re: re: st: How to test for heteroscedasticity with mediation model

["Ariel Linden" <ariel.linden@gmail.com>]

Hi Florian,

My response to you was directed at this statement  you had in your original
post:  "I can use postestimation commands after the regression for the
direct or indirect effect, but not for both at the same time." Thus, I
recommended these models as they can allow you use robust SE for the single
model.

As far as I can tell, there is no way of using the typical post-estimation
commands for these programs to test for heteroscedasticity (e.g., estat
hottest). 

While not foolproof, I would run the model both with and without robust SE
and see if there is a substantial difference in the results. If there is,
then you may have to do some manual checks on the data. Here is an example
using the data that comes with the -khb- command (findit khb). With this
example, there is no meaningful difference between the SEs. 

Ariel

*** code***
. use dlsy_khb.dta
. khb logit univ fses || abil

Decomposition using the KHB-Method

Model-Type:  logit                                 Number of obs     =
1896
Variables of Interest: fses                        Pseudo R2         =
0.16
Z-variable(s): abil
----------------------------------------------------------------------------
--
        univ |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
Interval]
-------------+--------------------------------------------------------------
--
fses         |
     Reduced |   .5471173   .0758482     7.21   0.000     .3984575
.6957771
        Full |   .3746087    .075724     4.95   0.000     .2261924
.5230249
        Diff |   .1725086     .03014     5.72   0.000     .1134353
.2315819
----------------------------------------------------------------------------
--

. khb logit univ fses || abil, vce(robust)

Decomposition using the KHB-Method

Model-Type:  logit                                 Number of obs     =
1896
Variables of Interest: fses                        Pseudo R2         =
0.16
Z-variable(s): abil
----------------------------------------------------------------------------
--
             |               Robust
        univ |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
Interval]
-------------+--------------------------------------------------------------
--
fses         |
     Reduced |   .5471173   .0852282     6.42   0.000      .380073
.7141615
        Full |   .3746087   .0857517     4.37   0.000     .2065385
.5426789
        Diff |   .1725086   .0305696     5.64   0.000     .1125932
.232424
----------------------------------------------------------------------------
--

**** end code****

.

Date: Sat, 16 Nov 2013 19:28:13 +0100
From: "Florian Christian Esser" <florian.esser@tu-dortmund.de>
Subject: re: st: How to test for heteroscedasticity with mediation model

Hi Ariel,

thanks a lot, my issue is not how to use robust standard errors but to
test whether or not I have to.

> I suggest you check out both the user-written program -khb- (findit khb)
> and
> the user-written program -medeff (findit medeff). Both programs allow you
> to
> choose robust SE as an option.
>
> Ariel
>
> Date: Fri, 15 Nov 2013 15:17:17 +0100
> From: "Florian Christian Esser" <florian.esser@tu-dortmund.de>
> Subject: st: How to test for heteroscedasticity with mediation model
>
> Hi everyone,
>
> I am calculating a (moderated) mediation model, and I need to check if I
> have to use heteroscedasticity adjusted standard errors in my model.
> How do I test for heteroscedasticity in such a mediation model?
> I can use postestimation commands after the regression for the direct or
> indirect effect, but not for both at the same time.
>
> Thanks in advance


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/