Hello community,
I have a database for several houndred households and about 500 days. The first 200 days are part of a benchmark period and during the following 300 days households received a treatment.
There are five groups consisting of one control group and four groups that received different sorts of treatment. Now i want to use Difference-in-Difference estimation to estimate the effect of the treatments.
The panel dataset has an identifier "id" and a time variable "days"
Therefore I used: xtset id days
My regression equation for the overal effect of treatment looks as follows:
xtreg output yperiod ytreatment yperiodtreatment, fe robust
where output is my dependent variable, yperiod is a dummy indicating whether we are in the benchmark or the treatment period, ytreatment is also a dummy indicating whether the household received treatment and yperiodtreatment is the interaction term (yperiod*ytreatment). While regressing all time-invariant variables are swept away, which is due to the fixed effects model.
To identify the effect of say group policy 1 I run the regression only considering observations from the control group and group 1.
My questions are now:
1. Is this a legitimate approach to test the effect of treatment
2. Is it okay to have multiple periods and not just two as in the basic diff-in-diff?
2. Can i use the complete observation period or do the benchmark and the treatment period have to have the same length?
3. Could i also use a regression model like
reg output yperiod ytreatment yperiodtreatment covariate1 covariate2 i.days i.id
where time-invariant variables are not swept away?
Many thanks in Advance!
I have a database for several houndred households and about 500 days. The first 200 days are part of a benchmark period and during the following 300 days households received a treatment.
There are five groups consisting of one control group and four groups that received different sorts of treatment. Now i want to use Difference-in-Difference estimation to estimate the effect of the treatments.
The panel dataset has an identifier "id" and a time variable "days"
Therefore I used: xtset id days
My regression equation for the overal effect of treatment looks as follows:
xtreg output yperiod ytreatment yperiodtreatment, fe robust
where output is my dependent variable, yperiod is a dummy indicating whether we are in the benchmark or the treatment period, ytreatment is also a dummy indicating whether the household received treatment and yperiodtreatment is the interaction term (yperiod*ytreatment). While regressing all time-invariant variables are swept away, which is due to the fixed effects model.
To identify the effect of say group policy 1 I run the regression only considering observations from the control group and group 1.
My questions are now:
1. Is this a legitimate approach to test the effect of treatment
2. Is it okay to have multiple periods and not just two as in the basic diff-in-diff?
2. Can i use the complete observation period or do the benchmark and the treatment period have to have the same length?
3. Could i also use a regression model like
reg output yperiod ytreatment yperiodtreatment covariate1 covariate2 i.days i.id
where time-invariant variables are not swept away?
Many thanks in Advance!
