Hello
I am attempting to run a structural break Weibull duration model with two
parameters for some social science (cabinet) duration data. I have a
small number of observations (13) and 6 covariates (though my problem
exists for any number of covariates I try).
I can compile the model, load the data, and load/gen inits without issue.
I cannot, however, avoid numerical overflow (undefined real result) traps
(I cannot even obtain one update without trap).
As per the winBUGS manual, I have tried:
1) using specified/reasonable priors (I obtained these from a slightly
simpler model, with more observations, for a lower number of covariates),
as in my inits below
2) standardisation of covariates to have mean 0 and standard deviation 1
(as in my data below)
3) put reasonable bounds on the distribution of Y (the duration)
3) using UpdaterSlice rather than DFreeARS
My model, inits and data follow.
Any help or advice is grately appreciated.
Arthur
model
{
for(i in 1:N){
Y[i] ~ dweib(r, mu[i])I(150,1900)
log(mu[i])<-m[i]## regression structure for covariates
m[i]<-alpha0
+alpha[1]*atts[i]
+ alpha[2]*maj[i]
+ alpha[3]*fract[i]
+ alpha[4]*pol[i]
+alpha[5]*seats[i]
+alpha[6]*votes[i]
+ J[i]*beta0
+ J[i]*beta[1]*atts[i]
+ J[i]*beta[2]*maj[i]
+ J[i]*beta[3]*fract[i]
+ J[i]*beta[4]*pol[i]
+J[i]*beta[5]*seats[i]
+J[i]*beta[6]*votes[i]
J[i] <- step(yr[i]-cp-eps);
}
for(i in 1:g){alpha[i] ~ dnorm(0,1.0E-6)}
for(i in 1:g){beta[i] ~ dnorm(0,1.0E-6)}
for(i in 1:T) {
priort[i] <- punif[i]/sum(punif[])
}
alpha0 ~ dnorm(0,1.0E-6)
beta0 ~ dnorm(0,1.0E-6)
cp ~ dcat(priort[])
r~dgamma(1,1.0E-3)
eps<-0.5
}
inits{
list(alpha=c(-19,9,-0.01,300,0,0),
beta=c(14,-16,0.04,265,0,0),
alpha0=18,
beta0=-20,
cp=10
)
}
##data
##all covariates have mean=0, sd=1
list(Y=c(1673, 604, 1307, 1595, 1832, 531, 1540, 1349, 219,
1610, 1497, 1463, 934),
atts=c(-0.716435813139072, -0.301342083389757, -0.794265861574658,
-1.20935957567396, -1.15747295288861, 0.217524978626755, -
0.768322561902083,
-0.379172147384419, 2.00761678167354, 1.46280633367525, 0.269411861231133,
0.658562071238475, 0.710448969507407),
maj=c(-0.716435822143307, -0.301342086325116, -0.794265923501635,
-1.20935967496984, -1.15747286455923, 0.217525183617782, -
0.768322506576223,
-0.379172172124362, 2.00761682583084, 1.46280622527176, 0.269411734209298,
0.658562273171539, 0.710448808098491),
fract=c(-0.645030899468411, -0.261636394513209, -0.516382537217905,
-0.777860722846884, -1.34066922032107, -0.463776103071283, -
0.88638719710475,
-0.267924814410628, 1.83344004750904, 1.65120105592168, 0.637093921578761,
-0.0898466717851229, 1.12777953572978),
pol=c(1.58671725865179, -0.919619660823142, -0.697697870724915,
0.117234223980945, 0.825104341051577, -0.851776144092821,
0.754916942620654,
-0.414667525585798, -1.50903175716309, -0.894441095902639, -
0.269628145402186,
1.49352949302410, 0.779359940365521),
seats=c(0.45495277259305, 0.391879881604168, 0.605308920144101,
0.822155559975886, 1.31015637515749, 0.618054174378085, 0.85668034587467,
0.319084368420061, -1.59604245101827, -1.73772821556916, -
0.627773501089918,
-0.117524511080664, -1.29920371938950),
votes=c(-0.629533336732271, -0.272644920429723, -0.528175842000219,
-0.783375286316831, -1.34539469287088, -0.477046851761567, -
0.885361776206008,
-0.272476190766197, 1.78484981014816, 1.66957095628813, 0.647686868196159,
-0.067302233736413, 1.15920349618766),
N=13,
T=13,
yr=c(1,2,3,4,5,6,7,8,9,10,11,12,13),
punif=c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1),
g=6))
-------------------------------------------------------------------
This list is for discussion of modelling issues and the BUGS software.
For help with crashes and error messages, first mail [log in to unmask]
To mail the BUGS list, mail to [log in to unmask]
Before mailing, please check the archive at www.jiscmail.ac.uk/lists/bugs.html
Please do not mail attachments to the list.
To leave the BUGS list, send LEAVE BUGS to [log in to unmask]
If this fails, mail [log in to unmask], NOT the whole list
