Making program's standard error results available for _se[var] call?

[Matt Manierre]

Hi everyone,

So I'm working on extracting output from a user generated program (traj). I can get the B coefficients to post as vector matrix e(b) without any trouble. This means I type something like:

local test=[eq1]_b[varname]

and it will return the regression coefficient.

How do I get stata to post the SE values (a 1 x 60 matrix) as a similar vector matrix that I can call then in the same fashion (local test=[eq1]_se[varname] )? Stata's programs do this but I can't figure out how to do for my own program. I see that the SEs are stored in r(table), but I can't figure out how to get stata to post results to that table and recognize _se.

The complication here that the SEs have to be entered manually- the B's are returned by the original program but not the SEs, and the program is run through a C plugin that is beyond my skill to even open at the moment.

This summarizes the situation I'm dealing with and my end goal:

program quicktest, eclass
regress price mpg headroom
tempvar b se
matrix define `b'=e(b)
*SE values from above regression must be entered manually in my case*
matrix define `se'=(58.42, 399.549, 2074.497)
ereturn post `b'
end

(the end goal is just):
di _b[mpg]/_se[mpg]

Thank you!

[Nick Cox]

There are indeed better ways of getting the standard errors after a regression. Here is one and I hope to be trumped with a yet better method:

. sysuse auto, clear 
(1978 Automobile Data)

. regress price mpg headroom 

      Source |       SS           df       MS      Number of obs   =        74
-------------+----------------------------------   F(2, 71)        =     10.44
       Model |   144280501         2  72140250.4   Prob > F        =    0.0001
    Residual |   490784895        71  6912463.32   R-squared       =    0.2272
-------------+----------------------------------   Adj R-squared   =    0.2054
       Total |   635065396        73  8699525.97   Root MSE        =    2629.2

------------------------------------------------------------------------------
       price |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
         mpg |  -259.1057   58.42485    -4.43   0.000    -375.6015   -142.6098
    headroom |  -334.0215   399.5499    -0.84   0.406    -1130.701    462.6585
       _cons |   12683.31   2074.497     6.11   0.000     8546.885    16819.74
------------------------------------------------------------------------------

. mata: st_matrix("myse", sqrt(diagonal(st_matrix("e(V)")))) 

. mat li myse 

myse[3,1]
           c1
r1   58.42485
r2  399.54994
r3  2074.4972

I don't know what this means for traj (provenance not stated), which I have never seen.

[Matt Manierre]

Thank you for that suggestion- generally a mata solution like this would work but in this case Traj (found here: https://www.andrew.cmu.edu/user/bjones/index.htm ) doesn't return an e(V) for the regression model at the end of the day. The model is handled within a plugin so I can't just tweak the traj code to get it to give me the values, unfortunately. I would just run the mlogit model outside of traj but that is inappropriate in this case because traj uses a mlogit model that is modified to account for uncertainty in group membership assignment. As such, I can't see any other option other than just manually entering the SEs into a matrix.

The solution I've thought of is just to bypass the _se[variable] idea and draw from the matrix I inputted.

Something along the lines of this pseudocode- I haven't gotten the chance to implement yet.

local list = var1 var 2 var 3
foreach var of varlist `list' {
local temp`var'= rownumb(SEmatrix,`var')
}