SpatioTemporal (version 0.9.2)

tstat: Basic diagnostic statistics and summaries for SpatioTemporal model output.

Description

Calculation of t-statistics and Hessian eigenvalues of model fits. 'tstat' prints out a standard R-style summary of point estimates, their SEs and t-statistics.

Usage

tstat(pardat,alphas=TRUE)

CVbasics(cvout)

Arguments

pardat

List which must include a 'par.all' vector and a square matrix 'hessian.all', of compatible sizes. Typically, this list would be the 'res.best' component of the output of

alphas

logical: should we add the point estimates as well in the return? Defaults to TRUE.

cvout

List, the output of estimateCV, run with the option hessian.all=TRUE.

Value

• tstat returns as a default a 3-column numerical data frame, with parameter names in rows, and point estimates, SEs and t-statistics in columns. If alphas=FALSE, it only returns a vector with the t-statistics. CVbasics returns a list with 3 matrices, each of dimension (# of parameters) X (# of CV groups). The matrix names are 'alphas' with point estimates, 'tees' with t-statistics, and 'eigens' with Hessian eigenvalues. The latter is sorted largest to smallest, rather than to match any particular parameter.

encoding

latin1

Details

These functions enable (until further package upgrade) a quick, standard summary of model performance. The tstat function returns, if alphas=TRUE (default), a 3-column summary commonly used in R, with parameter names in rows, and point estimates, SEs and t-statistics in columns. Shown are estimates for both LUR and kriging parameters. The CVbasics function is intended to diagnose cross-validation estimation. Besides point-estimates and t-statistics, the eigenvalues of the Hessian matrix (inverse of the parameter variance estimates) are also calculated and returned.

See Also

fit.mesa.model, estimateCV.

Examples

##load a model object
data(mesa.data.model)

##Set up initial parameter values for optimization
dimm <- loglike.dim(mesa.data.model)
x.init <- as.matrix(cbind(rep(2,dimm$nparam.cov),c(rep(c(1,-3),dimm$m+1),-3)))

################ tstat example
				
 ##estimate parameters
##This may take a while...
par.est <- fit.mesa.model(x.init, mesa.data.model, type="p",
      hessian.all=TRUE, control=list(trace=3,maxit=1000))
##Let's load precomputed results instead.
data(mesa.data.res)
par.est <- mesa.data.res$par.est

#### summary

tstat(par.est$res.best)