Michael Escobar

Professor and Interim Chair
Department of Public Health Sciences
University of Toronto

Department of Statistics (cross appointment)
University of Toronto

Contact Information:

Phone: (416) 978-8989 (voice mail, but email is better)
Fax:(416) 978-8299
email: m(dot)escobar(at)utoronto(dot)ca

Overview

Research Interest

My research has been a combination of theoretical and applied statistical research. You can click here to see my curriculum vitae. At the end of this web page, I give more information about my research interest here. My major theoretical and applied areas:

• Developing new methods to compute Dirichlet process models for nonparametric Bayesian methods.
• Developing mixture models using both frequentist and Bayesian techniques and applying these methods to model population heterogeneity.
• Applying statistical methods in psychiatric research, especially in the areas of schizophrenia, learning disabilities, and suicide.
• Applying and improving statistical techniques in the medical, biological, and public health sciences. Many of these projects are in conjunction with students and colleagues at the University of Toronto and places where I have previously worked.

Teaching

I teach and advise students primarily in the MSc/PhD program in Biostatistics
I also work with many students in other areas (such as epidemiology) on there statistical problems.

This year I am teaching an introductory course in applied Bayesian methods.

In previous years I had taught

• The introductory biostatistics course (CHL 5201 Introductory Biostatistics for Students in Biological Sciences I). The web page for this course is here.
• Categorical Data analysis (click here).
• Longitudinal Models course (Year 1, Year 2, Year 3).

Consulting

I have been a statistical consultant for a number of groups in a wide variety of areas. Please see my cv for an updated accounting of my experience.

Research notes:

Dirichlet processes, Nonparametric Bayesian Methods

• Here is a paper that I wrote with Mike West which contains my preferred method to compute these models. (Abstract, paper w/o figures, figures).
• Here is a directory that contains some Ugly code written in FORTRAN. You might find it helpful in starting your own code.
• Here is some recent work with George Tomlinson on a project that we call AnDe. This stands for Analysis of Densities. The basic idea is that instead of studying an observation for each person, we study the density of observations that each person has. The model look very similar to the Normal-Normal random effects model but instead of Normal distributions, we have Dirichlet processes. The slides of this talk presented at the SSC meeting and Joint Statistical meetings are here.

Computing Bayesian models

Here is a talk I gave at an FDA meeting which overviews Markov chain Monte Carlo methods. The talk is missing the overhead with computing output that was added to the talk later.

In August of 2003, I gave the course: "Introduction to Applied Bayesian Methods". Here and here are the slides for the graphs in that course.