02-19-2017 08:57 AM - edited 02-19-2017 08:58 AM
I am facing a case of heteroskedasticity of the errors in linear regression. I know there are different ways to account for it, but my professor told me to use PROC MIXED which can handle non-cosntant variance.
In case of categorical X2 the code should be the following:
proc mixed data=mydata; class X2; model Y = X2 / ddfm=satterth; repeated / group=X2; run;
Which iterations converge without problems.
But my independent variable X1 is continuous and I don't know if the code I wrote is right:
proc mixed data=mydata; model Y = X1; repeated / group=X1; run;
Which gives me back the warning of infinite likelihood, probably because of:
- too many parameters to estimate.
- no variation is some values of X.
- my code is not right.
Hence, in summary I don't have repeated measures, I just want a model which take into account different variances of the residuals.