Calibrating essay markers with the partial credit model

Geoff N. Masters
Siek Toon Khoo

Abstract

This paper shows how a range of differences in marker behaviour, including leniency/severity and central tendency, can be detected and investigated using the Rasch partial credit model. The method allows for variations in the use of rating categories by estimating a set of rating 'thresholds' for each assessor. By calibrating each assessor's rating behaviour in this way, the method automatically takes account of detected differences among assessors in the measurement process.