On the misuse of manifest variables in the detection of measurement bias (Q1205777)

The paper studies the measurement bias in some manifest variable \(Y\) which is believed to be related to a latent, unobserved variable \(W\) that cannot be directly measured. All random variables are assumed to be discrete and finite. The method used in the paper introduces a second manifest variable \(Z\), which is chosen to substitute \(W\). The authors present conditions for measurement invariance, for invariance of the conditional distribution of \(Y\) given \(Z\), Bayes sufficiency of \(Z\), stochastic independence of \(Y\) and \(Z\) and the reliability of \(Z\). The limitation of the presented procedure is discussed.