Causal Inference – Quiz 3

Quiz Question Explanations

Question 1

The beauty of diff-in-diff is that it doesn’t rely on the assumption that the two groups have the same outcomes at baseline and endline, only that the change in their outcomes would have been similar in the absence of the program. This is a more credible assumption than similarity in levels in certain circumstances.

Question 2

Matching designs often require larger samples than RCTs since you need to identify appropriate matches for the treatment & comparison groups. For this reason, matching designs can often be more expensive than RCTs to execute.

The key assumption in a matching design is that, conditional on covariates, treatment assignment is as-is random. In other words, you need to make the case that there are no unmeasured or unobserved differences between the matched treatment and control groups.

Pre-treatment or time-invariant data is always needed in a matching evaluation in order to conduct the match. Post-treatment data is not reliable for matching as it could have been affected by the intervention.

Matching designs do not ensure balance on both observable and unobservable characteristics. At best they improve balance on observables (at least those that are included in your matching algorithm). Beyond this, you need to make a credible argument that treatment and control groups are balanced on other characteristics.