Staggered DID, trend changes with different windows and end groups

[haiyan lin]

Dear All,

I use "eventstudyinteract" to do staggered DID, with multiple FEs and lag variables. I tried different choice of event windows and group distant relative events to the two ends. For example, when windows is [-5, 5], event time>=5 is assigned with 5, and event time<=-5 is assigned with -5. The weired thing is when I change event window, the trends change dramatically. I am thinking it might be something related with data structure, but could not figure it out.

My data is at firm level, from 1975-1995. The policy reform is at state level, from 1978-1993. I attached the results after tab relative event time (inter_yr) in the end.

Any advice is greatly appreciated!!!

Best,
Haiyan

.	tab inter_yr
	inter_yr	Freq.	Percent	Cum.
-18	1	0.00	0.00
-17	30	0.03	0.03
-16	56	0.05	0.08
-15	64	0.06	0.14
-14	87	0.08	0.22
-13	197	0.18	0.40
-12	1,166	1.08	1.48
-11	2,355	2.18	3.67
-10	3,174	2.94	6.61
-9	3,242	3.00	9.61
-8	3,582	3.32	12.93
-7	4,232	3.92	16.85
-6	4,335	4.02	20.87
-5	4,468	4.14	25.01
-4	4,598	4.26	29.27
-3	4,736	4.39	33.66
-2	4,792	4.44	38.11
-1	4,931	4.57	42.68
0	5,010	4.64	47.32
1	5,042	4.67	51.99
2	5,011	4.64	56.64
3	5,005	4.64	61.27
4	5,053	4.68	65.96
5	5,167	4.79	70.75
6	5,298	4.91	75.66
7	5,537	5.13	80.79
8	5,868	5.44	86.23
9	5,847	5.42	91.65
10	3,912	3.63	95.27
11	2,428	2.25	97.52
12	1,087	1.01	98.53
13	1,027	0.95	99.48
14	537	0.50	99.98
15	4	0.00	99.98
16	5	0.00	99.99
17	7	0.01	99.99
18	7	0.01	100.00
Total	107,898	100.00

[haiyan lin]

I upload the results where no covariates are included, except for multiple FEs. The only difference in dynamic graphs is the event window and correponding two end groups.