BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Multiply robust estimation of statistical interaction parameters -
  Stijn Vansteelandt\, University of Ghent\, Belgium
DTSTART:20081021T133000Z
DTEND:20081021T143000Z
UID:TALK10346@talks.cam.ac.uk
CONTACT:Nikolaos Demiris
DESCRIPTION:A primary focus of an increasing number of scientific studies 
 is to determine whether two exposures interact in the effect that they pro
 duce on an outcome of interest. Interaction is commonly assessed by fittin
 g regression models in which the linear predictor includes the product bet
 ween those exposures. When the main interest lies in the interaction\, thi
 s approach is not entirely satisfactory because it is prone to (possibly s
 evere) bias when the main exposure effects or the association between outc
 ome and extraneous factors are misspecified. In this talk\, I will therefo
 re consider conditional mean models with identity or log link which postul
 ate the statistical interaction in terms of a finite-dimensional parameter
 \, but which are otherwise unspecified. I will show that estimation of the
  interaction parameter is often not feasible in this model because it woul
 d require nonparametric estimation of auxiliary conditional expectations g
 iven high-dimensional variables. I will thus consider `multiply robust est
 imation'\, assuming at least one of several working submodels holds. The p
 roposed approach is novel in that it makes use of information on the joint
  distribution of the exposures conditional on the extraneous factors in ma
 king inferences about the interaction parameter of interest.\nAs such\, it
  essentially encompasses a `propensity score' approach to the estimation o
 f interaction parameters. In the special case of a randomized trial or a f
 amily-based genetic study in which the joint exposure distribution is know
 n by design or by Mendelian inheritance\, the procedure leads to asymptoti
 cally distribution-free tests of the null hypothesis of no interaction on 
 an additive scale. I will illustrate the methods via simulation and the an
 alysis of a randomized follow-up study.\nThis is based on joint work with 
 Tyler VanderWeele (University of Chicago) and James Robins (Harvard Univer
 sity).
LOCATION:Large Seminar Room\, 1st Floor\, Institute of Public Health\, Uni
 versity Forvie Site\, Robinson Way\, Cambridge
END:VEVENT
END:VCALENDAR