We hope you can join us this Wednesday, September 7, 2011 for the first Applied Statistics Workshop this semester. Jamie Robins, the Mitchell L. and Robin LaFoley Dong Professor of Epidemiology at the Harvard School of Public Health, will present his paper entitled "Parametrizations, Likelihoods, Semiparametrics, Causal Graphs, Model Selection and Discovery for Complex Causal Models". A light lunch will be served at 12 pm and the talk will begin at 12.15.

"Parametrizations, Likelihoods, Semiparametrics, Causal Graphs, Model Selection and Discovery for Complex Causal Models"
Jamie Robins
Harvard School of Public Health
CGIS K354 (1737 Cambridge St.)
Wednesday, September 7th, 2011 12.00 pm

Abstract:

I will discuss recent results on novel factorizations of the likelihood for both (i) semiparametric causal models (marginal structural models and structural nested models) and (ii) nonparametric causal graphical models with unmeasured confounders (hidden variables). I will show how the causal question of substantive interest dictates the choice of factorization (eg a causally complete MSM factorization is appropriate for inference on direct effects ) and a R(recursive)-factorization is appropriate for the construction of algorithms to perform model selection and causal discovery in the setting of nonparametric causal graphical models. Associated with each factorization is a parametrization . The parametrization dictates both the form of doubly robust estimators and the likelihood to be maximized in scoring algorithms such as BIC used for model selection. I will derive the relationship (ie mappings or diffeomorphisms ) between these alternate parameterizations. This is joint work with Thomas Richardson, Ilya Shpitser, Robin Evans, Eric Tchetgen and Andrea Rotnitzky.

Posted by Konstantin Kashin at September 5, 2011 12:47 AM