[gov3009-l] Wolos Pattanayak on "Propensity Score Analyses at the CDC"

Hello all, We hope you will join us this Wednesday, February 17th at the Applied Statistics workshop when we will be happy to have Cassandra Wolos Pattanayak (Department of Statistics). Details and an abstract are below. A light lunch will be served. Thanks! "Diagnostics for Covariate Balance: Propensity Score Analyses at the Centers for Disease Control and Prevention" (Joint work with Donald B. Rubin and the CDC) Cassandra Wolos Pattanayak Department of Statistics February 17th, 2010, 12 noon K354 CGIS Knafel (1737 Cambridge St) ABSTRACT: With propensity score analyses increasingly widespread, the crucial step of scrutinizing the balance of covariates in the selected treatment and control groups deserves emphasis. This talk focuses on two recent collaborations with the CDC that highlight the importance of diagnosing covariate balance at the design stage, without access to study outcomes. In the first application, we create matched treatment and control groups to examine the effect of transferring one v. two embryos during in vitro fertilization on pregnancy rates, multiple births, and other outcomes. The matching process demonstrates the importance of attention to detail in covariate selection and the use of balance checking to choose a matching protocol. The second application involves propensity score subclassification in a clinical trial evaluating a possible prevention for perinatal sepsis in South Africa. The clinical trial actually randomized patients to treatment or placebo, and our analysis illustrates the possibilities for improving covariate balance when faced with an "unlucky" randomization. Both projects have been conducted without access to study outcomes and with a focus on diagnostics. Cheers, matt. ~~~~~~~~~~~ Matthew Blackwell PhD Candidate Institute for Quantitative Social Science Department of Government Harvard University email: mblackwell(a)iq.harvard.edu url: http://people.fas.harvard.edu/~blackwel/