ordinal_superiority.bracl {brglm2}

Ordinal superiority scores of Agresti and Kateri (2017)

Description

ordinal_superiority() is a method for the estimation and inference about model-based ordinal superiority scores introduced in Agresti and Kateri (2017, Section 5) from fitted objects. The mean bias of the estimates of the ordinal superiority scores can be corrected.

Usage

## S3 method for class 'bracl'
ordinal_superiority(
  object,
  formula,
  data,
  measure = c("gamma", "Delta"),
  level = 0.95,
  bc = FALSE
)

Arguments

object

a fitted object from an ordinal regression model. Currently only models from class "bracl" are supported.

formula

a RHS formula indicating the group variable to use.

data

an optional data frame in which to look for variables with which to compute ordinal superiority measures. If omitted, an attempt is made to use the data that produced object.

measure

either "gamma" (default) or "Delta", specifying the ordinal superiority measure to be returned.

level

the confidence level required when computing confidence intervals for the ordinal superiority measures.

bc

logical. If FALSE (default) then the ordinal superiority measures are computed using the estimates in object. If TRUE then the ordinal superiority measure estimates are corrected for mean bias.

References

Agresti, A., Kateri, M. (2017). Ordinal probability effect measures for group comparisons in multinomial cumulative link models. Biometrics, 73 214-219. doi:10.1111/biom.12565.

Examples

data("stemcell", package = "brglm2")

# Adjacent category logit (proportional odds)
stem <- within(stemcell, {nreligion = as.numeric(religion)})
fit_bracl_p <- bracl(research ~ nreligion + gender, weights = frequency,
                     data = stem, type = "ML", parallel = TRUE)

# Estimates and 95% confidence intervals for the probabilities that the response
# category for gender "female" is higher than the response category for gender "male",
# while adjusting for religion.
ordinal_superiority(fit_bracl_p, ~ gender)

## Not run: 
# And their (very-similar in value here) bias corrected versions
# with 99% CIs
ordinal_superiority(fit_bracl_p, ~ gender, bc = TRUE, level = 0.99)
# Note that the object is refitted with type = "AS_mean"


## End(Not run)