turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Why the Poisson distribution is outputting wrong e...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

08-07-2014 11:48 AM

Hi,

The QQ Plot suggests that the studentized residuals of the response variables are not normally distributed.

When I run it with a poisson distributions (or others) it is outputting completely unfeasible estimates. It should be around 0.20 (which would be 20%) and it is estimating negative numbers.

I thought could be scaling issues (multiplied by 100) but it is not.

I have tried different estimation methods too (LaPlace, Quadrature). Not sure if it could be the linkfunction?

Any thoughts?

Thanks,

Marcio

Accepted Solutions

Solution

08-07-2014
03:24 PM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to marcioadg

08-07-2014 03:24 PM

All Replies

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to marcioadg

08-07-2014 11:58 AM

Without your data or code its almost impossible to say

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to Reeza

08-07-2014 12:02 PM

Thanks Reeza,

Please let me know if this helps. Note this is only one of the codes I tried. Thanks!

proc glimmix data=sows method=quad plots=residualpanel(conditional marginal) ;

CLASS week WeighGr90 ;

MODEL CV = WeighGr90 / dist=poisson ;

*random _residual_ / group=vargroup ;

random intercept / subject=week ;

lsmeans WeighGr90/pdiff;

output out=igausout pred=p resid=r STUDENT=std;

run;

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to marcioadg

08-07-2014 01:11 PM

The default link function for the Poissonis log. So, I think all your graphics are on the log linear predictor scale. You can add the ilink option to the plots statement to get the inverse line (original scale).

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

08-07-2014 03:14 PM

Thanks lvm, but the problem are the lsmeans estimates, even with the ilink option on the lsmeans statement it is retuning the negative values. Any thoughts?

Solution

08-07-2014
03:24 PM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Posted in reply to marcioadg

08-07-2014 03:24 PM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

08-07-2014 03:57 PM

lvm,

Using the ilink on the lsmeans statement was actually giving the right information, but you are right I was only looking on the left side of the table (so used to it). It is giving me on the right side like you said, the information that I needed. Thanks a lot! I really appreciate it! Problem solved.