아래와 같이 초청특강을 개최하오니 많은 참석 바랍니다.

 

1. 일시 : 2026년 7월 20일(월), 오후 3시~

 

2. 장소 : 통계학과 스마트강의실 (자연대연구실험동 222호)

 

3. 연사 : Prof. Dipak K. Dey (Department of Statistics, University of Connecticut)

 

4. 연제 : Power-Law Regression with Mixed Effects for Self-Organized Criticality

 

Abstract

 

  Extreme Value Theory (EVT) has long served as a fundamental framework for modeling rare and catastrophic events through extrapolation from observed exceedances. However, for systems exhibiting Self-Organized Criticality (SOC)—including power outages, earthquakes, wildfires, extreme precipitation, and storm intensities—the underlying generative mechanism is often governed by a power-law distribution rather than the classical EVT paradigm. In such systems, the tail behavior follows a linear decay on a log-log scale, where a single power-law exponent, , governs both moderate and extreme events across the entire support of the distribution.

Despite the central role of the power-law exponent in determining the frequency and severity of extremes, statistical methodologies that incorporate covariate effects into power-law behavior remain limited, particularly in settings where the exponent itself may vary across environmental, spatial, or temporal conditions. To address this gap, we propose a mixed-effects power-law regression framework in which the exponent parameter is modeled as a function of covariates. The proposed hierarchical structure incorporates both fixed and random effects, allowing partial pooling across related groups so that data-sparse regions can borrow strength from the broader population. This yields a flexible and context-sensitive modeling strategy for understanding how external drivers influence the occurrence and magnitude of extreme events.

Through extensive simulation studies and an application to extreme precipitation data across the United States, we demonstrate that the proposed approach provides substantially improved tail inference for SOC-driven phenomena. In particular, the model assigns scientifically meaningful probabilities to recent record-breaking events, including Hurricane Milton and the 2024–2025 California atmospheric river events, which conventional EVT-based approaches often regard as effectively impossible. These results highlight the importance of SOC-based statistical modeling for understanding complex systems characterized by cascading dynamics and heavy-tailed behavior.