2015-2016 WRF and Moore/Sloan Postdoctoral Fellow

Specializations
Astronomy & Physics, Neuroscience, Oceanography, Software Engineering
Tsuyoshi Kunihama was a Moore/Sloan Data Science and WRF Innovation in Data Science Postdoctoral Fellow from 2015-2016. He is currently an Assistant Professor at Nagoya University in Japan.
Biography
Assistant Professor, Nagoya University, Japan – started April 2016
Moore/Sloan Data Science and Washington Research Foundation Innovation in Data Science Postdoctoral Fellow, 2015-2016
Department of Sociology
UW mentors
Samuel Clark, Sociology
Tyler McCormick, Statistics and Sociology
Education history
Ph.D. Candidate, Department of Statistical Science, Duke University M.S., Economics, University of Tokyo, 2009
B.A., Economics, University of Tokyo, 2007
Research goals
My main research goal is to propose flexible Bayesian statistical models, which can incorporate various characteristics in social surveys and induce additional studies for deeper understanding of structures of societies. To understand the development of societies and monitor trends, extensive surveys have been conducted by private and government research institutions over a number of years. These surveys contain demographic, behavioral and attitudinal questions to investigate characteristics of population and subpopulations of special interest. The multivariate response variables are usually mixed-scale with highly complex dependencies. It is typical for these data sets to be high-dimensional, mixed-scale, time-indexed and heterogeneous with complex dependencies and massive missing values. The sample is collected via complex survey designs for which adjustments need to be incorporated into statistical models. For my research, I will develop joint modeling of large-scale social surveys. Associations among response variables in the population can be estimated in this framework. Also, the joint models can flexibly induce various conditional models because the conditional density can be expressed as a ratio of two joint densities in general. Hence, conditional associations for subpopulations can also be inferred.