R help - Feb 2016 - MCMCglmm and iteration

Hi everyone,
I'm running a bayesian regression using the package MCMCglmm (Hadfield 2010)
and to reach a normal posterior distribution of estimates, I increased the
number of iteration as well as the burnin threshold. However, it had unexpected
outcomes. Although it improved posterior distribution, it also increased
dramatically the value of estimates and decrease DIC.?
Here an example:
>head(spring)
pres ? large_road ? ?small_road ? ?cab0 ? ? ? ? ? ? 2011 ? ? ? ? ? ? ? ?32 ? ? ?
? ? 781 ? ? ? ? ? ? ?102 ? ? ? ? ? ? ? 179 ? ? ? ? ?2040 ? ? ? ? ? ? ?1256 ? ? ?
? ? ? 654 ? ? ? ? ?9841 ? ? ? ? ? ? ?187 ? ? ? ? ? ? ? 986 ? ? ? ? ?7560 ? ? ? ?
? ? ? 21 ? ? ? ? ? ? ? ?438 ? ? ? ? ? 571 ? ? ? ? ? ? ? 13 ? ? ? ? ? ? ? ? ? 5 ?
? ? ? ? 439 ? ?
>#pres is presence/absence data and other variable are distance to these
features
>## with?200,000 iteration and 30,000 burnin>prior <- list(R = list(V =
1, nu=0.002))>sp.simple <- MCMCglmm(pres ~ large_road + cab +
small_road,?family = "categorical", nitt = 200000, thin = 200, burnin
= 30000,?????????????????????????????????????????????data = spring,?prior =
prior, verbose = FALSE, pr = TRUE)
>summary(sp.simple)
?Iterations = 30001:199801?Thinning interval ?= 200?Sample size ?= 850?
?DIC: 14045.31?
?R-structure: ?~units
? ? ? ? ?post.mean ? ? l-95% ? ? CI u-95% ? ? CI eff.sampunits ? ? 294.7 ? ? ? ?
1.621 ? ? ? ?621.9 ? ? ? ? ? ?1.982
?Location effects: pres ~ large + cab + small + Coupe_0_5 + Regeneration +
Res_mature + DH + Autre + Eau + Pert_nonregen + MF + Coupe_6_20?
? ? ? ? ? ? ? ? ? ? ?post.mean ? ? l-95% ? ? ? CI ?u-95% ? ? CI eff.samp ? ?
pMCMC ??(Intercept) ? ? 5.76781 ? ? ?0.77622 ? ? ? 9.24375 ? ? ? ? ?1.829 ? ? ?
? ? ?<0.001 **large ? ? ? ? ? ? 0.37487 ? ? ?0.02692 ? ? ? 0.75282 ? ? ? ? ?
3.310 ? ? ? ? ? ?<0.001 **cab ? ? ? ? ? ? ? 0.94639 ? ? ?0.09906 ? ? ?
1.57939 ? ? ? ? ? 2.096 ? ? ? ? ? ?<0.001 **small ? ? ? ? ? -1.62192 ? ?
-2.60873 ? ? ?-0.20191 ? ? ? ? ? 2.002 ? ? ? ? ? ?<0.001 **

>## with?1,000,000 iteration and 500,000 burnin>prior <- list(R =
list(V = 1, nu=0.002))>sp.simple <- MCMCglmm(pres ~ large_road + cab +
small_road,?family = "categorical", nitt = 1000000, thin = 200, burnin
= 500000,?????????????????????????????????????????????data = spring,?prior =
prior, verbose = FALSE, pr = TRUE)
>summary(sp.simple)
?Iterations = 500001:999801?Thinning interval ?= 200?Sample size ?= 2500?
?DIC: 858.6316?
?R-structure: ?~units
? ? ? ? ?post.mean ? ?l-95% ? CI u-95% ? ? CI eff.sampunits ? ? 26764 ? ? ?17548
? ? ?34226 ? ? ? ? ? ? 124.5
?Location effects: pres ~ large_road + cab + small_road?
? ? ? ? ? ? ? ? ? ?post.mean ? ?l-95% ? ?CI u-95% ? ?CI eff.samp ? ? ?pMCMC ?
?(Intercept) ? ? 60.033 ? ? ?47.360 ? ? 70.042 ? ? ? ? ?137.9 ? ? ? ? ?
?<4e-04 ***large_road ? ? ?3.977 ? ? ? ?1.279 ? ? ? 6.616 ? ? ? ?1484.6 ? ? ?
? ? ?0.0080 **?cab_road ? ? ? ?9.913 ? ? ? ?6.761 ? ? 13.020 ? ? ? ? ?333.7 ? ?
? ? ? ?<4e-04 ***small ? ? ? ? ? -16.945 ? ? -20.694 ? ?-13.492 ? ? ? ?
?194.9 ? ? ? ? ? ?<4e-04 ***

?
I'm then wandering if it is because more iteration produce better estimates
and then a model that had a better fit with the data.
Anyone can help me??

R?mi


	[[alternative HTML version deleted]]

R?mi Lesmerises <remilesmerises <at> yahoo.ca> writes:
>  Hi everyone, I'm running a bayesian regression using the package
> MCMCglmm (Hadfield 2010) and to reach a normal posterior
> distribution of estimates, I increased the number of iteration as
> well as the burnin threshold. However, it had unexpected
> outcomes. Although it improved posterior distribution, it also
> increased dramatically the value of estimates and decrease DIC.?
> Here's an example:
head(spring)

pres large_road  small_road  cab 
0      2011         32         78 
1       102        179        204 
0      1256        654        984 
1       187        986        756 
0        21        438         57 
1        13          5        439 

# pres is presence/absence data and other variable are distance to these
features
# with?200,000 iteration and 30,000 burnin

prior <- list(R = list(V = 1, nu=0.002))
sp.simple <- MCMCglmm(pres ~ large_road + cab + small_road,
   family = "categorical", nitt = 200000, thin = 200,
    burnin = 30000, data = spring,?prior = prior, verbose = FALSE, pr = TRUE)
------------

(1) you will do much better with this kind of question on r-sig-mixed-models.
(2) it looks like your chain is mixing very, very badly.  If I'm reading
the output correctly, it looks like your effective sample sizes for the
first run (200K iterations) are 1-3 (!) -- you should be aiming for 
effective sample sizes of 100s to 1000s.  Even with a million iterations
you're only getting up to effective sample sizes of ~150 for some
parameters.  I would recommend (a) centring and scaling your parameters
to improve mixing and (b) cross-checking with a different method
(e.g. lme4 or glmmADMB) to make sure you're in the right ballpark.

  You shouldn't necessarily expect a Normal posterior as you increase
the number of iterations; the posterior distributions are only 
asymptotically Normal as the number of *observations* increases.