Jong Hee Park and ByungKoo Kim, 2020. "Why Your Neighbor Matters: Positions in Preferential Trade Agreement Networks and Export Growth in Global Value Chains." Economics & Politics, 32(3). pp. 381-410. [pdf]
Abstract: In rapidly expanding global and regional preferential trade agreements (PTA) networks, policy-makers are keen to situate their countries in a better position, believing that a better position in PTA networks will help their economies trade more and grow faster. In this paper, we provide a theory that explains how changes in countries' PTA network positions affect their trade performance. We argue that a dense and deep “neighbor network” provides a country with a wide access to global value chains, better protection to investment, and strong credibility to their policy commitments. To measure trade performance, we compute value-added exports at the country, year, and industry level across 43 countries, 56 industries, and 15 years (2000–14). The estimation of network position effects is done by panel fixed-effects methods and the sample-splitting and cross-fitting double machine-learning method. The findings show that as a country's neighbors have deeper and wider PTA networks, the country's value-added exports grow faster. Also, the industry-level analysis shows that sectors heavily engaging in the fragmentation of production stages exhibit faster growth with the improvement of neighbor networks.
ByungKoo Kim and Iain Osgood. 2020. "Pro-trade Blocs in the US Congress." The Forum. Vol.17. pp. 549-575. (Invited Contribution). [pdf]
Abstract: Who supports trade in the US Congress? We uncover the ideological space of trade voting, focusing on trade agreements and development policy as two fundamental cleavages around globalization. We then cluster members of Congress into coherent voting blocs, and identify the most pro-trade voting blocs in each Chamber. We find that these blocs: cross party lines; are ideologically heterogeneous; and are over-represented on the committees with jurisdiction over trade. We then examine two leading theories of Congressional voting – on constituency characteristics and campaign contributions – and find support for each using our learned voting blocs. Members of pro-trade blocs have defended their constituents’ and contributors’ interests by speaking out to confront the Trump administration’s protectionism. We conclude that unsupervised learning methods provide a valuable tool for exploring the multifaceted and dynamic divisions which characterize current debates over global economic integration.
ByungKoo Kim, Saki Kuzushima and Yuki Shiraito. "Paragraph-citation Topic Model for Corpora with Citations" [pdf][supplementary information]
Abstract: Topic modeling is one of the most popular approaches to statistical text analysis in many fields, especially in social sciences. An important feature of text data in social sciences is that many corpora consist of document networks in which documents cite other documents. However, existing topic models either ignore the network structure or make simplifying assumptions that do not reflect the structural properties of citation networks. In this paper, we propose a topic model that jointly analyzes both text and citations. In the proposed paragraph-citation topic model (PCTM), topics are assigned to paragraphs rather than tokens. The topic of a paragraph then shapes both the distribution of words and the paragraph’s decision to cite precedents. To model a paragraph’s citation decision, we introduce the latent citation utility that incorporates two established factors of strategic citation: popularity and topic similarity of the precedent. We demonstrate the utility of our model by applying it to two subsets of majority opinions of the Supreme Court of the United States: all opinions on Privacy and Voting Rights issues. We then use the results of the first application on all Supreme Court opinions on Privacy issues to predict the topic structure of the most recent case on reproductive rights, Dobbs v. Jackson Women’s Health Organization.
Stuart Benjamin, Kevin Quinn and ByungKoo Kim. "With Friends Like These: The Recent Rise in Partisan and Gender Differences in Circuit Judges? Treatments of Earlier Opinions." Under Review. [pdf available upon request]
Abstract: Judges shape the law with their votes and the reasoning in their opinions. An important element of the latter is which opinions they follow, and thus elevate, and which they cast doubt on, and thus diminish. Using a unique comprehensive dataset containing the substantive Shepard’s treatments of all circuit court published and unpublished opinions issued between 1974 and 2017, we examine the relationship between judges’ substantive treatments of earlier appellate opinions and their party, gender, and race. Are judges of a particular party, gender, or race more likely to positively treat (that is, follow), and less likely to negatively treat, opinions written by judges who share that attribute than are judges of a different party, gender or race? What we find is both surprising and nuanced. We have two major findings. First, we find no meaningful partisan differences in positively treating opinions in the early years of our study but significant partisan differences more recently. Those differences are sharpest for treatments in ideologically salient categories of cases. The partisan differences arise more for treatments of opinions written by Democratic appointees than for opinions written by Republican appointees, which we think is best explained by an accelerating movement among Republican appointees in a conservative direction compared to a steady move among Democratic appointees in a liberal direction. Notably, the increase in partisan differences is not a function of presidential cohorts or age cohorts. More recently appointed judges and judges appointed decades ago show similar patterns of increasing partisan differences in recent years. And this is not a function of less partisan judges retiring earlier: the recent partisan differences apply when we focus only on judges who served during the same extended period of time. Second, there are within-party gender differences in positively treating opinions similar to the partisan differences: male and female judges of the same party (particularly Democrats) diverge in their likelihood of following ideologically salient opinions written by female and male judges from their party, and those differences have risen in recent years. (We do not find meaningful differences based on race, but that reflects in significant part smaller numbers and thus a lack of statistical power.) These results defy easy explanation. They do not support the proposition that party and gender have always played a pervasive role for judges. But our results provide evidence of increasing partisan and gender-based behavior among judges in recent years. Those partisan and gender differences are best explained by political ideology. Thus it appears that judges’ substantive treatments of opinions have moved from lacking an ideological component to being significantly ideological. Many groups in the United States have recently become more ideologically polarized, and judges appear to be one of them.
ByungKoo Kim and Iain Osgood. "Democracy and Clustered Models of Global Economic Engagement." Under Review. [pdf][supplementary information]
Abstract: One of the most fundamental economic policy choices a society makes is how to order its global economic relations. What models do states use to structure this multi-faceted decision, and how do they choose among these alternatives? We combine data on trade policies, foreign investment, exchange rates, capital flows, and international treaties to discover states’ strategies of global economic engagement. We identify five distinct strategies through dynamic clustering. We then examine the economic and political drivers of states’ choices among these competing strategies, focusing on the tradeoffs between public and private goods activated by differing styles of openness. In particular, we uncover a production-focused and risk-heavy model of global integration favored by non-democracies, and cautious (or insular) models of semi-globalization favored by (large) democracies. Decisions over global economic engagement are clustered and multi-dimensional: uncovering this variety unlocks new findings about the role of democracy in shaping foreign economic policy.
Dynamic Spike and Slab Latent Space Model for Time-series Social Networks
Abstract: Latent space models are popular statistical approach to analyzing various networks. Recent studies extend the latent space models to study the structure of longitudinal networks. This paper contributes to the study of longitudinal networks by presenting a dynamic latent space model featured with recent advances in Bayesian analysis: a continuous dynamic spike and slab prior. More specifically, we build on the latent factor model by Hoff (2009) and assign dynamic spike and slab prior to the connectivity weight. The sparsity-inducing property of the spike and slab prior selects which dimensions of the latent space are active or inactive at a given time period. The set of active dimensions at a given time, or a regime, smoothly transitions over time, governing the evolution of networks. This paper offers two contributions to the existing literature on dynamic latent space models. First, we present an alternative way to model evolution dynamics of networks by focusing on the connectivity weights rather than latent positions. Our model offers a more holistic view on the evolution of networks as opposed to node-based approaches in existing network models. Second, our model provides a data-driven guidance on the dimensionality of the latent space. Conventionally studies using latent space models for networks have set the dimensions of the latent space at 2 for convenient visualizations. This could potentially be a problem when the true dimension is much larger. The sparsity-inducing property of our model informs the researchers how many dimensions of the latent space are relevant in network generation for each time period. We apply our model on the citation network of US appeals court and International trade networks.
Uncovering the Evolution of Legal Doctrines in the US Appeals Court: Hierarchical Latent Space Model for Directed Network with Signed Edges [with Kevin Quinn and Stuart Benjamin]
Abstract: How do we measure the evolution of legal doctrines over time? Existing studies took case-by-case approach to determine a tipping point in time at which legal doctrines shift. In this paper, we propose an empirical model to detect structural breaks in legal doctrines of the US law. Specifically, we focus on the citation networks of US court and examine a structural shift in the pattern of positive and negative treatments of precedents. We obtain from LexisNexis a large-scale data on the Shepardized citation network of all published and unpublished cases in the US appeals court from 1974 to 2017. We then model the Shepardized citation network with hierarchical dynamic latent space model to map cases on a low-dimensional doctrinal space. Then we apply machine learning methods to detect structural change point in the latent doctrinal space for each legal issue area.
Dynamic Biclustering for Dynamic Networks with Directed Edges
Abstract: Dimension reduction for network data has been most actively studied for simple networks with undirected and binary edges. However, many of the real world networks such as international trade networks feature more complex structure with directed and non-binary edges. Dimension reduction methods such as stochastic block models for simple networks cannot easily be applied to more complex networks. In the international capital flows network, for example, heavy capital-senders such as US or UK and heavy capital-receivers such as Luxembourg and Cayman islands should be grouped separately. In this project, I propose the dynamic biclustering approach to provide an effective summary of directed time-series network data. Biclustering aims to cluster rows and columns of the adjacency matrix respectively. Row-cluster, column-cluster and the intersection of those all summarize important information of the given network. I model each row and column cluster to evolve through individual Hidden Markov transitions. For estimation, I propose to employ Classification EM (CEM) combined with greedy algorithm for computational efficiency.
The Role of Economic Decline and Malaise in the Rise of Extreme-nationalist Populism [with Robert J. Franzese, Jr., Diogo Ferrari, Hayden Jackson, Patrick Wu and Wooseok Kim]
Abstract: In recent years, the support for extreme-nationalist populist politicians and parties has grown in developed European as well as in developing Latin American nations. Two “competing” explanations in the literature have been offered to account for the rise of populist, anti-elite, extreme nationalist attitudes: economic malaise and cultural or status threat. We view these two explanations as not at all competing; rather, they are deeply connected and intertwined. In this paper, we argue that individual reactions to economic malaise are shaped by their sociocultural perceptions nurtured in heterogeneous personal and neighborhood experiences. Our theoretical prediction suggests that there will be heterogeneous groups within the samples that vary in their reaction to economic malaise. To gain empirical leverage on these heterogeneous and intertwined causal relations, we employ a novel method called hdpGLM. In our preliminary analyses on the replication of Mutz (2018), hdpGLM confirms the presence of multiple latent clusters in the data that differ in how economic malaise relates to the support for extremist parties.