MCMCglmm and iteration behaviour (new attempt)

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

MCMCglmm and iteration behaviour (new attempt)

Rémi Lesmerises
Here a new attempt in trying to improve the visual of my request:

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  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)

>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.samp
units   294.7     1.621    621.9          1.982

Location effects: pres ~ large_road + cab + small_road

               post.mean   l-95%       CI    u-95%     CI    eff.samp    pMCMC
(Intercept)    5.76781     0.77622     9.24375     1.829       <0.001 **
large_road     0.37487     0.02692     0.75282     3.310       <0.001 **
cab            0.94639     0.09906     1.57939     2.096       <0.001 **
small_raod    -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.samp
units     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            9.913        6.761      13.020       333.7     <4e-04 ***
small_raod   -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?


Rémi Lesmerises
Université du Québec à Rimouski

______________________________________________
[hidden email] mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.