11-15-2011 02:21 PM
I have a tricky question regarding a dataset. The dataset has 4 categorical variables with various categories. The trick thing is that they have one identical level (ie. 100% percectly correlated for this level).
This caused a problem for running a linear model. SAS said the matrix is singular and the estimate is not unique.
Is there any good idea to deal with this level?
Thanks in advance.
11-15-2011 03:44 PM
Are you saying that, for one variable, every record has the same value and/or that everyone in the same group has the same value? If so, why not just drop that variable from your analysis?
11-16-2011 10:01 AM
Thanks for your suggestion, art297.
The variables are not 100% perfectly correlated. Just a level of them. So I cannot remove variables.
I found a way to deal with that:
1) recode the levels of variables into 0, -1, 1 (effect coding);
2) Only keep one of the correlated levels in the final model.
11-15-2011 04:10 PM
I'm going to guess that all variables in the model are class variables and that the "problem" is a note that says:
|Note:||The X'X matrix has been found to be singular, and a generalized inverse was used to solve the normal equations. Terms whose estimates are followed by the letter 'B' are not uniquely estimable.|
If so, there's nothing wrong with your model or data. You just need to understand how to interpret the parameter estimates. See this usage note: http://support.sas.com/kb/22/585.html
If this isn't the problem, post the error message that you get.