CSDID with unbalanced data

[hariom arora]

Hi All,
I am using CSDID command, I have unbalanced panel data on more than 3000 firms from 1988-2020, since some of the ids were treated in the first year itself i.e., 1988, i constructed the clone year of 1987 where no ids were treated. Now when I use the command some of the results for years omitted.
if possible can you please help me with this.
Below is the command i am using:
set seed 1
gen sample = runiform()<.9
csdid lnirddintensity lnipd if sample==1, ivar(id) time(year) gvar(treat) method(reg)

i have also used the default method but getting the issue.
@FernandoRios

[FernandoRios]

First things first
can you provide the following
tab year treat

That will show me if the data is correctly setup and which years could be potentially dropped

[hariom arora]

Thank you so much for your reply.
Here you go

year	treat
0	1988	1989	1991	1994	1995	1996	1997	1998	2005	2016	Total
1987	12	545	34	416	233	37	9	67	54	190	16	1613
1988	13	575	36	437	244	38	9	68	58	201	17	1696
1989	14	606	36	459	256	40	11	68	61	214	17	1782
1990	14	629	37	484	266	43	13	70	62	223	18	1859
1991	14	659	39	509	278	45	15	76	64	241	18	1958
1992	14	687	41	532	288	46	15	82	66	261	20	2052
1993	14	709	45	552	294	54	17	84	70	277	22	2138
1994	14	741	45	577	308	57	19	88	74	290	24	2237
1995	18	772	50	599	325	61	22	89	79	304	24	2343
1996	18	796	51	614	350	62	22	93	79	313	25	2423
1997	18	813	52	630	370	64	22	93	85	324	26	2497
1998	18	837	53	647	379	66	22	93	87	329	27	2558
1999	19	855	54	654	390	67	22	94	91	337	29	2612
2000	19	871	54	663	402	71	23	98	94	342	29	2666
2001	19	879	54	671	406	71	23	100	96	352	30	2701
2002	19	885	55	675	415	71	24	101	97	361	30	2733
2003	19	894	55	685	421	72	24	102	97	369	30	2768
2004	19	909	57	692	442	72	25	102	97	379	30	2824
2005	19	929	58	705	448	72	25	103	98	393	30	2880
2006	19	950	59	719	459	73	25	106	98	396	31	2935
2007	19	977	60	729	473	74	25	109	99	401	31	2997
2008	20	990	60	737	486	74	25	113	100	406	31	3042
2009	20	1000	60	745	488	74	26	113	101	407	31	3065
2010	20	1007	60	748	495	74	26	113	101	408	31	3083
2011	20	1022	60	755	502	74	26	113	101	409	31	3113
2012	20	1025	60	758	503	74	26	114	101	409	31	3121
2013	20	1035	60	763	504	74	27	114	101	412	31	3141
2014	20	1037	60	764	507	74	27	114	101	412	31	3147
2015	20	1041	61	765	509	74	27	114	102	414	31	3158
2016	20	1041	62	767	509	74	27	114	102	416	31	3163
2017	20	1041	62	767	509	74	27	114	102	417	31	3164
2018	20	1041	62	767	509	74	27	114	102	417	31	3164
2019	20	1041	62	767	509	74	27	114	102	417	31	3164
2020	20	1041	62	767	509	74	27	114	102	417	31	3164
Total	612	29880	1816	22519	13986	2218	757	3364	3024	11858	927	90961

[hariom arora]

@FernandoRios any update?

[FernandoRios]

There is nothing obvious about the problem
my guess is the panel is not appropriate or some
covRiates are missing

[hariom arora]

When I am using the default method then drops (omissions) are more but with reg produces good results [ can I use reg method instead of default method?].
Can you please guide me to plot the calendar ATT. I am not aware of the code for the same. @FernandoRios

[FernandoRios]

In that case other methods may be trying to force balance when there is none. This omitting cases.
for calendar plot
estat calendar
csdid_plot

[hariom arora]

Thanks @FernandoRios.
I have one another doubt.

When I am plotting graphs for different aggregation then what does the rectangular region indicates?
For example, in the following image what is the interpretation of the blue region since the triangles shows the ATT of particular aggregation? can one also observe the significance of the coefficients based on the plot?
Thanks once again.

[hariom arora]

@FernandoRios, sorry for disturbing you again, if possible reply.

[FernandoRios]

The regions are the 95% CI

[hariom arora]

Hi Dear FernandoRios,
When I am reproducing the results, I am getting no values. All variables are dropped, and zero observation is found while running the model.
I have also emailed you regarding this. Please help me to get the issue solved.

for your convenience, I am also attaching : tab year treatment.

year	treatment
0	1988	1989	1991	1994	1995	1996	1997	1998	2005	2016	Total
1987	1610	0	0	0	0	0	0	0	0	0	0	1610
1988	1121	572	0	0	0	0	0	0	0	0	0	1693
1989	1138	602	37	0	0	0	0	0	0	0	0	1777
1990	1192	625	37	0	0	0	0	0	0	0	0	1854
1991	783	655	38	477	0	0	0	0	0	0	0	1953
1992	825	683	40	499	0	0	0	0	0	0	0	2047
1993	867	704	44	515	0	0	0	0	0	0	0	2130
1994	601	737	44	539	308	0	0	0	0	0	0	2229
1995	572	769	49	560	325	61	0	0	0	0	0	2336
1996	570	793	50	570	349	62	22	0	0	0	0	2416
1997	497	810	51	585	368	64	22	93	0	0	0	2490
1998	420	833	52	600	377	66	22	93	87	0	0	2550
1999	431	851	53	607	388	67	22	94	91	0	0	2604
2000	438	867	53	614	400	71	23	98	94	0	0	2658
2001	450	875	53	621	404	71	23	100	96	0	0	2693
2002	459	882	54	625	413	71	24	101	97	0	0	2726
2003	468	891	54	634	419	72	24	102	97	0	0	2761
2004	479	907	56	640	440	72	25	102	97	0	0	2818
2005	102	927	57	652	446	72	25	103	98	392	0	2874
2006	104	948	58	665	457	73	25	106	98	395	0	2929
2007	104	976	59	675	472	74	25	109	99	401	0	2994
2008	107	989	59	682	485	74	25	113	100	406	0	3040
2009	107	999	59	690	488	74	26	113	101	407	0	3064
2010	108	1006	59	693	495	74	26	113	101	408	0	3083
2011	108	1021	60	700	502	74	26	113	101	409	0	3114
2012	108	1024	60	703	503	74	26	114	101	409	0	3122
2013	108	1034	60	708	504	74	27	114	101	412	0	3142
2014	108	1036	60	709	507	74	27	114	101	412	0	3148
2015	108	1040	61	710	509	74	27	114	102	414	0	3159
2016	77	1040	62	712	509	74	27	114	102	416	31	3164
2017	77	1040	62	712	509	74	27	114	102	417	31	3165
2018	77	1040	62	712	509	74	27	114	102	417	31	3165
2019	77	1040	62	712	509	74	27	114	102	417	31	3165
2020	77	1040	62	712	509	74	27	114	102	417	31	3165
Total	14478	29256	1727	19233	12104	1858	627	2579	2272	6549	155	90838

One more doubt: If an entity is treated in the year 1997, then the value for the treatment variable will be 1997 in the year 1997 and thereafter, but for all years before 1997, should it be zero or still 1997?


Thanks and Regards,
Hariom

[FernandoRios]

Still 1997
think of it as a more explicit test for anticipation.
You know they would be treated in 1997, so you can still treat them separately. If they do not know, PTA will hold. If they knew and reacted to it, PTA will be violated