Replacing regression constant with control group mean in esttab

[Gerishom Gimaiyo]

I would like to replace the constant with control group mean when running a regression. Currently, through use of estout's stats() option, I can add it as part of the table's statistics, but what I would like to do is place it where the constant would normally be in the regression results table. My current sample code is:

eg.
reg x Treatment a b c

sum x if Treatment == 0
estadd scalar n1 = r(mean): est*
esttab using "Test.csv", b(%4.2f) p star(* 0.10 ** 0.05 *** 0.01) stats(n1 N r2_a, fmt(%9.0f %9.0g %9.2f) labels ("Control Mean" "Obs" "Adjusted R-Squared")) nogap brackets label replace

What this does is to place a line before displaying the options in stats, yet I would like the Control Mean to appear where Constant in the regression table would. Kindly advice.

Thank you.
Gerry

[Andrew Musau]

You can try applying the following 3 changes

reg x Treatment a b c
sum x if Treatment == 0
estadd scalar n1 = r(mean)

esttab using "Test.csv", b(%4.2f) p star(* 0.10 ** 0.05 *** 0.01) \\
stats(N r2_a, fmt(%9.0f %9.0g %9.2f)  labels ("Control Mean" "Obs" "Adjusted R-Squared"))\\
nogap brackets drop(_cons) scalars(n1) label replace

[Gerishom Gimaiyo]

Thanks, Andrew. When I change the code as suggested, the Constant is indeed dropped, but it's not replaced by the Control Mean. When I run ereturn list, I actually see the scalar value n1 as part of the list, but it's not added to the bottom of the regression table. Any other tweaks to this?

[Andrew Musau]

It seems this problem has not been fixed, see my previous workaround here

https://www.statalist.org/forums/for...alar-pseudo-r2

In your case, don't use estadd scalar but

 reg x Treatment a b c
sum x if Treatment == 0
scalar n_1 = r(mean)
eststo, addscalars(n1 n_1)  
esttab using "Test.csv", b(%4.2f) p star(* 0.10 ** 0.05 *** 0.01) \\
stats(n1 N r2_a, fmt(%9.0f %9.0g %9.2f) labels ("Control Mean" "Obs" "Adjusted R-Squared")) \\
nogap brackets drop(_cons) scalars(n1) label replace

Now, the value of n1 cannot appear where the constant term is because it is not a coefficient from the model but a scalar. You can include the plain option and add other options gradually if you want to eliminate separating lines between the coefficients and the other statistics.

[Gerishom Gimaiyo]

Hi Andrew, many thanks. This solved it.

[Andrew Musau]

Just an update on why estadd does not work in #2 (what I wrongly thought was a bug). When specifying this command, you need to make sure that you first store estimates of the regression of interest. Then the workaround in #4 is not needed

eststo: reg x Treatment a b c sum x if Treatment == 0
estadd scalar n1 = r(mean)  
esttab using "Test.csv", b(%4.2f) p star(* 0.10 ** 0.05 *** 0.01) \\ \
stats(n1 N r2_a, fmt(%9.0f %9.0g %9.2f)  labels ("Control Mean" "Obs" "Adjusted R-Squared"))\\
nogap brackets drop(_cons) scalars(n1) label replace

[Denat Ephrem Negatu]

This was very helpful! Is there a way to display the same control mean across different models? One model with Fixed effect and another with no fixed effects which would have the same constant mean?

[Andrew Musau]

The first time (model)

local n1= r(mean)
estadd scalar n1 = `n1'

and subsequently after each estimation, add

estadd scalar n1 = `n1'