Average marginal effects with clustered standard errors

[Mary Hemler]

Hi everyone,

I have just run the code

logistic high_conflict i.ses i.single_parent i.gender i.immigrant i.age, vce(cluster individual_id)

followed by

margins, dydx(*)

, but I noticed that the standard errors for the margins command are robust standard errors rather than clustered standard errors. It seems that only the vce(robust) and the vce(unconditional) options are available for average marginal effects, which do not provide the correct confidence intervals for my study. My question is: is it possible to estimate average marginal effects with clustered standard errors in Stata?

Thank you!

[Mary Hemler]

Here is an example of the data:

      * Example generated by -dataex-. For more info, type help dataex
      clear
      input float(child ved_SES) long aleneforeldre byte gender long immigrant float age
      1 2 0 1 1  9
      1 2 0 1 1 10
      1 2 0 1 1 11
      2 1 1 1 3 12
      3 2 0 1 1  9
      4 2 1 1 1 10
      5 2 1 1 1 10
      6 2 0 0 1 11
      7 3 0 1 3  8
      7 3 0 1 3  8
      end
      label values aleneforeldre aleneforeldre
      label def aleneforeldre 0 "Partnerskap", modify
      label def aleneforeldre 1 "Aleneforeldre", modify
      label values immigrant immigrant
      label def immigrant 1 "Uten innvandringsbakgrunn", modify
      label def immigrant 3 "2 utenlandsfødte foreldre", modify

[FernandoRios]

That is probably a typographical error in the output.
If you do the math, you can check for yourself that margins is providing Clustered Standard errors after margins.

Specifically, margins doesn't really re-estimate Standard errors. It simply uses previously obtained ones on its estimations.

So if logit is estimated using Clustered Standard errors, Margins will simply apply those in the computation

[George Ford]

A thought: The dydx from margins is not the coefficient after logistic, so the SE and CI must change to reflect that fact.

sysuse auto, clear

logistic foreign mpg weight , vce(cluster rep78)
margins, dydx(*) 

reg foreign mpg weight , vce(cluster rep78)
margins, dydx(*)

logistic foreign mpg weight , vce(cluster rep78)

Logistic regression                                     Number of obs =     69
                                                        Wald chi2(2)  =  31.57
                                                        Prob > chi2   = 0.0000
Log pseudolikelihood = -22.677963                       Pseudo R2     = 0.4652

                                  (Std. err. adjusted for 5 clusters in rep78)
------------------------------------------------------------------------------
             |               Robust
     foreign | Odds ratio   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |   .8586435   .0232937    -5.62   0.000     .8141812    .9055339
      weight |   .9957808   .0007988    -5.27   0.000     .9942164    .9973478
       _cons |    1493883    3035310     7.00   0.000     27849.89    8.01e+07
------------------------------------------------------------------------------
Note: _cons estimates baseline odds.

. margins, dydx(*) 

Average marginal effects                                    Number of obs = 69
Model VCE: Robust

Expression: Pr(foreign), predict()
dy/dx wrt:  mpg weight

------------------------------------------------------------------------------
             |            Delta-method
             |      dy/dx   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |  -.0157453   .0012612   -12.48   0.000    -.0182173   -.0132734
      weight |  -.0004368   8.18e-06   -53.42   0.000    -.0004529   -.0004208
------------------------------------------------------------------------------



. reg foreign mpg weight , vce(cluster rep78)

Linear regression                               Number of obs     =         69
                                                F(2, 4)           =      27.15
                                                Prob > F          =     0.0047
                                                R-squared         =     0.4299
                                                Root MSE          =     .35523

                                  (Std. err. adjusted for 5 clusters in rep78)
------------------------------------------------------------------------------
             |               Robust
     foreign | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |  -.0149674   .0038453    -3.89   0.018    -.0256436   -.0042912
      weight |  -.0004669   .0000824    -5.66   0.005    -.0006957    -.000238
       _cons |   2.038506   .4047721     5.04   0.007     .9146782    3.162333
------------------------------------------------------------------------------

. margins, dydx(*) 

Average marginal effects                                    Number of obs = 69
Model VCE: Robust

Expression: Linear prediction, predict()
dy/dx wrt:  mpg weight

------------------------------------------------------------------------------
             |            Delta-method
             |      dy/dx   std. err.      t    P>|t|     [95% conf. interval]
-------------+----------------------------------------------------------------
         mpg |  -.0149674   .0038453    -3.89   0.018    -.0256436   -.0042912
      weight |  -.0004669   .0000824    -5.66   0.005    -.0006957    -.000238
------------------------------------------------------------------------------