Using Information Theory to Estimate Attribute Importance

• Main Authors: Jaw-Shen Wang, 王烑炫
• Other Authors: Dr. Kee-Kuo Chen
• Format: Others
• Language: en_US
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/41840205270947977795

博士 === 國立臺灣海洋大學 === 航運管理學系 === 99 === The purpose of this study is to provide an alternative approach for determining the relative importance of attributes in Importance-Performance Analysis (IPA). The traditional methods used to measure the importance of Attribute’s Importance (AI) in IPA can be classified into two broad approaches: those directly measured based on respondents and those indirectly estimated by statistical models. The current methods suffer from certain data and measurement problems. This dissertation proposes an information theory approach to estimate AI which will increase estimation quality. The proposed approach includes two models used sequentially. First, is the Generalized Maximum Entropy (GME) model that acts as a statistical model to indirectly estimate AI. Second, is the Generalized Cross Entropy (GCE) model which synthesizes information from the GME Model and from self-elicitation measures. The approach reparameterizes the IPA model using the GCE model. To assess the reduction in uncertainty in the GCE model over the GME model, we propose and define a Parameter Revision Index (PRI) that will be applied to the AI estimation results when the GCE model is used and compared those results to AI estimation results when the GME is used. The approach proposed in this dissertation and its effects are demonstrated in two ways. First, an illustration is provided composed of five hypothetical observations. This illustration will demonstrate the proposed approach on a step by step basis. Second, an empirical study is undertaken that analyzes the importance of the attributes of service encounter and price satisfaction to the customer purchase rate of carrier services of Taiwanese container shipping companies. The empirical study shows the ability of GCE to find the results more appropriate than those of the previous methods. Finally, the direction of future research is also discussed. The author believes that the approach will be promising for future business research.