did model

[Rasul Mamatov]

Hello all.
For my thesis, I use the difference-in-differences method to analyze data.
I analyze the effect of letters of appreciation issued to schools by the ministry on the entry of student data into an information system. My model, regression result and database will be attached.

Can you see if everything was done correctly?
What is not clear to me is that the value of post##treatment not changes when I use and don't use control variables and a fixed effect.

Thank you. Waiting for your reply.

Table 4 – Descriptive Statistics

Variable	Description	Obs	Mean	Std. Dev.	Min	Max
nstud	Dependent variable. number of students entered into EMIS in each period.	1000	606.667	562.642	0	4331
post	Binary variable: 1 if data is from the pos-treatment period.	1000	.5	.5	0	1
treatment	Binary variable: 1 if in the treatment group; 0 if in the control group.	1000	.5	.5	0	1
district	Variable 9 districts (fixed effect)	1000	4.276	2.284	1	9
region	Binary variable: 1 if the region is southern, 0 if it is northern.	1000	.592	.492	0	1
townership	Binary variable. 1 if the school is public, 0 if municipal.	1000	.794	.405	0	1
area	Binary variable. 1 if rural, 0 if urban.	1000	.81	.392	0	1
lang	Binary variable. 1, if the school is in Kyrgyz language, 0, if in a foreign language.	1000	.778	.416	0	1

---

my model: reg nstud post##treatment region townership area lang district_1-district_9 if period>1, vce (cluster region)

Table 8 – The impact of letters of gratitude on data entry results.

	(I)	(II)	(III)	(IV)
	nstud	nstud	nstud	nstud
1.letter	62.272*	62.272*	62.272*	62.272*
	(8.158)	(8.129)	(8.145)	(8.112)
1.treated	-96.698**	-101.346*	-71.134***	-74.120**
	(5.207)	(8.066)	(0.634)	(1.995)
1.letter#1. treated	23.548*	23.548*	23.548*	23.548**
	(1.861)	(1.854)	(1.858)	(1.851)
N	1000	1000	1000	1000
R2	0.267	0.251	0.168	0.008
adj. R2	0.256	0.246	0.158	0.005
FE	Yes	No	Yes	No
Controls	Yes	Yes	No	No

Robust standard errors in parentheses
*** p < 0.01, ** p < 0.05, * p < 0.1
Note: The number of observations is 1000 because this regression considers 2 periods (October 4 and December 3).

[George Ford]

you can't cluster on anything but code, which I think is the ID. Too few clusters in district, region. treatment is by code.

all X's are time invariant, so don't need them. Mundlak regression would allow you to get coefficients on the Xs if you need them.

reghdfe nstud c.post#c.treatment , absorb(code period) cluster(code)

[Rasul Mamatov]

Dear Mr. George,

I want to express my sincere gratitude for your invaluable advice. As someone new to the realm of data analysis and working with Stata, your guidance has been immensely helpful.

In light of your suggestions, I hope you wouldn't mind if I seek further clarification on a few points:

1. I've created a new variable that categorizes schools into 50 different groups. Would it be appropriate to utilize this variable as a cluster variable in my analysis?
2. When it comes to selecting variables for absorb FE, could you kindly advise me on the parameters I should consider? Additionally, apart from code and periods, are there any other variables you would recommend incorporating into this analysis?
3. Regarding the inclusion of control variables (X), I'm curious about how to implement Mudlak regression. Despite my efforts, I haven't been able to locate the necessary code to execute this regression model.

Your expertise in this field is greatly appreciated, and I genuinely value any insights you can provide. Please do not hesitate to let me know if my inquiries are becoming burdensome.

Warm regards, Rasul