skip to content
 

Estimation of Causal Effects in Network-Dependent Observational Data

Presented by: 
Oleg Sofrygin University of California, Berkeley
Date: 
Tuesday 12th July 2016 - 15:30 to 16:00
Venue: 
INI Seminar Room 1
Abstract: 
Co-author: Mark J. van der Laan (University of California, Berkeley, CA)

We outline the framework of targeted maximum likelihood estimation (TMLE) in observational network data. Consider a dataset in which each observational unit is causally connected to other units via a known social or geographical network. For each unit we observe their baseline covariates, their exposure and their outcome, and we are interested in estimating the effect of a single time-point stochastic intervention. We propose a semi-parametric statistical model that allows for between-unit dependencies: First, unit-level exposure can depend on the baseline covariates of other connected units. Second, the unit-level outcome can depend on the baseline covariates and exposures of other connected units. We impose some restrictions on our model, e.g., assuming that the unit's exposure and outcome depend on other units as some known (but otherwise arbitrary) summary measures of fixed dimensionality. A practical application of our approach is demonstrated in a large-scale networ k simulation study that applies two newly developed R packages: simcausal and tmlenet. We also discuss some extensions of our work towards estimation in longitudinal data.
The video for this talk should appear here if JavaScript is enabled.
If it doesn't, something may have gone wrong with our embedded player.
We'll get it fixed as soon as possible.
University of Cambridge Research Councils UK
    Clay Mathematics Institute London Mathematical Society NM Rothschild and Sons