Re: Wald test: alternatives and small sample sizes

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

OK....then try the following after -suest-:

test ([one_mean]x = [two_mean]x) ([one_mean]z = [two_mean]z)  
([one_mean]q = [two_mean]q) mtest

See the manual on -test-; the -mtest- gives you a multiple test option.

You might also want to control for multiple tests by doing the following:

test ([one_mean]x = [two_mean]x) ([one_mean]z = [two_mean]z)  
([one_mean]q = [two_mean]q) mtest(b)

-b- in this case is a Bonferonni adjustment (there are others; see the 
manual).

If you still don't find a significant result for the coefficient you are 
interested in you might not have enough power (precision) or the effect 
might not be there.

Best,
J.


____________________________________________________

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 24.06.2010 10:20, Collewaert V (MCFE) wrote:
> Hi John,
>
> First of all, I apologize for my mistake: indeed it should have been Y regressed on X Z and Q. 
>
> As you proposed, I ran the test on the three coefficients simultaneously: chi2(  3) =   12.35, Prob > chi2 =    0.0063. While this confirms that there are differences between the two models in terms of those three variables, I cannot tell which ones are different and which ones are not (which I need as I have three hypotheses, one for each variable, in which I claim that two effects will be the same and one will be different). 
>
> Kind regards,
>
> Veroniek
>
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of John Antonakis
> Sent: jeudi 24 juin 2010 9:34
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: Wald test: alternatives and small sample sizes
>
> Hi Veroniek:
>
> You might not have enough power.  Try testing all three coefficients 
> that are common simultaneously:
>
> test ([one_mean]x = [two_mean]x) ([one_mean]z = [two_mean]z) 
> ([one_mean]q = [two_mean]q)
>
> Note, you have y as a dependent variable and as an independent variable; 
> just to show you how to put more than 1 test in there I added q as a 
> predictor too:
>
> Regress y x z q + controls if group = 1
> Est store one
> Regress y x z q + controls if group = 0
> Est store two
> Suest one two, Cluster(Nr_Co)
>
> HTH,
> J.
>
> ____________________________________________________
>
> 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 24.06.2010 08:38, Collewaert V (MCFE) wrote:
>   
> > Dear Statalist,
> >
> > I am trying to estimate two models (on two subsamples) with SuEst and cluster option as both samples are related (they belong to the same ventures). Specifically:
> >
> > Regress Y X Y Z + controls if group = 1
> > Est store one
> > Regress Y X Y Z + controls if group = 0
> > Est store two
> > Suest one two, Cluster(Nr_Co)
> >
> > However (!) the control variables are different for each group (for instance I control for experience in group 1, but not in group 0, and control for tenure in group 0, but not in group 1), so I do not have the same model for both groups.
> >
> > X, Y and Z refer to three main constructs of interest to my study and are included in both models. One of my hypotheses is that construct X should have a stronger (and positive) effect on group 1's outcome than on group 0's outcome. I tried running a Wald test:
> >
> > Test [one_mean = two_mean] X
> >
> > However, results seem strange to me: X is highly significant in model (group) 1, but absolutely not significant in model (group) 2 and still the Wald test proclaims that both coefficients are equal (chi2(  1) =    1.09,  Prob > chi2 =    0.2966). Could the problem be my small sample sizes? (respectively 72 and 65) And if so, what alternatives could I try? Or should I use another test than the Wald test to test this kind of hypothesis?
> >
> > With kind regards,
> >
> > Veroniek
> >
> >
> >
> >
> >
> >
> > *
> > *   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/
>   
*
*   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/