Applied Statistics - Ben Hansen 

 This week the Applied Statistics Workshop will present a talk by  Ben Hansen , Assistant Professor of Statistics at the University of Michigan. Professor Hansen received his Ph.D. from the University of California at Berkeley and was an NSF Post-doctoral Fellow before joining the faculty at Michigan in 2003.  His research interests include optimal matching and stratification, causal inference in comparative studies, and length-optimal exact confidence procedures.  His work has appeared in  JASA  and the  Journal of Computational and Graphical Statistics , among others. 

 Professor Hansen will present a talk entitled "Matching with prognosis scores: A new method of adjustment for comparative studies."  The corresponding paper is available from the  course website . The presentation will be at noon on Wednesday, May 3 in Room N354, CGIS North, 1737 Cambridge St. Lunch will be provided.  An abstract of the paper appears on the jump: 


 
In one common route to causal inferences from observational data, the statistician builds a model to predict membership in treatment and control groups from pre-treatment variables, X, in order to obtain propensity scores, reductions f(X) of the covariate possessing certain favorable properties. The prediction of outcomes as a function of covariates, using control observations only, produces an alternate score, the prognosis score, with favorable properties of its own. As with propensity scores, stratification on the prognosis score brings to uncontrolled studies a concrete and desirable form of balance, a balance that is more familiar as an objective of experimental control. In parallel with the propensity score, prognosis scores reduce the dimension of the covariate; yet causal inferences conditional on them are as valid as are inferences conditional only on the unreduced covariate. They suggest themselves in certain studies for which propensity score adjustment is infeasible. Other settings call for a combination of prognosis and propensity scores; as compared to propensity scores alone, the pairing can be expected to reduce both the variance and bias of estimated treatment effects. Why have methodologists largely ignored the prognosis score, at a time of increasing popularity for propensity scores? The answer lies in part with older literature, in which a similar, somewhat atheoretical concept was first celebrated and then found to be flawed. Prognosis scores avoid this flaw, as emerges from theory presented herein.