Re: [Fwd: Re: st: sgmediation and panel data]

[John Antonakis <john.antonakis@unil.ch>]

Hi:

The Baron and Kenny method is flawed. It assumes two equations estimated 
by OLS:

x = a0 + b1z + u
y = b0 + b1x + e

If x is endogenous, then the cross equation correlation must be 
estimated either using maximum likelihood or an IV estimator (2SLS). If 
you don't use the predicted value of x in the y equation and if you 
don't correlated the disturbances, then you assuming that x is 
exogenous, which is a big mistake (if it isn't exogenous). Estimates in 
the y equation will be inconsistent.

So, I would use -cmp-; you can include dummy variables in there and also 
correct for clustering.

HTH,
John.

____________________________________________________

Prof. John Antonakis, Associate Dean 
Faculty of Business and Economics
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland

Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305

Faculty page:
http://www.hec.unil.ch/people/jantonakis

Personal page:
http://www.hec.unil.ch/jantonakis
____________________________________________________



On 16.05.2010 17:07, Konstantinos Pitsakis wrote:
> Dear John,
>
> I hadn’t realized that clustering the errors was the problem, so I now dug
> deeper into this and I think I got the point.
>
> Compared to your advice, I was wondering whether you think it would be OK to
> simply run fixed effects with clustering [vce(cl)] in Stata and then do the
> Baron & Kenny method on mediation (with the Sobel test). 
>
> This sounds easy and straightforward to me because I'm not sure why I should
> use cmp and nlcom.
>
> Any suggestions on this please?
>
> Best regards, 
>   
> Konstantinos Pitsakis
> T: +31 13466 2523
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of John Antonakis
> Sent: Wednesday, May 12, 2010 2:19 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: [Fwd: Re: st: sgmediation and panel data]
>
> Forgot to mention, -cmp- is a userwritten command. For information on 
> this see:
>
> ssc d cmp
>
> Best,
> J.
>
> -------- Original Message --------
> Subject: 	Re: st: sgmediation and panel data
> Date: 	Wed, 12 May 2010 13:22:43 +0200
> From: 	John Antonakis <john.antonakis@unil.ch>
> Reply-To: 	statalist@hsphsun2.harvard.edu
> Organization: 	University of Lausanne
> To: 	statalist@hsphsun2.harvard.edu
> References: 	<003701caf1bb$8c3b10e0$a4b132a0$@Pitsakis@uvt.nl>
>
>
>
> Hi Kostas:
>
> One solution would be with cmp, so you can cluster the standard errors, 
> e.g.,
>
> cmp (y = x z1 z2) (x = z1 z2 p1 p2), ind (1 1) cluster(firm)
>
> (I'd double check that these estimates and the overid and identification 
> stats with ivreg2)
>
> Then you can do the Sobel test using the delta method with nlcom. 
> Suppose you wanted to test the indirect effect of p1 on y through x. 
> You'd do:
>
> nlcom [x]p1*[y]x
>
> How about the indirect effects both of p1 and p2 via x:
>
> nlcom [x]p1*[y]x + [x]p2*[y]x
>
> HTH,
> John.
>
> ____________________________________________________
>
> Prof. John Antonakis, Associate Dean 
> Faculty of Business and Economics
> Department of Organizational Behavior
> University of Lausanne
> Internef #618
> CH-1015 Lausanne-Dorigny
> Switzerland
>
> Tel ++41 (0)21 692-3438
> Fax ++41 (0)21 692-3305
>
> Faculty page:
> http://www.hec.unil.ch/people/jantonakis
>
> Personal page:
> http://www.hec.unil.ch/jantonakis
> ____________________________________________________
>
>
>
> On 12.05.2010 12:11, Konstantinos Pitsakis wrote:
>   
> > Dear colleagues,
> >
> > I am trying to run a Sobel test for mediation on a panel database (15
> >     
> years)
>   
> > using the “sgmediation” command created by Phil Ender.
> >
> > According to a previous statalist thread, this test should NOT work with
> > panel data:
> > http://www.stata.com/statalist/archive/2009-03/msg01155.html 
> >
> > However, I don’t understand what this means: I declared my database to be
> > panel with xtset and then run the Sobel test without any problems…
> > Are the results I get useless/wrong and why? 
> >
> > Also, is there any alternative to the Sobel test for mediation (that
> > measures the exact strength and significance of the mediation)?
> >
> > Thank you very much for your help. 
> > Best regards, 
> >   
> > Dr Konstantinos Pitsakis 
> > Assistant Professor 
> >
> > Department of Organization Studies 
> > Faculty of Social and Behavioural Sciences 
> > Fellow, Center for Innovation Research (CiR) 
> > Tilburg University, Room P1.169 
> > PO Box 90153, 5000 LE 
> > Tilburg, the Netherlands 
> >
> > T: +31 13466 2523 
> > F: +31 13466 3002 
> > E: k.pitsakis@uvt.nl 
> >
> >
> >
> >
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/help.cgi?search
> > *   http://www.stata.com/support/statalist/faq
> > *   http://www.ats.ucla.edu/stat/stata/
> >   
> >     
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>
>
>   
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/