Good Afternoon,
I am estimating the effect of being offered a treatment across many states and between two time periods using a two-way fixed effects model specification. This is not the usual DID because there are states that offer the treatment in my “Pre” period. There are two treatment variables below that try to capture the observations that were offered between the two periods. My data looks like:
ID Year Offer Treat1 Treat2
1 2008 1 0 1
1 2010 1 1 1
2 2008 0 0 0
2 2010 1 1 1
3 2008 0 0 0
3 2010 0 0 0
The level of analysis is “ist” individuals(i) in states(s) across time(t). The “offer” is an indicator of whether the treatment was offered in period t. The “Treat1” is an indicator variable for being offered only in 2010. The “Treat2” is an indicator for being offered in 2008 or 2010. The code used to generate the two variables is:
gen Treat1=0
replace Treat1=1 if Offer==1 & Year==2010
gen Treat2=0
replace Treat2=1 if Offer==1 & Year==2008 | Offer==1 & Year==2010
I want to confirm that Treat1 captures the effect on individuals offered the treatment between 2008 and 2010. A concern becomes grouping those offered in 2008 into the control (making this slightly different than a DID). Secondly, Treat2 should capture the effect of being offered in 2008 or 2010. So, for Treat2 the control is simply not being offered treatment, but the time difference doesn’t necessarily capture newly offered observations. Can someone please suggest a way to treat observations offered between 2008 and 2010; and also, how to deal with the observations already offered treatment in the “Pre-2010” period.
My second question is how implement a two-way fixed effects model in STATA:
Y_ist= a_0 + a_1*Treat1_ist + time_fixed_t + state_fixed_s + u_ist
Naturally, I used this command…
xtreg Y Treat1, fe vce(cluster state)
Where state is just a dummy for state id. However, according to the manual, xtreg, fe uses regular Q-transformation with individual fixed effects (assuming only individual effects in the error term). Is there a way to run a two-way error component model where I can tell stata that the within transformation should be across state and time? Or is there a different command for two-way and three-way error component estimation?
Thank You in advance for your replies,
-Alan
I am estimating the effect of being offered a treatment across many states and between two time periods using a two-way fixed effects model specification. This is not the usual DID because there are states that offer the treatment in my “Pre” period. There are two treatment variables below that try to capture the observations that were offered between the two periods. My data looks like:
ID Year Offer Treat1 Treat2
1 2008 1 0 1
1 2010 1 1 1
2 2008 0 0 0
2 2010 1 1 1
3 2008 0 0 0
3 2010 0 0 0
The level of analysis is “ist” individuals(i) in states(s) across time(t). The “offer” is an indicator of whether the treatment was offered in period t. The “Treat1” is an indicator variable for being offered only in 2010. The “Treat2” is an indicator for being offered in 2008 or 2010. The code used to generate the two variables is:
gen Treat1=0
replace Treat1=1 if Offer==1 & Year==2010
gen Treat2=0
replace Treat2=1 if Offer==1 & Year==2008 | Offer==1 & Year==2010
I want to confirm that Treat1 captures the effect on individuals offered the treatment between 2008 and 2010. A concern becomes grouping those offered in 2008 into the control (making this slightly different than a DID). Secondly, Treat2 should capture the effect of being offered in 2008 or 2010. So, for Treat2 the control is simply not being offered treatment, but the time difference doesn’t necessarily capture newly offered observations. Can someone please suggest a way to treat observations offered between 2008 and 2010; and also, how to deal with the observations already offered treatment in the “Pre-2010” period.
My second question is how implement a two-way fixed effects model in STATA:
Y_ist= a_0 + a_1*Treat1_ist + time_fixed_t + state_fixed_s + u_ist
Naturally, I used this command…
xtreg Y Treat1, fe vce(cluster state)
Where state is just a dummy for state id. However, according to the manual, xtreg, fe uses regular Q-transformation with individual fixed effects (assuming only individual effects in the error term). Is there a way to run a two-way error component model where I can tell stata that the within transformation should be across state and time? Or is there a different command for two-way and three-way error component estimation?
Thank You in advance for your replies,
-Alan
