0
$\begingroup$

i am writing on a statistics thesis and i have an idea for it I would be thankful to have some input from.

Callaway and SantAnna have in their R package DiD an option to have only never-treated units, or not-yet (eventually) treated together with never-treated units, as controls.

The package also has the option to add covariates to a model, to enable the parallell trends assumption to be fulfilled conditional on covariate values. This is useful if there are differences in characteristics between treated and untreated units, and those characteristics at the same time affect how the outcome variable evolves over time.

My idea is that, if there are differences in characteristics between treated and untreated (never-treated) units, would it be reasonable to instead of adding covariates to the model, to have only not-yet (eventually) treated units as controls? I know that this would shrink the size of the control groups and that it would not be possible to estimate the treatment effect for the last treated cohort, but are there any other potential downsides with this approach?

/Robert

Improve this question
$\endgroup$
1
  • $\begingroup$ Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. $\endgroup$
    – Community Bot
    Commented Sep 5 at 13:42

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.