Re: st: Peer variable coefficient estimate nonsense

[Clyde B Schechter <clyde.schechter@einstein.yu.edu>]

Leaving aside the technical issues brought up by Nick Cox of how correlated variables might partition variance among themselves in a regression analysis, substantively, I wonder why Jian Zhang thinks it is nonsensical for peer age to be a stronger predictor of student score than the student's own age.  It isn't explained what kind of scores and peer groups these are, but if the "peer" group in question is the student's class, it wouldn't surprise me at all that mean peer group age, which is then a very strong proxy for grade level, would be a stronger predictor of, say, math achievement, than the child's individual age when the two are used together.  In fact, it _would_ surprise me if the opposite were true.  After all, we would expect 5th graders to have higher math scores than 4th graders, but there is no reason to think that an older 4th grader would outperform a younger 5th grader.

More generally, there are many other instances where a group attribute is a stronger predictor of an individual outcome than the analogous attribute of the individual. 

Clyde Schechter
Dept. of Family & Social Medicine
Albert Einstein College of Medicine
Bronx, NY, USA



*
*   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/