<html>
<head>
<style><!--
.hmmessage P
{
margin:0px;
padding:0px
}
body.hmmessage
{
font-size: 10pt;
font-family:Tahoma
}
--></style>
</head>
<body class='hmmessage'>
Hi, Hans,<br>thanks for the pointer, unfortunately, when I use a variable that has already been converted to an integer, I get the same error in all cases: "Incompatible array bounds". This occurs whether zeroInflation is TRUE or FALSE. Using the simplest model, with only one fixed variable, both glm.nb{MASS} and zeroinfl{pscl} converge, whereas glmm.ADMB does not. Therefore, I am inclined to ask if, in glmmADMB:<br>1. is there a way to designate an initial value of theta (to produce a geometric rather than a negative binomial distribution)?<br>2. are there constraints in picking a grouping factor?<br>Thank you for the advice,<br>Holly<br><br>glmmadmbZINBtotalCRATdensInt1fixGrpSrvy <- glmm.admb(<br>fixed=TotalCRATdensInt~<br>Dist,<br>group="fSurveyID",<br>data= Data110217noFlNaRVfallsub, <br>zeroInflation=FALSE,<br>family="nbinom")<br><br># Incompatible array bounds in dvar_vector operator * (_CONST dvar_matrix& m,_CONST dvar_vector& x)<br><br>library(MASS)<br><br>glmnb1totalCRATdensInt1fix <- glm.nb(<br>TotalCRATdensInt~<br>Dist,<br>data= Data110217noFlNaRVfallsub,<br>init.theta=1,<br>link = log,<br>control=glm.control(maxit = 25, trace = TRUE))<br><br>library(pscl)<br><br>psclZINB1totalCRATdensInt1fix <- zeroinfl(<br>formula= (TotalCRATdensInt~Dist),<br>dist="geometric",<br>link="logit",<br>data=Data110217noFlNaRVfallsub)<br><br>> Date: Wed, 30 Mar 2011 22:32:58 +0200<br>> Subject: Re: [ADMB Users] Error in a negative binomial, zero-inflated glmmADMB with random effect<br>> From: hskaug@gmail.com<br>> To: hollygoy@hotmail.com<br>> CC: users@admb-project.org<br>> <br>> Hi Holly,<br>> <br>> It seems that "round" cannot be used in the formula. Try to add<br>> a new variable to the dataframe, and use this as the response.<br>> <br>> Hans<br>> <br>> On Tue, Mar 29, 2011 at 5:44 PM, Holly G <hollygoy@hotmail.com> wrote:<br>> > Hello list,<br>> ><br>> > I have been working with a dataset that is large, overdispersed,<br>> > non-normally distributed, non-linear, spatially autocorrelated, and<br>> > zero-inflated. So far, I have been able to account for everything except for<br>> > the zero-inflation using GAMMs in R. To my knowledge, there is no general<br>> > software code available yet to incorporate zero- inflation into negative<br>> > binomial GAMMs, please correct me if I am wrong. As far as GLMMs, I have<br>> > modeled the data with glmmPQL (accounting for random effects and<br>> > autocorrelation structure), and zeroinfl{pscl} (accounting for<br>> > zero-inflation). For a zero-inflated model that includes random effects, I<br>> > am attempting glmm.admb (the example from the documentation ran smoothly),<br>> > but have encountered a few errors, all a variation of “The function<br>> > maximizer failed” (below). I am not sure if these errors are due to the<br>> > nature of the data or to a mistake that I overlooked, so I provide details<br>> > below. I would appreciate any advice. Apparently, the temporary nbmm.dat and<br>> > nbmm.pin files are not being created correctly, but I am unsure as to why or<br>> > how to fix it.<br>> ><br>> > Thank you,<br>> > Holly<br>> ><br> </body>
</html>