratic Backsliding

• In Event: Democratic Erosion

Abstract

Procedures within constitutions guide how legislatures are organized, how laws are enacted, and, in general, they structure democratic governance. We suspect these types of constitutional reforms impact democratic outcomes, but we also acknowledge that reforms do not occur at random. Therefore, we investigate the precursors of constitutional reforms such as meta-features of constitutions (age, amendment procedures, and institutional structure), along with the political, socioeconomic, and demographic characteristics of the country that also predict changes to the quality of democracy. To study this, we use machine learning prediction methods to learn what combinations of characteristics and changes to provisions in constitutions lead to democratic backsliding. We use data from V-DEM and the Comparative Constitutions Project. To discover the proper set of characteristics and provisions, we set up a two-stage design: first, we determine which pre-reform characteristics predict changes to the constitutional system and, second, we discover the types of procedural changes in constitutions that predict democratic backsliding. We then apply ensemble machine learning methods to discover which combination of relevant characteristics and constitutional provisions affect our democratic outcome variables most directly. This allows us to identify and test for heterogeneous treatment effects from both groups.

Authors

• Kiran Rose Auerbach, University of Bergen

• Joshua Yoshio Lerner, Northwestern University

• Kristen M. Renberg, Duke University