Hello Statalist community,
I have data where all observations became treated eventually but in different years. My data spans from 1987 to 2022 and the policy of interest has been passed by different states in different years, the first state adopted the policy in 1939 and the last state adopted the policy in 2014. I ran the xtreg using a binary variable to indicate whether the observation has adopted the policy to see the average effect of the policy on the outcome using this command: xtreg outcome_variable policy_start control_variables i.year_reporting, fe vce(cluster facility_id) where policy_start is a binary variable and it equals to 1 if that observation is in the treatment (adopted the policy) and 0 otherwise. After running the mentioned command, I got the coefficient of the policy to be negative and significant. So this suggests that the policy negatively affects the outcome variable. I then did the event study to estimate the effect relative to the time of policy adoption. I used 10 years before and 10 years after policy adoption (-10 to 10 is the event window). For example, when year = -10 means ten years before adopting the policy, year = 0 is the year of adoption, and year = 5 is 5 years after adopting the policy. I let -1 be the omitted reference year. I created the dummy for each year and ran the regression using this command:
xtreg outcome_variable dm_10 dm_9 dm_8 dm_7 dm_6 dm_5 dm_4 dm_3 dm_2 dm_zero dm1 dm2 dm3 dm4 dm5 dm6 dm7 dm8 dm9 dm10 control_variables i.year_reporting if time_since_treatment>= -10 & time_since_treatment<=10 , fe vce(cluster facility_id)
where dm_10 = dummy for 10 years before adopting policy (same for dm_9 to dm_1)
dm_zero = dummy for the year of adopting the policy
dm1 = dummy for 1 year after adopting policy (same for dm1 to dm10)
time_since_treatment = year_reporting - year_policy_occur
What confused me is when I ran the above command, I got coefficients for dm_10 to dm_2 which are years before adopting the policy to be negative numbers and all significant while coefficients for dm_zero, dm1 to dm10 are all positive numbers and significant. I was confused as to why this occurred since the average effect using the whole sample gave a negative and significant coefficient. I would expect the coefficients for years after the adoption (dm1 to dm10) to be negative instead of positive. I wonder why this is the case or I might misunderstood something. Please let me if you know why this happens. I greatly appreciate any help you can give.
Thank you.
I have data where all observations became treated eventually but in different years. My data spans from 1987 to 2022 and the policy of interest has been passed by different states in different years, the first state adopted the policy in 1939 and the last state adopted the policy in 2014. I ran the xtreg using a binary variable to indicate whether the observation has adopted the policy to see the average effect of the policy on the outcome using this command: xtreg outcome_variable policy_start control_variables i.year_reporting, fe vce(cluster facility_id) where policy_start is a binary variable and it equals to 1 if that observation is in the treatment (adopted the policy) and 0 otherwise. After running the mentioned command, I got the coefficient of the policy to be negative and significant. So this suggests that the policy negatively affects the outcome variable. I then did the event study to estimate the effect relative to the time of policy adoption. I used 10 years before and 10 years after policy adoption (-10 to 10 is the event window). For example, when year = -10 means ten years before adopting the policy, year = 0 is the year of adoption, and year = 5 is 5 years after adopting the policy. I let -1 be the omitted reference year. I created the dummy for each year and ran the regression using this command:
xtreg outcome_variable dm_10 dm_9 dm_8 dm_7 dm_6 dm_5 dm_4 dm_3 dm_2 dm_zero dm1 dm2 dm3 dm4 dm5 dm6 dm7 dm8 dm9 dm10 control_variables i.year_reporting if time_since_treatment>= -10 & time_since_treatment<=10 , fe vce(cluster facility_id)
where dm_10 = dummy for 10 years before adopting policy (same for dm_9 to dm_1)
dm_zero = dummy for the year of adopting the policy
dm1 = dummy for 1 year after adopting policy (same for dm1 to dm10)
time_since_treatment = year_reporting - year_policy_occur
What confused me is when I ran the above command, I got coefficients for dm_10 to dm_2 which are years before adopting the policy to be negative numbers and all significant while coefficients for dm_zero, dm1 to dm10 are all positive numbers and significant. I was confused as to why this occurred since the average effect using the whole sample gave a negative and significant coefficient. I would expect the coefficients for years after the adoption (dm1 to dm10) to be negative instead of positive. I wonder why this is the case or I might misunderstood something. Please let me if you know why this happens. I greatly appreciate any help you can give.
Thank you.
Comment