Hello Statalist! I have panel data on 50 states and 19 time periods, and I am attempting to isolate the impact of a policy change in a small number of the states. I have data on 5 time periods before the policy was implemented, and 14 periods afterwards. I am estimating the following DID model
areg dep interaction post_treat treatment [controls] i.year, absorb(state) robust
where post_treat is the time period dummy variable (post_treat=0 if in the pre-treatment period, post_treat=1 if in the post treatment period) treatment is the dummy variable indicating the treatment group (treatment=0 if in the control group, treatment=1 if in the treatment group), and interaction is the interaction term between post_treat and treatment. My principle concern is this: I would like to relax the constant treatment effect assumption, because the policy impacts a different number of people choose to participate in the policy in each time period (I have this information as well as a percent of the population). Is this a sensible course of action? How would I go about doing that? If DID is a waste of time, should I stick with my fixed effects OLS models and synthetic matching, or is there another methodology worth exploring?
Thanks for any help. Looking forward to hearing from you.
Jake
areg dep interaction post_treat treatment [controls] i.year, absorb(state) robust
where post_treat is the time period dummy variable (post_treat=0 if in the pre-treatment period, post_treat=1 if in the post treatment period) treatment is the dummy variable indicating the treatment group (treatment=0 if in the control group, treatment=1 if in the treatment group), and interaction is the interaction term between post_treat and treatment. My principle concern is this: I would like to relax the constant treatment effect assumption, because the policy impacts a different number of people choose to participate in the policy in each time period (I have this information as well as a percent of the population). Is this a sensible course of action? How would I go about doing that? If DID is a waste of time, should I stick with my fixed effects OLS models and synthetic matching, or is there another methodology worth exploring?
Thanks for any help. Looking forward to hearing from you.
Jake
