Date: Sun, 3 Nov 2002 07:56:09 -0800
Reply-To: Gregor <gregor.gorjanc@BFRO.UNI-LJ.SI>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Gregor <gregor.gorjanc@BFRO.UNI-LJ.SI>
Organization: http://groups.google.com/
Subject: The negative of the Hessian is not positive definite
Content-Type: text/plain; charset=ISO-8859-1
Hello!
I've posted this mail last week but nobody helped me. So I will try my luck
again.
I'm peroforming generalized linear model analysis with GENMOD
procedure in SAS package. My data (measurements from comet assay) is
supposed to come from gamma distribution. I've used identity link
instead of inverse which is the default canonical). Model contained
concentration (4 levels). I have two different datasets which come
from two similar experiments. I tried to model this effect with class
as well with regression. It worked fine with class for both datasets,
but I had problems with fitting regression in second dataset. The
warning message (WARNING: The negative of the Hessian is not positive
finite....) is produced. I'm sending the log and basic statistics
bellow.
Can anyone help me. Gregor
***************************************************************************
206 proc genmod data=brezE;
207 model tm1_olive = conc / dist=gamma link=id type3 wald;
208 run;
NOTE: Algorithm converged.
NOTE: The scale parameter was estimated by maximum likelihood.
NOTE: PROCEDURE GENMOD used:
real time 3.07 seconds
cpu time 0.11 seconds
209 proc genmod data=zE;
210 model tm1_olive = conc / dist=gamma link=id type3 wald;
211 run;
WARNING: The negative of the Hessian is not positive definite. The
convergence is questionable.
WARNING: The procedure is continuing but the validity of the model fit
is questionable.
WARNING: The specified model did not converge.
WARNING: Negative of Hessian not positive definite.
NOTE: The scale parameter was estimated by maximum likelihood.
NOTE: PROCEDURE GENMOD used:
real time 3.57 seconds
cpu time 0.05 seconds
The MEANS Procedure
Analysis Variable : tm1_olive
N Mean Std Dev Minimum Maximum
403 0.3853605 0.3139503 0.000100000 1.9701000
The MEANS Procedure
Analysis Variable : tm1_olive
N
conc Obs N Mean Std Dev Min Max
0 100 100 0.2057000 0.1388431 0.0001000
0.600100
10 100 100 0.3119000 0.2313438 0.000100
1.140100
50 100 100 0.3837000 0.2216218 0.03010
1.020100
100 103 103 0.6327214 0.4140142 0.000100
1.970100
With regrads,
Gregor
