Dimension reduction, propensity score analyses and double robustness for estimation of causal effects in observational studies

Add to your list(s) Download to your calendar using vCal

• Hui Guo, Statistical Laboratory, Cambridge
• Monday 26 January 2009, 16:00-17:30
• Centre for Mathematical Sciences, MR15.

If you have a question about this talk, please contact Dr Clive Bowsher.

In many medical studies, the focus is on estimating the average causal effect (ACE) of a treatment. If the available data are gathered from observational studies where randomisation is absent, then estimating the ACE becomes problematic. We aim to find a scalar propensity variable to represent subjects’ characteristics. Given that the response, characteristics and treatment are linearly related, identical (different) covariance matrices of the characteristics for the treated and untreated groups result in the same (different) estimated ACEs from regressing the response on the treatment and characteristics, and on the treatment and propensity variable. Moreover, if we construct an estimator by combining the response regression model and the propensity score model, there are two chances to obtain an unbiased estimator for the ACE , which is the property of double robustness.

This talk is part of the Causal Inference Seminar and Discussion Group series.

This talk is included in these lists:

• Causal Inference Seminar and Discussion Group
• Centre for Mathematical Sciences, MR15

Note that ex-directory lists are not shown.