Department of Statistics and Actuarial Science,
University of Waterloo
CANADA N2L 3G1
(519) 888-4567 Ext. 35543
Fax: (519) 746-1875
Email: Mary Thompson
Office: Math & Computer 6117
Rearch and Scholarly Interests
At present Professor Thompson works primarily in survey methodology. Through much of her career she worked on problems of inference in sampling theory. The early focus was on theorems establishing the optimality of certain design/estimation strategies under models for the survey population responses. She became very interested in the problem of the role of the design randomization in inference in the presence of models. Her book Theory of Sample Surveys contains a careful treatment of what ultimately came to be her view of the role of randomization, and the rationale for what are commonly called “‘model assisted” techniques. It also contains a systematic approach to using estimating functions in surveys, and a thorough discussion (with examples) of the role of the sampling design when survey data are used for analytic purposes. This has implications for the use of survey weights in analysis.
Another theme of her research has concerned estimation for partially observed Markov or semi-Markov models. Some of her more recent work in this area has had to do with the analysis of partially observed binary stochastic sequences. With former students R.J. Rosychuk and K. Shum, she has studied the impact of misclassification error. This has led to an increased understanding of identifiability and estimability issues associated with misclassification. With G.Y. Yi, she has worked on the treatment of missing values in longitudinal sequences.
These themes come together in aspects of inference from complex longitudinal surveys. Issues in the design of longitudinal surveys to support causal inference are central to work on the International Tobacco Control Survey, for example. Professor Thompson is also studying the application of multilevel models and longitudinal models with time-varying covariates to complex survey data, where the best ways to adapt the estimating functions systems for use with survey weights are still not well understood.
- Godambe, V. P. and Thompson, M. E. (2009) Estimating functions and survey sampling. To appear in C. R. Rao and D. Pfeffermann, Eds. (2009) Handbook of Statistics. Sample Surveys: Inference and Analysis (Volume 29b). Elsevier, 669-687.
- Pantoja-Galicia, N., Thompson, M. E. and Kovacevic, M. (2009) Assessing the temporal association of events using longitudinal complex survey data. In P. Lynn, Ed. (2009) Methodology of Longitudinal Surveys. John Wiley & Sons, Ltd, 333-350. (more)
Professor Thompson is Co-Director (with Professor John Goyder of Sociology) of the University of Waterloo Survey Research Centre. She is a member of Statistics Canada’s Advisory Committee on Statistical Methods. She has been active in the Statistical Society of Canada, serving as President in 2003-04. She is particularly interested in promoting collaboration among statisticians and health and social scientists. Since 2002 Professor Thompson has been involved with the International Tobacco Control project led by Professor Geoffrey Fong of Psychology.
Stat 890- Statistical Issues in Sample Survey Design and Analysis