[R-sig-ME] ZIP MCMCglmm - how to increase effective sample size?

[dani]

Dear list members,

I need some advice regarding this ZIP MCMCglmm model:

library(MCMCglmm)

priori <- list(R=list(V=diag(2), nu=0.002,fix=2),
               G=list(G1=list(V=diag(2), n=2),G2=list(V=diag(2), n=2)))

mj <- MCMCglmm(y ~ trait - 1 + at.level(trait,1):(x1+x2+x3+x4+ x5 +x6+x7+ offset),
                  random = ~idh(trait):group1 + idh(trait):group2,
                  family = "zipoisson",
                  prior = priori,
                  rcov = ~idh(trait):units,
               verbose=FALSE,
               thin   = 100,
               burnin = 3000,
               nitt   = 103000,
               saveX=TRUE, saveZ=TRUE, saveXL=TRUE, pr=TRUE, pl=FALSE,
               data = s25h)

summary(mj)

# Iterations = 3001:102901
# Thinning interval  = 100
# Sample size  = 1000
#
# DIC: 4811.791
#
# G-structure:  ~idh(trait):group1
#
#                                post.mean l-95% CI u-95% CI eff.samp
# traity.group1       0.4307   0.1351   0.9281    10.17
# traitzi_y. group1  4.3196   2.1216   7.4310    31.26
#
# ~idh(trait):group2
#
#                              post.mean l-95% CI u-95% CI eff.samp
# traity. group2       0.4233   0.2341   0.6781    30.81
# traitzi_y. group2    3.5497   1.2365   6.1525    26.39
#
# R-structure:  ~idh(trait):units
#
#                            post.mean l-95% CI u-95% CI eff.samp
# traity.units      0.02393 0.002833  0.06621    10.58
# traitzi_y.units   1.00000 1.000000  1.00000     0.00
#
# Location effects: y ~ trait - 1 + at.level(trait, 1):(x1 + x2 + x3 + x4 + x5 + x6 + x7 + offset)
#
#                                                            post.mean   l-95% CI   u-95% CI eff.samp  pMCMC
# traity                                               -4.3823820 -6.1496186 -2.6424402   23.592 <0.001 ***
# traitzi_y                                           3.4696204  2.6430476  4.1392235    1.922 <0.001 ***
# at.level(trait, 1):x1                         -0.0498043 -0.2192051  0.1097667   16.979  0.522
# at.level(trait, 1):x2M                     -0.2088408 -0.4535085  0.0440055    8.727  0.088 .
# at.level(trait, 1):x31                       0.1422342 -0.1473884  0.4199985   11.521  0.288
# at.level(trait, 1):x4                         0.0007054 -0.0030953  0.0043456   24.299  0.680
# at.level(trait, 1):x5                         0.1131704  0.0647184  0.1676469   26.621 <0.001 ***
# at.level(trait, 1):x6                        -0.0128734 -0.0483344  0.0306350   13.599  0.588
# at.level(trait, 1):x7                         0.0102356 -0.0276141  0.0540893   40.746  0.680
# at.level(trait, 1):offset                   1.3511873  0.6963525  2.1299075   13.216 <0.001 ***
#   ---
#   Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

I would like to increase my effective samples, but I am not sure which way to go. I tried increasing the NITT to 503000, but the effective samples actually got worse. Is there anything else I could do? I plan on dropping some variables from the model, but if I were to proceed with the model above, what could I have done better?

Thanks in advance!
DNM
<http://aka.ms/weboutlook>

	[[alternative HTML version deleted]]