st: Omitted variables due to collinearity in xtmixed

["Hauschildt, Kristina" <Hauschildt@his.de>]

Dear list, 

I have a problem understanding something that is happening in my analyses using -xtmixed- in Stata SE/12.1.

This is the command I use (simplified):

xtmixed depvar i.level1var1  i.level1var2 level1var3 level2var1 level2var2 level2var3 i.yearslevel1##c.level2var4 ||level2id: yearslevel1, var mle

When I run this, one interaction term (specifically: 5.yearslevel1#c.level2var4) gets dropped because of collinearity. I have trouble finding out where this collinearity could be coming from.  
What confuses me even more is the fact that the model runs fine with a different dependent variable. The only other difference in that model is that years has one level less, i.e., starts one year later because the dependent wasn't measured in the first year. All other predictors are exactly the same.

I have tried forcing Stata to include the interaction term  by adding "collinear" at the end of the command, but this doesn't help - once I input the command, the note regarding collinearity immediately appears nevertheless. 

I would appreciate any input -either ways to include the interaction term, or reasons for why this could be happening...

Btw: I am aware that normally I would need to set the covariances free by adding cov(uns) when including a random slope. If I do this, however, the model does not converge at all. 

Thanks for reading! 

Kind regards
Kristina Hauschildt




*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/