set_zeta: Calculate a value for the hyperparameter on the global...

Description Usage Arguments Details Value References See Also

View source: R/set_zeta.r


This function can be used to calculate reasonable values for the hyperparameter zeta, which controls the scale (and median) of the half-Cauchy prior on the global smoothing parameter for the latent field of trend parameters of an spmrf model.


set_zeta(yvec, mvec = NULL, linkfun = "identity", ncell = NULL,
  upBound = NULL, alpha = 0.05, order = 1)



A vector of observations on the original scale of measurement.


For binomial reponse variables only. Is a vector of 'trials' associated with each observed number of 'successes' represented in yvec.


The link function associated with the transformation of the expected value of the response (in a generalized linear models sense). Current options are "identity", "log", "logit", and "probit".


The number of grid cells. If there is only one observation per grid location (e.g., observation time or covariate value), then this is equal to the total number of observations. Otherwise is equal to the number of unique location values.


Upper bound on the expected value of the marginal standard deviations of the latent trend (field) parameters. This value is rarely known a priori and here is assumed to equal the standard deviation of the observed data unless otherwise specified.


The probability of exceeding upBound.


The order of the SPMRF model (1, 2, or 3).


Making alpha smaller will decrease the size of zeta, which will result in smoother latent trends if the information in the data does not overcome the prior information.

The methods for calculation of the hyperparameter zeta are outlined in Faulkner and Minin (2017) and are based on methods introduced by Sorbye and Rue (2014) for setting hyperparameters for the precision of Gaussian Markov random field priors.


A numeric scalar value for the hyperparmeter zeta, where zeta > 0.


Faulkner, J. R., and V. N. Minin. 2017. Locally adaptive smoothing with Markov random fields and shrinkage priors. Bayesian Analysis advance publication online.

Sorbye, S. and H. Rue. 2014. Scaling intrinsic Gaussian Markov random field priors in spatial modelling. Spatial Statistics 8:39-51.

See Also


faulknerjam/bnps documentation built on Oct. 18, 2018, 12:03 p.m.