Random Effects meta-analysis: JAGS quits with errors for data with only two studies

[Dale Steele]

Hi - I'm fitting a series of random effects meta-analyses using JAGS (using R2jags in R). I've pasted the model, the dataset, consisting of only two studies, and the output including error messages. Any ideas why this is failing? Works for comparisons with more than 2 studies and for other priors. Thanks. --Dale

My model is:

model
{
for( i in 1 : Nstud ) {
rA[i] ~ dbin(pA[i], nA[i])
rB[i] ~ dbin(pB[i], nB[i])
logit(pA[i]) <- mu[i]
logit(pB[i]) <- mu[i] + delta[i]
mu[i] ~ dnorm(0.0,1.0E-5)
delta[i] ~ dnorm(d, prec)
}
OR <- exp(d)
d ~ dnorm(0.0,1.0E-6)
prec ~ dgamma(0.001, 0.001)
tau <- 1/sqrt(prec)
tau.sq<-tau*tau
}

library(R2jags)
(c1 <- ombpoL2[[7]]) (R data frame) shown below:

events1 events2 noevents1 noevents2 N1 N2
7 2 71 73 78 75
0 6 147 141 147 147

library(R2jags)
Nstud <- dim(c1)[1]
rA <- c1$events2; rB <- c1$events1; nA <- c1$N2; nB <- c1$N1
data <- list("rA", "rB", "nA", "nB", "Nstud")
params <- c("OR", "d", "delta", "tau", "tau.sq")
init1 <- list("d"=0, "prec"=1, "delta"= rep(0, Nstud), "mu"=rep(0, Nstud))
init2 <- list("d"=0, "prec"=2, "delta"= rep(0, Nstud), "mu"=rep(0, Nstud))
init3 <- list("d"=0, "prec"=0.25, "delta"= rep(0, Nstud), "mu"=rep(0, Nstud))
inits <- list(init1, init2, init3)
load.module("glm")
m.c1 <- jags(data=data, inits=inits, parameters.to.save=params,
n.chains=3, n.iter=15000, n.burnin=1000, n.thin=1, model.file="invgamma.bug")

Output follows:

Compiling model graph
Resolving undeclared variables
Allocating nodes
Graph Size: 30

Initializing model

|********| 100%
Error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...) :
NA/NaN/Inf in 'y'
Error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...) :
NA/NaN/Inf in 'y'
Error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...) :
NA/NaN/Inf in 'y'
Error in if ((W > 1e-08) && (n.chains > 1)) { :
missing value where TRUE/FALSE needed

[Krzysztof Sakrejda]

The error is from R2jags, not JAGS itself. It might help if you run the example
with options(error=recover).

Krzysztof

On Wed, Aug 21, 2013 at 7:11 PM, Dale Steele eleetselad@users.sf.net wrote:

Hi - I'm fitting a series of random effects meta-analyses using JAGS (using
R2jags in R). I've pasted the model, the dataset, consisting of only two
studies, and the output including error messages. Any ideas why this is
failing? Works for comparisons with more than 2 studies and for other
priors. Thanks. --Dale

My model is:

model
{
for( i in 1 : Nstud ) {
rA[i] ~ dbin(pA[i], nA[i])
rB[i] ~ dbin(pB[i], nB[i])
logit(pA[i]) <- mu[i]
logit(pB[i]) <- mu[i] + delta[i]
mu[i] ~ dnorm(0.0,1.0E-5)
delta[i] ~ dnorm(d, prec)
}
OR <- exp(d)
d ~ dnorm(0.0,1.0E-6)
prec ~ dgamma(0.001, 0.001)
tau <- 1/sqrt(prec)
tau.sq<-tau*tau
}

library(R2jags)
(c1 <- ombpoL2[[7]]) (R data frame) shown below:

events1 events2 noevents1 noevents2 N1 N2
7 2 71 73 78 75
0 6 147 141 147 147

library(R2jags)
Nstud <- dim(c1)[1]
rA <- c1$events2; rB <- c1$events1; nA <- c1$N2; nB <- c1$N1
data <- list("rA", "rB", "nA", "nB", "Nstud")
params <- c("OR", "d", "delta", "tau", "tau.sq")
init1 <- list("d"=0, "prec"=1, "delta"= rep(0, Nstud), "mu"=rep(0, Nstud))
init2 <- list("d"=0, "prec"=2, "delta"= rep(0, Nstud), "mu"=rep(0, Nstud))
init3 <- list("d"=0, "prec"=0.25, "delta"= rep(0, Nstud), "mu"=rep(0,
Nstud))
inits <- list(init1, init2, init3)
load.module("glm")
m.c1 <- jags(data=data, inits=inits, parameters.to.save=params,
n.chains=3, n.iter=15000, n.burnin=1000, n.thin=1,
model.file="invgamma.bug")

Output follows:

Compiling model graph
Resolving undeclared variables
Allocating nodes
Graph Size: 30

Initializing model

|**| 100%
Error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...) :
NA/NaN/Inf in 'y'
Error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...) :
NA/NaN/Inf in 'y'
Error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...) :
NA/NaN/Inf in 'y'
Error in if ((W > 1e-08) && (n.chains > 1)) { :
missing value where TRUE/FALSE needed

---

Random Effects meta-analysis: JAGS quits with errors for data with only two
studies

---

Sent from sourceforge.net because you indicated interest in
https://sourceforge.net/p/mcmc-jags/discussion/610037/

To unsubscribe from further messages, please visit
https://sourceforge.net/auth/subscriptions/

--

Krzysztof Sakrejda

Organismic and Evolutionary Biology
University of Massachusetts, Amherst
319 Morrill Science Center South
611 N. Pleasant Street
Amherst, MA 01003

work #: 413-325-6555
email: sakrejda@cns.umass.edu

---