Combining different variables in one line in estout output

[Timofey Shvaya]

I have two different models stored into model1 and model2 (saved them using "estimates store model*"), now I want to make Tex output table where coefficients from models will be displayed in one line under same name. The problem is it is different coefficients - 1.dum_treatment#1.cj_march and dummy_treatment. When I use following syntax, I got output in two different lines even under one naming:

esttab model*  using "Table3.tex",  replace b(3) se(3)  label star(* 0.10 ** 0.05 *** 0.01) varlabels(1.dum_treatment#1.cj_march "CJ Visits X Post-Movement" dummy_treatment "CJ Visits X Post-Movement",)  keep(1.dum_treatment#1.cj_march dummy_treatment) .

Is there any way to fix it?

[Devra Golbe]

If I understand you correctly, something like this will do what you want:

clear
webuse auto
gen repdum=0
replace repdum=1 if rep78>3

rename foreign dummy
reg mpg weight dummy
estimates store m1

drop dummy
rename repdum dummy
reg mpg weight dummy
estimates store m2

esttab m*

The output it produces looks like this:

--------------------------------------------
                      (1)             (2)   
                      mpg             mpg   
--------------------------------------------
weight           -0.00659***     -0.00590***
                 (-10.34)        (-10.42)   

dummy              -1.650           0.416   
                  (-1.53)          (0.47)   

_cons               41.68***        38.93***
                  (19.25)         (19.98)   
--------------------------------------------
N                      74              74   
--------------------------------------------
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

That said, here's my question: How is the reader to understand what to infer from the coefficients, since the dummy is not the same in the two models?

[Timofey Shvaya]

That said, here's my question: How is the reader to understand what to infer from the coefficients, since the dummy is not the same in the two models?

Actually, it is the same independent variable, but in one model it is being constructed as interaction variable - 1.dum_treatment#1.cj_march ; in second model it is being constructed manually and called dummy_treatment. So the issue is the results are already stored and include one interaction variable and one non-interaction.

[Devra Golbe]

Sorry- the fact that the first "variable" is an interaction did not register. As you no doubt realized, my renaming strategy won't work for that.

[Andrew Musau]

estout is from SSC (FAQ Advice #12). Providing a reproducible example, as Devra Golbe did, would elicit faster replies. What you need is to rename the relevant coefficient in e(b) for one of the regressions to match the other. For this, consider erepost from SSC.

sysuse auto, clear

gen turndisp= c.turn#c.disp
eststo m1: reg mpg weight turndisp
local colnames "`:colnames e(b)'"

reg price weight c.turn#c.disp
mat b= e(b)
mat colnames b= `colnames'
erepost b=b, rename
estimates store m2

esttab m1 m2, coeflabel(turndisp "turn × displacement") label

Res.:

. esttab m1 m2, coeflabel(turndisp "turn × displacement") label

----------------------------------------------------
                              (1)             (2)   
                     Mileage (m~)           Price   
----------------------------------------------------
Weight (lbs.)            -0.00677***        1.943*  
                          (-5.23)          (2.06)   

turn × displacement      0.000144          0.0191   
                           (0.64)          (0.12)   

Constant                    40.58***        143.1   
                          (16.96)          (0.08)   
----------------------------------------------------
Observations                   74              74   
----------------------------------------------------
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

[Andrew Musau]

Sorry- the fact that the first "variable" is an interaction did not register. As you no doubt realized, my renaming strategy won't work for that.

The renaming would work if the variable to be renamed is the interaction term. Whatever you do with factor variables can be replicated manually.