andrew9lawson / bayesian-dm-code-examples

BDM Code examples from Win/OpenBUGS, CARBayes and INLA

Bayesian-DM-code-examples

BDM Code examples from Win/OpenBUGS, Nimble, CARBayes and INLA This repository is intended as an open source for code examples from a range of packages that fit hierarchical Bayesian models to small area health data. The examples represent a range of applications: spatial; Spatio-temporal; multivariate, multi-scale. Also they represnt a range of packages: MCMC -based packages such as Win/OpenBUGS, CARBayes, Nimble; and LAplace approximation via INLA.

Win/OpenBUGS:

Georgia_multiscale_WBugs.txt; Georgia_MVCAR_WBugs.txt; map_modelWB_pred.txt; R2WB_runfile_pred.txt; Ohio_resp_cancer_ST_Knorr_Held_model_WBugs.txt; Georgia_oral_cancer 2004_Convolution_Wbugs.txt

CARBayes:

carbayes_lipcancer_SCcongen90.txt; SCcongen90_CARBayesRcode

Nimble:

SC_congen_90_CONV_Nimble.txt; SC_congen90_lognormal_Nimble.txt; Gamma_Poisson_toy_data_Nimble.txt; STinT_models_Ohio_Nimble.txt; Georgia_MultiScale_CONV_Nimble.txt;

INLA:

SCcongen_INLA_models.txt; SCcongen_INLA_models_plusBYM.txt; FMD_INLA_ST_Rcode.txt; OHIO_STmapping_INLA_RcodeFAL

bayesian-dm-code-examples's Issues

The problem of binomial distribution modeling

Hi, excuse me. I have met some problems in learning spatio-temporal modeling by INLA.
My data includes year, the total number of trials(N), number of successes, number of failures, rate of success, longitude, latitude, and independent variable (water). as shown in the following figure.

I want to predict the rate of success in each region, Here is my code, but occured an error.

## Formula
myformula <- cbind(success,failure) ~ -1+Intercept+water
  f(spatial.field,
    model = myspde,
    group = spatial.field.group,
    control.group = list(model = "ar1",
                         hyper = myrho_hyper))

myoutput<-inla(myformula,
               data=inla.stack.data(mystack,spde=myspde,
                                    family="binomial(link=logit)"),
               control.predictor = list(A=inla.stack.A(mystack),
                                        compute=TRUE))

##Error in inla(myformula, data = inla.stack.data(mystack, spde = myspde, :
dim(y...orig)[2] == 1 is not TRUE
In addition, I'm confused that you just use count as the dependent variable, does that satisfy the binomial distribution?
Looking forward to your reply. Thank you!
sincerely,
Sue