I have panel data of year-firm observations. I am trying to estimate the effect of firm interactions with research and technology organisations (RTO's) on firm innovation and productivity. My data looks something like this:
Where 'rto' is 1 if the firm interacted with a research and technology organisation that year, 'inn' is 1 if the firm has made an innovation that year, 'prod' measures sales pr. employee, and 'ind' measures the firms industry, which should be included as a covariate, i.e., I can (with my full set of covariates in my real data) to some degree justify a conditional paralel trends assumption.
My question is how to estimate a staggered TWFE or DiD-model allowing for heterogenic effects when 1) the outcome is binary, as is the variable showing whether the firm made an innovation that year or not and 2) the same unit may get treated multiple times in the period.
In https://www.nber.org/papers/w29691 the authors review some papers providing TWFE estimators which allow for heterogenic effects across groups and time. These are e.g. provided in the Stata packages 'csdid' and 'did_multiplegt'. At least in my understanding these packages do not allow for binary outcomes. Any help in references to papers/Stata packages allowing TWFE with heterogenous effects and binary outcome would much appreciated.
Second, I have not come across any DiD/TWFE papers which discuss how to handle situations where units are treated multiple times. The staggered design discussions revolves around whether units can switch in/out of treatment, but not whether they can recieve treatments several times. Perhaps this is because same-unit multiple treatments should be modelled as a count variable instead of a binary variable? Any input here would also be much appreciated!
Code:
* Example generated by -dataex-. For more info, type help dataex clear input byte(year firmid rto inn prod ind) 1 1 0 0 10 2 2 1 0 0 10 2 3 1 1 1 12 2 4 1 0 0 12 2 5 1 0 0 13 2 1 2 0 0 9 3 2 2 0 0 9 3 3 2 0 0 9 3 4 2 0 0 8 3 5 2 0 0 9 3 1 3 0 0 7 5 2 3 1 0 7 5 3 3 0 0 7 5 4 3 1 1 10 5 5 3 0 0 12 5 1 4 0 0 6 3 2 4 0 0 6 3 3 4 0 0 6 3 4 4 1 1 8 3 5 4 0 0 10 3 end
My question is how to estimate a staggered TWFE or DiD-model allowing for heterogenic effects when 1) the outcome is binary, as is the variable showing whether the firm made an innovation that year or not and 2) the same unit may get treated multiple times in the period.
In https://www.nber.org/papers/w29691 the authors review some papers providing TWFE estimators which allow for heterogenic effects across groups and time. These are e.g. provided in the Stata packages 'csdid' and 'did_multiplegt'. At least in my understanding these packages do not allow for binary outcomes. Any help in references to papers/Stata packages allowing TWFE with heterogenous effects and binary outcome would much appreciated.
Second, I have not come across any DiD/TWFE papers which discuss how to handle situations where units are treated multiple times. The staggered design discussions revolves around whether units can switch in/out of treatment, but not whether they can recieve treatments several times. Perhaps this is because same-unit multiple treatments should be modelled as a count variable instead of a binary variable? Any input here would also be much appreciated!
Comment