In three papers, I consider two questions of nationalization in American politics, and one question of the methodology necessary to study them.
Nationalization is the process by which local politics become more like national politics on the basis of political issues and electoral engagement. It is usually measured using the difference in presidential and state-level electoral returns over time. To expand the study of nationalization, I use automated content analysis to derive new measures for the phenomenon’s study based on political text. In particular, I apply automated content analysis via latent dirichlet allocation to code for salient topics in text from national political agenda speech, local agenda speech, and state laws. The primary source for these local agenda codes is a novel database of State of the State addresses, which are like presidential State of the Union addresses, but are delivered by governors. I developed the database over the past seven years as part of this dissertation; it draws from all 50 States, and the earliest captured addresses date to the year 1893. The secondary sources for these codes are the State of the Union addresses and a corpus of laws passed by state legislatures. I utilize the codes from these naturally distinct text corpora to study the nationalization of the political agenda, and how nationalized elections relate to the production of salient laws. The comparison of naturally distinct texts, however, is problematic and requires further examination. To that end, the first paper, “A Theory and Method for Pooling Naturally Distinct Corpora” concerns the theory and method for why we should be able to use, pool, and compare the computer-generated codes from these naturally distinct text corpora to study nationalization. I propose a theoretical framework with which the researcher may defend the pooling of corpora, and introduce an empirical approach to testing for absolute comparability, the delta-statistic. While statistics like the Akaike Information Criterion (AIC) and penalized log likelihood can help the researcher to determine if a model fits the pooled corpora better than the corpora separately, the delta-statistic relies on a strong theory of latent traits to evaluate the absolute quality of a pooled model. This is especially important when it is impossible to evaluate ground truth fit because some data are unlabeled.
The second paper, “Have State Policy Agendas Become More Nationalized?” examines whether the nationalization of state policy agendas is related to the nationalization of gubernatorial elections. The analysis shows that State agendas, as laid out in the State of the State addresses, have become more similar to each other over time. It also shows that State agendas have become more similar to the national agenda, as laid out in the State of the Union addresses. Finally, I demonstrate an increasing relationship between the similarity in the agenda and the nationalization of elections. The findings suggest that the nationalization of the agenda is a significant and related factor to the nationalization of elections.
The third paper, “Can States Govern Effectively When Politics Are Nationalized?” considers the question of whether electoral nationalization moderates the relationship between divided government and legislative productivity in the states. I find a null effect of divided government on salient lawmaking ability, and that nationalization of state legislatures has generally decreased the production of salient laws. The result holds even though nationalization is unrelated to the ability of our state governments to take action on salient issues during times of divided government. The findings suggest that behavioral factors driving lawmaker decisions may be more to blame for lawmaking defects than institutional ones.
Taken together, the essays demonstrate the value of text analysis to the analysis of nationalization and other research topics in American politics.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/d8-wf0x-6k97 |
| Date | January 2020 |
| Creators | Sutherland, Joseph L. |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0025 seconds