Mountain Measurement, Inc.
the pinnacle of assessment

A More Informative Group Percentile Rank

By Brian Bontempo, Ph.D., Daniel Wilson, Philip Dickison, Ph.D., RN, & Ada Woo, Ph.D.

Paper to be presented at the 9th Conference of the International Test Commission in San Sebastian, Spain. July 2014.

Abstract

The purpose of this study was to explore a group performance reporting metric (aka group score) that usefully combines the concepts of ranking, scaling, and norming. This metric, called Group Percentile Ranks (GPR), is obtained by calculating a measure of central tendency for each group and determining the percentile rank for the group by comparing each groupís measure to the distribution of individual performance rather than the distribution of each groupís measure. The shape of the distribution of group percentiles is approximately normal. This technique is currently used by the US nursing licensure examination to provide feedback to nursing schools on the performance of their graduates.

The GPR method was evaluated from an empirical and theoretical perspective. Three years of data from the US nursing licensure exam were used to calculate traditional percentile ranks and group percentile ranks. The distributions were compared along the following criteria: ability to convey normative ordinal information, ability to convey normative interval information, impact of small program performance on other programs, quality of information at the extremes, and stability over time.

Given rounding, the traditional method provided more ordinal differentiation amongst moderately performing programs while GPR provided more at the extremes. Although neither method was able to successfully convey equal interval information, the GPR method conveyed distance information better than the traditional method. The traditional method was negatively impacted by the performance of small groups while GPR method was not. The GPR method was more stable than the traditional method, a result attributable to the number of individuals being far larger than the number of groups.

In conclusion, the GPR method provides a single metric which includes ranking, scaling, and norming information about a groupís performance. As with the traditional method, caution should be exercised when making inferences for small groups.