Hi everyone,
I am fairly new to stacked event studies regressions and have been having problems implementing the "stackedev" package. I have a dataset on minimum wages and employment rate across 50 different states with the time identifier being in quarters. So each state x quarter has a value for its minimum wage and employment rate. Over the time horizon, states increase their minimum wage and I have to treat each increase in minimum wage as an event. For example: an increase in minimum wage in California in 2010q1 is an event. I don't have a single event for each state and in some cases I have multiple states increasing their minimum wage in the same quarter. For eg: Both California and Oregon might have increased their minimum wage in 2010q1 and California could have minimum wage increases in 2010q1, 2010q2 and so on. I have tried to implement the stackedev package which creates each event(in this case the states that increased their minimum wage in a given quarter) and compares it to all the non-treated units. The helpfile suggests that I do something like this :
and so on.. Where the variable treat_year is the first_year under which the change takes effect (or the event happens).
The lead and lag variables are then created using gen rel=(year-treat_year)+1 .
In my case since the data is quarterly, both the above codes would be difficult to carry out. Specifically, treatment quarter will be different across the different states. Secondly the number of lags and leads are coming out to be too high since I am counting the difference between the quarters. The final idea is to implement the regression based on :
I am fairly new to stacked event studies regressions and have been having problems implementing the "stackedev" package. I have a dataset on minimum wages and employment rate across 50 different states with the time identifier being in quarters. So each state x quarter has a value for its minimum wage and employment rate. Over the time horizon, states increase their minimum wage and I have to treat each increase in minimum wage as an event. For example: an increase in minimum wage in California in 2010q1 is an event. I don't have a single event for each state and in some cases I have multiple states increasing their minimum wage in the same quarter. For eg: Both California and Oregon might have increased their minimum wage in 2010q1 and California could have minimum wage increases in 2010q1, 2010q2 and so on. I have tried to implement the stackedev package which creates each event(in this case the states that increased their minimum wage in a given quarter) and compares it to all the non-treated units. The helpfile suggests that I do something like this :
| gen treat_year=. | |
| replace treat_year=2006 if inrange(state,13,20) replace treat_year=2007 if inrange(state,21,25) |
The lead and lag variables are then created using gen rel=(year-treat_year)+1 .
In my case since the data is quarterly, both the above codes would be difficult to carry out. Specifically, treatment quarter will be different across the different states. Secondly the number of lags and leads are coming out to be too high since I am counting the difference between the quarters. The final idea is to implement the regression based on :
| stackedev outcome pre8 pre7 pre6 pre5 pre4 pre3 pre2 post0 post1 post2 post3 post4 ref, cohort(treat_year) time(year) never_treat(no_treat) unit_fe(state) clust_unit(state) covariates(cov) |
| But in my case I am confused as to whether I should do this above regression for one event at a time(say 2010q1 where CA and OR receive a treatment and other states dont) or whether it combines multiple events (this is what I thought stacked event regressions are supposed to mean). Any help on this would be very useful since I am not very well-versed with this. Thank you, Swayam |
