Understanding Society through Connections

How do local disruptions spread across an economy?

When a factory stops production, its customers may run short of essential inputs and reduce their own output. The effects can reach firms far from the original disruption, including those with no direct connection to the affected factory. Understanding these chains of dependence is essential to assessing economic losses and recovery.

We use observed supplier–customer networks to simulate production, inventories, procurement, and demand at the firm level. Research on the Great East Japan Earthquake shows how network concentration, difficulties in replacing inputs, and cycles of transactions influence the scale and persistence of losses. Incorporating establishment-level damage and power outages improves our reproduction of the initial decline in production after the disaster.

We have extended this framework to restrictions on economic activity and disruptions to international trade. By comparing scenarios, we examine when coordination across regions and changes in sourcing can reduce losses. These are conditional model-based assessments: their purpose is to compare preparedness and recovery options while making the underlying data and assumptions explicit.

What kinds of connections support the creation of new knowledge?

Invention depends on more than individual ability. Who collaborates with whom, where people meet, and which kinds of knowledge they encounter can all matter. We study these relationships using joint patent applications, citations, and the locations of inventors and establishments.

Geographic proximity remains important for collaboration, while links to distant and international partners provide access to diverse knowledge. Repeated collaboration can build experience, yet more repetition does not consistently produce better outcomes. Analyses of Japanese and US patent data show the importance of maintaining diversity in teams and technologies.

Historical evidence offers another way to investigate the role of interaction. Using the 1918 influenza pandemic in Japan, we find a relative decline in patent applications in collaboration-intensive technologies, driven largely by fewer new inventors entering patenting. Opportunities for interaction can therefore matter both for existing inventive activity and for the development of new inventors.

We also investigate the value and use of knowledge. This includes measuring regional economic complexity, evaluating Japan’s SBIR program, and predicting patent licensing with machine learning. In a recent study of 624,129 US patents, screening the top-ranked 10% captured about 90% of observed licensing events. This is a recall measure for prioritizing expert review, rather than a claim of 90% classification accuracy across all patents.

How do economic connections support activity and create systemic risk?

Firms exchange goods, borrow from banks, transfer money, and share knowledge. These relationships form several interconnected networks. We analyze supplier–customer links, bank–firm lending, money transfers, and international trade to understand the organization of economic activity.

Large Japanese production networks contain communities that cross conventional industry boundaries, with both upstream–downstream flows and feedback loops. We develop methods to separate and visualize these structures, investigate how trading relationships form, and study how the productivity of value-chain partners relates to growth. Work with regional bank-transfer data also examines how money flows reflect local economic activity.

Our financial research compares the correlation networks of bank loan portfolios and equity returns, showing that different layers can reveal different structures and dynamics. A multilayer model of cross-country finance and trade further shows how losses passing between layers can expose vulnerabilities missed by single-layer analyses. This work connects the condition of individual institutions with the stability of the wider system.

What can data reveal about epidemics and changing behavior?

Epidemic dynamics, mobility, and purchasing behavior vary across places and over time. We use aggregated location data, case counts, and online purchasing records to investigate patterns that broad averages can conceal.

Research on Tokyo separates population influx into patterns and examines their associations with subsequent epidemic dynamics. This approach does not identify individual transmission events or establish where people became infected; it helps identify mobility patterns that warrant closer attention. For case forecasting, we combine epidemic and time-series models to account for regional differences and changing conditions.

At the international level, analyses of lead–lag relationships among national case series identify two coexisting patterns whose prominence changes between 2020 and 2021–2022. Research on a Japanese online shopping platform also finds that purchasing increases associated with case growth and emergency declarations were temporary during the study period. Distinguishing short-lived responses from lasting behavioral change is a central part of this work.

Using data to investigate mechanisms and compare options

Our starting point is how to observe social systems. We identify the relationships represented in patent, transaction, financial, and mobility data, and study them as networks. Depending on the question, we combine statistical analysis, causal inference, network science, machine learning, and agent-based simulation.

We distinguish observed associations from causal effects, and empirical records from conditional simulation results. Through this combination, we seek to explain mechanisms and inform the comparison of policy and business options.

Research and laboratory videos

Explore further

• Publications
• Student research
• For prospective graduate students