Hi everyone,
I am using repeated cross sectional data to find out the impact of conflict on family planning. I have data for T0 (2008, before the conflict), T1 (2013, short term effects) and T2 (2018, long term affects). I constructed a variable to measure exposure to conflict. At T0, no observations are exposed to conflict, and at T1 and T2 the majority is not exposed but considerable group is exposed. Now, I want to do a Difference in Difference (DiD) analysis, combined with propensity score matching/weighting because I am using repeated cross sections. I will do seperate analyses, one to compare T0 with T2, and one to compare T0 and T1.
I tried to do so using the 'diff' command, by the following code:
diff unmetneed, treated(inc4) period(time13) id(baseid) cov(urban age religion children region) kernel ktype(gaussian) rcs support bw(0.05) cluster(baseid) robust report
Here, unmetneed is my outcome variable, inc4 is the treatment variable (which is 0 for everyone at t0), time13 indicates the year (missing for 2018, 0 for 2008, 1 for 2013). However, it isn't working and I'm getting the following error:
outcome does not vary; remember:
0 = negative outcome,
all other nonmissing values = positive outcome
This is (I think) due to the fact that no observations are treated at T0. However, this is supposed to be the case as this is the control'time'. I am thinking of matching the treated observations at T1 with very similar observations at T0, to create a 'treatment group' at T0 which can then be compared to treatment at T1. However, I have no idea how to do this yet.
Also, since there are essentially 4 groups (pre-treatment, pre-control, post-treatment, post-reatment), I initially wanted to match each of these 3 groups to the pre-control group, and then use these scores to weight the DiD regression. I am not sure if that would be similar/better/worse than using the diff command.
Any help would be very much appreciated. Thank you in advance!
I am using repeated cross sectional data to find out the impact of conflict on family planning. I have data for T0 (2008, before the conflict), T1 (2013, short term effects) and T2 (2018, long term affects). I constructed a variable to measure exposure to conflict. At T0, no observations are exposed to conflict, and at T1 and T2 the majority is not exposed but considerable group is exposed. Now, I want to do a Difference in Difference (DiD) analysis, combined with propensity score matching/weighting because I am using repeated cross sections. I will do seperate analyses, one to compare T0 with T2, and one to compare T0 and T1.
I tried to do so using the 'diff' command, by the following code:
diff unmetneed, treated(inc4) period(time13) id(baseid) cov(urban age religion children region) kernel ktype(gaussian) rcs support bw(0.05) cluster(baseid) robust report
Here, unmetneed is my outcome variable, inc4 is the treatment variable (which is 0 for everyone at t0), time13 indicates the year (missing for 2018, 0 for 2008, 1 for 2013). However, it isn't working and I'm getting the following error:
outcome does not vary; remember:
0 = negative outcome,
all other nonmissing values = positive outcome
This is (I think) due to the fact that no observations are treated at T0. However, this is supposed to be the case as this is the control'time'. I am thinking of matching the treated observations at T1 with very similar observations at T0, to create a 'treatment group' at T0 which can then be compared to treatment at T1. However, I have no idea how to do this yet.
Also, since there are essentially 4 groups (pre-treatment, pre-control, post-treatment, post-reatment), I initially wanted to match each of these 3 groups to the pre-control group, and then use these scores to weight the DiD regression. I am not sure if that would be similar/better/worse than using the diff command.
Any help would be very much appreciated. Thank you in advance!
