csdid only omitted variables

[Anja Ober]

Hello!

I have some issues with the csdid command. I am currently trying to use it with my dataset but I only get the message that all coefficients are zero (ommitted) and number of observations = 0 as well.
My dataset is a paneldataset (unbalanced) with a variable for the year, a personal identifier, income (used as outcome variable). My treatment is the year of birth of the child. Additional variables are sex, married and so on
Not every individual is observed every year.
My dataset does not contain any "."

I tried using: csdid gross_income sex, cluster(pid) time(syear) gvar(birthyearchild) method(reg)
birthyear child is defined as:
gen birthyearchild = 0
replace birthyearchild = syear if birth==1

I am quite new to this estimation technique so there might be some mistakes, but I cannot find them.
Any help or ideas on how I can improve the estimation would be greatly appreciated.

Thanks in advance!

[FernandoRios]

First thing first
tabúlate year gvar

[Anja Ober]

The result looks something like this:
birthyear ranging from 1999 to 2020
and surveyyear ranging from 1994 to 2020

year	0	1999	2000	2001	2002	2003
1998	9000	0	0	0	0	0
1999	9000	400	0	0	0	0
2000	14000	0	600	0	0	0
2001	14000	0	0	700	0	0
2002	15000	0	0	0	500	0
2003	15000	0	0	0	0	550

the other years before 1998 look the same and it continues after 2003
In total, there are a little over 400,000 observations

[FernandoRios]

That’s your problem
if you look at the example in helpfile you will see how the gvar year tabulation should look
specifically you do not see units treated day in 2000 before that year
hope this helps

[Jesse Haapanen]

Hi.

I am facing the same issue with my csdid command using panel data. I am running the command: csdid turnout_p, ivar(voting_area_id) time(year) gvar(gvar_zero) notyet,

where turnout_p is turnout percentage in different voting areas, voting_area_id is the unique id (string) for each of the voting area I am observing in my data, year is the panel data time variable (I only have three different years) and gvar_zero represents the time when each voting area was first treated.

My tabulation looks like this (I think it looks similar to the one provided in helpfile regarding csdid command):

. tab gvar_zero year

| year
gvar_zero | 2012 2017 2021 | Total
-----------+---------------------------------+----------
0 | 13 13 13 | 39
2012 | 226 226 226 | 678
2017 | 22 22 22 | 66
2021 | 8 8 8 | 24
-----------+---------------------------------+----------
Total | 269 269 269 | 807

Even if I restrict my gvar_zero so that it can only get values 2017 or 2021 the same problem occurs.

I am new to this command so basic mistakes are highly possible. Thanks in advance!

[FernandoRios]

Hi Jesse
Your problem is different.
CSDID expect the data to be equally spaced. Because it uses the space between years to determine what should be used as G-1. (Base period).
In you case...you cannot estimate the effect for 2012 (no pre-treatment period)
for 2017, you could use 2012 data (5 years previous), but then in 2021 you would have to use 2017 (last treated year), which is 4 years before.
This creates a conflict that explains your problem
F

[Jesse Haapanen]

Oh, now I see. In my case time between the years is almost similar but they are just because of the simplicity shown only as years. I can just change the way they are shown in my data and this fixed the problem. Thank you very much!

[Yinxin Fei]

Hello!
I meet with similar problem when I run the command "quietly csdid lnqty $X $Z i.mmodel_dum i.city_dum i.year_month, time(ymonth) gvar(trtmonth) notyet".
My research topic is how the opening of subway in a city affect vehicle sales. The data is at year-month-city-vehicle level. lnqty is the sales of each vehicle model in a city at each time period. $X and $Z are vehicle attributes and city characteristics respectively. i.mmodel_dum i.city_dum i.year_month represent vehicle model fixed effect, city fixed effect and year-month fixed effect. ymonth is the %tm format time variable indicating the year and month. trtmonth is the time when each city is treated. In the sample, there are no cities that are not treated.
Not every vehicle model is observed in each time period.
Part of the "tab ymonth trtmonth" looks like below (the whole table is too large since time spans from 2013m1 to 2015m12).
trtmonth
ymonth 2013m5 2013m6 2013m8 2013m9 2013m12 2014m4 Total

2013m1 6,062 2,144 3,882 3,542 7,262 1,903 38,946
2013m2 5,269 1,862 3,171 2,850 6,127 1,548 31,925
2013m3 5,266 1,948 3,661 3,121 6,556 1,644 34,112
2013m4 5,183 1,964 3,739 3,275 6,581 1,691 34,553
2013m5 5,293 1,993 3,763 3,437 6,897 1,665 35,644
2013m6 5,083 1,941 3,619 2,948 6,551 1,686 33,847
2013m7 5,391 1,949 3,765 3,371 6,828 1,644 35,395
2013m8 5,392 1,973 3,726 3,347 6,766 1,705 35,182
2013m9 5,501 2,088 3,741 3,339 6,899 1,725 36,168
2013m10 5,436 2,035 3,758 3,309 6,737 1,664 35,368
2013m11 5,636 2,065 3,774 3,340 6,949 1,755 36,429
2013m12 5,603 2,116 3,745 3,210 7,215 1,863 36,945
2014m1 6,172 2,278 4,088 3,782 7,836 1,976 40,911
2014m2 5,293 2,021 2,783 2,936 6,372 1,725 32,394
2014m3 5,658 2,132 3,150 3,347 7,021 1,750 35,777
2014m4 5,924 2,273 3,479 3,686 7,452 1,797 37,320
2014m5 6,036 2,269 3,285 3,679 7,648 1,854 38,684
2014m6 6,015 2,305 3,395 3,615 7,526 1,900 37,850
2014m7 6,235 2,271 3,589 3,656 7,536 1,843 38,317
2014m8 5,870 2,193 3,557 3,571 7,355 1,774 37,050
2014m9 6,171 2,472 3,663 3,781 7,827 1,941 40,196
2014m10 6,011 2,258 3,578 3,663 7,440 1,837 38,206
2014m11 6,201 2,231 3,763 3,592 7,629 1,869 39,007
2014m12 6,292 2,292 3,644 3,485 8,068 2,021 39,873

The command run smoothly, except that it returns to zero (omitted) ATT.
Thanks in advance.

[FernandoRios]

You cannot add year month fixed effects in themodel. That is causing the problems
And if you have repeated crossection, then I suggest you to move into csdid2 (from my site). csdid will do the work, but it may take some time to do all the work

[Yinxin Fei]

Thank you very much for the suggestion! I will see if I can revise the specification.
Also thanks again for the great public goods of csdid and csdid2 command.

[Yinxin Fei]

Hello!
I meet with similar problem when I run the command "quietly csdid lnqty $X $Z i.mmodel_dum i.city_dum i.year_month, time(ymonth) gvar(trtmonth) notyet".
My research topic is how the opening of subway in a city affect vehicle sales. The data is at year-month-city-vehicle level. lnqty is the sales of each vehicle model in a city at each time period. $X and $Z are vehicle attributes and city characteristics respectively. i.mmodel_dum i.city_dum i.year_month represent vehicle model fixed effect, city fixed effect and year-month fixed effect. ymonth is the %tm format time variable indicating the year and month. trtmonth is the time when each city is treated. In the sample, there are no cities that are not treated.
Not every vehicle model is observed in each time period.
Part of the "tab ymonth trtmonth" looks like below (the whole table is too large since time spans from 2013m1 to 2015m12).
trtmonth
ymonth 2013m5 2013m6 2013m8 2013m9 2013m12 2014m4 Total

2013m1 6,062 2,144 3,882 3,542 7,262 1,903 38,946
2013m2 5,269 1,862 3,171 2,850 6,127 1,548 31,925
2013m3 5,266 1,948 3,661 3,121 6,556 1,644 34,112
2013m4 5,183 1,964 3,739 3,275 6,581 1,691 34,553
2013m5 5,293 1,993 3,763 3,437 6,897 1,665 35,644
2013m6 5,083 1,941 3,619 2,948 6,551 1,686 33,847
2013m7 5,391 1,949 3,765 3,371 6,828 1,644 35,395
2013m8 5,392 1,973 3,726 3,347 6,766 1,705 35,182
2013m9 5,501 2,088 3,741 3,339 6,899 1,725 36,168
2013m10 5,436 2,035 3,758 3,309 6,737 1,664 35,368
2013m11 5,636 2,065 3,774 3,340 6,949 1,755 36,429
2013m12 5,603 2,116 3,745 3,210 7,215 1,863 36,945
2014m1 6,172 2,278 4,088 3,782 7,836 1,976 40,911
2014m2 5,293 2,021 2,783 2,936 6,372 1,725 32,394
2014m3 5,658 2,132 3,150 3,347 7,021 1,750 35,777
2014m4 5,924 2,273 3,479 3,686 7,452 1,797 37,320
2014m5 6,036 2,269 3,285 3,679 7,648 1,854 38,684
2014m6 6,015 2,305 3,395 3,615 7,526 1,900 37,850
2014m7 6,235 2,271 3,589 3,656 7,536 1,843 38,317
2014m8 5,870 2,193 3,557 3,571 7,355 1,774 37,050
2014m9 6,171 2,472 3,663 3,781 7,827 1,941 40,196
2014m10 6,011 2,258 3,578 3,663 7,440 1,837 38,206
2014m11 6,201 2,231 3,763 3,592 7,629 1,869 39,007
2014m12 6,292 2,292 3,644 3,485 8,068 2,021 39,873

The command run smoothly, except that it returns to zero (omitted) ATT.
Thanks in advance.

Sorry I have another related question. Still using this background, can I add year fixed effect or month fixed effect or city fixed effect? I try to change the year-month fixed effect to year fixed effect and month fixed effect. Now there are four sets of fixed effects: car model fixed effect, city fixed effect, year fixed effect and month fixed effect. I still get the omitted result.

[FernandoRios]

You can only add covariates that are observed across all treated groups. If a covariates is not present on even one treated group or. Calendar for all years, then you probably shouldn’t add that control to the model

[Yinxin Fei]

Got it! Thank you very much!

[Tiziana Giuliani]

Hello FernandoRios and everybody

I am trying to run csdid but the command returns me (omitted) for some of the years...tab tvar gvar doesn't show any problem apparently. I am attaching the results below.

Can you please help me in understanding what is the problem? Many thanks,

Tiziana

-.-.-.-.-.-.-.-.-.-.

csdid lvalesp lfatturato esperienza labint, cluster(codimp) time(anno) gvar(anno_primo_trattamento) method(dripw)
......x......x......x......x
Difference-in-difference with Multiple Time Periods

Number of obs = 144,965
Outcome model : least squares
Treatment model: inverse probability
(Std. err. adjusted for 60,135 clusters in codimp)
------------------------------------------------------------------------------
| Coefficient Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
g2017 |
t_2014_2015 | .1581282 .1641355 0.96 0.335 -.1635715 .4798279
t_2015_2016 | .2025789 .1571553 1.29 0.197 -.1054398 .5105976
t_2016_2017 | -.4973028 .2020681 -2.46 0.014 -.893349 -.1012565
t_2016_2018 | -.638093 .2054953 -3.11 0.002 -1.040856 -.2353296
t_2016_2019 | -.530153 .2123766 -2.50 0.013 -.9464035 -.1139026
t_2016_2020 | -.4231567 .2772522 -1.53 0.127 -.9665611 .1202477
t_2016_2021 | 0 (omitted)
-------------+----------------------------------------------------------------
g2018 |
t_2014_2015 | .0543936 .109197 0.50 0.618 -.1596285 .2684157
t_2015_2016 | -.0536442 .1653031 -0.32 0.746 -.3776324 .2703439
t_2016_2017 | -.6216778 .0605211 -10.27 0.000 -.740297 -.5030586
t_2017_2018 | -.1154007 .0543305 -2.12 0.034 -.2218865 -.0089149
t_2017_2019 | -.0874652 .0751574 -1.16 0.245 -.2347709 .0598406
t_2017_2020 | -.1858795 .1062882 -1.75 0.080 -.3942005 .0224416
t_2017_2021 | 0 (omitted)
-------------+----------------------------------------------------------------
g2019 |
t_2014_2015 | .0789758 .0989177 0.80 0.425 -.1148993 .2728509
t_2015_2016 | .0015087 .0922439 0.02 0.987 -.1792861 .1823034
t_2016_2017 | -.4819845 .11696 -4.12 0.000 -.7112219 -.2527472
t_2017_2018 | -.0301971 .1036247 -0.29 0.771 -.2332979 .1729036
t_2018_2019 | -.013581 .1022975 -0.13 0.894 -.2140803 .1869184
t_2018_2020 | .1295917 .1266511 1.02 0.306 -.1186398 .3778232
t_2018_2021 | 0 (omitted)
-------------+----------------------------------------------------------------
g2020 |
t_2014_2015 | .2691706 .2843052 0.95 0.344 -.2880574 .8263986
t_2015_2016 | -.1176946 .2240861 -0.53 0.599 -.5568953 .3215061
t_2016_2017 | -.2320971 .3089533 -0.75 0.453 -.8376345 .3734403
t_2017_2018 | -.7854793 .2462791 -3.19 0.001 -1.268178 -.3027811
t_2018_2019 | .4099231 .2524097 1.62 0.104 -.0847909 .904637
t_2019_2020 | -.1927531 .2308225 -0.84 0.404 -.645157 .2596507
t_2019_2021 | 0 (omitted)
------------------------------------------------------------------------------
Control: Never Treated

See Callaway and Sant'Anna (2021) for details

. sum anno anno_primo_trattamento

Variable | Obs Mean Std. dev. Min Max
-------------+---------------------------------------------------------
anno | 161,131 2016.966 2.2906 2014 2021
anno_primo~o | 161,131 114.0929 466.0988 0 2020

. tab anno anno_primo_trattamento

| Anno_Primo_trattamento
Anno | 0 2017 2018 2019 2020 | Total
-----------+-------------------------------------------------------+----------
2014 | 25,401 83 716 231 35 | 26,466
2015 | 26,155 86 792 248 34 | 27,315
2016 | 26,889 92 859 277 42 | 28,159
2017 | 15,115 71 755 234 29 | 16,204
2018 | 14,796 76 781 243 34 | 15,930
2019 | 15,360 74 775 263 36 | 16,508
2020 | 13,294 75 730 249 45 | 14,393
2021 | 15,012 78 776 253 37 | 16,156
-----------+-------------------------------------------------------+----------
Total | 152,022 635 6,184 1,998 292 | 161,131

[Tiziana Giuliani]

Sorry FernandoRios just noticed that the omitted error might be linked to missing values in covariates I am using...I dropped all observations with missing values but nothing happened and the results again shows omitted as above