[R-sig-ME] effective sample size in MCMCglmm
dani
orchidn at live.com
Tue Oct 10 19:44:37 CEST 2017
Hello everyone,
My question is:
do the effective samples I obtain in my MCMCglmm output (attached below) make sense?
I understand that the rule of thumb is to get effective samples of at least 100-1000. How should I tweak the thin, burnin, and the nitt specifications? My computer reaches its memory limit fast and I have barely been able to run the model below.
I have the following model:
k<-12 # number of fixed effects
prior2<-list(B=list(V=diag(k)*1e4, mu=rep(0,k)),
R=list(V=1, nu=0),
G=list(G1=list(V=1, nu=0),
G2=list(V=1, nu=0),
G3=list(V=1, nu=0)))
prior2$B$mu[k]<-1
prior2$B$V[k,k]<-1e-4
m3new <- MCMCglmm(y ~ f_newage_c+x2n+x8n+x9n+x5n+l_lfvcspo+x3n+x4n+x6n+x7n+offset,
random =~ studyid+class+idv(l_lfvcspn),
data = wo1,
family = "poisson", prior=prior2,
verbose=FALSE,
thin = 10,
burnin = 2000,
nitt = 200000,
saveX=TRUE, saveZ=TRUE, saveXL=TRUE, pr=TRUE, pl=TRUE)
Iterations = 2001:199991
Thinning interval = 10
Sample size = 19800
DIC: 2930.006
G-structure:
~studyid
post.mean l-95% CI u-95% CI eff.samp
studyid 0.1053 1.814e-11 0.5757 81.12
~class
post.mean l-95% CI u-95% CI eff.samp
class 0.7008 0.07577 1.207 382.1
~idv(l_lfvcspn)
post.mean l-95% CI u-95% CI eff.samp
l_lfvcspn. 705.7 37.33 2044 11852
R-structure:
~units
post.mean l-95% CI u-95% CI eff.samp
units 2.516 1.809 3.23 336.6
Location effects: y ~ f_newage_c + x2n + x8n + x9n + x5n + l_lfvcspo + x3n + x4n + x6n + x7n + offset
post.mean l-95% CI u-95% CI eff.samp pMCMC
(Intercept) -7.0395427 -7.5590206 -6.5187847 865.6 <5e-05 ***
f_newage_c 0.0099703 -0.0222981 0.0448880 3324.7 0.5615
x2nM -0.1068377 -0.3782251 0.1678528 3760.2 0.4462
x8n1 0.4103047 0.0920875 0.7179638 3884.3 0.0099 **
x9n1 -0.2784715 -0.5975232 0.0495615 3337.9 0.0901 .
x5n -0.0009378 -0.0064175 0.0044266 3528.4 0.7283
l_lfvcspo 0.4018810 -0.8271349 1.4468375 14536.6 0.4080
x3n 0.0789652 -0.0018683 0.1523108 3726.0 0.0438 *
x4n 0.0602655 -0.0643903 0.1859443 2711.9 0.3356
x6n -0.0137132 -0.0728385 0.0449804 3489.7 0.6453
Thank you all so much,
Dani
<http://aka.ms/weboutlook>
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list