R help - Jul 2012 - zeroinfl problem: cannot get standard errors, hessian has NaN

Hi!

I have three models.



In the first model, everything is fine.




However, in the second and third models, I have NA's for standard errors:





The hessians also have NaN's (same for m2 and m3). 



What should I do about it? It there a way to obtain the hessian without
transforming my variables? I will greatly appreciate your help!



--
View this message in context:
http://r.789695.n4.nabble.com/zeroinfl-problem-cannot-get-standard-errors-hessian-has-NaN-tp4637715.html
Sent from the R help mailing list archive at Nabble.com.

Hi,

This is not nearly enough information.  Please follow the posting
guide and provide us with a reproducible example, or at the bare
minimum, the code for your models.  Without more details we can only
wildly guess, but here are a few:

---You have more parameters than your data can support (the Hessian is
the matrix of partial second derivatives, and SEs are typically based
closely on the hessian (or rather the information matrix, the negative
of the hessian).
---you have some redundant effects?

Try a simpler model and see if you can find some particular parameter
that leads to the problem.  You could also try tweaking optimization
options, but I that is usually not the source of the problem.

We really need more information, with the ideal being putting data
that reproduces your problem (or your actual data) online somewhere
where we can easily download it, and providing us with all your model
code, so on our machines, we can exactly reproduce what is happening
for you and try to troubleshoot.

Cheers,

Josh

On Tue, Jul 24, 2012 at 10:50 PM, nikkks <nikita at sagowarrior.com>
wrote:
> Hi!
>
> I have three models.
>
>
>
> In the first model, everything is fine.
>
>
>
>
> However, in the second and third models, I have NA's for standard
errors:
>
>
>
>
>
> The hessians also have NaN's (same for m2 and m3).
>
>
>
> What should I do about it? It there a way to obtain the hessian without
> transforming my variables? I will greatly appreciate your help!
>
>
>
> --
> View this message in context:
http://r.789695.n4.nabble.com/zeroinfl-problem-cannot-get-standard-errors-hessian-has-NaN-tp4637715.html
> Sent from the R help mailing list archive at Nabble.com.
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.


-- 
Joshua Wiley
Ph.D. Student, Health Psychology
Programmer Analyst II, Statistical Consulting Group
University of California, Los Angeles
https://joshuawiley.com/

It's hard to say what exactly goes wrong without a reproducible example (see
the posting guide). However, a typical source of such problems is that some
regressors are scaled badly. Maybe you have one regressor that takes very large
values or squares of a regressor that takes moderately large values? The result
would be that the associated coefficients become tiny and the hessian is
conditioned badly. The simple solution is to scale the problematic regressor so
that it's coefficient is reasonably sized (e.g., divide by 1000 or 10000 or
so).


(sent from mobile phone)

-------- Original message --------
Subject: [R] zeroinfl problem: cannot get standard errors, hessian has NaN 
From: nikkks <nikita@sagowarrior.com> 
To: r-help@r-project.org 
CC:  

Hi!

I have three models.



In the first model, everything is fine.




However, in the second and third models, I have NA's for standard errors:





The hessians also have NaN's (same for m2 and m3). 



What should I do about it? It there a way to obtain the hessian without
transforming my variables? I will greatly appreciate your help!



--
View this message in context:
http://r.789695.n4.nabble.com/zeroinfl-problem-cannot-get-standard-errors-hessian-has-NaN-tp4637715.html
Sent from the R help mailing list archive at Nabble.com.

______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

	[[alternative HTML version deleted]]

Thank you, Achim! Rescaling worked and all is fine now!





--
View this message in context:
http://r.789695.n4.nabble.com/zeroinfl-problem-cannot-get-standard-errors-hessian-has-NaN-tp4637715p4637732.html
Sent from the R help mailing list archive at Nabble.com.