Brown University Center for Statistical Sciences Seminar
Abstract: When a test-treatment cannot be directly randomized because of ethical or practical constraints, recently attention has been drawn to designs that (i) randomize a surrogate intervention, the encouragement for the test-treatment, and (ii) apply this intervention to clusters (e.g., physicians) of subjects (e.g., patients) (e.g., McDonald, Hiu, and Tierney, 1992; Dexter et al., 1997). Unfortunately, this practical, clustered-encouragement-design (CED) does not prevent either noncompliance to the test-treatment, or indirect effects of the surrogate encouragement, which render, respectively, standard intention-to-treat analyses, and standard exclusion-based instrumental variables analyses (e.g., Angrist, Imbens, and Rubin, 1996) generally inappropriate or inapplicable for analyzing the effect of the test-treatment. For a surrogate intervention that is randomized at the subject level, the approach of Imbens and Rubin (1997) that does not require the exclusion restriction is appropriate, and models that incorporate covariate information in this setting are available (Hirano et al., 1999). In this work, using concepts from the phenomenologic Bayesian framework of Rubin (1978), we show how the more general, clustered structure of the CED can be analyzed in a way that improves upon current methods. Our method explores the information on the test-treatment of policy interest by simultaneously addressing (i) the surrogacy of the randomizer, (ii) the critical role of covariates in this setting, and (iii) the clustered character of the CED. We illustrate our methods by exploring the role of physician-consulting for increasing patients' completion rates of Advance Directive forms.
The work is joint with Professor Donald B. Rubin.
Brown Applied Mathematics Pattern Theory and Vision Seminar
Special Joint LCDS and Fluid Mechanics Seminar
Abstract: Performance of flow machines like jet engines, helicopters, and HVAC devices can be improved by flow control techniques, like separation control, mixing enhancement and noise reduction. These strategies can be related to control of mixing in shear flows. A general framework for vortex-based modeling, analysis, and control of shear flows is described using tools from dynamical systems and control theory. An example of application of the framework is presented. We identify areas of basic research in control of shear flows (separation, mixing, aeroacoustics) that can support technology transition in reasonably short time. This talk presents results of joint research with Andrzej Banaszuk (UTRC) and Igor Mezic (UCSB).
PDE Seminar
Department of Mathematics Colloquium
<--- 1999 Index