boot_hopit: Bootstrapping hopit model

In hopit: Hierarchical Ordered Probit Models with Application to Reporting Heterogeneity

Bootstrapping hopit model

Description

boot_hopit performs the bootstrap of a function dependent on a fitted model. In each of the bootstrap repetitions, a set of new model coefficients is drawn from the multivariate normal distribution, assuming the originally estimated model coefficients (see coef.hopit) as a mean and using the model variance-covariance matrix (see vcov.hopit). The drawn coefficients are then used to calculate the measure of interest using a function delivered by the func parameter.

Usage

boot_hopit(
  model,
  func,
  data = model$frame,
  nboot = 500,
  unlist = TRUE,
  boot.only.latent = TRUE,
  parallel.flag = FALSE,
  parallel.nb_cores = NULL,
  parallel.packages = NULL,
  parallel.variables = NULL,
  robust.vcov,
  ...
)

Arguments

model	a fitted hopit model.
func	a function to be bootstrapped of the form func(model, ...).
data	data used to fit the model.
nboot	a number of bootstrap replicates.
unlist	a logical indicating whether to unlist the boot object.
boot.only.latent	a logical indicating whether to perform the bootstrap on latent variables only.
parallel.flag	a logical if to use parallel computations.
parallel.nb_cores	number of cores (<= number of CPU cores on the current host).
parallel.packages	list of packages needed to run "func".
parallel.variables	list of global variables and functions needed to run "func".
robust.vcov	see vcov.hopit.
...	other parameters passed to the func.

Value

a list with bootstrapped elements.

Author(s)

Maciej J. Danko

See Also

percentile_CI, getLevels, getCutPoints, latentIndex, standardiseCoef, hopit.

Examples

# DATA
data(healthsurvey)

# the order of response levels decreases from the best health to
# the worst health; hence the hopit() parameter decreasing.levels
# is set to TRUE
levels(healthsurvey$health)

# fit a model
model1 <- hopit(latent.formula = health ~ hypertension + high_cholesterol +
                heart_attack_or_stroke + poor_mobility + very_poor_grip +
                depression + respiratory_problems +
                IADL_problems + obese + diabetes + other_diseases,
              thresh.formula = ~ sex + ageclass + country,
              decreasing.levels = TRUE,
              control = list(trace = FALSE),
              data = healthsurvey)

# Example 1 ---------------------
# bootstrapping cut-points

# a function to be bootstrapped
cutpoints <-  function(model) getCutPoints(model)$cutpoints
B <- boot_hopit(model = model1, func = cutpoints, nboot = 100)

# calculate lower and upper bounds using the percentile method
cutpoints.CI <- percentile_CI(B)

# print estimated cutpoints and their confidence intervals
cutpoints(model1)
cutpoints.CI

# Example 2 ---------------------
# bootstrapping differences in health levels

# a function to be bootstrapped
diff_BadHealth <- function(model) {
  hl <- getLevels(model = model, formula=~ sex + ageclass, sep=' ')
  hl$original[,1] + hl$original[,2] - hl$adjusted[,1]- hl$adjusted[,2]
}

# estimate the difference
est.org <- diff_BadHealth(model = model1)

# perform the bootstrap
B <- boot_hopit(model = model1, func = diff_BadHealth, nboot = 100)

# calculate lower and upper bounds using the percentile method
est.CI <- percentile_CI(B)

# plot the difference and its (asymmetrical) confidence intervals
pmar <- par('mar'); par(mar = c(9.5,pmar[2:4]))
m <- max(abs(est.CI))
pos <- barplot(est.org, names.arg = names(est.org), las = 3,
               ylab = 'Original - Adjusted',
               ylim=c(-m, m), density = 20, angle = c(45, -45),
               col = c('blue', 'orange'))
for (k in seq_along(pos)) lines(c(pos[k,1],pos[k,1]),
                                est.CI[,k], lwd = 2, col = 2)
abline(h = 0); box(); par(mar = pmar)

hopit documentation built on Oct. 1, 2022, 5:06 p.m.