skip to main content skip to footer

Evaluation of the Effects of Loglinear Smoothing Models on Equating Functions in the Presence of Structured Data Irregularities

Liu, Jinghua; Moses, Tim P.; Low, Albert C.
Publication Year:
Report Number:
ETS Research Report
Document Type:
Page Count:
Subject/Key Words:
Presmoothing, Loglinear Models, Percent Relative Error, Smoothness, Systematic Irregularities, Equating Function


Presmoothing a score distribution involves finding a balance between a good representation of the original data and smoothness. Smoothing reduces sampling variability, while a good data representation reduces the possibility of bias. When the test score distributions contain systematic irregularities that are due to scoring practice (e.g., formula scoring) rather than random irregularities, this trade-off becomes complicated. The current study explored whether the systematic irregularities should be preserved or smoothed and the corresponding impacts on score equating. Two properties of the equating function were evaluated: distribution matching and smoothness. The overall results did not exhibit a consistent pattern in terms of the impacts on the equating function caused by smoothing out or preserving the systematic irregularities.

Read More