jwdid

[Myat Thida Win]

Hello,

My research assesses the conflict effects on rice production by the rice mill cluster (spatial aggregation of rice mills in one location). The data is unbalanced and has 10 rounds. I am trying to implement staggered did. I run the regression manually using the following code:

reg ln_ttl_prod c.d1#c.time_dum1#c.nor_conflict_viol3#c.high_mod c.d1#c.time_dum2#c.nor_conflict_viol3#c.high_mod c.d1#c.time_dum3#c.nor_conflict_viol3#c.high_mod c.d1#c.time_dum4#c.nor_conflict_viol3#c.high_mod c.d1#c.time_dum5#c.nor_conflict_viol3#c.high_mod c.d1#c.time_dum6#c.nor_conflict_viol3#c.high_mod c.d1#c.time_dum7#c.nor_conflict_viol3#c.high_mod c.d2#c.time_dum2#c.nor_conflict_viol3#c.high_mod c.d2#c.time_dum3#c.nor_conflict_viol3#c.high_mod c.d2#c.time_dum4#c.nor_conflict_viol3#c.high_mod c.d2#c.time_dum5#c.nor_conflict_viol3#c.high_mod c.d2#c.time_dum6#c.nor_conflict_viol3#c.high_mod c.d2#c.time_dum7#c.nor_conflict_viol3#c.high_mod c.d3#c.time_dum3#c.nor_conflict_viol3#c.high_mod c.d3#c.time_dum4#c.nor_conflict_viol3#c.high_mod c.d3#c.time_dum5#c.nor_conflict_viol3#c.high_mod c.d3#c.time_dum6#c.nor_conflict_viol3#c.high_mod c.d3#c.time_dum7#c.nor_conflict_viol3#c.high_mod c.d4#c.time_dum4#c.nor_conflict_viol3#c.high_mod c.d4#c.time_dum5#c.nor_conflict_viol3#c.high_mod c.d4#c.time_dum6#c.nor_conflict_viol3#c.high_mod c.d4#c.time_dum7#c.nor_conflict_viol3#c.high_mod c.d5#c.time_dum5#c.nor_conflict_viol3#c.high_mod c.d5#c.time_dum6#c.nor_conflict_viol3#c.high_mod c.d5#c.time_dum7#c.nor_conflict_viol3#c.high_mod c.d6#c.time_dum6#c.nor_conflict_viol3#c.high_mod c.d6#c.time_dum7#c.nor_conflict_viol3#c.high_mod c.d7#c.time_dum7#c.nor_conflict_viol3#c.high_mod high_mod nor_conflict_viol3 i.ts i.time i.dt#c.time if open==1, cluster (vt)

where ln_ttl_prod is the log of total production, time_dum1-time_dum10 is the time dummies, d1-d7 are the cohort variables, nor_conflict_viol3 is the total conflict incidents, high_mod is the rice mill cluster. I control for township-fixed effects, time-fixed effects, and district-specific time trends. I clustered the standard error at the village tract level.

I have cohort variables from d1 to d10, but I noticed that after d7 all the samples were exposed to conflict so I did not include d8 to d10 in the regression. I would like to run a similar regression using jwdid as follows:

jwdid ln_ttl_prod, ivar(ts) tvar(time) gvar(first_treat)

where I created first_treat variable as follows:

gen first_treat=.
replace first_treat =1 if d1
replace first_treat =2 if d2
replace first_treat =3 if d3
replace first_treat =4 if d4
replace first_treat =5 if d5
replace first_treat =6 if d6
replace first_treat =7 if d7
replace first_treat =8 if d8
replace first_treat =9 if d9
replace first_treat =10 if d10

have the total observation of 1944 when I run manually, but I only have 644 in jwdid command. I am not sure why this happens and wondering if someone could help me.

I also would like to run the jwdid including the interaction of the rice mill cluster variable and total conflict incidents (intensity of conflict). But I am not sure how I could do that. If someone could provide me some guidance and tips, it would be really great. Thank you so much for your help in advance!

[FernandoRios]

I believe the problem is because units treated in period 1 do not have a pre-treatment period (are always treated). Those units are excluded from the analysis.
For other variables, just add it as controls

[Myat Thida Win]

Thank you so much, Dr. Fernando! My main research question is to see the heterogeneous effect of conflict by the rice mill clusters.

I added the rice mill cluster (high_mod) as a control variable and ran the jwdid command as follows:

jwdid ln_ttl_prod high_mod, ivar(ts) tvar(time) gvar(first_treat)


I am mainly interested in first_treat#time#high_mod results. I run estat group or estat calendar but I would like to see it specifically for first_treat#time#high_mod. Is there any way we could get the ATT of that interaction effects in particular and plot the results as we do in jwdid_plot?

My sample is at the rice mill level for a total of ten rounds. But I only want to control the fixed effects of the township instead of the mill's fixed effects. By specifying as township indicator instead of ID indicator as shown in the above command (ivar (ts)) could work the trick? Please let me know. Thank you so much!