Dear Students,

In the Additional Material section you could find the first commented paper for the month of November, which will cover causal effect estimation when one of the basic assumption of causal modeling is not met.

Today Dr. Marco Locatelli presented to you different techniques employing causal bounds to estimate causal effects in case of unmeasured confounding.

The proposed paper, "A simulation-based bias analysis to assess the impact of unmeasured confounding when designing non-randomized database studies", is a follow up on this topic considering a diverse approach to estimate causal effect in this framework, when some proxy variables are known to be associated with the unmeasured confounders. 

The authors show a novel method based on propensity score matching techniques estimating causal effects which adjust for the proxy variables in addition to the measured confounding factors. This approach makes few hypothesis on the distributions of the variables involved in the causal paths involving the treatment and the outcome of interest and it is applicable across different scenarios characterized by more complex Directed Acyclic Graphs.

Best Wishes,
Federico Pirola