Collinearity of Omitted Variables - Difference in Difference Model

[Layana Mukkadekat]

Hello All,

I'm currently doing my thesis on the impact evaluation of the Beti Bachao Beti Padhao policy in India. My aim is to use the difference in difference methodology based on a natural experiment which divides the treatment and control groups. I have a baseline year of 2015 and a final year of 2020. I am trying to run a difference in difference model with the command " xtreg v107 treateddistrict time treateddistrict# time, fe" Here v107 is the educational outcome. To give you some background, my unit of sample is girls aged between the age of 20-24 i.e. my respondents. My entire sample ( both 2015 and 2020) has 20,000 observations. 'treateddistrict' is a dummy variable which is 0 for districts in control group and '1' for districts in treatment group. 'time' is a dummy variable which is 0 in baseline year i.e. 2015 and 1 in end year i.e. 2020. I am interested in the interaction term to find the treatment effect.

However, this is the error I am getting.
" note: time omitted because of collinearity.
note: 1.treateddistrict#0b.time omitted because of collinearity.
note: 1.treateddistrict#1.time omitted because of collinearity. "

I need the effect of time and the interaction term for my thesis. How do I solve this?
In case you need it, here is the result table:

Fixed-effects (within) regression Number of obs = 15,630
Group variable: v024 Number of groups = 2

R-squared: Obs per group:
Within = 0.0008 min = 7,456
Between = 1.0000 avg = 7,815.0
Overall = 0.0000 max = 8,174

F(2,15626) = 6.13
corr(u_i, Xb) = -0.2967 Prob > F = 0.0022

---------------------------------------------------------------------------------
v107 | Coefficient Std. err. t P>|t| [95% conf. interval]
----------------+----------------------------------------------------------------
treateddistrict | -.1199561 .0371887 -3.23 0.001 -.1928503 -.0470618
time | 0 (omitted)
|
treateddistrict#|
time |
0 1 | -.0619084 .0565269 -1.10 0.273 -.1727077 .0488909
1 0 | 0 (omitted)
1 1 | 0 (omitted)
|
_cons | 4.398546 .0324213 135.67 0.000 4.334996 4.462095
----------------+----------------------------------------------------------------
sigma_u | .19573188
sigma_e | 1.6551209
rho | .01379216 (fraction of variance due to u_i)
---------------------------------------------------------------------------------
F test that all u_i=0: F(1, 15626) = 36.27 Prob > F = 0.0000

[Carlo Lazzaro]

Layana:
welcome to this forum.
The -fe- estimator wipes out time-invariant variables. Double-check your predictors.
See also -help didregress- and -help xtdidregress-.

[Layana Mukkadekat]

Thanks Carlo, I checked this out but I'm not able to figure it out. I think the only change in the result table I need is for it to give me the coefficient of treatement#time when it is 1 1, instead of 1,0 or 0,1
Is there any way I can tell STATA to omit the other two for collinearity but keep the interaction term when both are 1?

[Layana Mukkadekat]

Hey, By way of an update, I fixed it by removing time and treatment variables from the command! It gives me 3 coefficients for the interaction variable and I can see the time and treatment effects in there!