|Date:||Thursday, March 23, 2017|
In many scientific disciplines, it is common to find large number of studies addressing the same research question of interest. Meta-analysis can be used for combining or contrasting the results from these multiple studies. We develop a Bayesian approach for meta-analysis using Dirichlet process. The key aspect of the Dirichlet process in meta-analysis is the ability to assess the evidence of statistical heterogeneity in the underlying effects across studies while relaxing the distributional assumptions. Assuming that the study effects are generated from a Dirichlet process, the study effects parameters have support on a discrete space and enable borrowing of information across studies while facilitating clustering among studies. We also extend the approach for binary data in the presence of excessive zeros and propose a modified unconditional odds ratio which accounts for excessive zeros. Results from the data analyses, simulation studies, and the log-pseudo marginal likelihood (LPML) model selection procedure indicate that the proposed models perform better than conventional alternative methods. Some extensions to network meta-analysis will also be discussed.
Statistics seminar: Miroslaw Pawlak: “Nearest Neighbor Estimates for Nonlinear Time Series Systems” — Thursday, November 23 at 2:45 p.m., 301 Biological Sciences.
PIMS lecture: Anthony Bonato — Thursday, November 30 at 4 p.m., Robert Schultz Theatre.
Interdisciplinary seminar: Lena Kourkoutis — Friday, December 1 at 4 p.m., Robert Schultz Theatre.