Make a Command Esttab-able

[Jared Greathouse]

When you use my user written command fdid to do

clear *
cls
net install fdid, from("https://raw.githubusercontent.com/jgreathouse9/FDIDTutorial/main") replace

net get fdid, from("https://raw.githubusercontent.com/jgreathouse9/FDIDTutorial/main") replace

qui u smoking, clear

fdid cigsale, tr(treated) unitnames(state)

ereturn list

you get

scalars:
                 e(T1) =  19
                 e(T0) =  20
                 e(T2) =  12
                  e(T) =  31
                 e(r2) =  .9989
               e(DDr2) =  .6039
               e(rmse) =  1.2821
                e(N0U) =  4
              e(DDATT) =  -27.3491
             e(pDDATT) =  -31.1852
               e(pATT) =  -18.4423
               e(CILB) =  -14.5486
                e(ATT) =  -13.6467
               e(CIUB) =  -12.7448
                 e(se) =  .4602
              e(tstat) =  29.66

macros:
                e(cmd) : "."
                  e(U) : "Montana, Colorado, Nevada, Connecticut,"

matrices:
             e(series) :  31 x 9
            e(results) :  1 x 8

The table of interest, specifically, is

Forward Difference-in-Differences          T0 R2:     0.999     T0 RMSE:     1.282
-----------------------------------------------------------------------------------------
     cigsale |     ATT       Std. Err.      t       P>|t|    [95% Conf. Interval]
-------------+---------------------------------------------------------------------------
     treated | -13.64671     0.46016      29.66     0.000    -14.54861  -12.74481
-----------------------------------------------------------------------------------------
Treated Unit: California
FDID selects Montana, Colorado, Nevada, Connecticut, as the optimal donors.
See Li (2024) for technical details.

Okay, this is well and good. But, I want to make this esttab-able (by Ben Jann, of course, on ssc). At present, when I do

clear *
cls
//net install fdid, from("https://raw.githubusercontent.com/jgreathouse9/FDIDTutorial/main") replace

//net get fdid, from("https://raw.githubusercontent.com/jgreathouse9/FDIDTutorial/main") replace

qui u smoking, clear

fdid cigsale, tr(treated) unitnames(state)
esttab

I get

----------------------------
                      (1)   
                            
----------------------------
----------------------------
N                           
----------------------------
t statistics in parentheses
* p<0.05, ** p<0.01, *** p<0.001

Okay, well.... how would I be able to reproduce the table fdid prints, with esttab. Presumably, I must do something under the hood. For reference, the table to start with would be

mat l e(results)

e(results)[1,8]
                   ATT        PATT          SE           t          LB          UB          R2        RMSE
Statistics  -13.646711  -18.442321   .46016222   29.656305  -14.548612   -12.74481   .99890459     1.28215

Perhaps daniel klein or Bjarte Aagnes would be able to comment, should they be available?

[Andrew Musau]

The default for esttab is to use the coefficients matrix e(b) and the variance matrix e(V). Of course, the user can refer to any stored matrix via the -cells()- option, but you need the two aforementioned matrices to make the output available by simply running esttab.

[daniel klein]

Andrew has answered the question. esttab looks for e(b) and e(V). I cannot readily recall whether it re-computes the remaining elements (test statistics, p-values, CIs); it might, and in this case, you probably want to put auxiliary information, like degrees of freedom, into the right places, too. As Andrew notes, esttab and similar alternatives (that must not be named) typically have hooks to specify where the necessary information is stored.

When you do make e(b) and e(V) available, you might want to be sure other post-estimation commands, like test, predict, etc., work -- or fail -- as expected.

[Jared Greathouse]

Very well. I was looking through the help for sdid (a command quite similar to mine), and I (don't think) they did it like that, but I suspected it had something to do with it being e(b) and such.

So, I'll do that at some point before I submit to SJ. Thanks so much!


EDIT: yep, here it is for SDID

*for esttab and eststo
if "`vce'"!="noinference" {
matrix b=`ATT'
matrix V=`se'^2
matrix colnames b=`4'
matrix rownames b=`1'
matrix colnames V=`4'
matrix rownames V=`4'
ereturn post b V, depname(`1') obs(`Ntotal')
}

I'll adapt it to my own, then.

[Ben Jann]

Don't know it this is helpful, but you could also type:

fdid cigsale, tr(treated) unitnames(state)
esttab, cell(results)

ben

[Jared Greathouse]

I didn't know it could be done like that, but yeah that's helpful!