Model fit diagnostics

The function spDiag calculates measurements of model fit for objects of class NNGP and PGLogit.

Usage

spDiag(object, sub.sample, ...)

Arguments

object

an object of class NNGP or PGLogit.

sub.sample

an optional list that specifies the samples to included in the computations. Valid tags are start, end, and thin. Given the value associated with the tags, the sample subset is selected using seq(as.integer(start), as.integer(end), by=as.integer(thin)). The default values are start=floor(0.5*n.samples), end=n.samples and thin=1. If sub.samples is not specified, then it is taken from object, or, if not aviable in object the default values of start, end, and thin are used. Note, if the object is a NNGP response model and n is large, then computing the replicated data needed for GPD and GRS can take a long time.

...

currently no additional arguments.

Value

A list with the following tags:

DIC

a data frame holding Deviance information criterion (DIC) and associated values. Values in DIC include DIC the criterion (lower is better), D a goodness of fit, and pD the effective number of parameters, see Spiegelhalter et al. (2002) for details.

GPD

a data frame holding D=G+P and associated values. Values in GPD include G a goodness of fit, P a penalty term, and D the criterion (lower is better), see Gelfand and Ghosh (1998) for details.

GRS

a scoring rule, see Equation 27 in Gneiting and Raftery (2007) for details.

WAIC

a data frame hold Watanabe-Akaike information criteria (WAIC) and associated values. Values in WAIC include LPPD log pointwise predictive density, P.1 penalty term defined in unnumbered equation above Equation (11) in Gelman et al. (2014), P.2 an alternative penalty term defined in Equation (11), and the criteria WAIC.1 and WAIC.2 (lower is better) computed using P.1 and P.2, respectively.

y.rep.samples

if y.rep.samples in object were not used (or not available), then the newly computed y.rep.samples is returned.

y.fit.samples

if y.fit.samples in object were not used (or not available), then the newly computed y.fit.samples is returned.

s.indx

the index of samples used for the computations.

References

Finley, A.O., A. Datta, S. Banerjee (2022) spNNGP R Package for Nearest Neighbor Gaussian Process Models. Journal of Statistical Software, doi:10.18637/jss.v103.i05 .

Gelfand A.E. and Ghosh, S.K. (1998). Model choice: a minimum posterior predictive loss approach. Biometrika, 85:1-11.

Gelman, A., Hwang, J., and Vehtari, A. (2014). Understanding predictive information criteria for Bayesian models. Statistics and Computing, 24:997-1016.

Gneiting, T. and Raftery, A.E. (2007). Strictly proper scoring rules, prediction, and estimation. Journal of the American Statistical Association, 102:359-378.

Spiegelhalter, D.J., Best, N.G., Carlin, B.P., van der Linde, A. (2002). Bayesian measures of model complexity and fit (with discussion). Journal of the Royal Statistical Society, Series B., 64:583-639.

Author

Andrew O. Finley finleya@msu.edu,
Sudipto Banerjee sudipto@ucla.edu