ordinalTools is a package for working with output from ordinal regression models. As currently implemented, the package uses output from ordinal logistic regressions fit using the `polr()’ function from the package MASS to calculate means based on user-specified covariate constrasts. Standard errors are estimated either using the delta method or bootstrapping.

• The package contains a single function named ordinal_means() where the only required argument is object, an object of class polr.

• Methods for standard generics are provided, i.e., coef(), summary(), confint(), and print().

• Bootstrap standard errors and confidence intervals can be estimated using se.type=“bootstrap” and by specifying the data frame used to fit the original polr regression model.

• If two contrasts are specified, the function will also calculate the difference in means based on the two contrasts as well as its standard error and confidence interval.

Let y denote an ordinal outcome, and x1 and x2 covariates. An ordinal logistic regression model model with y as outcome and x1 and x2 as covariates is fitted using the polr package

library(MASS)
polr.fit <- polr(y ~ x1 + x2, data = DF, Hess=TRUE)

Using the polr.fit object, we can calculate the mean of y for various levels of the covariates x1 and x2

fit <- ordinal_means(polr.fit, contrast1=c(0,0), contrast2=c(1,0))

summary(fit)
confint(fit)

To calculate bootstrap standard errors and bootstrap confidence intervals, one must specify the data frame used in the polr call.

fit <- ordinal_means(polr.fit, data=DF, contrast1=c(0,0), contrast2=c(1,0),
                     se.type="bootstrap", R=100, conf.level = 0.95)

summary(fit)
confint(fit)

The development version of the package can be installed from GitHub using the devtools package:

devtools::install_github("junedsiddique/ordinalTools")