Re: NaN in Logistic Regression

[John Darrington]

On Sun, Mar 21, 2021 at 06:45:59PM +0000, Ricardo Mejias wrote:
     When running a logistic regression, I found that, under certain situations 
I get NaN symbols where most of the output numbers usually appear.
     
     
     The situation occurs when the independent variable Party20210112Npa is 
used with 13 other variables.
     
     It does not occur when used by itself or with a smaller number of the same 
other variables.
     
     
     The problem never happens when I am not using the variable Party20210112Npa
     
     
     A Description report does not show any missing values of any fields in 
this data.
     
     
     I suspect that this is a problem caused by the iteration process. I know 
that iterative solutions sometimes “get stuck” and cannot provide an answer.
     
     
     Do you agree that this is an iteration problem and if, so do you have any 
suggestions regarding how to avoid it?
     
     
     I can’t send you the data set over email due to its size. However, I am 
not looking for an actual solution to this problem from you, just for a general 
suggestion regarding how to avoid this kind of problem.


If the iterations fail to converge you will get a warning message about this.

But I think it is unlikely to happen unless there is something wrong with the
data.  For example if one of the predictor variables is not dichotomous then
it can cause this sort of problem.

If you compress your data and load it onto a public server, then perhaps
somebody can take a look at it.  But do include the complete syntax of the
command you are running.