[Rd] multicollinearity

[vincze orsolya]

Dear R users,

I'm analysing some data, and I'm using an lme function.
 I have a problem with choosing the  right order for three of my explanatory variables, which shows collinearity. Is there any rules to make the decision?(r.squared?) Or it's better if I choose the order,  that I think gives me more information about the data?
 
Say x1 is the variable with the highest r.squared, x3 is with the lowest.
If i use
      m1=lme(y~x1+x2+x3,...)
 x2, and x3 is not significant,

 but if i use 
       m2=lme(y~x2+x3+x1, ...) 
all of the 3 variable is significant.

 I would prefer the the m2, because it gives me more ionformation about the dat, but in this case I have to leave in the model x2 and x3, which causes the increase in AIC.

What's the solution?
Can anybody help me?

Cheers




________________________________________________________
„Olyan cikkeket akarunk, amelyek közelebb viszik az országot ahhoz, hogy népbutítás és alantas ösztönök helyett végiggondolt gondolatok irányítsák.” – komment.hu
http://ad.adverticum.net/b/cl,1,6022,318025,391319/click.prm