Re: st: Subgroup analysis

[David Bai <db555@mail.com>]

I checked the VIF test and didn't find any collinearity issues. 
Correlation test does not show high correlations either.
Would 28 predictors (mostly are dummy recoding variables of major 
variables found in the literature) be too many for a sample size of  
600?  Would you consider that there might be sth wrong with my analysis?
IAssuming the results are correct,  can I interpret this as because 
African Americans are very homogeneous in the distributions of  these 
predictors, and therefore it is hard for the analysis to distinguish 
any variations in the effects and therefore find non-significance in 
the results?  or other interpretations?
Some of the predictors that are often significant for student success 
(e.g., parental education, educational aspiration) are significant in 
my white subsample, but not in the African American student sample, 
which does not sound reasonable to me. So I am trying to figure out 
whether there is sth wrong. If nothing wrong, how to interpret the 
result....Thanks to Dave and anyone else who could provide any insight.
David

-----Original Message-----
From: David Bell <dcbell@iupui.edu>
To: statalist@hsphsun2.harvard.edu
Sent: Wed, Jul 7, 2010 4:13 pm
Subject: Re: st: Subgroup analysis


David-

How about collinearity issues in your African American subsample?  
Twenty-eight
predictors would make me nervous.

Dave
====================================
David C. Bell
Professor of Sociology
Indiana University Purdue University Indianapolis (IUPUI)
(317) 278-1336
====================================




On Jul 7, 2010, at 3:35 PM, David Bai wrote:

> Thank you, Dave. The African American group has 600 cases, and there 
are 28
predictors in the model. The same 28 predictors are used for all 
subgroup
analysis.
> David B
>
>
> -----Original Message-----
> From: David Bell <dcbell@iupui.edu>
> To: statalist@hsphsun2.harvard.edu
> Sent: Wed, Jul 7, 2010 3:18 pm
> Subject: Re: st: Subgroup analysis
>
>
> So how large is your African American subsample?  Maybe your power is 
much
lower
> in that subsample.  Or maybe your model is well specified for Whites 
but not
for
> African Americans.
>
> Dave
> ====================================
> David C. Bell
> Professor of Sociology
> Indiana University Purdue University Indianapolis (IUPUI)
> (317) 278-1336
> ====================================
>
>
>
>
> On Jul 7, 2010, at 2:35 PM, David Bai wrote:
>
> > Hi, all,
> >    I would like to compare predictors' effects across different
> racial/ethnic
> groups, so I first ran a comprehensive model including all groups, 
and then
used
> subpop function in stata to do subpopulation analyses for each ethnic 
group.
> >   What the results show is that many (not just a few) significant
> predictors
> in the comprehensive model (including all racial groups) become
non-significant
> in the African American group, while the White group's results are 
similar to
> the results of the comprehensive model. How can I interpret all 
these? Is it
> possibly because African Americans are very homogeneous in the 
distributions
of
> these predictors, and therefore it is hard for the analysis to 
distinguish any
> variations in the effects and therefore find non-significance in the 
results?
Or
> is it because the sample size of this group is relatively small 
compared with
> other groups (e.g., whites)  in the sample? Are there any other 
possible
> interpretations? Your insight will be appreciated.
> >     David B

*
*   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/