07-09-2016 07:56 PM
I’m looking at intra-individual variability for three variables (rate life in the past, present and the future) using the Proc Mixed command in SAS. My main predictors are Sample( nationality:Japan vs. U.S.), 5 Age groups and three time points (past, present and future). One of my level 2 variables(extraversion personality trait, which I am adjusting as a covariate) correlates very differently with the three main variables. My model only works when I also include this level 2 variable (Extraversion) in the random statement. I know we only put level 1 variables (slopes) in the random statements and not level 2 variables so I am not sure why the model only works when I put the level 2 variable in the random statement.
proc mixed noclprint covtest ; class sample AgeGroup5 time marriedyn gender ; model ratelife = gender NewChronNumb marriedyn extraversionnew|time NeuroticismNew ZEduHighest sample|time|AgeGroup5 /solution ddfm=bw; random intercept extraversionnew / type=un subject=id; lsmeans sample*time*agegroup5; run;
Thank you for your help in advance!
07-11-2016 02:07 PM
Just guessing here, but because extraversionnew and sample are both crossed with time, it may be that you have something like complete collinearity between some of the variables in your model. The other possibility is that you are overfitting your available data.