Skip to contents

For any fixed and random effect INLA supports a range of different priors of exponential distributions.

Currently supported for INLA in ibis.iSDM are the following priors that can be specified via "type":

  • "normal" or "gaussian": Priors on normal distributed and set to specified variable. Required parameters are a mean and a precision estimate provided to "hyper". Note that precision is not equivalent (rather the inverse) to typical standard deviation specified in Gaussian priors. Defaults are set to a mean of 0 and a precision of 0.001.

  • "clinear": Prior that places a constraint on the linear coefficients of a model so as that the coefficient is in a specified interval "c(lower,upper)". Specified through hyper these values can be negative, positive or infinite.

  • "spde", specifically 'prior.range' and 'prior.sigma': Specification of penalized complexity priors which can be added to a SPDE spatial random effect added via add_latent_spatial(). Here the range of the penalized complexity prior can be specified through 'prior.range' and the uncertainty via 'prior.sigma' both supplied to the options 'type' and 'hyper'.

Other priors available in INLA names(INLA::inla.models()$prior) ) might also work, but have not been tested!


INLAPrior(variable, type = "normal", hyper = c(0, 0.001), ...)

# S4 method for character,character
INLAPrior(variable, type = "normal", hyper = c(0, 0.001), ...)



A character matched against existing predictors or latent effects.


A character specifying the type of prior to be set.


A vector with numeric values to be used as hyper-parameters. See description. The default values are set to a mean of 0 and a precision of 0.001.


Variables passed on to prior object.


Compared to other engines, INLA does unfortunately does not support priors related to more stringent parameter regularization such as Laplace or Horseshoe priors, which limits the capability of engine_inla for regularization. That being said many of the default uninformative priors act already regularize the coefficients to some degree.


  • Rue, H., Riebler, A., Sørbye, S. H., Illian, J. B., Simpson, D. P., & Lindgren, F. K. (2017). Bayesian computing with INLA: a review. Annual Review of Statistics and Its Application, 4, 395-421.

  • Simpson, D., Rue, H., Riebler, A., Martins, T. G., & Sørbye, S. H. (2017). Penalising model component complexity: A principled, practical approach to constructing priors. Statistical science, 32(1), 1-28.