From: Strubbe Diederik <diederik.strubbe_at_ua.ac.be>

Date: Wed, 02 Jun 2010 12:56:00 +0200

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. Received on Wed 02 Jun 2010 - 11:04:43 GMT

Date: Wed, 02 Jun 2010 12:56:00 +0200

(I have no idea how to simulate a zero-inflated overdispersed Poisson variable, but the data used can be found at http://www.ua.ac.be/main.aspx?c=diederik.strubbe&n=23519).

First, I create a (strong) pattern in the dataset by: data$LFAP200 <- data$LFAP200 + (data$duck*data$duck)

I try to analyze these data by fitting several possible distributions (Poisson PO, zero-inflated Poisson ZIP, negative binomial type I and type II NBI NBII and zero-inflated negative binomial ZINBI) while using cubic splines with a df=3. The best fitting model will then be choses on the basis of its AIC.

However, these models frequently fail to converge, and I am not sure why, and what to do about it. For example:

> model_Poisson <- gamlss(duck ~ cs(HHCDI200,df=3) + cs(HHCDI1000,df=3) + cs(HHHDI200,df=3) + cs(HHHDI1000,df=3) + cs(LFAP200,df=3),data=data,family= PO)

GAMLSS-RS iteration 1: Global Deviance = 1350.623
GAMLSS-RS iteration 2: Global Deviance = 1350.623

> model_ZIPoisson <- gamlss(duck ~ cs(HHCDI200,df=3) + cs(HHCDI1000,df=3) + cs(HHHDI200,df=3) + cs(HHHDI1000,df=3) + cs(LFAP200,df=3),data=data,family= ZIP)

GAMLSS-RS iteration 1: Global Deviance = 326.3819 GAMLSS-RS iteration 2: Global Deviance = 225.1232 GAMLSS-RS iteration 3: Global Deviance = 319.9663Error in RS() : The global deviance is increasing Try different steps for the parameters or the model maybe inappropriate In addition: There were 14 warnings (use warnings() to see them)

> model_NBI <- gamlss(duck ~ cs(HHCDI200,df=3) + cs(HHCDI1000,df=3) + cs(HHHDI200,df=3) + cs(HHHDI1000,df=3) + cs(LFAP200,df=3),data=data,family= NBI)

GAMLSS-RS iteration 1: Global Deviance = 291.8607
GAMLSS-RS iteration 2: Global Deviance = 291.3291
######......######

GAMLSS-RS iteration 4: Global Deviance = 291.1135
GAMLSS-RS iteration 20: Global Deviance = 291.107
Warning message:

In RS() : Algorithm RS has not yet converged

> model_NBII <- gamlss(duck ~ cs(HHCDI200,df=3) + cs(HHCDI1000,df=3) + cs(HHHDI200,df=3) + cs(HHHDI1000,df=3) + cs(LFAP200,df=3),data=data,family= NBII)

GAMLSS-RS iteration 1: Global Deviance = 284.5993
GAMLSS-RS iteration 2: Global Deviance = 281.9548
######......######

GAMLSS-RS iteration 5: Global Deviance = 280.7311
GAMLSS-RS iteration 15: Global Deviance = 280.6343

> model_ZINBI <- gamlss(duck ~ cs(HHCDI200,df=3) + cs(HHCDI1000,df=3) + cs(HHHDI200,df=3) + cs(HHHDI1000,df=3) + cs(LFAP200,df=3),data=data,family= ZINBI)

GAMLSS-RS iteration 1: Global Deviance = 1672.234 GAMLSS-RS iteration 2: Global Deviance = 544.742 GAMLSS-RS iteration 3: Global Deviance = 598.9939Error in RS() : The global deviance is increasing Try different steps for the parameters or the model maybe inappropriate

Thus, in this case, only the Poisson (PO) and Negative Binomial type I (NBI)converge whereas all other models fail…

My first approach was to omit the smoothing factors for each model, or further reduce the number of variables but this does not solve the problem and most models fail, often yielding a “Error in RS() : The global deviance is increasing” message.

I would think that, given the fact that the dependent variable is zero-inflated and overdispersed, that the Zero-Inflated Negative Binomial (ZINBI) distribution would be the best fit, but the ZINBI even fails in the following very simple examples.

> model_ZINBI <- gamlss(duck ~ cs(LFAP200,df=3),data=data,family= ZINBI)

GAMLSS-RS iteration 1: Global Deviance = 3508.533 GAMLSS-RS iteration 2: Global Deviance = 1117.121 GAMLSS-RS iteration 3: Global Deviance = 652.5771 GAMLSS-RS iteration 4: Global Deviance = 632.8885 GAMLSS-RS iteration 5: Global Deviance = 645.1169Error in RS() : The global deviance is increasing Try different steps for the parameters or the model maybe inappropriate

> model_ZINBI <- gamlss(duck ~ LFAP200,data=data,family= ZINBI)

GAMLSS-RS iteration 1: Global Deviance = 3831.864 GAMLSS-RS iteration 2: Global Deviance = 1174.605 GAMLSS-RS iteration 3: Global Deviance = 562.5428 GAMLSS-RS iteration 4: Global Deviance = 344.0637 GAMLSS-RS iteration 5: Global Deviance = 1779.018Error in RS() : The global deviance is increasing Try different steps for the parameters or the model maybe inappropriate

Any suggestions on how to proceed with this?

Many thanks in advance,

Diederik

Diederik Strubbe

Evolutionary Ecology Group

Department of Biology

University of Antwerp

Groenenborgerlaan 171

2020 Antwerpen, Belgium

tel: +32 3 265 3464

[[alternative HTML version deleted]]

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. Received on Wed 02 Jun 2010 - 11:04:43 GMT

Archive maintained by Robert King, hosted by
the discipline of
statistics at the
University of Newcastle,
Australia.

Archive generated by hypermail 2.2.0, at Wed 02 Jun 2010 - 12:10:27 GMT.

*
Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-help.
Please read the posting
guide before posting to the list.
*