Faculty Profile
Walter Dempsey, PhD
- Associate Professor, Biostatistics
Walter Dempsey is an Assistant Professor of Biostatistics and Assistant Research Professor at the Institute for Social Research. His research focuses on statistical methods for digital and mobile health. His current work involves three complementary research themes: (1) experimental design and data analytic methods to inform multi-stage decision making in health; (2) statistical modeling of complex longitudinal and survival data; and (3) statistical modeling of complex relational structures such as interaction networks. In the coming years, he will continue to design and apply novel statistical methodologies to make sense of complex longitudinal, survival, and relational datasets. This work will inform decision making in health by aiding in intervention evaluation and development. Prior to joining, he was a postdoctoral fellow in the Department of Statistics at Harvard University where he worked within the Statistical Reinforcement Learning Lab under the supervision of Susan Murphy. He received his PhD in Statistics at the University of Chicago under the supervision of Peter McCullagh.
- PhD, Statistics, University of Chicago, 2015
- B.Sc., Mathematics, Statistics, and Economics, University of Chicago, 2009
Research Interests:
My research focuses on statistical methods and theory for digital and mobile health
as well as more traditional areas of biomedical research involving (multi-state) survival
analysis and temporal process regression. My theoretical research focuses on exchangeable
stochastic processes and associated statistical models. My methodological research
focuses on their associated inferential procedures. I believe a great statistician
marries their theory and methods to a strong understanding of the underlying science
of the data they are investigating. I try and work with applied scientists to adjust
my methods, improving our understanding, prediction, and treatment of scientific phenomena
in health.
Research Projects:
Statistical models for complex longitudinal and survival data: Mobile health (mHealth)
technologies are advancing at an accelerated pace, spurring rapid interest in digital
monitoring and mobile intervention. Data arising from mHealth intervention studies
is incredibly complex - mixed-type data collected at different time-scales with multiple
intervention components (i.e., factors) each designed to impact health outcomes over
a (potentially different) time-scale. Indeed, methods to (a) analyze this complex
data and (b) evaluate validity and efficacy of the intervention components lag well
behind scientific interest. My current research interest is in the development of
hierarchical latent variable models for complex longitudinal and survival data. The
work is motivated by two questions: (Statistical Methodology) How should we construct
compact representations of complex longitudinal data informed by scientific theory?
How can we use these representations to inform multi-stage decision making in health?
(Methodological foundations) How does one construct statistical models for state-space
data that satisfy natural invariance principles while reflecting known empirical behavior?
Statistical models for complex relational data: An open question is whether social networks can be harnessed to achieve health behavior change. Such work is in its infancy with modest preliminary results. My prior and current work aims to build statistical network models that satisfy natural invariance properties while simultaneously generating data that correspond with observed empirical properties. My future work will examine two important questions in this area. (Methodological foundations) How does one construct statistical network models for complex relational data that satisfy natural invariance principles while reflecting known empirical behavior? (Experimental design) How do we design a trial to test for nested causal effects of time-varying treatment in the presence of network interference?
W. Dempsey, P. Liao, S. Kumar, and S. A. Murphy, "The stratified micro-randomized trial design: sample size considerations for testing nested causal effects of time-varying treatments", Annals of Applied Statistics (to appear), 2019.
W. Dempsey, A. Moreno, C. Scott, M. Dennis, D. Gustafson, S. Murphy, and J. Rehg, "iSurvive: An Interpretable, Event-time Prediction Model for mHealth", Proceedings of the 34th International Conference on Machine Learning (ICML), 2017.
W. Dempsey, P. McCullagh, "Survival models and health sequences (with discussion)", Lifetime Data Analysis, 24(4):550-584, 2018.
H. Crane, W. Dempsey, "Edge exchangeable models for interaction networks", Journal of the American Statistical Association, Theory and Methods, 2018.
H. Crane, W. Dempsey, "A Framework for Statistical Network Modeling", Statistical Science (to appear), 2019+.
W. Dempsey, P. McCullagh, "Exchangeable Markov survival processes and weak continuity of predictive distributions.", Electronic Journal of Statistics, 11(2):5406-5451, 2017.
M4057 SPH II
1415 Washington Heights
Ann Arbor, MI 48103-2029
Phone: 734-615-9825
Email: wdem@umich.edu
For media inquiries: sph.media@umich.edu
Areas of Expertise: Biostatistics