Kernel-Smoothing Estimation of Item Characteristic Functions for Continuous Personality Items: An Empirical Comparison with the Linear and the Continuous-Response Models

Abstract

This study used kernel-smoothing procedures to estimate the item characteristic functions (ICFs) of a set of continuous personality items. The nonparametric ICFs were compared with the ICFs estimated (a) by the linear model and (b) by Samejima’s continuous-response model. The study was based on a conditioned approach and used an error-in-variables kernel model. The assumptions of the model were approximately correct in the present conditions. The results obtained with the kernel procedure suggested that (a) the ICFs of these items were essentially sigmoid curves, as is usually assumed in parametric item response theory, and (b) for most of the items, these curves were well approximated by both the linear and the continuous-response model.