Plot coefficients with bootstrapped confidence intervals for multiple dummies in one regression

[Pier Han]

Dear Statalist,

I am trying to make a graph of all 20 coefficients with the bootstrapped CIs of 20 (leader) dummies from the following regression:

forvalues i = 1/20 {
    gen leader`i' = (leaderID==`i')
}

reg Y leader* X1 X2 X3 X4, noconstant cluster(leaderID)

I try to use

wildbootstrap reg Y leader* X1 X2 X3 X4, noconstant cluster(leaderID)
coefplot, baselevels keep(leader*)

But it does not show bootstrapped CIs. It shows CIs from the regression: reg Y leader* X1 X2 X3 X4, noconstant cluster(leaderID).

I also try to use boottest following this post https://www.statalist.org/forums/for...ence-intervals, but I cannot find a way with boottest.
Any help would be appreciated.
Many thanks in advance.

[Hemanshu Kumar]

Try:

wildbootstrap reg Y leader* X1 X2 X3 X4, noconstant cluster(leaderID) coef(leader*)
matrix C = r(table)["b",.] \ r(table)["ll",.] \ r(table)["ul",.]
coefplot matrix(C), ci((2 3)) baselevels

[Pier Han]

Hi Kumar,

Thanks for your suggestion.
I try your codes, but I got an error C: invalid syntax in ci()

I slightly change the coefplot code to

coefplot matrix(C), ci((ci_bc[2] ci_bc[3]))

My code only show the coefficients. It does not show CIs.
Could you advise on how to resolve this?

[Hemanshu Kumar]

Can you show me the exact code you are using, starting with your wildbootstrap command? I had tested the code in #2 on another dataset, so it should not have produced that error. Also, as suggested in the forum FAQ, it always helps when you provide a data example using the dataex command.

[Pier Han]

Hi Kumar,

I follow your previous suggestion and got the error C: invalid syntax in ci()

wildbootstrap reg Y leader* X1 X2 X3 X4, noconstant cluster(leaderID) coef(leader*)
matrix C = r(table)["b",.] \ r(table)["ll",.] \ r(table)["ul",.]
coefplot matrix(C), ci((2 3)) baselevels

I then slightly change the codes regarding coefplot. The above error does not come out, but the graph does not show CIs.

wildbootstrap reg Y leader* X1 X2 X3 X4, noconstant cluster(leaderID) coef(leader*)
matrix C = r(table)["b",.] \ r(table)["ll",.] \ r(table)["ul",.]
coefplot matrix(C), ci((ci_bc[2] ci_bc[3]))

I understand that providing a data sample would be helpful for addressing questions more effectively. However, due to usage regulations regarding the data I'm working with, I’m unable to share a sample. I am sorry for any inconvenience this may cause.

[Hemanshu Kumar]

As suggested in the FAQ, when your data is sensitive, one alternative is using datasets provided by Stata. Here is a reproducible version of my solution:

sysuse nlsw88, clear

rename (wage industry age hours tenure collgrad) (Y leaderID X1 X2 X3 X4)

forvalues i = 1/12 {
    gen leader`i' = (leaderID==`i')
}

wildbootstrap reg Y leader* X1 X2 X3 X4, noconstant cluster(leaderID) coef(leader*)
matrix C = r(table)["b",.] \ r(table)["ll",.] \ r(table)["ul",.]

coefplot matrix(C), ci((2 3)) baselevels

Here the wildbootstrap output is:

Wild cluster bootstrap                             Number of obs      = 2,213
Linear regression                                  Number of clusters =    12
                                                   Cluster size:
Cluster variable: leaderID                                        min =     4
Error weight: Rademacher                                          avg = 184.4
                                                                  max =   815
-----------------------------------------------------------------------------
                       Y |   Estimate      t  p-value    [95% conf. interval]
-------------------------+---------------------------------------------------
constraints              |
            leaderID = 0 |   .6582282   11.80   0.004    .5566756    .8287446
             leader1 = 0 |   4.587576    6.70   0.010    2.616149    8.090896
             leader2 = 0 |   13.80054   26.32   0.006    8.756559    20.90045
             leader3 = 0 |   5.420142   10.14   0.008    3.782845    7.585453
             leader4 = 0 |   4.162726    9.55   0.000    3.329946    5.754276
             leader5 = 0 |   7.045373   18.48   0.000    6.371985    8.461219
             leader6 = 0 |    1.97545    5.09   0.014    1.087349    3.057915
             leader7 = 0 |   4.618751   16.18   0.004    3.936329    6.051695
             leader8 = 0 |   1.906992    6.68   0.028     .411966    2.728076
             leader9 = 0 |  -1.050857   -3.77   0.230    -2.82707    1.166074
            leader10 = 0 |  -.5883985   -2.87   0.380    -10.0006    9.534447
            leader11 = 0 |  -.9660532   -9.52   0.242   -1.618082    2.561457
-----------------------------------------------------------------------------

and the corresponding coefplot graph is

I am afraid I can't troubleshoot your code since I cannot replicate your problem. In your altered version of my code, I also don't know what ci_bc is.

Edit: I did just notice one problem -- because of the way you use wildcards to specify leader dummies, you will end up including the original leaderID variable as well. To avoid this, you should use leader1-leader20 instead of leader* in the wildbootstrap command.

[Hemanshu Kumar]

You may also want to check whether you have an old version of coefplot installed. Type

which coefplot

to see the version. In my case it is

*! version 1.8.6  22feb2023  Ben Jann

If you have an older version, you may want to run

adoupdate, update

to update it and other community-contributed packages.

[Pier Han]

Hi Kumar,
You're a lifesaver!
Your suggested codes worked perfectly after I update the coefplot package.
Thank you so much for your help!