[We’re pleased to welcome authors Philippe Duverger and Xiaoyin Wang of Towson University. Duverger and Wang recently published an article in Cornell Hospitality Quarterly entitled “Capturing Relative Importance of Customer Satisfaction Drivers using Bayesian Dominance Hierarchy,” which is currently free to read for a limited time. Below, they reflect on the inspiration for conducting this research:]
What motivated you to pursue this research?
We observed that most research on the drivers of customer satisfaction (CS) used large samples, often aggregated from several month and/or several properties. Although this is a fine method to look at CS trends it is not practical at the property level for immediate action. The current methods require large samples in order to achieve sufficient power and find significant estimates in models. Unfortunately, most hotel property monthly survey yield samples of less than 100 that will make driver analyses problematic and more likely most drivers will have non-significant estimates.
We asked ourselves if there would not be a method that could circumvent the problem of property managers that want and need to address CS drivers on a monthly basis, if not on a daily basis.
What has been the most challenging aspect of conducting your research? Were there any surprising findings?
We used a Bayesian statistical framework, borrowing from several literature areas to construct a model. Bayesian statistical analysis is still a fairly new method in practice, often not well understood, and can be computationally heavy. Therefore we first needed to explain the advantages of the method in a way that was pragmatic enough because our goal in this paper was to appeal to the hospitality manager.
Bayesian statistics work from the belief that the unknown parameter is a random variable and is associated with a probability distribution (prior distribution). The information in the sample data is used to adjust the prior perception of the unknown parameter and results in the final estimation of the parameter (posterior distribution). Therefore, even if the sample is small, significance can be determined.
Maybe more pragmatically, Bayesian analysis is “often a more direct way to tackle questions we usually want to know, such as: is the hypothesis likely to be true?” Bayesian analysis does not use double negatives, such as we often encounter, e.g., “we failed to reject the null hypothesis that there is no difference.” Bayesian analysis reports are straight forward: “given these data, it is likely that the difference is X% probable” (Chapman and McDonnell Fei 2015, p. 150).
There are many other advantages that we discuss in the paper.
In what ways is your research innovative, and how do you think it will impact the field?
We believe that our Bayesian model, for which we share the code at http://tinyurl.com/kdqjf4u, could be used by hospitality properties or hospitality corporate departments, to enhance monthly reporting along with other marketing metrics, and shared via dashboards.