csdid parallel trends

[sankavee karunakaran]

Hi All

I am attempting to run a staggered diff in diff. I am trying to see how supermarket store entry affects local retail rents and have panel data ranging from 2009-21 of retail rents across 101 locations. there is a store entry for every location and the store entries happen in the years between 2014-21, if i understand correctly, then my control group are the "not yet treated" locations

1, how do i ensure that there are parallel trends?

2. i typed in this code:

csdid rent vacancy_pct inventorybldgs , ivar(id) time(year) gvar(first_treat) method(dripw) wboot rseed(1) agg(event) notyet

and got the following output:

...........x...........x...........x...........x..
.........x...........x...........xxxxxxxxxxxxx
Difference-in-difference with Multiple Time Periods

Number of obs = 1,212
Outcome model : least squares
Treatment model: inverse probability
----------------------------------------------------------------------
| Coefficient Std. err. t [95% conf. interval]
-------------+--------------------------------------------------------
T-10 | -.1427156 .7556374 -0.19 -2.446862 2.161431
T-9 | .7684486 .9854719 0.78 -2.236527 3.773424
T-8 | 1.341737 .6823306 1.97 -.7388768 3.422351
T-7 | -1.477684 .9871418 -1.50 -4.487751 1.532382
T-6 | -.3142565 .6801118 -0.46 -2.388104 1.759591
T-5 | .9311965 .719714 1.29 -1.26341 3.125803
T-4 | -.6051048 .6499783 -0.93 -2.587068 1.376858
T-3 | .5028892 .7987295 0.63 -1.932657 2.938435
T-2 | -.3424532 .5576768 -0.61 -2.042963 1.358057
T-1 | .5310741 .6758089 0.79 -1.529653 2.591801
T+0 | .6123084 .6584088 0.93 -1.395361 2.619978
T+1 | .044663 1.162435 0.04 -3.499921 3.589247
T+2 | -1.30801 1.262548 -1.04 -5.157865 2.541846
T+3 | -.6199088 1.362374 -0.46 -4.774163 3.534345
T+4 | .1053078 1.21821 0.09 -3.609349 3.819965
T+5 | 2.377989 1.615073 1.47 -2.546814 7.302793
T+6 | 1.346427 1.144193 1.18 -2.142531 4.835385
----------------------------------------------------------------------
Control: Not yet Treated

See Callaway and Sant'Anna (2021) for details

how do i interpret this please?

thank you, sorry for any inconvenience!

[FernandoRios]

Two things
It is usually better to use
estat event
after the csdid to create the dynamic effects

Second:

----------------------------------------------------------------------
| Coefficient Std. err. t [95% conf. interval]
-------------+--------------------------------------------------------
T-10 | -.1427156 .7556374 -0.19 -2.446862 2.161431
T-9 | .7684486 .9854719 0.78 -2.236527 3.773424
T-8 | 1.341737 .6823306 1.97 -.7388768 3.422351
T-7 | -1.477684 .9871418 -1.50 -4.487751 1.532382
T-6 | -.3142565 .6801118 -0.46 -2.388104 1.759591
T-5 | .9311965 .719714 1.29 -1.26341 3.125803
T-4 | -.6051048 .6499783 -0.93 -2.587068 1.376858
T-3 | .5028892 .7987295 0.63 -1.932657 2.938435
T-2 | -.3424532 .5576768 -0.61 -2.042963 1.358057
T-1 | .5310741 .6758089 0.79 -1.529653 2.591801
T+0 | .6123084 .6584088 0.93 -1.395361 2.619978
T+1 | .044663 1.162435 0.04 -3.499921 3.589247
T+2 | -1.30801 1.262548 -1.04 -5.157865 2.541846
T+3 | -.6199088 1.362374 -0.46 -4.774163 3.534345
T+4 | .1053078 1.21821 0.09 -3.609349 3.819965
T+5 | 2.377989 1.615073 1.47 -2.546814 7.302793
T+6 | 1.346427 1.144193 1.18 -2.142531 4.835385
----------------------------------------------------------------------

If you look at your CI's all effects overlap over 0, suggesting parallel trends, or at least not enough precision to reject it. However, it also suggest no effects after treatment
HTH