get.or.results: Odds Ratios for Cluster Bootstrapped Multinomial Logistic...

In jenniferthompson/ClusterBootMultinom: Multinomial Logistic Regression Using Clustered Bootstrapping

Description

From a matrix of coefficient estimates from B vglm() model fits on cluster bootstrapped data, calculate and plot odds ratios and 95 used within multi.plot.ors(), but can be called explicitly. Note: does not currently support group effects of variables with nonlinear terms; will calculate one odds ratio per coefficient.

Usage

get.or.results(coef.matrix, remove.vars = NULL, round.vars = NULL,
  round.digits = NULL, out.strings = NULL)

Arguments

coef.matrix	Matrix of coefficient estimates, with rows = number of bootstrapped data sets, columns = number of coefficients.
remove.vars	Character vector of variable names to **not** include in calculations/plots. Defaults to NULL (show all variables).
round.vars	Character vector of variable names whose results should be rounded to something other than two decimal places. Useful for variables with very small changes in odds for one-unit change in variable. Defaults to NULL.
round.digits	Integer; number of digits to round [round.vars] to.
out.strings	List of character strings to label outcome comparisons. Defaults to B vs. A, C vs. A, etc. Note that onus is on the user to supply correct labels.

Value

Data frame with one record per coefficient, including odds ratio estimate, lower and upper confidence limits, character string of results (format: "OR (LCL, UCL)") and text describing comparison.

See Also

vglm, which this function assumes you are using; multi.plot.ors, which calculates p-values and plots all results using ggplot2.

Examples

## Create data frame
df <- data.frame(id = sample(1:20, size = 100, replace = TRUE),
                 x1 = rnorm(n = 100),
                 x2 = rbinom(p = 0.75, n = 100, size = 1),
                 y = sample(LETTERS[1:3], size = 100, replace = TRUE))
df <- df[order(df$id),]
df$time <- unlist(lapply(1:length(unique(df$id)),
                         FUN = function(idnum){ 1:nrow(df[df$id == unique(df$id)[idnum],]) }))

## Using create.sampdata(), generate list of cluster bootstrapped data sets
bootdata.list <- create.sampdata(org.data = df,
                                 id.var = 'id',
                                 n.sets = 25)

## Fit model to original and bootstrapped data frame, saving errors and warnings to .txt file
boot.fits.a <- multi.bootstrap(org.data = df,
                               data.sets = bootdata.list,
                               ref.outcome = grep('A', levels(df$y)),
                               multi.form = as.formula('y ~ x1 + x2'))

## Create matrices of coefficients for all bootstrap fits
boot.matrix.a <- do.call(rbind,
                         lapply(boot.fits.a$boot.models,
                                FUN = function(x){ x@coefficients }))

## Get odds ratios and CIs for x2
ors <- get.or.results(boot.matrix.a, remove.vars = 'x1')

jenniferthompson/ClusterBootMultinom documentation built on May 19, 2019, 4:03 a.m.