Hello-
I am writing an article that involves several models estimated with xtmixed.
My problem is that some of the independent variables have non-linear
relationships with the DV. Normally, I understood that to address this problem
one should include the IV plus the IV squared. However, my variables are all
standardized to mean zero so when I square them that obviously turns them into
garbage.
In addition, some of these variables with non-linear relationships are latent
variables derived from factor analysis so I can't just re-code them into all
positive values. I could create a new index variable from the particular
variables that comprise the loadings but that loses all of the value of
deriving latent variables in terms of error correlations.
In addition, the paper has 9 DVs and most of these IVs are relevant across
several DVs but are only non-linear with a few of them. Therefore I'm thinking
it might get too confusing presentation-wise to include the basic index version
of the IV in some models and the latent variable factor analysis version of the
IV in other models?
The 'simple' solution would be including the basic version of the variables
without any of the squared terms and then mentioning that I ran alternative
specifications. But those models would probably be terribly mis-specified?
Any advice would be much appreciated,
-Rahsaan
Rahsaan Maxwell
Assistant Professor
Department of Political Science
University of Massachusetts, Amherst
http://polsci.umass.edu/profiles/maxwell_rahsaan
----- End forwarded message -----
Rahsaan Maxwell
Assistant Professor
Department of Political Science
University of Massachusetts, Amherst
http://polsci.umass.edu/profiles/maxwell_rahsaan
*
* 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/
