02-29-2012 07:12 PM
I have ran a model using PROC LOGISTIC. However, SAS reported the message that: WARNING: The information matrix is singular and thus the convergence is questionable.
I am wondering how seriousness this problem is. Should I just ignore this warning messsage and use the results or is there any other procedure that I need to use?
02-29-2012 09:00 PM
Do not ignore that warning. It can occur when you have many independent variables and some are determined by the others (or nearly). For example, you may have categorical factor SITE with values "siteA" and "siteB" and continuous factor TEMPERATURE with value 20 at siteA and 25 at siteB. Then the two factors contain basically the same information and if you include them both in your regression, SAS will hit a mathematical problem and issue a warning. This is called overdetermination, it is only one kind of situation that can cause the warning you got. More importantly, the parameter estimates that were produced may be strongly biased and their standard errors grossly inflated. It is better that you identify the source of the problem and get rid of the warning.
PROC FREQ or PROC TABULATE are good tools for this kind of investigation.