02-19-2017 11:32 AM
I have to run a linear regression model where the residual don't have constant variance. I know there are different techniques to account for it but it is asked me to use PROC MIXED which allow non-constant variances. There are no repeated measures and my independent variable X is continuous. The code I wrote is the following:
proc mixed data=mydata; model Y = X / solution; repeated / group=X; run;
Is that right? It gives me back a warning of infinite likelihood maybe due to:
- too many variance parameters to estimates
- non variation on certain values of X
- my code is wrong
In addition, do I have do use scatterwhite degrees of freedom?