Non-staggered, continuous DiD
I have a dataset with 300 units and 4 periods. Treatment occurs in period 4 for all units (that is, treatment is not staggered), but it comes in a continuous "dosage" form.
I tried assessing the effects with did_multiplegt by Chaisemartin & D'Haultfoeuille (2024). In this context, there would be effects for one period and parallel trend assessment for two periods. The continuous option is set to linear polynomial (1), and as suggested, standard errors are to be estimated with the bootstrap method. However, I get the following error message:
This makes me think that the package is only useful for staggered treatment settings. I am also aware of Callaway, Goodman-Bacon and Sant’Anna (2024)'s paper on continuous treatment DiD, but I believe they do not have a Stata command out. I wonder what estimator do people suggest I pursue in this context other than TWFE?
I tried assessing the effects with did_multiplegt by Chaisemartin & D'Haultfoeuille (2024). In this context, there would be effects for one period and parallel trend assessment for two periods. The continuous option is set to linear polynomial (1), and as suggested, standard errors are to be estimated with the bootstrap method. However, I get the following error message:
Code:
did_multiplegt (dyn) Y id t D, effects(1) placebo(2) continuous(1) bootstrap(1000, 1) No treatment effect can be estimated. This is because Design Restriction 1 in de Chaisemartin & D'Haultfoeuille (2024) is not satisfie > d in the data, given the options requested. This may be due to the fact that groups' period-one treatment is continuous, or takes a > large number of values, and you have not specified the continuous option. If so, you can try to specify this option. If the issue pe > rsists even with this option, this means that all groups experience their first treatment change at the same date, a situation where > the estimators of de Chaisemartin & D'Haultfoeuille (2024) cannot be used.
