Causal Structures of Multidimensional QoE in Haptic-Audiovisual Communications: Bayesian Modeling

This article proposes a methodology for building and verifying plausible models that can express causation in multidimensional QoE for haptic-audiovisual interactive communications. For the modeling, we utilize subjective experimental data of five-point scores collected in a previous study where a pair of subjects carry out two kinds of interactive tasks (castanets hitting and object movement) in real space (not in virtual space). The multidimensional QoE is composed of 15 measures for the castanets hitting and 14 measures for the object movement. To reduce the dimension, we classify the QoE measures into three groups as indicators of three constructs (latent variables or factors): AVQ (AudioVisual Quality), HQ (Haptic Quality), and UXQ (User eXperience Quality). We then build two models: (1) a structural equation model in which AVQ and HQ correlated with each other give causal effects on UXQ, and (2) a confirmatory factor analysis model in which the three constructs are only correlated with each other. We refer to the former as 3C-SEM and the latter as 3C-CFA. We further introduce a CFA model with a single construct for which all QoE measures are its indicators (1C-CFA). We perform Bayesian analysis of the three models by means of Markov chain Monte Carlo simulation; in each model, the deviance information criterion is obtained for model comparison, and the posterior predictive p-value is calculated for model checking. As a result, we find that 3C-SEM is the most plausible and that HQ has a stronger causal effect on UXQ than AVQ. We also learn that the correlation between AVQ and UXQ is much higher than the direct causal effect and that the increase in the association as correlation is due to the causal effect of HQ on UXQ through the correlation of AVQ with HQ. Thus, it is suggested that improving haptic performance is more effective in enhancement of QoE than improving audiovisual performance.