s-news
[Top] [All Lists]

Problems with the CensorReg function

To: <s-news@wubios.wustl.edu>
Subject: Problems with the CensorReg function
From: Helena Mouriño <mhnunes@fc.ul.pt>
Date: Tue, 28 Nov 2006 16:07:05 -0000
Thread-index: AccTBz76uRI0pvCIR4aLpSC4sDHr6w==
Dear S-news maling list members,
 
 
I need to model the amount of chlorophyll-a in Portuguese waters as a function of the upwelling pulses felt in the previous days to the day when water samples were collected, water stratification and temperature at the surface of the ocean. As chlorophyll-a cannot take negative values, I thought that it would be a good idea to fit a Gaussian truncated regression model at the point zero. To attain this goal I've used the "censorReg" function in SPLUS 6.0, namely:
 
    adjustment<-censorReg(censor(dados.fito$clorophyll) ~     
    dados.fito$Stratification+dados.fito$Temperature+dados.fito$upwell0days+dados.fito$upwell1day+dados.fito$upwell2days+dados.fito$upwell3days+
    dados.fito$upwell4days+dados.fito$upwell5days+dados.fito$upwell6days+dados.fito$upwell7days,
    data="">
 
Unfortunately, this function did not work properly with my data set because I got a few warning messages:
 
          1: problems in covariance matrix computations; error code iervcv = 1 in: censorReg.mlest(Yi, censor.codesi, Xi, casewti, distribution, link,      
           parameter.fixed = parameter.fixed, fixed = fixed, control = ....

2: Maximum change in parameter does not change the likelihood in: censorReg.mlest(Yi, censor.codesi, NULL, casewti, distribution, link, parameter.fixed = parms.fixed, fixed = fix2p, control = control, ....

3: Information matrix is not positive definite in: censorReg.mlest(Yi, censor.codesi, NULL, casewti, distribution, link, parameter.fixed = parms.fixed, fixed = fix2p, control = control, ....

Has anybody experience with such warnings? I don't know what is the effect of such pitfalls in the estimated parameters.
 
 
I would also like to ask if anyone has ever tried to run a stepwise function afterwards? The reason I'm asking for this is because I would like to run a stepwise function (or a similiar one) to select a statistical significant subset of those variables. I've already tried the functions step, step.glm but neither of them worked properly. I get the following message:
 
    problems in names(z) <- names(wz) <- yn: supplied 207 names for a vector of lenght 414. Use traceback() to see the call stack in: remove(list).       
    Warning massages: Problem: internal error: trying to set internal exit action in frame , must be between 1 and 0.
 
After writing the command  "traceback()" in the command prompt, the SPLUS program aborts!
 
 
Has anybody ever had similar problems? Can anyone provide some hints on how to get through this matter?
 
 
Thanks in advanced.
 
 
Best regards,
 
Helena Mouriño.

Upgrade Your Email - Click here!

<Prev in Thread] Current Thread [Next in Thread>
  • Problems with the CensorReg function, Helena Mouriño <=