bootstrap.t.CI.lme: Bootstrapping for CIs and hypothesis test on kfunclme objects

In robertbagchi/ReplicatedPointPatterns: Analysis of replicated point patterns with covariates

Description

Carries out bootstrapping on kfunclme objects to get confidence intervals on parameter estimates and predictions.

bootstrap.compare.lme compares two nested model fits across a range of distances to determine the loss in explanatory power of a covariate and tests the null hypothesis of no effect of the covariate.

Usage

bootstrap.t.CI.lme(mods, lin.comb.Ct, nboot, alpha, ncore = 1, transform
= NULL)

bootstrap.compare.lme(mods, term, dists, nboot, ncore)
bootstrap.parallel.lme(mods, resids, lin.comb.Ct, nboot, ncore = 1)
compare.mods.bootstrap(modH0, modH1, res.r)

Arguments

mods	The models for each distance r.
lin.comb.Ct	The model matrix to be multiplied by the fixed effects to get predictions.
nboot	Number of iterations
alpha	Confidence level.
ncore	Number of cores to use in parallel computations
transform	Experimental and not fully implemented - function to transform the response
term	Term in model formula to be dropped in simpler, nested, model.
dists	Distances to be considered in model test.
resids	List of residuals to be used in bootstrapping. Not usually specified by user.
modH0	Simple model
modH1	Complex model
res.r	Residuals after randomisation.

Author(s)

Robert Bagchi Maintainer: Robert Bagchi <[e-mail]>

Examples

##---- Should be DIRECTLY executable !! ----
##-- ==>  Define data, use random,
##--	or do  help(data=index)  for the standard data sets.

## The function is currently defined as
function (mods, lin.comb.Ct, nboot, alpha, ncore = 1, transform = NULL) 
{
    require("abind")
    require("parallel")
    resids <- lapply(mods, residual.homogenise.lme)
    boot.esti <- bootstrap.parallel.lme(mods = mods, resids = resids, 
        lin.comb.Ct = lin.comb.Ct, nboot = nboot, ncore = ncore)
    sample.esti <- sapply(mods, getpars, lin.comb.Ct = lin.comb.Ct, 
        simplify = FALSE)
    boot.pars <- sapply(boot.esti, function(r, est) {
        mapply(function(x, est) {
            t.sim <- (x$beta.r - est$beta.r)/sqrt(diag(x$vcov.r))
            est$beta.r - t.sim * sqrt(diag(est$vcov.r))
        }, x = r, est = est, SIMPLIFY = TRUE)
    }, est = sample.esti, simplify = FALSE)
    boot.fix.cis <- apply(do.call("abind", args = list(what = boot.pars, 
        along = 3)), c(2, 1), quantile, c(alpha/2, 1 - alpha/2))
    sample.fix.cis <- aperm(sapply(sample.esti, function(x) x$beta.r), 
        c(2, 1))
    sample.fix.cis <- array(sample.fix.cis, dim = c(1, dim(sample.fix.cis)))
    modelpars <- abind(list(estimate = sample.fix.cis, boot.fix.cis), 
        along = 1)
    t.score <- lapply(boot.esti, function(bootsamp, obssamp) {
        mapply(function(sim, obs) {
            t.r <- (sim$pred.r - obs$pred.r)/sim$se.pred.r
            return(t.r)
        }, bootsamp, obssamp)
    }, obssamp = sample.esti)
    t.score <- do.call("abind", args = list(what = t.score, along = 3))
    uci <- apply(t.score, c(2, 1), quantile, alpha/2, na.rm = T)
    lci <- apply(t.score, c(2, 1), quantile, 1 - alpha/2, na.rm = T)
    uci <- split(uci, row(uci))
    lci <- split(lci, row(lci))
    CIs <- mapply(function(obs, ucl, lcl) {
        lower.CI <- obs$pred.r - lcl * obs$se.pred.r
        upper.CI <- obs$pred.r - ucl * obs$se.pred.r
        return(list(LCL = lower.CI, UCL = upper.CI))
    }, obs = sample.esti, ucl = uci, lcl = lci, SIMPLIFY = FALSE)
    estimator <- lapply(sample.esti, function(x) x$pred.r)
    lower.CI <- lapply(CIs, function(x) x$LCL)
    upper.CI <- lapply(CIs, function(x) x$UCL)
    bootstrap.CI <- list(estimator = estimator, lower = lower.CI, 
        upper = upper.CI, modelpars = modelpars)
    return(bootstrap.CI)
  }

robertbagchi/ReplicatedPointPatterns documentation built on May 27, 2019, 10:32 a.m.