Bayesian Estimation of Reliability

Japan Women’s University Journal:

Faculty of Integrated Arts and Social Sciences, 2017, 28, 59-78.

Bayesian Estimation of Ordinal Categorical Items’ Reliability Coefficients

Relationship Between True Values and Observed Scores

Yasuharu Okamoto

Abstract

A model and its associated algorithms are presented for an ordinal categorical items’ reliability coefficient

that denotes the predictive power of a regression of observed scores on true values. For categorical items, reliability coefficients

given with respect to the three types of reliability, precision, consistency, and predictive power, mutually disagree in general.

The predictive power reliability coefficient denotes the relationship between observed scores and true values,

and other reliability coefficients overestimate the predictive power. The algorithms proposed for predictive reliability

were applied to hypothetical datasets, and the results show that they are successful.

Keywords: reliability, predictive power, regression, ordered categorical items, Bayesian.