Collecting Confidence Intervals from regress using statsby

[Juan Pablo Mesa]

Hi. I want to use statsby to collect the lower and upper limits of the 95% confidence interval of a regression coefficient generated with regress. I'm using the following command to collect the coefficient and the standard error, but can't find a way to collect the confidence interval limits too. How can I collect the confidence interval limits?

 statsby _b _se, by(state): reg y x, nocons r

[Leonardo Guizzetti]

Here's a riff on the manual examples for -statsby-. Official regression commands return results in -r()- like the confidence interval in a matrix called -r(table)-. Here I show what that might look like following a linear regression.

use "https://www.stata-press.com/data/r17/auto2", clear
regress price weight length mpg if foreign==1, nohead
mat list r(table)

Result

. regress price weight length mpg if foreign==1, nohead
------------------------------------------------------------------------------
       price | Coefficient  Std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
      weight |   4.784841   1.670006     2.87   0.010     1.276288    8.293394
      length |   13.39052   50.70229     0.26   0.795    -93.13104    119.9121
         mpg |   -18.4072   59.37442    -0.31   0.760    -143.1482    106.3338
       _cons |   -6497.49   6337.952    -1.03   0.319    -19813.03    6818.052
------------------------------------------------------------------------------

. mat list r(table)

r(table)[9,4]
            weight      length         mpg       _cons
     b   4.7848411   13.390522  -18.407204  -6497.4902
    se   1.6700063   50.702293   59.374424   6337.9515
     t   2.8651635   .26410092  -.31001907   -1.025172
pvalue   .01029028    .7947022   .76010709   .31886286
    ll   1.2762881  -93.131042  -143.14824  -19813.032
    ul   8.2933942   119.91209   106.33383   6818.0518
    df          18          18          18          18
  crit    2.100922    2.100922    2.100922    2.100922
 eform           0           0           0           0

In recent versions of Stata you can access matrix results by their row and column names, or else you can always used the row and column number. Adding this to -statsby-, you can do something like this:

statsby weight_b=_b[weight] weight_ll=(r(table)["ll","weight"]) weight_ul=(r(table)["ul","weight"]), by(foreign) verbose nodots: regress price weight length mpg

Result

     +-------------------------------------------+
     |  foreign   weight_b   weigh~ll   weigh~ul |
     |-------------------------------------------|
  1. | Domestic   6.767233   4.301539   9.232928 |
  2. |  Foreign   4.784841   1.276288   8.293394 |
     +-------------------------------------------+

[Juan Pablo Mesa]

Thank you very much, Leonardo. Your code does just what I needed.

Here I leave a simplified version of your code for reference. It collects the coefficient, the standard error, the lower limit and the upper limit of a univariate linear regression:

 statsby (r(table)[1,1]) (r(table)[2,1]) (r(table)[5,1]) (r(table)[6,1]), by(state): reg y x, nocons r