f: Define general Gaussian models in the INLA formula

In inbo/INLA: Full Bayesian Analysis of Latent Gaussian Models using Integrated Nested Laplace Approximations

Description

Function used for defining of smooth and spatial terms within inla model formulae. The function does not evaluate anything - it exists purely to help set up a model. The function specifies one smooth function in the linear predictor (see inla.list.models) as

weight*f(var)

Usage

    f(...,
         model = "iid",
         copy=NULL,
         same.as = NULL,
         n=NULL,
         nrep = NULL,
         replicate = NULL,
         ngroup = NULL,
         group = NULL,
         control.group = inla.set.control.group.default(),
         hyper = NULL,
         initial=NULL,
         prior=NULL,
         param = NULL,
         fixed = NULL,
         season.length=NULL,
         constr = NULL,
         extraconstr=list(A=NULL, e=NULL),
         values=NULL,
         cyclic = NULL,
         diagonal = NULL,
         graph=NULL,
         graph.file=NULL,
         cdf=NULL,
         quantiles=NULL,
         Cmatrix=NULL,
         rankdef=NULL,
         Z = NULL,
         nrow = NULL,
         ncol = NULL,
         nu = NULL,
         bvalue = NULL,
         spde.prefix = NULL,
         spde2.prefix = NULL,
         spde2.transform = c("logit", "log", "identity"),
         spde3.prefix = NULL,
         spde3.transform = c("logit", "log", "identity"),
         mean.linear = inla.set.control.fixed.default()$mean,
         prec.linear = inla.set.control.fixed.default()$prec,
         compute = TRUE,
         of=NULL,
         precision = exp(14),
         range = NULL,
         adjust.for.con.comp = TRUE,
         order = NULL,
         scale = NULL,
         strata = NULL,
         rgeneric = NULL,
         scale.model = NULL,
         args.slm = list(rho.min = NULL, rho.max = NULL, 
                         X = NULL, W = NULL, Q.beta = NULL),
         args.ar1c = list(Z = NULL, Q.beta = NULL),
         args.intslope = list(subject = NULL, strata = NULL, covariates = NULL), 
         correct = NULL,
         locations = NULL, 
         debug = FALSE)

Arguments

...	Name of the covariate and, possibly of the weights vector. NB: order counts!!!! The first specified term is the covariate and the second one is the vector of weights (which can be negative).
model	A string indicating the choosen model. The default is iid. See names(inla.models()$latent) for a list of possible alternatives and inla.doc for detailed docs.
copy	TODO
same.as	TODO
n	An optional argument which defines the dimension of the model if this is different from length(sort(unique(covariate)))
nrep	TODO
replicate	We need to write documentation here
ngroup	TODO
group	TODO
control.group	TODO
hyper	Specification of the hyperparameter, fixed or random, initial values, priors and its parameters. See ?inla.models for the list of hyparameters for each model and its default options or use inla.doc() for detailed info on the family and supported prior distributions.
initial	THIS OPTION IS OBSOLETE; use hyper!!! Vector indicating the starting values for the optimization algorithm. The length of the vector depends on the number of hyperparamters in the choosen model. If fixed=T the value at which the parameters are fixed is determines through initial. See inla.models()$latent$'model name' to have info about the choosen model.
prior	THIS OPTION IS OBSOLETE; use hyper!!! Prior distribution(s) for the hyperparameters of the !random model. The default value depends on the type of model, see !www.r-inla.org for a detailed description of the models. See names(inla.models()$priors) for possible prior choices
param	THIS OPTION IS OBSOLETE; use hyper!!! Vector indicating the parameters a and b of the prior distribution for the hyperparameters. The length of the vector depends on the choosen model. See inla.models()$latent$'model name' to have info about the choosen model.
fixed	THIS OPTION IS OBSOLETE; use hyper!!! Vector of boolean variables indicating wheater the hyperparameters of the model are fixed or random. The length of the vector depends on the choosen model See inla.models()$latent$'model name' to have info about the choosen model.
season.length	Lenght of the seasonal compoment (ONLY if model="seasonal")
constr	A boolean variable indicating whater to set a sum to 0 constraint on the term. By default the sum to 0 constraint is imposed on all intrinsic models ("iid","rw1","rw1","besag", etc..).
extraconstr	This argument defines extra linear constraints. The argument is a list with two elements, a matrix A and a vector e, which defines the extra constraint Ax = e; for example extraconstr = list(A = A, e=e). The number of columns of A must correspond to the length of this f-model. Note that this constraint comes additional to the sum-to-zero constraint defined if constr = TRUE.
values	An optional vector giving all values assumed by the covariate for which we want estimated the effect. It must be a numeric vector, a vector of factors or NULL.
cyclic	A boolean specifying wheather the model is cyclical. Only valid for "rw1" and "rw2" models, is cyclic=T then the sum to 0 constraint is removed. For the correct form of the grah file see Martino and Rue (2008).
diagonal	An extra constant added to the diagonal of the precision matrix.
graph	Defines the graph-object either as a file with a graph-description, an inla.graph-object, or as a (sparse) symmetric matrix.
graph.file	THIS OPTION IS OBSOLETE AND REPLACED BY THE MORE GENERAL ARGUMENT graph. PLEASE CHANGE YOUR CODE. Name of the file containing the graph of the model; see www.r-inla.org/faq.
cdf	A vector of maximum 10 values between 0 and 1 x(0), x(1),…. The function returns, for each posterior marginal the probabilities Prob(X<x(p))
quantiles	A vector of maximum 10 quantiles, p(0), p(1),… to compute for each posterior marginal. The function returns, for each posterior marginal, the values x(0), x(1),… such that Prob(X<x)=p
Cmatrix	The specification of the precision matrix for the generic, generic3 or z models (up to a scaling constant). Cmatrix is either a (dense) matrix, a matrix created using Matrix::sparseMatrix(), or a filename which stores the non-zero elements of Cmatrix, in three columns: i, j and Qij. In case of the generic3 model, it is a list of such specifications.
rankdef	A number defining the rank deficiency of the model, with sum-to-zero constraint and possible extra-constraints taken into account. See details.
Z	The matrix for the z-model
nrow	Number of rows for 2d-models
ncol	Number of columns for 2d-models
nu	Smoothing parameter for the Matern2d-model, possible values are c(0, 1, 2, 3)
bvalue	TODO
spde.prefix	TODO
spde2.prefix	TODO
spde2.transform	TODO
spde3.prefix	TODO
spde3.transform	TODO
mean.linear	Prior mean for the linear component, only used if model="linear"
prec.linear	Prior precision for the linear component, only used if model="linear"
compute	A boolean variable indicating wheather the marginal posterior distribution for the nodes in the f() model should be computed or not. This is usefull for large models where we are only interested in some posterior marginals.
of	TODO
precision	The precision for the artifical noise added when creating a copy of a model and others.
range	A vector of size two giving the lower and upper range for the scaling parameter beta in the model COPY, CLINEAR, MEC and MEB. If low = high then the identity mapping is used.
adjust.for.con.comp	If TRUE (default), adjust some of the models (currently: besag, bym, bym2 and besag2) if the number of connected components in graph is larger than 1. If FALSE, do nothing.
order	Defines the order of the model: for model ar this defines the order p, in AR(p). Not used for other models at the time being.
scale	A scaling vector. Its meaning depends on the model.
strata	Currently not in use
rgeneric	A object of class inla.rgeneric which defines the model. (EXPERIMENTAL!)
scale.model	Logical. If TRUE then scale the RW1 and RW2 and BESAG and BYM and BESAG2 and RW2D models so the their (generlized) variance is 1. Default value is inla.getOption("scale.model.default")
args.slm	Required arguments to the model="slm"; see the documentation for further details.

,

args.ar1c

Required arguments to the model="ar1c"; see the documentation for further details.

,

args.intslope

A list with the subject (factor), strata (factor) and covariates (numeric) for the intslope model; see the documentation for further details.

,

correct

Add this model component to the list of variables to be used in the corrected Laplace approximation? If NULL use default choice, otherwise correct if TRUE and do not if FALSE. (This option is currently experimental.)

,

locations	A matrix with locations for the model dmatern. This also defines n.
debug	Enable local debug output

Details

There is no default value for rankdef, if it is not defined by the user then it is computed by the rank deficiency of the prior model (for the generic model, the default is zero), plus 1 for the sum-to-zero constraint if the prior model is proper, plus the number of extra constraints. Oops: This can be wrong, and then the user must define the rankdef explicitely.

Value

TODO

Author(s)

Havard Rue [e-mail]

See Also

inla, hyperpar.inla

inbo/INLA documentation built on Dec. 6, 2019, 9:51 a.m.