ALTERNATIVE LOGLINEAR SMOOTHING MODELS AND THEIR EFFECT ON EQUATING FUNCTION ACCURACY

Abstract

This simulation study evaluated the potential of alternative loglinear smoothing strategies for improving equipercentile equating function accuracy. These alternative strategies use cues from the sample data to make automatable and efficient improvements to model fit, either through the use of indicator functions for fitting large residuals or by averaging raw and smoothed frequencies. The strategies were studied across equating conditions based on rights-scored and formula-scored test data. Sample sizes were also manipulated. The results showed that the considered strategies produced equating functions with improved on-average accuracy but with added random variability. Of the considered alternative strategies, the frequency averaging strategy produced the most accurate equating functions for most of the evaluations done in the study. The frequency averaging strategy is recommended for circumstances where the desired loglinear model appears to fit the data poorly and where time constraints and/or data conditions make traditional modeling approaches unrealistic.