A comparison of Bayesian networks to principal component analysis to detect e-satisfaction factors

This research compares Bayesian networks to principal component analysis and regression models to detect the e-satisfaction attributes.The review of the e-marketing literature reveals that most researches rely on principal component and regression analyses to identify the e-satisfaction key attributes. Bayesian networks have yet been used in a marketing context to estimate service-profit chain for transport service satisfaction (Anderson, Mackoy, Thompson & Harrell, 2004) but nether applied to e-satisfaction.To compare the two methodologies, we developed two on-line questionnaires for a e-service company: Expedia, the world-wide e-travel leader. The first questionnaire deals with immediate post-purchase's satisfaction, the second one with customers' post-travel satisfaction. Respectively 8193 and 6653 people were interviewed in 2006. The interviewees had to report their overall and detailed satisfaction for the web site (first questionnaire) and their overall experience including the trip (second questionnaire). Two principal component and regressions analyses were run to identify the attributes contributing to customers' overall satisfaction and e-loyalty. We then ran Bayesian networks analyses on the two data sets. Our results confirm that Bayesian networks methodology has several advantages compared to the other approach. First it is able to explain dependent variable relationships. Secondly, it takes the non linear relationships into account. Besides, it can describe causal links using observable variables within a single multivariate model, and has the ability to conduct probabilistic inference for predictions, diagnostics. It can work with incomplete data. Finally, it provides a graphical representation which is of interest for both academics and managers.