-
Choose a paper; hopefully from another course. It should address a policy question/ assess a policy's impact
-
Provide a 'Critique' of its methods from a Causal perspective. 'Critique' = take it apart, look at it... you don't necessarily have to 'criticize' it. This project is explanatory, not about making some convincing argument. This should be the largest section of the assignment.
-
Present a proposal to 'remedy' any weak points/ what should have been done differently.
-
Suggest how to extend the research in a causally-relevant way. Provide detail and don't forget to motivate it (why would it be relevant/ important?). Focus less on saying "We want to collect 5000 people by random messaging from February to April in...", and instead say "having a Control Group would solve Bias".
-
Administrative: HTML or PDF only Writing style should be "Internal Technical Report", not a Policy Brief. The paper should have a technical focus and conversational tone. Assume Will has read the paper.
How you should think when you write the paper + things to include. They don't need to be in this order, but there is some sense to thinking this way.
-
Estimand: Put it writing, but also put it quantitatively. What are the parameters of your dependant variable? Binary? Continuous?
-
Data Generation Process: Where is the data coming from? Is it a Natural Experiment? Do we have Observational Data only? How is the data gathered/ recorded/ observed?
-
DAG: The DAG is an attempt to model the DGP that exists in nature. Draw some DAGs, make them simple, make them complex, experiment with cause and correlation.
-
Relational Identification Strategy: Control confounders, include other variables for precision, Colliders, etc...
-
Effect Identification Strategy: Fixes Effects, Matching, Propensity Scores, ATE, CATE, ITE, Difference in Differences... You should be explicit with what is being used/ observed.
-
Evaluation: Sensitivity Analysis, qualitative evaluation, parametric vs non-parametric models.