Date: Tue, 22 Oct 2002 17:34:53 -0700
Reply-To: Dale McLerran <stringplayer_2@YAHOO.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Dale McLerran <stringplayer_2@YAHOO.COM>
Subject: Re: FW: PROC NLMIXED Error Message
In-Reply-To: <4047227F34B1D51185B5000476727BEBD24289@roadrunner.uchc.edu>
Content-Type: text/plain; charset=us-ascii
Mike (and Narmatha),
It might be instructive to try fitting the model using the QPOINTS
option. This allows one to control the number of quadrature points
used by the procedure. I have not tried controling the number of
quadrature points directly in SAS, but several years ago when
investigating quadrature methods through simulation using some
other software, I found that the quadrature methods could be very
unstable when the size of the random effects is small. For odd
values of the number of quadrature points, convergence might be
achieved but not for an even number of quadrature points. Also,
for a larger number of quadrature points, the parameter estimates
and their standard errors would inflate to infinity. As I say,
this was in a situation where the random effects were small to
begin with. But my impression from these investigations was that
quadrature could not be relied upon for studies in which one does
not have large random effects. It is better to use the GLIMMIX
macro in such cases.
Fundamentally, it would seem that the integral that Narmatha is
attempting to fit through quadrature is poorly defined. The sort
of cycling which I observed in my simulations is probably occuring.
I don't know exactly how accuracy is measured, but the value 1.0
sounds suspiciously to me like no tendency toward stability
between successive numbers of quadrature points. In other words,
the sort of cycling between convergence and nonconvergence may
be occuring, of the estimates may be inflating with increased
number of quadrature points.
Dale
--- "Todd, Mike" <todd@PSYCHIATRY.UCHC.EDU> wrote:
> SAS-Lers:
>
> I'm forwarding this note on behalf of a non-subscriber, so be sure to
> Cc:
> her on with your replies, Dale...er...I mean, y'all. ;^) Thanks.
>
> -Mike
>
> ><><><><><><><><><><><><><><><><><><><><
> Michael Todd, Ph.D.
> University of Connecticut Health Center
> Alcohol Research Center-MC2103
> Department of Psychiatry
> 263 Farmington Avenue
> Farmington CT 06030-2103
>
> Voice: 860.679.5468
> Fax: 860.679.5464
> E-mail: todd@psychiatry.uchc.edu
>
>
>
> -----Original Message-----
> From: Narmatha Thanigasalam [mailto:nthaniga@GPU.SRV.UALBERTA.CA]
> Sent: Tuesday, October 22, 2002 5:18 PM
> To: MULTILEVEL@JISCMAIL.AC.UK
> Subject: PROC NLMIXED Error Message
>
>
> Dear Multilevelers,
>
> I am trying to run a multilevel logistic model using PROC NLMIXED and
> I got
> the following error message:
>
> ERROR: Quadrature accuracy of 0.000100 could not be achieved with 31
> points.
> The achieved accuracy was 1.000000.
>
> I tried increasing the quadrature tolerance (QTOL= option in PROC
> NLMIXED
> statement) and I am still getting the same error message. Am I
> overlooking
> something? What should I check for?
>
> Thanks in advance.
>
> Narmatha
>
>
>
> Narmatha Thanigasalam MSc
> Senior Data Analyst
> Centre for Health Outcomes Research and Utilization Studies
> Department of Public Health Sciences
> University of Alberta
> 13-103 Clinical Sciences Building
> Edmonton, Alberta
> Canada T6G 2G3
>
> Ph. (780) 492-7349
> Fax. (780) 492-0364
> Email: nthaniga@ualberta.ca
>
>
>
>
>
>
> -------------------------- Multilevel list --------------------------
> To leave, send leave multilevel to jiscmail@jiscmail.ac.uk
> For further info about the Multilevel list, please see
> http://www.nursing.teaching.man.ac.uk/staff/mcampbell/multilevel.html
=====
---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@fhcrc.org
Ph: (206) 667-2926
Fax: (206) 667-5977
---------------------------------------
__________________________________________________
Do you Yahoo!?
Y! Web Hosting - Let the expert host your web site
http://webhosting.yahoo.com/
| 
