A Bayesian Latent Variable Model for Analysis of Empathic Accuracy

Abstract

AbstractEmpathic accuracy (EA), defined as the ability to accurately understand the thoughts and emotions of others, has become a well-studied phenomenon in social and clinical psychology. A widely used computer-based EA paradigm compares perceivers’ ratings of targets’ feelings or affective states with the ratings of target themselves (the true ratings) and uses correlation or its monotonic transformation as a measure of EA. However, correlation has a number of notable limitations. In particular, perceivers may differ in their rating patterns, but still have similar overall correlations. To overcome the limitations, we propose a Bayesian latent variable model that decomposes EA into two separate dimensions—discrimination and variability. Discrimination measures perceivers’ sensitivity in relation to the true ratings, and variability measures the variance of random error in perceiver’s perceptions. Similar to the conventional correlation, the Bayesian model is able to measure the overall level of the association between perceiver and target, but more importantly, the Bayesian approach can provide insights into how perceivers differ in their EA. We demonstrate the advantages of the new EA measures in two case studies. The proposed Bayesian model has a simple specification and is easy to use in practice due to its straightforward implementation in popular software. The R code is included in the supplementary material.KeywordsBayesian methodologyBayesian modelBayesian estimateRandom effectRating agreementsContinuous responseEmpathic accuracyLinear modelLatent variableSubjective ratingsDiscriminationSubjective biasMonte Carlo Markov ChainJAGSModel selectionDICMusic empathySocial empathyCorrelationPsychologyPrior distribution