3.7 Discussion

In our discussion, we explored the causal effect through the lens of the potential outcome framework, emphasizing the crucial assumptions needed for its accurate identification. We particularly focused on the independence assumption and the unconfoundedness assumption, alongside the Stable Unit Treatment Value Assumption (SUTVA) and the overlap assumption. These assumptions play a pivotal role in ensuring valid causal inference, with the conditional independence assumption being highly effective in randomized controlled trials.

However, in observational studies where randomization is not possible, the conditional independence assumption can be quite stringent and challenging to meet as there might be unobserved variables driving the treatment assignment. Consequently, it is essential to leverage alternative methodologies designed for observational settings. Exploring approaches such as propensity score matching, instrumental variables, regression discontinuity design, and difference-in-differences can help overcome these challenges and improve the robustness of causal effect estimation when randomization is not feasible. By employing these methods, we can better navigate the complexities of observational data and draw more reliable causal inferences.