Comparison of different reliability estimation methods for single-item assessment: a simulation study.

Abstract

Single-item assessments have recently become popular in various fields, and researchers have developed methods for estimating the reliability of single-item assessments, some based on factor analysis and correction for attenuation, and others using the double monotonicity model, Guttman's λ₆, or the latent class model. However, no empirical study has investigated which method best estimates the reliability of single-item assessments. This study investigated this question using a simulation study. To represent assessments as they are found in practice, the simulation study varied several aspects: the item discrimination parameter, the test length of the multi-item assessment of the same construct, the sample size, and the correlation between the single-item assessment and the multi-item assessment of the same construct. The results suggest that by using the method based on the double monotonicity model and the method based on correction for attenuation simultaneously, researchers can obtain the most precise estimate of the range of reliability of a single-item assessment in 94.44% of cases. The test length of a multi-item assessment of the same construct, the item discrimination parameter, the sample size, and the correlation between the single-item assessment and the multi-item assessment of the same construct did not influence the choice of method choice.