st: Subgroup analysis or dummy interaction?

[Daniel Albert <daniel.albert@unisg.ch>]

Hi all,

I am currently working on a bibliographic data sample examining the 
effects of number of authors (AU) on article citation count (CC). In 
particular I argued for an inverted u shaped effect of number of authors

Besides many control variables I also coded articles in my sample as 
article type 0 or type 1 (dummy variable).

I then ran my OLS regression (including AU and AU_squared) for the (a) 
full sample; (b) the type "0" sub-sample; (c) and the type "1" sub-sample. 


While there was only a positive linear AU effect significant for the full 
sample and type "0" sub-sample, the squared effect became significant for 
the type "1" sub-sample.

However, if I run a one sample analysis with the interaction term type*AU 
respectively type*AU_squared the effect is not significant anymore. Same 
is true for the "xi" command.

My question now is, whether a one sample approach or the sub-sample 
analysis is the right approach? I was wondering if in the one sample 
approach the different effects might outweigh each other and make the 
effect apparently insignificant. 

The full sample contains approx 6000 observations and the type "1" 
constitutes approx 1400.

I mcentered all independent variables (except the dummy) togehter. Should 
I do this for the sub sample analysis instead?

Thank you very much for your help!

Daniel
*
*   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/