It’s not what you say, it’s how much you say it: Comparing Authoritarian and Democratic Constitutions Using Latent Text Analysis

While constitutions regulate political authority within a state, there is much we do not know about the interaction between political conditions and constitutional texts. Much of the literature relies upon hand-coded databases of constitutional texts. However, this approach raises both methodological and theoretical concerns. Advances in natural language processing, particularly latent text analysis, have the potential to provide a more reliable and refined means for coding constitutions. In this paper, I use a Latent Dirichlet Allocation topic model in order to compare constitutional topics in authoritarian and democratic regimes. Unlike previous articles, I find significant differences in the proportion of authoritarian and democratic constitution allocated to certain topics, particularly judicial and legislative institutions. Democratic constitutions are more likely to discuss some negative rights, while authoritarian constitutions are more likely to discuss positive rights. Democratic constitutions allocate proportionality greater text to elections, particularly district size. Finally, authoritarian constitutions are also more likely to allocate text to military affairs.