conjugateprior/hertielove

Spread the love

hertielove

You have a class full of terrible grades, scored out of 100 - maybe they are percentages. You want to curve them to fit Hertie's expectations about grade distributions using a smooth transformation that maintains monotonicity, and isn't especially ad-hoc. In particular you'd like not to have to threshold or make individual grade changes. We call this the #HertieLove.

This package applies #HertieLove a class of grades. It works by taking the raw grades as ability estimates from a 2PL IRT model, and then figuring out what values of the 2PL's parameters would generate the appropriate grade quantiles and applies those to the original score estimates. This curving function is also called hertielove.

The package also has a utility function that transforms grades out of 100 onto 'German' grades, which start at 1 (the best) and move in numerically uneven but conventional intervals to 5 (the worst). This function is called degrade since it generates grades suitable for DE.

Installation

remotes::install_github("conjugateprior/hertielove")

(you may have to install the remotes package first)

Usage

grades <- rbinom(25, prob = 0.66, size = 100) # fake some grades
loved <- hertielove(grades) # curved grades

deloved <- degrade(loved) # German grades for those curved grades